diff options
Diffstat (limited to 'src/3rdparty/masm/assembler')
25 files changed, 25353 insertions, 0 deletions
diff --git a/src/3rdparty/masm/assembler/ARMAssembler.cpp b/src/3rdparty/masm/assembler/ARMAssembler.cpp new file mode 100644 index 0000000000..6912d1ea39 --- /dev/null +++ b/src/3rdparty/masm/assembler/ARMAssembler.cpp @@ -0,0 +1,444 @@ +/* + * Copyright (C) 2009 University of Szeged + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL UNIVERSITY OF SZEGED OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "config.h" + +#if ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#include "ARMAssembler.h" + +namespace JSC { + +// Patching helpers + +void ARMAssembler::patchConstantPoolLoad(void* loadAddr, void* constPoolAddr) +{ + ARMWord *ldr = reinterpret_cast<ARMWord*>(loadAddr); + ARMWord diff = reinterpret_cast<ARMWord*>(constPoolAddr) - ldr; + ARMWord index = (*ldr & 0xfff) >> 1; + + ASSERT(diff >= 1); + if (diff >= 2 || index > 0) { + diff = (diff + index - 2) * sizeof(ARMWord); + ASSERT(diff <= 0xfff); + *ldr = (*ldr & ~0xfff) | diff; + } else + *ldr = (*ldr & ~(0xfff | ARMAssembler::DataTransferUp)) | sizeof(ARMWord); +} + +// Handle immediates + +ARMWord ARMAssembler::getOp2(ARMWord imm) +{ + int rol; + + if (imm <= 0xff) + return Op2Immediate | imm; + + if ((imm & 0xff000000) == 0) { + imm <<= 8; + rol = 8; + } + else { + imm = (imm << 24) | (imm >> 8); + rol = 0; + } + + if ((imm & 0xff000000) == 0) { + imm <<= 8; + rol += 4; + } + + if ((imm & 0xf0000000) == 0) { + imm <<= 4; + rol += 2; + } + + if ((imm & 0xc0000000) == 0) { + imm <<= 2; + rol += 1; + } + + if ((imm & 0x00ffffff) == 0) + return Op2Immediate | (imm >> 24) | (rol << 8); + + return InvalidImmediate; +} + +int ARMAssembler::genInt(int reg, ARMWord imm, bool positive) +{ + // Step1: Search a non-immediate part + ARMWord mask; + ARMWord imm1; + ARMWord imm2; + int rol; + + mask = 0xff000000; + rol = 8; + while(1) { + if ((imm & mask) == 0) { + imm = (imm << rol) | (imm >> (32 - rol)); + rol = 4 + (rol >> 1); + break; + } + rol += 2; + mask >>= 2; + if (mask & 0x3) { + // rol 8 + imm = (imm << 8) | (imm >> 24); + mask = 0xff00; + rol = 24; + while (1) { + if ((imm & mask) == 0) { + imm = (imm << rol) | (imm >> (32 - rol)); + rol = (rol >> 1) - 8; + break; + } + rol += 2; + mask >>= 2; + if (mask & 0x3) + return 0; + } + break; + } + } + + ASSERT((imm & 0xff) == 0); + + if ((imm & 0xff000000) == 0) { + imm1 = Op2Immediate | ((imm >> 16) & 0xff) | (((rol + 4) & 0xf) << 8); + imm2 = Op2Immediate | ((imm >> 8) & 0xff) | (((rol + 8) & 0xf) << 8); + } else if (imm & 0xc0000000) { + imm1 = Op2Immediate | ((imm >> 24) & 0xff) | ((rol & 0xf) << 8); + imm <<= 8; + rol += 4; + + if ((imm & 0xff000000) == 0) { + imm <<= 8; + rol += 4; + } + + if ((imm & 0xf0000000) == 0) { + imm <<= 4; + rol += 2; + } + + if ((imm & 0xc0000000) == 0) { + imm <<= 2; + rol += 1; + } + + if ((imm & 0x00ffffff) == 0) + imm2 = Op2Immediate | (imm >> 24) | ((rol & 0xf) << 8); + else + return 0; + } else { + if ((imm & 0xf0000000) == 0) { + imm <<= 4; + rol += 2; + } + + if ((imm & 0xc0000000) == 0) { + imm <<= 2; + rol += 1; + } + + imm1 = Op2Immediate | ((imm >> 24) & 0xff) | ((rol & 0xf) << 8); + imm <<= 8; + rol += 4; + + if ((imm & 0xf0000000) == 0) { + imm <<= 4; + rol += 2; + } + + if ((imm & 0xc0000000) == 0) { + imm <<= 2; + rol += 1; + } + + if ((imm & 0x00ffffff) == 0) + imm2 = Op2Immediate | (imm >> 24) | ((rol & 0xf) << 8); + else + return 0; + } + + if (positive) { + mov(reg, imm1); + orr(reg, reg, imm2); + } else { + mvn(reg, imm1); + bic(reg, reg, imm2); + } + + return 1; +} + +ARMWord ARMAssembler::getImm(ARMWord imm, int tmpReg, bool invert) +{ + ARMWord tmp; + + // Do it by 1 instruction + tmp = getOp2(imm); + if (tmp != InvalidImmediate) + return tmp; + + tmp = getOp2(~imm); + if (tmp != InvalidImmediate) { + if (invert) + return tmp | Op2InvertedImmediate; + mvn(tmpReg, tmp); + return tmpReg; + } + + return encodeComplexImm(imm, tmpReg); +} + +void ARMAssembler::moveImm(ARMWord imm, int dest) +{ + ARMWord tmp; + + // Do it by 1 instruction + tmp = getOp2(imm); + if (tmp != InvalidImmediate) { + mov(dest, tmp); + return; + } + + tmp = getOp2(~imm); + if (tmp != InvalidImmediate) { + mvn(dest, tmp); + return; + } + + encodeComplexImm(imm, dest); +} + +ARMWord ARMAssembler::encodeComplexImm(ARMWord imm, int dest) +{ +#if WTF_ARM_ARCH_AT_LEAST(7) + ARMWord tmp = getImm16Op2(imm); + if (tmp != InvalidImmediate) { + movw(dest, tmp); + return dest; + } + movw(dest, getImm16Op2(imm & 0xffff)); + movt(dest, getImm16Op2(imm >> 16)); + return dest; +#else + // Do it by 2 instruction + if (genInt(dest, imm, true)) + return dest; + if (genInt(dest, ~imm, false)) + return dest; + + ldrImmediate(dest, imm); + return dest; +#endif +} + +// Memory load/store helpers + +void ARMAssembler::dataTransfer32(DataTransferTypeA transferType, RegisterID srcDst, RegisterID base, int32_t offset) +{ + if (offset >= 0) { + if (offset <= 0xfff) + dtrUp(transferType, srcDst, base, offset); + else if (offset <= 0xfffff) { + add(ARMRegisters::S0, base, Op2Immediate | (offset >> 12) | (10 << 8)); + dtrUp(transferType, srcDst, ARMRegisters::S0, (offset & 0xfff)); + } else { + moveImm(offset, ARMRegisters::S0); + dtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); + } + } else { + if (offset >= -0xfff) + dtrDown(transferType, srcDst, base, -offset); + else if (offset >= -0xfffff) { + sub(ARMRegisters::S0, base, Op2Immediate | (-offset >> 12) | (10 << 8)); + dtrDown(transferType, srcDst, ARMRegisters::S0, (-offset & 0xfff)); + } else { + moveImm(offset, ARMRegisters::S0); + dtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); + } + } +} + +void ARMAssembler::baseIndexTransfer32(DataTransferTypeA transferType, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset) +{ + ASSERT(scale >= 0 && scale <= 3); + ARMWord op2 = lsl(index, scale); + + if (!offset) { + dtrUpRegister(transferType, srcDst, base, op2); + return; + } + + if (offset <= 0xfffff && offset >= -0xfffff) { + add(ARMRegisters::S0, base, op2); + dataTransfer32(transferType, srcDst, ARMRegisters::S0, offset); + return; + } + + moveImm(offset, ARMRegisters::S0); + add(ARMRegisters::S0, ARMRegisters::S0, op2); + dtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); +} + +void ARMAssembler::dataTransfer16(DataTransferTypeB transferType, RegisterID srcDst, RegisterID base, int32_t offset) +{ + if (offset >= 0) { + if (offset <= 0xff) + halfDtrUp(transferType, srcDst, base, getOp2Half(offset)); + else if (offset <= 0xffff) { + add(ARMRegisters::S0, base, Op2Immediate | (offset >> 8) | (12 << 8)); + halfDtrUp(transferType, srcDst, ARMRegisters::S0, getOp2Half(offset & 0xff)); + } else { + moveImm(offset, ARMRegisters::S0); + halfDtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); + } + } else { + if (offset >= -0xff) + halfDtrDown(transferType, srcDst, base, getOp2Half(-offset)); + else if (offset >= -0xffff) { + sub(ARMRegisters::S0, base, Op2Immediate | (-offset >> 8) | (12 << 8)); + halfDtrDown(transferType, srcDst, ARMRegisters::S0, getOp2Half(-offset & 0xff)); + } else { + moveImm(offset, ARMRegisters::S0); + halfDtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); + } + } +} + +void ARMAssembler::baseIndexTransfer16(DataTransferTypeB transferType, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset) +{ + if (!scale && !offset) { + halfDtrUpRegister(transferType, srcDst, base, index); + return; + } + + ARMWord op2 = lsl(index, scale); + + if (offset <= 0xffff && offset >= -0xffff) { + add(ARMRegisters::S0, base, op2); + dataTransfer16(transferType, srcDst, ARMRegisters::S0, offset); + return; + } + + moveImm(offset, ARMRegisters::S0); + add(ARMRegisters::S0, ARMRegisters::S0, op2); + halfDtrUpRegister(transferType, srcDst, base, ARMRegisters::S0); +} + +void ARMAssembler::dataTransferFloat(DataTransferTypeFloat transferType, FPRegisterID srcDst, RegisterID base, int32_t offset) +{ + // VFP cannot directly access memory that is not four-byte-aligned + if (!(offset & 0x3)) { + if (offset <= 0x3ff && offset >= 0) { + doubleDtrUp(transferType, srcDst, base, offset >> 2); + return; + } + if (offset <= 0x3ffff && offset >= 0) { + add(ARMRegisters::S0, base, Op2Immediate | (offset >> 10) | (11 << 8)); + doubleDtrUp(transferType, srcDst, ARMRegisters::S0, (offset >> 2) & 0xff); + return; + } + offset = -offset; + + if (offset <= 0x3ff && offset >= 0) { + doubleDtrDown(transferType, srcDst, base, offset >> 2); + return; + } + if (offset <= 0x3ffff && offset >= 0) { + sub(ARMRegisters::S0, base, Op2Immediate | (offset >> 10) | (11 << 8)); + doubleDtrDown(transferType, srcDst, ARMRegisters::S0, (offset >> 2) & 0xff); + return; + } + offset = -offset; + } + + moveImm(offset, ARMRegisters::S0); + add(ARMRegisters::S0, ARMRegisters::S0, base); + doubleDtrUp(transferType, srcDst, ARMRegisters::S0, 0); +} + +void ARMAssembler::baseIndexTransferFloat(DataTransferTypeFloat transferType, FPRegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset) +{ + add(ARMRegisters::S1, base, lsl(index, scale)); + dataTransferFloat(transferType, srcDst, ARMRegisters::S1, offset); +} + +PassRefPtr<ExecutableMemoryHandle> ARMAssembler::executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) +{ + // 64-bit alignment is required for next constant pool and JIT code as well + m_buffer.flushWithoutBarrier(true); + if (!m_buffer.isAligned(8)) + bkpt(0); + + RefPtr<ExecutableMemoryHandle> result = m_buffer.executableCopy(globalData, ownerUID, effort); + char* data = reinterpret_cast<char*>(result->start()); + + for (Jumps::Iterator iter = m_jumps.begin(); iter != m_jumps.end(); ++iter) { + // The last bit is set if the constant must be placed on constant pool. + int pos = (iter->m_offset) & (~0x1); + ARMWord* ldrAddr = reinterpret_cast_ptr<ARMWord*>(data + pos); + ARMWord* addr = getLdrImmAddress(ldrAddr); + if (*addr != InvalidBranchTarget) { + if (!(iter->m_offset & 1)) { + intptr_t difference = reinterpret_cast_ptr<ARMWord*>(data + *addr) - (ldrAddr + DefaultPrefetchOffset); + + if ((difference <= MaximumBranchOffsetDistance && difference >= MinimumBranchOffsetDistance)) { + *ldrAddr = B | getConditionalField(*ldrAddr) | (difference & BranchOffsetMask); + continue; + } + } + *addr = reinterpret_cast<ARMWord>(data + *addr); + } + } + + return result; +} + +#if OS(LINUX) && COMPILER(RVCT) + +__asm void ARMAssembler::cacheFlush(void* code, size_t size) +{ + ARM + push {r7} + add r1, r1, r0 + mov r7, #0xf0000 + add r7, r7, #0x2 + mov r2, #0x0 + svc #0x0 + pop {r7} + bx lr +} + +#endif + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) diff --git a/src/3rdparty/masm/assembler/ARMAssembler.h b/src/3rdparty/masm/assembler/ARMAssembler.h new file mode 100644 index 0000000000..3888226b21 --- /dev/null +++ b/src/3rdparty/masm/assembler/ARMAssembler.h @@ -0,0 +1,1129 @@ +/* + * Copyright (C) 2009, 2010 University of Szeged + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL UNIVERSITY OF SZEGED OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef ARMAssembler_h +#define ARMAssembler_h + +#if ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#include "AssemblerBufferWithConstantPool.h" +#include "JITCompilationEffort.h" +#include <wtf/Assertions.h> +namespace JSC { + + typedef uint32_t ARMWord; + + namespace ARMRegisters { + typedef enum { + r0 = 0, + r1, + r2, + r3, S0 = r3, /* Same as thumb assembler. */ + r4, + r5, + r6, + r7, + r8, + r9, + r10, + r11, + r12, S1 = r12, + r13, sp = r13, + r14, lr = r14, + r15, pc = r15 + } RegisterID; + + typedef enum { + d0, + d1, + d2, + d3, + d4, + d5, + d6, + d7, SD0 = d7, /* Same as thumb assembler. */ + d8, + d9, + d10, + d11, + d12, + d13, + d14, + d15, + d16, + d17, + d18, + d19, + d20, + d21, + d22, + d23, + d24, + d25, + d26, + d27, + d28, + d29, + d30, + d31 + } FPRegisterID; + + } // namespace ARMRegisters + + class ARMAssembler { + public: + typedef ARMRegisters::RegisterID RegisterID; + typedef ARMRegisters::FPRegisterID FPRegisterID; + typedef AssemblerBufferWithConstantPool<2048, 4, 4, ARMAssembler> ARMBuffer; + typedef SegmentedVector<AssemblerLabel, 64> Jumps; + + ARMAssembler() + : m_indexOfTailOfLastWatchpoint(1) + { + } + + // ARM conditional constants + typedef enum { + EQ = 0x00000000, // Zero + NE = 0x10000000, // Non-zero + CS = 0x20000000, + CC = 0x30000000, + MI = 0x40000000, + PL = 0x50000000, + VS = 0x60000000, + VC = 0x70000000, + HI = 0x80000000, + LS = 0x90000000, + GE = 0xa0000000, + LT = 0xb0000000, + GT = 0xc0000000, + LE = 0xd0000000, + AL = 0xe0000000 + } Condition; + + // ARM instruction constants + enum { + AND = (0x0 << 21), + EOR = (0x1 << 21), + SUB = (0x2 << 21), + RSB = (0x3 << 21), + ADD = (0x4 << 21), + ADC = (0x5 << 21), + SBC = (0x6 << 21), + RSC = (0x7 << 21), + TST = (0x8 << 21), + TEQ = (0x9 << 21), + CMP = (0xa << 21), + CMN = (0xb << 21), + ORR = (0xc << 21), + MOV = (0xd << 21), + BIC = (0xe << 21), + MVN = (0xf << 21), + MUL = 0x00000090, + MULL = 0x00c00090, + VMOV_F64 = 0x0eb00b40, + VADD_F64 = 0x0e300b00, + VDIV_F64 = 0x0e800b00, + VSUB_F64 = 0x0e300b40, + VMUL_F64 = 0x0e200b00, + VCMP_F64 = 0x0eb40b40, + VSQRT_F64 = 0x0eb10bc0, + VABS_F64 = 0x0eb00bc0, + VNEG_F64 = 0x0eb10b40, + STMDB = 0x09200000, + LDMIA = 0x08b00000, + B = 0x0a000000, + BL = 0x0b000000, + BX = 0x012fff10, + VMOV_VFP64 = 0x0c400a10, + VMOV_ARM64 = 0x0c500a10, + VMOV_VFP32 = 0x0e000a10, + VMOV_ARM32 = 0x0e100a10, + VCVT_F64_S32 = 0x0eb80bc0, + VCVT_S32_F64 = 0x0ebd0b40, + VCVT_U32_F64 = 0x0ebc0b40, + VCVT_F32_F64 = 0x0eb70bc0, + VCVT_F64_F32 = 0x0eb70ac0, + VMRS_APSR = 0x0ef1fa10, + CLZ = 0x016f0f10, + BKPT = 0xe1200070, + BLX = 0x012fff30, +#if WTF_ARM_ARCH_AT_LEAST(7) + MOVW = 0x03000000, + MOVT = 0x03400000, +#endif + NOP = 0xe1a00000, + }; + + enum { + Op2Immediate = (1 << 25), + ImmediateForHalfWordTransfer = (1 << 22), + Op2InvertedImmediate = (1 << 26), + SetConditionalCodes = (1 << 20), + Op2IsRegisterArgument = (1 << 25), + // Data transfer flags. + DataTransferUp = (1 << 23), + DataTransferWriteBack = (1 << 21), + DataTransferPostUpdate = (1 << 24), + DataTransferLoad = (1 << 20), + ByteDataTransfer = (1 << 22), + }; + + enum DataTransferTypeA { + LoadUint32 = 0x05000000 | DataTransferLoad, + LoadUint8 = 0x05400000 | DataTransferLoad, + StoreUint32 = 0x05000000, + StoreUint8 = 0x05400000, + }; + + enum DataTransferTypeB { + LoadUint16 = 0x010000b0 | DataTransferLoad, + LoadInt16 = 0x010000f0 | DataTransferLoad, + LoadInt8 = 0x010000d0 | DataTransferLoad, + StoreUint16 = 0x010000b0, + }; + + enum DataTransferTypeFloat { + LoadFloat = 0x0d000a00 | DataTransferLoad, + LoadDouble = 0x0d000b00 | DataTransferLoad, + StoreFloat = 0x0d000a00, + StoreDouble = 0x0d000b00, + }; + + // Masks of ARM instructions + enum { + BranchOffsetMask = 0x00ffffff, + ConditionalFieldMask = 0xf0000000, + DataTransferOffsetMask = 0xfff, + }; + + enum { + MinimumBranchOffsetDistance = -0x00800000, + MaximumBranchOffsetDistance = 0x007fffff, + }; + + enum { + padForAlign8 = 0x00, + padForAlign16 = 0x0000, + padForAlign32 = 0xe12fff7f // 'bkpt 0xffff' instruction. + }; + + static const ARMWord InvalidImmediate = 0xf0000000; + static const ARMWord InvalidBranchTarget = 0xffffffff; + static const int DefaultPrefetchOffset = 2; + + static const ARMWord BlxInstructionMask = 0x012fff30; + static const ARMWord LdrOrAddInstructionMask = 0x0ff00000; + static const ARMWord LdrPcImmediateInstructionMask = 0x0f7f0000; + + static const ARMWord AddImmediateInstruction = 0x02800000; + static const ARMWord BlxInstruction = 0x012fff30; + static const ARMWord LdrImmediateInstruction = 0x05900000; + static const ARMWord LdrPcImmediateInstruction = 0x051f0000; + + // Instruction formating + + void emitInstruction(ARMWord op, int rd, int rn, ARMWord op2) + { + ASSERT(((op2 & ~Op2Immediate) <= 0xfff) || (((op2 & ~ImmediateForHalfWordTransfer) <= 0xfff))); + m_buffer.putInt(op | RN(rn) | RD(rd) | op2); + } + + void emitDoublePrecisionInstruction(ARMWord op, int dd, int dn, int dm) + { + ASSERT((dd >= 0 && dd <= 31) && (dn >= 0 && dn <= 31) && (dm >= 0 && dm <= 31)); + m_buffer.putInt(op | ((dd & 0xf) << 12) | ((dd & 0x10) << (22 - 4)) + | ((dn & 0xf) << 16) | ((dn & 0x10) << (7 - 4)) + | (dm & 0xf) | ((dm & 0x10) << (5 - 4))); + } + + void emitSinglePrecisionInstruction(ARMWord op, int sd, int sn, int sm) + { + ASSERT((sd >= 0 && sd <= 31) && (sn >= 0 && sn <= 31) && (sm >= 0 && sm <= 31)); + m_buffer.putInt(op | ((sd >> 1) << 12) | ((sd & 0x1) << 22) + | ((sn >> 1) << 16) | ((sn & 0x1) << 7) + | (sm >> 1) | ((sm & 0x1) << 5)); + } + + void bitAnd(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | AND, rd, rn, op2); + } + + void bitAnds(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | AND | SetConditionalCodes, rd, rn, op2); + } + + void eor(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | EOR, rd, rn, op2); + } + + void eors(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | EOR | SetConditionalCodes, rd, rn, op2); + } + + void sub(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | SUB, rd, rn, op2); + } + + void subs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | SUB | SetConditionalCodes, rd, rn, op2); + } + + void rsb(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | RSB, rd, rn, op2); + } + + void rsbs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | RSB | SetConditionalCodes, rd, rn, op2); + } + + void add(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ADD, rd, rn, op2); + } + + void adds(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ADD | SetConditionalCodes, rd, rn, op2); + } + + void adc(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ADC, rd, rn, op2); + } + + void adcs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ADC | SetConditionalCodes, rd, rn, op2); + } + + void sbc(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | SBC, rd, rn, op2); + } + + void sbcs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | SBC | SetConditionalCodes, rd, rn, op2); + } + + void rsc(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | RSC, rd, rn, op2); + } + + void rscs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | RSC | SetConditionalCodes, rd, rn, op2); + } + + void tst(int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | TST | SetConditionalCodes, 0, rn, op2); + } + + void teq(int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | TEQ | SetConditionalCodes, 0, rn, op2); + } + + void cmp(int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | CMP | SetConditionalCodes, 0, rn, op2); + } + + void cmn(int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | CMN | SetConditionalCodes, 0, rn, op2); + } + + void orr(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ORR, rd, rn, op2); + } + + void orrs(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | ORR | SetConditionalCodes, rd, rn, op2); + } + + void mov(int rd, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | MOV, rd, ARMRegisters::r0, op2); + } + +#if WTF_ARM_ARCH_AT_LEAST(7) + void movw(int rd, ARMWord op2, Condition cc = AL) + { + ASSERT((op2 | 0xf0fff) == 0xf0fff); + m_buffer.putInt(toARMWord(cc) | MOVW | RD(rd) | op2); + } + + void movt(int rd, ARMWord op2, Condition cc = AL) + { + ASSERT((op2 | 0xf0fff) == 0xf0fff); + m_buffer.putInt(toARMWord(cc) | MOVT | RD(rd) | op2); + } +#endif + + void movs(int rd, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | MOV | SetConditionalCodes, rd, ARMRegisters::r0, op2); + } + + void bic(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | BIC, rd, rn, op2); + } + + void bics(int rd, int rn, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | BIC | SetConditionalCodes, rd, rn, op2); + } + + void mvn(int rd, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | MVN, rd, ARMRegisters::r0, op2); + } + + void mvns(int rd, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | MVN | SetConditionalCodes, rd, ARMRegisters::r0, op2); + } + + void mul(int rd, int rn, int rm, Condition cc = AL) + { + m_buffer.putInt(toARMWord(cc) | MUL | RN(rd) | RS(rn) | RM(rm)); + } + + void muls(int rd, int rn, int rm, Condition cc = AL) + { + m_buffer.putInt(toARMWord(cc) | MUL | SetConditionalCodes | RN(rd) | RS(rn) | RM(rm)); + } + + void mull(int rdhi, int rdlo, int rn, int rm, Condition cc = AL) + { + m_buffer.putInt(toARMWord(cc) | MULL | RN(rdhi) | RD(rdlo) | RS(rn) | RM(rm)); + } + + void vmov_f64(int dd, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VMOV_F64, dd, 0, dm); + } + + void vadd_f64(int dd, int dn, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VADD_F64, dd, dn, dm); + } + + void vdiv_f64(int dd, int dn, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VDIV_F64, dd, dn, dm); + } + + void vsub_f64(int dd, int dn, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VSUB_F64, dd, dn, dm); + } + + void vmul_f64(int dd, int dn, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VMUL_F64, dd, dn, dm); + } + + void vcmp_f64(int dd, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VCMP_F64, dd, 0, dm); + } + + void vsqrt_f64(int dd, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VSQRT_F64, dd, 0, dm); + } + + void vabs_f64(int dd, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VABS_F64, dd, 0, dm); + } + + void vneg_f64(int dd, int dm, Condition cc = AL) + { + emitDoublePrecisionInstruction(toARMWord(cc) | VNEG_F64, dd, 0, dm); + } + + void ldrImmediate(int rd, ARMWord imm, Condition cc = AL) + { + m_buffer.putIntWithConstantInt(toARMWord(cc) | LoadUint32 | DataTransferUp | RN(ARMRegisters::pc) | RD(rd), imm, true); + } + + void ldrUniqueImmediate(int rd, ARMWord imm, Condition cc = AL) + { + m_buffer.putIntWithConstantInt(toARMWord(cc) | LoadUint32 | DataTransferUp | RN(ARMRegisters::pc) | RD(rd), imm); + } + + void dtrUp(DataTransferTypeA transferType, int rd, int rb, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType | DataTransferUp, rd, rb, op2); + } + + void dtrUpRegister(DataTransferTypeA transferType, int rd, int rb, int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType | DataTransferUp | Op2IsRegisterArgument, rd, rb, rm); + } + + void dtrDown(DataTransferTypeA transferType, int rd, int rb, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType, rd, rb, op2); + } + + void dtrDownRegister(DataTransferTypeA transferType, int rd, int rb, int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType | Op2IsRegisterArgument, rd, rb, rm); + } + + void halfDtrUp(DataTransferTypeB transferType, int rd, int rb, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType | DataTransferUp, rd, rb, op2); + } + + void halfDtrUpRegister(DataTransferTypeB transferType, int rd, int rn, int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType | DataTransferUp, rd, rn, rm); + } + + void halfDtrDown(DataTransferTypeB transferType, int rd, int rb, ARMWord op2, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType, rd, rb, op2); + } + + void halfDtrDownRegister(DataTransferTypeB transferType, int rd, int rn, int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | transferType, rd, rn, rm); + } + + void doubleDtrUp(DataTransferTypeFloat type, int rd, int rb, ARMWord op2, Condition cc = AL) + { + ASSERT(op2 <= 0xff && rd <= 15); + /* Only d0-d15 and s0, s2, s4 ... s30 are supported. */ + m_buffer.putInt(toARMWord(cc) | DataTransferUp | type | (rd << 12) | RN(rb) | op2); + } + + void doubleDtrDown(DataTransferTypeFloat type, int rd, int rb, ARMWord op2, Condition cc = AL) + { + ASSERT(op2 <= 0xff && rd <= 15); + /* Only d0-d15 and s0, s2, s4 ... s30 are supported. */ + m_buffer.putInt(toARMWord(cc) | type | (rd << 12) | RN(rb) | op2); + } + + void push(int reg, Condition cc = AL) + { + ASSERT(ARMWord(reg) <= 0xf); + m_buffer.putInt(toARMWord(cc) | StoreUint32 | DataTransferWriteBack | RN(ARMRegisters::sp) | RD(reg) | 0x4); + } + + void pop(int reg, Condition cc = AL) + { + ASSERT(ARMWord(reg) <= 0xf); + m_buffer.putInt(toARMWord(cc) | (LoadUint32 ^ DataTransferPostUpdate) | DataTransferUp | RN(ARMRegisters::sp) | RD(reg) | 0x4); + } + + inline void poke(int reg, Condition cc = AL) + { + dtrDown(StoreUint32, ARMRegisters::sp, 0, reg, cc); + } + + inline void peek(int reg, Condition cc = AL) + { + dtrUp(LoadUint32, reg, ARMRegisters::sp, 0, cc); + } + + void vmov_vfp64(int sm, int rt, int rt2, Condition cc = AL) + { + ASSERT(rt != rt2); + m_buffer.putInt(toARMWord(cc) | VMOV_VFP64 | RN(rt2) | RD(rt) | (sm & 0xf) | ((sm & 0x10) << (5 - 4))); + } + + void vmov_arm64(int rt, int rt2, int sm, Condition cc = AL) + { + ASSERT(rt != rt2); + m_buffer.putInt(toARMWord(cc) | VMOV_ARM64 | RN(rt2) | RD(rt) | (sm & 0xf) | ((sm & 0x10) << (5 - 4))); + } + + void vmov_vfp32(int sn, int rt, Condition cc = AL) + { + ASSERT(rt <= 15); + emitSinglePrecisionInstruction(toARMWord(cc) | VMOV_VFP32, rt << 1, sn, 0); + } + + void vmov_arm32(int rt, int sn, Condition cc = AL) + { + ASSERT(rt <= 15); + emitSinglePrecisionInstruction(toARMWord(cc) | VMOV_ARM32, rt << 1, sn, 0); + } + + void vcvt_f64_s32(int dd, int sm, Condition cc = AL) + { + ASSERT(!(sm & 0x1)); // sm must be divisible by 2 + emitDoublePrecisionInstruction(toARMWord(cc) | VCVT_F64_S32, dd, 0, (sm >> 1)); + } + + void vcvt_s32_f64(int sd, int dm, Condition cc = AL) + { + ASSERT(!(sd & 0x1)); // sd must be divisible by 2 + emitDoublePrecisionInstruction(toARMWord(cc) | VCVT_S32_F64, (sd >> 1), 0, dm); + } + + void vcvt_u32_f64(int sd, int dm, Condition cc = AL) + { + ASSERT(!(sd & 0x1)); // sd must be divisible by 2 + emitDoublePrecisionInstruction(toARMWord(cc) | VCVT_U32_F64, (sd >> 1), 0, dm); + } + + void vcvt_f64_f32(int dd, int sm, Condition cc = AL) + { + ASSERT(dd <= 15 && sm <= 15); + emitDoublePrecisionInstruction(toARMWord(cc) | VCVT_F64_F32, dd, 0, sm); + } + + void vcvt_f32_f64(int dd, int sm, Condition cc = AL) + { + ASSERT(dd <= 15 && sm <= 15); + emitDoublePrecisionInstruction(toARMWord(cc) | VCVT_F32_F64, dd, 0, sm); + } + + void vmrs_apsr(Condition cc = AL) + { + m_buffer.putInt(toARMWord(cc) | VMRS_APSR); + } + + void clz(int rd, int rm, Condition cc = AL) + { + m_buffer.putInt(toARMWord(cc) | CLZ | RD(rd) | RM(rm)); + } + + void bkpt(ARMWord value) + { + m_buffer.putInt(BKPT | ((value & 0xff0) << 4) | (value & 0xf)); + } + + void nop() + { + m_buffer.putInt(NOP); + } + + void bx(int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | BX, 0, 0, RM(rm)); + } + + AssemblerLabel blx(int rm, Condition cc = AL) + { + emitInstruction(toARMWord(cc) | BLX, 0, 0, RM(rm)); + return m_buffer.label(); + } + + static ARMWord lsl(int reg, ARMWord value) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(value <= 0x1f); + return reg | (value << 7) | 0x00; + } + + static ARMWord lsr(int reg, ARMWord value) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(value <= 0x1f); + return reg | (value << 7) | 0x20; + } + + static ARMWord asr(int reg, ARMWord value) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(value <= 0x1f); + return reg | (value << 7) | 0x40; + } + + static ARMWord lslRegister(int reg, int shiftReg) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(shiftReg <= ARMRegisters::pc); + return reg | (shiftReg << 8) | 0x10; + } + + static ARMWord lsrRegister(int reg, int shiftReg) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(shiftReg <= ARMRegisters::pc); + return reg | (shiftReg << 8) | 0x30; + } + + static ARMWord asrRegister(int reg, int shiftReg) + { + ASSERT(reg <= ARMRegisters::pc); + ASSERT(shiftReg <= ARMRegisters::pc); + return reg | (shiftReg << 8) | 0x50; + } + + // General helpers + + size_t codeSize() const + { + return m_buffer.codeSize(); + } + + void ensureSpace(int insnSpace, int constSpace) + { + m_buffer.ensureSpace(insnSpace, constSpace); + } + + int sizeOfConstantPool() + { + return m_buffer.sizeOfConstantPool(); + } + + AssemblerLabel labelIgnoringWatchpoints() + { + m_buffer.ensureSpaceForAnyInstruction(); + return m_buffer.label(); + } + + AssemblerLabel labelForWatchpoint() + { + m_buffer.ensureSpaceForAnyInstruction(maxJumpReplacementSize() / sizeof(ARMWord)); + AssemblerLabel result = m_buffer.label(); + if (result.m_offset != (m_indexOfTailOfLastWatchpoint - maxJumpReplacementSize())) + result = label(); + m_indexOfTailOfLastWatchpoint = result.m_offset + maxJumpReplacementSize(); + return label(); + } + + AssemblerLabel label() + { + AssemblerLabel result = labelIgnoringWatchpoints(); + while (result.m_offset + 1 < m_indexOfTailOfLastWatchpoint) { + nop(); + // The available number of instructions are ensured by labelForWatchpoint. + result = m_buffer.label(); + } + return result; + } + + AssemblerLabel align(int alignment) + { + while (!m_buffer.isAligned(alignment)) + mov(ARMRegisters::r0, ARMRegisters::r0); + + return label(); + } + + AssemblerLabel loadBranchTarget(int rd, Condition cc = AL, int useConstantPool = 0) + { + ensureSpace(sizeof(ARMWord), sizeof(ARMWord)); + m_jumps.append(m_buffer.codeSize() | (useConstantPool & 0x1)); + ldrUniqueImmediate(rd, InvalidBranchTarget, cc); + return m_buffer.label(); + } + + AssemblerLabel jmp(Condition cc = AL, int useConstantPool = 0) + { + return loadBranchTarget(ARMRegisters::pc, cc, useConstantPool); + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData&, void* ownerUID, JITCompilationEffort); + + unsigned debugOffset() { return m_buffer.debugOffset(); } + + // DFG assembly helpers for moving data between fp and registers. + void vmov(RegisterID rd1, RegisterID rd2, FPRegisterID rn) + { + vmov_arm64(rd1, rd2, rn); + } + + void vmov(FPRegisterID rd, RegisterID rn1, RegisterID rn2) + { + vmov_vfp64(rd, rn1, rn2); + } + + // Patching helpers + + static ARMWord* getLdrImmAddress(ARMWord* insn) + { + // Check for call + if ((*insn & LdrPcImmediateInstructionMask) != LdrPcImmediateInstruction) { + // Must be BLX + ASSERT((*insn & BlxInstructionMask) == BlxInstruction); + insn--; + } + + // Must be an ldr ..., [pc +/- imm] + ASSERT((*insn & LdrPcImmediateInstructionMask) == LdrPcImmediateInstruction); + + ARMWord addr = reinterpret_cast<ARMWord>(insn) + DefaultPrefetchOffset * sizeof(ARMWord); + if (*insn & DataTransferUp) + return reinterpret_cast<ARMWord*>(addr + (*insn & DataTransferOffsetMask)); + return reinterpret_cast<ARMWord*>(addr - (*insn & DataTransferOffsetMask)); + } + + static ARMWord* getLdrImmAddressOnPool(ARMWord* insn, uint32_t* constPool) + { + // Must be an ldr ..., [pc +/- imm] + ASSERT((*insn & LdrPcImmediateInstructionMask) == LdrPcImmediateInstruction); + + if (*insn & 0x1) + return reinterpret_cast<ARMWord*>(constPool + ((*insn & DataTransferOffsetMask) >> 1)); + return getLdrImmAddress(insn); + } + + static void patchPointerInternal(intptr_t from, void* to) + { + ARMWord* insn = reinterpret_cast<ARMWord*>(from); + ARMWord* addr = getLdrImmAddress(insn); + *addr = reinterpret_cast<ARMWord>(to); + } + + static ARMWord patchConstantPoolLoad(ARMWord load, ARMWord value) + { + value = (value << 1) + 1; + ASSERT(!(value & ~DataTransferOffsetMask)); + return (load & ~DataTransferOffsetMask) | value; + } + + static void patchConstantPoolLoad(void* loadAddr, void* constPoolAddr); + + // Read pointers + static void* readPointer(void* from) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(from); + ARMWord* address = getLdrImmAddress(instruction); + return *reinterpret_cast<void**>(address); + } + + // Patch pointers + + static void linkPointer(void* code, AssemblerLabel from, void* to) + { + patchPointerInternal(reinterpret_cast<intptr_t>(code) + from.m_offset, to); + } + + static void repatchInt32(void* where, int32_t to) + { + patchPointerInternal(reinterpret_cast<intptr_t>(where), reinterpret_cast<void*>(to)); + } + + static void repatchCompact(void* where, int32_t value) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(where); + ASSERT((*instruction & 0x0f700000) == LoadUint32); + if (value >= 0) + *instruction = (*instruction & 0xff7ff000) | DataTransferUp | value; + else + *instruction = (*instruction & 0xff7ff000) | -value; + cacheFlush(instruction, sizeof(ARMWord)); + } + + static void repatchPointer(void* from, void* to) + { + patchPointerInternal(reinterpret_cast<intptr_t>(from), to); + } + + // Linkers + static intptr_t getAbsoluteJumpAddress(void* base, int offset = 0) + { + return reinterpret_cast<intptr_t>(base) + offset - sizeof(ARMWord); + } + + void linkJump(AssemblerLabel from, AssemblerLabel to) + { + ARMWord* insn = reinterpret_cast<ARMWord*>(getAbsoluteJumpAddress(m_buffer.data(), from.m_offset)); + ARMWord* addr = getLdrImmAddressOnPool(insn, m_buffer.poolAddress()); + *addr = toARMWord(to.m_offset); + } + + static void linkJump(void* code, AssemblerLabel from, void* to) + { + patchPointerInternal(getAbsoluteJumpAddress(code, from.m_offset), to); + } + + static void relinkJump(void* from, void* to) + { + patchPointerInternal(getAbsoluteJumpAddress(from), to); + } + + static void linkCall(void* code, AssemblerLabel from, void* to) + { + patchPointerInternal(getAbsoluteJumpAddress(code, from.m_offset), to); + } + + static void relinkCall(void* from, void* to) + { + patchPointerInternal(getAbsoluteJumpAddress(from), to); + } + + static void* readCallTarget(void* from) + { + return reinterpret_cast<void*>(readPointer(reinterpret_cast<void*>(getAbsoluteJumpAddress(from)))); + } + + static void replaceWithJump(void* instructionStart, void* to) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(instructionStart); + intptr_t difference = reinterpret_cast<intptr_t>(to) - (reinterpret_cast<intptr_t>(instruction) + DefaultPrefetchOffset * sizeof(ARMWord)); + + if (!(difference & 1)) { + difference >>= 2; + if ((difference <= MaximumBranchOffsetDistance && difference >= MinimumBranchOffsetDistance)) { + // Direct branch. + instruction[0] = B | AL | (difference & BranchOffsetMask); + cacheFlush(instruction, sizeof(ARMWord)); + return; + } + } + + // Load target. + instruction[0] = LoadUint32 | AL | RN(ARMRegisters::pc) | RD(ARMRegisters::pc) | 4; + instruction[1] = reinterpret_cast<ARMWord>(to); + cacheFlush(instruction, sizeof(ARMWord) * 2); + } + + static ptrdiff_t maxJumpReplacementSize() + { + return sizeof(ARMWord) * 2; + } + + static void replaceWithLoad(void* instructionStart) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(instructionStart); + cacheFlush(instruction, sizeof(ARMWord)); + + ASSERT((*instruction & LdrOrAddInstructionMask) == AddImmediateInstruction || (*instruction & LdrOrAddInstructionMask) == LdrImmediateInstruction); + if ((*instruction & LdrOrAddInstructionMask) == AddImmediateInstruction) { + *instruction = (*instruction & ~LdrOrAddInstructionMask) | LdrImmediateInstruction; + cacheFlush(instruction, sizeof(ARMWord)); + } + } + + static void replaceWithAddressComputation(void* instructionStart) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(instructionStart); + cacheFlush(instruction, sizeof(ARMWord)); + + ASSERT((*instruction & LdrOrAddInstructionMask) == AddImmediateInstruction || (*instruction & LdrOrAddInstructionMask) == LdrImmediateInstruction); + if ((*instruction & LdrOrAddInstructionMask) == LdrImmediateInstruction) { + *instruction = (*instruction & ~LdrOrAddInstructionMask) | AddImmediateInstruction; + cacheFlush(instruction, sizeof(ARMWord)); + } + } + + static void revertBranchPtrWithPatch(void* instructionStart, RegisterID rn, ARMWord imm) + { + ARMWord* instruction = reinterpret_cast<ARMWord*>(instructionStart); + + ASSERT((instruction[2] & LdrPcImmediateInstructionMask) == LdrPcImmediateInstruction); + instruction[0] = toARMWord(AL) | ((instruction[2] & 0x0fff0fff) + sizeof(ARMWord)) | RD(ARMRegisters::S1); + *getLdrImmAddress(instruction) = imm; + instruction[1] = toARMWord(AL) | CMP | SetConditionalCodes | RN(rn) | RM(ARMRegisters::S1); + cacheFlush(instruction, 2 * sizeof(ARMWord)); + } + + // Address operations + + static void* getRelocatedAddress(void* code, AssemblerLabel label) + { + return reinterpret_cast<void*>(reinterpret_cast<char*>(code) + label.m_offset); + } + + // Address differences + + static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b) + { + return b.m_offset - a.m_offset; + } + + static unsigned getCallReturnOffset(AssemblerLabel call) + { + return call.m_offset; + } + + // Handle immediates + + static ARMWord getOp2(ARMWord imm); + + // Fast case if imm is known to be between 0 and 0xff + static ARMWord getOp2Byte(ARMWord imm) + { + ASSERT(imm <= 0xff); + return Op2Immediate | imm; + } + + static ARMWord getOp2Half(ARMWord imm) + { + ASSERT(imm <= 0xff); + return ImmediateForHalfWordTransfer | (imm & 0x0f) | ((imm & 0xf0) << 4); + } + +#if WTF_ARM_ARCH_AT_LEAST(7) + static ARMWord getImm16Op2(ARMWord imm) + { + if (imm <= 0xffff) + return (imm & 0xf000) << 4 | (imm & 0xfff); + return InvalidImmediate; + } +#endif + ARMWord getImm(ARMWord imm, int tmpReg, bool invert = false); + void moveImm(ARMWord imm, int dest); + ARMWord encodeComplexImm(ARMWord imm, int dest); + + // Memory load/store helpers + + void dataTransfer32(DataTransferTypeA, RegisterID srcDst, RegisterID base, int32_t offset); + void baseIndexTransfer32(DataTransferTypeA, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset); + void dataTransfer16(DataTransferTypeB, RegisterID srcDst, RegisterID base, int32_t offset); + void baseIndexTransfer16(DataTransferTypeB, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset); + void dataTransferFloat(DataTransferTypeFloat, FPRegisterID srcDst, RegisterID base, int32_t offset); + void baseIndexTransferFloat(DataTransferTypeFloat, FPRegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset); + + // Constant pool hnadlers + + static ARMWord placeConstantPoolBarrier(int offset) + { + offset = (offset - sizeof(ARMWord)) >> 2; + ASSERT((offset <= MaximumBranchOffsetDistance && offset >= MinimumBranchOffsetDistance)); + return AL | B | (offset & BranchOffsetMask); + } + +#if OS(LINUX) && COMPILER(GCC) + static inline void linuxPageFlush(uintptr_t begin, uintptr_t end) + { + asm volatile( + "push {r7}\n" + "mov r0, %0\n" + "mov r1, %1\n" + "mov r7, #0xf0000\n" + "add r7, r7, #0x2\n" + "mov r2, #0x0\n" + "svc 0x0\n" + "pop {r7}\n" + : + : "r" (begin), "r" (end) + : "r0", "r1", "r2"); + } +#endif + +#if OS(LINUX) && COMPILER(RVCT) + static __asm void cacheFlush(void* code, size_t); +#else + static void cacheFlush(void* code, size_t size) + { +#if OS(LINUX) && COMPILER(GCC) + size_t page = pageSize(); + uintptr_t current = reinterpret_cast<uintptr_t>(code); + uintptr_t end = current + size; + uintptr_t firstPageEnd = (current & ~(page - 1)) + page; + + if (end <= firstPageEnd) { + linuxPageFlush(current, end); + return; + } + + linuxPageFlush(current, firstPageEnd); + + for (current = firstPageEnd; current + page < end; current += page) + linuxPageFlush(current, current + page); + + linuxPageFlush(current, end); +#elif OS(WINCE) + CacheRangeFlush(code, size, CACHE_SYNC_ALL); +#elif OS(QNX) && ENABLE(ASSEMBLER_WX_EXCLUSIVE) + UNUSED_PARAM(code); + UNUSED_PARAM(size); +#elif OS(QNX) + msync(code, size, MS_INVALIDATE_ICACHE); +#else +#error "The cacheFlush support is missing on this platform." +#endif + } +#endif + + private: + static ARMWord RM(int reg) + { + ASSERT(reg <= ARMRegisters::pc); + return reg; + } + + static ARMWord RS(int reg) + { + ASSERT(reg <= ARMRegisters::pc); + return reg << 8; + } + + static ARMWord RD(int reg) + { + ASSERT(reg <= ARMRegisters::pc); + return reg << 12; + } + + static ARMWord RN(int reg) + { + ASSERT(reg <= ARMRegisters::pc); + return reg << 16; + } + + static ARMWord getConditionalField(ARMWord i) + { + return i & ConditionalFieldMask; + } + + static ARMWord toARMWord(Condition cc) + { + return static_cast<ARMWord>(cc); + } + + static ARMWord toARMWord(uint32_t u) + { + return static_cast<ARMWord>(u); + } + + int genInt(int reg, ARMWord imm, bool positive); + + ARMBuffer m_buffer; + Jumps m_jumps; + uint32_t m_indexOfTailOfLastWatchpoint; + }; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#endif // ARMAssembler_h diff --git a/src/3rdparty/masm/assembler/ARMv7Assembler.cpp b/src/3rdparty/masm/assembler/ARMv7Assembler.cpp new file mode 100644 index 0000000000..faca66421b --- /dev/null +++ b/src/3rdparty/masm/assembler/ARMv7Assembler.cpp @@ -0,0 +1,36 @@ +/* + * Copyright (C) 2010 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS'' + * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, + * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS + * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF + * THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "config.h" + +#if ENABLE(ASSEMBLER) && CPU(ARM_THUMB2) + +#include "ARMv7Assembler.h" + +namespace JSC { + +} + +#endif diff --git a/src/3rdparty/masm/assembler/ARMv7Assembler.h b/src/3rdparty/masm/assembler/ARMv7Assembler.h new file mode 100644 index 0000000000..7dcf656921 --- /dev/null +++ b/src/3rdparty/masm/assembler/ARMv7Assembler.h @@ -0,0 +1,2790 @@ +/* + * Copyright (C) 2009, 2010, 2012, 2013 Apple Inc. All rights reserved. + * Copyright (C) 2010 University of Szeged + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef ARMAssembler_h +#define ARMAssembler_h + +#if ENABLE(ASSEMBLER) && CPU(ARM_THUMB2) + +#include "AssemblerBuffer.h" +#include <wtf/Assertions.h> +#include <wtf/Vector.h> +#include <stdint.h> + +namespace JSC { + +namespace ARMRegisters { + typedef enum { + r0, + r1, + r2, + r3, + r4, + r5, + r6, + r7, wr = r7, // thumb work register + r8, + r9, sb = r9, // static base + r10, sl = r10, // stack limit + r11, fp = r11, // frame pointer + r12, ip = r12, + r13, sp = r13, + r14, lr = r14, + r15, pc = r15, + } RegisterID; + + typedef enum { + s0, + s1, + s2, + s3, + s4, + s5, + s6, + s7, + s8, + s9, + s10, + s11, + s12, + s13, + s14, + s15, + s16, + s17, + s18, + s19, + s20, + s21, + s22, + s23, + s24, + s25, + s26, + s27, + s28, + s29, + s30, + s31, + } FPSingleRegisterID; + + typedef enum { + d0, + d1, + d2, + d3, + d4, + d5, + d6, + d7, + d8, + d9, + d10, + d11, + d12, + d13, + d14, + d15, + d16, + d17, + d18, + d19, + d20, + d21, + d22, + d23, + d24, + d25, + d26, + d27, + d28, + d29, + d30, + d31, + } FPDoubleRegisterID; + + typedef enum { + q0, + q1, + q2, + q3, + q4, + q5, + q6, + q7, + q8, + q9, + q10, + q11, + q12, + q13, + q14, + q15, + q16, + q17, + q18, + q19, + q20, + q21, + q22, + q23, + q24, + q25, + q26, + q27, + q28, + q29, + q30, + q31, + } FPQuadRegisterID; + + inline FPSingleRegisterID asSingle(FPDoubleRegisterID reg) + { + ASSERT(reg < d16); + return (FPSingleRegisterID)(reg << 1); + } + + inline FPDoubleRegisterID asDouble(FPSingleRegisterID reg) + { + ASSERT(!(reg & 1)); + return (FPDoubleRegisterID)(reg >> 1); + } +} + +class ARMv7Assembler; +class ARMThumbImmediate { + friend class ARMv7Assembler; + + typedef uint8_t ThumbImmediateType; + static const ThumbImmediateType TypeInvalid = 0; + static const ThumbImmediateType TypeEncoded = 1; + static const ThumbImmediateType TypeUInt16 = 2; + + typedef union { + int16_t asInt; + struct { + unsigned imm8 : 8; + unsigned imm3 : 3; + unsigned i : 1; + unsigned imm4 : 4; + }; + // If this is an encoded immediate, then it may describe a shift, or a pattern. + struct { + unsigned shiftValue7 : 7; + unsigned shiftAmount : 5; + }; + struct { + unsigned immediate : 8; + unsigned pattern : 4; + }; + } ThumbImmediateValue; + + // byte0 contains least significant bit; not using an array to make client code endian agnostic. + typedef union { + int32_t asInt; + struct { + uint8_t byte0; + uint8_t byte1; + uint8_t byte2; + uint8_t byte3; + }; + } PatternBytes; + + ALWAYS_INLINE static void countLeadingZerosPartial(uint32_t& value, int32_t& zeros, const int N) + { + if (value & ~((1 << N) - 1)) /* check for any of the top N bits (of 2N bits) are set */ + value >>= N; /* if any were set, lose the bottom N */ + else /* if none of the top N bits are set, */ + zeros += N; /* then we have identified N leading zeros */ + } + + static int32_t countLeadingZeros(uint32_t value) + { + if (!value) + return 32; + + int32_t zeros = 0; + countLeadingZerosPartial(value, zeros, 16); + countLeadingZerosPartial(value, zeros, 8); + countLeadingZerosPartial(value, zeros, 4); + countLeadingZerosPartial(value, zeros, 2); + countLeadingZerosPartial(value, zeros, 1); + return zeros; + } + + ARMThumbImmediate() + : m_type(TypeInvalid) + { + m_value.asInt = 0; + } + + ARMThumbImmediate(ThumbImmediateType type, ThumbImmediateValue value) + : m_type(type) + , m_value(value) + { + } + + ARMThumbImmediate(ThumbImmediateType type, uint16_t value) + : m_type(TypeUInt16) + { + // Make sure this constructor is only reached with type TypeUInt16; + // this extra parameter makes the code a little clearer by making it + // explicit at call sites which type is being constructed + ASSERT_UNUSED(type, type == TypeUInt16); + + m_value.asInt = value; + } + +public: + static ARMThumbImmediate makeEncodedImm(uint32_t value) + { + ThumbImmediateValue encoding; + encoding.asInt = 0; + + // okay, these are easy. + if (value < 256) { + encoding.immediate = value; + encoding.pattern = 0; + return ARMThumbImmediate(TypeEncoded, encoding); + } + + int32_t leadingZeros = countLeadingZeros(value); + // if there were 24 or more leading zeros, then we'd have hit the (value < 256) case. + ASSERT(leadingZeros < 24); + + // Given a number with bit fields Z:B:C, where count(Z)+count(B)+count(C) == 32, + // Z are the bits known zero, B is the 8-bit immediate, C are the bits to check for + // zero. count(B) == 8, so the count of bits to be checked is 24 - count(Z). + int32_t rightShiftAmount = 24 - leadingZeros; + if (value == ((value >> rightShiftAmount) << rightShiftAmount)) { + // Shift the value down to the low byte position. The assign to + // shiftValue7 drops the implicit top bit. + encoding.shiftValue7 = value >> rightShiftAmount; + // The endoded shift amount is the magnitude of a right rotate. + encoding.shiftAmount = 8 + leadingZeros; + return ARMThumbImmediate(TypeEncoded, encoding); + } + + PatternBytes bytes; + bytes.asInt = value; + + if ((bytes.byte0 == bytes.byte1) && (bytes.byte0 == bytes.byte2) && (bytes.byte0 == bytes.byte3)) { + encoding.immediate = bytes.byte0; + encoding.pattern = 3; + return ARMThumbImmediate(TypeEncoded, encoding); + } + + if ((bytes.byte0 == bytes.byte2) && !(bytes.byte1 | bytes.byte3)) { + encoding.immediate = bytes.byte0; + encoding.pattern = 1; + return ARMThumbImmediate(TypeEncoded, encoding); + } + + if ((bytes.byte1 == bytes.byte3) && !(bytes.byte0 | bytes.byte2)) { + encoding.immediate = bytes.byte1; + encoding.pattern = 2; + return ARMThumbImmediate(TypeEncoded, encoding); + } + + return ARMThumbImmediate(); + } + + static ARMThumbImmediate makeUInt12(int32_t value) + { + return (!(value & 0xfffff000)) + ? ARMThumbImmediate(TypeUInt16, (uint16_t)value) + : ARMThumbImmediate(); + } + + static ARMThumbImmediate makeUInt12OrEncodedImm(int32_t value) + { + // If this is not a 12-bit unsigned it, try making an encoded immediate. + return (!(value & 0xfffff000)) + ? ARMThumbImmediate(TypeUInt16, (uint16_t)value) + : makeEncodedImm(value); + } + + // The 'make' methods, above, return a !isValid() value if the argument + // cannot be represented as the requested type. This methods is called + // 'get' since the argument can always be represented. + static ARMThumbImmediate makeUInt16(uint16_t value) + { + return ARMThumbImmediate(TypeUInt16, value); + } + + bool isValid() + { + return m_type != TypeInvalid; + } + + uint16_t asUInt16() const { return m_value.asInt; } + + // These methods rely on the format of encoded byte values. + bool isUInt3() { return !(m_value.asInt & 0xfff8); } + bool isUInt4() { return !(m_value.asInt & 0xfff0); } + bool isUInt5() { return !(m_value.asInt & 0xffe0); } + bool isUInt6() { return !(m_value.asInt & 0xffc0); } + bool isUInt7() { return !(m_value.asInt & 0xff80); } + bool isUInt8() { return !(m_value.asInt & 0xff00); } + bool isUInt9() { return (m_type == TypeUInt16) && !(m_value.asInt & 0xfe00); } + bool isUInt10() { return (m_type == TypeUInt16) && !(m_value.asInt & 0xfc00); } + bool isUInt12() { return (m_type == TypeUInt16) && !(m_value.asInt & 0xf000); } + bool isUInt16() { return m_type == TypeUInt16; } + uint8_t getUInt3() { ASSERT(isUInt3()); return m_value.asInt; } + uint8_t getUInt4() { ASSERT(isUInt4()); return m_value.asInt; } + uint8_t getUInt5() { ASSERT(isUInt5()); return m_value.asInt; } + uint8_t getUInt6() { ASSERT(isUInt6()); return m_value.asInt; } + uint8_t getUInt7() { ASSERT(isUInt7()); return m_value.asInt; } + uint8_t getUInt8() { ASSERT(isUInt8()); return m_value.asInt; } + uint16_t getUInt9() { ASSERT(isUInt9()); return m_value.asInt; } + uint16_t getUInt10() { ASSERT(isUInt10()); return m_value.asInt; } + uint16_t getUInt12() { ASSERT(isUInt12()); return m_value.asInt; } + uint16_t getUInt16() { ASSERT(isUInt16()); return m_value.asInt; } + + bool isEncodedImm() { return m_type == TypeEncoded; } + +private: + ThumbImmediateType m_type; + ThumbImmediateValue m_value; +}; + +typedef enum { + SRType_LSL, + SRType_LSR, + SRType_ASR, + SRType_ROR, + + SRType_RRX = SRType_ROR +} ARMShiftType; + +class ShiftTypeAndAmount { + friend class ARMv7Assembler; + +public: + ShiftTypeAndAmount() + { + m_u.type = (ARMShiftType)0; + m_u.amount = 0; + } + + ShiftTypeAndAmount(ARMShiftType type, unsigned amount) + { + m_u.type = type; + m_u.amount = amount & 31; + } + + unsigned lo4() { return m_u.lo4; } + unsigned hi4() { return m_u.hi4; } + +private: + union { + struct { + unsigned lo4 : 4; + unsigned hi4 : 4; + }; + struct { + unsigned type : 2; + unsigned amount : 6; + }; + } m_u; +}; + +class ARMv7Assembler { +public: + typedef ARMRegisters::RegisterID RegisterID; + typedef ARMRegisters::FPSingleRegisterID FPSingleRegisterID; + typedef ARMRegisters::FPDoubleRegisterID FPDoubleRegisterID; + typedef ARMRegisters::FPQuadRegisterID FPQuadRegisterID; + + // (HS, LO, HI, LS) -> (AE, B, A, BE) + // (VS, VC) -> (O, NO) + typedef enum { + ConditionEQ, + ConditionNE, + ConditionHS, ConditionCS = ConditionHS, + ConditionLO, ConditionCC = ConditionLO, + ConditionMI, + ConditionPL, + ConditionVS, + ConditionVC, + ConditionHI, + ConditionLS, + ConditionGE, + ConditionLT, + ConditionGT, + ConditionLE, + ConditionAL, + ConditionInvalid + } Condition; + +#define JUMP_ENUM_WITH_SIZE(index, value) (((value) << 3) | (index)) +#define JUMP_ENUM_SIZE(jump) ((jump) >> 3) + enum JumpType { JumpFixed = JUMP_ENUM_WITH_SIZE(0, 0), + JumpNoCondition = JUMP_ENUM_WITH_SIZE(1, 5 * sizeof(uint16_t)), + JumpCondition = JUMP_ENUM_WITH_SIZE(2, 6 * sizeof(uint16_t)), + JumpNoConditionFixedSize = JUMP_ENUM_WITH_SIZE(3, 5 * sizeof(uint16_t)), + JumpConditionFixedSize = JUMP_ENUM_WITH_SIZE(4, 6 * sizeof(uint16_t)) + }; + enum JumpLinkType { + LinkInvalid = JUMP_ENUM_WITH_SIZE(0, 0), + LinkJumpT1 = JUMP_ENUM_WITH_SIZE(1, sizeof(uint16_t)), + LinkJumpT2 = JUMP_ENUM_WITH_SIZE(2, sizeof(uint16_t)), + LinkJumpT3 = JUMP_ENUM_WITH_SIZE(3, 2 * sizeof(uint16_t)), + LinkJumpT4 = JUMP_ENUM_WITH_SIZE(4, 2 * sizeof(uint16_t)), + LinkConditionalJumpT4 = JUMP_ENUM_WITH_SIZE(5, 3 * sizeof(uint16_t)), + LinkBX = JUMP_ENUM_WITH_SIZE(6, 5 * sizeof(uint16_t)), + LinkConditionalBX = JUMP_ENUM_WITH_SIZE(7, 6 * sizeof(uint16_t)) + }; + + class LinkRecord { + public: + LinkRecord(intptr_t from, intptr_t to, JumpType type, Condition condition) + { + data.realTypes.m_from = from; + data.realTypes.m_to = to; + data.realTypes.m_type = type; + data.realTypes.m_linkType = LinkInvalid; + data.realTypes.m_condition = condition; + } + void operator=(const LinkRecord& other) + { + data.copyTypes.content[0] = other.data.copyTypes.content[0]; + data.copyTypes.content[1] = other.data.copyTypes.content[1]; + data.copyTypes.content[2] = other.data.copyTypes.content[2]; + } + intptr_t from() const { return data.realTypes.m_from; } + void setFrom(intptr_t from) { data.realTypes.m_from = from; } + intptr_t to() const { return data.realTypes.m_to; } + JumpType type() const { return data.realTypes.m_type; } + JumpLinkType linkType() const { return data.realTypes.m_linkType; } + void setLinkType(JumpLinkType linkType) { ASSERT(data.realTypes.m_linkType == LinkInvalid); data.realTypes.m_linkType = linkType; } + Condition condition() const { return data.realTypes.m_condition; } + private: + union { + struct RealTypes { + intptr_t m_from : 31; + intptr_t m_to : 31; + JumpType m_type : 8; + JumpLinkType m_linkType : 8; + Condition m_condition : 16; + } realTypes; + struct CopyTypes { + uint32_t content[3]; + } copyTypes; + COMPILE_ASSERT(sizeof(RealTypes) == sizeof(CopyTypes), LinkRecordCopyStructSizeEqualsRealStruct); + } data; + }; + + ARMv7Assembler() + : m_indexOfLastWatchpoint(INT_MIN) + , m_indexOfTailOfLastWatchpoint(INT_MIN) + { + } + +private: + + // ARMv7, Appx-A.6.3 + static bool BadReg(RegisterID reg) + { + return (reg == ARMRegisters::sp) || (reg == ARMRegisters::pc); + } + + uint32_t singleRegisterMask(FPSingleRegisterID rdNum, int highBitsShift, int lowBitShift) + { + uint32_t rdMask = (rdNum >> 1) << highBitsShift; + if (rdNum & 1) + rdMask |= 1 << lowBitShift; + return rdMask; + } + + uint32_t doubleRegisterMask(FPDoubleRegisterID rdNum, int highBitShift, int lowBitsShift) + { + uint32_t rdMask = (rdNum & 0xf) << lowBitsShift; + if (rdNum & 16) + rdMask |= 1 << highBitShift; + return rdMask; + } + + typedef enum { + OP_ADD_reg_T1 = 0x1800, + OP_SUB_reg_T1 = 0x1A00, + OP_ADD_imm_T1 = 0x1C00, + OP_SUB_imm_T1 = 0x1E00, + OP_MOV_imm_T1 = 0x2000, + OP_CMP_imm_T1 = 0x2800, + OP_ADD_imm_T2 = 0x3000, + OP_SUB_imm_T2 = 0x3800, + OP_AND_reg_T1 = 0x4000, + OP_EOR_reg_T1 = 0x4040, + OP_TST_reg_T1 = 0x4200, + OP_RSB_imm_T1 = 0x4240, + OP_CMP_reg_T1 = 0x4280, + OP_ORR_reg_T1 = 0x4300, + OP_MVN_reg_T1 = 0x43C0, + OP_ADD_reg_T2 = 0x4400, + OP_MOV_reg_T1 = 0x4600, + OP_BLX = 0x4700, + OP_BX = 0x4700, + OP_STR_reg_T1 = 0x5000, + OP_STRH_reg_T1 = 0x5200, + OP_STRB_reg_T1 = 0x5400, + OP_LDRSB_reg_T1 = 0x5600, + OP_LDR_reg_T1 = 0x5800, + OP_LDRH_reg_T1 = 0x5A00, + OP_LDRB_reg_T1 = 0x5C00, + OP_LDRSH_reg_T1 = 0x5E00, + OP_STR_imm_T1 = 0x6000, + OP_LDR_imm_T1 = 0x6800, + OP_STRB_imm_T1 = 0x7000, + OP_LDRB_imm_T1 = 0x7800, + OP_STRH_imm_T1 = 0x8000, + OP_LDRH_imm_T1 = 0x8800, + OP_STR_imm_T2 = 0x9000, + OP_LDR_imm_T2 = 0x9800, + OP_ADD_SP_imm_T1 = 0xA800, + OP_ADD_SP_imm_T2 = 0xB000, + OP_SUB_SP_imm_T1 = 0xB080, + OP_BKPT = 0xBE00, + OP_IT = 0xBF00, + OP_NOP_T1 = 0xBF00, + } OpcodeID; + + typedef enum { + OP_B_T1 = 0xD000, + OP_B_T2 = 0xE000, + OP_AND_reg_T2 = 0xEA00, + OP_TST_reg_T2 = 0xEA10, + OP_ORR_reg_T2 = 0xEA40, + OP_ORR_S_reg_T2 = 0xEA50, + OP_ASR_imm_T1 = 0xEA4F, + OP_LSL_imm_T1 = 0xEA4F, + OP_LSR_imm_T1 = 0xEA4F, + OP_ROR_imm_T1 = 0xEA4F, + OP_MVN_reg_T2 = 0xEA6F, + OP_EOR_reg_T2 = 0xEA80, + OP_ADD_reg_T3 = 0xEB00, + OP_ADD_S_reg_T3 = 0xEB10, + OP_SUB_reg_T2 = 0xEBA0, + OP_SUB_S_reg_T2 = 0xEBB0, + OP_CMP_reg_T2 = 0xEBB0, + OP_VMOV_CtoD = 0xEC00, + OP_VMOV_DtoC = 0xEC10, + OP_FSTS = 0xED00, + OP_VSTR = 0xED00, + OP_FLDS = 0xED10, + OP_VLDR = 0xED10, + OP_VMOV_CtoS = 0xEE00, + OP_VMOV_StoC = 0xEE10, + OP_VMUL_T2 = 0xEE20, + OP_VADD_T2 = 0xEE30, + OP_VSUB_T2 = 0xEE30, + OP_VDIV = 0xEE80, + OP_VABS_T2 = 0xEEB0, + OP_VCMP = 0xEEB0, + OP_VCVT_FPIVFP = 0xEEB0, + OP_VMOV_T2 = 0xEEB0, + OP_VMOV_IMM_T2 = 0xEEB0, + OP_VMRS = 0xEEB0, + OP_VNEG_T2 = 0xEEB0, + OP_VSQRT_T1 = 0xEEB0, + OP_VCVTSD_T1 = 0xEEB0, + OP_VCVTDS_T1 = 0xEEB0, + OP_B_T3a = 0xF000, + OP_B_T4a = 0xF000, + OP_AND_imm_T1 = 0xF000, + OP_TST_imm = 0xF010, + OP_ORR_imm_T1 = 0xF040, + OP_MOV_imm_T2 = 0xF040, + OP_MVN_imm = 0xF060, + OP_EOR_imm_T1 = 0xF080, + OP_ADD_imm_T3 = 0xF100, + OP_ADD_S_imm_T3 = 0xF110, + OP_CMN_imm = 0xF110, + OP_ADC_imm = 0xF140, + OP_SUB_imm_T3 = 0xF1A0, + OP_SUB_S_imm_T3 = 0xF1B0, + OP_CMP_imm_T2 = 0xF1B0, + OP_RSB_imm_T2 = 0xF1C0, + OP_RSB_S_imm_T2 = 0xF1D0, + OP_ADD_imm_T4 = 0xF200, + OP_MOV_imm_T3 = 0xF240, + OP_SUB_imm_T4 = 0xF2A0, + OP_MOVT = 0xF2C0, + OP_UBFX_T1 = 0xF3C0, + OP_NOP_T2a = 0xF3AF, + OP_STRB_imm_T3 = 0xF800, + OP_STRB_reg_T2 = 0xF800, + OP_LDRB_imm_T3 = 0xF810, + OP_LDRB_reg_T2 = 0xF810, + OP_STRH_imm_T3 = 0xF820, + OP_STRH_reg_T2 = 0xF820, + OP_LDRH_reg_T2 = 0xF830, + OP_LDRH_imm_T3 = 0xF830, + OP_STR_imm_T4 = 0xF840, + OP_STR_reg_T2 = 0xF840, + OP_LDR_imm_T4 = 0xF850, + OP_LDR_reg_T2 = 0xF850, + OP_STRB_imm_T2 = 0xF880, + OP_LDRB_imm_T2 = 0xF890, + OP_STRH_imm_T2 = 0xF8A0, + OP_LDRH_imm_T2 = 0xF8B0, + OP_STR_imm_T3 = 0xF8C0, + OP_LDR_imm_T3 = 0xF8D0, + OP_LDRSB_reg_T2 = 0xF910, + OP_LDRSH_reg_T2 = 0xF930, + OP_LSL_reg_T2 = 0xFA00, + OP_LSR_reg_T2 = 0xFA20, + OP_ASR_reg_T2 = 0xFA40, + OP_ROR_reg_T2 = 0xFA60, + OP_CLZ = 0xFAB0, + OP_SMULL_T1 = 0xFB80, +#if CPU(APPLE_ARMV7S) + OP_SDIV_T1 = 0xFB90, + OP_UDIV_T1 = 0xFBB0, +#endif + } OpcodeID1; + + typedef enum { + OP_VADD_T2b = 0x0A00, + OP_VDIVb = 0x0A00, + OP_FLDSb = 0x0A00, + OP_VLDRb = 0x0A00, + OP_VMOV_IMM_T2b = 0x0A00, + OP_VMOV_T2b = 0x0A40, + OP_VMUL_T2b = 0x0A00, + OP_FSTSb = 0x0A00, + OP_VSTRb = 0x0A00, + OP_VMOV_StoCb = 0x0A10, + OP_VMOV_CtoSb = 0x0A10, + OP_VMOV_DtoCb = 0x0A10, + OP_VMOV_CtoDb = 0x0A10, + OP_VMRSb = 0x0A10, + OP_VABS_T2b = 0x0A40, + OP_VCMPb = 0x0A40, + OP_VCVT_FPIVFPb = 0x0A40, + OP_VNEG_T2b = 0x0A40, + OP_VSUB_T2b = 0x0A40, + OP_VSQRT_T1b = 0x0A40, + OP_VCVTSD_T1b = 0x0A40, + OP_VCVTDS_T1b = 0x0A40, + OP_NOP_T2b = 0x8000, + OP_B_T3b = 0x8000, + OP_B_T4b = 0x9000, + } OpcodeID2; + + struct FourFours { + FourFours(unsigned f3, unsigned f2, unsigned f1, unsigned f0) + { + m_u.f0 = f0; + m_u.f1 = f1; + m_u.f2 = f2; + m_u.f3 = f3; + } + + union { + unsigned value; + struct { + unsigned f0 : 4; + unsigned f1 : 4; + unsigned f2 : 4; + unsigned f3 : 4; + }; + } m_u; + }; + + class ARMInstructionFormatter; + + // false means else! + bool ifThenElseConditionBit(Condition condition, bool isIf) + { + return isIf ? (condition & 1) : !(condition & 1); + } + uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if, bool inst4if) + { + int mask = (ifThenElseConditionBit(condition, inst2if) << 3) + | (ifThenElseConditionBit(condition, inst3if) << 2) + | (ifThenElseConditionBit(condition, inst4if) << 1) + | 1; + ASSERT((condition != ConditionAL) || !(mask & (mask - 1))); + return (condition << 4) | mask; + } + uint8_t ifThenElse(Condition condition, bool inst2if, bool inst3if) + { + int mask = (ifThenElseConditionBit(condition, inst2if) << 3) + | (ifThenElseConditionBit(condition, inst3if) << 2) + | 2; + ASSERT((condition != ConditionAL) || !(mask & (mask - 1))); + return (condition << 4) | mask; + } + uint8_t ifThenElse(Condition condition, bool inst2if) + { + int mask = (ifThenElseConditionBit(condition, inst2if) << 3) + | 4; + ASSERT((condition != ConditionAL) || !(mask & (mask - 1))); + return (condition << 4) | mask; + } + + uint8_t ifThenElse(Condition condition) + { + int mask = 8; + return (condition << 4) | mask; + } + +public: + + void adc(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + // Rd can only be SP if Rn is also SP. + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isEncodedImm()); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_ADC_imm, rn, rd, imm); + } + + void add(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + // Rd can only be SP if Rn is also SP. + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isValid()); + + if (rn == ARMRegisters::sp) { + ASSERT(!(imm.getUInt16() & 3)); + if (!(rd & 8) && imm.isUInt10()) { + m_formatter.oneWordOp5Reg3Imm8(OP_ADD_SP_imm_T1, rd, static_cast<uint8_t>(imm.getUInt10() >> 2)); + return; + } else if ((rd == ARMRegisters::sp) && imm.isUInt9()) { + m_formatter.oneWordOp9Imm7(OP_ADD_SP_imm_T2, static_cast<uint8_t>(imm.getUInt9() >> 2)); + return; + } + } else if (!((rd | rn) & 8)) { + if (imm.isUInt3()) { + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_ADD_imm_T1, (RegisterID)imm.getUInt3(), rn, rd); + return; + } else if ((rd == rn) && imm.isUInt8()) { + m_formatter.oneWordOp5Reg3Imm8(OP_ADD_imm_T2, rd, imm.getUInt8()); + return; + } + } + + if (imm.isEncodedImm()) + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_ADD_imm_T3, rn, rd, imm); + else { + ASSERT(imm.isUInt12()); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_ADD_imm_T4, rn, rd, imm); + } + } + + ALWAYS_INLINE void add(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ADD_reg_T3, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + // NOTE: In an IT block, add doesn't modify the flags register. + ALWAYS_INLINE void add(RegisterID rd, RegisterID rn, RegisterID rm) + { + if (rd == rn) + m_formatter.oneWordOp8RegReg143(OP_ADD_reg_T2, rm, rd); + else if (rd == rm) + m_formatter.oneWordOp8RegReg143(OP_ADD_reg_T2, rn, rd); + else if (!((rd | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_ADD_reg_T1, rm, rn, rd); + else + add(rd, rn, rm, ShiftTypeAndAmount()); + } + + // Not allowed in an IT (if then) block. + ALWAYS_INLINE void add_S(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + // Rd can only be SP if Rn is also SP. + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isEncodedImm()); + + if (!((rd | rn) & 8)) { + if (imm.isUInt3()) { + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_ADD_imm_T1, (RegisterID)imm.getUInt3(), rn, rd); + return; + } else if ((rd == rn) && imm.isUInt8()) { + m_formatter.oneWordOp5Reg3Imm8(OP_ADD_imm_T2, rd, imm.getUInt8()); + return; + } + } + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_ADD_S_imm_T3, rn, rd, imm); + } + + // Not allowed in an IT (if then) block? + ALWAYS_INLINE void add_S(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ADD_S_reg_T3, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + // Not allowed in an IT (if then) block. + ALWAYS_INLINE void add_S(RegisterID rd, RegisterID rn, RegisterID rm) + { + if (!((rd | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_ADD_reg_T1, rm, rn, rd); + else + add_S(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void ARM_and(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(imm.isEncodedImm()); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_AND_imm_T1, rn, rd, imm); + } + + ALWAYS_INLINE void ARM_and(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_AND_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void ARM_and(RegisterID rd, RegisterID rn, RegisterID rm) + { + if ((rd == rn) && !((rd | rm) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_AND_reg_T1, rm, rd); + else if ((rd == rm) && !((rd | rn) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_AND_reg_T1, rn, rd); + else + ARM_and(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void asr(RegisterID rd, RegisterID rm, int32_t shiftAmount) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + ShiftTypeAndAmount shift(SRType_ASR, shiftAmount); + m_formatter.twoWordOp16FourFours(OP_ASR_imm_T1, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void asr(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ASR_reg_T2, rn, FourFours(0xf, rd, 0, rm)); + } + + // Only allowed in IT (if then) block if last instruction. + ALWAYS_INLINE AssemblerLabel b() + { + m_formatter.twoWordOp16Op16(OP_B_T4a, OP_B_T4b); + return m_formatter.label(); + } + + // Only allowed in IT (if then) block if last instruction. + ALWAYS_INLINE AssemblerLabel blx(RegisterID rm) + { + ASSERT(rm != ARMRegisters::pc); + m_formatter.oneWordOp8RegReg143(OP_BLX, rm, (RegisterID)8); + return m_formatter.label(); + } + + // Only allowed in IT (if then) block if last instruction. + ALWAYS_INLINE AssemblerLabel bx(RegisterID rm) + { + m_formatter.oneWordOp8RegReg143(OP_BX, rm, (RegisterID)0); + return m_formatter.label(); + } + + void bkpt(uint8_t imm = 0) + { + m_formatter.oneWordOp8Imm8(OP_BKPT, imm); + } + + ALWAYS_INLINE void clz(RegisterID rd, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_CLZ, rm, FourFours(0xf, rd, 8, rm)); + } + + ALWAYS_INLINE void cmn(RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isEncodedImm()); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_CMN_imm, rn, (RegisterID)0xf, imm); + } + + ALWAYS_INLINE void cmp(RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isEncodedImm()); + + if (!(rn & 8) && imm.isUInt8()) + m_formatter.oneWordOp5Reg3Imm8(OP_CMP_imm_T1, rn, imm.getUInt8()); + else + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_CMP_imm_T2, rn, (RegisterID)0xf, imm); + } + + ALWAYS_INLINE void cmp(RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_CMP_reg_T2, rn, FourFours(shift.hi4(), 0xf, shift.lo4(), rm)); + } + + ALWAYS_INLINE void cmp(RegisterID rn, RegisterID rm) + { + if ((rn | rm) & 8) + cmp(rn, rm, ShiftTypeAndAmount()); + else + m_formatter.oneWordOp10Reg3Reg3(OP_CMP_reg_T1, rm, rn); + } + + // xor is not spelled with an 'e'. :-( + ALWAYS_INLINE void eor(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(imm.isEncodedImm()); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_EOR_imm_T1, rn, rd, imm); + } + + // xor is not spelled with an 'e'. :-( + ALWAYS_INLINE void eor(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_EOR_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + // xor is not spelled with an 'e'. :-( + void eor(RegisterID rd, RegisterID rn, RegisterID rm) + { + if ((rd == rn) && !((rd | rm) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_EOR_reg_T1, rm, rd); + else if ((rd == rm) && !((rd | rn) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_EOR_reg_T1, rn, rd); + else + eor(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void it(Condition cond) + { + m_formatter.oneWordOp8Imm8(OP_IT, ifThenElse(cond)); + } + + ALWAYS_INLINE void it(Condition cond, bool inst2if) + { + m_formatter.oneWordOp8Imm8(OP_IT, ifThenElse(cond, inst2if)); + } + + ALWAYS_INLINE void it(Condition cond, bool inst2if, bool inst3if) + { + m_formatter.oneWordOp8Imm8(OP_IT, ifThenElse(cond, inst2if, inst3if)); + } + + ALWAYS_INLINE void it(Condition cond, bool inst2if, bool inst3if, bool inst4if) + { + m_formatter.oneWordOp8Imm8(OP_IT, ifThenElse(cond, inst2if, inst3if, inst4if)); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void ldr(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt7()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDR_imm_T1, imm.getUInt7() >> 2, rn, rt); + else if ((rn == ARMRegisters::sp) && !(rt & 8) && imm.isUInt10()) + m_formatter.oneWordOp5Reg3Imm8(OP_LDR_imm_T2, rt, static_cast<uint8_t>(imm.getUInt10() >> 2)); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDR_imm_T3, rn, rt, imm.getUInt12()); + } + + ALWAYS_INLINE void ldrWide8BitImmediate(RegisterID rt, RegisterID rn, uint8_t immediate) + { + ASSERT(rn != ARMRegisters::pc); + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDR_imm_T3, rn, rt, immediate); + } + + ALWAYS_INLINE void ldrCompact(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(imm.isUInt7()); + ASSERT(!((rt | rn) & 8)); + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDR_imm_T1, imm.getUInt7() >> 2, rn, rt); + } + + // If index is set, this is a regular offset or a pre-indexed load; + // if index is not set then is is a post-index load. + // + // If wback is set rn is updated - this is a pre or post index load, + // if wback is not set this is a regular offset memory access. + // + // (-255 <= offset <= 255) + // _reg = REG[rn] + // _tmp = _reg + offset + // MEM[index ? _tmp : _reg] = REG[rt] + // if (wback) REG[rn] = _tmp + ALWAYS_INLINE void ldr(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + ASSERT((offset & ~0xff) == 0); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDR_imm_T4, rn, rt, offset); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void ldr(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_LDR_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_LDR_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void ldrh(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt6()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDRH_imm_T1, imm.getUInt6() >> 2, rn, rt); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDRH_imm_T2, rn, rt, imm.getUInt12()); + } + + // If index is set, this is a regular offset or a pre-indexed load; + // if index is not set then is is a post-index load. + // + // If wback is set rn is updated - this is a pre or post index load, + // if wback is not set this is a regular offset memory access. + // + // (-255 <= offset <= 255) + // _reg = REG[rn] + // _tmp = _reg + offset + // MEM[index ? _tmp : _reg] = REG[rt] + // if (wback) REG[rn] = _tmp + ALWAYS_INLINE void ldrh(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + ASSERT((offset & ~0xff) == 0); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDRH_imm_T3, rn, rt, offset); + } + + ALWAYS_INLINE void ldrh(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(!BadReg(rt)); // Memory hint + ASSERT(rn != ARMRegisters::pc); // LDRH (literal) + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_LDRH_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_LDRH_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + void ldrb(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt5()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_LDRB_imm_T1, imm.getUInt5(), rn, rt); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDRB_imm_T2, rn, rt, imm.getUInt12()); + } + + void ldrb(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + + ASSERT(!(offset & ~0xff)); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_LDRB_imm_T3, rn, rt, offset); + } + + ALWAYS_INLINE void ldrb(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); // LDR (literal) + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_LDRB_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_LDRB_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + void ldrsb(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_LDRSB_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_LDRSB_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + void ldrsh(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_LDRSH_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_LDRSH_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + void lsl(RegisterID rd, RegisterID rm, int32_t shiftAmount) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + ShiftTypeAndAmount shift(SRType_LSL, shiftAmount); + m_formatter.twoWordOp16FourFours(OP_LSL_imm_T1, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void lsl(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_LSL_reg_T2, rn, FourFours(0xf, rd, 0, rm)); + } + + ALWAYS_INLINE void lsr(RegisterID rd, RegisterID rm, int32_t shiftAmount) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + ShiftTypeAndAmount shift(SRType_LSR, shiftAmount); + m_formatter.twoWordOp16FourFours(OP_LSR_imm_T1, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void lsr(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_LSR_reg_T2, rn, FourFours(0xf, rd, 0, rm)); + } + + ALWAYS_INLINE void movT3(RegisterID rd, ARMThumbImmediate imm) + { + ASSERT(imm.isValid()); + ASSERT(!imm.isEncodedImm()); + ASSERT(!BadReg(rd)); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_MOV_imm_T3, imm.m_value.imm4, rd, imm); + } + +#if OS(LINUX) || OS(QNX) + static void revertJumpTo_movT3movtcmpT2(void* instructionStart, RegisterID left, RegisterID right, uintptr_t imm) + { + uint16_t* address = static_cast<uint16_t*>(instructionStart); + ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(imm)); + ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(imm >> 16)); + address[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16); + address[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(right, lo16); + address[2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16); + address[3] = twoWordOp5i6Imm4Reg4EncodedImmSecond(right, hi16); + address[4] = OP_CMP_reg_T2 | left; + cacheFlush(address, sizeof(uint16_t) * 5); + } +#else + static void revertJumpTo_movT3(void* instructionStart, RegisterID rd, ARMThumbImmediate imm) + { + ASSERT(imm.isValid()); + ASSERT(!imm.isEncodedImm()); + ASSERT(!BadReg(rd)); + + uint16_t* address = static_cast<uint16_t*>(instructionStart); + address[0] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, imm); + address[1] = twoWordOp5i6Imm4Reg4EncodedImmSecond(rd, imm); + cacheFlush(address, sizeof(uint16_t) * 2); + } +#endif + + ALWAYS_INLINE void mov(RegisterID rd, ARMThumbImmediate imm) + { + ASSERT(imm.isValid()); + ASSERT(!BadReg(rd)); + + if ((rd < 8) && imm.isUInt8()) + m_formatter.oneWordOp5Reg3Imm8(OP_MOV_imm_T1, rd, imm.getUInt8()); + else if (imm.isEncodedImm()) + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_MOV_imm_T2, 0xf, rd, imm); + else + movT3(rd, imm); + } + + ALWAYS_INLINE void mov(RegisterID rd, RegisterID rm) + { + m_formatter.oneWordOp8RegReg143(OP_MOV_reg_T1, rm, rd); + } + + ALWAYS_INLINE void movt(RegisterID rd, ARMThumbImmediate imm) + { + ASSERT(imm.isUInt16()); + ASSERT(!BadReg(rd)); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_MOVT, imm.m_value.imm4, rd, imm); + } + + ALWAYS_INLINE void mvn(RegisterID rd, ARMThumbImmediate imm) + { + ASSERT(imm.isEncodedImm()); + ASSERT(!BadReg(rd)); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_MVN_imm, 0xf, rd, imm); + } + + ALWAYS_INLINE void mvn(RegisterID rd, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp16FourFours(OP_MVN_reg_T2, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void mvn(RegisterID rd, RegisterID rm) + { + if (!((rd | rm) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_MVN_reg_T1, rm, rd); + else + mvn(rd, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void neg(RegisterID rd, RegisterID rm) + { + ARMThumbImmediate zero = ARMThumbImmediate::makeUInt12(0); + sub(rd, zero, rm); + } + + ALWAYS_INLINE void orr(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(imm.isEncodedImm()); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_ORR_imm_T1, rn, rd, imm); + } + + ALWAYS_INLINE void orr(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ORR_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + void orr(RegisterID rd, RegisterID rn, RegisterID rm) + { + if ((rd == rn) && !((rd | rm) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_ORR_reg_T1, rm, rd); + else if ((rd == rm) && !((rd | rn) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_ORR_reg_T1, rn, rd); + else + orr(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void orr_S(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ORR_S_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + void orr_S(RegisterID rd, RegisterID rn, RegisterID rm) + { + if ((rd == rn) && !((rd | rm) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_ORR_reg_T1, rm, rd); + else if ((rd == rm) && !((rd | rn) & 8)) + m_formatter.oneWordOp10Reg3Reg3(OP_ORR_reg_T1, rn, rd); + else + orr_S(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void ror(RegisterID rd, RegisterID rm, int32_t shiftAmount) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rm)); + ShiftTypeAndAmount shift(SRType_ROR, shiftAmount); + m_formatter.twoWordOp16FourFours(OP_ROR_imm_T1, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + ALWAYS_INLINE void ror(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_ROR_reg_T2, rn, FourFours(0xf, rd, 0, rm)); + } + +#if CPU(APPLE_ARMV7S) + ALWAYS_INLINE void sdiv(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_SDIV_T1, rn, FourFours(0xf, rd, 0xf, rm)); + } +#endif + + ALWAYS_INLINE void smull(RegisterID rdLo, RegisterID rdHi, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rdLo)); + ASSERT(!BadReg(rdHi)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + ASSERT(rdLo != rdHi); + m_formatter.twoWordOp12Reg4FourFours(OP_SMULL_T1, rn, FourFours(rdLo, rdHi, 0, rm)); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void str(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt7()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_STR_imm_T1, imm.getUInt7() >> 2, rn, rt); + else if ((rn == ARMRegisters::sp) && !(rt & 8) && imm.isUInt10()) + m_formatter.oneWordOp5Reg3Imm8(OP_STR_imm_T2, rt, static_cast<uint8_t>(imm.getUInt10() >> 2)); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STR_imm_T3, rn, rt, imm.getUInt12()); + } + + // If index is set, this is a regular offset or a pre-indexed store; + // if index is not set then is is a post-index store. + // + // If wback is set rn is updated - this is a pre or post index store, + // if wback is not set this is a regular offset memory access. + // + // (-255 <= offset <= 255) + // _reg = REG[rn] + // _tmp = _reg + offset + // MEM[index ? _tmp : _reg] = REG[rt] + // if (wback) REG[rn] = _tmp + ALWAYS_INLINE void str(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + ASSERT((offset & ~0xff) == 0); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STR_imm_T4, rn, rt, offset); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void str(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_STR_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_STR_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void strb(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt7()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_STRB_imm_T1, imm.getUInt7() >> 2, rn, rt); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STRB_imm_T2, rn, rt, imm.getUInt12()); + } + + // If index is set, this is a regular offset or a pre-indexed store; + // if index is not set then is is a post-index store. + // + // If wback is set rn is updated - this is a pre or post index store, + // if wback is not set this is a regular offset memory access. + // + // (-255 <= offset <= 255) + // _reg = REG[rn] + // _tmp = _reg + offset + // MEM[index ? _tmp : _reg] = REG[rt] + // if (wback) REG[rn] = _tmp + ALWAYS_INLINE void strb(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + ASSERT((offset & ~0xff) == 0); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STRB_imm_T3, rn, rt, offset); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void strb(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_STRB_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_STRB_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void strh(RegisterID rt, RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isUInt12()); + + if (!((rt | rn) & 8) && imm.isUInt7()) + m_formatter.oneWordOp5Imm5Reg3Reg3(OP_STRH_imm_T1, imm.getUInt7() >> 2, rn, rt); + else + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STRH_imm_T2, rn, rt, imm.getUInt12()); + } + + // If index is set, this is a regular offset or a pre-indexed store; + // if index is not set then is is a post-index store. + // + // If wback is set rn is updated - this is a pre or post index store, + // if wback is not set this is a regular offset memory access. + // + // (-255 <= offset <= 255) + // _reg = REG[rn] + // _tmp = _reg + offset + // MEM[index ? _tmp : _reg] = REG[rt] + // if (wback) REG[rn] = _tmp + ALWAYS_INLINE void strh(RegisterID rt, RegisterID rn, int offset, bool index, bool wback) + { + ASSERT(rt != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(index || wback); + ASSERT(!wback | (rt != rn)); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + ASSERT(!(offset & ~0xff)); + + offset |= (wback << 8); + offset |= (add << 9); + offset |= (index << 10); + offset |= (1 << 11); + + m_formatter.twoWordOp12Reg4Reg4Imm12(OP_STRH_imm_T3, rn, rt, offset); + } + + // rt == ARMRegisters::pc only allowed if last instruction in IT (if then) block. + ALWAYS_INLINE void strh(RegisterID rt, RegisterID rn, RegisterID rm, unsigned shift = 0) + { + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + ASSERT(shift <= 3); + + if (!shift && !((rt | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_STRH_reg_T1, rm, rn, rt); + else + m_formatter.twoWordOp12Reg4FourFours(OP_STRH_reg_T2, rn, FourFours(rt, 0, shift, rm)); + } + + ALWAYS_INLINE void sub(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + // Rd can only be SP if Rn is also SP. + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isValid()); + + if ((rn == ARMRegisters::sp) && (rd == ARMRegisters::sp) && imm.isUInt9()) { + ASSERT(!(imm.getUInt16() & 3)); + m_formatter.oneWordOp9Imm7(OP_SUB_SP_imm_T1, static_cast<uint8_t>(imm.getUInt9() >> 2)); + return; + } else if (!((rd | rn) & 8)) { + if (imm.isUInt3()) { + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_SUB_imm_T1, (RegisterID)imm.getUInt3(), rn, rd); + return; + } else if ((rd == rn) && imm.isUInt8()) { + m_formatter.oneWordOp5Reg3Imm8(OP_SUB_imm_T2, rd, imm.getUInt8()); + return; + } + } + + if (imm.isEncodedImm()) + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_SUB_imm_T3, rn, rd, imm); + else { + ASSERT(imm.isUInt12()); + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_SUB_imm_T4, rn, rd, imm); + } + } + + ALWAYS_INLINE void sub(RegisterID rd, ARMThumbImmediate imm, RegisterID rn) + { + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isValid()); + ASSERT(imm.isUInt12()); + + if (!((rd | rn) & 8) && !imm.getUInt12()) + m_formatter.oneWordOp10Reg3Reg3(OP_RSB_imm_T1, rn, rd); + else + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_RSB_imm_T2, rn, rd, imm); + } + + ALWAYS_INLINE void sub(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_SUB_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + // NOTE: In an IT block, add doesn't modify the flags register. + ALWAYS_INLINE void sub(RegisterID rd, RegisterID rn, RegisterID rm) + { + if (!((rd | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_SUB_reg_T1, rm, rn, rd); + else + sub(rd, rn, rm, ShiftTypeAndAmount()); + } + + // Not allowed in an IT (if then) block. + void sub_S(RegisterID rd, RegisterID rn, ARMThumbImmediate imm) + { + // Rd can only be SP if Rn is also SP. + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isValid()); + + if ((rn == ARMRegisters::sp) && (rd == ARMRegisters::sp) && imm.isUInt9()) { + ASSERT(!(imm.getUInt16() & 3)); + m_formatter.oneWordOp9Imm7(OP_SUB_SP_imm_T1, static_cast<uint8_t>(imm.getUInt9() >> 2)); + return; + } else if (!((rd | rn) & 8)) { + if (imm.isUInt3()) { + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_SUB_imm_T1, (RegisterID)imm.getUInt3(), rn, rd); + return; + } else if ((rd == rn) && imm.isUInt8()) { + m_formatter.oneWordOp5Reg3Imm8(OP_SUB_imm_T2, rd, imm.getUInt8()); + return; + } + } + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_SUB_S_imm_T3, rn, rd, imm); + } + + ALWAYS_INLINE void sub_S(RegisterID rd, ARMThumbImmediate imm, RegisterID rn) + { + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(imm.isValid()); + ASSERT(imm.isUInt12()); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_RSB_S_imm_T2, rn, rd, imm); + } + + // Not allowed in an IT (if then) block? + ALWAYS_INLINE void sub_S(RegisterID rd, RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT((rd != ARMRegisters::sp) || (rn == ARMRegisters::sp)); + ASSERT(rd != ARMRegisters::pc); + ASSERT(rn != ARMRegisters::pc); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_SUB_S_reg_T2, rn, FourFours(shift.hi4(), rd, shift.lo4(), rm)); + } + + // Not allowed in an IT (if then) block. + ALWAYS_INLINE void sub_S(RegisterID rd, RegisterID rn, RegisterID rm) + { + if (!((rd | rn | rm) & 8)) + m_formatter.oneWordOp7Reg3Reg3Reg3(OP_SUB_reg_T1, rm, rn, rd); + else + sub_S(rd, rn, rm, ShiftTypeAndAmount()); + } + + ALWAYS_INLINE void tst(RegisterID rn, ARMThumbImmediate imm) + { + ASSERT(!BadReg(rn)); + ASSERT(imm.isEncodedImm()); + + m_formatter.twoWordOp5i6Imm4Reg4EncodedImm(OP_TST_imm, rn, (RegisterID)0xf, imm); + } + + ALWAYS_INLINE void tst(RegisterID rn, RegisterID rm, ShiftTypeAndAmount shift) + { + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_TST_reg_T2, rn, FourFours(shift.hi4(), 0xf, shift.lo4(), rm)); + } + + ALWAYS_INLINE void tst(RegisterID rn, RegisterID rm) + { + if ((rn | rm) & 8) + tst(rn, rm, ShiftTypeAndAmount()); + else + m_formatter.oneWordOp10Reg3Reg3(OP_TST_reg_T1, rm, rn); + } + + ALWAYS_INLINE void ubfx(RegisterID rd, RegisterID rn, unsigned lsb, unsigned width) + { + ASSERT(lsb < 32); + ASSERT((width >= 1) && (width <= 32)); + ASSERT((lsb + width) <= 32); + m_formatter.twoWordOp12Reg40Imm3Reg4Imm20Imm5(OP_UBFX_T1, rd, rn, (lsb & 0x1c) << 10, (lsb & 0x3) << 6, (width - 1) & 0x1f); + } + +#if CPU(APPLE_ARMV7S) + ALWAYS_INLINE void udiv(RegisterID rd, RegisterID rn, RegisterID rm) + { + ASSERT(!BadReg(rd)); + ASSERT(!BadReg(rn)); + ASSERT(!BadReg(rm)); + m_formatter.twoWordOp12Reg4FourFours(OP_UDIV_T1, rn, FourFours(0xf, rd, 0xf, rm)); + } +#endif + + void vadd(FPDoubleRegisterID rd, FPDoubleRegisterID rn, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VADD_T2, OP_VADD_T2b, true, rn, rd, rm); + } + + void vcmp(FPDoubleRegisterID rd, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VCMP, OP_VCMPb, true, VFPOperand(4), rd, rm); + } + + void vcmpz(FPDoubleRegisterID rd) + { + m_formatter.vfpOp(OP_VCMP, OP_VCMPb, true, VFPOperand(5), rd, VFPOperand(0)); + } + + void vcvt_signedToFloatingPoint(FPDoubleRegisterID rd, FPSingleRegisterID rm) + { + // boolean values are 64bit (toInt, unsigned, roundZero) + m_formatter.vfpOp(OP_VCVT_FPIVFP, OP_VCVT_FPIVFPb, true, vcvtOp(false, false, false), rd, rm); + } + + void vcvt_floatingPointToSigned(FPSingleRegisterID rd, FPDoubleRegisterID rm) + { + // boolean values are 64bit (toInt, unsigned, roundZero) + m_formatter.vfpOp(OP_VCVT_FPIVFP, OP_VCVT_FPIVFPb, true, vcvtOp(true, false, true), rd, rm); + } + + void vcvt_floatingPointToUnsigned(FPSingleRegisterID rd, FPDoubleRegisterID rm) + { + // boolean values are 64bit (toInt, unsigned, roundZero) + m_formatter.vfpOp(OP_VCVT_FPIVFP, OP_VCVT_FPIVFPb, true, vcvtOp(true, true, true), rd, rm); + } + + void vdiv(FPDoubleRegisterID rd, FPDoubleRegisterID rn, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VDIV, OP_VDIVb, true, rn, rd, rm); + } + + void vldr(FPDoubleRegisterID rd, RegisterID rn, int32_t imm) + { + m_formatter.vfpMemOp(OP_VLDR, OP_VLDRb, true, rn, rd, imm); + } + + void flds(FPSingleRegisterID rd, RegisterID rn, int32_t imm) + { + m_formatter.vfpMemOp(OP_FLDS, OP_FLDSb, false, rn, rd, imm); + } + + void vmov(RegisterID rd, FPSingleRegisterID rn) + { + ASSERT(!BadReg(rd)); + m_formatter.vfpOp(OP_VMOV_StoC, OP_VMOV_StoCb, false, rn, rd, VFPOperand(0)); + } + + void vmov(FPSingleRegisterID rd, RegisterID rn) + { + ASSERT(!BadReg(rn)); + m_formatter.vfpOp(OP_VMOV_CtoS, OP_VMOV_CtoSb, false, rd, rn, VFPOperand(0)); + } + + void vmov(RegisterID rd1, RegisterID rd2, FPDoubleRegisterID rn) + { + ASSERT(!BadReg(rd1)); + ASSERT(!BadReg(rd2)); + m_formatter.vfpOp(OP_VMOV_DtoC, OP_VMOV_DtoCb, true, rd2, VFPOperand(rd1 | 16), rn); + } + + void vmov(FPDoubleRegisterID rd, RegisterID rn1, RegisterID rn2) + { + ASSERT(!BadReg(rn1)); + ASSERT(!BadReg(rn2)); + m_formatter.vfpOp(OP_VMOV_CtoD, OP_VMOV_CtoDb, true, rn2, VFPOperand(rn1 | 16), rd); + } + + void vmov(FPDoubleRegisterID rd, FPDoubleRegisterID rn) + { + m_formatter.vfpOp(OP_VMOV_T2, OP_VMOV_T2b, true, VFPOperand(0), rd, rn); + } + + void vmrs(RegisterID reg = ARMRegisters::pc) + { + ASSERT(reg != ARMRegisters::sp); + m_formatter.vfpOp(OP_VMRS, OP_VMRSb, false, VFPOperand(1), VFPOperand(0x10 | reg), VFPOperand(0)); + } + + void vmul(FPDoubleRegisterID rd, FPDoubleRegisterID rn, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VMUL_T2, OP_VMUL_T2b, true, rn, rd, rm); + } + + void vstr(FPDoubleRegisterID rd, RegisterID rn, int32_t imm) + { + m_formatter.vfpMemOp(OP_VSTR, OP_VSTRb, true, rn, rd, imm); + } + + void fsts(FPSingleRegisterID rd, RegisterID rn, int32_t imm) + { + m_formatter.vfpMemOp(OP_FSTS, OP_FSTSb, false, rn, rd, imm); + } + + void vsub(FPDoubleRegisterID rd, FPDoubleRegisterID rn, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VSUB_T2, OP_VSUB_T2b, true, rn, rd, rm); + } + + void vabs(FPDoubleRegisterID rd, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VABS_T2, OP_VABS_T2b, true, VFPOperand(16), rd, rm); + } + + void vneg(FPDoubleRegisterID rd, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VNEG_T2, OP_VNEG_T2b, true, VFPOperand(1), rd, rm); + } + + void vsqrt(FPDoubleRegisterID rd, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VSQRT_T1, OP_VSQRT_T1b, true, VFPOperand(17), rd, rm); + } + + void vcvtds(FPDoubleRegisterID rd, FPSingleRegisterID rm) + { + m_formatter.vfpOp(OP_VCVTDS_T1, OP_VCVTDS_T1b, false, VFPOperand(23), rd, rm); + } + + void vcvtsd(FPSingleRegisterID rd, FPDoubleRegisterID rm) + { + m_formatter.vfpOp(OP_VCVTSD_T1, OP_VCVTSD_T1b, true, VFPOperand(23), rd, rm); + } + + void nop() + { + m_formatter.oneWordOp8Imm8(OP_NOP_T1, 0); + } + + void nopw() + { + m_formatter.twoWordOp16Op16(OP_NOP_T2a, OP_NOP_T2b); + } + + AssemblerLabel labelIgnoringWatchpoints() + { + return m_formatter.label(); + } + + AssemblerLabel labelForWatchpoint() + { + AssemblerLabel result = m_formatter.label(); + if (static_cast<int>(result.m_offset) != m_indexOfLastWatchpoint) + result = label(); + m_indexOfLastWatchpoint = result.m_offset; + m_indexOfTailOfLastWatchpoint = result.m_offset + maxJumpReplacementSize(); + return result; + } + + AssemblerLabel label() + { + AssemblerLabel result = m_formatter.label(); + while (UNLIKELY(static_cast<int>(result.m_offset) < m_indexOfTailOfLastWatchpoint)) { + if (UNLIKELY(static_cast<int>(result.m_offset) + 4 <= m_indexOfTailOfLastWatchpoint)) + nopw(); + else + nop(); + result = m_formatter.label(); + } + return result; + } + + AssemblerLabel align(int alignment) + { + while (!m_formatter.isAligned(alignment)) + bkpt(); + + return label(); + } + + static void* getRelocatedAddress(void* code, AssemblerLabel label) + { + ASSERT(label.isSet()); + return reinterpret_cast<void*>(reinterpret_cast<ptrdiff_t>(code) + label.m_offset); + } + + static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b) + { + return b.m_offset - a.m_offset; + } + + int executableOffsetFor(int location) + { + if (!location) + return 0; + return static_cast<int32_t*>(m_formatter.data())[location / sizeof(int32_t) - 1]; + } + + int jumpSizeDelta(JumpType jumpType, JumpLinkType jumpLinkType) { return JUMP_ENUM_SIZE(jumpType) - JUMP_ENUM_SIZE(jumpLinkType); } + + // Assembler admin methods: + + static ALWAYS_INLINE bool linkRecordSourceComparator(const LinkRecord& a, const LinkRecord& b) + { + return a.from() < b.from(); + } + + bool canCompact(JumpType jumpType) + { + // The following cannot be compacted: + // JumpFixed: represents custom jump sequence + // JumpNoConditionFixedSize: represents unconditional jump that must remain a fixed size + // JumpConditionFixedSize: represents conditional jump that must remain a fixed size + return (jumpType == JumpNoCondition) || (jumpType == JumpCondition); + } + + JumpLinkType computeJumpType(JumpType jumpType, const uint8_t* from, const uint8_t* to) + { + if (jumpType == JumpFixed) + return LinkInvalid; + + // for patchable jump we must leave space for the longest code sequence + if (jumpType == JumpNoConditionFixedSize) + return LinkBX; + if (jumpType == JumpConditionFixedSize) + return LinkConditionalBX; + + const int paddingSize = JUMP_ENUM_SIZE(jumpType); + + if (jumpType == JumpCondition) { + // 2-byte conditional T1 + const uint16_t* jumpT1Location = reinterpret_cast_ptr<const uint16_t*>(from - (paddingSize - JUMP_ENUM_SIZE(LinkJumpT1))); + if (canBeJumpT1(jumpT1Location, to)) + return LinkJumpT1; + // 4-byte conditional T3 + const uint16_t* jumpT3Location = reinterpret_cast_ptr<const uint16_t*>(from - (paddingSize - JUMP_ENUM_SIZE(LinkJumpT3))); + if (canBeJumpT3(jumpT3Location, to)) + return LinkJumpT3; + // 4-byte conditional T4 with IT + const uint16_t* conditionalJumpT4Location = + reinterpret_cast_ptr<const uint16_t*>(from - (paddingSize - JUMP_ENUM_SIZE(LinkConditionalJumpT4))); + if (canBeJumpT4(conditionalJumpT4Location, to)) + return LinkConditionalJumpT4; + } else { + // 2-byte unconditional T2 + const uint16_t* jumpT2Location = reinterpret_cast_ptr<const uint16_t*>(from - (paddingSize - JUMP_ENUM_SIZE(LinkJumpT2))); + if (canBeJumpT2(jumpT2Location, to)) + return LinkJumpT2; + // 4-byte unconditional T4 + const uint16_t* jumpT4Location = reinterpret_cast_ptr<const uint16_t*>(from - (paddingSize - JUMP_ENUM_SIZE(LinkJumpT4))); + if (canBeJumpT4(jumpT4Location, to)) + return LinkJumpT4; + // use long jump sequence + return LinkBX; + } + + ASSERT(jumpType == JumpCondition); + return LinkConditionalBX; + } + + JumpLinkType computeJumpType(LinkRecord& record, const uint8_t* from, const uint8_t* to) + { + JumpLinkType linkType = computeJumpType(record.type(), from, to); + record.setLinkType(linkType); + return linkType; + } + + void recordLinkOffsets(int32_t regionStart, int32_t regionEnd, int32_t offset) + { + int32_t ptr = regionStart / sizeof(int32_t); + const int32_t end = regionEnd / sizeof(int32_t); + int32_t* offsets = static_cast<int32_t*>(m_formatter.data()); + while (ptr < end) + offsets[ptr++] = offset; + } + + Vector<LinkRecord, 0, UnsafeVectorOverflow>& jumpsToLink() + { + std::sort(m_jumpsToLink.begin(), m_jumpsToLink.end(), linkRecordSourceComparator); + return m_jumpsToLink; + } + + void ALWAYS_INLINE link(LinkRecord& record, uint8_t* from, uint8_t* to) + { + switch (record.linkType()) { + case LinkJumpT1: + linkJumpT1(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkJumpT2: + linkJumpT2(reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkJumpT3: + linkJumpT3(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkJumpT4: + linkJumpT4(reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkConditionalJumpT4: + linkConditionalJumpT4(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkConditionalBX: + linkConditionalBX(record.condition(), reinterpret_cast_ptr<uint16_t*>(from), to); + break; + case LinkBX: + linkBX(reinterpret_cast_ptr<uint16_t*>(from), to); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + break; + } + } + + void* unlinkedCode() { return m_formatter.data(); } + size_t codeSize() const { return m_formatter.codeSize(); } + + static unsigned getCallReturnOffset(AssemblerLabel call) + { + ASSERT(call.isSet()); + return call.m_offset; + } + + // Linking & patching: + // + // 'link' and 'patch' methods are for use on unprotected code - such as the code + // within the AssemblerBuffer, and code being patched by the patch buffer. Once + // code has been finalized it is (platform support permitting) within a non- + // writable region of memory; to modify the code in an execute-only execuable + // pool the 'repatch' and 'relink' methods should be used. + + void linkJump(AssemblerLabel from, AssemblerLabel to, JumpType type, Condition condition) + { + ASSERT(to.isSet()); + ASSERT(from.isSet()); + m_jumpsToLink.append(LinkRecord(from.m_offset, to.m_offset, type, condition)); + } + + static void linkJump(void* code, AssemblerLabel from, void* to) + { + ASSERT(from.isSet()); + + uint16_t* location = reinterpret_cast<uint16_t*>(reinterpret_cast<intptr_t>(code) + from.m_offset); + linkJumpAbsolute(location, to); + } + + static void linkCall(void* code, AssemblerLabel from, void* to) + { + ASSERT(!(reinterpret_cast<intptr_t>(code) & 1)); + ASSERT(from.isSet()); + ASSERT(reinterpret_cast<intptr_t>(to) & 1); + + setPointer(reinterpret_cast<uint16_t*>(reinterpret_cast<intptr_t>(code) + from.m_offset) - 1, to, false); + } + + static void linkPointer(void* code, AssemblerLabel where, void* value) + { + setPointer(reinterpret_cast<char*>(code) + where.m_offset, value, false); + } + + static void relinkJump(void* from, void* to) + { + ASSERT(!(reinterpret_cast<intptr_t>(from) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(to) & 1)); + + linkJumpAbsolute(reinterpret_cast<uint16_t*>(from), to); + + cacheFlush(reinterpret_cast<uint16_t*>(from) - 5, 5 * sizeof(uint16_t)); + } + + static void relinkCall(void* from, void* to) + { + ASSERT(!(reinterpret_cast<intptr_t>(from) & 1)); + ASSERT(reinterpret_cast<intptr_t>(to) & 1); + + setPointer(reinterpret_cast<uint16_t*>(from) - 1, to, true); + } + + static void* readCallTarget(void* from) + { + return readPointer(reinterpret_cast<uint16_t*>(from) - 1); + } + + static void repatchInt32(void* where, int32_t value) + { + ASSERT(!(reinterpret_cast<intptr_t>(where) & 1)); + + setInt32(where, value, true); + } + + static void repatchCompact(void* where, int32_t offset) + { + ASSERT(offset >= -255 && offset <= 255); + + bool add = true; + if (offset < 0) { + add = false; + offset = -offset; + } + + offset |= (add << 9); + offset |= (1 << 10); + offset |= (1 << 11); + + uint16_t* location = reinterpret_cast<uint16_t*>(where); + location[1] &= ~((1 << 12) - 1); + location[1] |= offset; + cacheFlush(location, sizeof(uint16_t) * 2); + } + + static void repatchPointer(void* where, void* value) + { + ASSERT(!(reinterpret_cast<intptr_t>(where) & 1)); + + setPointer(where, value, true); + } + + static void* readPointer(void* where) + { + return reinterpret_cast<void*>(readInt32(where)); + } + + static void replaceWithJump(void* instructionStart, void* to) + { + ASSERT(!(bitwise_cast<uintptr_t>(instructionStart) & 1)); + ASSERT(!(bitwise_cast<uintptr_t>(to) & 1)); + +#if OS(LINUX) || OS(QNX) + if (canBeJumpT4(reinterpret_cast<uint16_t*>(instructionStart), to)) { + uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 2; + linkJumpT4(ptr, to); + cacheFlush(ptr - 2, sizeof(uint16_t) * 2); + } else { + uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 5; + linkBX(ptr, to); + cacheFlush(ptr - 5, sizeof(uint16_t) * 5); + } +#else + uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart) + 2; + linkJumpT4(ptr, to); + cacheFlush(ptr - 2, sizeof(uint16_t) * 2); +#endif + } + + static ptrdiff_t maxJumpReplacementSize() + { +#if OS(LINUX) || OS(QNX) + return 10; +#else + return 4; +#endif + } + + static void replaceWithLoad(void* instructionStart) + { + ASSERT(!(bitwise_cast<uintptr_t>(instructionStart) & 1)); + uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart); + switch (ptr[0] & 0xFFF0) { + case OP_LDR_imm_T3: + break; + case OP_ADD_imm_T3: + ASSERT(!(ptr[1] & 0xF000)); + ptr[0] &= 0x000F; + ptr[0] |= OP_LDR_imm_T3; + ptr[1] |= (ptr[1] & 0x0F00) << 4; + ptr[1] &= 0xF0FF; + cacheFlush(ptr, sizeof(uint16_t) * 2); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + static void replaceWithAddressComputation(void* instructionStart) + { + ASSERT(!(bitwise_cast<uintptr_t>(instructionStart) & 1)); + uint16_t* ptr = reinterpret_cast<uint16_t*>(instructionStart); + switch (ptr[0] & 0xFFF0) { + case OP_LDR_imm_T3: + ASSERT(!(ptr[1] & 0x0F00)); + ptr[0] &= 0x000F; + ptr[0] |= OP_ADD_imm_T3; + ptr[1] |= (ptr[1] & 0xF000) >> 4; + ptr[1] &= 0x0FFF; + cacheFlush(ptr, sizeof(uint16_t) * 2); + break; + case OP_ADD_imm_T3: + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + unsigned debugOffset() { return m_formatter.debugOffset(); } + +#if OS(LINUX) + static inline void linuxPageFlush(uintptr_t begin, uintptr_t end) + { + asm volatile( + "push {r7}\n" + "mov r0, %0\n" + "mov r1, %1\n" + "movw r7, #0x2\n" + "movt r7, #0xf\n" + "movs r2, #0x0\n" + "svc 0x0\n" + "pop {r7}\n" + : + : "r" (begin), "r" (end) + : "r0", "r1", "r2"); + } +#endif + + static void cacheFlush(void* code, size_t size) + { +#if OS(IOS) + sys_cache_control(kCacheFunctionPrepareForExecution, code, size); +#elif OS(LINUX) + size_t page = pageSize(); + uintptr_t current = reinterpret_cast<uintptr_t>(code); + uintptr_t end = current + size; + uintptr_t firstPageEnd = (current & ~(page - 1)) + page; + + if (end <= firstPageEnd) { + linuxPageFlush(current, end); + return; + } + + linuxPageFlush(current, firstPageEnd); + + for (current = firstPageEnd; current + page < end; current += page) + linuxPageFlush(current, current + page); + + linuxPageFlush(current, end); +#elif OS(WINCE) + CacheRangeFlush(code, size, CACHE_SYNC_ALL); +#elif OS(QNX) +#if !ENABLE(ASSEMBLER_WX_EXCLUSIVE) + msync(code, size, MS_INVALIDATE_ICACHE); +#else + UNUSED_PARAM(code); + UNUSED_PARAM(size); +#endif +#else +#error "The cacheFlush support is missing on this platform." +#endif + } + +private: + // VFP operations commonly take one or more 5-bit operands, typically representing a + // floating point register number. This will commonly be encoded in the instruction + // in two parts, with one single bit field, and one 4-bit field. In the case of + // double precision operands the high bit of the register number will be encoded + // separately, and for single precision operands the high bit of the register number + // will be encoded individually. + // VFPOperand encapsulates a 5-bit VFP operand, with bits 0..3 containing the 4-bit + // field to be encoded together in the instruction (the low 4-bits of a double + // register number, or the high 4-bits of a single register number), and bit 4 + // contains the bit value to be encoded individually. + struct VFPOperand { + explicit VFPOperand(uint32_t value) + : m_value(value) + { + ASSERT(!(m_value & ~0x1f)); + } + + VFPOperand(FPDoubleRegisterID reg) + : m_value(reg) + { + } + + VFPOperand(RegisterID reg) + : m_value(reg) + { + } + + VFPOperand(FPSingleRegisterID reg) + : m_value(((reg & 1) << 4) | (reg >> 1)) // rotate the lowest bit of 'reg' to the top. + { + } + + uint32_t bits1() + { + return m_value >> 4; + } + + uint32_t bits4() + { + return m_value & 0xf; + } + + uint32_t m_value; + }; + + VFPOperand vcvtOp(bool toInteger, bool isUnsigned, bool isRoundZero) + { + // Cannot specify rounding when converting to float. + ASSERT(toInteger || !isRoundZero); + + uint32_t op = 0x8; + if (toInteger) { + // opc2 indicates both toInteger & isUnsigned. + op |= isUnsigned ? 0x4 : 0x5; + // 'op' field in instruction is isRoundZero + if (isRoundZero) + op |= 0x10; + } else { + ASSERT(!isRoundZero); + // 'op' field in instruction is isUnsigned + if (!isUnsigned) + op |= 0x10; + } + return VFPOperand(op); + } + + static void setInt32(void* code, uint32_t value, bool flush) + { + uint16_t* location = reinterpret_cast<uint16_t*>(code); + ASSERT(isMOV_imm_T3(location - 4) && isMOVT(location - 2)); + + ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(value)); + ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(value >> 16)); + location[-4] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16); + location[-3] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-3] >> 8) & 0xf, lo16); + location[-2] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16); + location[-1] = twoWordOp5i6Imm4Reg4EncodedImmSecond((location[-1] >> 8) & 0xf, hi16); + + if (flush) + cacheFlush(location - 4, 4 * sizeof(uint16_t)); + } + + static int32_t readInt32(void* code) + { + uint16_t* location = reinterpret_cast<uint16_t*>(code); + ASSERT(isMOV_imm_T3(location - 4) && isMOVT(location - 2)); + + ARMThumbImmediate lo16; + ARMThumbImmediate hi16; + decodeTwoWordOp5i6Imm4Reg4EncodedImmFirst(lo16, location[-4]); + decodeTwoWordOp5i6Imm4Reg4EncodedImmSecond(lo16, location[-3]); + decodeTwoWordOp5i6Imm4Reg4EncodedImmFirst(hi16, location[-2]); + decodeTwoWordOp5i6Imm4Reg4EncodedImmSecond(hi16, location[-1]); + uint32_t result = hi16.asUInt16(); + result <<= 16; + result |= lo16.asUInt16(); + return static_cast<int32_t>(result); + } + + static void setUInt7ForLoad(void* code, ARMThumbImmediate imm) + { + // Requires us to have planted a LDR_imm_T1 + ASSERT(imm.isValid()); + ASSERT(imm.isUInt7()); + uint16_t* location = reinterpret_cast<uint16_t*>(code); + location[0] &= ~((static_cast<uint16_t>(0x7f) >> 2) << 6); + location[0] |= (imm.getUInt7() >> 2) << 6; + cacheFlush(location, sizeof(uint16_t)); + } + + static void setPointer(void* code, void* value, bool flush) + { + setInt32(code, reinterpret_cast<uint32_t>(value), flush); + } + + static bool isB(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return ((instruction[0] & 0xf800) == OP_B_T4a) && ((instruction[1] & 0xd000) == OP_B_T4b); + } + + static bool isBX(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return (instruction[0] & 0xff87) == OP_BX; + } + + static bool isMOV_imm_T3(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return ((instruction[0] & 0xFBF0) == OP_MOV_imm_T3) && ((instruction[1] & 0x8000) == 0); + } + + static bool isMOVT(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return ((instruction[0] & 0xFBF0) == OP_MOVT) && ((instruction[1] & 0x8000) == 0); + } + + static bool isNOP_T1(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return instruction[0] == OP_NOP_T1; + } + + static bool isNOP_T2(void* address) + { + uint16_t* instruction = static_cast<uint16_t*>(address); + return (instruction[0] == OP_NOP_T2a) && (instruction[1] == OP_NOP_T2b); + } + + static bool canBeJumpT1(const uint16_t* instruction, const void* target) + { + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + // It does not appear to be documented in the ARM ARM (big surprise), but + // for OP_B_T1 the branch displacement encoded in the instruction is 2 + // less than the actual displacement. + relative -= 2; + return ((relative << 23) >> 23) == relative; + } + + static bool canBeJumpT2(const uint16_t* instruction, const void* target) + { + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + // It does not appear to be documented in the ARM ARM (big surprise), but + // for OP_B_T2 the branch displacement encoded in the instruction is 2 + // less than the actual displacement. + relative -= 2; + return ((relative << 20) >> 20) == relative; + } + + static bool canBeJumpT3(const uint16_t* instruction, const void* target) + { + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + return ((relative << 11) >> 11) == relative; + } + + static bool canBeJumpT4(const uint16_t* instruction, const void* target) + { + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + return ((relative << 7) >> 7) == relative; + } + + void linkJumpT1(Condition cond, uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + ASSERT(canBeJumpT1(instruction, target)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + // It does not appear to be documented in the ARM ARM (big surprise), but + // for OP_B_T1 the branch displacement encoded in the instruction is 2 + // less than the actual displacement. + relative -= 2; + + // All branch offsets should be an even distance. + ASSERT(!(relative & 1)); + instruction[-1] = OP_B_T1 | ((cond & 0xf) << 8) | ((relative & 0x1fe) >> 1); + } + + static void linkJumpT2(uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + ASSERT(canBeJumpT2(instruction, target)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + // It does not appear to be documented in the ARM ARM (big surprise), but + // for OP_B_T2 the branch displacement encoded in the instruction is 2 + // less than the actual displacement. + relative -= 2; + + // All branch offsets should be an even distance. + ASSERT(!(relative & 1)); + instruction[-1] = OP_B_T2 | ((relative & 0xffe) >> 1); + } + + void linkJumpT3(Condition cond, uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + ASSERT(canBeJumpT3(instruction, target)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + + // All branch offsets should be an even distance. + ASSERT(!(relative & 1)); + instruction[-2] = OP_B_T3a | ((relative & 0x100000) >> 10) | ((cond & 0xf) << 6) | ((relative & 0x3f000) >> 12); + instruction[-1] = OP_B_T3b | ((relative & 0x80000) >> 8) | ((relative & 0x40000) >> 5) | ((relative & 0xffe) >> 1); + } + + static void linkJumpT4(uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + ASSERT(canBeJumpT4(instruction, target)); + + intptr_t relative = reinterpret_cast<intptr_t>(target) - (reinterpret_cast<intptr_t>(instruction)); + // ARM encoding for the top two bits below the sign bit is 'peculiar'. + if (relative >= 0) + relative ^= 0xC00000; + + // All branch offsets should be an even distance. + ASSERT(!(relative & 1)); + instruction[-2] = OP_B_T4a | ((relative & 0x1000000) >> 14) | ((relative & 0x3ff000) >> 12); + instruction[-1] = OP_B_T4b | ((relative & 0x800000) >> 10) | ((relative & 0x400000) >> 11) | ((relative & 0xffe) >> 1); + } + + void linkConditionalJumpT4(Condition cond, uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + instruction[-3] = ifThenElse(cond) | OP_IT; + linkJumpT4(instruction, target); + } + + static void linkBX(uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + const uint16_t JUMP_TEMPORARY_REGISTER = ARMRegisters::ip; + ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) + 1)); + ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) >> 16)); + instruction[-5] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16); + instruction[-4] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16); + instruction[-3] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16); + instruction[-2] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16); + instruction[-1] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3); + } + + void linkConditionalBX(Condition cond, uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + linkBX(instruction, target); + instruction[-6] = ifThenElse(cond, true, true) | OP_IT; + } + + static void linkJumpAbsolute(uint16_t* instruction, void* target) + { + // FIMXE: this should be up in the MacroAssembler layer. :-( + ASSERT(!(reinterpret_cast<intptr_t>(instruction) & 1)); + ASSERT(!(reinterpret_cast<intptr_t>(target) & 1)); + + ASSERT((isMOV_imm_T3(instruction - 5) && isMOVT(instruction - 3) && isBX(instruction - 1)) + || (isNOP_T1(instruction - 5) && isNOP_T2(instruction - 4) && isB(instruction - 2))); + + if (canBeJumpT4(instruction, target)) { + // There may be a better way to fix this, but right now put the NOPs first, since in the + // case of an conditional branch this will be coming after an ITTT predicating *three* + // instructions! Looking backwards to modify the ITTT to an IT is not easy, due to + // variable wdith encoding - the previous instruction might *look* like an ITTT but + // actually be the second half of a 2-word op. + instruction[-5] = OP_NOP_T1; + instruction[-4] = OP_NOP_T2a; + instruction[-3] = OP_NOP_T2b; + linkJumpT4(instruction, target); + } else { + const uint16_t JUMP_TEMPORARY_REGISTER = ARMRegisters::ip; + ARMThumbImmediate lo16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) + 1)); + ARMThumbImmediate hi16 = ARMThumbImmediate::makeUInt16(static_cast<uint16_t>(reinterpret_cast<uint32_t>(target) >> 16)); + instruction[-5] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOV_imm_T3, lo16); + instruction[-4] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, lo16); + instruction[-3] = twoWordOp5i6Imm4Reg4EncodedImmFirst(OP_MOVT, hi16); + instruction[-2] = twoWordOp5i6Imm4Reg4EncodedImmSecond(JUMP_TEMPORARY_REGISTER, hi16); + instruction[-1] = OP_BX | (JUMP_TEMPORARY_REGISTER << 3); + } + } + + static uint16_t twoWordOp5i6Imm4Reg4EncodedImmFirst(uint16_t op, ARMThumbImmediate imm) + { + return op | (imm.m_value.i << 10) | imm.m_value.imm4; + } + + static void decodeTwoWordOp5i6Imm4Reg4EncodedImmFirst(ARMThumbImmediate& result, uint16_t value) + { + result.m_value.i = (value >> 10) & 1; + result.m_value.imm4 = value & 15; + } + + static uint16_t twoWordOp5i6Imm4Reg4EncodedImmSecond(uint16_t rd, ARMThumbImmediate imm) + { + return (imm.m_value.imm3 << 12) | (rd << 8) | imm.m_value.imm8; + } + + static void decodeTwoWordOp5i6Imm4Reg4EncodedImmSecond(ARMThumbImmediate& result, uint16_t value) + { + result.m_value.imm3 = (value >> 12) & 7; + result.m_value.imm8 = value & 255; + } + + class ARMInstructionFormatter { + public: + ALWAYS_INLINE void oneWordOp5Reg3Imm8(OpcodeID op, RegisterID rd, uint8_t imm) + { + m_buffer.putShort(op | (rd << 8) | imm); + } + + ALWAYS_INLINE void oneWordOp5Imm5Reg3Reg3(OpcodeID op, uint8_t imm, RegisterID reg1, RegisterID reg2) + { + m_buffer.putShort(op | (imm << 6) | (reg1 << 3) | reg2); + } + + ALWAYS_INLINE void oneWordOp7Reg3Reg3Reg3(OpcodeID op, RegisterID reg1, RegisterID reg2, RegisterID reg3) + { + m_buffer.putShort(op | (reg1 << 6) | (reg2 << 3) | reg3); + } + + ALWAYS_INLINE void oneWordOp8Imm8(OpcodeID op, uint8_t imm) + { + m_buffer.putShort(op | imm); + } + + ALWAYS_INLINE void oneWordOp8RegReg143(OpcodeID op, RegisterID reg1, RegisterID reg2) + { + m_buffer.putShort(op | ((reg2 & 8) << 4) | (reg1 << 3) | (reg2 & 7)); + } + + ALWAYS_INLINE void oneWordOp9Imm7(OpcodeID op, uint8_t imm) + { + m_buffer.putShort(op | imm); + } + + ALWAYS_INLINE void oneWordOp10Reg3Reg3(OpcodeID op, RegisterID reg1, RegisterID reg2) + { + m_buffer.putShort(op | (reg1 << 3) | reg2); + } + + ALWAYS_INLINE void twoWordOp12Reg4FourFours(OpcodeID1 op, RegisterID reg, FourFours ff) + { + m_buffer.putShort(op | reg); + m_buffer.putShort(ff.m_u.value); + } + + ALWAYS_INLINE void twoWordOp16FourFours(OpcodeID1 op, FourFours ff) + { + m_buffer.putShort(op); + m_buffer.putShort(ff.m_u.value); + } + + ALWAYS_INLINE void twoWordOp16Op16(OpcodeID1 op1, OpcodeID2 op2) + { + m_buffer.putShort(op1); + m_buffer.putShort(op2); + } + + ALWAYS_INLINE void twoWordOp5i6Imm4Reg4EncodedImm(OpcodeID1 op, int imm4, RegisterID rd, ARMThumbImmediate imm) + { + ARMThumbImmediate newImm = imm; + newImm.m_value.imm4 = imm4; + + m_buffer.putShort(ARMv7Assembler::twoWordOp5i6Imm4Reg4EncodedImmFirst(op, newImm)); + m_buffer.putShort(ARMv7Assembler::twoWordOp5i6Imm4Reg4EncodedImmSecond(rd, newImm)); + } + + ALWAYS_INLINE void twoWordOp12Reg4Reg4Imm12(OpcodeID1 op, RegisterID reg1, RegisterID reg2, uint16_t imm) + { + m_buffer.putShort(op | reg1); + m_buffer.putShort((reg2 << 12) | imm); + } + + ALWAYS_INLINE void twoWordOp12Reg40Imm3Reg4Imm20Imm5(OpcodeID1 op, RegisterID reg1, RegisterID reg2, uint16_t imm1, uint16_t imm2, uint16_t imm3) + { + m_buffer.putShort(op | reg1); + m_buffer.putShort((imm1 << 12) | (reg2 << 8) | (imm2 << 6) | imm3); + } + + // Formats up instructions of the pattern: + // 111111111B11aaaa:bbbb222SA2C2cccc + // Where 1s in the pattern come from op1, 2s in the pattern come from op2, S is the provided size bit. + // Operands provide 5 bit values of the form Aaaaa, Bbbbb, Ccccc. + ALWAYS_INLINE void vfpOp(OpcodeID1 op1, OpcodeID2 op2, bool size, VFPOperand a, VFPOperand b, VFPOperand c) + { + ASSERT(!(op1 & 0x004f)); + ASSERT(!(op2 & 0xf1af)); + m_buffer.putShort(op1 | b.bits1() << 6 | a.bits4()); + m_buffer.putShort(op2 | b.bits4() << 12 | size << 8 | a.bits1() << 7 | c.bits1() << 5 | c.bits4()); + } + + // Arm vfp addresses can be offset by a 9-bit ones-comp immediate, left shifted by 2. + // (i.e. +/-(0..255) 32-bit words) + ALWAYS_INLINE void vfpMemOp(OpcodeID1 op1, OpcodeID2 op2, bool size, RegisterID rn, VFPOperand rd, int32_t imm) + { + bool up = true; + if (imm < 0) { + imm = -imm; + up = false; + } + + uint32_t offset = imm; + ASSERT(!(offset & ~0x3fc)); + offset >>= 2; + + m_buffer.putShort(op1 | (up << 7) | rd.bits1() << 6 | rn); + m_buffer.putShort(op2 | rd.bits4() << 12 | size << 8 | offset); + } + + // Administrative methods: + + size_t codeSize() const { return m_buffer.codeSize(); } + AssemblerLabel label() const { return m_buffer.label(); } + bool isAligned(int alignment) const { return m_buffer.isAligned(alignment); } + void* data() const { return m_buffer.data(); } + + unsigned debugOffset() { return m_buffer.debugOffset(); } + + private: + AssemblerBuffer m_buffer; + } m_formatter; + + Vector<LinkRecord, 0, UnsafeVectorOverflow> m_jumpsToLink; + int m_indexOfLastWatchpoint; + int m_indexOfTailOfLastWatchpoint; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(ARM_THUMB2) + +#endif // ARMAssembler_h diff --git a/src/3rdparty/masm/assembler/AbstractMacroAssembler.h b/src/3rdparty/masm/assembler/AbstractMacroAssembler.h new file mode 100644 index 0000000000..95eaf7d99d --- /dev/null +++ b/src/3rdparty/masm/assembler/AbstractMacroAssembler.h @@ -0,0 +1,842 @@ +/* + * Copyright (C) 2008, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef AbstractMacroAssembler_h +#define AbstractMacroAssembler_h + +#include "AssemblerBuffer.h" +#include "CodeLocation.h" +#include "MacroAssemblerCodeRef.h" +#include <wtf/CryptographicallyRandomNumber.h> +#include <wtf/Noncopyable.h> +#include <wtf/UnusedParam.h> + +#if ENABLE(ASSEMBLER) + + +#if PLATFORM(QT) +#define ENABLE_JIT_CONSTANT_BLINDING 0 +#endif + +#ifndef ENABLE_JIT_CONSTANT_BLINDING +#define ENABLE_JIT_CONSTANT_BLINDING 1 +#endif + +namespace JSC { + +class JumpReplacementWatchpoint; +class LinkBuffer; +class RepatchBuffer; +class Watchpoint; +namespace DFG { +struct OSRExit; +} + +template <class AssemblerType> +class AbstractMacroAssembler { +public: + friend class JITWriteBarrierBase; + typedef AssemblerType AssemblerType_T; + + typedef MacroAssemblerCodePtr CodePtr; + typedef MacroAssemblerCodeRef CodeRef; + + class Jump; + + typedef typename AssemblerType::RegisterID RegisterID; + + // Section 1: MacroAssembler operand types + // + // The following types are used as operands to MacroAssembler operations, + // describing immediate and memory operands to the instructions to be planted. + + enum Scale { + TimesOne, + TimesTwo, + TimesFour, + TimesEight, + }; + + // Address: + // + // Describes a simple base-offset address. + struct Address { + explicit Address(RegisterID base, int32_t offset = 0) + : base(base) + , offset(offset) + { + } + + RegisterID base; + int32_t offset; + }; + + struct ExtendedAddress { + explicit ExtendedAddress(RegisterID base, intptr_t offset = 0) + : base(base) + , offset(offset) + { + } + + RegisterID base; + intptr_t offset; + }; + + // ImplicitAddress: + // + // This class is used for explicit 'load' and 'store' operations + // (as opposed to situations in which a memory operand is provided + // to a generic operation, such as an integer arithmetic instruction). + // + // In the case of a load (or store) operation we want to permit + // addresses to be implicitly constructed, e.g. the two calls: + // + // load32(Address(addrReg), destReg); + // load32(addrReg, destReg); + // + // Are equivalent, and the explicit wrapping of the Address in the former + // is unnecessary. + struct ImplicitAddress { + ImplicitAddress(RegisterID base) + : base(base) + , offset(0) + { + } + + ImplicitAddress(Address address) + : base(address.base) + , offset(address.offset) + { + } + + RegisterID base; + int32_t offset; + }; + + // BaseIndex: + // + // Describes a complex addressing mode. + struct BaseIndex { + BaseIndex(RegisterID base, RegisterID index, Scale scale, int32_t offset = 0) + : base(base) + , index(index) + , scale(scale) + , offset(offset) + { + } + + RegisterID base; + RegisterID index; + Scale scale; + int32_t offset; + }; + + // AbsoluteAddress: + // + // Describes an memory operand given by a pointer. For regular load & store + // operations an unwrapped void* will be used, rather than using this. + struct AbsoluteAddress { + explicit AbsoluteAddress(const void* ptr) + : m_ptr(ptr) + { + } + + const void* m_ptr; + }; + + // TrustedImmPtr: + // + // A pointer sized immediate operand to an instruction - this is wrapped + // in a class requiring explicit construction in order to differentiate + // from pointers used as absolute addresses to memory operations + struct TrustedImmPtr { + TrustedImmPtr() { } + + explicit TrustedImmPtr(const void* value) + : m_value(value) + { + } + + // This is only here so that TrustedImmPtr(0) does not confuse the C++ + // overload handling rules. + explicit TrustedImmPtr(int value) + : m_value(0) + { + ASSERT_UNUSED(value, !value); + } + + explicit TrustedImmPtr(size_t value) + : m_value(reinterpret_cast<void*>(value)) + { + } + + intptr_t asIntptr() + { + return reinterpret_cast<intptr_t>(m_value); + } + + const void* m_value; + }; + + struct ImmPtr : +#if ENABLE(JIT_CONSTANT_BLINDING) + private TrustedImmPtr +#else + public TrustedImmPtr +#endif + { + explicit ImmPtr(const void* value) + : TrustedImmPtr(value) + { + } + + TrustedImmPtr asTrustedImmPtr() { return *this; } + }; + + // TrustedImm32: + // + // A 32bit immediate operand to an instruction - this is wrapped in a + // class requiring explicit construction in order to prevent RegisterIDs + // (which are implemented as an enum) from accidentally being passed as + // immediate values. + struct TrustedImm32 { + TrustedImm32() { } + + explicit TrustedImm32(int32_t value) + : m_value(value) + { + } + +#if !CPU(X86_64) + explicit TrustedImm32(TrustedImmPtr ptr) + : m_value(ptr.asIntptr()) + { + } +#endif + + int32_t m_value; + }; + + + struct Imm32 : +#if ENABLE(JIT_CONSTANT_BLINDING) + private TrustedImm32 +#else + public TrustedImm32 +#endif + { + explicit Imm32(int32_t value) + : TrustedImm32(value) + { + } +#if !CPU(X86_64) + explicit Imm32(TrustedImmPtr ptr) + : TrustedImm32(ptr) + { + } +#endif + const TrustedImm32& asTrustedImm32() const { return *this; } + + }; + + // TrustedImm64: + // + // A 64bit immediate operand to an instruction - this is wrapped in a + // class requiring explicit construction in order to prevent RegisterIDs + // (which are implemented as an enum) from accidentally being passed as + // immediate values. + struct TrustedImm64 { + TrustedImm64() { } + + explicit TrustedImm64(int64_t value) + : m_value(value) + { + } + +#if CPU(X86_64) + explicit TrustedImm64(TrustedImmPtr ptr) + : m_value(ptr.asIntptr()) + { + } +#endif + + int64_t m_value; + }; + + struct Imm64 : +#if ENABLE(JIT_CONSTANT_BLINDING) + private TrustedImm64 +#else + public TrustedImm64 +#endif + { + explicit Imm64(int64_t value) + : TrustedImm64(value) + { + } +#if CPU(X86_64) + explicit Imm64(TrustedImmPtr ptr) + : TrustedImm64(ptr) + { + } +#endif + const TrustedImm64& asTrustedImm64() const { return *this; } + }; + + // Section 2: MacroAssembler code buffer handles + // + // The following types are used to reference items in the code buffer + // during JIT code generation. For example, the type Jump is used to + // track the location of a jump instruction so that it may later be + // linked to a label marking its destination. + + + // Label: + // + // A Label records a point in the generated instruction stream, typically such that + // it may be used as a destination for a jump. + class Label { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend struct DFG::OSRExit; + friend class Jump; + friend class JumpReplacementWatchpoint; + friend class MacroAssemblerCodeRef; + friend class LinkBuffer; + friend class Watchpoint; + + public: + Label() + { + } + + Label(AbstractMacroAssembler<AssemblerType>* masm) + : m_label(masm->m_assembler.label()) + { + } + + bool isSet() const { return m_label.isSet(); } + private: + AssemblerLabel m_label; + }; + + // ConvertibleLoadLabel: + // + // A ConvertibleLoadLabel records a loadPtr instruction that can be patched to an addPtr + // so that: + // + // loadPtr(Address(a, i), b) + // + // becomes: + // + // addPtr(TrustedImmPtr(i), a, b) + class ConvertibleLoadLabel { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend class LinkBuffer; + + public: + ConvertibleLoadLabel() + { + } + + ConvertibleLoadLabel(AbstractMacroAssembler<AssemblerType>* masm) + : m_label(masm->m_assembler.labelIgnoringWatchpoints()) + { + } + + bool isSet() const { return m_label.isSet(); } + private: + AssemblerLabel m_label; + }; + + // DataLabelPtr: + // + // A DataLabelPtr is used to refer to a location in the code containing a pointer to be + // patched after the code has been generated. + class DataLabelPtr { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend class LinkBuffer; + public: + DataLabelPtr() + { + } + + DataLabelPtr(AbstractMacroAssembler<AssemblerType>* masm) + : m_label(masm->m_assembler.label()) + { + } + + bool isSet() const { return m_label.isSet(); } + + private: + AssemblerLabel m_label; + }; + + // DataLabel32: + // + // A DataLabelPtr is used to refer to a location in the code containing a pointer to be + // patched after the code has been generated. + class DataLabel32 { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend class LinkBuffer; + public: + DataLabel32() + { + } + + DataLabel32(AbstractMacroAssembler<AssemblerType>* masm) + : m_label(masm->m_assembler.label()) + { + } + + AssemblerLabel label() const { return m_label; } + + private: + AssemblerLabel m_label; + }; + + // DataLabelCompact: + // + // A DataLabelCompact is used to refer to a location in the code containing a + // compact immediate to be patched after the code has been generated. + class DataLabelCompact { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend class LinkBuffer; + public: + DataLabelCompact() + { + } + + DataLabelCompact(AbstractMacroAssembler<AssemblerType>* masm) + : m_label(masm->m_assembler.label()) + { + } + + DataLabelCompact(AssemblerLabel label) + : m_label(label) + { + } + + private: + AssemblerLabel m_label; + }; + + // Call: + // + // A Call object is a reference to a call instruction that has been planted + // into the code buffer - it is typically used to link the call, setting the + // relative offset such that when executed it will call to the desired + // destination. + class Call { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + + public: + enum Flags { + None = 0x0, + Linkable = 0x1, + Near = 0x2, + LinkableNear = 0x3, + }; + + Call() + : m_flags(None) + { + } + + Call(AssemblerLabel jmp, Flags flags) + : m_label(jmp) + , m_flags(flags) + { + } + + bool isFlagSet(Flags flag) + { + return m_flags & flag; + } + + static Call fromTailJump(Jump jump) + { + return Call(jump.m_label, Linkable); + } + + AssemblerLabel m_label; + private: + Flags m_flags; + }; + + // Jump: + // + // A jump object is a reference to a jump instruction that has been planted + // into the code buffer - it is typically used to link the jump, setting the + // relative offset such that when executed it will jump to the desired + // destination. + class Jump { + template<class TemplateAssemblerType> + friend class AbstractMacroAssembler; + friend class Call; + friend struct DFG::OSRExit; + friend class LinkBuffer; + public: + Jump() + { + } + +#if CPU(ARM_THUMB2) + // Fixme: this information should be stored in the instruction stream, not in the Jump object. + Jump(AssemblerLabel jmp, ARMv7Assembler::JumpType type = ARMv7Assembler::JumpNoCondition, ARMv7Assembler::Condition condition = ARMv7Assembler::ConditionInvalid) + : m_label(jmp) + , m_type(type) + , m_condition(condition) + { + } +#elif CPU(SH4) + Jump(AssemblerLabel jmp, SH4Assembler::JumpType type = SH4Assembler::JumpFar) + : m_label(jmp) + , m_type(type) + { + } +#else + Jump(AssemblerLabel jmp) + : m_label(jmp) + { + } +#endif + + Label label() const + { + Label result; + result.m_label = m_label; + return result; + } + + void link(AbstractMacroAssembler<AssemblerType>* masm) const + { +#if ENABLE(DFG_REGISTER_ALLOCATION_VALIDATION) + masm->checkRegisterAllocationAgainstBranchRange(m_label.m_offset, masm->debugOffset()); +#endif + +#if CPU(ARM_THUMB2) + masm->m_assembler.linkJump(m_label, masm->m_assembler.label(), m_type, m_condition); +#elif CPU(SH4) + masm->m_assembler.linkJump(m_label, masm->m_assembler.label(), m_type); +#else + masm->m_assembler.linkJump(m_label, masm->m_assembler.label()); +#endif + } + + void linkTo(Label label, AbstractMacroAssembler<AssemblerType>* masm) const + { +#if ENABLE(DFG_REGISTER_ALLOCATION_VALIDATION) + masm->checkRegisterAllocationAgainstBranchRange(label.m_label.m_offset, m_label.m_offset); +#endif + +#if CPU(ARM_THUMB2) + masm->m_assembler.linkJump(m_label, label.m_label, m_type, m_condition); +#else + masm->m_assembler.linkJump(m_label, label.m_label); +#endif + } + + bool isSet() const { return m_label.isSet(); } + + private: + AssemblerLabel m_label; +#if CPU(ARM_THUMB2) + ARMv7Assembler::JumpType m_type; + ARMv7Assembler::Condition m_condition; +#endif +#if CPU(SH4) + SH4Assembler::JumpType m_type; +#endif + }; + + struct PatchableJump { + PatchableJump() + { + } + + explicit PatchableJump(Jump jump) + : m_jump(jump) + { + } + + operator Jump&() { return m_jump; } + + Jump m_jump; + }; + + // JumpList: + // + // A JumpList is a set of Jump objects. + // All jumps in the set will be linked to the same destination. + class JumpList { + friend class LinkBuffer; + + public: + typedef Vector<Jump, 2> JumpVector; + + JumpList() { } + + JumpList(Jump jump) + { + append(jump); + } + + void link(AbstractMacroAssembler<AssemblerType>* masm) + { + size_t size = m_jumps.size(); + for (size_t i = 0; i < size; ++i) + m_jumps[i].link(masm); + m_jumps.clear(); + } + + void linkTo(Label label, AbstractMacroAssembler<AssemblerType>* masm) + { + size_t size = m_jumps.size(); + for (size_t i = 0; i < size; ++i) + m_jumps[i].linkTo(label, masm); + m_jumps.clear(); + } + + void append(Jump jump) + { + m_jumps.append(jump); + } + + void append(const JumpList& other) + { + m_jumps.append(other.m_jumps.begin(), other.m_jumps.size()); + } + + bool empty() + { + return !m_jumps.size(); + } + + void clear() + { + m_jumps.clear(); + } + + const JumpVector& jumps() const { return m_jumps; } + + private: + JumpVector m_jumps; + }; + + + // Section 3: Misc admin methods +#if ENABLE(DFG_JIT) + Label labelIgnoringWatchpoints() + { + Label result; + result.m_label = m_assembler.labelIgnoringWatchpoints(); + return result; + } +#else + Label labelIgnoringWatchpoints() + { + return label(); + } +#endif + + Label label() + { + return Label(this); + } + + void padBeforePatch() + { + // Rely on the fact that asking for a label already does the padding. + (void)label(); + } + + Label watchpointLabel() + { + Label result; + result.m_label = m_assembler.labelForWatchpoint(); + return result; + } + + Label align() + { + m_assembler.align(16); + return Label(this); + } + +#if ENABLE(DFG_REGISTER_ALLOCATION_VALIDATION) + class RegisterAllocationOffset { + public: + RegisterAllocationOffset(unsigned offset) + : m_offset(offset) + { + } + + void check(unsigned low, unsigned high) + { + RELEASE_ASSERT_WITH_MESSAGE(!(low <= m_offset && m_offset <= high), "Unsafe branch over register allocation at instruction offset %u in jump offset range %u..%u", m_offset, low, high); + } + + private: + unsigned m_offset; + }; + + void addRegisterAllocationAtOffset(unsigned offset) + { + m_registerAllocationForOffsets.append(RegisterAllocationOffset(offset)); + } + + void clearRegisterAllocationOffsets() + { + m_registerAllocationForOffsets.clear(); + } + + void checkRegisterAllocationAgainstBranchRange(unsigned offset1, unsigned offset2) + { + if (offset1 > offset2) + std::swap(offset1, offset2); + + size_t size = m_registerAllocationForOffsets.size(); + for (size_t i = 0; i < size; ++i) + m_registerAllocationForOffsets[i].check(offset1, offset2); + } +#endif + + template<typename T, typename U> + static ptrdiff_t differenceBetween(T from, U to) + { + return AssemblerType::getDifferenceBetweenLabels(from.m_label, to.m_label); + } + + static ptrdiff_t differenceBetweenCodePtr(const MacroAssemblerCodePtr& a, const MacroAssemblerCodePtr& b) + { + return reinterpret_cast<ptrdiff_t>(b.executableAddress()) - reinterpret_cast<ptrdiff_t>(a.executableAddress()); + } + + unsigned debugOffset() { return m_assembler.debugOffset(); } + + ALWAYS_INLINE static void cacheFlush(void* code, size_t size) + { + AssemblerType::cacheFlush(code, size); + } +protected: + AbstractMacroAssembler() + : m_randomSource(cryptographicallyRandomNumber()) + { + } + + AssemblerType m_assembler; + + uint32_t random() + { + return m_randomSource.getUint32(); + } + + WeakRandom m_randomSource; + +#if ENABLE(DFG_REGISTER_ALLOCATION_VALIDATION) + Vector<RegisterAllocationOffset, 10> m_registerAllocationForOffsets; +#endif + +#if ENABLE(JIT_CONSTANT_BLINDING) + static bool scratchRegisterForBlinding() { return false; } + static bool shouldBlindForSpecificArch(uint32_t) { return true; } + static bool shouldBlindForSpecificArch(uint64_t) { return true; } +#endif + + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkJump(void* code, Jump jump, CodeLocationLabel target) + { + AssemblerType::linkJump(code, jump.m_label, target.dataLocation()); + } + + static void linkPointer(void* code, AssemblerLabel label, void* value) + { + AssemblerType::linkPointer(code, label, value); + } + + static void* getLinkerAddress(void* code, AssemblerLabel label) + { + return AssemblerType::getRelocatedAddress(code, label); + } + + static unsigned getLinkerCallReturnOffset(Call call) + { + return AssemblerType::getCallReturnOffset(call.m_label); + } + + static void repatchJump(CodeLocationJump jump, CodeLocationLabel destination) + { + AssemblerType::relinkJump(jump.dataLocation(), destination.dataLocation()); + } + + static void repatchNearCall(CodeLocationNearCall nearCall, CodeLocationLabel destination) + { + AssemblerType::relinkCall(nearCall.dataLocation(), destination.executableAddress()); + } + + static void repatchCompact(CodeLocationDataLabelCompact dataLabelCompact, int32_t value) + { + AssemblerType::repatchCompact(dataLabelCompact.dataLocation(), value); + } + + static void repatchInt32(CodeLocationDataLabel32 dataLabel32, int32_t value) + { + AssemblerType::repatchInt32(dataLabel32.dataLocation(), value); + } + + static void repatchPointer(CodeLocationDataLabelPtr dataLabelPtr, void* value) + { + AssemblerType::repatchPointer(dataLabelPtr.dataLocation(), value); + } + + static void* readPointer(CodeLocationDataLabelPtr dataLabelPtr) + { + return AssemblerType::readPointer(dataLabelPtr.dataLocation()); + } + + static void replaceWithLoad(CodeLocationConvertibleLoad label) + { + AssemblerType::replaceWithLoad(label.dataLocation()); + } + + static void replaceWithAddressComputation(CodeLocationConvertibleLoad label) + { + AssemblerType::replaceWithAddressComputation(label.dataLocation()); + } +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // AbstractMacroAssembler_h diff --git a/src/3rdparty/masm/assembler/AssemblerBuffer.h b/src/3rdparty/masm/assembler/AssemblerBuffer.h new file mode 100644 index 0000000000..277ec1043c --- /dev/null +++ b/src/3rdparty/masm/assembler/AssemblerBuffer.h @@ -0,0 +1,181 @@ +/* + * Copyright (C) 2008, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef AssemblerBuffer_h +#define AssemblerBuffer_h + +#if ENABLE(ASSEMBLER) + +#include "ExecutableAllocator.h" +#include "JITCompilationEffort.h" +#include "JSGlobalData.h" +#include "stdint.h" +#include <string.h> +#include <wtf/Assertions.h> +#include <wtf/FastMalloc.h> +#include <wtf/StdLibExtras.h> + +namespace JSC { + + struct AssemblerLabel { + AssemblerLabel() + : m_offset(std::numeric_limits<uint32_t>::max()) + { + } + + explicit AssemblerLabel(uint32_t offset) + : m_offset(offset) + { + } + + bool isSet() const { return (m_offset != std::numeric_limits<uint32_t>::max()); } + + AssemblerLabel labelAtOffset(int offset) const + { + return AssemblerLabel(m_offset + offset); + } + + uint32_t m_offset; + }; + + class AssemblerBuffer { + static const int inlineCapacity = 128; + public: + AssemblerBuffer() + : m_storage(inlineCapacity) + , m_buffer(&(*m_storage.begin())) + , m_capacity(inlineCapacity) + , m_index(0) + { + } + + ~AssemblerBuffer() + { + } + + bool isAvailable(int space) + { + return m_index + space <= m_capacity; + } + + void ensureSpace(int space) + { + if (!isAvailable(space)) + grow(); + } + + bool isAligned(int alignment) const + { + return !(m_index & (alignment - 1)); + } + + template<typename IntegralType> + void putIntegral(IntegralType value) + { + ensureSpace(sizeof(IntegralType)); + putIntegralUnchecked(value); + } + + template<typename IntegralType> + void putIntegralUnchecked(IntegralType value) + { + ASSERT(isAvailable(sizeof(IntegralType))); + *reinterpret_cast_ptr<IntegralType*>(m_buffer + m_index) = value; + m_index += sizeof(IntegralType); + } + + void putByteUnchecked(int8_t value) { putIntegralUnchecked(value); } + void putByte(int8_t value) { putIntegral(value); } + void putShortUnchecked(int16_t value) { putIntegralUnchecked(value); } + void putShort(int16_t value) { putIntegral(value); } + void putIntUnchecked(int32_t value) { putIntegralUnchecked(value); } + void putInt(int32_t value) { putIntegral(value); } + void putInt64Unchecked(int64_t value) { putIntegralUnchecked(value); } + void putInt64(int64_t value) { putIntegral(value); } + + void* data() const + { + return m_buffer; + } + + size_t codeSize() const + { + return m_index; + } + + AssemblerLabel label() const + { + return AssemblerLabel(m_index); + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + if (!m_index) + return 0; + + RefPtr<ExecutableMemoryHandle> result = globalData.executableAllocator.allocate(globalData, m_index, ownerUID, effort); + + if (!result) + return 0; + + ExecutableAllocator::makeWritable(result->start(), result->sizeInBytes()); + + memcpy(result->start(), m_buffer, m_index); + + return result.release(); + } + + unsigned debugOffset() { return m_index; } + + protected: + void append(const char* data, int size) + { + if (!isAvailable(size)) + grow(size); + + memcpy(m_buffer + m_index, data, size); + m_index += size; + } + + void grow(int extraCapacity = 0) + { + m_capacity += m_capacity / 2 + extraCapacity; + + m_storage.grow(m_capacity); + m_buffer = &(*m_storage.begin()); + } + + private: + Vector<char, inlineCapacity, UnsafeVectorOverflow> m_storage; + char* m_buffer; + int m_capacity; + int m_index; + }; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // AssemblerBuffer_h diff --git a/src/3rdparty/masm/assembler/AssemblerBufferWithConstantPool.h b/src/3rdparty/masm/assembler/AssemblerBufferWithConstantPool.h new file mode 100644 index 0000000000..5377ef0c7a --- /dev/null +++ b/src/3rdparty/masm/assembler/AssemblerBufferWithConstantPool.h @@ -0,0 +1,342 @@ +/* + * Copyright (C) 2009 University of Szeged + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL UNIVERSITY OF SZEGED OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef AssemblerBufferWithConstantPool_h +#define AssemblerBufferWithConstantPool_h + +#if ENABLE(ASSEMBLER) + +#include "AssemblerBuffer.h" +#include <wtf/SegmentedVector.h> + +#define ASSEMBLER_HAS_CONSTANT_POOL 1 + +namespace JSC { + +/* + On a constant pool 4 or 8 bytes data can be stored. The values can be + constants or addresses. The addresses should be 32 or 64 bits. The constants + should be double-precisions float or integer numbers which are hard to be + encoded as few machine instructions. + + TODO: The pool is desinged to handle both 32 and 64 bits values, but + currently only the 4 bytes constants are implemented and tested. + + The AssemblerBuffer can contain multiple constant pools. Each pool is inserted + into the instruction stream - protected by a jump instruction from the + execution flow. + + The flush mechanism is called when no space remain to insert the next instruction + into the pool. Three values are used to determine when the constant pool itself + have to be inserted into the instruction stream (Assembler Buffer): + + - maxPoolSize: size of the constant pool in bytes, this value cannot be + larger than the maximum offset of a PC relative memory load + + - barrierSize: size of jump instruction in bytes which protects the + constant pool from execution + + - maxInstructionSize: maximum length of a machine instruction in bytes + + There are some callbacks which solve the target architecture specific + address handling: + + - TYPE patchConstantPoolLoad(TYPE load, int value): + patch the 'load' instruction with the index of the constant in the + constant pool and return the patched instruction. + + - void patchConstantPoolLoad(void* loadAddr, void* constPoolAddr): + patch the a PC relative load instruction at 'loadAddr' address with the + final relative offset. The offset can be computed with help of + 'constPoolAddr' (the address of the constant pool) and index of the + constant (which is stored previously in the load instruction itself). + + - TYPE placeConstantPoolBarrier(int size): + return with a constant pool barrier instruction which jumps over the + constant pool. + + The 'put*WithConstant*' functions should be used to place a data into the + constant pool. +*/ + +template <int maxPoolSize, int barrierSize, int maxInstructionSize, class AssemblerType> +class AssemblerBufferWithConstantPool : public AssemblerBuffer { + typedef SegmentedVector<uint32_t, 512> LoadOffsets; + using AssemblerBuffer::putIntegral; + using AssemblerBuffer::putIntegralUnchecked; +public: + typedef struct { + short high; + short low; + } TwoShorts; + + enum { + UniqueConst, + ReusableConst, + UnusedEntry, + }; + + AssemblerBufferWithConstantPool() + : AssemblerBuffer() + , m_numConsts(0) + , m_maxDistance(maxPoolSize) + , m_lastConstDelta(0) + { + m_pool = static_cast<uint32_t*>(fastMalloc(maxPoolSize)); + m_mask = static_cast<char*>(fastMalloc(maxPoolSize / sizeof(uint32_t))); + } + + ~AssemblerBufferWithConstantPool() + { + fastFree(m_mask); + fastFree(m_pool); + } + + void ensureSpace(int space) + { + flushIfNoSpaceFor(space); + AssemblerBuffer::ensureSpace(space); + } + + void ensureSpace(int insnSpace, int constSpace) + { + flushIfNoSpaceFor(insnSpace, constSpace); + AssemblerBuffer::ensureSpace(insnSpace); + } + + void ensureSpaceForAnyInstruction(int amount = 1) + { + flushIfNoSpaceFor(amount * maxInstructionSize, amount * sizeof(uint64_t)); + } + + bool isAligned(int alignment) + { + flushIfNoSpaceFor(alignment); + return AssemblerBuffer::isAligned(alignment); + } + + void putByteUnchecked(int value) + { + AssemblerBuffer::putByteUnchecked(value); + correctDeltas(1); + } + + void putByte(int value) + { + flushIfNoSpaceFor(1); + AssemblerBuffer::putByte(value); + correctDeltas(1); + } + + void putShortUnchecked(int value) + { + AssemblerBuffer::putShortUnchecked(value); + correctDeltas(2); + } + + void putShort(int value) + { + flushIfNoSpaceFor(2); + AssemblerBuffer::putShort(value); + correctDeltas(2); + } + + void putIntUnchecked(int value) + { + AssemblerBuffer::putIntUnchecked(value); + correctDeltas(4); + } + + void putInt(int value) + { + flushIfNoSpaceFor(4); + AssemblerBuffer::putInt(value); + correctDeltas(4); + } + + void putInt64Unchecked(int64_t value) + { + AssemblerBuffer::putInt64Unchecked(value); + correctDeltas(8); + } + + void putIntegral(TwoShorts value) + { + putIntegral(value.high); + putIntegral(value.low); + } + + void putIntegralUnchecked(TwoShorts value) + { + putIntegralUnchecked(value.high); + putIntegralUnchecked(value.low); + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + flushConstantPool(false); + return AssemblerBuffer::executableCopy(globalData, ownerUID, effort); + } + + void putShortWithConstantInt(uint16_t insn, uint32_t constant, bool isReusable = false) + { + putIntegralWithConstantInt(insn, constant, isReusable); + } + + void putIntWithConstantInt(uint32_t insn, uint32_t constant, bool isReusable = false) + { + putIntegralWithConstantInt(insn, constant, isReusable); + } + + // This flushing mechanism can be called after any unconditional jumps. + void flushWithoutBarrier(bool isForced = false) + { + // Flush if constant pool is more than 60% full to avoid overuse of this function. + if (isForced || 5 * static_cast<uint32_t>(m_numConsts) > 3 * maxPoolSize / sizeof(uint32_t)) + flushConstantPool(false); + } + + uint32_t* poolAddress() + { + return m_pool; + } + + int sizeOfConstantPool() + { + return m_numConsts; + } + +private: + void correctDeltas(int insnSize) + { + m_maxDistance -= insnSize; + m_lastConstDelta -= insnSize; + if (m_lastConstDelta < 0) + m_lastConstDelta = 0; + } + + void correctDeltas(int insnSize, int constSize) + { + correctDeltas(insnSize); + + m_maxDistance -= m_lastConstDelta; + m_lastConstDelta = constSize; + } + + template<typename IntegralType> + void putIntegralWithConstantInt(IntegralType insn, uint32_t constant, bool isReusable) + { + if (!m_numConsts) + m_maxDistance = maxPoolSize; + flushIfNoSpaceFor(sizeof(IntegralType), 4); + + m_loadOffsets.append(codeSize()); + if (isReusable) { + for (int i = 0; i < m_numConsts; ++i) { + if (m_mask[i] == ReusableConst && m_pool[i] == constant) { + putIntegral(static_cast<IntegralType>(AssemblerType::patchConstantPoolLoad(insn, i))); + correctDeltas(sizeof(IntegralType)); + return; + } + } + } + + m_pool[m_numConsts] = constant; + m_mask[m_numConsts] = static_cast<char>(isReusable ? ReusableConst : UniqueConst); + + putIntegral(static_cast<IntegralType>(AssemblerType::patchConstantPoolLoad(insn, m_numConsts))); + ++m_numConsts; + + correctDeltas(sizeof(IntegralType), 4); + } + + void flushConstantPool(bool useBarrier = true) + { + if (m_numConsts == 0) + return; + int alignPool = (codeSize() + (useBarrier ? barrierSize : 0)) & (sizeof(uint64_t) - 1); + + if (alignPool) + alignPool = sizeof(uint64_t) - alignPool; + + // Callback to protect the constant pool from execution + if (useBarrier) + putIntegral(AssemblerType::placeConstantPoolBarrier(m_numConsts * sizeof(uint32_t) + alignPool)); + + if (alignPool) { + if (alignPool & 1) + AssemblerBuffer::putByte(AssemblerType::padForAlign8); + if (alignPool & 2) + AssemblerBuffer::putShort(AssemblerType::padForAlign16); + if (alignPool & 4) + AssemblerBuffer::putInt(AssemblerType::padForAlign32); + } + + int constPoolOffset = codeSize(); + append(reinterpret_cast<char*>(m_pool), m_numConsts * sizeof(uint32_t)); + + // Patch each PC relative load + for (LoadOffsets::Iterator iter = m_loadOffsets.begin(); iter != m_loadOffsets.end(); ++iter) { + void* loadAddr = reinterpret_cast<char*>(data()) + *iter; + AssemblerType::patchConstantPoolLoad(loadAddr, reinterpret_cast<char*>(data()) + constPoolOffset); + } + + m_loadOffsets.clear(); + m_numConsts = 0; + } + + void flushIfNoSpaceFor(int nextInsnSize) + { + if (m_numConsts == 0) + return; + int lastConstDelta = m_lastConstDelta > nextInsnSize ? m_lastConstDelta - nextInsnSize : 0; + if ((m_maxDistance < nextInsnSize + lastConstDelta + barrierSize + (int)sizeof(uint32_t))) + flushConstantPool(); + } + + void flushIfNoSpaceFor(int nextInsnSize, int nextConstSize) + { + if (m_numConsts == 0) + return; + if ((m_maxDistance < nextInsnSize + m_lastConstDelta + nextConstSize + barrierSize + (int)sizeof(uint32_t)) || + (m_numConsts * sizeof(uint32_t) + nextConstSize >= maxPoolSize)) + flushConstantPool(); + } + + uint32_t* m_pool; + char* m_mask; + LoadOffsets m_loadOffsets; + + int m_numConsts; + int m_maxDistance; + int m_lastConstDelta; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // AssemblerBufferWithConstantPool_h diff --git a/src/3rdparty/masm/assembler/CodeLocation.h b/src/3rdparty/masm/assembler/CodeLocation.h new file mode 100644 index 0000000000..86d1f2b755 --- /dev/null +++ b/src/3rdparty/masm/assembler/CodeLocation.h @@ -0,0 +1,218 @@ +/* + * Copyright (C) 2009 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef CodeLocation_h +#define CodeLocation_h + +#include "MacroAssemblerCodeRef.h" + +#if ENABLE(ASSEMBLER) + +namespace JSC { + +class CodeLocationInstruction; +class CodeLocationLabel; +class CodeLocationJump; +class CodeLocationCall; +class CodeLocationNearCall; +class CodeLocationDataLabelCompact; +class CodeLocationDataLabel32; +class CodeLocationDataLabelPtr; +class CodeLocationConvertibleLoad; + +// The CodeLocation* types are all pretty much do-nothing wrappers around +// CodePtr (or MacroAssemblerCodePtr, to give it its full name). These +// classes only exist to provide type-safety when linking and patching code. +// +// The one new piece of functionallity introduced by these classes is the +// ability to create (or put another way, to re-discover) another CodeLocation +// at an offset from one you already know. When patching code to optimize it +// we often want to patch a number of instructions that are short, fixed +// offsets apart. To reduce memory overhead we will only retain a pointer to +// one of the instructions, and we will use the *AtOffset methods provided by +// CodeLocationCommon to find the other points in the code to modify. +class CodeLocationCommon : public MacroAssemblerCodePtr { +public: + CodeLocationInstruction instructionAtOffset(int offset); + CodeLocationLabel labelAtOffset(int offset); + CodeLocationJump jumpAtOffset(int offset); + CodeLocationCall callAtOffset(int offset); + CodeLocationNearCall nearCallAtOffset(int offset); + CodeLocationDataLabelPtr dataLabelPtrAtOffset(int offset); + CodeLocationDataLabel32 dataLabel32AtOffset(int offset); + CodeLocationDataLabelCompact dataLabelCompactAtOffset(int offset); + CodeLocationConvertibleLoad convertibleLoadAtOffset(int offset); + +protected: + CodeLocationCommon() + { + } + + CodeLocationCommon(MacroAssemblerCodePtr location) + : MacroAssemblerCodePtr(location) + { + } +}; + +class CodeLocationInstruction : public CodeLocationCommon { +public: + CodeLocationInstruction() {} + explicit CodeLocationInstruction(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationInstruction(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationLabel : public CodeLocationCommon { +public: + CodeLocationLabel() {} + explicit CodeLocationLabel(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationLabel(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationJump : public CodeLocationCommon { +public: + CodeLocationJump() {} + explicit CodeLocationJump(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationJump(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationCall : public CodeLocationCommon { +public: + CodeLocationCall() {} + explicit CodeLocationCall(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationCall(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationNearCall : public CodeLocationCommon { +public: + CodeLocationNearCall() {} + explicit CodeLocationNearCall(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationNearCall(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationDataLabel32 : public CodeLocationCommon { +public: + CodeLocationDataLabel32() {} + explicit CodeLocationDataLabel32(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationDataLabel32(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationDataLabelCompact : public CodeLocationCommon { +public: + CodeLocationDataLabelCompact() { } + explicit CodeLocationDataLabelCompact(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) { } + explicit CodeLocationDataLabelCompact(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) { } +}; + +class CodeLocationDataLabelPtr : public CodeLocationCommon { +public: + CodeLocationDataLabelPtr() {} + explicit CodeLocationDataLabelPtr(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) {} + explicit CodeLocationDataLabelPtr(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) {} +}; + +class CodeLocationConvertibleLoad : public CodeLocationCommon { +public: + CodeLocationConvertibleLoad() { } + explicit CodeLocationConvertibleLoad(MacroAssemblerCodePtr location) + : CodeLocationCommon(location) { } + explicit CodeLocationConvertibleLoad(void* location) + : CodeLocationCommon(MacroAssemblerCodePtr(location)) { } +}; + +inline CodeLocationInstruction CodeLocationCommon::instructionAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationInstruction(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationLabel CodeLocationCommon::labelAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationLabel(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationJump CodeLocationCommon::jumpAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationJump(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationCall CodeLocationCommon::callAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationCall(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationNearCall CodeLocationCommon::nearCallAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationNearCall(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationDataLabelPtr CodeLocationCommon::dataLabelPtrAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationDataLabelPtr(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationDataLabel32 CodeLocationCommon::dataLabel32AtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationDataLabel32(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationDataLabelCompact CodeLocationCommon::dataLabelCompactAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationDataLabelCompact(reinterpret_cast<char*>(dataLocation()) + offset); +} + +inline CodeLocationConvertibleLoad CodeLocationCommon::convertibleLoadAtOffset(int offset) +{ + ASSERT_VALID_CODE_OFFSET(offset); + return CodeLocationConvertibleLoad(reinterpret_cast<char*>(dataLocation()) + offset); +} + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // CodeLocation_h diff --git a/src/3rdparty/masm/assembler/LinkBuffer.cpp b/src/3rdparty/masm/assembler/LinkBuffer.cpp new file mode 100644 index 0000000000..645eba5380 --- /dev/null +++ b/src/3rdparty/masm/assembler/LinkBuffer.cpp @@ -0,0 +1,230 @@ +/* + * Copyright (C) 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "config.h" +#include "LinkBuffer.h" + +#if ENABLE(ASSEMBLER) + +#include "Options.h" + +namespace JSC { + +LinkBuffer::CodeRef LinkBuffer::finalizeCodeWithoutDisassembly() +{ + performFinalization(); + + return CodeRef(m_executableMemory); +} + +LinkBuffer::CodeRef LinkBuffer::finalizeCodeWithDisassembly(const char* format, ...) +{ + ASSERT(Options::showDisassembly() || Options::showDFGDisassembly()); + + CodeRef result = finalizeCodeWithoutDisassembly(); + + dataLogF("Generated JIT code for "); + va_list argList; + va_start(argList, format); + WTF::dataLogFV(format, argList); + va_end(argList); + dataLogF(":\n"); + + dataLogF(" Code at [%p, %p):\n", result.code().executableAddress(), static_cast<char*>(result.code().executableAddress()) + result.size()); + disassemble(result.code(), m_size, " ", WTF::dataFile()); + + return result; +} + +void LinkBuffer::linkCode(void* ownerUID, JITCompilationEffort effort) +{ + ASSERT(!m_code); +#if !ENABLE(BRANCH_COMPACTION) + m_executableMemory = m_assembler->m_assembler.executableCopy(*m_globalData, ownerUID, effort); + if (!m_executableMemory) + return; + m_code = m_executableMemory->start(); + m_size = m_assembler->m_assembler.codeSize(); + ASSERT(m_code); +#else + m_initialSize = m_assembler->m_assembler.codeSize(); + m_executableMemory = m_globalData->executableAllocator.allocate(*m_globalData, m_initialSize, ownerUID, effort); + if (!m_executableMemory) + return; + m_code = (uint8_t*)m_executableMemory->start(); + ASSERT(m_code); + ExecutableAllocator::makeWritable(m_code, m_initialSize); + uint8_t* inData = (uint8_t*)m_assembler->unlinkedCode(); + uint8_t* outData = reinterpret_cast<uint8_t*>(m_code); + int readPtr = 0; + int writePtr = 0; + Vector<LinkRecord, 0, UnsafeVectorOverflow>& jumpsToLink = m_assembler->jumpsToLink(); + unsigned jumpCount = jumpsToLink.size(); + for (unsigned i = 0; i < jumpCount; ++i) { + int offset = readPtr - writePtr; + ASSERT(!(offset & 1)); + + // Copy the instructions from the last jump to the current one. + size_t regionSize = jumpsToLink[i].from() - readPtr; + uint16_t* copySource = reinterpret_cast_ptr<uint16_t*>(inData + readPtr); + uint16_t* copyEnd = reinterpret_cast_ptr<uint16_t*>(inData + readPtr + regionSize); + uint16_t* copyDst = reinterpret_cast_ptr<uint16_t*>(outData + writePtr); + ASSERT(!(regionSize % 2)); + ASSERT(!(readPtr % 2)); + ASSERT(!(writePtr % 2)); + while (copySource != copyEnd) + *copyDst++ = *copySource++; + m_assembler->recordLinkOffsets(readPtr, jumpsToLink[i].from(), offset); + readPtr += regionSize; + writePtr += regionSize; + + // Calculate absolute address of the jump target, in the case of backwards + // branches we need to be precise, forward branches we are pessimistic + const uint8_t* target; + if (jumpsToLink[i].to() >= jumpsToLink[i].from()) + target = outData + jumpsToLink[i].to() - offset; // Compensate for what we have collapsed so far + else + target = outData + jumpsToLink[i].to() - m_assembler->executableOffsetFor(jumpsToLink[i].to()); + + JumpLinkType jumpLinkType = m_assembler->computeJumpType(jumpsToLink[i], outData + writePtr, target); + // Compact branch if we can... + if (m_assembler->canCompact(jumpsToLink[i].type())) { + // Step back in the write stream + int32_t delta = m_assembler->jumpSizeDelta(jumpsToLink[i].type(), jumpLinkType); + if (delta) { + writePtr -= delta; + m_assembler->recordLinkOffsets(jumpsToLink[i].from() - delta, readPtr, readPtr - writePtr); + } + } + jumpsToLink[i].setFrom(writePtr); + } + // Copy everything after the last jump + memcpy(outData + writePtr, inData + readPtr, m_initialSize - readPtr); + m_assembler->recordLinkOffsets(readPtr, m_initialSize, readPtr - writePtr); + + for (unsigned i = 0; i < jumpCount; ++i) { + uint8_t* location = outData + jumpsToLink[i].from(); + uint8_t* target = outData + jumpsToLink[i].to() - m_assembler->executableOffsetFor(jumpsToLink[i].to()); + m_assembler->link(jumpsToLink[i], location, target); + } + + jumpsToLink.clear(); + m_size = writePtr + m_initialSize - readPtr; + m_executableMemory->shrink(m_size); + +#if DUMP_LINK_STATISTICS + dumpLinkStatistics(m_code, m_initialSize, m_size); +#endif +#if DUMP_CODE + dumpCode(m_code, m_size); +#endif +#endif +} + +void LinkBuffer::performFinalization() +{ +#ifndef NDEBUG + ASSERT(!m_completed); + ASSERT(isValid()); + m_completed = true; +#endif + +#if ENABLE(BRANCH_COMPACTION) + ExecutableAllocator::makeExecutable(code(), m_initialSize); +#else + ExecutableAllocator::makeExecutable(code(), m_size); +#endif + MacroAssembler::cacheFlush(code(), m_size); +} + +#if DUMP_LINK_STATISTICS +void LinkBuffer::dumpLinkStatistics(void* code, size_t initializeSize, size_t finalSize) +{ + static unsigned linkCount = 0; + static unsigned totalInitialSize = 0; + static unsigned totalFinalSize = 0; + linkCount++; + totalInitialSize += initialSize; + totalFinalSize += finalSize; + dataLogF("link %p: orig %u, compact %u (delta %u, %.2f%%)\n", + code, static_cast<unsigned>(initialSize), static_cast<unsigned>(finalSize), + static_cast<unsigned>(initialSize - finalSize), + 100.0 * (initialSize - finalSize) / initialSize); + dataLogF("\ttotal %u: orig %u, compact %u (delta %u, %.2f%%)\n", + linkCount, totalInitialSize, totalFinalSize, totalInitialSize - totalFinalSize, + 100.0 * (totalInitialSize - totalFinalSize) / totalInitialSize); +} +#endif + +#if DUMP_CODE +void LinkBuffer::dumpCode(void* code, size_t size) +{ +#if CPU(ARM_THUMB2) + // Dump the generated code in an asm file format that can be assembled and then disassembled + // for debugging purposes. For example, save this output as jit.s: + // gcc -arch armv7 -c jit.s + // otool -tv jit.o + static unsigned codeCount = 0; + unsigned short* tcode = static_cast<unsigned short*>(code); + size_t tsize = size / sizeof(short); + char nameBuf[128]; + snprintf(nameBuf, sizeof(nameBuf), "_jsc_jit%u", codeCount++); + dataLogF("\t.syntax unified\n" + "\t.section\t__TEXT,__text,regular,pure_instructions\n" + "\t.globl\t%s\n" + "\t.align 2\n" + "\t.code 16\n" + "\t.thumb_func\t%s\n" + "# %p\n" + "%s:\n", nameBuf, nameBuf, code, nameBuf); + + for (unsigned i = 0; i < tsize; i++) + dataLogF("\t.short\t0x%x\n", tcode[i]); +#elif CPU(ARM_TRADITIONAL) + // gcc -c jit.s + // objdump -D jit.o + static unsigned codeCount = 0; + unsigned int* tcode = static_cast<unsigned int*>(code); + size_t tsize = size / sizeof(unsigned int); + char nameBuf[128]; + snprintf(nameBuf, sizeof(nameBuf), "_jsc_jit%u", codeCount++); + dataLogF("\t.globl\t%s\n" + "\t.align 4\n" + "\t.code 32\n" + "\t.text\n" + "# %p\n" + "%s:\n", nameBuf, code, nameBuf); + + for (unsigned i = 0; i < tsize; i++) + dataLogF("\t.long\t0x%x\n", tcode[i]); +#endif +} +#endif + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + + diff --git a/src/3rdparty/masm/assembler/LinkBuffer.h b/src/3rdparty/masm/assembler/LinkBuffer.h new file mode 100644 index 0000000000..e1882433c1 --- /dev/null +++ b/src/3rdparty/masm/assembler/LinkBuffer.h @@ -0,0 +1,297 @@ +/* + * Copyright (C) 2009, 2010, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef LinkBuffer_h +#define LinkBuffer_h + +#if ENABLE(ASSEMBLER) + +#define DUMP_LINK_STATISTICS 0 +#define DUMP_CODE 0 + +#define GLOBAL_THUNK_ID reinterpret_cast<void*>(static_cast<intptr_t>(-1)) +#define REGEXP_CODE_ID reinterpret_cast<void*>(static_cast<intptr_t>(-2)) + +#include "JITCompilationEffort.h" +#include "MacroAssembler.h" +#include <wtf/DataLog.h> +#include <wtf/Noncopyable.h> + +namespace JSC { + +class JSGlobalData; + +// LinkBuffer: +// +// This class assists in linking code generated by the macro assembler, once code generation +// has been completed, and the code has been copied to is final location in memory. At this +// time pointers to labels within the code may be resolved, and relative offsets to external +// addresses may be fixed. +// +// Specifically: +// * Jump objects may be linked to external targets, +// * The address of Jump objects may taken, such that it can later be relinked. +// * The return address of a Call may be acquired. +// * The address of a Label pointing into the code may be resolved. +// * The value referenced by a DataLabel may be set. +// +class LinkBuffer { + WTF_MAKE_NONCOPYABLE(LinkBuffer); + typedef MacroAssemblerCodeRef CodeRef; + typedef MacroAssemblerCodePtr CodePtr; + typedef MacroAssembler::Label Label; + typedef MacroAssembler::Jump Jump; + typedef MacroAssembler::PatchableJump PatchableJump; + typedef MacroAssembler::JumpList JumpList; + typedef MacroAssembler::Call Call; + typedef MacroAssembler::DataLabelCompact DataLabelCompact; + typedef MacroAssembler::DataLabel32 DataLabel32; + typedef MacroAssembler::DataLabelPtr DataLabelPtr; + typedef MacroAssembler::ConvertibleLoadLabel ConvertibleLoadLabel; +#if ENABLE(BRANCH_COMPACTION) + typedef MacroAssembler::LinkRecord LinkRecord; + typedef MacroAssembler::JumpLinkType JumpLinkType; +#endif + +public: + LinkBuffer(JSGlobalData& globalData, MacroAssembler* masm, void* ownerUID, JITCompilationEffort effort = JITCompilationMustSucceed) + : m_size(0) +#if ENABLE(BRANCH_COMPACTION) + , m_initialSize(0) +#endif + , m_code(0) + , m_assembler(masm) + , m_globalData(&globalData) +#ifndef NDEBUG + , m_completed(false) + , m_effort(effort) +#endif + { + linkCode(ownerUID, effort); + } + + ~LinkBuffer() + { + ASSERT(m_completed || (!m_executableMemory && m_effort == JITCompilationCanFail)); + } + + bool didFailToAllocate() const + { + return !m_executableMemory; + } + + bool isValid() const + { + return !didFailToAllocate(); + } + + // These methods are used to link or set values at code generation time. + + void link(Call call, FunctionPtr function) + { + ASSERT(call.isFlagSet(Call::Linkable)); + call.m_label = applyOffset(call.m_label); + MacroAssembler::linkCall(code(), call, function); + } + + void link(Jump jump, CodeLocationLabel label) + { + jump.m_label = applyOffset(jump.m_label); + MacroAssembler::linkJump(code(), jump, label); + } + + void link(JumpList list, CodeLocationLabel label) + { + for (unsigned i = 0; i < list.m_jumps.size(); ++i) + link(list.m_jumps[i], label); + } + + void patch(DataLabelPtr label, void* value) + { + AssemblerLabel target = applyOffset(label.m_label); + MacroAssembler::linkPointer(code(), target, value); + } + + void patch(DataLabelPtr label, CodeLocationLabel value) + { + AssemblerLabel target = applyOffset(label.m_label); + MacroAssembler::linkPointer(code(), target, value.executableAddress()); + } + + // These methods are used to obtain handles to allow the code to be relinked / repatched later. + + CodeLocationCall locationOf(Call call) + { + ASSERT(call.isFlagSet(Call::Linkable)); + ASSERT(!call.isFlagSet(Call::Near)); + return CodeLocationCall(MacroAssembler::getLinkerAddress(code(), applyOffset(call.m_label))); + } + + CodeLocationNearCall locationOfNearCall(Call call) + { + ASSERT(call.isFlagSet(Call::Linkable)); + ASSERT(call.isFlagSet(Call::Near)); + return CodeLocationNearCall(MacroAssembler::getLinkerAddress(code(), applyOffset(call.m_label))); + } + + CodeLocationLabel locationOf(PatchableJump jump) + { + return CodeLocationLabel(MacroAssembler::getLinkerAddress(code(), applyOffset(jump.m_jump.m_label))); + } + + CodeLocationLabel locationOf(Label label) + { + return CodeLocationLabel(MacroAssembler::getLinkerAddress(code(), applyOffset(label.m_label))); + } + + CodeLocationDataLabelPtr locationOf(DataLabelPtr label) + { + return CodeLocationDataLabelPtr(MacroAssembler::getLinkerAddress(code(), applyOffset(label.m_label))); + } + + CodeLocationDataLabel32 locationOf(DataLabel32 label) + { + return CodeLocationDataLabel32(MacroAssembler::getLinkerAddress(code(), applyOffset(label.m_label))); + } + + CodeLocationDataLabelCompact locationOf(DataLabelCompact label) + { + return CodeLocationDataLabelCompact(MacroAssembler::getLinkerAddress(code(), applyOffset(label.m_label))); + } + + CodeLocationConvertibleLoad locationOf(ConvertibleLoadLabel label) + { + return CodeLocationConvertibleLoad(MacroAssembler::getLinkerAddress(code(), applyOffset(label.m_label))); + } + + // This method obtains the return address of the call, given as an offset from + // the start of the code. + unsigned returnAddressOffset(Call call) + { + call.m_label = applyOffset(call.m_label); + return MacroAssembler::getLinkerCallReturnOffset(call); + } + + uint32_t offsetOf(Label label) + { + return applyOffset(label.m_label).m_offset; + } + + // Upon completion of all patching 'FINALIZE_CODE()' should be called once to + // complete generation of the code. Alternatively, call + // finalizeCodeWithoutDisassembly() directly if you have your own way of + // displaying disassembly. + + CodeRef finalizeCodeWithoutDisassembly(); + CodeRef finalizeCodeWithDisassembly(const char* format, ...) WTF_ATTRIBUTE_PRINTF(2, 3); + + CodePtr trampolineAt(Label label) + { + return CodePtr(MacroAssembler::AssemblerType_T::getRelocatedAddress(code(), applyOffset(label.m_label))); + } + + void* debugAddress() + { + return m_code; + } + + size_t debugSize() + { + return m_size; + } + +private: + template <typename T> T applyOffset(T src) + { +#if ENABLE(BRANCH_COMPACTION) + src.m_offset -= m_assembler->executableOffsetFor(src.m_offset); +#endif + return src; + } + + // Keep this private! - the underlying code should only be obtained externally via finalizeCode(). + void* code() + { + return m_code; + } + + void linkCode(void* ownerUID, JITCompilationEffort); + + void performFinalization(); + +#if DUMP_LINK_STATISTICS + static void dumpLinkStatistics(void* code, size_t initialSize, size_t finalSize); +#endif + +#if DUMP_CODE + static void dumpCode(void* code, size_t); +#endif + + RefPtr<ExecutableMemoryHandle> m_executableMemory; + size_t m_size; +#if ENABLE(BRANCH_COMPACTION) + size_t m_initialSize; +#endif + void* m_code; + MacroAssembler* m_assembler; + JSGlobalData* m_globalData; +#ifndef NDEBUG + bool m_completed; + JITCompilationEffort m_effort; +#endif +}; + +#define FINALIZE_CODE_IF(condition, linkBufferReference, dataLogFArgumentsForHeading) \ + (UNLIKELY((condition)) \ + ? ((linkBufferReference).finalizeCodeWithDisassembly dataLogFArgumentsForHeading) \ + : (linkBufferReference).finalizeCodeWithoutDisassembly()) + +// Use this to finalize code, like so: +// +// CodeRef code = FINALIZE_CODE(linkBuffer, ("my super thingy number %d", number)); +// +// Which, in disassembly mode, will print: +// +// Generated JIT code for my super thingy number 42: +// Code at [0x123456, 0x234567]: +// 0x123456: mov $0, 0 +// 0x12345a: ret +// +// ... and so on. +// +// Note that the dataLogFArgumentsForHeading are only evaluated when showDisassembly +// is true, so you can hide expensive disassembly-only computations inside there. + +#define FINALIZE_CODE(linkBufferReference, dataLogFArgumentsForHeading) \ + FINALIZE_CODE_IF(Options::showDisassembly(), linkBufferReference, dataLogFArgumentsForHeading) + +#define FINALIZE_DFG_CODE(linkBufferReference, dataLogFArgumentsForHeading) \ + FINALIZE_CODE_IF((Options::showDisassembly() || Options::showDFGDisassembly()), linkBufferReference, dataLogFArgumentsForHeading) + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // LinkBuffer_h diff --git a/src/3rdparty/masm/assembler/MIPSAssembler.h b/src/3rdparty/masm/assembler/MIPSAssembler.h new file mode 100644 index 0000000000..7f553bb9a1 --- /dev/null +++ b/src/3rdparty/masm/assembler/MIPSAssembler.h @@ -0,0 +1,1107 @@ +/* + * Copyright (C) 2009 Apple Inc. All rights reserved. + * Copyright (C) 2009 University of Szeged + * All rights reserved. + * Copyright (C) 2010 MIPS Technologies, Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY MIPS TECHNOLOGIES, INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL MIPS TECHNOLOGIES, INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MIPSAssembler_h +#define MIPSAssembler_h + +#if ENABLE(ASSEMBLER) && CPU(MIPS) + +#include "AssemblerBuffer.h" +#include "JITCompilationEffort.h" +#include <wtf/Assertions.h> +#include <wtf/SegmentedVector.h> + +namespace JSC { + +typedef uint32_t MIPSWord; + +namespace MIPSRegisters { +typedef enum { + r0 = 0, + r1, + r2, + r3, + r4, + r5, + r6, + r7, + r8, + r9, + r10, + r11, + r12, + r13, + r14, + r15, + r16, + r17, + r18, + r19, + r20, + r21, + r22, + r23, + r24, + r25, + r26, + r27, + r28, + r29, + r30, + r31, + zero = r0, + at = r1, + v0 = r2, + v1 = r3, + a0 = r4, + a1 = r5, + a2 = r6, + a3 = r7, + t0 = r8, + t1 = r9, + t2 = r10, + t3 = r11, + t4 = r12, + t5 = r13, + t6 = r14, + t7 = r15, + s0 = r16, + s1 = r17, + s2 = r18, + s3 = r19, + s4 = r20, + s5 = r21, + s6 = r22, + s7 = r23, + t8 = r24, + t9 = r25, + k0 = r26, + k1 = r27, + gp = r28, + sp = r29, + fp = r30, + ra = r31 +} RegisterID; + +typedef enum { + f0, + f1, + f2, + f3, + f4, + f5, + f6, + f7, + f8, + f9, + f10, + f11, + f12, + f13, + f14, + f15, + f16, + f17, + f18, + f19, + f20, + f21, + f22, + f23, + f24, + f25, + f26, + f27, + f28, + f29, + f30, + f31 +} FPRegisterID; + +} // namespace MIPSRegisters + +class MIPSAssembler { +public: + typedef MIPSRegisters::RegisterID RegisterID; + typedef MIPSRegisters::FPRegisterID FPRegisterID; + typedef SegmentedVector<AssemblerLabel, 64> Jumps; + + MIPSAssembler() + : m_indexOfLastWatchpoint(INT_MIN) + , m_indexOfTailOfLastWatchpoint(INT_MIN) + { + } + + // MIPS instruction opcode field position + enum { + OP_SH_RD = 11, + OP_SH_RT = 16, + OP_SH_RS = 21, + OP_SH_SHAMT = 6, + OP_SH_CODE = 16, + OP_SH_FD = 6, + OP_SH_FS = 11, + OP_SH_FT = 16 + }; + + void emitInst(MIPSWord op) + { + void* oldBase = m_buffer.data(); + + m_buffer.putInt(op); + + void* newBase = m_buffer.data(); + if (oldBase != newBase) + relocateJumps(oldBase, newBase); + } + + void nop() + { + emitInst(0x00000000); + } + + /* Need to insert one load data delay nop for mips1. */ + void loadDelayNop() + { +#if WTF_MIPS_ISA(1) + nop(); +#endif + } + + /* Need to insert one coprocessor access delay nop for mips1. */ + void copDelayNop() + { +#if WTF_MIPS_ISA(1) + nop(); +#endif + } + + void move(RegisterID rd, RegisterID rs) + { + /* addu */ + emitInst(0x00000021 | (rd << OP_SH_RD) | (rs << OP_SH_RS)); + } + + /* Set an immediate value to a register. This may generate 1 or 2 + instructions. */ + void li(RegisterID dest, int imm) + { + if (imm >= -32768 && imm <= 32767) + addiu(dest, MIPSRegisters::zero, imm); + else if (imm >= 0 && imm < 65536) + ori(dest, MIPSRegisters::zero, imm); + else { + lui(dest, imm >> 16); + if (imm & 0xffff) + ori(dest, dest, imm); + } + } + + void lui(RegisterID rt, int imm) + { + emitInst(0x3c000000 | (rt << OP_SH_RT) | (imm & 0xffff)); + } + + void addiu(RegisterID rt, RegisterID rs, int imm) + { + emitInst(0x24000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void addu(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000021 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void subu(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000023 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void mult(RegisterID rs, RegisterID rt) + { + emitInst(0x00000018 | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void div(RegisterID rs, RegisterID rt) + { + emitInst(0x0000001a | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void mfhi(RegisterID rd) + { + emitInst(0x00000010 | (rd << OP_SH_RD)); + } + + void mflo(RegisterID rd) + { + emitInst(0x00000012 | (rd << OP_SH_RD)); + } + + void mul(RegisterID rd, RegisterID rs, RegisterID rt) + { +#if WTF_MIPS_ISA_AT_LEAST(32) + emitInst(0x70000002 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); +#else + mult(rs, rt); + mflo(rd); +#endif + } + + void andInsn(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000024 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void andi(RegisterID rt, RegisterID rs, int imm) + { + emitInst(0x30000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void nor(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000027 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void orInsn(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000025 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void ori(RegisterID rt, RegisterID rs, int imm) + { + emitInst(0x34000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void xorInsn(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x00000026 | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void xori(RegisterID rt, RegisterID rs, int imm) + { + emitInst(0x38000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void slt(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x0000002a | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void sltu(RegisterID rd, RegisterID rs, RegisterID rt) + { + emitInst(0x0000002b | (rd << OP_SH_RD) | (rs << OP_SH_RS) | (rt << OP_SH_RT)); + } + + void sltiu(RegisterID rt, RegisterID rs, int imm) + { + emitInst(0x2c000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void sll(RegisterID rd, RegisterID rt, int shamt) + { + emitInst(0x00000000 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | ((shamt & 0x1f) << OP_SH_SHAMT)); + } + + void sllv(RegisterID rd, RegisterID rt, RegisterID rs) + { + emitInst(0x00000004 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | (rs << OP_SH_RS)); + } + + void sra(RegisterID rd, RegisterID rt, int shamt) + { + emitInst(0x00000003 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | ((shamt & 0x1f) << OP_SH_SHAMT)); + } + + void srav(RegisterID rd, RegisterID rt, RegisterID rs) + { + emitInst(0x00000007 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | (rs << OP_SH_RS)); + } + + void srl(RegisterID rd, RegisterID rt, int shamt) + { + emitInst(0x00000002 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | ((shamt & 0x1f) << OP_SH_SHAMT)); + } + + void srlv(RegisterID rd, RegisterID rt, RegisterID rs) + { + emitInst(0x00000006 | (rd << OP_SH_RD) | (rt << OP_SH_RT) | (rs << OP_SH_RS)); + } + + void lb(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x80000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lbu(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x90000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lw(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x8c000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lwl(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x88000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lwr(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x98000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lh(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x84000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void lhu(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0x94000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + loadDelayNop(); + } + + void sb(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0xa0000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void sh(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0xa4000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void sw(RegisterID rt, RegisterID rs, int offset) + { + emitInst(0xac000000 | (rt << OP_SH_RT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void jr(RegisterID rs) + { + emitInst(0x00000008 | (rs << OP_SH_RS)); + } + + void jalr(RegisterID rs) + { + emitInst(0x0000f809 | (rs << OP_SH_RS)); + } + + void jal() + { + emitInst(0x0c000000); + } + + void bkpt() + { + int value = 512; /* BRK_BUG */ + emitInst(0x0000000d | ((value & 0x3ff) << OP_SH_CODE)); + } + + void bgez(RegisterID rs, int imm) + { + emitInst(0x04010000 | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void bltz(RegisterID rs, int imm) + { + emitInst(0x04000000 | (rs << OP_SH_RS) | (imm & 0xffff)); + } + + void beq(RegisterID rs, RegisterID rt, int imm) + { + emitInst(0x10000000 | (rs << OP_SH_RS) | (rt << OP_SH_RT) | (imm & 0xffff)); + } + + void bne(RegisterID rs, RegisterID rt, int imm) + { + emitInst(0x14000000 | (rs << OP_SH_RS) | (rt << OP_SH_RT) | (imm & 0xffff)); + } + + void bc1t() + { + emitInst(0x45010000); + } + + void bc1f() + { + emitInst(0x45000000); + } + + void appendJump() + { + m_jumps.append(m_buffer.label()); + } + + void addd(FPRegisterID fd, FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200000 | (fd << OP_SH_FD) | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + } + + void subd(FPRegisterID fd, FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200001 | (fd << OP_SH_FD) | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + } + + void muld(FPRegisterID fd, FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200002 | (fd << OP_SH_FD) | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + } + + void divd(FPRegisterID fd, FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200003 | (fd << OP_SH_FD) | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + } + + void lwc1(FPRegisterID ft, RegisterID rs, int offset) + { + emitInst(0xc4000000 | (ft << OP_SH_FT) | (rs << OP_SH_RS) | (offset & 0xffff)); + copDelayNop(); + } + + void ldc1(FPRegisterID ft, RegisterID rs, int offset) + { + emitInst(0xd4000000 | (ft << OP_SH_FT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void swc1(FPRegisterID ft, RegisterID rs, int offset) + { + emitInst(0xe4000000 | (ft << OP_SH_FT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void sdc1(FPRegisterID ft, RegisterID rs, int offset) + { + emitInst(0xf4000000 | (ft << OP_SH_FT) | (rs << OP_SH_RS) | (offset & 0xffff)); + } + + void mtc1(RegisterID rt, FPRegisterID fs) + { + emitInst(0x44800000 | (fs << OP_SH_FS) | (rt << OP_SH_RT)); + copDelayNop(); + } + + void mthc1(RegisterID rt, FPRegisterID fs) + { + emitInst(0x44e00000 | (fs << OP_SH_FS) | (rt << OP_SH_RT)); + copDelayNop(); + } + + void mfc1(RegisterID rt, FPRegisterID fs) + { + emitInst(0x44000000 | (fs << OP_SH_FS) | (rt << OP_SH_RT)); + copDelayNop(); + } + + void sqrtd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46200004 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void movd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46200006 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void negd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46200007 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void truncwd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x4620000d | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void cvtdw(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46800021 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void cvtds(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46000021 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void cvtwd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46200024 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void cvtsd(FPRegisterID fd, FPRegisterID fs) + { + emitInst(0x46200020 | (fd << OP_SH_FD) | (fs << OP_SH_FS)); + } + + void ceqd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200032 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cngtd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x4620003f | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cnged(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x4620003d | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cltd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x4620003c | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cled(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x4620003e | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cueqd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200033 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void coled(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200036 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void coltd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200034 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void culed(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200037 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + void cultd(FPRegisterID fs, FPRegisterID ft) + { + emitInst(0x46200035 | (fs << OP_SH_FS) | (ft << OP_SH_FT)); + copDelayNop(); + } + + // General helpers + + AssemblerLabel labelIgnoringWatchpoints() + { + return m_buffer.label(); + } + + AssemblerLabel labelForWatchpoint() + { + AssemblerLabel result = m_buffer.label(); + if (static_cast<int>(result.m_offset) != m_indexOfLastWatchpoint) + result = label(); + m_indexOfLastWatchpoint = result.m_offset; + m_indexOfTailOfLastWatchpoint = result.m_offset + maxJumpReplacementSize(); + return result; + } + + AssemblerLabel label() + { + AssemblerLabel result = m_buffer.label(); + while (UNLIKELY(static_cast<int>(result.m_offset) < m_indexOfTailOfLastWatchpoint)) { + nop(); + result = m_buffer.label(); + } + return result; + } + + AssemblerLabel align(int alignment) + { + while (!m_buffer.isAligned(alignment)) + bkpt(); + + return label(); + } + + static void* getRelocatedAddress(void* code, AssemblerLabel label) + { + return reinterpret_cast<void*>(reinterpret_cast<char*>(code) + label.m_offset); + } + + static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b) + { + return b.m_offset - a.m_offset; + } + + // Assembler admin methods: + + size_t codeSize() const + { + return m_buffer.codeSize(); + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + RefPtr<ExecutableMemoryHandle> result = m_buffer.executableCopy(globalData, ownerUID, effort); + if (!result) + return 0; + + relocateJumps(m_buffer.data(), result->start()); + return result.release(); + } + + unsigned debugOffset() { return m_buffer.debugOffset(); } + + // Assembly helpers for moving data between fp and registers. + void vmov(RegisterID rd1, RegisterID rd2, FPRegisterID rn) + { +#if WTF_MIPS_ISA_REV(2) && WTF_MIPS_FP64 + mfc1(rd1, rn); + mfhc1(rd2, rn); +#else + mfc1(rd1, rn); + mfc1(rd2, FPRegisterID(rn + 1)); +#endif + } + + void vmov(FPRegisterID rd, RegisterID rn1, RegisterID rn2) + { +#if WTF_MIPS_ISA_REV(2) && WTF_MIPS_FP64 + mtc1(rn1, rd); + mthc1(rn2, rd); +#else + mtc1(rn1, rd); + mtc1(rn2, FPRegisterID(rd + 1)); +#endif + } + + static unsigned getCallReturnOffset(AssemblerLabel call) + { + // The return address is after a call and a delay slot instruction + return call.m_offset; + } + + // Linking & patching: + // + // 'link' and 'patch' methods are for use on unprotected code - such as the code + // within the AssemblerBuffer, and code being patched by the patch buffer. Once + // code has been finalized it is (platform support permitting) within a non- + // writable region of memory; to modify the code in an execute-only execuable + // pool the 'repatch' and 'relink' methods should be used. + + static size_t linkDirectJump(void* code, void* to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(code)); + size_t ops = 0; + int32_t slotAddr = reinterpret_cast<int>(insn) + 4; + int32_t toAddr = reinterpret_cast<int>(to); + + if ((slotAddr & 0xf0000000) != (toAddr & 0xf0000000)) { + // lui + *insn = 0x3c000000 | (MIPSRegisters::t9 << OP_SH_RT) | ((toAddr >> 16) & 0xffff); + ++insn; + // ori + *insn = 0x34000000 | (MIPSRegisters::t9 << OP_SH_RT) | (MIPSRegisters::t9 << OP_SH_RS) | (toAddr & 0xffff); + ++insn; + // jr + *insn = 0x00000008 | (MIPSRegisters::t9 << OP_SH_RS); + ++insn; + ops = 4 * sizeof(MIPSWord); + } else { + // j + *insn = 0x08000000 | ((toAddr & 0x0fffffff) >> 2); + ++insn; + ops = 2 * sizeof(MIPSWord); + } + // nop + *insn = 0x00000000; + return ops; + } + + void linkJump(AssemblerLabel from, AssemblerLabel to) + { + ASSERT(to.isSet()); + ASSERT(from.isSet()); + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(m_buffer.data()) + from.m_offset); + MIPSWord* toPos = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(m_buffer.data()) + to.m_offset); + + ASSERT(!(*(insn - 1)) && !(*(insn - 2)) && !(*(insn - 3)) && !(*(insn - 5))); + insn = insn - 6; + linkWithOffset(insn, toPos); + } + + static void linkJump(void* code, AssemblerLabel from, void* to) + { + ASSERT(from.isSet()); + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(code) + from.m_offset); + + ASSERT(!(*(insn - 1)) && !(*(insn - 2)) && !(*(insn - 3)) && !(*(insn - 5))); + insn = insn - 6; + linkWithOffset(insn, to); + } + + static void linkCall(void* code, AssemblerLabel from, void* to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(code) + from.m_offset); + linkCallInternal(insn, to); + } + + static void linkPointer(void* code, AssemblerLabel from, void* to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(code) + from.m_offset); + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + *insn = (*insn & 0xffff0000) | ((reinterpret_cast<intptr_t>(to) >> 16) & 0xffff); + insn++; + ASSERT((*insn & 0xfc000000) == 0x34000000); // ori + *insn = (*insn & 0xffff0000) | (reinterpret_cast<intptr_t>(to) & 0xffff); + } + + static void relinkJump(void* from, void* to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(from); + + ASSERT(!(*(insn - 1)) && !(*(insn - 5))); + insn = insn - 6; + int flushSize = linkWithOffset(insn, to); + + cacheFlush(insn, flushSize); + } + + static void relinkCall(void* from, void* to) + { + void* start; + int size = linkCallInternal(from, to); + if (size == sizeof(MIPSWord)) + start = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(from) - 2 * sizeof(MIPSWord)); + else + start = reinterpret_cast<void*>(reinterpret_cast<intptr_t>(from) - 4 * sizeof(MIPSWord)); + + cacheFlush(start, size); + } + + static void repatchInt32(void* from, int32_t to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(from); + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + *insn = (*insn & 0xffff0000) | ((to >> 16) & 0xffff); + insn++; + ASSERT((*insn & 0xfc000000) == 0x34000000); // ori + *insn = (*insn & 0xffff0000) | (to & 0xffff); + insn--; + cacheFlush(insn, 2 * sizeof(MIPSWord)); + } + + static int32_t readInt32(void* from) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(from); + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + int32_t result = (*insn & 0x0000ffff) << 16; + insn++; + ASSERT((*insn & 0xfc000000) == 0x34000000); // ori + result |= *insn & 0x0000ffff; + return result; + } + + static void repatchCompact(void* where, int32_t value) + { + repatchInt32(where, value); + } + + static void repatchPointer(void* from, void* to) + { + repatchInt32(from, reinterpret_cast<int32_t>(to)); + } + + static void* readPointer(void* from) + { + return reinterpret_cast<void*>(readInt32(from)); + } + + static void* readCallTarget(void* from) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(from); + insn -= 4; + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + int32_t result = (*insn & 0x0000ffff) << 16; + insn++; + ASSERT((*insn & 0xfc000000) == 0x34000000); // ori + result |= *insn & 0x0000ffff; + return reinterpret_cast<void*>(result); + } + + static void cacheFlush(void* code, size_t size) + { +#if GCC_VERSION_AT_LEAST(4, 3, 0) +#if WTF_MIPS_ISA_REV(2) && !GCC_VERSION_AT_LEAST(4, 4, 3) + int lineSize; + asm("rdhwr %0, $1" : "=r" (lineSize)); + // + // Modify "start" and "end" to avoid GCC 4.3.0-4.4.2 bug in + // mips_expand_synci_loop that may execute synci one more time. + // "start" points to the fisrt byte of the cache line. + // "end" points to the last byte of the line before the last cache line. + // Because size is always a multiple of 4, this is safe to set + // "end" to the last byte. + // + intptr_t start = reinterpret_cast<intptr_t>(code) & (-lineSize); + intptr_t end = ((reinterpret_cast<intptr_t>(code) + size - 1) & (-lineSize)) - 1; + __builtin___clear_cache(reinterpret_cast<char*>(start), reinterpret_cast<char*>(end)); +#else + intptr_t end = reinterpret_cast<intptr_t>(code) + size; + __builtin___clear_cache(reinterpret_cast<char*>(code), reinterpret_cast<char*>(end)); +#endif +#else + _flush_cache(reinterpret_cast<char*>(code), size, BCACHE); +#endif + } + + static ptrdiff_t maxJumpReplacementSize() + { + return sizeof(MIPSWord) * 4; + } + + static void revertJumpToMove(void* instructionStart, RegisterID rt, int imm) + { + MIPSWord* insn = static_cast<MIPSWord*>(instructionStart); + size_t codeSize = 2 * sizeof(MIPSWord); + + // lui + *insn = 0x3c000000 | (rt << OP_SH_RT) | ((imm >> 16) & 0xffff); + ++insn; + // ori + *insn = 0x34000000 | (rt << OP_SH_RS) | (rt << OP_SH_RT) | (imm & 0xffff); + ++insn; + // if jr $t9 + if (*insn == 0x03200008) { + *insn = 0x00000000; + codeSize += sizeof(MIPSWord); + } + cacheFlush(insn, codeSize); + } + + static void replaceWithJump(void* instructionStart, void* to) + { + ASSERT(!(bitwise_cast<uintptr_t>(instructionStart) & 3)); + ASSERT(!(bitwise_cast<uintptr_t>(to) & 3)); + size_t ops = linkDirectJump(instructionStart, to); + cacheFlush(instructionStart, ops); + } + + static void replaceWithLoad(void* instructionStart) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(instructionStart); + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + insn++; + ASSERT((*insn & 0xfc0007ff) == 0x00000021); // addu + insn++; + *insn = 0x8c000000 | ((*insn) & 0x3ffffff); // lw + cacheFlush(insn, 4); + } + + static void replaceWithAddressComputation(void* instructionStart) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(instructionStart); + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + insn++; + ASSERT((*insn & 0xfc0007ff) == 0x00000021); // addu + insn++; + *insn = 0x24000000 | ((*insn) & 0x3ffffff); // addiu + cacheFlush(insn, 4); + } + +private: + /* Update each jump in the buffer of newBase. */ + void relocateJumps(void* oldBase, void* newBase) + { + // Check each jump + for (Jumps::Iterator iter = m_jumps.begin(); iter != m_jumps.end(); ++iter) { + int pos = iter->m_offset; + MIPSWord* insn = reinterpret_cast<MIPSWord*>(reinterpret_cast<intptr_t>(newBase) + pos); + insn = insn + 2; + // Need to make sure we have 5 valid instructions after pos + if ((unsigned)pos >= m_buffer.codeSize() - 5 * sizeof(MIPSWord)) + continue; + + if ((*insn & 0xfc000000) == 0x08000000) { // j + int offset = *insn & 0x03ffffff; + int oldInsnAddress = (int)insn - (int)newBase + (int)oldBase; + int topFourBits = (oldInsnAddress + 4) >> 28; + int oldTargetAddress = (topFourBits << 28) | (offset << 2); + int newTargetAddress = oldTargetAddress - (int)oldBase + (int)newBase; + int newInsnAddress = (int)insn; + if (((newInsnAddress + 4) >> 28) == (newTargetAddress >> 28)) + *insn = 0x08000000 | ((newTargetAddress >> 2) & 0x3ffffff); + else { + /* lui */ + *insn = 0x3c000000 | (MIPSRegisters::t9 << OP_SH_RT) | ((newTargetAddress >> 16) & 0xffff); + /* ori */ + *(insn + 1) = 0x34000000 | (MIPSRegisters::t9 << OP_SH_RT) | (MIPSRegisters::t9 << OP_SH_RS) | (newTargetAddress & 0xffff); + /* jr */ + *(insn + 2) = 0x00000008 | (MIPSRegisters::t9 << OP_SH_RS); + } + } else if ((*insn & 0xffe00000) == 0x3c000000) { // lui + int high = (*insn & 0xffff) << 16; + int low = *(insn + 1) & 0xffff; + int oldTargetAddress = high | low; + int newTargetAddress = oldTargetAddress - (int)oldBase + (int)newBase; + /* lui */ + *insn = 0x3c000000 | (MIPSRegisters::t9 << OP_SH_RT) | ((newTargetAddress >> 16) & 0xffff); + /* ori */ + *(insn + 1) = 0x34000000 | (MIPSRegisters::t9 << OP_SH_RT) | (MIPSRegisters::t9 << OP_SH_RS) | (newTargetAddress & 0xffff); + } + } + } + + static int linkWithOffset(MIPSWord* insn, void* to) + { + ASSERT((*insn & 0xfc000000) == 0x10000000 // beq + || (*insn & 0xfc000000) == 0x14000000 // bne + || (*insn & 0xffff0000) == 0x45010000 // bc1t + || (*insn & 0xffff0000) == 0x45000000); // bc1f + intptr_t diff = (reinterpret_cast<intptr_t>(to) - reinterpret_cast<intptr_t>(insn) - 4) >> 2; + + if (diff < -32768 || diff > 32767 || *(insn + 2) != 0x10000003) { + /* + Convert the sequence: + beq $2, $3, target + nop + b 1f + nop + nop + nop + 1: + + to the new sequence if possible: + bne $2, $3, 1f + nop + j target + nop + nop + nop + 1: + + OR to the new sequence: + bne $2, $3, 1f + nop + lui $25, target >> 16 + ori $25, $25, target & 0xffff + jr $25 + nop + 1: + + Note: beq/bne/bc1t/bc1f are converted to bne/beq/bc1f/bc1t. + */ + + if (*(insn + 2) == 0x10000003) { + if ((*insn & 0xfc000000) == 0x10000000) // beq + *insn = (*insn & 0x03ff0000) | 0x14000005; // bne + else if ((*insn & 0xfc000000) == 0x14000000) // bne + *insn = (*insn & 0x03ff0000) | 0x10000005; // beq + else if ((*insn & 0xffff0000) == 0x45010000) // bc1t + *insn = 0x45000005; // bc1f + else if ((*insn & 0xffff0000) == 0x45000000) // bc1f + *insn = 0x45010005; // bc1t + else + ASSERT(0); + } + + insn = insn + 2; + if ((reinterpret_cast<intptr_t>(insn) + 4) >> 28 + == reinterpret_cast<intptr_t>(to) >> 28) { + *insn = 0x08000000 | ((reinterpret_cast<intptr_t>(to) >> 2) & 0x3ffffff); + *(insn + 1) = 0; + return 4 * sizeof(MIPSWord); + } + + intptr_t newTargetAddress = reinterpret_cast<intptr_t>(to); + /* lui */ + *insn = 0x3c000000 | (MIPSRegisters::t9 << OP_SH_RT) | ((newTargetAddress >> 16) & 0xffff); + /* ori */ + *(insn + 1) = 0x34000000 | (MIPSRegisters::t9 << OP_SH_RT) | (MIPSRegisters::t9 << OP_SH_RS) | (newTargetAddress & 0xffff); + /* jr */ + *(insn + 2) = 0x00000008 | (MIPSRegisters::t9 << OP_SH_RS); + return 5 * sizeof(MIPSWord); + } + + *insn = (*insn & 0xffff0000) | (diff & 0xffff); + return sizeof(MIPSWord); + } + + static int linkCallInternal(void* from, void* to) + { + MIPSWord* insn = reinterpret_cast<MIPSWord*>(from); + insn = insn - 4; + + if ((*(insn + 2) & 0xfc000000) == 0x0c000000) { // jal + if ((reinterpret_cast<intptr_t>(from) - 4) >> 28 + == reinterpret_cast<intptr_t>(to) >> 28) { + *(insn + 2) = 0x0c000000 | ((reinterpret_cast<intptr_t>(to) >> 2) & 0x3ffffff); + return sizeof(MIPSWord); + } + + /* lui $25, (to >> 16) & 0xffff */ + *insn = 0x3c000000 | (MIPSRegisters::t9 << OP_SH_RT) | ((reinterpret_cast<intptr_t>(to) >> 16) & 0xffff); + /* ori $25, $25, to & 0xffff */ + *(insn + 1) = 0x34000000 | (MIPSRegisters::t9 << OP_SH_RT) | (MIPSRegisters::t9 << OP_SH_RS) | (reinterpret_cast<intptr_t>(to) & 0xffff); + /* jalr $25 */ + *(insn + 2) = 0x0000f809 | (MIPSRegisters::t9 << OP_SH_RS); + return 3 * sizeof(MIPSWord); + } + + ASSERT((*insn & 0xffe00000) == 0x3c000000); // lui + ASSERT((*(insn + 1) & 0xfc000000) == 0x34000000); // ori + + /* lui */ + *insn = (*insn & 0xffff0000) | ((reinterpret_cast<intptr_t>(to) >> 16) & 0xffff); + /* ori */ + *(insn + 1) = (*(insn + 1) & 0xffff0000) | (reinterpret_cast<intptr_t>(to) & 0xffff); + return 2 * sizeof(MIPSWord); + } + + AssemblerBuffer m_buffer; + Jumps m_jumps; + int m_indexOfLastWatchpoint; + int m_indexOfTailOfLastWatchpoint; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(MIPS) + +#endif // MIPSAssembler_h diff --git a/src/3rdparty/masm/assembler/MacroAssembler.h b/src/3rdparty/masm/assembler/MacroAssembler.h new file mode 100644 index 0000000000..f74680d7fc --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssembler.h @@ -0,0 +1,1465 @@ +/* + * Copyright (C) 2008, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssembler_h +#define MacroAssembler_h + +#include <wtf/Platform.h> + +#if ENABLE(ASSEMBLER) + +#if CPU(ARM_THUMB2) +#include "MacroAssemblerARMv7.h" +namespace JSC { typedef MacroAssemblerARMv7 MacroAssemblerBase; }; + +#elif CPU(ARM_TRADITIONAL) +#include "MacroAssemblerARM.h" +namespace JSC { typedef MacroAssemblerARM MacroAssemblerBase; }; + +#elif CPU(MIPS) +#include "MacroAssemblerMIPS.h" +namespace JSC { +typedef MacroAssemblerMIPS MacroAssemblerBase; +}; + +#elif CPU(X86) +#include "MacroAssemblerX86.h" +namespace JSC { typedef MacroAssemblerX86 MacroAssemblerBase; }; + +#elif CPU(X86_64) +#include "MacroAssemblerX86_64.h" +namespace JSC { typedef MacroAssemblerX86_64 MacroAssemblerBase; }; + +#elif CPU(SH4) +#include "MacroAssemblerSH4.h" +namespace JSC { +typedef MacroAssemblerSH4 MacroAssemblerBase; +}; + +#else +#error "The MacroAssembler is not supported on this platform." +#endif + +namespace JSC { + +class MacroAssembler : public MacroAssemblerBase { +public: + + using MacroAssemblerBase::pop; + using MacroAssemblerBase::jump; + using MacroAssemblerBase::branch32; + using MacroAssemblerBase::move; + +#if ENABLE(JIT_CONSTANT_BLINDING) + using MacroAssemblerBase::add32; + using MacroAssemblerBase::and32; + using MacroAssemblerBase::branchAdd32; + using MacroAssemblerBase::branchMul32; + using MacroAssemblerBase::branchSub32; + using MacroAssemblerBase::lshift32; + using MacroAssemblerBase::or32; + using MacroAssemblerBase::rshift32; + using MacroAssemblerBase::store32; + using MacroAssemblerBase::sub32; + using MacroAssemblerBase::urshift32; + using MacroAssemblerBase::xor32; +#endif + + static const double twoToThe32; // This is super useful for some double code. + + // Utilities used by the DFG JIT. +#if ENABLE(DFG_JIT) + using MacroAssemblerBase::invert; + + static DoubleCondition invert(DoubleCondition cond) + { + switch (cond) { + case DoubleEqual: + return DoubleNotEqualOrUnordered; + case DoubleNotEqual: + return DoubleEqualOrUnordered; + case DoubleGreaterThan: + return DoubleLessThanOrEqualOrUnordered; + case DoubleGreaterThanOrEqual: + return DoubleLessThanOrUnordered; + case DoubleLessThan: + return DoubleGreaterThanOrEqualOrUnordered; + case DoubleLessThanOrEqual: + return DoubleGreaterThanOrUnordered; + case DoubleEqualOrUnordered: + return DoubleNotEqual; + case DoubleNotEqualOrUnordered: + return DoubleEqual; + case DoubleGreaterThanOrUnordered: + return DoubleLessThanOrEqual; + case DoubleGreaterThanOrEqualOrUnordered: + return DoubleLessThan; + case DoubleLessThanOrUnordered: + return DoubleGreaterThanOrEqual; + case DoubleLessThanOrEqualOrUnordered: + return DoubleGreaterThan; + default: + RELEASE_ASSERT_NOT_REACHED(); + return DoubleEqual; // make compiler happy + } + } + + static bool isInvertible(ResultCondition cond) + { + switch (cond) { + case Zero: + case NonZero: + return true; + default: + return false; + } + } + + static ResultCondition invert(ResultCondition cond) + { + switch (cond) { + case Zero: + return NonZero; + case NonZero: + return Zero; + default: + RELEASE_ASSERT_NOT_REACHED(); + return Zero; // Make compiler happy for release builds. + } + } +#endif + + // Platform agnostic onvenience functions, + // described in terms of other macro assembly methods. + void pop() + { + addPtr(TrustedImm32(sizeof(void*)), stackPointerRegister); + } + + void peek(RegisterID dest, int index = 0) + { + loadPtr(Address(stackPointerRegister, (index * sizeof(void*))), dest); + } + + Address addressForPoke(int index) + { + return Address(stackPointerRegister, (index * sizeof(void*))); + } + + void poke(RegisterID src, int index = 0) + { + storePtr(src, addressForPoke(index)); + } + + void poke(TrustedImm32 value, int index = 0) + { + store32(value, addressForPoke(index)); + } + + void poke(TrustedImmPtr imm, int index = 0) + { + storePtr(imm, addressForPoke(index)); + } + +#if CPU(X86_64) + void peek64(RegisterID dest, int index = 0) + { + load64(Address(stackPointerRegister, (index * sizeof(void*))), dest); + } + + void poke(TrustedImm64 value, int index = 0) + { + store64(value, addressForPoke(index)); + } + + void poke64(RegisterID src, int index = 0) + { + store64(src, addressForPoke(index)); + } +#endif + +#if CPU(MIPS) + void poke(FPRegisterID src, int index = 0) + { + ASSERT(!(index & 1)); + storeDouble(src, addressForPoke(index)); + } +#endif + + // Backwards banches, these are currently all implemented using existing forwards branch mechanisms. + void branchPtr(RelationalCondition cond, RegisterID op1, TrustedImmPtr imm, Label target) + { + branchPtr(cond, op1, imm).linkTo(target, this); + } + void branchPtr(RelationalCondition cond, RegisterID op1, ImmPtr imm, Label target) + { + branchPtr(cond, op1, imm).linkTo(target, this); + } + + void branch32(RelationalCondition cond, RegisterID op1, RegisterID op2, Label target) + { + branch32(cond, op1, op2).linkTo(target, this); + } + + void branch32(RelationalCondition cond, RegisterID op1, TrustedImm32 imm, Label target) + { + branch32(cond, op1, imm).linkTo(target, this); + } + + void branch32(RelationalCondition cond, RegisterID op1, Imm32 imm, Label target) + { + branch32(cond, op1, imm).linkTo(target, this); + } + + void branch32(RelationalCondition cond, RegisterID left, Address right, Label target) + { + branch32(cond, left, right).linkTo(target, this); + } + + Jump branch32(RelationalCondition cond, TrustedImm32 left, RegisterID right) + { + return branch32(commute(cond), right, left); + } + + Jump branch32(RelationalCondition cond, Imm32 left, RegisterID right) + { + return branch32(commute(cond), right, left); + } + + void branchTestPtr(ResultCondition cond, RegisterID reg, Label target) + { + branchTestPtr(cond, reg).linkTo(target, this); + } + +#if !CPU(ARM_THUMB2) + PatchableJump patchableBranchPtr(RelationalCondition cond, Address left, TrustedImmPtr right = TrustedImmPtr(0)) + { + return PatchableJump(branchPtr(cond, left, right)); + } + + PatchableJump patchableBranchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + return PatchableJump(branchPtrWithPatch(cond, left, dataLabel, initialRightValue)); + } + + PatchableJump patchableJump() + { + return PatchableJump(jump()); + } + + PatchableJump patchableBranchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + return PatchableJump(branchTest32(cond, reg, mask)); + } +#endif // !CPU(ARM_THUMB2) + +#if !CPU(ARM) + PatchableJump patchableBranch32(RelationalCondition cond, RegisterID reg, TrustedImm32 imm) + { + return PatchableJump(branch32(cond, reg, imm)); + } +#endif // !(CPU(ARM) + + void jump(Label target) + { + jump().linkTo(target, this); + } + + // Commute a relational condition, returns a new condition that will produce + // the same results given the same inputs but with their positions exchanged. + static RelationalCondition commute(RelationalCondition condition) + { + switch (condition) { + case Above: + return Below; + case AboveOrEqual: + return BelowOrEqual; + case Below: + return Above; + case BelowOrEqual: + return AboveOrEqual; + case GreaterThan: + return LessThan; + case GreaterThanOrEqual: + return LessThanOrEqual; + case LessThan: + return GreaterThan; + case LessThanOrEqual: + return GreaterThanOrEqual; + default: + break; + } + + ASSERT(condition == Equal || condition == NotEqual); + return condition; + } + + static const unsigned BlindingModulus = 64; + bool shouldConsiderBlinding() + { + return !(random() & (BlindingModulus - 1)); + } + + // Ptr methods + // On 32-bit platforms (i.e. x86), these methods directly map onto their 32-bit equivalents. + // FIXME: should this use a test for 32-bitness instead of this specific exception? +#if !CPU(X86_64) + void addPtr(Address src, RegisterID dest) + { + add32(src, dest); + } + + void addPtr(AbsoluteAddress src, RegisterID dest) + { + add32(src, dest); + } + + void addPtr(RegisterID src, RegisterID dest) + { + add32(src, dest); + } + + void addPtr(TrustedImm32 imm, RegisterID srcDest) + { + add32(imm, srcDest); + } + + void addPtr(TrustedImmPtr imm, RegisterID dest) + { + add32(TrustedImm32(imm), dest); + } + + void addPtr(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + add32(imm, src, dest); + } + + void addPtr(TrustedImm32 imm, AbsoluteAddress address) + { + add32(imm, address); + } + + void andPtr(RegisterID src, RegisterID dest) + { + and32(src, dest); + } + + void andPtr(TrustedImm32 imm, RegisterID srcDest) + { + and32(imm, srcDest); + } + + void negPtr(RegisterID dest) + { + neg32(dest); + } + + void orPtr(RegisterID src, RegisterID dest) + { + or32(src, dest); + } + + void orPtr(RegisterID op1, RegisterID op2, RegisterID dest) + { + or32(op1, op2, dest); + } + + void orPtr(TrustedImmPtr imm, RegisterID dest) + { + or32(TrustedImm32(imm), dest); + } + + void orPtr(TrustedImm32 imm, RegisterID dest) + { + or32(imm, dest); + } + + void subPtr(RegisterID src, RegisterID dest) + { + sub32(src, dest); + } + + void subPtr(TrustedImm32 imm, RegisterID dest) + { + sub32(imm, dest); + } + + void subPtr(TrustedImmPtr imm, RegisterID dest) + { + sub32(TrustedImm32(imm), dest); + } + + void xorPtr(RegisterID src, RegisterID dest) + { + xor32(src, dest); + } + + void xorPtr(TrustedImm32 imm, RegisterID srcDest) + { + xor32(imm, srcDest); + } + + + void loadPtr(ImplicitAddress address, RegisterID dest) + { + load32(address, dest); + } + + void loadPtr(BaseIndex address, RegisterID dest) + { + load32(address, dest); + } + + void loadPtr(const void* address, RegisterID dest) + { + load32(address, dest); + } + + DataLabel32 loadPtrWithAddressOffsetPatch(Address address, RegisterID dest) + { + return load32WithAddressOffsetPatch(address, dest); + } + + DataLabelCompact loadPtrWithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + return load32WithCompactAddressOffsetPatch(address, dest); + } + + void move(ImmPtr imm, RegisterID dest) + { + move(Imm32(imm.asTrustedImmPtr()), dest); + } + + void comparePtr(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + compare32(cond, left, right, dest); + } + + void storePtr(RegisterID src, ImplicitAddress address) + { + store32(src, address); + } + + void storePtr(RegisterID src, BaseIndex address) + { + store32(src, address); + } + + void storePtr(RegisterID src, void* address) + { + store32(src, address); + } + + void storePtr(TrustedImmPtr imm, ImplicitAddress address) + { + store32(TrustedImm32(imm), address); + } + + void storePtr(ImmPtr imm, Address address) + { + store32(Imm32(imm.asTrustedImmPtr()), address); + } + + void storePtr(TrustedImmPtr imm, void* address) + { + store32(TrustedImm32(imm), address); + } + + DataLabel32 storePtrWithAddressOffsetPatch(RegisterID src, Address address) + { + return store32WithAddressOffsetPatch(src, address); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, RegisterID right) + { + return branch32(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, TrustedImmPtr right) + { + return branch32(cond, left, TrustedImm32(right)); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, ImmPtr right) + { + return branch32(cond, left, Imm32(right.asTrustedImmPtr())); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, Address right) + { + return branch32(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, Address left, RegisterID right) + { + return branch32(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + return branch32(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, Address left, TrustedImmPtr right) + { + return branch32(cond, left, TrustedImm32(right)); + } + + Jump branchPtr(RelationalCondition cond, AbsoluteAddress left, TrustedImmPtr right) + { + return branch32(cond, left, TrustedImm32(right)); + } + + Jump branchSubPtr(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchSub32(cond, src, dest); + } + + Jump branchTestPtr(ResultCondition cond, RegisterID reg, RegisterID mask) + { + return branchTest32(cond, reg, mask); + } + + Jump branchTestPtr(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest32(cond, reg, mask); + } + + Jump branchTestPtr(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest32(cond, address, mask); + } + + Jump branchTestPtr(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest32(cond, address, mask); + } + + Jump branchAddPtr(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchAdd32(cond, src, dest); + } + + Jump branchSubPtr(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + return branchSub32(cond, imm, dest); + } + using MacroAssemblerBase::branchTest8; + Jump branchTest8(ResultCondition cond, ExtendedAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + return MacroAssemblerBase::branchTest8(cond, Address(address.base, address.offset), mask); + } +#else + void addPtr(RegisterID src, RegisterID dest) + { + add64(src, dest); + } + + void addPtr(Address src, RegisterID dest) + { + add64(src, dest); + } + + void addPtr(TrustedImm32 imm, RegisterID srcDest) + { + add64(imm, srcDest); + } + + void addPtr(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + add64(imm, src, dest); + } + + void addPtr(TrustedImm32 imm, Address address) + { + add64(imm, address); + } + + void addPtr(AbsoluteAddress src, RegisterID dest) + { + add64(src, dest); + } + + void addPtr(TrustedImmPtr imm, RegisterID dest) + { + add64(TrustedImm64(imm), dest); + } + + void addPtr(TrustedImm32 imm, AbsoluteAddress address) + { + add64(imm, address); + } + + void andPtr(RegisterID src, RegisterID dest) + { + and64(src, dest); + } + + void andPtr(TrustedImm32 imm, RegisterID srcDest) + { + and64(imm, srcDest); + } + + void negPtr(RegisterID dest) + { + neg64(dest); + } + + void orPtr(RegisterID src, RegisterID dest) + { + or64(src, dest); + } + + void orPtr(TrustedImm32 imm, RegisterID dest) + { + or64(imm, dest); + } + + void orPtr(TrustedImmPtr imm, RegisterID dest) + { + or64(TrustedImm64(imm), dest); + } + + void orPtr(RegisterID op1, RegisterID op2, RegisterID dest) + { + or64(op1, op2, dest); + } + + void orPtr(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + or64(imm, src, dest); + } + + void rotateRightPtr(TrustedImm32 imm, RegisterID srcDst) + { + rotateRight64(imm, srcDst); + } + + void subPtr(RegisterID src, RegisterID dest) + { + sub64(src, dest); + } + + void subPtr(TrustedImm32 imm, RegisterID dest) + { + sub64(imm, dest); + } + + void subPtr(TrustedImmPtr imm, RegisterID dest) + { + sub64(TrustedImm64(imm), dest); + } + + void xorPtr(RegisterID src, RegisterID dest) + { + xor64(src, dest); + } + + void xorPtr(RegisterID src, Address dest) + { + xor64(src, dest); + } + + void xorPtr(TrustedImm32 imm, RegisterID srcDest) + { + xor64(imm, srcDest); + } + + void loadPtr(ImplicitAddress address, RegisterID dest) + { + load64(address, dest); + } + + void loadPtr(BaseIndex address, RegisterID dest) + { + load64(address, dest); + } + + void loadPtr(const void* address, RegisterID dest) + { + load64(address, dest); + } + + DataLabel32 loadPtrWithAddressOffsetPatch(Address address, RegisterID dest) + { + return load64WithAddressOffsetPatch(address, dest); + } + + DataLabelCompact loadPtrWithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + return load64WithCompactAddressOffsetPatch(address, dest); + } + + void storePtr(RegisterID src, ImplicitAddress address) + { + store64(src, address); + } + + void storePtr(RegisterID src, BaseIndex address) + { + store64(src, address); + } + + void storePtr(RegisterID src, void* address) + { + store64(src, address); + } + + void storePtr(TrustedImmPtr imm, ImplicitAddress address) + { + store64(TrustedImm64(imm), address); + } + + void storePtr(TrustedImmPtr imm, BaseIndex address) + { + store64(TrustedImm64(imm), address); + } + + DataLabel32 storePtrWithAddressOffsetPatch(RegisterID src, Address address) + { + return store64WithAddressOffsetPatch(src, address); + } + + void comparePtr(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + compare64(cond, left, right, dest); + } + + void comparePtr(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + compare64(cond, left, right, dest); + } + + void testPtr(ResultCondition cond, RegisterID reg, TrustedImm32 mask, RegisterID dest) + { + test64(cond, reg, mask, dest); + } + + void testPtr(ResultCondition cond, RegisterID reg, RegisterID mask, RegisterID dest) + { + test64(cond, reg, mask, dest); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, RegisterID right) + { + return branch64(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, TrustedImmPtr right) + { + return branch64(cond, left, TrustedImm64(right)); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, Address right) + { + return branch64(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, Address left, RegisterID right) + { + return branch64(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + return branch64(cond, left, right); + } + + Jump branchPtr(RelationalCondition cond, Address left, TrustedImmPtr right) + { + return branch64(cond, left, TrustedImm64(right)); + } + + Jump branchTestPtr(ResultCondition cond, RegisterID reg, RegisterID mask) + { + return branchTest64(cond, reg, mask); + } + + Jump branchTestPtr(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest64(cond, reg, mask); + } + + Jump branchTestPtr(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest64(cond, address, mask); + } + + Jump branchTestPtr(ResultCondition cond, Address address, RegisterID reg) + { + return branchTest64(cond, address, reg); + } + + Jump branchTestPtr(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest64(cond, address, mask); + } + + Jump branchTestPtr(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + return branchTest64(cond, address, mask); + } + + Jump branchAddPtr(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + return branchAdd64(cond, imm, dest); + } + + Jump branchAddPtr(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchAdd64(cond, src, dest); + } + + Jump branchSubPtr(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + return branchSub64(cond, imm, dest); + } + + Jump branchSubPtr(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchSub64(cond, src, dest); + } + + Jump branchSubPtr(ResultCondition cond, RegisterID src1, TrustedImm32 src2, RegisterID dest) + { + return branchSub64(cond, src1, src2, dest); + } + +#if ENABLE(JIT_CONSTANT_BLINDING) + using MacroAssemblerBase::and64; + using MacroAssemblerBase::convertInt32ToDouble; + using MacroAssemblerBase::store64; + bool shouldBlindDouble(double value) + { + // Don't trust NaN or +/-Infinity + if (!std::isfinite(value)) + return shouldConsiderBlinding(); + + // Try to force normalisation, and check that there's no change + // in the bit pattern + if (bitwise_cast<uint64_t>(value * 1.0) != bitwise_cast<uint64_t>(value)) + return shouldConsiderBlinding(); + + value = abs(value); + // Only allow a limited set of fractional components + double scaledValue = value * 8; + if (scaledValue / 8 != value) + return shouldConsiderBlinding(); + double frac = scaledValue - floor(scaledValue); + if (frac != 0.0) + return shouldConsiderBlinding(); + + return value > 0xff; + } + + bool shouldBlind(ImmPtr imm) + { +#if ENABLE(FORCED_JIT_BLINDING) + UNUSED_PARAM(imm); + // Debug always blind all constants, if only so we know + // if we've broken blinding during patch development. + return true; +#endif + + // First off we'll special case common, "safe" values to avoid hurting + // performance too much + uintptr_t value = imm.asTrustedImmPtr().asIntptr(); + switch (value) { + case 0xffff: + case 0xffffff: + case 0xffffffffL: + case 0xffffffffffL: + case 0xffffffffffffL: + case 0xffffffffffffffL: + case 0xffffffffffffffffL: + return false; + default: { + if (value <= 0xff) + return false; + if (~value <= 0xff) + return false; + } + } + + if (!shouldConsiderBlinding()) + return false; + + return shouldBlindForSpecificArch(value); + } + + struct RotatedImmPtr { + RotatedImmPtr(uintptr_t v1, uint8_t v2) + : value(v1) + , rotation(v2) + { + } + TrustedImmPtr value; + TrustedImm32 rotation; + }; + + RotatedImmPtr rotationBlindConstant(ImmPtr imm) + { + uint8_t rotation = random() % (sizeof(void*) * 8); + uintptr_t value = imm.asTrustedImmPtr().asIntptr(); + value = (value << rotation) | (value >> (sizeof(void*) * 8 - rotation)); + return RotatedImmPtr(value, rotation); + } + + void loadRotationBlindedConstant(RotatedImmPtr constant, RegisterID dest) + { + move(constant.value, dest); + rotateRightPtr(constant.rotation, dest); + } + + bool shouldBlind(Imm64 imm) + { +#if ENABLE(FORCED_JIT_BLINDING) + UNUSED_PARAM(imm); + // Debug always blind all constants, if only so we know + // if we've broken blinding during patch development. + return true; +#endif + + // First off we'll special case common, "safe" values to avoid hurting + // performance too much + uint64_t value = imm.asTrustedImm64().m_value; + switch (value) { + case 0xffff: + case 0xffffff: + case 0xffffffffL: + case 0xffffffffffL: + case 0xffffffffffffL: + case 0xffffffffffffffL: + case 0xffffffffffffffffL: + return false; + default: { + if (value <= 0xff) + return false; + if (~value <= 0xff) + return false; + + JSValue jsValue = JSValue::decode(value); + if (jsValue.isInt32()) + return shouldBlind(Imm32(jsValue.asInt32())); + if (jsValue.isDouble() && !shouldBlindDouble(jsValue.asDouble())) + return false; + + if (!shouldBlindDouble(bitwise_cast<double>(value))) + return false; + } + } + + if (!shouldConsiderBlinding()) + return false; + + return shouldBlindForSpecificArch(value); + } + + struct RotatedImm64 { + RotatedImm64(uint64_t v1, uint8_t v2) + : value(v1) + , rotation(v2) + { + } + TrustedImm64 value; + TrustedImm32 rotation; + }; + + RotatedImm64 rotationBlindConstant(Imm64 imm) + { + uint8_t rotation = random() % (sizeof(int64_t) * 8); + uint64_t value = imm.asTrustedImm64().m_value; + value = (value << rotation) | (value >> (sizeof(int64_t) * 8 - rotation)); + return RotatedImm64(value, rotation); + } + + void loadRotationBlindedConstant(RotatedImm64 constant, RegisterID dest) + { + move(constant.value, dest); + rotateRight64(constant.rotation, dest); + } + + void convertInt32ToDouble(Imm32 imm, FPRegisterID dest) + { + if (shouldBlind(imm)) { + RegisterID scratchRegister = scratchRegisterForBlinding(); + loadXorBlindedConstant(xorBlindConstant(imm), scratchRegister); + convertInt32ToDouble(scratchRegister, dest); + } else + convertInt32ToDouble(imm.asTrustedImm32(), dest); + } + + void move(ImmPtr imm, RegisterID dest) + { + if (shouldBlind(imm)) + loadRotationBlindedConstant(rotationBlindConstant(imm), dest); + else + move(imm.asTrustedImmPtr(), dest); + } + + void move(Imm64 imm, RegisterID dest) + { + if (shouldBlind(imm)) + loadRotationBlindedConstant(rotationBlindConstant(imm), dest); + else + move(imm.asTrustedImm64(), dest); + } + + void and64(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = andBlindedConstant(imm); + and64(key.value1, dest); + and64(key.value2, dest); + } else + and64(imm.asTrustedImm32(), dest); + } + + Jump branchPtr(RelationalCondition cond, RegisterID left, ImmPtr right) + { + if (shouldBlind(right)) { + RegisterID scratchRegister = scratchRegisterForBlinding(); + loadRotationBlindedConstant(rotationBlindConstant(right), scratchRegister); + return branchPtr(cond, left, scratchRegister); + } + return branchPtr(cond, left, right.asTrustedImmPtr()); + } + + void storePtr(ImmPtr imm, Address dest) + { + if (shouldBlind(imm)) { + RegisterID scratchRegister = scratchRegisterForBlinding(); + loadRotationBlindedConstant(rotationBlindConstant(imm), scratchRegister); + storePtr(scratchRegister, dest); + } else + storePtr(imm.asTrustedImmPtr(), dest); + } + + void store64(Imm64 imm, Address dest) + { + if (shouldBlind(imm)) { + RegisterID scratchRegister = scratchRegisterForBlinding(); + loadRotationBlindedConstant(rotationBlindConstant(imm), scratchRegister); + store64(scratchRegister, dest); + } else + store64(imm.asTrustedImm64(), dest); + } + +#endif + +#endif // !CPU(X86_64) + +#if ENABLE(JIT_CONSTANT_BLINDING) + bool shouldBlind(Imm32 imm) + { +#if ENABLE(FORCED_JIT_BLINDING) + UNUSED_PARAM(imm); + // Debug always blind all constants, if only so we know + // if we've broken blinding during patch development. + return true; +#else + + // First off we'll special case common, "safe" values to avoid hurting + // performance too much + uint32_t value = imm.asTrustedImm32().m_value; + switch (value) { + case 0xffff: + case 0xffffff: + case 0xffffffff: + return false; + default: + if (value <= 0xff) + return false; + if (~value <= 0xff) + return false; + } + + if (!shouldConsiderBlinding()) + return false; + + return shouldBlindForSpecificArch(value); +#endif + } + + struct BlindedImm32 { + BlindedImm32(int32_t v1, int32_t v2) + : value1(v1) + , value2(v2) + { + } + TrustedImm32 value1; + TrustedImm32 value2; + }; + + uint32_t keyForConstant(uint32_t value, uint32_t& mask) + { + uint32_t key = random(); + if (value <= 0xff) + mask = 0xff; + else if (value <= 0xffff) + mask = 0xffff; + else if (value <= 0xffffff) + mask = 0xffffff; + else + mask = 0xffffffff; + return key & mask; + } + + uint32_t keyForConstant(uint32_t value) + { + uint32_t mask = 0; + return keyForConstant(value, mask); + } + + BlindedImm32 xorBlindConstant(Imm32 imm) + { + uint32_t baseValue = imm.asTrustedImm32().m_value; + uint32_t key = keyForConstant(baseValue); + return BlindedImm32(baseValue ^ key, key); + } + + BlindedImm32 additionBlindedConstant(Imm32 imm) + { + // The addition immediate may be used as a pointer offset. Keep aligned based on "imm". + static uint32_t maskTable[4] = { 0xfffffffc, 0xffffffff, 0xfffffffe, 0xffffffff }; + + uint32_t baseValue = imm.asTrustedImm32().m_value; + uint32_t key = keyForConstant(baseValue) & maskTable[baseValue & 3]; + if (key > baseValue) + key = key - baseValue; + return BlindedImm32(baseValue - key, key); + } + + BlindedImm32 andBlindedConstant(Imm32 imm) + { + uint32_t baseValue = imm.asTrustedImm32().m_value; + uint32_t mask = 0; + uint32_t key = keyForConstant(baseValue, mask); + ASSERT((baseValue & mask) == baseValue); + return BlindedImm32(((baseValue & key) | ~key) & mask, ((baseValue & ~key) | key) & mask); + } + + BlindedImm32 orBlindedConstant(Imm32 imm) + { + uint32_t baseValue = imm.asTrustedImm32().m_value; + uint32_t mask = 0; + uint32_t key = keyForConstant(baseValue, mask); + ASSERT((baseValue & mask) == baseValue); + return BlindedImm32((baseValue & key) & mask, (baseValue & ~key) & mask); + } + + void loadXorBlindedConstant(BlindedImm32 constant, RegisterID dest) + { + move(constant.value1, dest); + xor32(constant.value2, dest); + } + + void add32(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = additionBlindedConstant(imm); + add32(key.value1, dest); + add32(key.value2, dest); + } else + add32(imm.asTrustedImm32(), dest); + } + + void addPtr(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = additionBlindedConstant(imm); + addPtr(key.value1, dest); + addPtr(key.value2, dest); + } else + addPtr(imm.asTrustedImm32(), dest); + } + + void and32(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = andBlindedConstant(imm); + and32(key.value1, dest); + and32(key.value2, dest); + } else + and32(imm.asTrustedImm32(), dest); + } + + void andPtr(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = andBlindedConstant(imm); + andPtr(key.value1, dest); + andPtr(key.value2, dest); + } else + andPtr(imm.asTrustedImm32(), dest); + } + + void and32(Imm32 imm, RegisterID src, RegisterID dest) + { + if (shouldBlind(imm)) { + if (src == dest) + return and32(imm.asTrustedImm32(), dest); + loadXorBlindedConstant(xorBlindConstant(imm), dest); + and32(src, dest); + } else + and32(imm.asTrustedImm32(), src, dest); + } + + void move(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) + loadXorBlindedConstant(xorBlindConstant(imm), dest); + else + move(imm.asTrustedImm32(), dest); + } + + void or32(Imm32 imm, RegisterID src, RegisterID dest) + { + if (shouldBlind(imm)) { + if (src == dest) + return or32(imm, dest); + loadXorBlindedConstant(xorBlindConstant(imm), dest); + or32(src, dest); + } else + or32(imm.asTrustedImm32(), src, dest); + } + + void or32(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = orBlindedConstant(imm); + or32(key.value1, dest); + or32(key.value2, dest); + } else + or32(imm.asTrustedImm32(), dest); + } + + void poke(Imm32 value, int index = 0) + { + store32(value, addressForPoke(index)); + } + + void poke(ImmPtr value, int index = 0) + { + storePtr(value, addressForPoke(index)); + } + +#if CPU(X86_64) + void poke(Imm64 value, int index = 0) + { + store64(value, addressForPoke(index)); + } +#endif + + void store32(Imm32 imm, Address dest) + { + if (shouldBlind(imm)) { +#if CPU(X86) || CPU(X86_64) + BlindedImm32 blind = xorBlindConstant(imm); + store32(blind.value1, dest); + xor32(blind.value2, dest); +#else + if (RegisterID scratchRegister = (RegisterID)scratchRegisterForBlinding()) { + loadXorBlindedConstant(xorBlindConstant(imm), scratchRegister); + store32(scratchRegister, dest); + } else { + // If we don't have a scratch register available for use, we'll just + // place a random number of nops. + uint32_t nopCount = random() & 3; + while (nopCount--) + nop(); + store32(imm.asTrustedImm32(), dest); + } +#endif + } else + store32(imm.asTrustedImm32(), dest); + } + + void sub32(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = additionBlindedConstant(imm); + sub32(key.value1, dest); + sub32(key.value2, dest); + } else + sub32(imm.asTrustedImm32(), dest); + } + + void subPtr(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 key = additionBlindedConstant(imm); + subPtr(key.value1, dest); + subPtr(key.value2, dest); + } else + subPtr(imm.asTrustedImm32(), dest); + } + + void xor32(Imm32 imm, RegisterID src, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 blind = xorBlindConstant(imm); + xor32(blind.value1, src, dest); + xor32(blind.value2, dest); + } else + xor32(imm.asTrustedImm32(), src, dest); + } + + void xor32(Imm32 imm, RegisterID dest) + { + if (shouldBlind(imm)) { + BlindedImm32 blind = xorBlindConstant(imm); + xor32(blind.value1, dest); + xor32(blind.value2, dest); + } else + xor32(imm.asTrustedImm32(), dest); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Imm32 right) + { + if (shouldBlind(right)) { + if (RegisterID scratchRegister = (RegisterID)scratchRegisterForBlinding()) { + loadXorBlindedConstant(xorBlindConstant(right), scratchRegister); + return branch32(cond, left, scratchRegister); + } + // If we don't have a scratch register available for use, we'll just + // place a random number of nops. + uint32_t nopCount = random() & 3; + while (nopCount--) + nop(); + return branch32(cond, left, right.asTrustedImm32()); + } + + return branch32(cond, left, right.asTrustedImm32()); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, Imm32 imm, RegisterID dest) + { + if (src == dest) + ASSERT(scratchRegisterForBlinding()); + + if (shouldBlind(imm)) { + if (src == dest) { + if (RegisterID scratchRegister = (RegisterID)scratchRegisterForBlinding()) { + move(src, scratchRegister); + src = scratchRegister; + } + } + loadXorBlindedConstant(xorBlindConstant(imm), dest); + return branchAdd32(cond, src, dest); + } + return branchAdd32(cond, src, imm.asTrustedImm32(), dest); + } + + Jump branchMul32(ResultCondition cond, Imm32 imm, RegisterID src, RegisterID dest) + { + if (src == dest) + ASSERT(scratchRegisterForBlinding()); + + if (shouldBlind(imm)) { + if (src == dest) { + if (RegisterID scratchRegister = (RegisterID)scratchRegisterForBlinding()) { + move(src, scratchRegister); + src = scratchRegister; + } + } + loadXorBlindedConstant(xorBlindConstant(imm), dest); + return branchMul32(cond, src, dest); + } + return branchMul32(cond, imm.asTrustedImm32(), src, dest); + } + + // branchSub32 takes a scratch register as 32 bit platforms make use of this, + // with src == dst, and on x86-32 we don't have a platform scratch register. + Jump branchSub32(ResultCondition cond, RegisterID src, Imm32 imm, RegisterID dest, RegisterID scratch) + { + if (shouldBlind(imm)) { + ASSERT(scratch != dest); + ASSERT(scratch != src); + loadXorBlindedConstant(xorBlindConstant(imm), scratch); + return branchSub32(cond, src, scratch, dest); + } + return branchSub32(cond, src, imm.asTrustedImm32(), dest); + } + + // Immediate shifts only have 5 controllable bits + // so we'll consider them safe for now. + TrustedImm32 trustedImm32ForShift(Imm32 imm) + { + return TrustedImm32(imm.asTrustedImm32().m_value & 31); + } + + void lshift32(Imm32 imm, RegisterID dest) + { + lshift32(trustedImm32ForShift(imm), dest); + } + + void lshift32(RegisterID src, Imm32 amount, RegisterID dest) + { + lshift32(src, trustedImm32ForShift(amount), dest); + } + + void rshift32(Imm32 imm, RegisterID dest) + { + rshift32(trustedImm32ForShift(imm), dest); + } + + void rshift32(RegisterID src, Imm32 amount, RegisterID dest) + { + rshift32(src, trustedImm32ForShift(amount), dest); + } + + void urshift32(Imm32 imm, RegisterID dest) + { + urshift32(trustedImm32ForShift(imm), dest); + } + + void urshift32(RegisterID src, Imm32 amount, RegisterID dest) + { + urshift32(src, trustedImm32ForShift(amount), dest); + } +#endif +}; + +} // namespace JSC + +#else // ENABLE(ASSEMBLER) + +// If there is no assembler for this platform, at least allow code to make references to +// some of the things it would otherwise define, albeit without giving that code any way +// of doing anything useful. +class MacroAssembler { +private: + MacroAssembler() { } + +public: + + enum RegisterID { NoRegister }; + enum FPRegisterID { NoFPRegister }; +}; + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssembler_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerARM.cpp b/src/3rdparty/masm/assembler/MacroAssemblerARM.cpp new file mode 100644 index 0000000000..98dc3e9879 --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerARM.cpp @@ -0,0 +1,99 @@ +/* + * Copyright (C) 2009 University of Szeged + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY UNIVERSITY OF SZEGED ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL UNIVERSITY OF SZEGED OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "config.h" + +#if ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#include "MacroAssemblerARM.h" + +#if OS(LINUX) +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <unistd.h> +#include <elf.h> +#include <asm/hwcap.h> +#endif + +namespace JSC { + +static bool isVFPPresent() +{ +#if OS(LINUX) + int fd = open("/proc/self/auxv", O_RDONLY); + if (fd > 0) { + Elf32_auxv_t aux; + while (read(fd, &aux, sizeof(Elf32_auxv_t))) { + if (aux.a_type == AT_HWCAP) { + close(fd); + return aux.a_un.a_val & HWCAP_VFP; + } + } + close(fd); + } +#endif + +#if (COMPILER(RVCT) && defined(__TARGET_FPU_VFP)) || (COMPILER(GCC) && defined(__VFP_FP__)) + return true; +#else + return false; +#endif +} + +const bool MacroAssemblerARM::s_isVFPPresent = isVFPPresent(); + +#if CPU(ARMV5_OR_LOWER) +/* On ARMv5 and below, natural alignment is required. */ +void MacroAssemblerARM::load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) +{ + ARMWord op2; + + ASSERT(address.scale >= 0 && address.scale <= 3); + op2 = m_assembler.lsl(address.index, static_cast<int>(address.scale)); + + if (address.offset >= 0 && address.offset + 0x2 <= 0xff) { + m_assembler.add(ARMRegisters::S0, address.base, op2); + m_assembler.halfDtrUp(ARMAssembler::LoadUint16, dest, ARMRegisters::S0, ARMAssembler::getOp2Half(address.offset)); + m_assembler.halfDtrUp(ARMAssembler::LoadUint16, ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Half(address.offset + 0x2)); + } else if (address.offset < 0 && address.offset >= -0xff) { + m_assembler.add(ARMRegisters::S0, address.base, op2); + m_assembler.halfDtrDown(ARMAssembler::LoadUint16, dest, ARMRegisters::S0, ARMAssembler::getOp2Half(-address.offset)); + m_assembler.halfDtrDown(ARMAssembler::LoadUint16, ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Half(-address.offset - 0x2)); + } else { + m_assembler.moveImm(address.offset, ARMRegisters::S0); + m_assembler.add(ARMRegisters::S0, ARMRegisters::S0, op2); + m_assembler.halfDtrUpRegister(ARMAssembler::LoadUint16, dest, address.base, ARMRegisters::S0); + m_assembler.add(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::Op2Immediate | 0x2); + m_assembler.halfDtrUpRegister(ARMAssembler::LoadUint16, ARMRegisters::S0, address.base, ARMRegisters::S0); + } + m_assembler.orr(dest, dest, m_assembler.lsl(ARMRegisters::S0, 16)); +} +#endif + +} + +#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) diff --git a/src/3rdparty/masm/assembler/MacroAssemblerARM.h b/src/3rdparty/masm/assembler/MacroAssemblerARM.h new file mode 100644 index 0000000000..01e34c97cd --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerARM.h @@ -0,0 +1,1386 @@ +/* + * Copyright (C) 2008 Apple Inc. + * Copyright (C) 2009, 2010 University of Szeged + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerARM_h +#define MacroAssemblerARM_h + +#if ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#include "ARMAssembler.h" +#include "AbstractMacroAssembler.h" + +namespace JSC { + +class MacroAssemblerARM : public AbstractMacroAssembler<ARMAssembler> { + static const int DoubleConditionMask = 0x0f; + static const int DoubleConditionBitSpecial = 0x10; + COMPILE_ASSERT(!(DoubleConditionBitSpecial & DoubleConditionMask), DoubleConditionBitSpecial_should_not_interfere_with_ARMAssembler_Condition_codes); +public: + typedef ARMRegisters::FPRegisterID FPRegisterID; + + enum RelationalCondition { + Equal = ARMAssembler::EQ, + NotEqual = ARMAssembler::NE, + Above = ARMAssembler::HI, + AboveOrEqual = ARMAssembler::CS, + Below = ARMAssembler::CC, + BelowOrEqual = ARMAssembler::LS, + GreaterThan = ARMAssembler::GT, + GreaterThanOrEqual = ARMAssembler::GE, + LessThan = ARMAssembler::LT, + LessThanOrEqual = ARMAssembler::LE + }; + + enum ResultCondition { + Overflow = ARMAssembler::VS, + Signed = ARMAssembler::MI, + Zero = ARMAssembler::EQ, + NonZero = ARMAssembler::NE + }; + + enum DoubleCondition { + // These conditions will only evaluate to true if the comparison is ordered - i.e. neither operand is NaN. + DoubleEqual = ARMAssembler::EQ, + DoubleNotEqual = ARMAssembler::NE | DoubleConditionBitSpecial, + DoubleGreaterThan = ARMAssembler::GT, + DoubleGreaterThanOrEqual = ARMAssembler::GE, + DoubleLessThan = ARMAssembler::CC, + DoubleLessThanOrEqual = ARMAssembler::LS, + // If either operand is NaN, these conditions always evaluate to true. + DoubleEqualOrUnordered = ARMAssembler::EQ | DoubleConditionBitSpecial, + DoubleNotEqualOrUnordered = ARMAssembler::NE, + DoubleGreaterThanOrUnordered = ARMAssembler::HI, + DoubleGreaterThanOrEqualOrUnordered = ARMAssembler::CS, + DoubleLessThanOrUnordered = ARMAssembler::LT, + DoubleLessThanOrEqualOrUnordered = ARMAssembler::LE, + }; + + static const RegisterID stackPointerRegister = ARMRegisters::sp; + static const RegisterID linkRegister = ARMRegisters::lr; + + static const Scale ScalePtr = TimesFour; + + void add32(RegisterID src, RegisterID dest) + { + m_assembler.adds(dest, dest, src); + } + + void add32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.adds(dest, op1, op2); + } + + void add32(TrustedImm32 imm, Address address) + { + load32(address, ARMRegisters::S1); + add32(imm, ARMRegisters::S1); + store32(ARMRegisters::S1, address); + } + + void add32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.adds(dest, dest, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void add32(AbsoluteAddress src, RegisterID dest) + { + move(TrustedImmPtr(src.m_ptr), ARMRegisters::S1); + m_assembler.dtrUp(ARMAssembler::LoadUint32, ARMRegisters::S1, ARMRegisters::S1, 0); + add32(ARMRegisters::S1, dest); + } + + void add32(Address src, RegisterID dest) + { + load32(src, ARMRegisters::S1); + add32(ARMRegisters::S1, dest); + } + + void add32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.adds(dest, src, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void and32(RegisterID src, RegisterID dest) + { + m_assembler.bitAnds(dest, dest, src); + } + + void and32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.bitAnds(dest, op1, op2); + } + + void and32(TrustedImm32 imm, RegisterID dest) + { + ARMWord w = m_assembler.getImm(imm.m_value, ARMRegisters::S0, true); + if (w & ARMAssembler::Op2InvertedImmediate) + m_assembler.bics(dest, dest, w & ~ARMAssembler::Op2InvertedImmediate); + else + m_assembler.bitAnds(dest, dest, w); + } + + void and32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ARMWord w = m_assembler.getImm(imm.m_value, ARMRegisters::S0, true); + if (w & ARMAssembler::Op2InvertedImmediate) + m_assembler.bics(dest, src, w & ~ARMAssembler::Op2InvertedImmediate); + else + m_assembler.bitAnds(dest, src, w); + } + + void and32(Address src, RegisterID dest) + { + load32(src, ARMRegisters::S1); + and32(ARMRegisters::S1, dest); + } + + void lshift32(RegisterID shiftAmount, RegisterID dest) + { + lshift32(dest, shiftAmount, dest); + } + + void lshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + ARMWord w = ARMAssembler::getOp2Byte(0x1f); + m_assembler.bitAnd(ARMRegisters::S0, shiftAmount, w); + + m_assembler.movs(dest, m_assembler.lslRegister(src, ARMRegisters::S0)); + } + + void lshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.movs(dest, m_assembler.lsl(dest, imm.m_value & 0x1f)); + } + + void lshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.movs(dest, m_assembler.lsl(src, imm.m_value & 0x1f)); + } + + void mul32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op2 == dest) { + if (op1 == dest) { + move(op2, ARMRegisters::S0); + op2 = ARMRegisters::S0; + } else { + // Swap the operands. + RegisterID tmp = op1; + op1 = op2; + op2 = tmp; + } + } + m_assembler.muls(dest, op1, op2); + } + + void mul32(RegisterID src, RegisterID dest) + { + mul32(src, dest, dest); + } + + void mul32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(imm, ARMRegisters::S0); + m_assembler.muls(dest, src, ARMRegisters::S0); + } + + void neg32(RegisterID srcDest) + { + m_assembler.rsbs(srcDest, srcDest, ARMAssembler::getOp2Byte(0)); + } + + void or32(RegisterID src, RegisterID dest) + { + m_assembler.orrs(dest, dest, src); + } + + void or32(RegisterID src, AbsoluteAddress dest) + { + move(TrustedImmPtr(dest.m_ptr), ARMRegisters::S0); + load32(Address(ARMRegisters::S0), ARMRegisters::S1); + or32(src, ARMRegisters::S1); + store32(ARMRegisters::S1, ARMRegisters::S0); + } + + void or32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.orrs(dest, dest, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void or32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.orrs(dest, src, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void or32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.orrs(dest, op1, op2); + } + + void rshift32(RegisterID shiftAmount, RegisterID dest) + { + rshift32(dest, shiftAmount, dest); + } + + void rshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + ARMWord w = ARMAssembler::getOp2Byte(0x1f); + m_assembler.bitAnd(ARMRegisters::S0, shiftAmount, w); + + m_assembler.movs(dest, m_assembler.asrRegister(src, ARMRegisters::S0)); + } + + void rshift32(TrustedImm32 imm, RegisterID dest) + { + rshift32(dest, imm, dest); + } + + void rshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.movs(dest, m_assembler.asr(src, imm.m_value & 0x1f)); + } + + void urshift32(RegisterID shiftAmount, RegisterID dest) + { + urshift32(dest, shiftAmount, dest); + } + + void urshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + ARMWord w = ARMAssembler::getOp2Byte(0x1f); + m_assembler.bitAnd(ARMRegisters::S0, shiftAmount, w); + + m_assembler.movs(dest, m_assembler.lsrRegister(src, ARMRegisters::S0)); + } + + void urshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.movs(dest, m_assembler.lsr(dest, imm.m_value & 0x1f)); + } + + void urshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.movs(dest, m_assembler.lsr(src, imm.m_value & 0x1f)); + } + + void sub32(RegisterID src, RegisterID dest) + { + m_assembler.subs(dest, dest, src); + } + + void sub32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.subs(dest, dest, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void sub32(TrustedImm32 imm, Address address) + { + load32(address, ARMRegisters::S1); + sub32(imm, ARMRegisters::S1); + store32(ARMRegisters::S1, address); + } + + void sub32(Address src, RegisterID dest) + { + load32(src, ARMRegisters::S1); + sub32(ARMRegisters::S1, dest); + } + + void sub32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.subs(dest, src, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void xor32(RegisterID src, RegisterID dest) + { + m_assembler.eors(dest, dest, src); + } + + void xor32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.eors(dest, op1, op2); + } + + void xor32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value == -1) + m_assembler.mvns(dest, dest); + else + m_assembler.eors(dest, dest, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void xor32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (imm.m_value == -1) + m_assembler.mvns(dest, src); + else + m_assembler.eors(dest, src, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void countLeadingZeros32(RegisterID src, RegisterID dest) + { +#if WTF_ARM_ARCH_AT_LEAST(5) + m_assembler.clz(dest, src); +#else + UNUSED_PARAM(src); + UNUSED_PARAM(dest); + RELEASE_ASSERT_NOT_REACHED(); +#endif + } + + void load8(ImplicitAddress address, RegisterID dest) + { + m_assembler.dataTransfer32(ARMAssembler::LoadUint8, dest, address.base, address.offset); + } + + void load8(BaseIndex address, RegisterID dest) + { + m_assembler.baseIndexTransfer32(ARMAssembler::LoadUint8, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void load8Signed(BaseIndex address, RegisterID dest) + { + m_assembler.baseIndexTransfer16(ARMAssembler::LoadInt8, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void load16(ImplicitAddress address, RegisterID dest) + { + m_assembler.dataTransfer16(ARMAssembler::LoadUint16, dest, address.base, address.offset); + } + + void load16(BaseIndex address, RegisterID dest) + { + m_assembler.baseIndexTransfer16(ARMAssembler::LoadUint16, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void load16Signed(BaseIndex address, RegisterID dest) + { + m_assembler.baseIndexTransfer16(ARMAssembler::LoadInt16, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void load32(ImplicitAddress address, RegisterID dest) + { + m_assembler.dataTransfer32(ARMAssembler::LoadUint32, dest, address.base, address.offset); + } + + void load32(BaseIndex address, RegisterID dest) + { + m_assembler.baseIndexTransfer32(ARMAssembler::LoadUint32, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + +#if CPU(ARMV5_OR_LOWER) + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest); +#else + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) + { + load32(address, dest); + } +#endif + + void load16Unaligned(BaseIndex address, RegisterID dest) + { + load16(address, dest); + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result(this); + ASSERT(address.offset >= 0 && address.offset <= 255); + m_assembler.dtrUp(ARMAssembler::LoadUint32, dest, address.base, address.offset); + return result; + } + + DataLabel32 load32WithAddressOffsetPatch(Address address, RegisterID dest) + { + DataLabel32 dataLabel(this); + m_assembler.ldrUniqueImmediate(ARMRegisters::S0, 0); + m_assembler.dtrUpRegister(ARMAssembler::LoadUint32, dest, address.base, ARMRegisters::S0); + return dataLabel; + } + + static bool isCompactPtrAlignedAddressOffset(ptrdiff_t value) + { + return value >= -4095 && value <= 4095; + } + + DataLabelCompact load32WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + DataLabelCompact dataLabel(this); + ASSERT(isCompactPtrAlignedAddressOffset(address.offset)); + if (address.offset >= 0) + m_assembler.dtrUp(ARMAssembler::LoadUint32, dest, address.base, address.offset); + else + m_assembler.dtrDown(ARMAssembler::LoadUint32, dest, address.base, address.offset); + return dataLabel; + } + + DataLabel32 store32WithAddressOffsetPatch(RegisterID src, Address address) + { + DataLabel32 dataLabel(this); + m_assembler.ldrUniqueImmediate(ARMRegisters::S0, 0); + m_assembler.dtrUpRegister(ARMAssembler::StoreUint32, src, address.base, ARMRegisters::S0); + return dataLabel; + } + + void store8(RegisterID src, BaseIndex address) + { + m_assembler.baseIndexTransfer32(ARMAssembler::StoreUint8, src, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void store8(TrustedImm32 imm, const void* address) + { + move(TrustedImm32(reinterpret_cast<ARMWord>(address)), ARMRegisters::S0); + move(imm, ARMRegisters::S1); + m_assembler.dtrUp(ARMAssembler::StoreUint8, ARMRegisters::S1, ARMRegisters::S0, 0); + } + + void store16(RegisterID src, BaseIndex address) + { + m_assembler.baseIndexTransfer16(ARMAssembler::StoreUint16, src, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void store32(RegisterID src, ImplicitAddress address) + { + m_assembler.dataTransfer32(ARMAssembler::StoreUint32, src, address.base, address.offset); + } + + void store32(RegisterID src, BaseIndex address) + { + m_assembler.baseIndexTransfer32(ARMAssembler::StoreUint32, src, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void store32(TrustedImm32 imm, ImplicitAddress address) + { + move(imm, ARMRegisters::S1); + store32(ARMRegisters::S1, address); + } + + void store32(TrustedImm32 imm, BaseIndex address) + { + move(imm, ARMRegisters::S1); + m_assembler.baseIndexTransfer32(ARMAssembler::StoreUint32, ARMRegisters::S1, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void store32(RegisterID src, const void* address) + { + m_assembler.ldrUniqueImmediate(ARMRegisters::S0, reinterpret_cast<ARMWord>(address)); + m_assembler.dtrUp(ARMAssembler::StoreUint32, src, ARMRegisters::S0, 0); + } + + void store32(TrustedImm32 imm, const void* address) + { + m_assembler.ldrUniqueImmediate(ARMRegisters::S0, reinterpret_cast<ARMWord>(address)); + m_assembler.moveImm(imm.m_value, ARMRegisters::S1); + m_assembler.dtrUp(ARMAssembler::StoreUint32, ARMRegisters::S1, ARMRegisters::S0, 0); + } + + void pop(RegisterID dest) + { + m_assembler.pop(dest); + } + + void push(RegisterID src) + { + m_assembler.push(src); + } + + void push(Address address) + { + load32(address, ARMRegisters::S1); + push(ARMRegisters::S1); + } + + void push(TrustedImm32 imm) + { + move(imm, ARMRegisters::S0); + push(ARMRegisters::S0); + } + + void move(TrustedImm32 imm, RegisterID dest) + { + m_assembler.moveImm(imm.m_value, dest); + } + + void move(RegisterID src, RegisterID dest) + { + if (src != dest) + m_assembler.mov(dest, src); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + move(TrustedImm32(imm), dest); + } + + void swap(RegisterID reg1, RegisterID reg2) + { + move(reg1, ARMRegisters::S0); + move(reg2, reg1); + move(ARMRegisters::S0, reg2); + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + if (src != dest) + move(src, dest); + } + + void zeroExtend32ToPtr(RegisterID src, RegisterID dest) + { + if (src != dest) + move(src, dest); + } + + Jump branch8(RelationalCondition cond, Address left, TrustedImm32 right) + { + load8(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch8(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + ASSERT(!(right.m_value & 0xFFFFFF00)); + load8(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch32(RelationalCondition cond, RegisterID left, RegisterID right, int useConstantPool = 0) + { + m_assembler.cmp(left, right); + return Jump(m_assembler.jmp(ARMCondition(cond), useConstantPool)); + } + + Jump branch32(RelationalCondition cond, RegisterID left, TrustedImm32 right, int useConstantPool = 0) + { + internalCompare32(left, right); + return Jump(m_assembler.jmp(ARMCondition(cond), useConstantPool)); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Address right) + { + load32(right, ARMRegisters::S1); + return branch32(cond, left, ARMRegisters::S1); + } + + Jump branch32(RelationalCondition cond, Address left, RegisterID right) + { + load32(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch32(RelationalCondition cond, Address left, TrustedImm32 right) + { + load32(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch32(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + load32(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch32WithUnalignedHalfWords(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + load32WithUnalignedHalfWords(left, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branchTest8(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + load8(address, ARMRegisters::S1); + return branchTest32(cond, ARMRegisters::S1, mask); + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + move(TrustedImmPtr(address.m_ptr), ARMRegisters::S1); + load8(Address(ARMRegisters::S1), ARMRegisters::S1); + return branchTest32(cond, ARMRegisters::S1, mask); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, RegisterID mask) + { + ASSERT((cond == Zero) || (cond == NonZero)); + m_assembler.tst(reg, mask); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + ASSERT((cond == Zero) || (cond == NonZero)); + ARMWord w = m_assembler.getImm(mask.m_value, ARMRegisters::S0, true); + if (w & ARMAssembler::Op2InvertedImmediate) + m_assembler.bics(ARMRegisters::S0, reg, w & ~ARMAssembler::Op2InvertedImmediate); + else + m_assembler.tst(reg, w); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchTest32(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + load32(address, ARMRegisters::S1); + return branchTest32(cond, ARMRegisters::S1, mask); + } + + Jump branchTest32(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + load32(address, ARMRegisters::S1); + return branchTest32(cond, ARMRegisters::S1, mask); + } + + Jump jump() + { + return Jump(m_assembler.jmp()); + } + + void jump(RegisterID target) + { + m_assembler.bx(target); + } + + void jump(Address address) + { + load32(address, ARMRegisters::pc); + } + + void jump(AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), ARMRegisters::S0); + load32(Address(ARMRegisters::S0, 0), ARMRegisters::pc); + } + + void moveDoubleToInts(FPRegisterID src, RegisterID dest1, RegisterID dest2) + { + m_assembler.vmov(dest1, dest2, src); + } + + void moveIntsToDouble(RegisterID src1, RegisterID src2, FPRegisterID dest, FPRegisterID) + { + m_assembler.vmov(dest, src1, src2); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + add32(src, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchAdd32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + add32(op1, op2, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + add32(imm, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + add32(src, imm, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, AbsoluteAddress dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + add32(imm, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + void mull32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op2 == dest) { + if (op1 == dest) { + move(op2, ARMRegisters::S0); + op2 = ARMRegisters::S0; + } else { + // Swap the operands. + RegisterID tmp = op1; + op1 = op2; + op2 = tmp; + } + } + m_assembler.mull(ARMRegisters::S1, dest, op1, op2); + m_assembler.cmp(ARMRegisters::S1, m_assembler.asr(dest, 31)); + } + + Jump branchMul32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + mull32(src1, src2, dest); + cond = NonZero; + } + else + mul32(src1, src2, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchMul32(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchMul32(cond, src, dest, dest); + } + + Jump branchMul32(ResultCondition cond, TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + move(imm, ARMRegisters::S0); + mull32(ARMRegisters::S0, src, dest); + cond = NonZero; + } + else + mul32(imm, src, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + sub32(src, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + sub32(imm, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + sub32(src, imm, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchSub32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + m_assembler.subs(dest, op1, op2); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchNeg32(ResultCondition cond, RegisterID srcDest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + neg32(srcDest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + Jump branchOr32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Signed) || (cond == Zero) || (cond == NonZero)); + or32(src, dest); + return Jump(m_assembler.jmp(ARMCondition(cond))); + } + + PatchableJump patchableBranch32(RelationalCondition cond, RegisterID reg, TrustedImm32 imm) + { + internalCompare32(reg, imm); + Jump jump(m_assembler.loadBranchTarget(ARMRegisters::S1, ARMCondition(cond), true)); + m_assembler.bx(ARMRegisters::S1, ARMCondition(cond)); + return PatchableJump(jump); + } + + void breakpoint() + { + m_assembler.bkpt(0); + } + + Call nearCall() + { + m_assembler.loadBranchTarget(ARMRegisters::S1, ARMAssembler::AL, true); + return Call(m_assembler.blx(ARMRegisters::S1), Call::LinkableNear); + } + + Call call(RegisterID target) + { + return Call(m_assembler.blx(target), Call::None); + } + + void call(Address address) + { + call32(address.base, address.offset); + } + + void ret() + { + m_assembler.bx(linkRegister); + } + + void compare32(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + m_assembler.cmp(left, right); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(0)); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(1), ARMCondition(cond)); + } + + void compare32(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + m_assembler.cmp(left, m_assembler.getImm(right.m_value, ARMRegisters::S0)); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(0)); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(1), ARMCondition(cond)); + } + + void compare8(RelationalCondition cond, Address left, TrustedImm32 right, RegisterID dest) + { + load8(left, ARMRegisters::S1); + compare32(cond, ARMRegisters::S1, right, dest); + } + + void test32(ResultCondition cond, RegisterID reg, TrustedImm32 mask, RegisterID dest) + { + if (mask.m_value == -1) + m_assembler.cmp(0, reg); + else + m_assembler.tst(reg, m_assembler.getImm(mask.m_value, ARMRegisters::S0)); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(0)); + m_assembler.mov(dest, ARMAssembler::getOp2Byte(1), ARMCondition(cond)); + } + + void test32(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + load32(address, ARMRegisters::S1); + test32(cond, ARMRegisters::S1, mask, dest); + } + + void test8(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + load8(address, ARMRegisters::S1); + test32(cond, ARMRegisters::S1, mask, dest); + } + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.add(dest, src, m_assembler.getImm(imm.m_value, ARMRegisters::S0)); + } + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + load32(address.m_ptr, ARMRegisters::S1); + add32(imm, ARMRegisters::S1); + store32(ARMRegisters::S1, address.m_ptr); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + ARMWord tmp; + + move(TrustedImmPtr(address.m_ptr), ARMRegisters::S1); + m_assembler.dtrUp(ARMAssembler::LoadUint32, ARMRegisters::S0, ARMRegisters::S1, 0); + + if ((tmp = ARMAssembler::getOp2(imm.m_value)) != ARMAssembler::InvalidImmediate) + m_assembler.adds(ARMRegisters::S0, ARMRegisters::S0, tmp); + else if ((tmp = ARMAssembler::getOp2(-imm.m_value)) != ARMAssembler::InvalidImmediate) + m_assembler.subs(ARMRegisters::S0, ARMRegisters::S0, tmp); + else { + m_assembler.adds(ARMRegisters::S0, ARMRegisters::S0, m_assembler.getImm(imm.m_value, ARMRegisters::S1)); + move(TrustedImmPtr(address.m_ptr), ARMRegisters::S1); + } + m_assembler.dtrUp(ARMAssembler::StoreUint32, ARMRegisters::S0, ARMRegisters::S1, 0); + + m_assembler.dtrUp(ARMAssembler::LoadUint32, ARMRegisters::S0, ARMRegisters::S1, sizeof(ARMWord)); + if (imm.m_value >= 0) + m_assembler.adc(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Byte(0)); + else + m_assembler.sbc(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Byte(0)); + m_assembler.dtrUp(ARMAssembler::StoreUint32, ARMRegisters::S0, ARMRegisters::S1, sizeof(ARMWord)); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + load32(address.m_ptr, ARMRegisters::S1); + sub32(imm, ARMRegisters::S1); + store32(ARMRegisters::S1, address.m_ptr); + } + + void load32(const void* address, RegisterID dest) + { + m_assembler.ldrUniqueImmediate(ARMRegisters::S0, reinterpret_cast<ARMWord>(address)); + m_assembler.dtrUp(ARMAssembler::LoadUint32, dest, ARMRegisters::S0, 0); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + load32(left.m_ptr, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, TrustedImm32 right) + { + load32(left.m_ptr, ARMRegisters::S1); + return branch32(cond, ARMRegisters::S1, right); + } + + void relativeTableJump(RegisterID index, int scale) + { + ASSERT(scale >= 0 && scale <= 31); + m_assembler.add(ARMRegisters::pc, ARMRegisters::pc, m_assembler.lsl(index, scale)); + + // NOP the default prefetching + m_assembler.mov(ARMRegisters::r0, ARMRegisters::r0); + } + + Call call() + { + ensureSpace(2 * sizeof(ARMWord), sizeof(ARMWord)); + m_assembler.loadBranchTarget(ARMRegisters::S1, ARMAssembler::AL, true); + return Call(m_assembler.blx(ARMRegisters::S1), Call::Linkable); + } + + Call tailRecursiveCall() + { + return Call::fromTailJump(jump()); + } + + Call makeTailRecursiveCall(Jump oldJump) + { + return Call::fromTailJump(oldJump); + } + + DataLabelPtr moveWithPatch(TrustedImmPtr initialValue, RegisterID dest) + { + DataLabelPtr dataLabel(this); + m_assembler.ldrUniqueImmediate(dest, reinterpret_cast<ARMWord>(initialValue.m_value)); + return dataLabel; + } + + Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + ensureSpace(3 * sizeof(ARMWord), 2 * sizeof(ARMWord)); + dataLabel = moveWithPatch(initialRightValue, ARMRegisters::S1); + Jump jump = branch32(cond, left, ARMRegisters::S1, true); + return jump; + } + + Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + load32(left, ARMRegisters::S1); + ensureSpace(3 * sizeof(ARMWord), 2 * sizeof(ARMWord)); + dataLabel = moveWithPatch(initialRightValue, ARMRegisters::S0); + Jump jump = branch32(cond, ARMRegisters::S0, ARMRegisters::S1, true); + return jump; + } + + DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + DataLabelPtr dataLabel = moveWithPatch(initialValue, ARMRegisters::S1); + store32(ARMRegisters::S1, address); + return dataLabel; + } + + DataLabelPtr storePtrWithPatch(ImplicitAddress address) + { + return storePtrWithPatch(TrustedImmPtr(0), address); + } + + // Floating point operators + static bool supportsFloatingPoint() + { + return s_isVFPPresent; + } + + static bool supportsFloatingPointTruncate() + { + return false; + } + + static bool supportsFloatingPointSqrt() + { + return s_isVFPPresent; + } + static bool supportsFloatingPointAbs() { return false; } + + void loadFloat(BaseIndex address, FPRegisterID dest) + { + m_assembler.baseIndexTransferFloat(ARMAssembler::LoadFloat, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void loadDouble(ImplicitAddress address, FPRegisterID dest) + { + m_assembler.dataTransferFloat(ARMAssembler::LoadDouble, dest, address.base, address.offset); + } + + void loadDouble(BaseIndex address, FPRegisterID dest) + { + m_assembler.baseIndexTransferFloat(ARMAssembler::LoadDouble, dest, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void loadDouble(const void* address, FPRegisterID dest) + { + move(TrustedImm32(reinterpret_cast<ARMWord>(address)), ARMRegisters::S0); + m_assembler.doubleDtrUp(ARMAssembler::LoadDouble, dest, ARMRegisters::S0, 0); + } + + void storeFloat(FPRegisterID src, BaseIndex address) + { + m_assembler.baseIndexTransferFloat(ARMAssembler::StoreFloat, src, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void storeDouble(FPRegisterID src, ImplicitAddress address) + { + m_assembler.dataTransferFloat(ARMAssembler::StoreDouble, src, address.base, address.offset); + } + + void storeDouble(FPRegisterID src, BaseIndex address) + { + m_assembler.baseIndexTransferFloat(ARMAssembler::StoreDouble, src, address.base, address.index, static_cast<int>(address.scale), address.offset); + } + + void storeDouble(FPRegisterID src, const void* address) + { + move(TrustedImm32(reinterpret_cast<ARMWord>(address)), ARMRegisters::S0); + m_assembler.dataTransferFloat(ARMAssembler::StoreDouble, src, ARMRegisters::S0, 0); + } + + void moveDouble(FPRegisterID src, FPRegisterID dest) + { + if (src != dest) + m_assembler.vmov_f64(dest, src); + } + + void addDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vadd_f64(dest, dest, src); + } + + void addDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vadd_f64(dest, op1, op2); + } + + void addDouble(Address src, FPRegisterID dest) + { + loadDouble(src, ARMRegisters::SD0); + addDouble(ARMRegisters::SD0, dest); + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + loadDouble(address.m_ptr, ARMRegisters::SD0); + addDouble(ARMRegisters::SD0, dest); + } + + void divDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vdiv_f64(dest, dest, src); + } + + void divDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vdiv_f64(dest, op1, op2); + } + + void divDouble(Address src, FPRegisterID dest) + { + RELEASE_ASSERT_NOT_REACHED(); // Untested + loadDouble(src, ARMRegisters::SD0); + divDouble(ARMRegisters::SD0, dest); + } + + void subDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vsub_f64(dest, dest, src); + } + + void subDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vsub_f64(dest, op1, op2); + } + + void subDouble(Address src, FPRegisterID dest) + { + loadDouble(src, ARMRegisters::SD0); + subDouble(ARMRegisters::SD0, dest); + } + + void mulDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vmul_f64(dest, dest, src); + } + + void mulDouble(Address src, FPRegisterID dest) + { + loadDouble(src, ARMRegisters::SD0); + mulDouble(ARMRegisters::SD0, dest); + } + + void mulDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vmul_f64(dest, op1, op2); + } + + void sqrtDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vsqrt_f64(dest, src); + } + + void absDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vabs_f64(dest, src); + } + + void negateDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vneg_f64(dest, src); + } + + void convertInt32ToDouble(RegisterID src, FPRegisterID dest) + { + m_assembler.vmov_vfp32(dest << 1, src); + m_assembler.vcvt_f64_s32(dest, dest << 1); + } + + void convertInt32ToDouble(Address src, FPRegisterID dest) + { + load32(src, ARMRegisters::S1); + convertInt32ToDouble(ARMRegisters::S1, dest); + } + + void convertInt32ToDouble(AbsoluteAddress src, FPRegisterID dest) + { + move(TrustedImmPtr(src.m_ptr), ARMRegisters::S1); + load32(Address(ARMRegisters::S1), ARMRegisters::S1); + convertInt32ToDouble(ARMRegisters::S1, dest); + } + + void convertFloatToDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.vcvt_f64_f32(dst, src); + } + + void convertDoubleToFloat(FPRegisterID src, FPRegisterID dst) + { + m_assembler.vcvt_f32_f64(dst, src); + } + + Jump branchDouble(DoubleCondition cond, FPRegisterID left, FPRegisterID right) + { + m_assembler.vcmp_f64(left, right); + m_assembler.vmrs_apsr(); + if (cond & DoubleConditionBitSpecial) + m_assembler.cmp(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::VS); + return Jump(m_assembler.jmp(static_cast<ARMAssembler::Condition>(cond & ~DoubleConditionMask))); + } + + // Truncates 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, INT_MIN). + enum BranchTruncateType { BranchIfTruncateFailed, BranchIfTruncateSuccessful }; + Jump branchTruncateDoubleToInt32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + truncateDoubleToInt32(src, dest); + + m_assembler.add(ARMRegisters::S0, dest, ARMAssembler::getOp2Byte(1)); + m_assembler.bic(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Byte(1)); + + ARMWord w = ARMAssembler::getOp2(0x80000000); + ASSERT(w != ARMAssembler::InvalidImmediate); + m_assembler.cmp(ARMRegisters::S0, w); + return Jump(m_assembler.jmp(branchType == BranchIfTruncateFailed ? ARMAssembler::EQ : ARMAssembler::NE)); + } + + Jump branchTruncateDoubleToUint32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + truncateDoubleToUint32(src, dest); + + m_assembler.add(ARMRegisters::S0, dest, ARMAssembler::getOp2Byte(1)); + m_assembler.bic(ARMRegisters::S0, ARMRegisters::S0, ARMAssembler::getOp2Byte(1)); + + m_assembler.cmp(ARMRegisters::S0, ARMAssembler::getOp2Byte(0)); + return Jump(m_assembler.jmp(branchType == BranchIfTruncateFailed ? ARMAssembler::EQ : ARMAssembler::NE)); + } + + // Result is undefined if the value is outside of the integer range. + void truncateDoubleToInt32(FPRegisterID src, RegisterID dest) + { + m_assembler.vcvt_s32_f64(ARMRegisters::SD0 << 1, src); + m_assembler.vmov_arm32(dest, ARMRegisters::SD0 << 1); + } + + void truncateDoubleToUint32(FPRegisterID src, RegisterID dest) + { + m_assembler.vcvt_u32_f64(ARMRegisters::SD0 << 1, src); + m_assembler.vmov_arm32(dest, ARMRegisters::SD0 << 1); + } + + // Convert 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, 0). + void branchConvertDoubleToInt32(FPRegisterID src, RegisterID dest, JumpList& failureCases, FPRegisterID) + { + m_assembler.vcvt_s32_f64(ARMRegisters::SD0 << 1, src); + m_assembler.vmov_arm32(dest, ARMRegisters::SD0 << 1); + + // Convert the integer result back to float & compare to the original value - if not equal or unordered (NaN) then jump. + m_assembler.vcvt_f64_s32(ARMRegisters::SD0, ARMRegisters::SD0 << 1); + failureCases.append(branchDouble(DoubleNotEqualOrUnordered, src, ARMRegisters::SD0)); + + // If the result is zero, it might have been -0.0, and 0.0 equals to -0.0 + failureCases.append(branchTest32(Zero, dest)); + } + + Jump branchDoubleNonZero(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.mov(ARMRegisters::S0, ARMAssembler::getOp2Byte(0)); + convertInt32ToDouble(ARMRegisters::S0, scratch); + return branchDouble(DoubleNotEqual, reg, scratch); + } + + Jump branchDoubleZeroOrNaN(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.mov(ARMRegisters::S0, ARMAssembler::getOp2Byte(0)); + convertInt32ToDouble(ARMRegisters::S0, scratch); + return branchDouble(DoubleEqualOrUnordered, reg, scratch); + } + + // Invert a relational condition, e.g. == becomes !=, < becomes >=, etc. + static RelationalCondition invert(RelationalCondition cond) + { + ASSERT((static_cast<uint32_t>(cond & 0x0fffffff)) == 0 && static_cast<uint32_t>(cond) < static_cast<uint32_t>(ARMAssembler::AL)); + return static_cast<RelationalCondition>(cond ^ 0x10000000); + } + + void nop() + { + m_assembler.nop(); + } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + return FunctionPtr(reinterpret_cast<void(*)()>(ARMAssembler::readCallTarget(call.dataLocation()))); + } + + static void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + ARMAssembler::replaceWithJump(instructionStart.dataLocation(), destination.dataLocation()); + } + + static ptrdiff_t maxJumpReplacementSize() + { + ARMAssembler::maxJumpReplacementSize(); + return 0; + } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return false; } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr) + { + UNREACHABLE_FOR_PLATFORM(); + return CodeLocationLabel(); + } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + return label.labelAtOffset(0); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID reg, void* initialValue) + { + ARMAssembler::revertBranchPtrWithPatch(instructionStart.dataLocation(), reg, reinterpret_cast<uintptr_t>(initialValue) & 0xffff); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel, Address, void*) + { + UNREACHABLE_FOR_PLATFORM(); + } + +protected: + ARMAssembler::Condition ARMCondition(RelationalCondition cond) + { + return static_cast<ARMAssembler::Condition>(cond); + } + + ARMAssembler::Condition ARMCondition(ResultCondition cond) + { + return static_cast<ARMAssembler::Condition>(cond); + } + + void ensureSpace(int insnSpace, int constSpace) + { + m_assembler.ensureSpace(insnSpace, constSpace); + } + + int sizeOfConstantPool() + { + return m_assembler.sizeOfConstantPool(); + } + + void call32(RegisterID base, int32_t offset) + { + load32(Address(base, offset), ARMRegisters::S1); + m_assembler.blx(ARMRegisters::S1); + } + +private: + friend class LinkBuffer; + friend class RepatchBuffer; + + void internalCompare32(RegisterID left, TrustedImm32 right) + { + ARMWord tmp = (static_cast<unsigned>(right.m_value) == 0x80000000) ? ARMAssembler::InvalidImmediate : m_assembler.getOp2(-right.m_value); + if (tmp != ARMAssembler::InvalidImmediate) + m_assembler.cmn(left, tmp); + else + m_assembler.cmp(left, m_assembler.getImm(right.m_value, ARMRegisters::S0)); + } + + static void linkCall(void* code, Call call, FunctionPtr function) + { + ARMAssembler::linkCall(code, call.m_label, function.value()); + } + + static void repatchCall(CodeLocationCall call, CodeLocationLabel destination) + { + ARMAssembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + static void repatchCall(CodeLocationCall call, FunctionPtr destination) + { + ARMAssembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + static const bool s_isVFPPresent; +}; + +} + +#endif // ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL) + +#endif // MacroAssemblerARM_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerARMv7.h b/src/3rdparty/masm/assembler/MacroAssemblerARMv7.h new file mode 100644 index 0000000000..81c1d7e08a --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerARMv7.h @@ -0,0 +1,1914 @@ +/* + * Copyright (C) 2009, 2010 Apple Inc. All rights reserved. + * Copyright (C) 2010 University of Szeged + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerARMv7_h +#define MacroAssemblerARMv7_h + +#if ENABLE(ASSEMBLER) + +#include "ARMv7Assembler.h" +#include "AbstractMacroAssembler.h" + +namespace JSC { + +class MacroAssemblerARMv7 : public AbstractMacroAssembler<ARMv7Assembler> { + // FIXME: switch dataTempRegister & addressTempRegister, or possibly use r7? + // - dTR is likely used more than aTR, and we'll get better instruction + // encoding if it's in the low 8 registers. + static const RegisterID dataTempRegister = ARMRegisters::ip; + static const RegisterID addressTempRegister = ARMRegisters::r3; + + static const ARMRegisters::FPDoubleRegisterID fpTempRegister = ARMRegisters::d7; + inline ARMRegisters::FPSingleRegisterID fpTempRegisterAsSingle() { return ARMRegisters::asSingle(fpTempRegister); } + +public: + MacroAssemblerARMv7() + : m_makeJumpPatchable(false) + { + } + + typedef ARMv7Assembler::LinkRecord LinkRecord; + typedef ARMv7Assembler::JumpType JumpType; + typedef ARMv7Assembler::JumpLinkType JumpLinkType; + + static bool isCompactPtrAlignedAddressOffset(ptrdiff_t value) + { + return value >= -255 && value <= 255; + } + + Vector<LinkRecord, 0, UnsafeVectorOverflow>& jumpsToLink() { return m_assembler.jumpsToLink(); } + void* unlinkedCode() { return m_assembler.unlinkedCode(); } + bool canCompact(JumpType jumpType) { return m_assembler.canCompact(jumpType); } + JumpLinkType computeJumpType(JumpType jumpType, const uint8_t* from, const uint8_t* to) { return m_assembler.computeJumpType(jumpType, from, to); } + JumpLinkType computeJumpType(LinkRecord& record, const uint8_t* from, const uint8_t* to) { return m_assembler.computeJumpType(record, from, to); } + void recordLinkOffsets(int32_t regionStart, int32_t regionEnd, int32_t offset) {return m_assembler.recordLinkOffsets(regionStart, regionEnd, offset); } + int jumpSizeDelta(JumpType jumpType, JumpLinkType jumpLinkType) { return m_assembler.jumpSizeDelta(jumpType, jumpLinkType); } + void link(LinkRecord& record, uint8_t* from, uint8_t* to) { return m_assembler.link(record, from, to); } + + struct ArmAddress { + enum AddressType { + HasOffset, + HasIndex, + } type; + RegisterID base; + union { + int32_t offset; + struct { + RegisterID index; + Scale scale; + }; + } u; + + explicit ArmAddress(RegisterID base, int32_t offset = 0) + : type(HasOffset) + , base(base) + { + u.offset = offset; + } + + explicit ArmAddress(RegisterID base, RegisterID index, Scale scale = TimesOne) + : type(HasIndex) + , base(base) + { + u.index = index; + u.scale = scale; + } + }; + +public: + typedef ARMRegisters::FPDoubleRegisterID FPRegisterID; + + static const Scale ScalePtr = TimesFour; + + enum RelationalCondition { + Equal = ARMv7Assembler::ConditionEQ, + NotEqual = ARMv7Assembler::ConditionNE, + Above = ARMv7Assembler::ConditionHI, + AboveOrEqual = ARMv7Assembler::ConditionHS, + Below = ARMv7Assembler::ConditionLO, + BelowOrEqual = ARMv7Assembler::ConditionLS, + GreaterThan = ARMv7Assembler::ConditionGT, + GreaterThanOrEqual = ARMv7Assembler::ConditionGE, + LessThan = ARMv7Assembler::ConditionLT, + LessThanOrEqual = ARMv7Assembler::ConditionLE + }; + + enum ResultCondition { + Overflow = ARMv7Assembler::ConditionVS, + Signed = ARMv7Assembler::ConditionMI, + Zero = ARMv7Assembler::ConditionEQ, + NonZero = ARMv7Assembler::ConditionNE + }; + + enum DoubleCondition { + // These conditions will only evaluate to true if the comparison is ordered - i.e. neither operand is NaN. + DoubleEqual = ARMv7Assembler::ConditionEQ, + DoubleNotEqual = ARMv7Assembler::ConditionVC, // Not the right flag! check for this & handle differently. + DoubleGreaterThan = ARMv7Assembler::ConditionGT, + DoubleGreaterThanOrEqual = ARMv7Assembler::ConditionGE, + DoubleLessThan = ARMv7Assembler::ConditionLO, + DoubleLessThanOrEqual = ARMv7Assembler::ConditionLS, + // If either operand is NaN, these conditions always evaluate to true. + DoubleEqualOrUnordered = ARMv7Assembler::ConditionVS, // Not the right flag! check for this & handle differently. + DoubleNotEqualOrUnordered = ARMv7Assembler::ConditionNE, + DoubleGreaterThanOrUnordered = ARMv7Assembler::ConditionHI, + DoubleGreaterThanOrEqualOrUnordered = ARMv7Assembler::ConditionHS, + DoubleLessThanOrUnordered = ARMv7Assembler::ConditionLT, + DoubleLessThanOrEqualOrUnordered = ARMv7Assembler::ConditionLE, + }; + + static const RegisterID stackPointerRegister = ARMRegisters::sp; + static const RegisterID linkRegister = ARMRegisters::lr; + + // Integer arithmetic operations: + // + // Operations are typically two operand - operation(source, srcDst) + // For many operations the source may be an TrustedImm32, the srcDst operand + // may often be a memory location (explictly described using an Address + // object). + + void add32(RegisterID src, RegisterID dest) + { + m_assembler.add(dest, dest, src); + } + + void add32(TrustedImm32 imm, RegisterID dest) + { + add32(imm, dest, dest); + } + + void add32(AbsoluteAddress src, RegisterID dest) + { + load32(src.m_ptr, dataTempRegister); + add32(dataTempRegister, dest); + } + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add(dest, src, armImm); + else { + move(imm, dataTempRegister); + m_assembler.add(dest, src, dataTempRegister); + } + } + + void add32(TrustedImm32 imm, Address address) + { + load32(address, dataTempRegister); + + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add(dataTempRegister, dataTempRegister, armImm); + else { + // Hrrrm, since dataTempRegister holds the data loaded, + // use addressTempRegister to hold the immediate. + move(imm, addressTempRegister); + m_assembler.add(dataTempRegister, dataTempRegister, addressTempRegister); + } + + store32(dataTempRegister, address); + } + + void add32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + add32(dataTempRegister, dest); + } + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + load32(address.m_ptr, dataTempRegister); + + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add(dataTempRegister, dataTempRegister, armImm); + else { + // Hrrrm, since dataTempRegister holds the data loaded, + // use addressTempRegister to hold the immediate. + move(imm, addressTempRegister); + m_assembler.add(dataTempRegister, dataTempRegister, addressTempRegister); + } + + store32(dataTempRegister, address.m_ptr); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), addressTempRegister); + + m_assembler.ldr(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt12(0)); + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add_S(dataTempRegister, dataTempRegister, armImm); + else { + move(imm, addressTempRegister); + m_assembler.add_S(dataTempRegister, dataTempRegister, addressTempRegister); + move(TrustedImmPtr(address.m_ptr), addressTempRegister); + } + m_assembler.str(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt12(0)); + + m_assembler.ldr(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt12(4)); + m_assembler.adc(dataTempRegister, dataTempRegister, ARMThumbImmediate::makeEncodedImm(imm.m_value >> 31)); + m_assembler.str(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt12(4)); + } + + void and32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.ARM_and(dest, op1, op2); + } + + void and32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.ARM_and(dest, src, armImm); + else { + move(imm, dataTempRegister); + m_assembler.ARM_and(dest, src, dataTempRegister); + } + } + + void and32(RegisterID src, RegisterID dest) + { + and32(dest, src, dest); + } + + void and32(TrustedImm32 imm, RegisterID dest) + { + and32(imm, dest, dest); + } + + void and32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + and32(dataTempRegister, dest); + } + + void countLeadingZeros32(RegisterID src, RegisterID dest) + { + m_assembler.clz(dest, src); + } + + void lshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + // Clamp the shift to the range 0..31 + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(0x1f); + ASSERT(armImm.isValid()); + m_assembler.ARM_and(dataTempRegister, shiftAmount, armImm); + + m_assembler.lsl(dest, src, dataTempRegister); + } + + void lshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.lsl(dest, src, imm.m_value & 0x1f); + } + + void lshift32(RegisterID shiftAmount, RegisterID dest) + { + lshift32(dest, shiftAmount, dest); + } + + void lshift32(TrustedImm32 imm, RegisterID dest) + { + lshift32(dest, imm, dest); + } + + void mul32(RegisterID src, RegisterID dest) + { + m_assembler.smull(dest, dataTempRegister, dest, src); + } + + void mul32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(imm, dataTempRegister); + m_assembler.smull(dest, dataTempRegister, src, dataTempRegister); + } + + void neg32(RegisterID srcDest) + { + m_assembler.neg(srcDest, srcDest); + } + + void or32(RegisterID src, RegisterID dest) + { + m_assembler.orr(dest, dest, src); + } + + void or32(RegisterID src, AbsoluteAddress dest) + { + move(TrustedImmPtr(dest.m_ptr), addressTempRegister); + load32(addressTempRegister, dataTempRegister); + or32(src, dataTempRegister); + store32(dataTempRegister, addressTempRegister); + } + + void or32(TrustedImm32 imm, RegisterID dest) + { + or32(imm, dest, dest); + } + + void or32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.orr(dest, op1, op2); + } + + void or32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.orr(dest, src, armImm); + else { + move(imm, dataTempRegister); + m_assembler.orr(dest, src, dataTempRegister); + } + } + + void rshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + // Clamp the shift to the range 0..31 + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(0x1f); + ASSERT(armImm.isValid()); + m_assembler.ARM_and(dataTempRegister, shiftAmount, armImm); + + m_assembler.asr(dest, src, dataTempRegister); + } + + void rshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.asr(dest, src, imm.m_value & 0x1f); + } + + void rshift32(RegisterID shiftAmount, RegisterID dest) + { + rshift32(dest, shiftAmount, dest); + } + + void rshift32(TrustedImm32 imm, RegisterID dest) + { + rshift32(dest, imm, dest); + } + + void urshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + // Clamp the shift to the range 0..31 + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(0x1f); + ASSERT(armImm.isValid()); + m_assembler.ARM_and(dataTempRegister, shiftAmount, armImm); + + m_assembler.lsr(dest, src, dataTempRegister); + } + + void urshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.lsr(dest, src, imm.m_value & 0x1f); + } + + void urshift32(RegisterID shiftAmount, RegisterID dest) + { + urshift32(dest, shiftAmount, dest); + } + + void urshift32(TrustedImm32 imm, RegisterID dest) + { + urshift32(dest, imm, dest); + } + + void sub32(RegisterID src, RegisterID dest) + { + m_assembler.sub(dest, dest, src); + } + + void sub32(TrustedImm32 imm, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.sub(dest, dest, armImm); + else { + move(imm, dataTempRegister); + m_assembler.sub(dest, dest, dataTempRegister); + } + } + + void sub32(TrustedImm32 imm, Address address) + { + load32(address, dataTempRegister); + + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.sub(dataTempRegister, dataTempRegister, armImm); + else { + // Hrrrm, since dataTempRegister holds the data loaded, + // use addressTempRegister to hold the immediate. + move(imm, addressTempRegister); + m_assembler.sub(dataTempRegister, dataTempRegister, addressTempRegister); + } + + store32(dataTempRegister, address); + } + + void sub32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + sub32(dataTempRegister, dest); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + load32(address.m_ptr, dataTempRegister); + + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12OrEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.sub(dataTempRegister, dataTempRegister, armImm); + else { + // Hrrrm, since dataTempRegister holds the data loaded, + // use addressTempRegister to hold the immediate. + move(imm, addressTempRegister); + m_assembler.sub(dataTempRegister, dataTempRegister, addressTempRegister); + } + + store32(dataTempRegister, address.m_ptr); + } + + void xor32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.eor(dest, op1, op2); + } + + void xor32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (imm.m_value == -1) { + m_assembler.mvn(dest, src); + return; + } + + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.eor(dest, src, armImm); + else { + move(imm, dataTempRegister); + m_assembler.eor(dest, src, dataTempRegister); + } + } + + void xor32(RegisterID src, RegisterID dest) + { + xor32(dest, src, dest); + } + + void xor32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value == -1) + m_assembler.mvn(dest, dest); + else + xor32(imm, dest, dest); + } + + + // Memory access operations: + // + // Loads are of the form load(address, destination) and stores of the form + // store(source, address). The source for a store may be an TrustedImm32. Address + // operand objects to loads and store will be implicitly constructed if a + // register is passed. + +private: + void load32(ArmAddress address, RegisterID dest) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.ldr(dest, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.ldr(dest, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.ldr(dest, address.base, address.u.offset, true, false); + } + } + + void load16(ArmAddress address, RegisterID dest) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.ldrh(dest, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.ldrh(dest, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.ldrh(dest, address.base, address.u.offset, true, false); + } + } + + void load16Signed(ArmAddress address, RegisterID dest) + { + ASSERT(address.type == ArmAddress::HasIndex); + m_assembler.ldrsh(dest, address.base, address.u.index, address.u.scale); + } + + void load8(ArmAddress address, RegisterID dest) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.ldrb(dest, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.ldrb(dest, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.ldrb(dest, address.base, address.u.offset, true, false); + } + } + + void load8Signed(ArmAddress address, RegisterID dest) + { + ASSERT(address.type == ArmAddress::HasIndex); + m_assembler.ldrsb(dest, address.base, address.u.index, address.u.scale); + } + +protected: + void store32(RegisterID src, ArmAddress address) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.str(src, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.str(src, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.str(src, address.base, address.u.offset, true, false); + } + } + +private: + void store8(RegisterID src, ArmAddress address) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.strb(src, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.strb(src, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.strb(src, address.base, address.u.offset, true, false); + } + } + + void store16(RegisterID src, ArmAddress address) + { + if (address.type == ArmAddress::HasIndex) + m_assembler.strh(src, address.base, address.u.index, address.u.scale); + else if (address.u.offset >= 0) { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.u.offset); + ASSERT(armImm.isValid()); + m_assembler.strh(src, address.base, armImm); + } else { + ASSERT(address.u.offset >= -255); + m_assembler.strh(src, address.base, address.u.offset, true, false); + } + } + +public: + void load32(ImplicitAddress address, RegisterID dest) + { + load32(setupArmAddress(address), dest); + } + + void load32(BaseIndex address, RegisterID dest) + { + load32(setupArmAddress(address), dest); + } + + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) + { + load32(setupArmAddress(address), dest); + } + + void load16Unaligned(BaseIndex address, RegisterID dest) + { + load16(setupArmAddress(address), dest); + } + + void load32(const void* address, RegisterID dest) + { + move(TrustedImmPtr(address), addressTempRegister); + m_assembler.ldr(dest, addressTempRegister, ARMThumbImmediate::makeUInt16(0)); + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result(this); + ASSERT(address.offset >= 0 && address.offset <= 255); + m_assembler.ldrWide8BitImmediate(dest, address.base, address.offset); + return result; + } + + void load8(ImplicitAddress address, RegisterID dest) + { + load8(setupArmAddress(address), dest); + } + + void load8Signed(ImplicitAddress, RegisterID) + { + UNREACHABLE_FOR_PLATFORM(); + } + + void load8(BaseIndex address, RegisterID dest) + { + load8(setupArmAddress(address), dest); + } + + void load8Signed(BaseIndex address, RegisterID dest) + { + load8Signed(setupArmAddress(address), dest); + } + + DataLabel32 load32WithAddressOffsetPatch(Address address, RegisterID dest) + { + DataLabel32 label = moveWithPatch(TrustedImm32(address.offset), dataTempRegister); + load32(ArmAddress(address.base, dataTempRegister), dest); + return label; + } + + DataLabelCompact load32WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + + RegisterID base = address.base; + + DataLabelCompact label(this); + ASSERT(isCompactPtrAlignedAddressOffset(address.offset)); + + m_assembler.ldr(dest, base, address.offset, true, false); + return label; + } + + void load16(BaseIndex address, RegisterID dest) + { + m_assembler.ldrh(dest, makeBaseIndexBase(address), address.index, address.scale); + } + + void load16Signed(BaseIndex address, RegisterID dest) + { + load16Signed(setupArmAddress(address), dest); + } + + void load16(ImplicitAddress address, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeUInt12(address.offset); + if (armImm.isValid()) + m_assembler.ldrh(dest, address.base, armImm); + else { + move(TrustedImm32(address.offset), dataTempRegister); + m_assembler.ldrh(dest, address.base, dataTempRegister); + } + } + + void load16Signed(ImplicitAddress, RegisterID) + { + UNREACHABLE_FOR_PLATFORM(); + } + + DataLabel32 store32WithAddressOffsetPatch(RegisterID src, Address address) + { + DataLabel32 label = moveWithPatch(TrustedImm32(address.offset), dataTempRegister); + store32(src, ArmAddress(address.base, dataTempRegister)); + return label; + } + + void store32(RegisterID src, ImplicitAddress address) + { + store32(src, setupArmAddress(address)); + } + + void store32(RegisterID src, BaseIndex address) + { + store32(src, setupArmAddress(address)); + } + + void store32(TrustedImm32 imm, ImplicitAddress address) + { + move(imm, dataTempRegister); + store32(dataTempRegister, setupArmAddress(address)); + } + + void store32(TrustedImm32 imm, BaseIndex address) + { + move(imm, dataTempRegister); + store32(dataTempRegister, setupArmAddress(address)); + } + + void store32(RegisterID src, const void* address) + { + move(TrustedImmPtr(address), addressTempRegister); + m_assembler.str(src, addressTempRegister, ARMThumbImmediate::makeUInt16(0)); + } + + void store32(TrustedImm32 imm, const void* address) + { + move(imm, dataTempRegister); + store32(dataTempRegister, address); + } + + void store8(RegisterID src, BaseIndex address) + { + store8(src, setupArmAddress(address)); + } + + void store8(RegisterID src, void* address) + { + move(TrustedImmPtr(address), addressTempRegister); + store8(src, ArmAddress(addressTempRegister, 0)); + } + + void store8(TrustedImm32 imm, void* address) + { + move(imm, dataTempRegister); + store8(dataTempRegister, address); + } + + void store16(RegisterID src, BaseIndex address) + { + store16(src, setupArmAddress(address)); + } + + // Possibly clobbers src, but not on this architecture. + void moveDoubleToInts(FPRegisterID src, RegisterID dest1, RegisterID dest2) + { + m_assembler.vmov(dest1, dest2, src); + } + + void moveIntsToDouble(RegisterID src1, RegisterID src2, FPRegisterID dest, FPRegisterID scratch) + { + UNUSED_PARAM(scratch); + m_assembler.vmov(dest, src1, src2); + } + +#if ENABLE(JIT_CONSTANT_BLINDING) + static bool shouldBlindForSpecificArch(uint32_t value) + { + ARMThumbImmediate immediate = ARMThumbImmediate::makeEncodedImm(value); + + // Couldn't be encoded as an immediate, so assume it's untrusted. + if (!immediate.isValid()) + return true; + + // If we can encode the immediate, we have less than 16 attacker + // controlled bits. + if (immediate.isEncodedImm()) + return false; + + // Don't let any more than 12 bits of an instruction word + // be controlled by an attacker. + return !immediate.isUInt12(); + } +#endif + + // Floating-point operations: + + static bool supportsFloatingPoint() { return true; } + static bool supportsFloatingPointTruncate() { return true; } + static bool supportsFloatingPointSqrt() { return true; } + static bool supportsFloatingPointAbs() { return true; } + + void loadDouble(ImplicitAddress address, FPRegisterID dest) + { + RegisterID base = address.base; + int32_t offset = address.offset; + + // Arm vfp addresses can be offset by a 9-bit ones-comp immediate, left shifted by 2. + if ((offset & 3) || (offset > (255 * 4)) || (offset < -(255 * 4))) { + add32(TrustedImm32(offset), base, addressTempRegister); + base = addressTempRegister; + offset = 0; + } + + m_assembler.vldr(dest, base, offset); + } + + void loadFloat(ImplicitAddress address, FPRegisterID dest) + { + RegisterID base = address.base; + int32_t offset = address.offset; + + // Arm vfp addresses can be offset by a 9-bit ones-comp immediate, left shifted by 2. + if ((offset & 3) || (offset > (255 * 4)) || (offset < -(255 * 4))) { + add32(TrustedImm32(offset), base, addressTempRegister); + base = addressTempRegister; + offset = 0; + } + + m_assembler.flds(ARMRegisters::asSingle(dest), base, offset); + } + + void loadDouble(BaseIndex address, FPRegisterID dest) + { + move(address.index, addressTempRegister); + lshift32(TrustedImm32(address.scale), addressTempRegister); + add32(address.base, addressTempRegister); + loadDouble(Address(addressTempRegister, address.offset), dest); + } + + void loadFloat(BaseIndex address, FPRegisterID dest) + { + move(address.index, addressTempRegister); + lshift32(TrustedImm32(address.scale), addressTempRegister); + add32(address.base, addressTempRegister); + loadFloat(Address(addressTempRegister, address.offset), dest); + } + + void moveDouble(FPRegisterID src, FPRegisterID dest) + { + if (src != dest) + m_assembler.vmov(dest, src); + } + + void loadDouble(const void* address, FPRegisterID dest) + { + move(TrustedImmPtr(address), addressTempRegister); + m_assembler.vldr(dest, addressTempRegister, 0); + } + + void storeDouble(FPRegisterID src, ImplicitAddress address) + { + RegisterID base = address.base; + int32_t offset = address.offset; + + // Arm vfp addresses can be offset by a 9-bit ones-comp immediate, left shifted by 2. + if ((offset & 3) || (offset > (255 * 4)) || (offset < -(255 * 4))) { + add32(TrustedImm32(offset), base, addressTempRegister); + base = addressTempRegister; + offset = 0; + } + + m_assembler.vstr(src, base, offset); + } + + void storeFloat(FPRegisterID src, ImplicitAddress address) + { + RegisterID base = address.base; + int32_t offset = address.offset; + + // Arm vfp addresses can be offset by a 9-bit ones-comp immediate, left shifted by 2. + if ((offset & 3) || (offset > (255 * 4)) || (offset < -(255 * 4))) { + add32(TrustedImm32(offset), base, addressTempRegister); + base = addressTempRegister; + offset = 0; + } + + m_assembler.fsts(ARMRegisters::asSingle(src), base, offset); + } + + void storeDouble(FPRegisterID src, const void* address) + { + move(TrustedImmPtr(address), addressTempRegister); + storeDouble(src, addressTempRegister); + } + + void storeDouble(FPRegisterID src, BaseIndex address) + { + move(address.index, addressTempRegister); + lshift32(TrustedImm32(address.scale), addressTempRegister); + add32(address.base, addressTempRegister); + storeDouble(src, Address(addressTempRegister, address.offset)); + } + + void storeFloat(FPRegisterID src, BaseIndex address) + { + move(address.index, addressTempRegister); + lshift32(TrustedImm32(address.scale), addressTempRegister); + add32(address.base, addressTempRegister); + storeFloat(src, Address(addressTempRegister, address.offset)); + } + + void addDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vadd(dest, dest, src); + } + + void addDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + addDouble(fpTempRegister, dest); + } + + void addDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vadd(dest, op1, op2); + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + loadDouble(address.m_ptr, fpTempRegister); + m_assembler.vadd(dest, dest, fpTempRegister); + } + + void divDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vdiv(dest, dest, src); + } + + void divDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vdiv(dest, op1, op2); + } + + void subDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vsub(dest, dest, src); + } + + void subDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + subDouble(fpTempRegister, dest); + } + + void subDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vsub(dest, op1, op2); + } + + void mulDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vmul(dest, dest, src); + } + + void mulDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + mulDouble(fpTempRegister, dest); + } + + void mulDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.vmul(dest, op1, op2); + } + + void sqrtDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vsqrt(dest, src); + } + + void absDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vabs(dest, src); + } + + void negateDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.vneg(dest, src); + } + + void convertInt32ToDouble(RegisterID src, FPRegisterID dest) + { + m_assembler.vmov(fpTempRegister, src, src); + m_assembler.vcvt_signedToFloatingPoint(dest, fpTempRegisterAsSingle()); + } + + void convertInt32ToDouble(Address address, FPRegisterID dest) + { + // Fixme: load directly into the fpr! + load32(address, dataTempRegister); + m_assembler.vmov(fpTempRegister, dataTempRegister, dataTempRegister); + m_assembler.vcvt_signedToFloatingPoint(dest, fpTempRegisterAsSingle()); + } + + void convertInt32ToDouble(AbsoluteAddress address, FPRegisterID dest) + { + // Fixme: load directly into the fpr! + load32(address.m_ptr, dataTempRegister); + m_assembler.vmov(fpTempRegister, dataTempRegister, dataTempRegister); + m_assembler.vcvt_signedToFloatingPoint(dest, fpTempRegisterAsSingle()); + } + + void convertFloatToDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.vcvtds(dst, ARMRegisters::asSingle(src)); + } + + void convertDoubleToFloat(FPRegisterID src, FPRegisterID dst) + { + m_assembler.vcvtsd(ARMRegisters::asSingle(dst), src); + } + + Jump branchDouble(DoubleCondition cond, FPRegisterID left, FPRegisterID right) + { + m_assembler.vcmp(left, right); + m_assembler.vmrs(); + + if (cond == DoubleNotEqual) { + // ConditionNE jumps if NotEqual *or* unordered - force the unordered cases not to jump. + Jump unordered = makeBranch(ARMv7Assembler::ConditionVS); + Jump result = makeBranch(ARMv7Assembler::ConditionNE); + unordered.link(this); + return result; + } + if (cond == DoubleEqualOrUnordered) { + Jump unordered = makeBranch(ARMv7Assembler::ConditionVS); + Jump notEqual = makeBranch(ARMv7Assembler::ConditionNE); + unordered.link(this); + // We get here if either unordered or equal. + Jump result = jump(); + notEqual.link(this); + return result; + } + return makeBranch(cond); + } + + enum BranchTruncateType { BranchIfTruncateFailed, BranchIfTruncateSuccessful }; + Jump branchTruncateDoubleToInt32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + // Convert into dest. + m_assembler.vcvt_floatingPointToSigned(fpTempRegisterAsSingle(), src); + m_assembler.vmov(dest, fpTempRegisterAsSingle()); + + // Calculate 2x dest. If the value potentially underflowed, it will have + // clamped to 0x80000000, so 2x dest is zero in this case. In the case of + // overflow the result will be equal to -2. + Jump underflow = branchAdd32(Zero, dest, dest, dataTempRegister); + Jump noOverflow = branch32(NotEqual, dataTempRegister, TrustedImm32(-2)); + + // For BranchIfTruncateSuccessful, we branch if 'noOverflow' jumps. + underflow.link(this); + if (branchType == BranchIfTruncateSuccessful) + return noOverflow; + + // We'll reach the current point in the code on failure, so plant a + // jump here & link the success case. + Jump failure = jump(); + noOverflow.link(this); + return failure; + } + + Jump branchTruncateDoubleToUint32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + m_assembler.vcvt_floatingPointToSigned(fpTempRegisterAsSingle(), src); + m_assembler.vmov(dest, fpTempRegisterAsSingle()); + + Jump overflow = branch32(Equal, dest, TrustedImm32(0x7fffffff)); + Jump success = branch32(GreaterThanOrEqual, dest, TrustedImm32(0)); + overflow.link(this); + + if (branchType == BranchIfTruncateSuccessful) + return success; + + Jump failure = jump(); + success.link(this); + return failure; + } + + // Result is undefined if the value is outside of the integer range. + void truncateDoubleToInt32(FPRegisterID src, RegisterID dest) + { + m_assembler.vcvt_floatingPointToSigned(fpTempRegisterAsSingle(), src); + m_assembler.vmov(dest, fpTempRegisterAsSingle()); + } + + void truncateDoubleToUint32(FPRegisterID src, RegisterID dest) + { + m_assembler.vcvt_floatingPointToUnsigned(fpTempRegisterAsSingle(), src); + m_assembler.vmov(dest, fpTempRegisterAsSingle()); + } + + // Convert 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, 0). + void branchConvertDoubleToInt32(FPRegisterID src, RegisterID dest, JumpList& failureCases, FPRegisterID) + { + m_assembler.vcvt_floatingPointToSigned(fpTempRegisterAsSingle(), src); + m_assembler.vmov(dest, fpTempRegisterAsSingle()); + + // Convert the integer result back to float & compare to the original value - if not equal or unordered (NaN) then jump. + m_assembler.vcvt_signedToFloatingPoint(fpTempRegister, fpTempRegisterAsSingle()); + failureCases.append(branchDouble(DoubleNotEqualOrUnordered, src, fpTempRegister)); + + // If the result is zero, it might have been -0.0, and the double comparison won't catch this! + failureCases.append(branchTest32(Zero, dest)); + } + + Jump branchDoubleNonZero(FPRegisterID reg, FPRegisterID) + { + m_assembler.vcmpz(reg); + m_assembler.vmrs(); + Jump unordered = makeBranch(ARMv7Assembler::ConditionVS); + Jump result = makeBranch(ARMv7Assembler::ConditionNE); + unordered.link(this); + return result; + } + + Jump branchDoubleZeroOrNaN(FPRegisterID reg, FPRegisterID) + { + m_assembler.vcmpz(reg); + m_assembler.vmrs(); + Jump unordered = makeBranch(ARMv7Assembler::ConditionVS); + Jump notEqual = makeBranch(ARMv7Assembler::ConditionNE); + unordered.link(this); + // We get here if either unordered or equal. + Jump result = jump(); + notEqual.link(this); + return result; + } + + // Stack manipulation operations: + // + // The ABI is assumed to provide a stack abstraction to memory, + // containing machine word sized units of data. Push and pop + // operations add and remove a single register sized unit of data + // to or from the stack. Peek and poke operations read or write + // values on the stack, without moving the current stack position. + + void pop(RegisterID dest) + { + // store postindexed with writeback + m_assembler.ldr(dest, ARMRegisters::sp, sizeof(void*), false, true); + } + + void push(RegisterID src) + { + // store preindexed with writeback + m_assembler.str(src, ARMRegisters::sp, -sizeof(void*), true, true); + } + + void push(Address address) + { + load32(address, dataTempRegister); + push(dataTempRegister); + } + + void push(TrustedImm32 imm) + { + move(imm, dataTempRegister); + push(dataTempRegister); + } + + // Register move operations: + // + // Move values in registers. + + void move(TrustedImm32 imm, RegisterID dest) + { + uint32_t value = imm.m_value; + + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(value); + + if (armImm.isValid()) + m_assembler.mov(dest, armImm); + else if ((armImm = ARMThumbImmediate::makeEncodedImm(~value)).isValid()) + m_assembler.mvn(dest, armImm); + else { + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(value)); + if (value & 0xffff0000) + m_assembler.movt(dest, ARMThumbImmediate::makeUInt16(value >> 16)); + } + } + + void move(RegisterID src, RegisterID dest) + { + if (src != dest) + m_assembler.mov(dest, src); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + move(TrustedImm32(imm), dest); + } + + void swap(RegisterID reg1, RegisterID reg2) + { + move(reg1, dataTempRegister); + move(reg2, reg1); + move(dataTempRegister, reg2); + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + move(src, dest); + } + + void zeroExtend32ToPtr(RegisterID src, RegisterID dest) + { + move(src, dest); + } + + // Invert a relational condition, e.g. == becomes !=, < becomes >=, etc. + static RelationalCondition invert(RelationalCondition cond) + { + return static_cast<RelationalCondition>(cond ^ 1); + } + + void nop() + { + m_assembler.nop(); + } + + static void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + ARMv7Assembler::replaceWithJump(instructionStart.dataLocation(), destination.dataLocation()); + } + + static ptrdiff_t maxJumpReplacementSize() + { + return ARMv7Assembler::maxJumpReplacementSize(); + } + + // Forwards / external control flow operations: + // + // This set of jump and conditional branch operations return a Jump + // object which may linked at a later point, allow forwards jump, + // or jumps that will require external linkage (after the code has been + // relocated). + // + // For branches, signed <, >, <= and >= are denoted as l, g, le, and ge + // respecitvely, for unsigned comparisons the names b, a, be, and ae are + // used (representing the names 'below' and 'above'). + // + // Operands to the comparision are provided in the expected order, e.g. + // jle32(reg1, TrustedImm32(5)) will branch if the value held in reg1, when + // treated as a signed 32bit value, is less than or equal to 5. + // + // jz and jnz test whether the first operand is equal to zero, and take + // an optional second operand of a mask under which to perform the test. +private: + + // Should we be using TEQ for equal/not-equal? + void compare32(RegisterID left, TrustedImm32 right) + { + int32_t imm = right.m_value; + if (!imm) + m_assembler.tst(left, left); + else { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm); + if (armImm.isValid()) + m_assembler.cmp(left, armImm); + else if ((armImm = ARMThumbImmediate::makeEncodedImm(-imm)).isValid()) + m_assembler.cmn(left, armImm); + else { + move(TrustedImm32(imm), dataTempRegister); + m_assembler.cmp(left, dataTempRegister); + } + } + } + + void test32(RegisterID reg, TrustedImm32 mask) + { + int32_t imm = mask.m_value; + + if (imm == -1) + m_assembler.tst(reg, reg); + else { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm); + if (armImm.isValid()) + m_assembler.tst(reg, armImm); + else { + move(mask, dataTempRegister); + m_assembler.tst(reg, dataTempRegister); + } + } + } + +public: + Jump branch32(RelationalCondition cond, RegisterID left, RegisterID right) + { + m_assembler.cmp(left, right); + return Jump(makeBranch(cond)); + } + + Jump branch32(RelationalCondition cond, RegisterID left, TrustedImm32 right) + { + compare32(left, right); + return Jump(makeBranch(cond)); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Address right) + { + load32(right, dataTempRegister); + return branch32(cond, left, dataTempRegister); + } + + Jump branch32(RelationalCondition cond, Address left, RegisterID right) + { + load32(left, dataTempRegister); + return branch32(cond, dataTempRegister, right); + } + + Jump branch32(RelationalCondition cond, Address left, TrustedImm32 right) + { + // use addressTempRegister incase the branch32 we call uses dataTempRegister. :-/ + load32(left, addressTempRegister); + return branch32(cond, addressTempRegister, right); + } + + Jump branch32(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + // use addressTempRegister incase the branch32 we call uses dataTempRegister. :-/ + load32(left, addressTempRegister); + return branch32(cond, addressTempRegister, right); + } + + Jump branch32WithUnalignedHalfWords(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + // use addressTempRegister incase the branch32 we call uses dataTempRegister. :-/ + load32WithUnalignedHalfWords(left, addressTempRegister); + return branch32(cond, addressTempRegister, right); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + load32(left.m_ptr, dataTempRegister); + return branch32(cond, dataTempRegister, right); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, TrustedImm32 right) + { + // use addressTempRegister incase the branch32 we call uses dataTempRegister. :-/ + load32(left.m_ptr, addressTempRegister); + return branch32(cond, addressTempRegister, right); + } + + Jump branch8(RelationalCondition cond, RegisterID left, TrustedImm32 right) + { + compare32(left, right); + return Jump(makeBranch(cond)); + } + + Jump branch8(RelationalCondition cond, Address left, TrustedImm32 right) + { + ASSERT(!(0xffffff00 & right.m_value)); + // use addressTempRegister incase the branch8 we call uses dataTempRegister. :-/ + load8(left, addressTempRegister); + return branch8(cond, addressTempRegister, right); + } + + Jump branch8(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + ASSERT(!(0xffffff00 & right.m_value)); + // use addressTempRegister incase the branch32 we call uses dataTempRegister. :-/ + load8(left, addressTempRegister); + return branch32(cond, addressTempRegister, right); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, RegisterID mask) + { + m_assembler.tst(reg, mask); + return Jump(makeBranch(cond)); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + test32(reg, mask); + return Jump(makeBranch(cond)); + } + + Jump branchTest32(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + // use addressTempRegister incase the branchTest32 we call uses dataTempRegister. :-/ + load32(address, addressTempRegister); + return branchTest32(cond, addressTempRegister, mask); + } + + Jump branchTest32(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + // use addressTempRegister incase the branchTest32 we call uses dataTempRegister. :-/ + load32(address, addressTempRegister); + return branchTest32(cond, addressTempRegister, mask); + } + + Jump branchTest8(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + // use addressTempRegister incase the branchTest8 we call uses dataTempRegister. :-/ + load8(address, addressTempRegister); + return branchTest32(cond, addressTempRegister, mask); + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + // use addressTempRegister incase the branchTest8 we call uses dataTempRegister. :-/ + move(TrustedImmPtr(address.m_ptr), addressTempRegister); + load8(Address(addressTempRegister), addressTempRegister); + return branchTest32(cond, addressTempRegister, mask); + } + + void jump(RegisterID target) + { + m_assembler.bx(target); + } + + // Address is a memory location containing the address to jump to + void jump(Address address) + { + load32(address, dataTempRegister); + m_assembler.bx(dataTempRegister); + } + + void jump(AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), dataTempRegister); + load32(Address(dataTempRegister), dataTempRegister); + m_assembler.bx(dataTempRegister); + } + + + // Arithmetic control flow operations: + // + // This set of conditional branch operations branch based + // on the result of an arithmetic operation. The operation + // is performed as normal, storing the result. + // + // * jz operations branch if the result is zero. + // * jo operations branch if the (signed) arithmetic + // operation caused an overflow to occur. + + Jump branchAdd32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.add_S(dest, op1, op2); + return Jump(makeBranch(cond)); + } + + Jump branchAdd32(ResultCondition cond, RegisterID op1, TrustedImm32 imm, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add_S(dest, op1, armImm); + else { + move(imm, dataTempRegister); + m_assembler.add_S(dest, op1, dataTempRegister); + } + return Jump(makeBranch(cond)); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchAdd32(cond, dest, src, dest); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + return branchAdd32(cond, dest, imm, dest); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, AbsoluteAddress dest) + { + // Move the high bits of the address into addressTempRegister, + // and load the value into dataTempRegister. + move(TrustedImmPtr(dest.m_ptr), addressTempRegister); + m_assembler.ldr(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt16(0)); + + // Do the add. + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.add_S(dataTempRegister, dataTempRegister, armImm); + else { + // If the operand does not fit into an immediate then load it temporarily + // into addressTempRegister; since we're overwriting addressTempRegister + // we'll need to reload it with the high bits of the address afterwards. + move(imm, addressTempRegister); + m_assembler.add_S(dataTempRegister, dataTempRegister, addressTempRegister); + move(TrustedImmPtr(dest.m_ptr), addressTempRegister); + } + + // Store the result. + m_assembler.str(dataTempRegister, addressTempRegister, ARMThumbImmediate::makeUInt16(0)); + + return Jump(makeBranch(cond)); + } + + Jump branchMul32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + m_assembler.smull(dest, dataTempRegister, src1, src2); + + if (cond == Overflow) { + m_assembler.asr(addressTempRegister, dest, 31); + return branch32(NotEqual, addressTempRegister, dataTempRegister); + } + + return branchTest32(cond, dest); + } + + Jump branchMul32(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchMul32(cond, src, dest, dest); + } + + Jump branchMul32(ResultCondition cond, TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(imm, dataTempRegister); + return branchMul32(cond, dataTempRegister, src, dest); + } + + Jump branchNeg32(ResultCondition cond, RegisterID srcDest) + { + ARMThumbImmediate zero = ARMThumbImmediate::makeUInt12(0); + m_assembler.sub_S(srcDest, zero, srcDest); + return Jump(makeBranch(cond)); + } + + Jump branchOr32(ResultCondition cond, RegisterID src, RegisterID dest) + { + m_assembler.orr_S(dest, dest, src); + return Jump(makeBranch(cond)); + } + + Jump branchSub32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.sub_S(dest, op1, op2); + return Jump(makeBranch(cond)); + } + + Jump branchSub32(ResultCondition cond, RegisterID op1, TrustedImm32 imm, RegisterID dest) + { + ARMThumbImmediate armImm = ARMThumbImmediate::makeEncodedImm(imm.m_value); + if (armImm.isValid()) + m_assembler.sub_S(dest, op1, armImm); + else { + move(imm, dataTempRegister); + m_assembler.sub_S(dest, op1, dataTempRegister); + } + return Jump(makeBranch(cond)); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, RegisterID dest) + { + return branchSub32(cond, dest, src, dest); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + return branchSub32(cond, dest, imm, dest); + } + + void relativeTableJump(RegisterID index, int scale) + { + ASSERT(scale >= 0 && scale <= 31); + + // dataTempRegister will point after the jump if index register contains zero + move(ARMRegisters::pc, dataTempRegister); + m_assembler.add(dataTempRegister, dataTempRegister, ARMThumbImmediate::makeEncodedImm(9)); + + ShiftTypeAndAmount shift(SRType_LSL, scale); + m_assembler.add(dataTempRegister, dataTempRegister, index, shift); + jump(dataTempRegister); + } + + // Miscellaneous operations: + + void breakpoint(uint8_t imm = 0) + { + m_assembler.bkpt(imm); + } + + ALWAYS_INLINE Call nearCall() + { + moveFixedWidthEncoding(TrustedImm32(0), dataTempRegister); + return Call(m_assembler.blx(dataTempRegister), Call::LinkableNear); + } + + ALWAYS_INLINE Call call() + { + moveFixedWidthEncoding(TrustedImm32(0), dataTempRegister); + return Call(m_assembler.blx(dataTempRegister), Call::Linkable); + } + + ALWAYS_INLINE Call call(RegisterID target) + { + return Call(m_assembler.blx(target), Call::None); + } + + ALWAYS_INLINE Call call(Address address) + { + load32(address, dataTempRegister); + return Call(m_assembler.blx(dataTempRegister), Call::None); + } + + ALWAYS_INLINE void ret() + { + m_assembler.bx(linkRegister); + } + + void compare32(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + m_assembler.cmp(left, right); + m_assembler.it(armV7Condition(cond), false); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(1)); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(0)); + } + + void compare32(RelationalCondition cond, Address left, RegisterID right, RegisterID dest) + { + load32(left, dataTempRegister); + compare32(cond, dataTempRegister, right, dest); + } + + void compare8(RelationalCondition cond, Address left, TrustedImm32 right, RegisterID dest) + { + load8(left, addressTempRegister); + compare32(cond, addressTempRegister, right, dest); + } + + void compare32(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + compare32(left, right); + m_assembler.it(armV7Condition(cond), false); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(1)); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(0)); + } + + // FIXME: + // The mask should be optional... paerhaps the argument order should be + // dest-src, operations always have a dest? ... possibly not true, considering + // asm ops like test, or pseudo ops like pop(). + void test32(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + load32(address, dataTempRegister); + test32(dataTempRegister, mask); + m_assembler.it(armV7Condition(cond), false); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(1)); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(0)); + } + + void test8(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + load8(address, dataTempRegister); + test32(dataTempRegister, mask); + m_assembler.it(armV7Condition(cond), false); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(1)); + m_assembler.mov(dest, ARMThumbImmediate::makeUInt16(0)); + } + + ALWAYS_INLINE DataLabel32 moveWithPatch(TrustedImm32 imm, RegisterID dst) + { + padBeforePatch(); + moveFixedWidthEncoding(imm, dst); + return DataLabel32(this); + } + + ALWAYS_INLINE DataLabelPtr moveWithPatch(TrustedImmPtr imm, RegisterID dst) + { + padBeforePatch(); + moveFixedWidthEncoding(TrustedImm32(imm), dst); + return DataLabelPtr(this); + } + + ALWAYS_INLINE Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + dataLabel = moveWithPatch(initialRightValue, dataTempRegister); + return branch32(cond, left, dataTempRegister); + } + + ALWAYS_INLINE Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + load32(left, addressTempRegister); + dataLabel = moveWithPatch(initialRightValue, dataTempRegister); + return branch32(cond, addressTempRegister, dataTempRegister); + } + + PatchableJump patchableBranchPtr(RelationalCondition cond, Address left, TrustedImmPtr right = TrustedImmPtr(0)) + { + m_makeJumpPatchable = true; + Jump result = branch32(cond, left, TrustedImm32(right)); + m_makeJumpPatchable = false; + return PatchableJump(result); + } + + PatchableJump patchableBranchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + m_makeJumpPatchable = true; + Jump result = branchTest32(cond, reg, mask); + m_makeJumpPatchable = false; + return PatchableJump(result); + } + + PatchableJump patchableBranch32(RelationalCondition cond, RegisterID reg, TrustedImm32 imm) + { + m_makeJumpPatchable = true; + Jump result = branch32(cond, reg, imm); + m_makeJumpPatchable = false; + return PatchableJump(result); + } + + PatchableJump patchableBranchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + m_makeJumpPatchable = true; + Jump result = branchPtrWithPatch(cond, left, dataLabel, initialRightValue); + m_makeJumpPatchable = false; + return PatchableJump(result); + } + + PatchableJump patchableJump() + { + padBeforePatch(); + m_makeJumpPatchable = true; + Jump result = jump(); + m_makeJumpPatchable = false; + return PatchableJump(result); + } + + ALWAYS_INLINE DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + DataLabelPtr label = moveWithPatch(initialValue, dataTempRegister); + store32(dataTempRegister, address); + return label; + } + ALWAYS_INLINE DataLabelPtr storePtrWithPatch(ImplicitAddress address) { return storePtrWithPatch(TrustedImmPtr(0), address); } + + + ALWAYS_INLINE Call tailRecursiveCall() + { + // Like a normal call, but don't link. + moveFixedWidthEncoding(TrustedImm32(0), dataTempRegister); + return Call(m_assembler.bx(dataTempRegister), Call::Linkable); + } + + ALWAYS_INLINE Call makeTailRecursiveCall(Jump oldJump) + { + oldJump.link(this); + return tailRecursiveCall(); + } + + + int executableOffsetFor(int location) + { + return m_assembler.executableOffsetFor(location); + } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + return FunctionPtr(reinterpret_cast<void(*)()>(ARMv7Assembler::readCallTarget(call.dataLocation()))); + } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return false; } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + const unsigned twoWordOpSize = 4; + return label.labelAtOffset(-twoWordOpSize * 2); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID rd, void* initialValue) + { +#if OS(LINUX) || OS(QNX) + ARMv7Assembler::revertJumpTo_movT3movtcmpT2(instructionStart.dataLocation(), rd, dataTempRegister, reinterpret_cast<uintptr_t>(initialValue)); +#else + UNUSED_PARAM(rd); + ARMv7Assembler::revertJumpTo_movT3(instructionStart.dataLocation(), dataTempRegister, ARMThumbImmediate::makeUInt16(reinterpret_cast<uintptr_t>(initialValue) & 0xffff)); +#endif + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr) + { + UNREACHABLE_FOR_PLATFORM(); + return CodeLocationLabel(); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel, Address, void*) + { + UNREACHABLE_FOR_PLATFORM(); + } + +protected: + ALWAYS_INLINE Jump jump() + { + m_assembler.label(); // Force nop-padding if we're in the middle of a watchpoint. + moveFixedWidthEncoding(TrustedImm32(0), dataTempRegister); + return Jump(m_assembler.bx(dataTempRegister), m_makeJumpPatchable ? ARMv7Assembler::JumpNoConditionFixedSize : ARMv7Assembler::JumpNoCondition); + } + + ALWAYS_INLINE Jump makeBranch(ARMv7Assembler::Condition cond) + { + m_assembler.label(); // Force nop-padding if we're in the middle of a watchpoint. + m_assembler.it(cond, true, true); + moveFixedWidthEncoding(TrustedImm32(0), dataTempRegister); + return Jump(m_assembler.bx(dataTempRegister), m_makeJumpPatchable ? ARMv7Assembler::JumpConditionFixedSize : ARMv7Assembler::JumpCondition, cond); + } + ALWAYS_INLINE Jump makeBranch(RelationalCondition cond) { return makeBranch(armV7Condition(cond)); } + ALWAYS_INLINE Jump makeBranch(ResultCondition cond) { return makeBranch(armV7Condition(cond)); } + ALWAYS_INLINE Jump makeBranch(DoubleCondition cond) { return makeBranch(armV7Condition(cond)); } + + ArmAddress setupArmAddress(BaseIndex address) + { + if (address.offset) { + ARMThumbImmediate imm = ARMThumbImmediate::makeUInt12OrEncodedImm(address.offset); + if (imm.isValid()) + m_assembler.add(addressTempRegister, address.base, imm); + else { + move(TrustedImm32(address.offset), addressTempRegister); + m_assembler.add(addressTempRegister, addressTempRegister, address.base); + } + + return ArmAddress(addressTempRegister, address.index, address.scale); + } else + return ArmAddress(address.base, address.index, address.scale); + } + + ArmAddress setupArmAddress(Address address) + { + if ((address.offset >= -0xff) && (address.offset <= 0xfff)) + return ArmAddress(address.base, address.offset); + + move(TrustedImm32(address.offset), addressTempRegister); + return ArmAddress(address.base, addressTempRegister); + } + + ArmAddress setupArmAddress(ImplicitAddress address) + { + if ((address.offset >= -0xff) && (address.offset <= 0xfff)) + return ArmAddress(address.base, address.offset); + + move(TrustedImm32(address.offset), addressTempRegister); + return ArmAddress(address.base, addressTempRegister); + } + + RegisterID makeBaseIndexBase(BaseIndex address) + { + if (!address.offset) + return address.base; + + ARMThumbImmediate imm = ARMThumbImmediate::makeUInt12OrEncodedImm(address.offset); + if (imm.isValid()) + m_assembler.add(addressTempRegister, address.base, imm); + else { + move(TrustedImm32(address.offset), addressTempRegister); + m_assembler.add(addressTempRegister, addressTempRegister, address.base); + } + + return addressTempRegister; + } + + void moveFixedWidthEncoding(TrustedImm32 imm, RegisterID dst) + { + uint32_t value = imm.m_value; + m_assembler.movT3(dst, ARMThumbImmediate::makeUInt16(value & 0xffff)); + m_assembler.movt(dst, ARMThumbImmediate::makeUInt16(value >> 16)); + } + + ARMv7Assembler::Condition armV7Condition(RelationalCondition cond) + { + return static_cast<ARMv7Assembler::Condition>(cond); + } + + ARMv7Assembler::Condition armV7Condition(ResultCondition cond) + { + return static_cast<ARMv7Assembler::Condition>(cond); + } + + ARMv7Assembler::Condition armV7Condition(DoubleCondition cond) + { + return static_cast<ARMv7Assembler::Condition>(cond); + } + +private: + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkCall(void* code, Call call, FunctionPtr function) + { + ARMv7Assembler::linkCall(code, call.m_label, function.value()); + } + + static void repatchCall(CodeLocationCall call, CodeLocationLabel destination) + { + ARMv7Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + static void repatchCall(CodeLocationCall call, FunctionPtr destination) + { + ARMv7Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + bool m_makeJumpPatchable; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssemblerARMv7_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerCodeRef.h b/src/3rdparty/masm/assembler/MacroAssemblerCodeRef.h new file mode 100644 index 0000000000..89cffb1278 --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerCodeRef.h @@ -0,0 +1,406 @@ +/* + * Copyright (C) 2009, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerCodeRef_h +#define MacroAssemblerCodeRef_h + +#include "Disassembler.h" +#include "ExecutableAllocator.h" +#include "LLIntData.h" +#include <wtf/DataLog.h> +#include <wtf/PassRefPtr.h> +#include <wtf/RefPtr.h> +#include <wtf/UnusedParam.h> + +// ASSERT_VALID_CODE_POINTER checks that ptr is a non-null pointer, and that it is a valid +// instruction address on the platform (for example, check any alignment requirements). +#if CPU(ARM_THUMB2) +// ARM/thumb instructions must be 16-bit aligned, but all code pointers to be loaded +// into the processor are decorated with the bottom bit set, indicating that this is +// thumb code (as oposed to 32-bit traditional ARM). The first test checks for both +// decorated and undectorated null, and the second test ensures that the pointer is +// decorated. +#define ASSERT_VALID_CODE_POINTER(ptr) \ + ASSERT(reinterpret_cast<intptr_t>(ptr) & ~1); \ + ASSERT(reinterpret_cast<intptr_t>(ptr) & 1) +#define ASSERT_VALID_CODE_OFFSET(offset) \ + ASSERT(!(offset & 1)) // Must be multiple of 2. +#else +#define ASSERT_VALID_CODE_POINTER(ptr) \ + ASSERT(ptr) +#define ASSERT_VALID_CODE_OFFSET(offset) // Anything goes! +#endif + +#if CPU(X86) && OS(WINDOWS) +#define CALLING_CONVENTION_IS_STDCALL 1 +#ifndef CDECL +#if COMPILER(MSVC) +#define CDECL __cdecl +#else +#define CDECL __attribute__ ((__cdecl)) +#endif // COMPILER(MSVC) +#endif // CDECL +#else +#define CALLING_CONVENTION_IS_STDCALL 0 +#endif + +#if CPU(X86) +#define HAS_FASTCALL_CALLING_CONVENTION 1 +#ifndef FASTCALL +#if COMPILER(MSVC) +#define FASTCALL __fastcall +#else +#define FASTCALL __attribute__ ((fastcall)) +#endif // COMPILER(MSVC) +#endif // FASTCALL +#else +#define HAS_FASTCALL_CALLING_CONVENTION 0 +#endif // CPU(X86) + +namespace JSC { + +// FunctionPtr: +// +// FunctionPtr should be used to wrap pointers to C/C++ functions in JSC +// (particularly, the stub functions). +class FunctionPtr { +public: + FunctionPtr() + : m_value(0) + { + } + + template<typename returnType> + FunctionPtr(returnType(*value)()) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1> + FunctionPtr(returnType(*value)(argType1)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2> + FunctionPtr(returnType(*value)(argType1, argType2)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3> + FunctionPtr(returnType(*value)(argType1, argType2, argType3)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3, typename argType4> + FunctionPtr(returnType(*value)(argType1, argType2, argType3, argType4)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3, typename argType4, typename argType5> + FunctionPtr(returnType(*value)(argType1, argType2, argType3, argType4, argType5)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3, typename argType4, typename argType5, typename argType6> + FunctionPtr(returnType(*value)(argType1, argType2, argType3, argType4, argType5, argType6)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + +// MSVC doesn't seem to treat functions with different calling conventions as +// different types; these methods already defined for fastcall, below. +#if CALLING_CONVENTION_IS_STDCALL && !OS(WINDOWS) + + template<typename returnType> + FunctionPtr(returnType (CDECL *value)()) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1> + FunctionPtr(returnType (CDECL *value)(argType1)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2> + FunctionPtr(returnType (CDECL *value)(argType1, argType2)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3> + FunctionPtr(returnType (CDECL *value)(argType1, argType2, argType3)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3, typename argType4> + FunctionPtr(returnType (CDECL *value)(argType1, argType2, argType3, argType4)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } +#endif + +#if HAS_FASTCALL_CALLING_CONVENTION + + template<typename returnType> + FunctionPtr(returnType (FASTCALL *value)()) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1> + FunctionPtr(returnType (FASTCALL *value)(argType1)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2> + FunctionPtr(returnType (FASTCALL *value)(argType1, argType2)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3> + FunctionPtr(returnType (FASTCALL *value)(argType1, argType2, argType3)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + template<typename returnType, typename argType1, typename argType2, typename argType3, typename argType4> + FunctionPtr(returnType (FASTCALL *value)(argType1, argType2, argType3, argType4)) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } +#endif + + template<typename FunctionType> + explicit FunctionPtr(FunctionType* value) + // Using a C-ctyle cast here to avoid compiler error on RVTC: + // Error: #694: reinterpret_cast cannot cast away const or other type qualifiers + // (I guess on RVTC function pointers have a different constness to GCC/MSVC?) + : m_value((void*)value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + void* value() const { return m_value; } + void* executableAddress() const { return m_value; } + + +private: + void* m_value; +}; + +// ReturnAddressPtr: +// +// ReturnAddressPtr should be used to wrap return addresses generated by processor +// 'call' instructions exectued in JIT code. We use return addresses to look up +// exception and optimization information, and to repatch the call instruction +// that is the source of the return address. +class ReturnAddressPtr { +public: + ReturnAddressPtr() + : m_value(0) + { + } + + explicit ReturnAddressPtr(void* value) + : m_value(value) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + explicit ReturnAddressPtr(FunctionPtr function) + : m_value(function.value()) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + void* value() const { return m_value; } + +private: + void* m_value; +}; + +// MacroAssemblerCodePtr: +// +// MacroAssemblerCodePtr should be used to wrap pointers to JIT generated code. +class MacroAssemblerCodePtr { +public: + MacroAssemblerCodePtr() + : m_value(0) + { + } + + explicit MacroAssemblerCodePtr(void* value) +#if CPU(ARM_THUMB2) + // Decorate the pointer as a thumb code pointer. + : m_value(reinterpret_cast<char*>(value) + 1) +#else + : m_value(value) +#endif + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + static MacroAssemblerCodePtr createFromExecutableAddress(void* value) + { + ASSERT_VALID_CODE_POINTER(value); + MacroAssemblerCodePtr result; + result.m_value = value; + return result; + } + +#if ENABLE(LLINT) + static MacroAssemblerCodePtr createLLIntCodePtr(LLIntCode codeId) + { + return createFromExecutableAddress(LLInt::getCodePtr(codeId)); + } +#endif + + explicit MacroAssemblerCodePtr(ReturnAddressPtr ra) + : m_value(ra.value()) + { + ASSERT_VALID_CODE_POINTER(m_value); + } + + void* executableAddress() const { return m_value; } +#if CPU(ARM_THUMB2) + // To use this pointer as a data address remove the decoration. + void* dataLocation() const { ASSERT_VALID_CODE_POINTER(m_value); return reinterpret_cast<char*>(m_value) - 1; } +#else + void* dataLocation() const { ASSERT_VALID_CODE_POINTER(m_value); return m_value; } +#endif + + bool operator!() const + { + return !m_value; + } + +private: + void* m_value; +}; + +// MacroAssemblerCodeRef: +// +// A reference to a section of JIT generated code. A CodeRef consists of a +// pointer to the code, and a ref pointer to the pool from within which it +// was allocated. +class MacroAssemblerCodeRef { +private: + // This is private because it's dangerous enough that we want uses of it + // to be easy to find - hence the static create method below. + explicit MacroAssemblerCodeRef(MacroAssemblerCodePtr codePtr) + : m_codePtr(codePtr) + { + ASSERT(m_codePtr); + } + +public: + MacroAssemblerCodeRef() + { + } + + MacroAssemblerCodeRef(PassRefPtr<ExecutableMemoryHandle> executableMemory) + : m_codePtr(executableMemory->start()) + , m_executableMemory(executableMemory) + { + ASSERT(m_executableMemory->isManaged()); + ASSERT(m_executableMemory->start()); + ASSERT(m_codePtr); + } + + // Use this only when you know that the codePtr refers to code that is + // already being kept alive through some other means. Typically this means + // that codePtr is immortal. + static MacroAssemblerCodeRef createSelfManagedCodeRef(MacroAssemblerCodePtr codePtr) + { + return MacroAssemblerCodeRef(codePtr); + } + +#if ENABLE(LLINT) + // Helper for creating self-managed code refs from LLInt. + static MacroAssemblerCodeRef createLLIntCodeRef(LLIntCode codeId) + { + return createSelfManagedCodeRef(MacroAssemblerCodePtr::createFromExecutableAddress(LLInt::getCodePtr(codeId))); + } +#endif + + ExecutableMemoryHandle* executableMemory() const + { + return m_executableMemory.get(); + } + + MacroAssemblerCodePtr code() const + { + return m_codePtr; + } + + size_t size() const + { + if (!m_executableMemory) + return 0; + return m_executableMemory->sizeInBytes(); + } + + bool tryToDisassemble(const char* prefix) const + { + return JSC::tryToDisassemble(m_codePtr, size(), prefix, WTF::dataFile()); + } + + bool operator!() const { return !m_codePtr; } + +private: + MacroAssemblerCodePtr m_codePtr; + RefPtr<ExecutableMemoryHandle> m_executableMemory; +}; + +} // namespace JSC + +#endif // MacroAssemblerCodeRef_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerMIPS.h b/src/3rdparty/masm/assembler/MacroAssemblerMIPS.h new file mode 100644 index 0000000000..e18d86c5b3 --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerMIPS.h @@ -0,0 +1,2751 @@ +/* + * Copyright (C) 2008 Apple Inc. All rights reserved. + * Copyright (C) 2010 MIPS Technologies, Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY MIPS TECHNOLOGIES, INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL MIPS TECHNOLOGIES, INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerMIPS_h +#define MacroAssemblerMIPS_h + +#if ENABLE(ASSEMBLER) && CPU(MIPS) + +#include "AbstractMacroAssembler.h" +#include "MIPSAssembler.h" + +namespace JSC { + +class MacroAssemblerMIPS : public AbstractMacroAssembler<MIPSAssembler> { +public: + typedef MIPSRegisters::FPRegisterID FPRegisterID; + + MacroAssemblerMIPS() + : m_fixedWidth(false) + { + } + + static bool isCompactPtrAlignedAddressOffset(ptrdiff_t value) + { + return value >= -2147483647 - 1 && value <= 2147483647; + } + + static const Scale ScalePtr = TimesFour; + + // For storing immediate number + static const RegisterID immTempRegister = MIPSRegisters::t0; + // For storing data loaded from the memory + static const RegisterID dataTempRegister = MIPSRegisters::t1; + // For storing address base + static const RegisterID addrTempRegister = MIPSRegisters::t2; + // For storing compare result + static const RegisterID cmpTempRegister = MIPSRegisters::t3; + + // FP temp register + static const FPRegisterID fpTempRegister = MIPSRegisters::f16; + + static const int MaximumCompactPtrAlignedAddressOffset = 0x7FFFFFFF; + + enum RelationalCondition { + Equal, + NotEqual, + Above, + AboveOrEqual, + Below, + BelowOrEqual, + GreaterThan, + GreaterThanOrEqual, + LessThan, + LessThanOrEqual + }; + + enum ResultCondition { + Overflow, + Signed, + Zero, + NonZero + }; + + enum DoubleCondition { + DoubleEqual, + DoubleNotEqual, + DoubleGreaterThan, + DoubleGreaterThanOrEqual, + DoubleLessThan, + DoubleLessThanOrEqual, + DoubleEqualOrUnordered, + DoubleNotEqualOrUnordered, + DoubleGreaterThanOrUnordered, + DoubleGreaterThanOrEqualOrUnordered, + DoubleLessThanOrUnordered, + DoubleLessThanOrEqualOrUnordered + }; + + static const RegisterID stackPointerRegister = MIPSRegisters::sp; + static const RegisterID returnAddressRegister = MIPSRegisters::ra; + + // Integer arithmetic operations: + // + // Operations are typically two operand - operation(source, srcDst) + // For many operations the source may be an TrustedImm32, the srcDst operand + // may often be a memory location (explictly described using an Address + // object). + + void add32(RegisterID src, RegisterID dest) + { + m_assembler.addu(dest, dest, src); + } + + void add32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.addu(dest, op1, op2); + } + + void add32(TrustedImm32 imm, RegisterID dest) + { + add32(imm, dest, dest); + } + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (imm.m_value >= -32768 && imm.m_value <= 32767 + && !m_fixedWidth) { + /* + addiu dest, src, imm + */ + m_assembler.addiu(dest, src, imm.m_value); + } else { + /* + li immTemp, imm + addu dest, src, immTemp + */ + move(imm, immTempRegister); + m_assembler.addu(dest, src, immTempRegister); + } + } + + void add32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + add32(imm, src, dest); + } + + void add32(TrustedImm32 imm, Address address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + lw dataTemp, offset(base) + li immTemp, imm + addu dataTemp, dataTemp, immTemp + sw dataTemp, offset(base) + */ + m_assembler.lw(dataTempRegister, address.base, address.offset); + if (imm.m_value >= -32768 && imm.m_value <= 32767 + && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, dataTempRegister, imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.addu(dataTempRegister, dataTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, address.base, address.offset); + } else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lw dataTemp, (offset & 0xffff)(addrTemp) + li immtemp, imm + addu dataTemp, dataTemp, immTemp + sw dataTemp, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dataTempRegister, addrTempRegister, address.offset); + + if (imm.m_value >= -32768 && imm.m_value <= 32767 && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, dataTempRegister, imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.addu(dataTempRegister, dataTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, addrTempRegister, address.offset); + } + } + + void add32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + add32(dataTempRegister, dest); + } + + void add32(AbsoluteAddress src, RegisterID dest) + { + load32(src.m_ptr, dataTempRegister); + add32(dataTempRegister, dest); + } + + void add32(RegisterID src, Address dest) + { + if (dest.offset >= -32768 && dest.offset <= 32767 && !m_fixedWidth) { + /* + lw dataTemp, offset(base) + addu dataTemp, dataTemp, src + sw dataTemp, offset(base) + */ + m_assembler.lw(dataTempRegister, dest.base, dest.offset); + m_assembler.addu(dataTempRegister, dataTempRegister, src); + m_assembler.sw(dataTempRegister, dest.base, dest.offset); + } else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lw dataTemp, (offset & 0xffff)(addrTemp) + addu dataTemp, dataTemp, src + sw dataTemp, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (dest.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, dest.base); + m_assembler.lw(dataTempRegister, addrTempRegister, dest.offset); + m_assembler.addu(dataTempRegister, dataTempRegister, src); + m_assembler.sw(dataTempRegister, addrTempRegister, dest.offset); + } + } + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + /* + li addrTemp, address + li immTemp, imm + lw cmpTemp, 0(addrTemp) + addu dataTemp, cmpTemp, immTemp + sw dataTemp, 0(addrTemp) + */ + move(TrustedImmPtr(address.m_ptr), addrTempRegister); + m_assembler.lw(cmpTempRegister, addrTempRegister, 0); + if (imm.m_value >= -32768 && imm.m_value <= 32767 && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, cmpTempRegister, imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.addu(dataTempRegister, cmpTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, addrTempRegister, 0); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + /* + add32(imm, address) + sltu immTemp, dataTemp, cmpTemp # set carry-in bit + lw dataTemp, 4(addrTemp) + addiu dataTemp, imm.m_value >> 31 ? -1 : 0 + addu dataTemp, dataTemp, immTemp + sw dataTemp, 4(addrTemp) + */ + add32(imm, address); + m_assembler.sltu(immTempRegister, dataTempRegister, cmpTempRegister); + m_assembler.lw(dataTempRegister, addrTempRegister, 4); + if (imm.m_value >> 31) + m_assembler.addiu(dataTempRegister, dataTempRegister, -1); + m_assembler.addu(dataTempRegister, dataTempRegister, immTempRegister); + m_assembler.sw(dataTempRegister, addrTempRegister, 4); + } + + void and32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + and32(dataTempRegister, dest); + } + + void and32(RegisterID src, RegisterID dest) + { + m_assembler.andInsn(dest, dest, src); + } + + void and32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.andInsn(dest, op1, op2); + } + + void and32(TrustedImm32 imm, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + move(MIPSRegisters::zero, dest); + else if (imm.m_value > 0 && imm.m_value < 65535 && !m_fixedWidth) + m_assembler.andi(dest, dest, imm.m_value); + else { + /* + li immTemp, imm + and dest, dest, immTemp + */ + move(imm, immTempRegister); + m_assembler.andInsn(dest, dest, immTempRegister); + } + } + + void and32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + move(MIPSRegisters::zero, dest); + else if (imm.m_value > 0 && imm.m_value < 65535 && !m_fixedWidth) + m_assembler.andi(dest, src, imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.andInsn(dest, src, immTempRegister); + } + } + + void lshift32(RegisterID shiftAmount, RegisterID dest) + { + m_assembler.sllv(dest, dest, shiftAmount); + } + + void lshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + m_assembler.sllv(dest, src, shiftAmount); + } + + void lshift32(TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + m_assembler.sllv(dest, dest, immTempRegister); + } + + void lshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + m_assembler.sllv(dest, src, immTempRegister); + } + + void mul32(RegisterID src, RegisterID dest) + { + m_assembler.mul(dest, dest, src); + } + + void mul32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.mul(dest, op1, op2); + } + + void mul32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + move(MIPSRegisters::zero, dest); + else if (imm.m_value == 1 && !m_fixedWidth) + move(src, dest); + else { + /* + li dataTemp, imm + mul dest, src, dataTemp + */ + move(imm, dataTempRegister); + m_assembler.mul(dest, src, dataTempRegister); + } + } + + void neg32(RegisterID srcDest) + { + m_assembler.subu(srcDest, MIPSRegisters::zero, srcDest); + } + + void or32(RegisterID src, RegisterID dest) + { + m_assembler.orInsn(dest, dest, src); + } + + void or32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.orInsn(dest, op1, op2); + } + + void or32(TrustedImm32 imm, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + return; + + if (imm.m_value > 0 && imm.m_value < 65535 + && !m_fixedWidth) { + m_assembler.ori(dest, dest, imm.m_value); + return; + } + + /* + li dataTemp, imm + or dest, dest, dataTemp + */ + move(imm, dataTempRegister); + m_assembler.orInsn(dest, dest, dataTempRegister); + } + + void or32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + return; + + if (imm.m_value > 0 && imm.m_value < 65535 && !m_fixedWidth) { + m_assembler.ori(dest, src, imm.m_value); + return; + } + + /* + li dataTemp, imm + or dest, src, dataTemp + */ + move(imm, dataTempRegister); + m_assembler.orInsn(dest, src, dataTempRegister); + } + + void or32(RegisterID src, AbsoluteAddress dest) + { + load32(dest.m_ptr, dataTempRegister); + m_assembler.orInsn(dataTempRegister, dataTempRegister, src); + store32(dataTempRegister, dest.m_ptr); + } + + void rshift32(RegisterID shiftAmount, RegisterID dest) + { + m_assembler.srav(dest, dest, shiftAmount); + } + + void rshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + m_assembler.srav(dest, src, shiftAmount); + } + + void rshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.sra(dest, dest, imm.m_value); + } + + void rshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.sra(dest, src, imm.m_value); + } + + void urshift32(RegisterID shiftAmount, RegisterID dest) + { + m_assembler.srlv(dest, dest, shiftAmount); + } + + void urshift32(RegisterID src, RegisterID shiftAmount, RegisterID dest) + { + m_assembler.srlv(dest, src, shiftAmount); + } + + void urshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.srl(dest, dest, imm.m_value); + } + + void urshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + m_assembler.srl(dest, src, imm.m_value); + } + + void sub32(RegisterID src, RegisterID dest) + { + m_assembler.subu(dest, dest, src); + } + + void sub32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.subu(dest, op1, op2); + } + + void sub32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value >= -32767 && imm.m_value <= 32768 + && !m_fixedWidth) { + /* + addiu dest, src, imm + */ + m_assembler.addiu(dest, dest, -imm.m_value); + } else { + /* + li immTemp, imm + subu dest, src, immTemp + */ + move(imm, immTempRegister); + m_assembler.subu(dest, dest, immTempRegister); + } + } + + void sub32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value >= -32767 && imm.m_value <= 32768 + && !m_fixedWidth) { + /* + addiu dest, src, imm + */ + m_assembler.addiu(dest, src, -imm.m_value); + } else { + /* + li immTemp, imm + subu dest, src, immTemp + */ + move(imm, immTempRegister); + m_assembler.subu(dest, src, immTempRegister); + } + } + + void sub32(TrustedImm32 imm, Address address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + lw dataTemp, offset(base) + li immTemp, imm + subu dataTemp, dataTemp, immTemp + sw dataTemp, offset(base) + */ + m_assembler.lw(dataTempRegister, address.base, address.offset); + if (imm.m_value >= -32767 && imm.m_value <= 32768 && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, dataTempRegister, -imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.subu(dataTempRegister, dataTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, address.base, address.offset); + } else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lw dataTemp, (offset & 0xffff)(addrTemp) + li immtemp, imm + subu dataTemp, dataTemp, immTemp + sw dataTemp, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dataTempRegister, addrTempRegister, address.offset); + + if (imm.m_value >= -32767 && imm.m_value <= 32768 + && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, dataTempRegister, -imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.subu(dataTempRegister, dataTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, addrTempRegister, address.offset); + } + } + + void sub32(Address src, RegisterID dest) + { + load32(src, dataTempRegister); + sub32(dataTempRegister, dest); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + /* + li addrTemp, address + li immTemp, imm + lw dataTemp, 0(addrTemp) + subu dataTemp, dataTemp, immTemp + sw dataTemp, 0(addrTemp) + */ + move(TrustedImmPtr(address.m_ptr), addrTempRegister); + m_assembler.lw(dataTempRegister, addrTempRegister, 0); + + if (imm.m_value >= -32767 && imm.m_value <= 32768 && !m_fixedWidth) + m_assembler.addiu(dataTempRegister, dataTempRegister, -imm.m_value); + else { + move(imm, immTempRegister); + m_assembler.subu(dataTempRegister, dataTempRegister, immTempRegister); + } + m_assembler.sw(dataTempRegister, addrTempRegister, 0); + } + + void xor32(RegisterID src, RegisterID dest) + { + m_assembler.xorInsn(dest, dest, src); + } + + void xor32(RegisterID op1, RegisterID op2, RegisterID dest) + { + m_assembler.xorInsn(dest, op1, op2); + } + + void xor32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value == -1) { + m_assembler.nor(dest, dest, MIPSRegisters::zero); + return; + } + + /* + li immTemp, imm + xor dest, dest, immTemp + */ + move(imm, immTempRegister); + m_assembler.xorInsn(dest, dest, immTempRegister); + } + + void xor32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (imm.m_value == -1) { + m_assembler.nor(dest, src, MIPSRegisters::zero); + return; + } + + /* + li immTemp, imm + xor dest, dest, immTemp + */ + move(imm, immTempRegister); + m_assembler.xorInsn(dest, src, immTempRegister); + } + + void sqrtDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.sqrtd(dst, src); + } + + void absDouble(FPRegisterID, FPRegisterID) + { + RELEASE_ASSERT_NOT_REACHED(); + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result(this); + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lw dest, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dest, addrTempRegister, address.offset); + return result; + } + + // Memory access operations: + // + // Loads are of the form load(address, destination) and stores of the form + // store(source, address). The source for a store may be an TrustedImm32. Address + // operand objects to loads and store will be implicitly constructed if a + // register is passed. + + /* Need to use zero-extened load byte for load8. */ + void load8(ImplicitAddress address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) + m_assembler.lbu(dest, address.base, address.offset); + else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lbu dest, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lbu(dest, addrTempRegister, address.offset); + } + } + + void load8(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lbu dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lbu(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lbu dest, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lbu(dest, addrTempRegister, address.offset); + } + } + + void load8Signed(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lb dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lb(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lb dest, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lb(dest, addrTempRegister, address.offset); + } + } + + void load32(ImplicitAddress address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) + m_assembler.lw(dest, address.base, address.offset); + else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lw dest, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dest, addrTempRegister, address.offset); + } + } + + void load32(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lw dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lw dest, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lw(dest, addrTempRegister, address.offset); + } + } + + void load16Unaligned(BaseIndex address, RegisterID dest) + { + load16(address, dest); + } + + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32764 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + (Big-Endian) + lwl dest, address.offset(addrTemp) + lwr dest, address.offset+3(addrTemp) + (Little-Endian) + lwl dest, address.offset+3(addrTemp) + lwr dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); +#if CPU(BIG_ENDIAN) + m_assembler.lwl(dest, addrTempRegister, address.offset); + m_assembler.lwr(dest, addrTempRegister, address.offset + 3); +#else + m_assembler.lwl(dest, addrTempRegister, address.offset + 3); + m_assembler.lwr(dest, addrTempRegister, address.offset); + +#endif + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, address.offset >> 16 + ori immTemp, immTemp, address.offset & 0xffff + addu addrTemp, addrTemp, immTemp + (Big-Endian) + lw dest, 0(at) + lw dest, 3(at) + (Little-Endian) + lw dest, 3(at) + lw dest, 0(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, address.offset >> 16); + m_assembler.ori(immTempRegister, immTempRegister, address.offset); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); +#if CPU(BIG_ENDIAN) + m_assembler.lwl(dest, addrTempRegister, 0); + m_assembler.lwr(dest, addrTempRegister, 3); +#else + m_assembler.lwl(dest, addrTempRegister, 3); + m_assembler.lwr(dest, addrTempRegister, 0); +#endif + } + } + + void load32(const void* address, RegisterID dest) + { + /* + li addrTemp, address + lw dest, 0(addrTemp) + */ + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.lw(dest, addrTempRegister, 0); + } + + DataLabel32 load32WithAddressOffsetPatch(Address address, RegisterID dest) + { + m_fixedWidth = true; + /* + lui addrTemp, address.offset >> 16 + ori addrTemp, addrTemp, address.offset & 0xffff + addu addrTemp, addrTemp, address.base + lw dest, 0(addrTemp) + */ + DataLabel32 dataLabel(this); + move(TrustedImm32(address.offset), addrTempRegister); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lw(dest, addrTempRegister, 0); + m_fixedWidth = false; + return dataLabel; + } + + DataLabelCompact load32WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + DataLabelCompact dataLabel(this); + load32WithAddressOffsetPatch(address, dest); + return dataLabel; + } + + /* Need to use zero-extened load half-word for load16. */ + void load16(ImplicitAddress address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) + m_assembler.lhu(dest, address.base, address.offset); + else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + lhu dest, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lhu(dest, addrTempRegister, address.offset); + } + } + + /* Need to use zero-extened load half-word for load16. */ + void load16(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lhu dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lhu(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lhu dest, (address.offset & 0xffff)(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lhu(dest, addrTempRegister, address.offset); + } + } + + void load16Signed(BaseIndex address, RegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lh dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lh(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lh dest, (address.offset & 0xffff)(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lh(dest, addrTempRegister, address.offset); + } + } + + DataLabel32 store32WithAddressOffsetPatch(RegisterID src, Address address) + { + m_fixedWidth = true; + /* + lui addrTemp, address.offset >> 16 + ori addrTemp, addrTemp, address.offset & 0xffff + addu addrTemp, addrTemp, address.base + sw src, 0(addrTemp) + */ + DataLabel32 dataLabel(this); + move(TrustedImm32(address.offset), addrTempRegister); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sw(src, addrTempRegister, 0); + m_fixedWidth = false; + return dataLabel; + } + + void store8(RegisterID src, BaseIndex address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + sb src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sb(src, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + sb src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.sb(src, addrTempRegister, address.offset); + } + } + + void store8(TrustedImm32 imm, void* address) + { + /* + li immTemp, imm + li addrTemp, address + sb src, 0(addrTemp) + */ + if (!imm.m_value && !m_fixedWidth) { + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sb(MIPSRegisters::zero, addrTempRegister, 0); + } else { + move(imm, immTempRegister); + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sb(immTempRegister, addrTempRegister, 0); + } + } + + void store16(RegisterID src, BaseIndex address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + sh src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sh(src, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + sh src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.sh(src, addrTempRegister, address.offset); + } + } + + void store32(RegisterID src, ImplicitAddress address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) + m_assembler.sw(src, address.base, address.offset); + else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + sw src, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sw(src, addrTempRegister, address.offset); + } + } + + void store32(RegisterID src, BaseIndex address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + sw src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sw(src, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + sw src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.sw(src, addrTempRegister, address.offset); + } + } + + void store32(TrustedImm32 imm, ImplicitAddress address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + if (!imm.m_value) + m_assembler.sw(MIPSRegisters::zero, address.base, address.offset); + else { + move(imm, immTempRegister); + m_assembler.sw(immTempRegister, address.base, address.offset); + } + } else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + sw immTemp, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + if (!imm.m_value && !m_fixedWidth) + m_assembler.sw(MIPSRegisters::zero, addrTempRegister, address.offset); + else { + move(imm, immTempRegister); + m_assembler.sw(immTempRegister, addrTempRegister, address.offset); + } + } + } + + void store32(TrustedImm32 imm, BaseIndex address) + { + if (address.offset >= -32768 && address.offset <= 32767 && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + sw src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + if (!imm.m_value) + m_assembler.sw(MIPSRegisters::zero, addrTempRegister, address.offset); + else { + move(imm, immTempRegister); + m_assembler.sw(immTempRegister, addrTempRegister, address.offset); + } + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + sw src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + if (!imm.m_value && !m_fixedWidth) + m_assembler.sw(MIPSRegisters::zero, addrTempRegister, address.offset); + else { + move(imm, immTempRegister); + m_assembler.sw(immTempRegister, addrTempRegister, address.offset); + } + } + } + + + void store32(RegisterID src, const void* address) + { + /* + li addrTemp, address + sw src, 0(addrTemp) + */ + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sw(src, addrTempRegister, 0); + } + + void store32(TrustedImm32 imm, const void* address) + { + /* + li immTemp, imm + li addrTemp, address + sw src, 0(addrTemp) + */ + if (!imm.m_value && !m_fixedWidth) { + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sw(MIPSRegisters::zero, addrTempRegister, 0); + } else { + move(imm, immTempRegister); + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sw(immTempRegister, addrTempRegister, 0); + } + } + + // Floating-point operations: + + static bool supportsFloatingPoint() + { +#if WTF_MIPS_DOUBLE_FLOAT + return true; +#else + return false; +#endif + } + + static bool supportsFloatingPointTruncate() + { +#if WTF_MIPS_DOUBLE_FLOAT && WTF_MIPS_ISA_AT_LEAST(2) + return true; +#else + return false; +#endif + } + + static bool supportsFloatingPointSqrt() + { +#if WTF_MIPS_DOUBLE_FLOAT && WTF_MIPS_ISA_AT_LEAST(2) + return true; +#else + return false; +#endif + } + static bool supportsFloatingPointAbs() { return false; } + + // Stack manipulation operations: + // + // The ABI is assumed to provide a stack abstraction to memory, + // containing machine word sized units of data. Push and pop + // operations add and remove a single register sized unit of data + // to or from the stack. Peek and poke operations read or write + // values on the stack, without moving the current stack position. + + void pop(RegisterID dest) + { + m_assembler.lw(dest, MIPSRegisters::sp, 0); + m_assembler.addiu(MIPSRegisters::sp, MIPSRegisters::sp, 4); + } + + void push(RegisterID src) + { + m_assembler.addiu(MIPSRegisters::sp, MIPSRegisters::sp, -4); + m_assembler.sw(src, MIPSRegisters::sp, 0); + } + + void push(Address address) + { + load32(address, dataTempRegister); + push(dataTempRegister); + } + + void push(TrustedImm32 imm) + { + move(imm, immTempRegister); + push(immTempRegister); + } + + // Register move operations: + // + // Move values in registers. + + void move(TrustedImm32 imm, RegisterID dest) + { + if (!imm.m_value && !m_fixedWidth) + move(MIPSRegisters::zero, dest); + else if (m_fixedWidth) { + m_assembler.lui(dest, imm.m_value >> 16); + m_assembler.ori(dest, dest, imm.m_value); + } else + m_assembler.li(dest, imm.m_value); + } + + void move(RegisterID src, RegisterID dest) + { + if (src != dest || m_fixedWidth) + m_assembler.move(dest, src); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + move(TrustedImm32(imm), dest); + } + + void swap(RegisterID reg1, RegisterID reg2) + { + move(reg1, immTempRegister); + move(reg2, reg1); + move(immTempRegister, reg2); + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + if (src != dest || m_fixedWidth) + move(src, dest); + } + + void zeroExtend32ToPtr(RegisterID src, RegisterID dest) + { + if (src != dest || m_fixedWidth) + move(src, dest); + } + + // Forwards / external control flow operations: + // + // This set of jump and conditional branch operations return a Jump + // object which may linked at a later point, allow forwards jump, + // or jumps that will require external linkage (after the code has been + // relocated). + // + // For branches, signed <, >, <= and >= are denoted as l, g, le, and ge + // respecitvely, for unsigned comparisons the names b, a, be, and ae are + // used (representing the names 'below' and 'above'). + // + // Operands to the comparision are provided in the expected order, e.g. + // jle32(reg1, TrustedImm32(5)) will branch if the value held in reg1, when + // treated as a signed 32bit value, is less than or equal to 5. + // + // jz and jnz test whether the first operand is equal to zero, and take + // an optional second operand of a mask under which to perform the test. + + Jump branch8(RelationalCondition cond, Address left, TrustedImm32 right) + { + // Make sure the immediate value is unsigned 8 bits. + ASSERT(!(right.m_value & 0xFFFFFF00)); + load8(left, dataTempRegister); + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + void compare8(RelationalCondition cond, Address left, TrustedImm32 right, RegisterID dest) + { + // Make sure the immediate value is unsigned 8 bits. + ASSERT(!(right.m_value & 0xFFFFFF00)); + load8(left, dataTempRegister); + move(right, immTempRegister); + compare32(cond, dataTempRegister, immTempRegister, dest); + } + + Jump branch8(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + ASSERT(!(right.m_value & 0xFFFFFF00)); + load8(left, dataTempRegister); + // Be careful that the previous load8() uses immTempRegister. + // So, we need to put move() after load8(). + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + Jump branch32(RelationalCondition cond, RegisterID left, RegisterID right) + { + if (cond == Equal) + return branchEqual(left, right); + if (cond == NotEqual) + return branchNotEqual(left, right); + if (cond == Above) { + m_assembler.sltu(cmpTempRegister, right, left); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == AboveOrEqual) { + m_assembler.sltu(cmpTempRegister, left, right); + return branchEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Below) { + m_assembler.sltu(cmpTempRegister, left, right); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == BelowOrEqual) { + m_assembler.sltu(cmpTempRegister, right, left); + return branchEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == GreaterThan) { + m_assembler.slt(cmpTempRegister, right, left); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == GreaterThanOrEqual) { + m_assembler.slt(cmpTempRegister, left, right); + return branchEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == LessThan) { + m_assembler.slt(cmpTempRegister, left, right); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == LessThanOrEqual) { + m_assembler.slt(cmpTempRegister, right, left); + return branchEqual(cmpTempRegister, MIPSRegisters::zero); + } + ASSERT(0); + + return Jump(); + } + + Jump branch32(RelationalCondition cond, RegisterID left, TrustedImm32 right) + { + move(right, immTempRegister); + return branch32(cond, left, immTempRegister); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Address right) + { + load32(right, dataTempRegister); + return branch32(cond, left, dataTempRegister); + } + + Jump branch32(RelationalCondition cond, Address left, RegisterID right) + { + load32(left, dataTempRegister); + return branch32(cond, dataTempRegister, right); + } + + Jump branch32(RelationalCondition cond, Address left, TrustedImm32 right) + { + load32(left, dataTempRegister); + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + Jump branch32(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + load32(left, dataTempRegister); + // Be careful that the previous load32() uses immTempRegister. + // So, we need to put move() after load32(). + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + Jump branch32WithUnalignedHalfWords(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + load32WithUnalignedHalfWords(left, dataTempRegister); + // Be careful that the previous load32WithUnalignedHalfWords() + // uses immTempRegister. + // So, we need to put move() after load32WithUnalignedHalfWords(). + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + load32(left.m_ptr, dataTempRegister); + return branch32(cond, dataTempRegister, right); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, TrustedImm32 right) + { + load32(left.m_ptr, dataTempRegister); + move(right, immTempRegister); + return branch32(cond, dataTempRegister, immTempRegister); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, RegisterID mask) + { + ASSERT((cond == Zero) || (cond == NonZero)); + m_assembler.andInsn(cmpTempRegister, reg, mask); + if (cond == Zero) + return branchEqual(cmpTempRegister, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + ASSERT((cond == Zero) || (cond == NonZero)); + if (mask.m_value == -1 && !m_fixedWidth) { + if (cond == Zero) + return branchEqual(reg, MIPSRegisters::zero); + return branchNotEqual(reg, MIPSRegisters::zero); + } + move(mask, immTempRegister); + return branchTest32(cond, reg, immTempRegister); + } + + Jump branchTest32(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + load32(address, dataTempRegister); + return branchTest32(cond, dataTempRegister, mask); + } + + Jump branchTest32(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + load32(address, dataTempRegister); + return branchTest32(cond, dataTempRegister, mask); + } + + Jump branchTest8(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + load8(address, dataTempRegister); + return branchTest32(cond, dataTempRegister, mask); + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + move(TrustedImmPtr(address.m_ptr), dataTempRegister); + load8(Address(dataTempRegister), dataTempRegister); + return branchTest32(cond, dataTempRegister, mask); + } + + Jump jump() + { + return branchEqual(MIPSRegisters::zero, MIPSRegisters::zero); + } + + void jump(RegisterID target) + { + move(target, MIPSRegisters::t9); + m_assembler.jr(MIPSRegisters::t9); + m_assembler.nop(); + } + + void jump(Address address) + { + m_fixedWidth = true; + load32(address, MIPSRegisters::t9); + m_assembler.jr(MIPSRegisters::t9); + m_assembler.nop(); + m_fixedWidth = false; + } + + void jump(AbsoluteAddress address) + { + m_fixedWidth = true; + load32(address.m_ptr, MIPSRegisters::t9); + m_assembler.jr(MIPSRegisters::t9); + m_assembler.nop(); + m_fixedWidth = false; + } + + void moveDoubleToInts(FPRegisterID src, RegisterID dest1, RegisterID dest2) + { + m_assembler.vmov(dest1, dest2, src); + } + + void moveIntsToDouble(RegisterID src1, RegisterID src2, FPRegisterID dest, FPRegisterID scratch) + { + UNUSED_PARAM(scratch); + m_assembler.vmov(dest, src1, src2); + } + + // Arithmetic control flow operations: + // + // This set of conditional branch operations branch based + // on the result of an arithmetic operation. The operation + // is performed as normal, storing the result. + // + // * jz operations branch if the result is zero. + // * jo operations branch if the (signed) arithmetic + // operation caused an overflow to occur. + + Jump branchAdd32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + move dest, dataTemp + xor cmpTemp, dataTemp, src + bltz cmpTemp, No_overflow # diff sign bit -> no overflow + addu dest, dataTemp, src + xor cmpTemp, dest, dataTemp + bgez cmpTemp, No_overflow # same sign big -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + move(dest, dataTempRegister); + m_assembler.xorInsn(cmpTempRegister, dataTempRegister, src); + m_assembler.bltz(cmpTempRegister, 10); + m_assembler.addu(dest, dataTempRegister, src); + m_assembler.xorInsn(cmpTempRegister, dest, dataTempRegister); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + add32(src, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + add32(src, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + add32(src, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchAdd32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + move dataTemp, op1 + xor cmpTemp, dataTemp, op2 + bltz cmpTemp, No_overflow # diff sign bit -> no overflow + addu dest, dataTemp, op2 + xor cmpTemp, dest, dataTemp + bgez cmpTemp, No_overflow # same sign big -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + move(op1, dataTempRegister); + m_assembler.xorInsn(cmpTempRegister, dataTempRegister, op2); + m_assembler.bltz(cmpTempRegister, 10); + m_assembler.addu(dest, dataTempRegister, op2); + m_assembler.xorInsn(cmpTempRegister, dest, dataTempRegister); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + add32(op1, op2, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + add32(op1, op2, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + add32(op1, op2, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + return branchAdd32(cond, immTempRegister, dest); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + move(src, dest); + return branchAdd32(cond, immTempRegister, dest); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, AbsoluteAddress dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + move dataTemp, dest + xori cmpTemp, dataTemp, imm + bltz cmpTemp, No_overflow # diff sign bit -> no overflow + addiu dataTemp, dataTemp, imm + move dest, dataTemp + xori cmpTemp, dataTemp, imm + bgez cmpTemp, No_overflow # same sign big -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + if (imm.m_value >= -32768 && imm.m_value <= 32767 && !m_fixedWidth) { + load32(dest.m_ptr, dataTempRegister); + m_assembler.xori(cmpTempRegister, dataTempRegister, imm.m_value); + m_assembler.bltz(cmpTempRegister, 10); + m_assembler.addiu(dataTempRegister, dataTempRegister, imm.m_value); + store32(dataTempRegister, dest.m_ptr); + m_assembler.xori(cmpTempRegister, dataTempRegister, imm.m_value); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + } else { + load32(dest.m_ptr, dataTempRegister); + move(imm, immTempRegister); + m_assembler.xorInsn(cmpTempRegister, dataTempRegister, immTempRegister); + m_assembler.bltz(cmpTempRegister, 10); + m_assembler.addiu(dataTempRegister, dataTempRegister, immTempRegister); + store32(dataTempRegister, dest.m_ptr); + m_assembler.xori(cmpTempRegister, dataTempRegister, immTempRegister); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + } + return jump(); + } + move(imm, immTempRegister); + load32(dest.m_ptr, dataTempRegister); + add32(immTempRegister, dataTempRegister); + store32(dataTempRegister, dest.m_ptr); + if (cond == Signed) { + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dataTempRegister, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) + return branchEqual(dataTempRegister, MIPSRegisters::zero); + if (cond == NonZero) + return branchNotEqual(dataTempRegister, MIPSRegisters::zero); + ASSERT(0); + return Jump(); + } + + Jump branchMul32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + mult src, dest + mfhi dataTemp + mflo dest + sra addrTemp, dest, 31 + beq dataTemp, addrTemp, No_overflow # all sign bits (bit 63 to bit 31) are the same -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + m_assembler.mult(src1, src2); + m_assembler.mfhi(dataTempRegister); + m_assembler.mflo(dest); + m_assembler.sra(addrTempRegister, dest, 31); + m_assembler.beq(dataTempRegister, addrTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + mul32(src1, src2, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + mul32(src1, src2, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + mul32(src1, src2, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchMul32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + mult src, dest + mfhi dataTemp + mflo dest + sra addrTemp, dest, 31 + beq dataTemp, addrTemp, No_overflow # all sign bits (bit 63 to bit 31) are the same -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + m_assembler.mult(src, dest); + m_assembler.mfhi(dataTempRegister); + m_assembler.mflo(dest); + m_assembler.sra(addrTempRegister, dest, 31); + m_assembler.beq(dataTempRegister, addrTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + mul32(src, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + mul32(src, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + mul32(src, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchMul32(ResultCondition cond, TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(imm, immTempRegister); + return branchMul32(cond, immTempRegister, src, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + move dest, dataTemp + xor cmpTemp, dataTemp, src + bgez cmpTemp, No_overflow # same sign bit -> no overflow + subu dest, dataTemp, src + xor cmpTemp, dest, dataTemp + bgez cmpTemp, No_overflow # same sign bit -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + move(dest, dataTempRegister); + m_assembler.xorInsn(cmpTempRegister, dataTempRegister, src); + m_assembler.bgez(cmpTempRegister, 10); + m_assembler.subu(dest, dataTempRegister, src); + m_assembler.xorInsn(cmpTempRegister, dest, dataTempRegister); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + sub32(src, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + sub32(src, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + sub32(src, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + return branchSub32(cond, immTempRegister, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + move(imm, immTempRegister); + return branchSub32(cond, src, immTempRegister, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID op1, RegisterID op2, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Overflow) { + /* + move dataTemp, op1 + xor cmpTemp, dataTemp, op2 + bgez cmpTemp, No_overflow # same sign bit -> no overflow + subu dest, dataTemp, op2 + xor cmpTemp, dest, dataTemp + bgez cmpTemp, No_overflow # same sign bit -> no overflow + nop + b Overflow + nop + nop + nop + nop + nop + No_overflow: + */ + move(op1, dataTempRegister); + m_assembler.xorInsn(cmpTempRegister, dataTempRegister, op2); + m_assembler.bgez(cmpTempRegister, 10); + m_assembler.subu(dest, dataTempRegister, op2); + m_assembler.xorInsn(cmpTempRegister, dest, dataTempRegister); + m_assembler.bgez(cmpTempRegister, 7); + m_assembler.nop(); + return jump(); + } + if (cond == Signed) { + sub32(op1, op2, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + sub32(op1, op2, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + sub32(op1, op2, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + Jump branchNeg32(ResultCondition cond, RegisterID srcDest) + { + m_assembler.li(dataTempRegister, -1); + return branchMul32(cond, dataTempRegister, srcDest); + } + + Jump branchOr32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Signed) || (cond == Zero) || (cond == NonZero)); + if (cond == Signed) { + or32(src, dest); + // Check if dest is negative. + m_assembler.slt(cmpTempRegister, dest, MIPSRegisters::zero); + return branchNotEqual(cmpTempRegister, MIPSRegisters::zero); + } + if (cond == Zero) { + or32(src, dest); + return branchEqual(dest, MIPSRegisters::zero); + } + if (cond == NonZero) { + or32(src, dest); + return branchNotEqual(dest, MIPSRegisters::zero); + } + ASSERT(0); + return Jump(); + } + + // Miscellaneous operations: + + void breakpoint() + { + m_assembler.bkpt(); + } + + Call nearCall() + { + /* We need two words for relaxation. */ + m_assembler.nop(); + m_assembler.nop(); + m_assembler.jal(); + m_assembler.nop(); + return Call(m_assembler.label(), Call::LinkableNear); + } + + Call call() + { + m_assembler.lui(MIPSRegisters::t9, 0); + m_assembler.ori(MIPSRegisters::t9, MIPSRegisters::t9, 0); + m_assembler.jalr(MIPSRegisters::t9); + m_assembler.nop(); + return Call(m_assembler.label(), Call::Linkable); + } + + Call call(RegisterID target) + { + move(target, MIPSRegisters::t9); + m_assembler.jalr(MIPSRegisters::t9); + m_assembler.nop(); + return Call(m_assembler.label(), Call::None); + } + + Call call(Address address) + { + m_fixedWidth = true; + load32(address, MIPSRegisters::t9); + m_assembler.jalr(MIPSRegisters::t9); + m_assembler.nop(); + m_fixedWidth = false; + return Call(m_assembler.label(), Call::None); + } + + void ret() + { + m_assembler.jr(MIPSRegisters::ra); + m_assembler.nop(); + } + + void compare32(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + if (cond == Equal) { + m_assembler.xorInsn(dest, left, right); + m_assembler.sltiu(dest, dest, 1); + } else if (cond == NotEqual) { + m_assembler.xorInsn(dest, left, right); + m_assembler.sltu(dest, MIPSRegisters::zero, dest); + } else if (cond == Above) + m_assembler.sltu(dest, right, left); + else if (cond == AboveOrEqual) { + m_assembler.sltu(dest, left, right); + m_assembler.xori(dest, dest, 1); + } else if (cond == Below) + m_assembler.sltu(dest, left, right); + else if (cond == BelowOrEqual) { + m_assembler.sltu(dest, right, left); + m_assembler.xori(dest, dest, 1); + } else if (cond == GreaterThan) + m_assembler.slt(dest, right, left); + else if (cond == GreaterThanOrEqual) { + m_assembler.slt(dest, left, right); + m_assembler.xori(dest, dest, 1); + } else if (cond == LessThan) + m_assembler.slt(dest, left, right); + else if (cond == LessThanOrEqual) { + m_assembler.slt(dest, right, left); + m_assembler.xori(dest, dest, 1); + } + } + + void compare32(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + move(right, immTempRegister); + compare32(cond, left, immTempRegister, dest); + } + + void test8(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + ASSERT((cond == Zero) || (cond == NonZero)); + load8(address, dataTempRegister); + if (mask.m_value == -1 && !m_fixedWidth) { + if (cond == Zero) + m_assembler.sltiu(dest, dataTempRegister, 1); + else + m_assembler.sltu(dest, MIPSRegisters::zero, dataTempRegister); + } else { + move(mask, immTempRegister); + m_assembler.andInsn(cmpTempRegister, dataTempRegister, immTempRegister); + if (cond == Zero) + m_assembler.sltiu(dest, cmpTempRegister, 1); + else + m_assembler.sltu(dest, MIPSRegisters::zero, cmpTempRegister); + } + } + + void test32(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + ASSERT((cond == Zero) || (cond == NonZero)); + load32(address, dataTempRegister); + if (mask.m_value == -1 && !m_fixedWidth) { + if (cond == Zero) + m_assembler.sltiu(dest, dataTempRegister, 1); + else + m_assembler.sltu(dest, MIPSRegisters::zero, dataTempRegister); + } else { + move(mask, immTempRegister); + m_assembler.andInsn(cmpTempRegister, dataTempRegister, immTempRegister); + if (cond == Zero) + m_assembler.sltiu(dest, cmpTempRegister, 1); + else + m_assembler.sltu(dest, MIPSRegisters::zero, cmpTempRegister); + } + } + + DataLabel32 moveWithPatch(TrustedImm32 imm, RegisterID dest) + { + m_fixedWidth = true; + DataLabel32 label(this); + move(imm, dest); + m_fixedWidth = false; + return label; + } + + DataLabelPtr moveWithPatch(TrustedImmPtr initialValue, RegisterID dest) + { + m_fixedWidth = true; + DataLabelPtr label(this); + move(initialValue, dest); + m_fixedWidth = false; + return label; + } + + Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + m_fixedWidth = true; + dataLabel = moveWithPatch(initialRightValue, immTempRegister); + Jump temp = branch32(cond, left, immTempRegister); + m_fixedWidth = false; + return temp; + } + + Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + m_fixedWidth = true; + load32(left, dataTempRegister); + dataLabel = moveWithPatch(initialRightValue, immTempRegister); + Jump temp = branch32(cond, dataTempRegister, immTempRegister); + m_fixedWidth = false; + return temp; + } + + DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + m_fixedWidth = true; + DataLabelPtr dataLabel = moveWithPatch(initialValue, dataTempRegister); + store32(dataTempRegister, address); + m_fixedWidth = false; + return dataLabel; + } + + DataLabelPtr storePtrWithPatch(ImplicitAddress address) + { + return storePtrWithPatch(TrustedImmPtr(0), address); + } + + Call tailRecursiveCall() + { + // Like a normal call, but don't update the returned address register + m_fixedWidth = true; + move(TrustedImm32(0), MIPSRegisters::t9); + m_assembler.jr(MIPSRegisters::t9); + m_assembler.nop(); + m_fixedWidth = false; + return Call(m_assembler.label(), Call::Linkable); + } + + Call makeTailRecursiveCall(Jump oldJump) + { + oldJump.link(this); + return tailRecursiveCall(); + } + + void loadFloat(BaseIndex address, FPRegisterID dest) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lwc1 dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lwc1(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lwc1 dest, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lwc1(dest, addrTempRegister, address.offset); + } + } + + void loadDouble(ImplicitAddress address, FPRegisterID dest) + { +#if WTF_MIPS_ISA(1) + /* + li addrTemp, address.offset + addu addrTemp, addrTemp, base + lwc1 dest, 0(addrTemp) + lwc1 dest+1, 4(addrTemp) + */ + move(TrustedImm32(address.offset), addrTempRegister); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lwc1(dest, addrTempRegister, 0); + m_assembler.lwc1(FPRegisterID(dest + 1), addrTempRegister, 4); +#else + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + m_assembler.ldc1(dest, address.base, address.offset); + } else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + ldc1 dest, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.ldc1(dest, addrTempRegister, address.offset); + } +#endif + } + + void loadDouble(BaseIndex address, FPRegisterID dest) + { +#if WTF_MIPS_ISA(1) + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lwc1 dest, address.offset(addrTemp) + lwc1 dest+1, (address.offset+4)(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lwc1(dest, addrTempRegister, address.offset); + m_assembler.lwc1(FPRegisterID(dest + 1), addrTempRegister, address.offset + 4); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + lwc1 dest, (address.offset & 0xffff)(at) + lwc1 dest+1, (address.offset & 0xffff + 4)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.lwc1(dest, addrTempRegister, address.offset); + m_assembler.lwc1(FPRegisterID(dest + 1), addrTempRegister, address.offset + 4); + } +#else + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + ldc1 dest, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.ldc1(dest, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + ldc1 dest, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.ldc1(dest, addrTempRegister, address.offset); + } +#endif + } + + void loadDouble(const void* address, FPRegisterID dest) + { +#if WTF_MIPS_ISA(1) + /* + li addrTemp, address + lwc1 dest, 0(addrTemp) + lwc1 dest+1, 4(addrTemp) + */ + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.lwc1(dest, addrTempRegister, 0); + m_assembler.lwc1(FPRegisterID(dest + 1), addrTempRegister, 4); +#else + /* + li addrTemp, address + ldc1 dest, 0(addrTemp) + */ + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.ldc1(dest, addrTempRegister, 0); +#endif + } + + void storeFloat(FPRegisterID src, BaseIndex address) + { + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + swc1 src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.swc1(src, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + swc1 src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.swc1(src, addrTempRegister, address.offset); + } + } + + void storeDouble(FPRegisterID src, ImplicitAddress address) + { +#if WTF_MIPS_ISA(1) + /* + li addrTemp, address.offset + addu addrTemp, addrTemp, base + swc1 dest, 0(addrTemp) + swc1 dest+1, 4(addrTemp) + */ + move(TrustedImm32(address.offset), addrTempRegister); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.swc1(src, addrTempRegister, 0); + m_assembler.swc1(FPRegisterID(src + 1), addrTempRegister, 4); +#else + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) + m_assembler.sdc1(src, address.base, address.offset); + else { + /* + lui addrTemp, (offset + 0x8000) >> 16 + addu addrTemp, addrTemp, base + sdc1 src, (offset & 0xffff)(addrTemp) + */ + m_assembler.lui(addrTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sdc1(src, addrTempRegister, address.offset); + } +#endif + } + + void storeDouble(FPRegisterID src, BaseIndex address) + { +#if WTF_MIPS_ISA(1) + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + swc1 src, address.offset(addrTemp) + swc1 src+1, (address.offset + 4)(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.swc1(src, addrTempRegister, address.offset); + m_assembler.swc1(FPRegisterID(src + 1), addrTempRegister, address.offset + 4); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + swc1 src, (address.offset & 0xffff)(at) + swc1 src+1, (address.offset & 0xffff + 4)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.swc1(src, addrTempRegister, address.offset); + m_assembler.swc1(FPRegisterID(src + 1), addrTempRegister, address.offset + 4); + } +#else + if (address.offset >= -32768 && address.offset <= 32767 + && !m_fixedWidth) { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + sdc1 src, address.offset(addrTemp) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.sdc1(src, addrTempRegister, address.offset); + } else { + /* + sll addrTemp, address.index, address.scale + addu addrTemp, addrTemp, address.base + lui immTemp, (address.offset + 0x8000) >> 16 + addu addrTemp, addrTemp, immTemp + sdc1 src, (address.offset & 0xffff)(at) + */ + m_assembler.sll(addrTempRegister, address.index, address.scale); + m_assembler.addu(addrTempRegister, addrTempRegister, address.base); + m_assembler.lui(immTempRegister, (address.offset + 0x8000) >> 16); + m_assembler.addu(addrTempRegister, addrTempRegister, immTempRegister); + m_assembler.sdc1(src, addrTempRegister, address.offset); + } +#endif + } + + void storeDouble(FPRegisterID src, const void* address) + { +#if WTF_MIPS_ISA(1) + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.swc1(src, addrTempRegister, 0); + m_assembler.swc1(FPRegisterID(src + 1), addrTempRegister, 4); +#else + move(TrustedImmPtr(address), addrTempRegister); + m_assembler.sdc1(src, addrTempRegister, 0); +#endif + } + + void moveDouble(FPRegisterID src, FPRegisterID dest) + { + if (src != dest || m_fixedWidth) + m_assembler.movd(dest, src); + } + + void swapDouble(FPRegisterID fr1, FPRegisterID fr2) + { + moveDouble(fr1, fpTempRegister); + moveDouble(fr2, fr1); + moveDouble(fpTempRegister, fr2); + } + + void addDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.addd(dest, dest, src); + } + + void addDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.addd(dest, op1, op2); + } + + void addDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + m_assembler.addd(dest, dest, fpTempRegister); + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + loadDouble(address.m_ptr, fpTempRegister); + m_assembler.addd(dest, dest, fpTempRegister); + } + + void subDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.subd(dest, dest, src); + } + + void subDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.subd(dest, op1, op2); + } + + void subDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + m_assembler.subd(dest, dest, fpTempRegister); + } + + void mulDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.muld(dest, dest, src); + } + + void mulDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + m_assembler.muld(dest, dest, fpTempRegister); + } + + void mulDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.muld(dest, op1, op2); + } + + void divDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.divd(dest, dest, src); + } + + void divDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + m_assembler.divd(dest, op1, op2); + } + + void divDouble(Address src, FPRegisterID dest) + { + loadDouble(src, fpTempRegister); + m_assembler.divd(dest, dest, fpTempRegister); + } + + void negateDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.negd(dest, src); + } + + void convertInt32ToDouble(RegisterID src, FPRegisterID dest) + { + m_assembler.mtc1(src, fpTempRegister); + m_assembler.cvtdw(dest, fpTempRegister); + } + + void convertInt32ToDouble(Address src, FPRegisterID dest) + { + load32(src, dataTempRegister); + m_assembler.mtc1(dataTempRegister, fpTempRegister); + m_assembler.cvtdw(dest, fpTempRegister); + } + + void convertInt32ToDouble(AbsoluteAddress src, FPRegisterID dest) + { + load32(src.m_ptr, dataTempRegister); + m_assembler.mtc1(dataTempRegister, fpTempRegister); + m_assembler.cvtdw(dest, fpTempRegister); + } + + void convertFloatToDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.cvtds(dst, src); + } + + void convertDoubleToFloat(FPRegisterID src, FPRegisterID dst) + { + m_assembler.cvtsd(dst, src); + } + + void insertRelaxationWords() + { + /* We need four words for relaxation. */ + m_assembler.beq(MIPSRegisters::zero, MIPSRegisters::zero, 3); // Jump over nops; + m_assembler.nop(); + m_assembler.nop(); + m_assembler.nop(); + } + + Jump branchTrue() + { + m_assembler.appendJump(); + m_assembler.bc1t(); + m_assembler.nop(); + insertRelaxationWords(); + return Jump(m_assembler.label()); + } + + Jump branchFalse() + { + m_assembler.appendJump(); + m_assembler.bc1f(); + m_assembler.nop(); + insertRelaxationWords(); + return Jump(m_assembler.label()); + } + + Jump branchEqual(RegisterID rs, RegisterID rt) + { + m_assembler.nop(); + m_assembler.nop(); + m_assembler.appendJump(); + m_assembler.beq(rs, rt, 0); + m_assembler.nop(); + insertRelaxationWords(); + return Jump(m_assembler.label()); + } + + Jump branchNotEqual(RegisterID rs, RegisterID rt) + { + m_assembler.nop(); + m_assembler.nop(); + m_assembler.appendJump(); + m_assembler.bne(rs, rt, 0); + m_assembler.nop(); + insertRelaxationWords(); + return Jump(m_assembler.label()); + } + + Jump branchDouble(DoubleCondition cond, FPRegisterID left, FPRegisterID right) + { + if (cond == DoubleEqual) { + m_assembler.ceqd(left, right); + return branchTrue(); + } + if (cond == DoubleNotEqual) { + m_assembler.cueqd(left, right); + return branchFalse(); // false + } + if (cond == DoubleGreaterThan) { + m_assembler.cngtd(left, right); + return branchFalse(); // false + } + if (cond == DoubleGreaterThanOrEqual) { + m_assembler.cnged(left, right); + return branchFalse(); // false + } + if (cond == DoubleLessThan) { + m_assembler.cltd(left, right); + return branchTrue(); + } + if (cond == DoubleLessThanOrEqual) { + m_assembler.cled(left, right); + return branchTrue(); + } + if (cond == DoubleEqualOrUnordered) { + m_assembler.cueqd(left, right); + return branchTrue(); + } + if (cond == DoubleNotEqualOrUnordered) { + m_assembler.ceqd(left, right); + return branchFalse(); // false + } + if (cond == DoubleGreaterThanOrUnordered) { + m_assembler.coled(left, right); + return branchFalse(); // false + } + if (cond == DoubleGreaterThanOrEqualOrUnordered) { + m_assembler.coltd(left, right); + return branchFalse(); // false + } + if (cond == DoubleLessThanOrUnordered) { + m_assembler.cultd(left, right); + return branchTrue(); + } + if (cond == DoubleLessThanOrEqualOrUnordered) { + m_assembler.culed(left, right); + return branchTrue(); + } + ASSERT(0); + + return Jump(); + } + + // Truncates 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, INT_MAX 0x7fffffff). + enum BranchTruncateType { BranchIfTruncateFailed, BranchIfTruncateSuccessful }; + Jump branchTruncateDoubleToInt32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + m_assembler.truncwd(fpTempRegister, src); + m_assembler.mfc1(dest, fpTempRegister); + return branch32(branchType == BranchIfTruncateFailed ? Equal : NotEqual, dest, TrustedImm32(0x7fffffff)); + } + + Jump branchTruncateDoubleToUint32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + m_assembler.truncwd(fpTempRegister, src); + m_assembler.mfc1(dest, fpTempRegister); + return branch32(branchType == BranchIfTruncateFailed ? Equal : NotEqual, dest, TrustedImm32(0)); + } + + // Result is undefined if the value is outside of the integer range. + void truncateDoubleToInt32(FPRegisterID src, RegisterID dest) + { + m_assembler.truncwd(fpTempRegister, src); + m_assembler.mfc1(dest, fpTempRegister); + } + + // Result is undefined if src > 2^31 + void truncateDoubleToUint32(FPRegisterID src, RegisterID dest) + { + m_assembler.truncwd(fpTempRegister, src); + m_assembler.mfc1(dest, fpTempRegister); + } + + // Convert 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, 0). + void branchConvertDoubleToInt32(FPRegisterID src, RegisterID dest, JumpList& failureCases, FPRegisterID fpTemp) + { + m_assembler.cvtwd(fpTempRegister, src); + m_assembler.mfc1(dest, fpTempRegister); + + // If the result is zero, it might have been -0.0, and the double comparison won't catch this! + failureCases.append(branch32(Equal, dest, MIPSRegisters::zero)); + + // Convert the integer result back to float & compare to the original value - if not equal or unordered (NaN) then jump. + convertInt32ToDouble(dest, fpTemp); + failureCases.append(branchDouble(DoubleNotEqualOrUnordered, fpTemp, src)); + } + + Jump branchDoubleNonZero(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.vmov(scratch, MIPSRegisters::zero, MIPSRegisters::zero); + return branchDouble(DoubleNotEqual, reg, scratch); + } + + Jump branchDoubleZeroOrNaN(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.vmov(scratch, MIPSRegisters::zero, MIPSRegisters::zero); + return branchDouble(DoubleEqualOrUnordered, reg, scratch); + } + + // Invert a relational condition, e.g. == becomes !=, < becomes >=, etc. + static RelationalCondition invert(RelationalCondition cond) + { + RelationalCondition r; + if (cond == Equal) + r = NotEqual; + else if (cond == NotEqual) + r = Equal; + else if (cond == Above) + r = BelowOrEqual; + else if (cond == AboveOrEqual) + r = Below; + else if (cond == Below) + r = AboveOrEqual; + else if (cond == BelowOrEqual) + r = Above; + else if (cond == GreaterThan) + r = LessThanOrEqual; + else if (cond == GreaterThanOrEqual) + r = LessThan; + else if (cond == LessThan) + r = GreaterThanOrEqual; + else if (cond == LessThanOrEqual) + r = GreaterThan; + return r; + } + + void nop() + { + m_assembler.nop(); + } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + return FunctionPtr(reinterpret_cast<void(*)()>(MIPSAssembler::readCallTarget(call.dataLocation()))); + } + + static void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + MIPSAssembler::replaceWithJump(instructionStart.dataLocation(), destination.dataLocation()); + } + + static ptrdiff_t maxJumpReplacementSize() + { + MIPSAssembler::maxJumpReplacementSize(); + return 0; + } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return false; } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + return label.labelAtOffset(0); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID, void* initialValue) + { + MIPSAssembler::revertJumpToMove(instructionStart.dataLocation(), immTempRegister, reinterpret_cast<int>(initialValue) & 0xffff); + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr) + { + UNREACHABLE_FOR_PLATFORM(); + return CodeLocationLabel(); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel instructionStart, Address, void* initialValue) + { + UNREACHABLE_FOR_PLATFORM(); + } + + +private: + // If m_fixedWidth is true, we will generate a fixed number of instructions. + // Otherwise, we can emit any number of instructions. + bool m_fixedWidth; + + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkCall(void* code, Call call, FunctionPtr function) + { + MIPSAssembler::linkCall(code, call.m_label, function.value()); + } + + static void repatchCall(CodeLocationCall call, CodeLocationLabel destination) + { + MIPSAssembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + static void repatchCall(CodeLocationCall call, FunctionPtr destination) + { + MIPSAssembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + +}; + +} + +#endif // ENABLE(ASSEMBLER) && CPU(MIPS) + +#endif // MacroAssemblerMIPS_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerSH4.cpp b/src/3rdparty/masm/assembler/MacroAssemblerSH4.cpp new file mode 100644 index 0000000000..59de3ff48c --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerSH4.cpp @@ -0,0 +1,52 @@ +/* + * Copyright (C) 2011 STMicroelectronics. All rights reserved. + * Copyright (C) 2008 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "config.h" + +#if ENABLE(ASSEMBLER) && CPU(SH4) + +#include "MacroAssemblerSH4.h" + +namespace JSC { + +void MacroAssemblerSH4::linkCall(void* code, Call call, FunctionPtr function) +{ + SH4Assembler::linkCall(code, call.m_label, function.value()); +} + +void MacroAssemblerSH4::repatchCall(CodeLocationCall call, CodeLocationLabel destination) +{ + SH4Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); +} + +void MacroAssemblerSH4::repatchCall(CodeLocationCall call, FunctionPtr destination) +{ + SH4Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); +} + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) diff --git a/src/3rdparty/masm/assembler/MacroAssemblerSH4.h b/src/3rdparty/masm/assembler/MacroAssemblerSH4.h new file mode 100644 index 0000000000..56fb74d45b --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerSH4.h @@ -0,0 +1,2293 @@ +/* + * Copyright (C) 2009-2011 STMicroelectronics. All rights reserved. + * Copyright (C) 2008 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MacroAssemblerSH4_h +#define MacroAssemblerSH4_h + +#if ENABLE(ASSEMBLER) && CPU(SH4) + +#include "SH4Assembler.h" +#include "AbstractMacroAssembler.h" +#include <wtf/Assertions.h> + +namespace JSC { + +class MacroAssemblerSH4 : public AbstractMacroAssembler<SH4Assembler> { +public: + typedef SH4Assembler::FPRegisterID FPRegisterID; + + static const Scale ScalePtr = TimesFour; + static const FPRegisterID fscratch = SH4Registers::fr10; + static const RegisterID stackPointerRegister = SH4Registers::sp; + static const RegisterID linkRegister = SH4Registers::pr; + static const RegisterID scratchReg3 = SH4Registers::r13; + + static const int MaximumCompactPtrAlignedAddressOffset = 60; + + static bool isCompactPtrAlignedAddressOffset(ptrdiff_t value) + { + return (value >= 0) && (value <= MaximumCompactPtrAlignedAddressOffset); + } + + enum RelationalCondition { + Equal = SH4Assembler::EQ, + NotEqual = SH4Assembler::NE, + Above = SH4Assembler::HI, + AboveOrEqual = SH4Assembler::HS, + Below = SH4Assembler::LI, + BelowOrEqual = SH4Assembler::LS, + GreaterThan = SH4Assembler::GT, + GreaterThanOrEqual = SH4Assembler::GE, + LessThan = SH4Assembler::LT, + LessThanOrEqual = SH4Assembler::LE + }; + + enum ResultCondition { + Overflow = SH4Assembler::OF, + Signed = SH4Assembler::SI, + Zero = SH4Assembler::EQ, + NonZero = SH4Assembler::NE + }; + + enum DoubleCondition { + // These conditions will only evaluate to true if the comparison is ordered - i.e. neither operand is NaN. + DoubleEqual = SH4Assembler::EQ, + DoubleNotEqual = SH4Assembler::NE, + DoubleGreaterThan = SH4Assembler::GT, + DoubleGreaterThanOrEqual = SH4Assembler::GE, + DoubleLessThan = SH4Assembler::LT, + DoubleLessThanOrEqual = SH4Assembler::LE, + // If either operand is NaN, these conditions always evaluate to true. + DoubleEqualOrUnordered = SH4Assembler::EQU, + DoubleNotEqualOrUnordered = SH4Assembler::NEU, + DoubleGreaterThanOrUnordered = SH4Assembler::GTU, + DoubleGreaterThanOrEqualOrUnordered = SH4Assembler::GEU, + DoubleLessThanOrUnordered = SH4Assembler::LTU, + DoubleLessThanOrEqualOrUnordered = SH4Assembler::LEU, + }; + + RegisterID claimScratch() + { + return m_assembler.claimScratch(); + } + + void releaseScratch(RegisterID reg) + { + m_assembler.releaseScratch(reg); + } + + // Integer arithmetic operations + + void add32(RegisterID src, RegisterID dest) + { + m_assembler.addlRegReg(src, dest); + } + + void add32(TrustedImm32 imm, RegisterID dest) + { + if (m_assembler.isImmediate(imm.m_value)) { + m_assembler.addlImm8r(imm.m_value, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm.m_value, scr); + m_assembler.addlRegReg(scr, dest); + releaseScratch(scr); + } + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (src != dest) + m_assembler.movlRegReg(src, dest); + add32(imm, dest); + } + + void add32(TrustedImm32 imm, Address address) + { + RegisterID scr = claimScratch(); + load32(address, scr); + add32(imm, scr); + store32(scr, address); + releaseScratch(scr); + } + + void add32(Address src, RegisterID dest) + { + RegisterID scr = claimScratch(); + load32(src, scr); + m_assembler.addlRegReg(scr, dest); + releaseScratch(scr); + } + + void add32(AbsoluteAddress src, RegisterID dest) + { + RegisterID scr = claimScratch(); + load32(src.m_ptr, scr); + m_assembler.addlRegReg(scr, dest); + releaseScratch(scr); + } + + void and32(RegisterID src, RegisterID dest) + { + m_assembler.andlRegReg(src, dest); + } + + void and32(TrustedImm32 imm, RegisterID dest) + { + if ((imm.m_value <= 255) && (imm.m_value >= 0) && (dest == SH4Registers::r0)) { + m_assembler.andlImm8r(imm.m_value, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant((imm.m_value), scr); + m_assembler.andlRegReg(scr, dest); + releaseScratch(scr); + } + + void and32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (src != dest) { + move(imm, dest); + and32(src, dest); + return; + } + + and32(imm, dest); + } + + void lshift32(RegisterID shiftamount, RegisterID dest) + { + if (shiftamount == SH4Registers::r0) + m_assembler.andlImm8r(0x1f, shiftamount); + else { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(0x1f, scr); + m_assembler.andlRegReg(scr, shiftamount); + releaseScratch(scr); + } + m_assembler.shllRegReg(dest, shiftamount); + } + + void rshift32(int imm, RegisterID dest) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(-imm, scr); + m_assembler.shaRegReg(dest, scr); + releaseScratch(scr); + } + + void lshift32(TrustedImm32 imm, RegisterID dest) + { + if (!imm.m_value) + return; + + if ((imm.m_value == 1) || (imm.m_value == 2) || (imm.m_value == 8) || (imm.m_value == 16)) { + m_assembler.shllImm8r(imm.m_value, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant((imm.m_value & 0x1f) , scr); + m_assembler.shllRegReg(dest, scr); + releaseScratch(scr); + } + + void lshift32(RegisterID src, TrustedImm32 shiftamount, RegisterID dest) + { + if (src != dest) + move(src, dest); + + lshift32(shiftamount, dest); + } + + void mul32(RegisterID src, RegisterID dest) + { + m_assembler.imullRegReg(src, dest); + m_assembler.stsmacl(dest); + } + + void mul32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + RegisterID scr = claimScratch(); + move(imm, scr); + if (src != dest) + move(src, dest); + mul32(scr, dest); + releaseScratch(scr); + } + + void or32(RegisterID src, RegisterID dest) + { + m_assembler.orlRegReg(src, dest); + } + + void or32(TrustedImm32 imm, RegisterID dest) + { + if ((imm.m_value <= 255) && (imm.m_value >= 0) && (dest == SH4Registers::r0)) { + m_assembler.orlImm8r(imm.m_value, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm.m_value, scr); + m_assembler.orlRegReg(scr, dest); + releaseScratch(scr); + } + + void or32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op1 == op2) + move(op1, dest); + else if (op1 == dest) + or32(op2, dest); + else { + move(op2, dest); + or32(op1, dest); + } + } + + +void or32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (src != dest) { + move(imm, dest); + or32(src, dest); + return; + } + + or32(imm, dest); + } + + void xor32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + if (src != dest) { + move(imm, dest); + xor32(src, dest); + return; + } + + xor32(imm, dest); + } + + void rshift32(RegisterID shiftamount, RegisterID dest) + { + if (shiftamount == SH4Registers::r0) + m_assembler.andlImm8r(0x1f, shiftamount); + else { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(0x1f, scr); + m_assembler.andlRegReg(scr, shiftamount); + releaseScratch(scr); + } + m_assembler.neg(shiftamount, shiftamount); + m_assembler.shaRegReg(dest, shiftamount); + } + + void rshift32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value & 0x1f) + rshift32(imm.m_value & 0x1f, dest); + } + + void rshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + if (src != dest) + move(src, dest); + rshift32(imm, dest); + } + + void sub32(RegisterID src, RegisterID dest) + { + m_assembler.sublRegReg(src, dest); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address, RegisterID scratchReg) + { + RegisterID result = claimScratch(); + + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scratchReg); + m_assembler.movlMemReg(scratchReg, result); + + if (m_assembler.isImmediate(-imm.m_value)) + m_assembler.addlImm8r(-imm.m_value, result); + else { + m_assembler.loadConstant(imm.m_value, scratchReg3); + m_assembler.sublRegReg(scratchReg3, result); + } + + store32(result, scratchReg); + releaseScratch(result); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + RegisterID result = claimScratch(); + RegisterID scratchReg = claimScratch(); + + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scratchReg); + m_assembler.movlMemReg(scratchReg, result); + + if (m_assembler.isImmediate(-imm.m_value)) + m_assembler.addlImm8r(-imm.m_value, result); + else { + m_assembler.loadConstant(imm.m_value, scratchReg3); + m_assembler.sublRegReg(scratchReg3, result); + } + + store32(result, scratchReg); + releaseScratch(result); + releaseScratch(scratchReg); + } + + void add32(TrustedImm32 imm, AbsoluteAddress address, RegisterID scratchReg) + { + RegisterID result = claimScratch(); + + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scratchReg); + m_assembler.movlMemReg(scratchReg, result); + + if (m_assembler.isImmediate(imm.m_value)) + m_assembler.addlImm8r(imm.m_value, result); + else { + m_assembler.loadConstant(imm.m_value, scratchReg3); + m_assembler.addlRegReg(scratchReg3, result); + } + + store32(result, scratchReg); + releaseScratch(result); + } + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + RegisterID result = claimScratch(); + RegisterID scratchReg = claimScratch(); + + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scratchReg); + m_assembler.movlMemReg(scratchReg, result); + + if (m_assembler.isImmediate(imm.m_value)) + m_assembler.addlImm8r(imm.m_value, result); + else { + m_assembler.loadConstant(imm.m_value, scratchReg3); + m_assembler.addlRegReg(scratchReg3, result); + } + + store32(result, scratchReg); + releaseScratch(result); + releaseScratch(scratchReg); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + RegisterID scr1 = claimScratch(); + RegisterID scr2 = claimScratch(); + + // Add 32-bit LSB first. + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scr1); + m_assembler.movlMemReg(scr1, scr1); // scr1 = 32-bit LSB of int64 @ address + m_assembler.loadConstant(imm.m_value, scr2); + m_assembler.clrt(); + m_assembler.addclRegReg(scr1, scr2); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr), scr1); + m_assembler.movlRegMem(scr2, scr1); // Update address with 32-bit LSB result. + + // Then add 32-bit MSB. + m_assembler.addlImm8r(4, scr1); + m_assembler.movlMemReg(scr1, scr1); // scr1 = 32-bit MSB of int64 @ address + m_assembler.movt(scr2); + if (imm.m_value < 0) + m_assembler.addlImm8r(-1, scr2); // Sign extend imm value if needed. + m_assembler.addvlRegReg(scr2, scr1); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address.m_ptr) + 4, scr2); + m_assembler.movlRegMem(scr1, scr2); // Update (address + 4) with 32-bit MSB result. + + releaseScratch(scr2); + releaseScratch(scr1); + } + + void sub32(TrustedImm32 imm, RegisterID dest) + { + if (m_assembler.isImmediate(-imm.m_value)) { + m_assembler.addlImm8r(-imm.m_value, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm.m_value, scr); + m_assembler.sublRegReg(scr, dest); + releaseScratch(scr); + } + + void sub32(Address src, RegisterID dest) + { + RegisterID scr = claimScratch(); + load32(src, scr); + m_assembler.sublRegReg(scr, dest); + releaseScratch(scr); + } + + void xor32(RegisterID src, RegisterID dest) + { + m_assembler.xorlRegReg(src, dest); + } + + void xor32(TrustedImm32 imm, RegisterID srcDest) + { + if (imm.m_value == -1) { + m_assembler.notlReg(srcDest, srcDest); + return; + } + + if ((srcDest != SH4Registers::r0) || (imm.m_value > 255) || (imm.m_value < 0)) { + RegisterID scr = claimScratch(); + m_assembler.loadConstant((imm.m_value), scr); + m_assembler.xorlRegReg(scr, srcDest); + releaseScratch(scr); + return; + } + + m_assembler.xorlImm8r(imm.m_value, srcDest); + } + + void compare32(int imm, RegisterID dst, RelationalCondition cond) + { + if (((cond == Equal) || (cond == NotEqual)) && (dst == SH4Registers::r0) && m_assembler.isImmediate(imm)) { + m_assembler.cmpEqImmR0(imm, dst); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm, scr); + m_assembler.cmplRegReg(scr, dst, SH4Condition(cond)); + releaseScratch(scr); + } + + void compare32(int offset, RegisterID base, RegisterID left, RelationalCondition cond) + { + RegisterID scr = claimScratch(); + if (!offset) { + m_assembler.movlMemReg(base, scr); + m_assembler.cmplRegReg(scr, left, SH4Condition(cond)); + releaseScratch(scr); + return; + } + + if ((offset < 0) || (offset >= 64)) { + m_assembler.loadConstant(offset, scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movlMemReg(scr, scr); + m_assembler.cmplRegReg(scr, left, SH4Condition(cond)); + releaseScratch(scr); + return; + } + + m_assembler.movlMemReg(offset >> 2, base, scr); + m_assembler.cmplRegReg(scr, left, SH4Condition(cond)); + releaseScratch(scr); + } + + void testImm(int imm, int offset, RegisterID base) + { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + + if ((offset < 0) || (offset >= 64)) { + m_assembler.loadConstant(offset, scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movlMemReg(scr, scr); + } else if (offset) + m_assembler.movlMemReg(offset >> 2, base, scr); + else + m_assembler.movlMemReg(base, scr); + if (m_assembler.isImmediate(imm)) + m_assembler.movImm8(imm, scr1); + else + m_assembler.loadConstant(imm, scr1); + + m_assembler.testlRegReg(scr, scr1); + releaseScratch(scr); + releaseScratch(scr1); + } + + void testlImm(int imm, RegisterID dst) + { + if ((dst == SH4Registers::r0) && (imm <= 255) && (imm >= 0)) { + m_assembler.testlImm8r(imm, dst); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm, scr); + m_assembler.testlRegReg(scr, dst); + releaseScratch(scr); + } + + void compare32(RegisterID right, int offset, RegisterID base, RelationalCondition cond) + { + if (!offset) { + RegisterID scr = claimScratch(); + m_assembler.movlMemReg(base, scr); + m_assembler.cmplRegReg(right, scr, SH4Condition(cond)); + releaseScratch(scr); + return; + } + + if ((offset < 0) || (offset >= 64)) { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(offset, scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movlMemReg(scr, scr); + m_assembler.cmplRegReg(right, scr, SH4Condition(cond)); + releaseScratch(scr); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.movlMemReg(offset >> 2, base, scr); + m_assembler.cmplRegReg(right, scr, SH4Condition(cond)); + releaseScratch(scr); + } + + void compare32(int imm, int offset, RegisterID base, RelationalCondition cond) + { + if (!offset) { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + m_assembler.movlMemReg(base, scr); + m_assembler.loadConstant(imm, scr1); + m_assembler.cmplRegReg(scr1, scr, SH4Condition(cond)); + releaseScratch(scr1); + releaseScratch(scr); + return; + } + + if ((offset < 0) || (offset >= 64)) { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + m_assembler.loadConstant(offset, scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movlMemReg(scr, scr); + m_assembler.loadConstant(imm, scr1); + m_assembler.cmplRegReg(scr1, scr, SH4Condition(cond)); + releaseScratch(scr1); + releaseScratch(scr); + return; + } + + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + m_assembler.movlMemReg(offset >> 2, base, scr); + m_assembler.loadConstant(imm, scr1); + m_assembler.cmplRegReg(scr1, scr, SH4Condition(cond)); + releaseScratch(scr1); + releaseScratch(scr); + } + + // Memory access operation + + void load32(ImplicitAddress address, RegisterID dest) + { + load32(address.base, address.offset, dest); + } + + void load8(ImplicitAddress address, RegisterID dest) + { + load8(address.base, address.offset, dest); + } + + void load8(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + load8(scr, address.offset, dest); + releaseScratch(scr); + } + + void load8Signed(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + load8Signed(scr, address.offset, dest); + releaseScratch(scr); + } + + void load32(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + load32(scr, address.offset, dest); + releaseScratch(scr); + } + + void load32(const void* address, RegisterID dest) + { + m_assembler.loadConstant(reinterpret_cast<uint32_t>(const_cast<void*>(address)), dest); + m_assembler.movlMemReg(dest, dest); + } + + void load32(RegisterID base, int offset, RegisterID dest) + { + if (!offset) { + m_assembler.movlMemReg(base, dest); + return; + } + + if ((offset >= 0) && (offset < 64)) { + m_assembler.movlMemReg(offset >> 2, base, dest); + return; + } + + if ((dest == SH4Registers::r0) && (dest != base)) { + m_assembler.loadConstant((offset), dest); + m_assembler.movlR0mr(base, dest); + return; + } + + RegisterID scr; + if (dest == base) + scr = claimScratch(); + else + scr = dest; + m_assembler.loadConstant((offset), scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movlMemReg(scr, dest); + + if (dest == base) + releaseScratch(scr); + } + + void load8Signed(RegisterID base, int offset, RegisterID dest) + { + if (!offset) { + m_assembler.movbMemReg(base, dest); + return; + } + + if ((offset > 0) && (offset < 64) && (dest == SH4Registers::r0)) { + m_assembler.movbMemReg(offset, base, dest); + return; + } + + if (base != dest) { + m_assembler.loadConstant((offset), dest); + m_assembler.addlRegReg(base, dest); + m_assembler.movbMemReg(dest, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant((offset), scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movbMemReg(scr, dest); + releaseScratch(scr); + } + + void load8(RegisterID base, int offset, RegisterID dest) + { + if (!offset) { + m_assembler.movbMemReg(base, dest); + m_assembler.extub(dest, dest); + return; + } + + if ((offset > 0) && (offset < 64) && (dest == SH4Registers::r0)) { + m_assembler.movbMemReg(offset, base, dest); + m_assembler.extub(dest, dest); + return; + } + + if (base != dest) { + m_assembler.loadConstant((offset), dest); + m_assembler.addlRegReg(base, dest); + m_assembler.movbMemReg(dest, dest); + m_assembler.extub(dest, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant((offset), scr); + m_assembler.addlRegReg(base, scr); + m_assembler.movbMemReg(scr, dest); + m_assembler.extub(dest, dest); + releaseScratch(scr); + } + + void load32(RegisterID r0, RegisterID src, RegisterID dst) + { + ASSERT(r0 == SH4Registers::r0); + m_assembler.movlR0mr(src, dst); + } + + void load32(RegisterID src, RegisterID dst) + { + m_assembler.movlMemReg(src, dst); + } + + void load16(ImplicitAddress address, RegisterID dest) + { + if (!address.offset) { + m_assembler.movwMemReg(address.base, dest); + extuw(dest, dest); + return; + } + + if ((address.offset > 0) && (address.offset < 64) && (dest == SH4Registers::r0)) { + m_assembler.movwMemReg(address.offset, address.base, dest); + extuw(dest, dest); + return; + } + + if (address.base != dest) { + m_assembler.loadConstant((address.offset), dest); + m_assembler.addlRegReg(address.base, dest); + m_assembler.movwMemReg(dest, dest); + extuw(dest, dest); + return; + } + + RegisterID scr = claimScratch(); + m_assembler.loadConstant((address.offset), scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movwMemReg(scr, dest); + extuw(dest, dest); + releaseScratch(scr); + } + + void load16Unaligned(BaseIndex address, RegisterID dest) + { + + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + add32(address.base, scr); + load8(scr, scr1); + add32(TrustedImm32(1), scr); + load8(scr, dest); + m_assembler.shllImm8r(8, dest); + or32(scr1, dest); + + releaseScratch(scr); + releaseScratch(scr1); + } + + void load16(RegisterID src, RegisterID dest) + { + m_assembler.movwMemReg(src, dest); + extuw(dest, dest); + } + + void load16Signed(RegisterID src, RegisterID dest) + { + m_assembler.movwMemReg(src, dest); + } + + void load16(RegisterID r0, RegisterID src, RegisterID dest) + { + ASSERT(r0 == SH4Registers::r0); + m_assembler.movwR0mr(src, dest); + extuw(dest, dest); + } + + void load16Signed(RegisterID r0, RegisterID src, RegisterID dest) + { + ASSERT(r0 == SH4Registers::r0); + m_assembler.movwR0mr(src, dest); + } + + void load16(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + + if (address.offset) + add32(TrustedImm32(address.offset), scr); + if (address.base == SH4Registers::r0) + load16(address.base, scr, dest); + else { + add32(address.base, scr); + load16(scr, dest); + } + + releaseScratch(scr); + } + + void load16Signed(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + + if (address.offset) + add32(TrustedImm32(address.offset), scr); + if (address.base == SH4Registers::r0) + load16Signed(address.base, scr, dest); + else { + add32(address.base, scr); + load16Signed(scr, dest); + } + + releaseScratch(scr); + } + + void store8(RegisterID src, BaseIndex address) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + + m_assembler.movbRegMem(src, scr); + + releaseScratch(scr); + } + + void store16(RegisterID src, BaseIndex address) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + + m_assembler.movwRegMem(src, scr); + + releaseScratch(scr); + } + + void store32(RegisterID src, ImplicitAddress address) + { + RegisterID scr = claimScratch(); + store32(src, address.offset, address.base, scr); + releaseScratch(scr); + } + + void store32(RegisterID src, int offset, RegisterID base, RegisterID scr) + { + if (!offset) { + m_assembler.movlRegMem(src, base); + return; + } + + if ((offset >=0) && (offset < 64)) { + m_assembler.movlRegMem(src, offset >> 2, base); + return; + } + + m_assembler.loadConstant((offset), scr); + if (scr == SH4Registers::r0) { + m_assembler.movlRegMemr0(src, base); + return; + } + + m_assembler.addlRegReg(base, scr); + m_assembler.movlRegMem(src, scr); + } + + void store32(RegisterID src, RegisterID offset, RegisterID base) + { + ASSERT(offset == SH4Registers::r0); + m_assembler.movlRegMemr0(src, base); + } + + void store32(RegisterID src, RegisterID dst) + { + m_assembler.movlRegMem(src, dst); + } + + void store32(TrustedImm32 imm, ImplicitAddress address) + { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + m_assembler.loadConstant((imm.m_value), scr); + store32(scr, address.offset, address.base, scr1); + releaseScratch(scr); + releaseScratch(scr1); + } + + void store32(RegisterID src, BaseIndex address) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + store32(src, Address(scr, address.offset)); + + releaseScratch(scr); + } + + void store32(TrustedImm32 imm, void* address) + { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + m_assembler.loadConstant((imm.m_value), scr); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address), scr1); + m_assembler.movlRegMem(scr, scr1); + releaseScratch(scr); + releaseScratch(scr1); + } + + void store32(RegisterID src, void* address) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address), scr); + m_assembler.movlRegMem(src, scr); + releaseScratch(scr); + } + + DataLabel32 load32WithAddressOffsetPatch(Address address, RegisterID dest) + { + RegisterID scr = claimScratch(); + DataLabel32 label(this); + m_assembler.loadConstantUnReusable(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movlMemReg(scr, dest); + releaseScratch(scr); + return label; + } + + DataLabel32 store32WithAddressOffsetPatch(RegisterID src, Address address) + { + RegisterID scr = claimScratch(); + DataLabel32 label(this); + m_assembler.loadConstantUnReusable(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movlRegMem(src, scr); + releaseScratch(scr); + return label; + } + + DataLabelCompact load32WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + DataLabelCompact dataLabel(this); + ASSERT(address.offset <= MaximumCompactPtrAlignedAddressOffset); + ASSERT(address.offset >= 0); + m_assembler.movlMemRegCompact(address.offset >> 2, address.base, dest); + return dataLabel; + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result(this); + + RegisterID scr = claimScratch(); + m_assembler.movImm8(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movlMemReg(scr, dest); + releaseScratch(scr); + + return result; + } + + // Floating-point operations + + static bool supportsFloatingPoint() { return true; } + static bool supportsFloatingPointTruncate() { return true; } + static bool supportsFloatingPointSqrt() { return true; } + static bool supportsFloatingPointAbs() { return false; } + + void moveDoubleToInts(FPRegisterID src, RegisterID dest1, RegisterID dest2) + { + m_assembler.fldsfpul((FPRegisterID)(src + 1)); + m_assembler.stsfpulReg(dest1); + m_assembler.fldsfpul(src); + m_assembler.stsfpulReg(dest2); + } + + void moveIntsToDouble(RegisterID src1, RegisterID src2, FPRegisterID dest, FPRegisterID scratch) + { + UNUSED_PARAM(scratch); + m_assembler.ldsrmfpul(src1); + m_assembler.fstsfpul((FPRegisterID)(dest + 1)); + m_assembler.ldsrmfpul(src2); + m_assembler.fstsfpul(dest); + } + + void loadFloat(BaseIndex address, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + m_assembler.fmovsReadrm(scr, dest); + releaseScratch(scr); + } + + void loadDouble(BaseIndex address, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + m_assembler.fmovsReadrminc(scr, (FPRegisterID)(dest + 1)); + m_assembler.fmovsReadrm(scr, dest); + releaseScratch(scr); + } + + void loadDouble(ImplicitAddress address, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + + m_assembler.loadConstant(address.offset, scr); + if (address.base == SH4Registers::r0) { + m_assembler.fmovsReadr0r(scr, (FPRegisterID)(dest + 1)); + m_assembler.addlImm8r(4, scr); + m_assembler.fmovsReadr0r(scr, dest); + releaseScratch(scr); + return; + } + + m_assembler.addlRegReg(address.base, scr); + m_assembler.fmovsReadrminc(scr, (FPRegisterID)(dest + 1)); + m_assembler.fmovsReadrm(scr, dest); + releaseScratch(scr); + } + + void loadDouble(const void* address, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(address), scr); + m_assembler.fmovsReadrminc(scr, (FPRegisterID)(dest + 1)); + m_assembler.fmovsReadrm(scr, dest); + releaseScratch(scr); + } + + void storeFloat(FPRegisterID src, BaseIndex address) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + m_assembler.fmovsWriterm(src, scr); + + releaseScratch(scr); + } + + void storeDouble(FPRegisterID src, ImplicitAddress address) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.fmovsWriterm((FPRegisterID)(src + 1), scr); + m_assembler.addlImm8r(4, scr); + m_assembler.fmovsWriterm(src, scr); + releaseScratch(scr); + } + + void storeDouble(FPRegisterID src, BaseIndex address) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + m_assembler.fmovsWriterm((FPRegisterID)(src + 1), scr); + m_assembler.addlImm8r(4, scr); + m_assembler.fmovsWriterm(src, scr); + + releaseScratch(scr); + } + + void addDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + if (op1 == dest) + m_assembler.daddRegReg(op2, dest); + else { + m_assembler.dmovRegReg(op1, dest); + m_assembler.daddRegReg(op2, dest); + } + } + + void addDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.daddRegReg(src, dest); + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + loadDouble(address.m_ptr, fscratch); + addDouble(fscratch, dest); + } + + void addDouble(Address address, FPRegisterID dest) + { + loadDouble(address, fscratch); + addDouble(fscratch, dest); + } + + void subDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.dsubRegReg(src, dest); + } + + void subDouble(Address address, FPRegisterID dest) + { + loadDouble(address, fscratch); + subDouble(fscratch, dest); + } + + void mulDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.dmulRegReg(src, dest); + } + + void mulDouble(Address address, FPRegisterID dest) + { + loadDouble(address, fscratch); + mulDouble(fscratch, dest); + } + + void divDouble(FPRegisterID src, FPRegisterID dest) + { + m_assembler.ddivRegReg(src, dest); + } + + void convertFloatToDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.fldsfpul(src); + m_assembler.dcnvsd(dst); + } + + void convertDoubleToFloat(FPRegisterID src, FPRegisterID dst) + { + m_assembler.dcnvds(src); + m_assembler.fstsfpul(dst); + } + + void convertInt32ToDouble(RegisterID src, FPRegisterID dest) + { + m_assembler.ldsrmfpul(src); + m_assembler.floatfpulDreg(dest); + } + + void convertInt32ToDouble(AbsoluteAddress src, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(reinterpret_cast<uint32_t>(src.m_ptr), scr); + convertInt32ToDouble(scr, dest); + releaseScratch(scr); + } + + void convertInt32ToDouble(Address src, FPRegisterID dest) + { + RegisterID scr = claimScratch(); + load32(src, scr); + convertInt32ToDouble(scr, dest); + releaseScratch(scr); + } + + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) + { + RegisterID scr = claimScratch(); + RegisterID scr1 = claimScratch(); + Jump m_jump; + JumpList end; + + if (dest != SH4Registers::r0) + move(SH4Registers::r0, scr1); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + + if (address.offset) + add32(TrustedImm32(address.offset), scr); + + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 68, sizeof(uint32_t)); + move(scr, SH4Registers::r0); + m_assembler.andlImm8r(0x3, SH4Registers::r0); + m_assembler.cmpEqImmR0(0x0, SH4Registers::r0); + m_jump = Jump(m_assembler.jne(), SH4Assembler::JumpNear); + if (dest != SH4Registers::r0) + move(scr1, SH4Registers::r0); + + load32(scr, dest); + end.append(Jump(m_assembler.bra(), SH4Assembler::JumpNear)); + m_assembler.nop(); + m_jump.link(this); + m_assembler.andlImm8r(0x1, SH4Registers::r0); + m_assembler.cmpEqImmR0(0x0, SH4Registers::r0); + + if (dest != SH4Registers::r0) + move(scr1, SH4Registers::r0); + + m_jump = Jump(m_assembler.jne(), SH4Assembler::JumpNear); + load16(scr, scr1); + add32(TrustedImm32(2), scr); + load16(scr, dest); + m_assembler.shllImm8r(16, dest); + or32(scr1, dest); + end.append(Jump(m_assembler.bra(), SH4Assembler::JumpNear)); + m_assembler.nop(); + m_jump.link(this); + load8(scr, scr1); + add32(TrustedImm32(1), scr); + load16(scr, dest); + m_assembler.shllImm8r(8, dest); + or32(dest, scr1); + add32(TrustedImm32(2), scr); + load8(scr, dest); + m_assembler.shllImm8r(8, dest); + m_assembler.shllImm8r(16, dest); + or32(scr1, dest); + end.link(this); + + releaseScratch(scr); + releaseScratch(scr1); + } + + Jump branch32WithUnalignedHalfWords(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + RegisterID scr = scratchReg3; + load32WithUnalignedHalfWords(left, scr); + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) + m_assembler.testlRegReg(scr, scr); + else + compare32(right.m_value, scr, cond); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branchDoubleNonZero(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.movImm8(0, scratchReg3); + convertInt32ToDouble(scratchReg3, scratch); + return branchDouble(DoubleNotEqual, reg, scratch); + } + + Jump branchDoubleZeroOrNaN(FPRegisterID reg, FPRegisterID scratch) + { + m_assembler.movImm8(0, scratchReg3); + convertInt32ToDouble(scratchReg3, scratch); + return branchDouble(DoubleEqualOrUnordered, reg, scratch); + } + + Jump branchDouble(DoubleCondition cond, FPRegisterID left, FPRegisterID right) + { + if (cond == DoubleEqual) { + m_assembler.dcmppeq(right, left); + return branchTrue(); + } + + if (cond == DoubleNotEqual) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppeq(right, left); + releaseScratch(scr); + Jump m_jump = branchFalse(); + end.link(this); + return m_jump; + } + + if (cond == DoubleGreaterThan) { + m_assembler.dcmppgt(right, left); + return branchTrue(); + } + + if (cond == DoubleGreaterThanOrEqual) { + m_assembler.dcmppgt(left, right); + return branchFalse(); + } + + if (cond == DoubleLessThan) { + m_assembler.dcmppgt(left, right); + return branchTrue(); + } + + if (cond == DoubleLessThanOrEqual) { + m_assembler.dcmppgt(right, left); + return branchFalse(); + } + + if (cond == DoubleEqualOrUnordered) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppeq(left, right); + Jump m_jump = Jump(m_assembler.je()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + if (cond == DoubleGreaterThanOrUnordered) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppgt(right, left); + Jump m_jump = Jump(m_assembler.je()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + if (cond == DoubleGreaterThanOrEqualOrUnordered) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppgt(left, right); + Jump m_jump = Jump(m_assembler.jne()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + if (cond == DoubleLessThanOrUnordered) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppgt(left, right); + Jump m_jump = Jump(m_assembler.je()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + if (cond == DoubleLessThanOrEqualOrUnordered) { + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppgt(right, left); + Jump m_jump = Jump(m_assembler.jne()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + ASSERT(cond == DoubleNotEqualOrUnordered); + RegisterID scr = claimScratch(); + JumpList end; + m_assembler.loadConstant(0x7fbfffff, scratchReg3); + m_assembler.dcnvds(right); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 22, sizeof(uint32_t)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcnvds(left); + m_assembler.stsfpulReg(scr); + m_assembler.cmplRegReg(scratchReg3, scr, SH4Condition(Equal)); + end.append(Jump(m_assembler.je(), SH4Assembler::JumpNear)); + m_assembler.dcmppeq(right, left); + Jump m_jump = Jump(m_assembler.jne()); + end.link(this); + m_assembler.extraInstrForBranch(scr); + releaseScratch(scr); + return m_jump; + } + + Jump branchTrue() + { + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 6, sizeof(uint32_t)); + Jump m_jump = Jump(m_assembler.je()); + m_assembler.extraInstrForBranch(scratchReg3); + return m_jump; + } + + Jump branchFalse() + { + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 6, sizeof(uint32_t)); + Jump m_jump = Jump(m_assembler.jne()); + m_assembler.extraInstrForBranch(scratchReg3); + return m_jump; + } + + Jump branch32(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + RegisterID scr = claimScratch(); + move(left.index, scr); + lshift32(TrustedImm32(left.scale), scr); + add32(left.base, scr); + load32(scr, left.offset, scr); + compare32(right.m_value, scr, cond); + releaseScratch(scr); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + void sqrtDouble(FPRegisterID src, FPRegisterID dest) + { + if (dest != src) + m_assembler.dmovRegReg(src, dest); + m_assembler.dsqrt(dest); + } + + void absDouble(FPRegisterID, FPRegisterID) + { + RELEASE_ASSERT_NOT_REACHED(); + } + + Jump branchTest8(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + RegisterID addressTempRegister = claimScratch(); + load8(address, addressTempRegister); + Jump jmp = branchTest32(cond, addressTempRegister, mask); + releaseScratch(addressTempRegister); + return jmp; + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + RegisterID addressTempRegister = claimScratch(); + move(TrustedImmPtr(address.m_ptr), addressTempRegister); + load8(Address(addressTempRegister), addressTempRegister); + Jump jmp = branchTest32(cond, addressTempRegister, mask); + releaseScratch(addressTempRegister); + return jmp; + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + if (src != dest) + move(src, dest); + } + + Jump branch8(RelationalCondition cond, Address left, TrustedImm32 right) + { + RegisterID addressTempRegister = claimScratch(); + load8(left, addressTempRegister); + Jump jmp = branch32(cond, addressTempRegister, right); + releaseScratch(addressTempRegister); + return jmp; + } + + void compare8(RelationalCondition cond, Address left, TrustedImm32 right, RegisterID dest) + { + RegisterID addressTempRegister = claimScratch(); + load8(left, addressTempRegister); + compare32(cond, addressTempRegister, right, dest); + releaseScratch(addressTempRegister); + } + + Jump branchTruncateDoubleToInt32(FPRegisterID src, RegisterID dest) + { + m_assembler.ftrcdrmfpul(src); + m_assembler.stsfpulReg(dest); + m_assembler.loadConstant(0x7fffffff, scratchReg3); + m_assembler.cmplRegReg(dest, scratchReg3, SH4Condition(Equal)); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 14, sizeof(uint32_t)); + m_assembler.branch(BT_OPCODE, 2); + m_assembler.addlImm8r(1, scratchReg3); + m_assembler.cmplRegReg(dest, scratchReg3, SH4Condition(Equal)); + return branchTrue(); + } + + // Stack manipulation operations + + void pop(RegisterID dest) + { + m_assembler.popReg(dest); + } + + void push(RegisterID src) + { + m_assembler.pushReg(src); + } + + void push(Address address) + { + if (!address.offset) { + push(address.base); + return; + } + + if ((address.offset < 0) || (address.offset >= 64)) { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movlMemReg(scr, SH4Registers::sp); + m_assembler.addlImm8r(-4, SH4Registers::sp); + releaseScratch(scr); + return; + } + + m_assembler.movlMemReg(address.offset >> 2, address.base, SH4Registers::sp); + m_assembler.addlImm8r(-4, SH4Registers::sp); + } + + void push(TrustedImm32 imm) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(imm.m_value, scr); + push(scr); + releaseScratch(scr); + } + + // Register move operations + + void move(TrustedImm32 imm, RegisterID dest) + { + m_assembler.loadConstant(imm.m_value, dest); + } + + DataLabelPtr moveWithPatch(TrustedImmPtr initialValue, RegisterID dest) + { + m_assembler.ensureSpace(m_assembler.maxInstructionSize, sizeof(uint32_t)); + DataLabelPtr dataLabel(this); + m_assembler.loadConstantUnReusable(reinterpret_cast<uint32_t>(initialValue.m_value), dest); + return dataLabel; + } + + void move(RegisterID src, RegisterID dest) + { + if (src != dest) + m_assembler.movlRegReg(src, dest); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + m_assembler.loadConstant(imm.asIntptr(), dest); + } + + void extuw(RegisterID src, RegisterID dst) + { + m_assembler.extuw(src, dst); + } + + void compare32(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + m_assembler.cmplRegReg(right, left, SH4Condition(cond)); + if (cond != NotEqual) { + m_assembler.movt(dest); + return; + } + + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 4); + m_assembler.movImm8(0, dest); + m_assembler.branch(BT_OPCODE, 0); + m_assembler.movImm8(1, dest); + } + + void compare32(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + if (left != dest) { + move(right, dest); + compare32(cond, left, dest, dest); + return; + } + + RegisterID scr = claimScratch(); + move(right, scr); + compare32(cond, left, scr, dest); + releaseScratch(scr); + } + + void test8(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + ASSERT((cond == Zero) || (cond == NonZero)); + + load8(address, dest); + if (mask.m_value == -1) + compare32(0, dest, static_cast<RelationalCondition>(cond)); + else + testlImm(mask.m_value, dest); + if (cond != NonZero) { + m_assembler.movt(dest); + return; + } + + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 4); + m_assembler.movImm8(0, dest); + m_assembler.branch(BT_OPCODE, 0); + m_assembler.movImm8(1, dest); + } + + void loadPtrLinkReg(ImplicitAddress address) + { + RegisterID scr = claimScratch(); + load32(address, scr); + m_assembler.ldspr(scr); + releaseScratch(scr); + } + + Jump branch32(RelationalCondition cond, RegisterID left, RegisterID right) + { + m_assembler.cmplRegReg(right, left, SH4Condition(cond)); + /* BT label => BF off + nop LDR reg + nop braf @reg + nop nop + */ + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, RegisterID left, TrustedImm32 right) + { + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) + m_assembler.testlRegReg(left, left); + else + compare32(right.m_value, left, cond); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Address right) + { + compare32(right.offset, right.base, left, cond); + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, Address left, RegisterID right) + { + compare32(right, left.offset, left.base, cond); + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, Address left, TrustedImm32 right) + { + compare32(right.m_value, left.offset, left.base, cond); + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + RegisterID scr = claimScratch(); + + move(TrustedImm32(reinterpret_cast<uint32_t>(left.m_ptr)), scr); + m_assembler.cmplRegReg(right, scr, SH4Condition(cond)); + releaseScratch(scr); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, TrustedImm32 right) + { + RegisterID addressTempRegister = claimScratch(); + + m_assembler.loadConstant(reinterpret_cast<uint32_t>(left.m_ptr), addressTempRegister); + m_assembler.movlMemReg(addressTempRegister, addressTempRegister); + compare32(right.m_value, addressTempRegister, cond); + releaseScratch(addressTempRegister); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branch8(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + ASSERT(!(right.m_value & 0xFFFFFF00)); + RegisterID scr = claimScratch(); + + move(left.index, scr); + lshift32(TrustedImm32(left.scale), scr); + + if (left.offset) + add32(TrustedImm32(left.offset), scr); + add32(left.base, scr); + load8(scr, scr); + RegisterID scr1 = claimScratch(); + m_assembler.loadConstant(right.m_value, scr1); + releaseScratch(scr); + releaseScratch(scr1); + + return branch32(cond, scr, scr1); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, RegisterID mask) + { + ASSERT((cond == Zero) || (cond == NonZero)); + + m_assembler.testlRegReg(reg, mask); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + ASSERT((cond == Zero) || (cond == NonZero)); + + if (mask.m_value == -1) + m_assembler.testlRegReg(reg, reg); + else + testlImm(mask.m_value, reg); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchTest32(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + ASSERT((cond == Zero) || (cond == NonZero)); + + if (mask.m_value == -1) + compare32(0, address.offset, address.base, static_cast<RelationalCondition>(cond)); + else + testImm(mask.m_value, address.offset, address.base); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchTest32(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + RegisterID scr = claimScratch(); + + move(address.index, scr); + lshift32(TrustedImm32(address.scale), scr); + add32(address.base, scr); + load32(scr, address.offset, scr); + + if (mask.m_value == -1) + m_assembler.testlRegReg(scr, scr); + else + testlImm(mask.m_value, scr); + + releaseScratch(scr); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump jump() + { + return Jump(m_assembler.jmp()); + } + + void jump(RegisterID target) + { + m_assembler.jmpReg(target); + } + + void jump(Address address) + { + RegisterID scr = claimScratch(); + + if ((address.offset < 0) || (address.offset >= 64)) { + m_assembler.loadConstant(address.offset, scr); + m_assembler.addlRegReg(address.base, scr); + m_assembler.movlMemReg(scr, scr); + } else if (address.offset) + m_assembler.movlMemReg(address.offset >> 2, address.base, scr); + else + m_assembler.movlMemReg(address.base, scr); + m_assembler.jmpReg(scr); + + releaseScratch(scr); + } + + // Arithmetic control flow operations + + Jump branchAdd32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + if (cond == Overflow) { + m_assembler.addvlRegReg(src, dest); + return branchTrue(); + } + + if (cond == Signed) { + m_assembler.addlRegReg(src, dest); + // Check if dest is negative + m_assembler.cmppz(dest); + return branchFalse(); + } + + m_assembler.addlRegReg(src, dest); + compare32(0, dest, Equal); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + move(imm, scratchReg3); + return branchAdd32(cond, scratchReg3, dest); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + if (src != dest) + move(src, dest); + + if (cond == Overflow) { + move(imm, scratchReg3); + m_assembler.addvlRegReg(scratchReg3, dest); + return branchTrue(); + } + + add32(imm, dest); + + if (cond == Signed) { + m_assembler.cmppz(dest); + return branchFalse(); + } + + compare32(0, dest, Equal); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchMul32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + if (cond == Overflow) { + RegisterID scr1 = claimScratch(); + RegisterID scr = claimScratch(); + m_assembler.dmullRegReg(src, dest); + m_assembler.stsmacl(dest); + m_assembler.movImm8(-31, scr); + m_assembler.movlRegReg(dest, scr1); + m_assembler.shaRegReg(scr1, scr); + m_assembler.stsmach(scr); + m_assembler.cmplRegReg(scr, scr1, SH4Condition(Equal)); + releaseScratch(scr1); + releaseScratch(scr); + return branchFalse(); + } + + m_assembler.imullRegReg(src, dest); + m_assembler.stsmacl(dest); + if (cond == Signed) { + // Check if dest is negative + m_assembler.cmppz(dest); + return branchFalse(); + } + + compare32(0, dest, static_cast<RelationalCondition>(cond)); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchMul32(ResultCondition cond, TrustedImm32 imm, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + move(imm, scratchReg3); + if (src != dest) + move(src, dest); + + return branchMul32(cond, scratchReg3, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + if (cond == Overflow) { + m_assembler.subvlRegReg(src, dest); + return branchTrue(); + } + + if (cond == Signed) { + // Check if dest is negative + m_assembler.sublRegReg(src, dest); + compare32(0, dest, LessThan); + return branchTrue(); + } + + sub32(src, dest); + compare32(0, dest, static_cast<RelationalCondition>(cond)); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + ASSERT((cond == Overflow) || (cond == Signed) || (cond == Zero) || (cond == NonZero)); + + move(imm, scratchReg3); + return branchSub32(cond, scratchReg3, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + move(imm, scratchReg3); + if (src != dest) + move(src, dest); + return branchSub32(cond, scratchReg3, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + if (src1 != dest) + move(src1, dest); + return branchSub32(cond, src2, dest); + } + + Jump branchOr32(ResultCondition cond, RegisterID src, RegisterID dest) + { + ASSERT((cond == Signed) || (cond == Zero) || (cond == NonZero)); + + if (cond == Signed) { + or32(src, dest); + compare32(0, dest, static_cast<RelationalCondition>(LessThan)); + return branchTrue(); + } + + or32(src, dest); + compare32(0, dest, static_cast<RelationalCondition>(cond)); + + if (cond == NonZero) // NotEqual + return branchFalse(); + return branchTrue(); + } + + void branchConvertDoubleToInt32(FPRegisterID src, RegisterID dest, JumpList& failureCases, FPRegisterID fpTemp) + { + m_assembler.ftrcdrmfpul(src); + m_assembler.stsfpulReg(dest); + convertInt32ToDouble(dest, fscratch); + failureCases.append(branchDouble(DoubleNotEqualOrUnordered, fscratch, src)); + + if (dest == SH4Registers::r0) + m_assembler.cmpEqImmR0(0, dest); + else { + m_assembler.movImm8(0, scratchReg3); + m_assembler.cmplRegReg(scratchReg3, dest, SH4Condition(Equal)); + } + failureCases.append(branchTrue()); + } + + void neg32(RegisterID dst) + { + m_assembler.neg(dst, dst); + } + + void urshift32(RegisterID shiftamount, RegisterID dest) + { + if (shiftamount == SH4Registers::r0) + m_assembler.andlImm8r(0x1f, shiftamount); + else { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(0x1f, scr); + m_assembler.andlRegReg(scr, shiftamount); + releaseScratch(scr); + } + m_assembler.neg(shiftamount, shiftamount); + m_assembler.shllRegReg(dest, shiftamount); + } + + void urshift32(TrustedImm32 imm, RegisterID dest) + { + RegisterID scr = claimScratch(); + m_assembler.loadConstant(-(imm.m_value & 0x1f), scr); + m_assembler.shaRegReg(dest, scr); + releaseScratch(scr); + } + + void urshift32(RegisterID src, TrustedImm32 shiftamount, RegisterID dest) + { + if (src != dest) + move(src, dest); + + urshift32(shiftamount, dest); + } + + Call call() + { + return Call(m_assembler.call(), Call::Linkable); + } + + Call nearCall() + { + return Call(m_assembler.call(), Call::LinkableNear); + } + + Call call(RegisterID target) + { + return Call(m_assembler.call(target), Call::None); + } + + void call(Address address, RegisterID target) + { + load32(address.base, address.offset, target); + m_assembler.ensureSpace(m_assembler.maxInstructionSize + 2); + m_assembler.branch(JSR_OPCODE, target); + m_assembler.nop(); + } + + void breakpoint() + { + m_assembler.bkpt(); + m_assembler.nop(); + } + + Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + RegisterID dataTempRegister = claimScratch(); + + dataLabel = moveWithPatch(initialRightValue, dataTempRegister); + m_assembler.cmplRegReg(dataTempRegister, left, SH4Condition(cond)); + releaseScratch(dataTempRegister); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + RegisterID scr = claimScratch(); + + m_assembler.loadConstant(left.offset, scr); + m_assembler.addlRegReg(left.base, scr); + m_assembler.movlMemReg(scr, scr); + RegisterID scr1 = claimScratch(); + dataLabel = moveWithPatch(initialRightValue, scr1); + m_assembler.cmplRegReg(scr1, scr, SH4Condition(cond)); + releaseScratch(scr); + releaseScratch(scr1); + + if (cond == NotEqual) + return branchFalse(); + return branchTrue(); + } + + void ret() + { + m_assembler.ret(); + m_assembler.nop(); + } + + DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + RegisterID scr = claimScratch(); + DataLabelPtr label = moveWithPatch(initialValue, scr); + store32(scr, address); + releaseScratch(scr); + return label; + } + + DataLabelPtr storePtrWithPatch(ImplicitAddress address) { return storePtrWithPatch(TrustedImmPtr(0), address); } + + int sizeOfConstantPool() + { + return m_assembler.sizeOfConstantPool(); + } + + Call tailRecursiveCall() + { + RegisterID scr = claimScratch(); + + m_assembler.loadConstantUnReusable(0x0, scr, true); + Jump m_jump = Jump(m_assembler.jmp(scr)); + releaseScratch(scr); + + return Call::fromTailJump(m_jump); + } + + Call makeTailRecursiveCall(Jump oldJump) + { + oldJump.link(this); + return tailRecursiveCall(); + } + + void nop() + { + m_assembler.nop(); + } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + return FunctionPtr(reinterpret_cast<void(*)()>(SH4Assembler::readCallTarget(call.dataLocation()))); + } + + static void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + RELEASE_ASSERT_NOT_REACHED(); + } + + static ptrdiff_t maxJumpReplacementSize() + { + RELEASE_ASSERT_NOT_REACHED(); + return 0; + } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return false; } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + return label.labelAtOffset(0); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID, void* initialValue) + { + SH4Assembler::revertJump(instructionStart.dataLocation(), reinterpret_cast<uintptr_t>(initialValue) & 0xffff); + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr) + { + UNREACHABLE_FOR_PLATFORM(); + return CodeLocationLabel(); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel instructionStart, Address, void* initialValue) + { + UNREACHABLE_FOR_PLATFORM(); + } + +protected: + SH4Assembler::Condition SH4Condition(RelationalCondition cond) + { + return static_cast<SH4Assembler::Condition>(cond); + } + + SH4Assembler::Condition SH4Condition(ResultCondition cond) + { + return static_cast<SH4Assembler::Condition>(cond); + } +private: + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkCall(void*, Call, FunctionPtr); + static void repatchCall(CodeLocationCall, CodeLocationLabel); + static void repatchCall(CodeLocationCall, FunctionPtr); +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssemblerSH4_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerX86.h b/src/3rdparty/masm/assembler/MacroAssemblerX86.h new file mode 100644 index 0000000000..27a030edfd --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerX86.h @@ -0,0 +1,314 @@ +/* + * Copyright (C) 2008 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerX86_h +#define MacroAssemblerX86_h + +#if ENABLE(ASSEMBLER) && CPU(X86) + +#include "MacroAssemblerX86Common.h" + +namespace JSC { + +class MacroAssemblerX86 : public MacroAssemblerX86Common { +public: + static const Scale ScalePtr = TimesFour; + + using MacroAssemblerX86Common::add32; + using MacroAssemblerX86Common::and32; + using MacroAssemblerX86Common::branchAdd32; + using MacroAssemblerX86Common::branchSub32; + using MacroAssemblerX86Common::sub32; + using MacroAssemblerX86Common::or32; + using MacroAssemblerX86Common::load32; + using MacroAssemblerX86Common::store32; + using MacroAssemblerX86Common::store8; + using MacroAssemblerX86Common::branch32; + using MacroAssemblerX86Common::call; + using MacroAssemblerX86Common::jump; + using MacroAssemblerX86Common::addDouble; + using MacroAssemblerX86Common::loadDouble; + using MacroAssemblerX86Common::storeDouble; + using MacroAssemblerX86Common::convertInt32ToDouble; + using MacroAssemblerX86Common::branchTest8; + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.leal_mr(imm.m_value, src, dest); + } + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + m_assembler.addl_im(imm.m_value, address.m_ptr); + } + + void add32(AbsoluteAddress address, RegisterID dest) + { + m_assembler.addl_mr(address.m_ptr, dest); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + m_assembler.addl_im(imm.m_value, address.m_ptr); + m_assembler.adcl_im(imm.m_value >> 31, reinterpret_cast<const char*>(address.m_ptr) + sizeof(int32_t)); + } + + void and32(TrustedImm32 imm, AbsoluteAddress address) + { + m_assembler.andl_im(imm.m_value, address.m_ptr); + } + + void or32(TrustedImm32 imm, AbsoluteAddress address) + { + m_assembler.orl_im(imm.m_value, address.m_ptr); + } + + void or32(RegisterID reg, AbsoluteAddress address) + { + m_assembler.orl_rm(reg, address.m_ptr); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + m_assembler.subl_im(imm.m_value, address.m_ptr); + } + + void load32(const void* address, RegisterID dest) + { + m_assembler.movl_mr(address, dest); + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result = ConvertibleLoadLabel(this); + m_assembler.movl_mr(address.offset, address.base, dest); + return result; + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + m_assembler.addsd_mr(address.m_ptr, dest); + } + + void storeDouble(FPRegisterID src, const void* address) + { + ASSERT(isSSE2Present()); + m_assembler.movsd_rm(src, address); + } + + void convertInt32ToDouble(AbsoluteAddress src, FPRegisterID dest) + { + m_assembler.cvtsi2sd_mr(src.m_ptr, dest); + } + + void store32(TrustedImm32 imm, void* address) + { + m_assembler.movl_i32m(imm.m_value, address); + } + + void store32(RegisterID src, void* address) + { + m_assembler.movl_rm(src, address); + } + + void store8(TrustedImm32 imm, void* address) + { + ASSERT(-128 <= imm.m_value && imm.m_value < 128); + m_assembler.movb_i8m(imm.m_value, address); + } + + // Possibly clobbers src. + void moveDoubleToInts(FPRegisterID src, RegisterID dest1, RegisterID dest2) + { + movePackedToInt32(src, dest1); + rshiftPacked(TrustedImm32(32), src); + movePackedToInt32(src, dest2); + } + + void moveIntsToDouble(RegisterID src1, RegisterID src2, FPRegisterID dest, FPRegisterID scratch) + { + moveInt32ToPacked(src1, dest); + moveInt32ToPacked(src2, scratch); + lshiftPacked(TrustedImm32(32), scratch); + orPacked(scratch, dest); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, AbsoluteAddress dest) + { + m_assembler.addl_im(imm.m_value, dest.m_ptr); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, AbsoluteAddress dest) + { + m_assembler.subl_im(imm.m_value, dest.m_ptr); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + m_assembler.cmpl_rm(right, left.m_ptr); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, AbsoluteAddress left, TrustedImm32 right) + { + m_assembler.cmpl_im(right.m_value, left.m_ptr); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Call call() + { + return Call(m_assembler.call(), Call::Linkable); + } + + // Address is a memory location containing the address to jump to + void jump(AbsoluteAddress address) + { + m_assembler.jmp_m(address.m_ptr); + } + + Call tailRecursiveCall() + { + return Call::fromTailJump(jump()); + } + + Call makeTailRecursiveCall(Jump oldJump) + { + return Call::fromTailJump(oldJump); + } + + + DataLabelPtr moveWithPatch(TrustedImmPtr initialValue, RegisterID dest) + { + padBeforePatch(); + m_assembler.movl_i32r(initialValue.asIntptr(), dest); + return DataLabelPtr(this); + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + ASSERT(mask.m_value >= -128 && mask.m_value <= 255); + if (mask.m_value == -1) + m_assembler.cmpb_im(0, address.m_ptr); + else + m_assembler.testb_im(mask.m_value, address.m_ptr); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + padBeforePatch(); + m_assembler.cmpl_ir_force32(initialRightValue.asIntptr(), left); + dataLabel = DataLabelPtr(this); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + padBeforePatch(); + m_assembler.cmpl_im_force32(initialRightValue.asIntptr(), left.offset, left.base); + dataLabel = DataLabelPtr(this); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + padBeforePatch(); + m_assembler.movl_i32m(initialValue.asIntptr(), address.offset, address.base); + return DataLabelPtr(this); + } + + static bool supportsFloatingPoint() { return isSSE2Present(); } + // See comment on MacroAssemblerARMv7::supportsFloatingPointTruncate() + static bool supportsFloatingPointTruncate() { return isSSE2Present(); } + static bool supportsFloatingPointSqrt() { return isSSE2Present(); } + static bool supportsFloatingPointAbs() { return isSSE2Present(); } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + intptr_t offset = reinterpret_cast<int32_t*>(call.dataLocation())[-1]; + return FunctionPtr(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(call.dataLocation()) + offset)); + } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return true; } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + const int opcodeBytes = 1; + const int modRMBytes = 1; + const int immediateBytes = 4; + const int totalBytes = opcodeBytes + modRMBytes + immediateBytes; + ASSERT(totalBytes >= maxJumpReplacementSize()); + return label.labelAtOffset(-totalBytes); + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr label) + { + const int opcodeBytes = 1; + const int modRMBytes = 1; + const int offsetBytes = 0; + const int immediateBytes = 4; + const int totalBytes = opcodeBytes + modRMBytes + offsetBytes + immediateBytes; + ASSERT(totalBytes >= maxJumpReplacementSize()); + return label.labelAtOffset(-totalBytes); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID reg, void* initialValue) + { + X86Assembler::revertJumpTo_cmpl_ir_force32(instructionStart.executableAddress(), reinterpret_cast<intptr_t>(initialValue), reg); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel instructionStart, Address address, void* initialValue) + { + ASSERT(!address.offset); + X86Assembler::revertJumpTo_cmpl_im_force32(instructionStart.executableAddress(), reinterpret_cast<intptr_t>(initialValue), 0, address.base); + } + +private: + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkCall(void* code, Call call, FunctionPtr function) + { + X86Assembler::linkCall(code, call.m_label, function.value()); + } + + static void repatchCall(CodeLocationCall call, CodeLocationLabel destination) + { + X86Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } + + static void repatchCall(CodeLocationCall call, FunctionPtr destination) + { + X86Assembler::relinkCall(call.dataLocation(), destination.executableAddress()); + } +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssemblerX86_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerX86Common.h b/src/3rdparty/masm/assembler/MacroAssemblerX86Common.h new file mode 100644 index 0000000000..53cb80c210 --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerX86Common.h @@ -0,0 +1,1541 @@ +/* + * Copyright (C) 2008 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerX86Common_h +#define MacroAssemblerX86Common_h + +#if ENABLE(ASSEMBLER) + +#include "X86Assembler.h" +#include "AbstractMacroAssembler.h" + +namespace JSC { + +class MacroAssemblerX86Common : public AbstractMacroAssembler<X86Assembler> { +protected: +#if CPU(X86_64) + static const X86Registers::RegisterID scratchRegister = X86Registers::r11; +#endif + + static const int DoubleConditionBitInvert = 0x10; + static const int DoubleConditionBitSpecial = 0x20; + static const int DoubleConditionBits = DoubleConditionBitInvert | DoubleConditionBitSpecial; + +public: + typedef X86Assembler::FPRegisterID FPRegisterID; + typedef X86Assembler::XMMRegisterID XMMRegisterID; + + static bool isCompactPtrAlignedAddressOffset(ptrdiff_t value) + { + return value >= -128 && value <= 127; + } + + enum RelationalCondition { + Equal = X86Assembler::ConditionE, + NotEqual = X86Assembler::ConditionNE, + Above = X86Assembler::ConditionA, + AboveOrEqual = X86Assembler::ConditionAE, + Below = X86Assembler::ConditionB, + BelowOrEqual = X86Assembler::ConditionBE, + GreaterThan = X86Assembler::ConditionG, + GreaterThanOrEqual = X86Assembler::ConditionGE, + LessThan = X86Assembler::ConditionL, + LessThanOrEqual = X86Assembler::ConditionLE + }; + + enum ResultCondition { + Overflow = X86Assembler::ConditionO, + Signed = X86Assembler::ConditionS, + Zero = X86Assembler::ConditionE, + NonZero = X86Assembler::ConditionNE + }; + + enum DoubleCondition { + // These conditions will only evaluate to true if the comparison is ordered - i.e. neither operand is NaN. + DoubleEqual = X86Assembler::ConditionE | DoubleConditionBitSpecial, + DoubleNotEqual = X86Assembler::ConditionNE, + DoubleGreaterThan = X86Assembler::ConditionA, + DoubleGreaterThanOrEqual = X86Assembler::ConditionAE, + DoubleLessThan = X86Assembler::ConditionA | DoubleConditionBitInvert, + DoubleLessThanOrEqual = X86Assembler::ConditionAE | DoubleConditionBitInvert, + // If either operand is NaN, these conditions always evaluate to true. + DoubleEqualOrUnordered = X86Assembler::ConditionE, + DoubleNotEqualOrUnordered = X86Assembler::ConditionNE | DoubleConditionBitSpecial, + DoubleGreaterThanOrUnordered = X86Assembler::ConditionB | DoubleConditionBitInvert, + DoubleGreaterThanOrEqualOrUnordered = X86Assembler::ConditionBE | DoubleConditionBitInvert, + DoubleLessThanOrUnordered = X86Assembler::ConditionB, + DoubleLessThanOrEqualOrUnordered = X86Assembler::ConditionBE, + }; + COMPILE_ASSERT( + !((X86Assembler::ConditionE | X86Assembler::ConditionNE | X86Assembler::ConditionA | X86Assembler::ConditionAE | X86Assembler::ConditionB | X86Assembler::ConditionBE) & DoubleConditionBits), + DoubleConditionBits_should_not_interfere_with_X86Assembler_Condition_codes); + + static const RegisterID stackPointerRegister = X86Registers::esp; + +#if ENABLE(JIT_CONSTANT_BLINDING) + static bool shouldBlindForSpecificArch(uint32_t value) { return value >= 0x00ffffff; } +#if CPU(X86_64) + static bool shouldBlindForSpecificArch(uint64_t value) { return value >= 0x00ffffff; } +#if OS(DARWIN) // On 64-bit systems other than DARWIN uint64_t and uintptr_t are the same type so overload is prohibited. + static bool shouldBlindForSpecificArch(uintptr_t value) { return value >= 0x00ffffff; } +#endif +#endif +#endif + + // Integer arithmetic operations: + // + // Operations are typically two operand - operation(source, srcDst) + // For many operations the source may be an TrustedImm32, the srcDst operand + // may often be a memory location (explictly described using an Address + // object). + + void add32(RegisterID src, RegisterID dest) + { + m_assembler.addl_rr(src, dest); + } + + void add32(TrustedImm32 imm, Address address) + { + m_assembler.addl_im(imm.m_value, address.offset, address.base); + } + + void add32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.addl_ir(imm.m_value, dest); + } + + void add32(Address src, RegisterID dest) + { + m_assembler.addl_mr(src.offset, src.base, dest); + } + + void add32(RegisterID src, Address dest) + { + m_assembler.addl_rm(src, dest.offset, dest.base); + } + + void add32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.leal_mr(imm.m_value, src, dest); + } + + void and32(RegisterID src, RegisterID dest) + { + m_assembler.andl_rr(src, dest); + } + + void and32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.andl_ir(imm.m_value, dest); + } + + void and32(RegisterID src, Address dest) + { + m_assembler.andl_rm(src, dest.offset, dest.base); + } + + void and32(Address src, RegisterID dest) + { + m_assembler.andl_mr(src.offset, src.base, dest); + } + + void and32(TrustedImm32 imm, Address address) + { + m_assembler.andl_im(imm.m_value, address.offset, address.base); + } + + void and32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op1 == op2) + zeroExtend32ToPtr(op1, dest); + else if (op1 == dest) + and32(op2, dest); + else { + move(op2, dest); + and32(op1, dest); + } + } + + void and32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(src, dest); + and32(imm, dest); + } + + void lshift32(RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (shift_amount == X86Registers::ecx) + m_assembler.shll_CLr(dest); + else { + // On x86 we can only shift by ecx; if asked to shift by another register we'll + // need rejig the shift amount into ecx first, and restore the registers afterwards. + // If we dest is ecx, then shift the swapped register! + swap(shift_amount, X86Registers::ecx); + m_assembler.shll_CLr(dest == X86Registers::ecx ? shift_amount : dest); + swap(shift_amount, X86Registers::ecx); + } + } + + void lshift32(RegisterID src, RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (src != dest) + move(src, dest); + lshift32(shift_amount, dest); + } + + void lshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.shll_i8r(imm.m_value, dest); + } + + void lshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + if (src != dest) + move(src, dest); + lshift32(imm, dest); + } + + void mul32(RegisterID src, RegisterID dest) + { + m_assembler.imull_rr(src, dest); + } + + void mul32(Address src, RegisterID dest) + { + m_assembler.imull_mr(src.offset, src.base, dest); + } + + void mul32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.imull_i32r(src, imm.m_value, dest); + } + + void neg32(RegisterID srcDest) + { + m_assembler.negl_r(srcDest); + } + + void neg32(Address srcDest) + { + m_assembler.negl_m(srcDest.offset, srcDest.base); + } + + void or32(RegisterID src, RegisterID dest) + { + m_assembler.orl_rr(src, dest); + } + + void or32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.orl_ir(imm.m_value, dest); + } + + void or32(RegisterID src, Address dest) + { + m_assembler.orl_rm(src, dest.offset, dest.base); + } + + void or32(Address src, RegisterID dest) + { + m_assembler.orl_mr(src.offset, src.base, dest); + } + + void or32(TrustedImm32 imm, Address address) + { + m_assembler.orl_im(imm.m_value, address.offset, address.base); + } + + void or32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op1 == op2) + zeroExtend32ToPtr(op1, dest); + else if (op1 == dest) + or32(op2, dest); + else { + move(op2, dest); + or32(op1, dest); + } + } + + void or32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(src, dest); + or32(imm, dest); + } + + void rshift32(RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (shift_amount == X86Registers::ecx) + m_assembler.sarl_CLr(dest); + else { + // On x86 we can only shift by ecx; if asked to shift by another register we'll + // need rejig the shift amount into ecx first, and restore the registers afterwards. + // If we dest is ecx, then shift the swapped register! + swap(shift_amount, X86Registers::ecx); + m_assembler.sarl_CLr(dest == X86Registers::ecx ? shift_amount : dest); + swap(shift_amount, X86Registers::ecx); + } + } + + void rshift32(RegisterID src, RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (src != dest) + move(src, dest); + rshift32(shift_amount, dest); + } + + void rshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.sarl_i8r(imm.m_value, dest); + } + + void rshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + if (src != dest) + move(src, dest); + rshift32(imm, dest); + } + + void urshift32(RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (shift_amount == X86Registers::ecx) + m_assembler.shrl_CLr(dest); + else { + // On x86 we can only shift by ecx; if asked to shift by another register we'll + // need rejig the shift amount into ecx first, and restore the registers afterwards. + // If we dest is ecx, then shift the swapped register! + swap(shift_amount, X86Registers::ecx); + m_assembler.shrl_CLr(dest == X86Registers::ecx ? shift_amount : dest); + swap(shift_amount, X86Registers::ecx); + } + } + + void urshift32(RegisterID src, RegisterID shift_amount, RegisterID dest) + { + ASSERT(shift_amount != dest); + + if (src != dest) + move(src, dest); + urshift32(shift_amount, dest); + } + + void urshift32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.shrl_i8r(imm.m_value, dest); + } + + void urshift32(RegisterID src, TrustedImm32 imm, RegisterID dest) + { + if (src != dest) + move(src, dest); + urshift32(imm, dest); + } + + void sub32(RegisterID src, RegisterID dest) + { + m_assembler.subl_rr(src, dest); + } + + void sub32(TrustedImm32 imm, RegisterID dest) + { + m_assembler.subl_ir(imm.m_value, dest); + } + + void sub32(TrustedImm32 imm, Address address) + { + m_assembler.subl_im(imm.m_value, address.offset, address.base); + } + + void sub32(Address src, RegisterID dest) + { + m_assembler.subl_mr(src.offset, src.base, dest); + } + + void sub32(RegisterID src, Address dest) + { + m_assembler.subl_rm(src, dest.offset, dest.base); + } + + void xor32(RegisterID src, RegisterID dest) + { + m_assembler.xorl_rr(src, dest); + } + + void xor32(TrustedImm32 imm, Address dest) + { + if (imm.m_value == -1) + m_assembler.notl_m(dest.offset, dest.base); + else + m_assembler.xorl_im(imm.m_value, dest.offset, dest.base); + } + + void xor32(TrustedImm32 imm, RegisterID dest) + { + if (imm.m_value == -1) + m_assembler.notl_r(dest); + else + m_assembler.xorl_ir(imm.m_value, dest); + } + + void xor32(RegisterID src, Address dest) + { + m_assembler.xorl_rm(src, dest.offset, dest.base); + } + + void xor32(Address src, RegisterID dest) + { + m_assembler.xorl_mr(src.offset, src.base, dest); + } + + void xor32(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op1 == op2) + move(TrustedImm32(0), dest); + else if (op1 == dest) + xor32(op2, dest); + else { + move(op2, dest); + xor32(op1, dest); + } + } + + void xor32(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(src, dest); + xor32(imm, dest); + } + + void sqrtDouble(FPRegisterID src, FPRegisterID dst) + { + m_assembler.sqrtsd_rr(src, dst); + } + + void absDouble(FPRegisterID src, FPRegisterID dst) + { + ASSERT(src != dst); + static const double negativeZeroConstant = -0.0; + loadDouble(&negativeZeroConstant, dst); + m_assembler.andnpd_rr(src, dst); + } + + void negateDouble(FPRegisterID src, FPRegisterID dst) + { + ASSERT(src != dst); + static const double negativeZeroConstant = -0.0; + loadDouble(&negativeZeroConstant, dst); + m_assembler.xorpd_rr(src, dst); + } + + + // Memory access operations: + // + // Loads are of the form load(address, destination) and stores of the form + // store(source, address). The source for a store may be an TrustedImm32. Address + // operand objects to loads and store will be implicitly constructed if a + // register is passed. + + void load32(ImplicitAddress address, RegisterID dest) + { + m_assembler.movl_mr(address.offset, address.base, dest); + } + + void load32(BaseIndex address, RegisterID dest) + { + m_assembler.movl_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load32WithUnalignedHalfWords(BaseIndex address, RegisterID dest) + { + load32(address, dest); + } + + void load16Unaligned(BaseIndex address, RegisterID dest) + { + load16(address, dest); + } + + DataLabel32 load32WithAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + m_assembler.movl_mr_disp32(address.offset, address.base, dest); + return DataLabel32(this); + } + + DataLabelCompact load32WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + m_assembler.movl_mr_disp8(address.offset, address.base, dest); + return DataLabelCompact(this); + } + + static void repatchCompact(CodeLocationDataLabelCompact dataLabelCompact, int32_t value) + { + ASSERT(isCompactPtrAlignedAddressOffset(value)); + AssemblerType_T::repatchCompact(dataLabelCompact.dataLocation(), value); + } + + DataLabelCompact loadCompactWithAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + m_assembler.movl_mr_disp8(address.offset, address.base, dest); + return DataLabelCompact(this); + } + + void load8(BaseIndex address, RegisterID dest) + { + m_assembler.movzbl_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load8(ImplicitAddress address, RegisterID dest) + { + m_assembler.movzbl_mr(address.offset, address.base, dest); + } + + void load8Signed(BaseIndex address, RegisterID dest) + { + m_assembler.movsbl_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load8Signed(ImplicitAddress address, RegisterID dest) + { + m_assembler.movsbl_mr(address.offset, address.base, dest); + } + + void load16(BaseIndex address, RegisterID dest) + { + m_assembler.movzwl_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load16(Address address, RegisterID dest) + { + m_assembler.movzwl_mr(address.offset, address.base, dest); + } + + void load16Signed(BaseIndex address, RegisterID dest) + { + m_assembler.movswl_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load16Signed(Address address, RegisterID dest) + { + m_assembler.movswl_mr(address.offset, address.base, dest); + } + + DataLabel32 store32WithAddressOffsetPatch(RegisterID src, Address address) + { + padBeforePatch(); + m_assembler.movl_rm_disp32(src, address.offset, address.base); + return DataLabel32(this); + } + + void store32(RegisterID src, ImplicitAddress address) + { + m_assembler.movl_rm(src, address.offset, address.base); + } + + void store32(RegisterID src, BaseIndex address) + { + m_assembler.movl_rm(src, address.offset, address.base, address.index, address.scale); + } + + void store32(TrustedImm32 imm, ImplicitAddress address) + { + m_assembler.movl_i32m(imm.m_value, address.offset, address.base); + } + + void store32(TrustedImm32 imm, BaseIndex address) + { + m_assembler.movl_i32m(imm.m_value, address.offset, address.base, address.index, address.scale); + } + + void store8(TrustedImm32 imm, Address address) + { + ASSERT(-128 <= imm.m_value && imm.m_value < 128); + m_assembler.movb_i8m(imm.m_value, address.offset, address.base); + } + + void store8(TrustedImm32 imm, BaseIndex address) + { + ASSERT(-128 <= imm.m_value && imm.m_value < 128); + m_assembler.movb_i8m(imm.m_value, address.offset, address.base, address.index, address.scale); + } + + void store8(RegisterID src, BaseIndex address) + { +#if CPU(X86) + // On 32-bit x86 we can only store from the first 4 registers; + // esp..edi are mapped to the 'h' registers! + if (src >= 4) { + // Pick a temporary register. + RegisterID temp; + if (address.base != X86Registers::eax && address.index != X86Registers::eax) + temp = X86Registers::eax; + else if (address.base != X86Registers::ebx && address.index != X86Registers::ebx) + temp = X86Registers::ebx; + else { + ASSERT(address.base != X86Registers::ecx && address.index != X86Registers::ecx); + temp = X86Registers::ecx; + } + + // Swap to the temporary register to perform the store. + swap(src, temp); + m_assembler.movb_rm(temp, address.offset, address.base, address.index, address.scale); + swap(src, temp); + return; + } +#endif + m_assembler.movb_rm(src, address.offset, address.base, address.index, address.scale); + } + + void store16(RegisterID src, BaseIndex address) + { +#if CPU(X86) + // On 32-bit x86 we can only store from the first 4 registers; + // esp..edi are mapped to the 'h' registers! + if (src >= 4) { + // Pick a temporary register. + RegisterID temp; + if (address.base != X86Registers::eax && address.index != X86Registers::eax) + temp = X86Registers::eax; + else if (address.base != X86Registers::ebx && address.index != X86Registers::ebx) + temp = X86Registers::ebx; + else { + ASSERT(address.base != X86Registers::ecx && address.index != X86Registers::ecx); + temp = X86Registers::ecx; + } + + // Swap to the temporary register to perform the store. + swap(src, temp); + m_assembler.movw_rm(temp, address.offset, address.base, address.index, address.scale); + swap(src, temp); + return; + } +#endif + m_assembler.movw_rm(src, address.offset, address.base, address.index, address.scale); + } + + + // Floating-point operation: + // + // Presently only supports SSE, not x87 floating point. + + void moveDouble(FPRegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + if (src != dest) + m_assembler.movsd_rr(src, dest); + } + + void loadDouble(const void* address, FPRegisterID dest) + { +#if CPU(X86) + ASSERT(isSSE2Present()); + m_assembler.movsd_mr(address, dest); +#else + move(TrustedImmPtr(address), scratchRegister); + loadDouble(scratchRegister, dest); +#endif + } + + void loadDouble(ImplicitAddress address, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.movsd_mr(address.offset, address.base, dest); + } + + void loadDouble(BaseIndex address, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.movsd_mr(address.offset, address.base, address.index, address.scale, dest); + } + void loadFloat(BaseIndex address, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.movss_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void storeDouble(FPRegisterID src, ImplicitAddress address) + { + ASSERT(isSSE2Present()); + m_assembler.movsd_rm(src, address.offset, address.base); + } + + void storeDouble(FPRegisterID src, BaseIndex address) + { + ASSERT(isSSE2Present()); + m_assembler.movsd_rm(src, address.offset, address.base, address.index, address.scale); + } + + void storeFloat(FPRegisterID src, BaseIndex address) + { + ASSERT(isSSE2Present()); + m_assembler.movss_rm(src, address.offset, address.base, address.index, address.scale); + } + + void convertDoubleToFloat(FPRegisterID src, FPRegisterID dst) + { + ASSERT(isSSE2Present()); + m_assembler.cvtsd2ss_rr(src, dst); + } + + void convertFloatToDouble(FPRegisterID src, FPRegisterID dst) + { + ASSERT(isSSE2Present()); + m_assembler.cvtss2sd_rr(src, dst); + } + + void addDouble(FPRegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.addsd_rr(src, dest); + } + + void addDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + if (op1 == dest) + addDouble(op2, dest); + else { + moveDouble(op2, dest); + addDouble(op1, dest); + } + } + + void addDouble(Address src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.addsd_mr(src.offset, src.base, dest); + } + + void divDouble(FPRegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.divsd_rr(src, dest); + } + + void divDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + // B := A / B is invalid. + ASSERT(op1 == dest || op2 != dest); + + moveDouble(op1, dest); + divDouble(op2, dest); + } + + void divDouble(Address src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.divsd_mr(src.offset, src.base, dest); + } + + void subDouble(FPRegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.subsd_rr(src, dest); + } + + void subDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + // B := A - B is invalid. + ASSERT(op1 == dest || op2 != dest); + + moveDouble(op1, dest); + subDouble(op2, dest); + } + + void subDouble(Address src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.subsd_mr(src.offset, src.base, dest); + } + + void mulDouble(FPRegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.mulsd_rr(src, dest); + } + + void mulDouble(FPRegisterID op1, FPRegisterID op2, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + if (op1 == dest) + mulDouble(op2, dest); + else { + moveDouble(op2, dest); + mulDouble(op1, dest); + } + } + + void mulDouble(Address src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.mulsd_mr(src.offset, src.base, dest); + } + + void convertInt32ToDouble(RegisterID src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.cvtsi2sd_rr(src, dest); + } + + void convertInt32ToDouble(Address src, FPRegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.cvtsi2sd_mr(src.offset, src.base, dest); + } + + Jump branchDouble(DoubleCondition cond, FPRegisterID left, FPRegisterID right) + { + ASSERT(isSSE2Present()); + + if (cond & DoubleConditionBitInvert) + m_assembler.ucomisd_rr(left, right); + else + m_assembler.ucomisd_rr(right, left); + + if (cond == DoubleEqual) { + if (left == right) + return Jump(m_assembler.jnp()); + Jump isUnordered(m_assembler.jp()); + Jump result = Jump(m_assembler.je()); + isUnordered.link(this); + return result; + } else if (cond == DoubleNotEqualOrUnordered) { + if (left == right) + return Jump(m_assembler.jp()); + Jump isUnordered(m_assembler.jp()); + Jump isEqual(m_assembler.je()); + isUnordered.link(this); + Jump result = jump(); + isEqual.link(this); + return result; + } + + ASSERT(!(cond & DoubleConditionBitSpecial)); + return Jump(m_assembler.jCC(static_cast<X86Assembler::Condition>(cond & ~DoubleConditionBits))); + } + + // Truncates 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, INT_MIN). + enum BranchTruncateType { BranchIfTruncateFailed, BranchIfTruncateSuccessful }; + Jump branchTruncateDoubleToInt32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + ASSERT(isSSE2Present()); + m_assembler.cvttsd2si_rr(src, dest); + return branch32(branchType ? NotEqual : Equal, dest, TrustedImm32(0x80000000)); + } + + Jump branchTruncateDoubleToUint32(FPRegisterID src, RegisterID dest, BranchTruncateType branchType = BranchIfTruncateFailed) + { + ASSERT(isSSE2Present()); + m_assembler.cvttsd2si_rr(src, dest); + return branch32(branchType ? GreaterThanOrEqual : LessThan, dest, TrustedImm32(0)); + } + + void truncateDoubleToInt32(FPRegisterID src, RegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.cvttsd2si_rr(src, dest); + } + +#if CPU(X86_64) + void truncateDoubleToUint32(FPRegisterID src, RegisterID dest) + { + ASSERT(isSSE2Present()); + m_assembler.cvttsd2siq_rr(src, dest); + } +#endif + + // Convert 'src' to an integer, and places the resulting 'dest'. + // If the result is not representable as a 32 bit value, branch. + // May also branch for some values that are representable in 32 bits + // (specifically, in this case, 0). + void branchConvertDoubleToInt32(FPRegisterID src, RegisterID dest, JumpList& failureCases, FPRegisterID fpTemp) + { + ASSERT(isSSE2Present()); + m_assembler.cvttsd2si_rr(src, dest); + + // If the result is zero, it might have been -0.0, and the double comparison won't catch this! + failureCases.append(branchTest32(Zero, dest)); + + // Convert the integer result back to float & compare to the original value - if not equal or unordered (NaN) then jump. + convertInt32ToDouble(dest, fpTemp); + m_assembler.ucomisd_rr(fpTemp, src); + failureCases.append(m_assembler.jp()); + failureCases.append(m_assembler.jne()); + } + + Jump branchDoubleNonZero(FPRegisterID reg, FPRegisterID scratch) + { + ASSERT(isSSE2Present()); + m_assembler.xorpd_rr(scratch, scratch); + return branchDouble(DoubleNotEqual, reg, scratch); + } + + Jump branchDoubleZeroOrNaN(FPRegisterID reg, FPRegisterID scratch) + { + ASSERT(isSSE2Present()); + m_assembler.xorpd_rr(scratch, scratch); + return branchDouble(DoubleEqualOrUnordered, reg, scratch); + } + + void lshiftPacked(TrustedImm32 imm, XMMRegisterID reg) + { + ASSERT(isSSE2Present()); + m_assembler.psllq_i8r(imm.m_value, reg); + } + + void rshiftPacked(TrustedImm32 imm, XMMRegisterID reg) + { + ASSERT(isSSE2Present()); + m_assembler.psrlq_i8r(imm.m_value, reg); + } + + void orPacked(XMMRegisterID src, XMMRegisterID dst) + { + ASSERT(isSSE2Present()); + m_assembler.por_rr(src, dst); + } + + void moveInt32ToPacked(RegisterID src, XMMRegisterID dst) + { + ASSERT(isSSE2Present()); + m_assembler.movd_rr(src, dst); + } + + void movePackedToInt32(XMMRegisterID src, RegisterID dst) + { + ASSERT(isSSE2Present()); + m_assembler.movd_rr(src, dst); + } + + // Stack manipulation operations: + // + // The ABI is assumed to provide a stack abstraction to memory, + // containing machine word sized units of data. Push and pop + // operations add and remove a single register sized unit of data + // to or from the stack. Peek and poke operations read or write + // values on the stack, without moving the current stack position. + + void pop(RegisterID dest) + { + m_assembler.pop_r(dest); + } + + void push(RegisterID src) + { + m_assembler.push_r(src); + } + + void push(Address address) + { + m_assembler.push_m(address.offset, address.base); + } + + void push(TrustedImm32 imm) + { + m_assembler.push_i32(imm.m_value); + } + + + // Register move operations: + // + // Move values in registers. + + void move(TrustedImm32 imm, RegisterID dest) + { + // Note: on 64-bit the TrustedImm32 value is zero extended into the register, it + // may be useful to have a separate version that sign extends the value? + if (!imm.m_value) + m_assembler.xorl_rr(dest, dest); + else + m_assembler.movl_i32r(imm.m_value, dest); + } + +#if CPU(X86_64) + void move(RegisterID src, RegisterID dest) + { + // Note: on 64-bit this is is a full register move; perhaps it would be + // useful to have separate move32 & movePtr, with move32 zero extending? + if (src != dest) + m_assembler.movq_rr(src, dest); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + m_assembler.movq_i64r(imm.asIntptr(), dest); + } + + void move(TrustedImm64 imm, RegisterID dest) + { + m_assembler.movq_i64r(imm.m_value, dest); + } + + void swap(RegisterID reg1, RegisterID reg2) + { + if (reg1 != reg2) + m_assembler.xchgq_rr(reg1, reg2); + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + m_assembler.movsxd_rr(src, dest); + } + + void zeroExtend32ToPtr(RegisterID src, RegisterID dest) + { + m_assembler.movl_rr(src, dest); + } +#else + void move(RegisterID src, RegisterID dest) + { + if (src != dest) + m_assembler.movl_rr(src, dest); + } + + void move(TrustedImmPtr imm, RegisterID dest) + { + m_assembler.movl_i32r(imm.asIntptr(), dest); + } + + void swap(RegisterID reg1, RegisterID reg2) + { + if (reg1 != reg2) + m_assembler.xchgl_rr(reg1, reg2); + } + + void signExtend32ToPtr(RegisterID src, RegisterID dest) + { + move(src, dest); + } + + void zeroExtend32ToPtr(RegisterID src, RegisterID dest) + { + move(src, dest); + } +#endif + + + // Forwards / external control flow operations: + // + // This set of jump and conditional branch operations return a Jump + // object which may linked at a later point, allow forwards jump, + // or jumps that will require external linkage (after the code has been + // relocated). + // + // For branches, signed <, >, <= and >= are denoted as l, g, le, and ge + // respecitvely, for unsigned comparisons the names b, a, be, and ae are + // used (representing the names 'below' and 'above'). + // + // Operands to the comparision are provided in the expected order, e.g. + // jle32(reg1, TrustedImm32(5)) will branch if the value held in reg1, when + // treated as a signed 32bit value, is less than or equal to 5. + // + // jz and jnz test whether the first operand is equal to zero, and take + // an optional second operand of a mask under which to perform the test. + +public: + Jump branch8(RelationalCondition cond, Address left, TrustedImm32 right) + { + m_assembler.cmpb_im(right.m_value, left.offset, left.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, RegisterID left, RegisterID right) + { + m_assembler.cmpl_rr(right, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, RegisterID left, TrustedImm32 right) + { + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) + m_assembler.testl_rr(left, left); + else + m_assembler.cmpl_ir(right.m_value, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, RegisterID left, Address right) + { + m_assembler.cmpl_mr(right.offset, right.base, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, Address left, RegisterID right) + { + m_assembler.cmpl_rm(right, left.offset, left.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, Address left, TrustedImm32 right) + { + m_assembler.cmpl_im(right.m_value, left.offset, left.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + m_assembler.cmpl_im(right.m_value, left.offset, left.base, left.index, left.scale); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch32WithUnalignedHalfWords(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + return branch32(cond, left, right); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, RegisterID mask) + { + m_assembler.testl_rr(reg, mask); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest32(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + // if we are only interested in the low seven bits, this can be tested with a testb + if (mask.m_value == -1) + m_assembler.testl_rr(reg, reg); + else + m_assembler.testl_i32r(mask.m_value, reg); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest32(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + if (mask.m_value == -1) + m_assembler.cmpl_im(0, address.offset, address.base); + else + m_assembler.testl_i32m(mask.m_value, address.offset, address.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest32(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + if (mask.m_value == -1) + m_assembler.cmpl_im(0, address.offset, address.base, address.index, address.scale); + else + m_assembler.testl_i32m(mask.m_value, address.offset, address.base, address.index, address.scale); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest8(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + // Byte in TrustedImm32 is not well defined, so be a little permisive here, but don't accept nonsense values. + ASSERT(mask.m_value >= -128 && mask.m_value <= 255); + if (mask.m_value == -1) + m_assembler.cmpb_im(0, address.offset, address.base); + else + m_assembler.testb_im(mask.m_value, address.offset, address.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest8(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + // Byte in TrustedImm32 is not well defined, so be a little permisive here, but don't accept nonsense values. + ASSERT(mask.m_value >= -128 && mask.m_value <= 255); + if (mask.m_value == -1) + m_assembler.cmpb_im(0, address.offset, address.base, address.index, address.scale); + else + m_assembler.testb_im(mask.m_value, address.offset, address.base, address.index, address.scale); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch8(RelationalCondition cond, BaseIndex left, TrustedImm32 right) + { + ASSERT(!(right.m_value & 0xFFFFFF00)); + + m_assembler.cmpb_im(right.m_value, left.offset, left.base, left.index, left.scale); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump jump() + { + return Jump(m_assembler.jmp()); + } + + void jump(RegisterID target) + { + m_assembler.jmp_r(target); + } + + // Address is a memory location containing the address to jump to + void jump(Address address) + { + m_assembler.jmp_m(address.offset, address.base); + } + + + // Arithmetic control flow operations: + // + // This set of conditional branch operations branch based + // on the result of an arithmetic operation. The operation + // is performed as normal, storing the result. + // + // * jz operations branch if the result is zero. + // * jo operations branch if the (signed) arithmetic + // operation caused an overflow to occur. + + Jump branchAdd32(ResultCondition cond, RegisterID src, RegisterID dest) + { + add32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + add32(imm, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 src, Address dest) + { + add32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, Address dest) + { + add32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd32(ResultCondition cond, Address src, RegisterID dest) + { + add32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + if (src1 == dest) + return branchAdd32(cond, src2, dest); + move(src2, dest); + return branchAdd32(cond, src1, dest); + } + + Jump branchAdd32(ResultCondition cond, RegisterID src, TrustedImm32 imm, RegisterID dest) + { + move(src, dest); + return branchAdd32(cond, imm, dest); + } + + Jump branchMul32(ResultCondition cond, RegisterID src, RegisterID dest) + { + mul32(src, dest); + if (cond != Overflow) + m_assembler.testl_rr(dest, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchMul32(ResultCondition cond, Address src, RegisterID dest) + { + mul32(src, dest); + if (cond != Overflow) + m_assembler.testl_rr(dest, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchMul32(ResultCondition cond, TrustedImm32 imm, RegisterID src, RegisterID dest) + { + mul32(imm, src, dest); + if (cond != Overflow) + m_assembler.testl_rr(dest, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchMul32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + if (src1 == dest) + return branchMul32(cond, src2, dest); + move(src2, dest); + return branchMul32(cond, src1, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, RegisterID dest) + { + sub32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + sub32(imm, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, TrustedImm32 imm, Address dest) + { + sub32(imm, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, RegisterID src, Address dest) + { + sub32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, Address src, RegisterID dest) + { + sub32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub32(ResultCondition cond, RegisterID src1, RegisterID src2, RegisterID dest) + { + // B := A - B is invalid. + ASSERT(src1 == dest || src2 != dest); + + move(src1, dest); + return branchSub32(cond, src2, dest); + } + + Jump branchSub32(ResultCondition cond, RegisterID src1, TrustedImm32 src2, RegisterID dest) + { + move(src1, dest); + return branchSub32(cond, src2, dest); + } + + Jump branchNeg32(ResultCondition cond, RegisterID srcDest) + { + neg32(srcDest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchOr32(ResultCondition cond, RegisterID src, RegisterID dest) + { + or32(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + + // Miscellaneous operations: + + void breakpoint() + { + m_assembler.int3(); + } + + Call nearCall() + { + return Call(m_assembler.call(), Call::LinkableNear); + } + + Call call(RegisterID target) + { + return Call(m_assembler.call(target), Call::None); + } + + void call(Address address) + { + m_assembler.call_m(address.offset, address.base); + } + + void ret() + { + m_assembler.ret(); + } + + void compare8(RelationalCondition cond, Address left, TrustedImm32 right, RegisterID dest) + { + m_assembler.cmpb_im(right.m_value, left.offset, left.base); + set32(x86Condition(cond), dest); + } + + void compare32(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + m_assembler.cmpl_rr(right, left); + set32(x86Condition(cond), dest); + } + + void compare32(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) + m_assembler.testl_rr(left, left); + else + m_assembler.cmpl_ir(right.m_value, left); + set32(x86Condition(cond), dest); + } + + // FIXME: + // The mask should be optional... perhaps the argument order should be + // dest-src, operations always have a dest? ... possibly not true, considering + // asm ops like test, or pseudo ops like pop(). + + void test8(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + if (mask.m_value == -1) + m_assembler.cmpb_im(0, address.offset, address.base); + else + m_assembler.testb_im(mask.m_value, address.offset, address.base); + set32(x86Condition(cond), dest); + } + + void test32(ResultCondition cond, Address address, TrustedImm32 mask, RegisterID dest) + { + if (mask.m_value == -1) + m_assembler.cmpl_im(0, address.offset, address.base); + else + m_assembler.testl_i32m(mask.m_value, address.offset, address.base); + set32(x86Condition(cond), dest); + } + + // Invert a relational condition, e.g. == becomes !=, < becomes >=, etc. + static RelationalCondition invert(RelationalCondition cond) + { + return static_cast<RelationalCondition>(cond ^ 1); + } + + void nop() + { + m_assembler.nop(); + } + + static void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + X86Assembler::replaceWithJump(instructionStart.executableAddress(), destination.executableAddress()); + } + + static ptrdiff_t maxJumpReplacementSize() + { + return X86Assembler::maxJumpReplacementSize(); + } + +protected: + X86Assembler::Condition x86Condition(RelationalCondition cond) + { + return static_cast<X86Assembler::Condition>(cond); + } + + X86Assembler::Condition x86Condition(ResultCondition cond) + { + return static_cast<X86Assembler::Condition>(cond); + } + + void set32(X86Assembler::Condition cond, RegisterID dest) + { +#if CPU(X86) + // On 32-bit x86 we can only set the first 4 registers; + // esp..edi are mapped to the 'h' registers! + if (dest >= 4) { + m_assembler.xchgl_rr(dest, X86Registers::eax); + m_assembler.setCC_r(cond, X86Registers::eax); + m_assembler.movzbl_rr(X86Registers::eax, X86Registers::eax); + m_assembler.xchgl_rr(dest, X86Registers::eax); + return; + } +#endif + m_assembler.setCC_r(cond, dest); + m_assembler.movzbl_rr(dest, dest); + } + +private: + // Only MacroAssemblerX86 should be using the following method; SSE2 is always available on + // x86_64, and clients & subclasses of MacroAssembler should be using 'supportsFloatingPoint()'. + friend class MacroAssemblerX86; + +#if CPU(X86) +#if OS(MAC_OS_X) + + // All X86 Macs are guaranteed to support at least SSE2, + static bool isSSE2Present() + { + return true; + } + +#else // OS(MAC_OS_X) + + enum SSE2CheckState { + NotCheckedSSE2, + HasSSE2, + NoSSE2 + }; + + static bool isSSE2Present() + { + if (s_sse2CheckState == NotCheckedSSE2) { + // Default the flags value to zero; if the compiler is + // not MSVC or GCC we will read this as SSE2 not present. + int flags = 0; +#if COMPILER(MSVC) + _asm { + mov eax, 1 // cpuid function 1 gives us the standard feature set + cpuid; + mov flags, edx; + } +#elif COMPILER(GCC) + asm ( + "movl $0x1, %%eax;" + "pushl %%ebx;" + "cpuid;" + "popl %%ebx;" + "movl %%edx, %0;" + : "=g" (flags) + : + : "%eax", "%ecx", "%edx" + ); +#endif + static const int SSE2FeatureBit = 1 << 26; + s_sse2CheckState = (flags & SSE2FeatureBit) ? HasSSE2 : NoSSE2; + } + // Only check once. + ASSERT(s_sse2CheckState != NotCheckedSSE2); + + return s_sse2CheckState == HasSSE2; + } + + static SSE2CheckState s_sse2CheckState; + +#endif // OS(MAC_OS_X) +#elif !defined(NDEBUG) // CPU(X86) + + // On x86-64 we should never be checking for SSE2 in a non-debug build, + // but non debug add this method to keep the asserts above happy. + static bool isSSE2Present() + { + return true; + } + +#endif +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssemblerX86Common_h diff --git a/src/3rdparty/masm/assembler/MacroAssemblerX86_64.h b/src/3rdparty/masm/assembler/MacroAssemblerX86_64.h new file mode 100644 index 0000000000..c711e6f8da --- /dev/null +++ b/src/3rdparty/masm/assembler/MacroAssemblerX86_64.h @@ -0,0 +1,643 @@ +/* + * Copyright (C) 2008, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef MacroAssemblerX86_64_h +#define MacroAssemblerX86_64_h + +#if ENABLE(ASSEMBLER) && CPU(X86_64) + +#include "MacroAssemblerX86Common.h" + +#define REPTACH_OFFSET_CALL_R11 3 + +namespace JSC { + +class MacroAssemblerX86_64 : public MacroAssemblerX86Common { +public: + static const Scale ScalePtr = TimesEight; + + using MacroAssemblerX86Common::add32; + using MacroAssemblerX86Common::and32; + using MacroAssemblerX86Common::branchAdd32; + using MacroAssemblerX86Common::or32; + using MacroAssemblerX86Common::sub32; + using MacroAssemblerX86Common::load32; + using MacroAssemblerX86Common::store32; + using MacroAssemblerX86Common::store8; + using MacroAssemblerX86Common::call; + using MacroAssemblerX86Common::jump; + using MacroAssemblerX86Common::addDouble; + using MacroAssemblerX86Common::loadDouble; + using MacroAssemblerX86Common::convertInt32ToDouble; + + void add32(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + add32(imm, Address(scratchRegister)); + } + + void and32(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + and32(imm, Address(scratchRegister)); + } + + void add32(AbsoluteAddress address, RegisterID dest) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + add32(Address(scratchRegister), dest); + } + + void or32(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + or32(imm, Address(scratchRegister)); + } + + void or32(RegisterID reg, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + or32(reg, Address(scratchRegister)); + } + + void sub32(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + sub32(imm, Address(scratchRegister)); + } + + void load32(const void* address, RegisterID dest) + { + if (dest == X86Registers::eax) + m_assembler.movl_mEAX(address); + else { + move(TrustedImmPtr(address), dest); + load32(dest, dest); + } + } + + void addDouble(AbsoluteAddress address, FPRegisterID dest) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + m_assembler.addsd_mr(0, scratchRegister, dest); + } + + void convertInt32ToDouble(TrustedImm32 imm, FPRegisterID dest) + { + move(imm, scratchRegister); + m_assembler.cvtsi2sd_rr(scratchRegister, dest); + } + + void store32(TrustedImm32 imm, void* address) + { + move(TrustedImmPtr(address), scratchRegister); + store32(imm, scratchRegister); + } + + void store8(TrustedImm32 imm, void* address) + { + move(TrustedImmPtr(address), scratchRegister); + store8(imm, Address(scratchRegister)); + } + + Call call() + { + DataLabelPtr label = moveWithPatch(TrustedImmPtr(0), scratchRegister); + Call result = Call(m_assembler.call(scratchRegister), Call::Linkable); + ASSERT_UNUSED(label, differenceBetween(label, result) == REPTACH_OFFSET_CALL_R11); + return result; + } + + // Address is a memory location containing the address to jump to + void jump(AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + jump(Address(scratchRegister)); + } + + Call tailRecursiveCall() + { + DataLabelPtr label = moveWithPatch(TrustedImmPtr(0), scratchRegister); + Jump newJump = Jump(m_assembler.jmp_r(scratchRegister)); + ASSERT_UNUSED(label, differenceBetween(label, newJump) == REPTACH_OFFSET_CALL_R11); + return Call::fromTailJump(newJump); + } + + Call makeTailRecursiveCall(Jump oldJump) + { + oldJump.link(this); + DataLabelPtr label = moveWithPatch(TrustedImmPtr(0), scratchRegister); + Jump newJump = Jump(m_assembler.jmp_r(scratchRegister)); + ASSERT_UNUSED(label, differenceBetween(label, newJump) == REPTACH_OFFSET_CALL_R11); + return Call::fromTailJump(newJump); + } + + Jump branchAdd32(ResultCondition cond, TrustedImm32 src, AbsoluteAddress dest) + { + move(TrustedImmPtr(dest.m_ptr), scratchRegister); + add32(src, Address(scratchRegister)); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + void add64(RegisterID src, RegisterID dest) + { + m_assembler.addq_rr(src, dest); + } + + void add64(Address src, RegisterID dest) + { + m_assembler.addq_mr(src.offset, src.base, dest); + } + + void add64(AbsoluteAddress src, RegisterID dest) + { + move(TrustedImmPtr(src.m_ptr), scratchRegister); + add64(Address(scratchRegister), dest); + } + + void add64(TrustedImm32 imm, RegisterID srcDest) + { + m_assembler.addq_ir(imm.m_value, srcDest); + } + + void add64(TrustedImm64 imm, RegisterID dest) + { + move(imm, scratchRegister); + add64(scratchRegister, dest); + } + + void add64(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + m_assembler.leaq_mr(imm.m_value, src, dest); + } + + void add64(TrustedImm32 imm, Address address) + { + m_assembler.addq_im(imm.m_value, address.offset, address.base); + } + + void add64(TrustedImm32 imm, AbsoluteAddress address) + { + move(TrustedImmPtr(address.m_ptr), scratchRegister); + add64(imm, Address(scratchRegister)); + } + + void and64(RegisterID src, RegisterID dest) + { + m_assembler.andq_rr(src, dest); + } + + void and64(TrustedImm32 imm, RegisterID srcDest) + { + m_assembler.andq_ir(imm.m_value, srcDest); + } + + void neg64(RegisterID dest) + { + m_assembler.negq_r(dest); + } + + void or64(RegisterID src, RegisterID dest) + { + m_assembler.orq_rr(src, dest); + } + + void or64(TrustedImm64 imm, RegisterID dest) + { + move(imm, scratchRegister); + or64(scratchRegister, dest); + } + + void or64(TrustedImm32 imm, RegisterID dest) + { + m_assembler.orq_ir(imm.m_value, dest); + } + + void or64(RegisterID op1, RegisterID op2, RegisterID dest) + { + if (op1 == op2) + move(op1, dest); + else if (op1 == dest) + or64(op2, dest); + else { + move(op2, dest); + or64(op1, dest); + } + } + + void or64(TrustedImm32 imm, RegisterID src, RegisterID dest) + { + move(src, dest); + or64(imm, dest); + } + + void rotateRight64(TrustedImm32 imm, RegisterID srcDst) + { + m_assembler.rorq_i8r(imm.m_value, srcDst); + } + + void sub64(RegisterID src, RegisterID dest) + { + m_assembler.subq_rr(src, dest); + } + + void sub64(TrustedImm32 imm, RegisterID dest) + { + m_assembler.subq_ir(imm.m_value, dest); + } + + void sub64(TrustedImm64 imm, RegisterID dest) + { + move(imm, scratchRegister); + sub64(scratchRegister, dest); + } + + void xor64(RegisterID src, RegisterID dest) + { + m_assembler.xorq_rr(src, dest); + } + + void xor64(RegisterID src, Address dest) + { + m_assembler.xorq_rm(src, dest.offset, dest.base); + } + + void xor64(TrustedImm32 imm, RegisterID srcDest) + { + m_assembler.xorq_ir(imm.m_value, srcDest); + } + + void load64(ImplicitAddress address, RegisterID dest) + { + m_assembler.movq_mr(address.offset, address.base, dest); + } + + void load64(BaseIndex address, RegisterID dest) + { + m_assembler.movq_mr(address.offset, address.base, address.index, address.scale, dest); + } + + void load64(const void* address, RegisterID dest) + { + if (dest == X86Registers::eax) + m_assembler.movq_mEAX(address); + else { + move(TrustedImmPtr(address), dest); + load64(dest, dest); + } + } + + DataLabel32 load64WithAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + m_assembler.movq_mr_disp32(address.offset, address.base, dest); + return DataLabel32(this); + } + + DataLabelCompact load64WithCompactAddressOffsetPatch(Address address, RegisterID dest) + { + padBeforePatch(); + m_assembler.movq_mr_disp8(address.offset, address.base, dest); + return DataLabelCompact(this); + } + + void store64(RegisterID src, ImplicitAddress address) + { + m_assembler.movq_rm(src, address.offset, address.base); + } + + void store64(RegisterID src, BaseIndex address) + { + m_assembler.movq_rm(src, address.offset, address.base, address.index, address.scale); + } + + void store64(RegisterID src, void* address) + { + if (src == X86Registers::eax) + m_assembler.movq_EAXm(address); + else { + move(TrustedImmPtr(address), scratchRegister); + store64(src, scratchRegister); + } + } + + void store64(TrustedImm64 imm, ImplicitAddress address) + { + move(imm, scratchRegister); + store64(scratchRegister, address); + } + + void store64(TrustedImm64 imm, BaseIndex address) + { + move(imm, scratchRegister); + m_assembler.movq_rm(scratchRegister, address.offset, address.base, address.index, address.scale); + } + + DataLabel32 store64WithAddressOffsetPatch(RegisterID src, Address address) + { + padBeforePatch(); + m_assembler.movq_rm_disp32(src, address.offset, address.base); + return DataLabel32(this); + } + + void move64ToDouble(RegisterID src, FPRegisterID dest) + { + m_assembler.movq_rr(src, dest); + } + + void moveDoubleTo64(FPRegisterID src, RegisterID dest) + { + m_assembler.movq_rr(src, dest); + } + + void compare64(RelationalCondition cond, RegisterID left, TrustedImm32 right, RegisterID dest) + { + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) + m_assembler.testq_rr(left, left); + else + m_assembler.cmpq_ir(right.m_value, left); + m_assembler.setCC_r(x86Condition(cond), dest); + m_assembler.movzbl_rr(dest, dest); + } + + void compare64(RelationalCondition cond, RegisterID left, RegisterID right, RegisterID dest) + { + m_assembler.cmpq_rr(right, left); + m_assembler.setCC_r(x86Condition(cond), dest); + m_assembler.movzbl_rr(dest, dest); + } + + Jump branch64(RelationalCondition cond, RegisterID left, RegisterID right) + { + m_assembler.cmpq_rr(right, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch64(RelationalCondition cond, RegisterID left, TrustedImm64 right) + { + if (((cond == Equal) || (cond == NotEqual)) && !right.m_value) { + m_assembler.testq_rr(left, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + move(right, scratchRegister); + return branch64(cond, left, scratchRegister); + } + + Jump branch64(RelationalCondition cond, RegisterID left, Address right) + { + m_assembler.cmpq_mr(right.offset, right.base, left); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch64(RelationalCondition cond, AbsoluteAddress left, RegisterID right) + { + move(TrustedImmPtr(left.m_ptr), scratchRegister); + return branch64(cond, Address(scratchRegister), right); + } + + Jump branch64(RelationalCondition cond, Address left, RegisterID right) + { + m_assembler.cmpq_rm(right, left.offset, left.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branch64(RelationalCondition cond, Address left, TrustedImm64 right) + { + move(right, scratchRegister); + return branch64(cond, left, scratchRegister); + } + + Jump branchTest64(ResultCondition cond, RegisterID reg, RegisterID mask) + { + m_assembler.testq_rr(reg, mask); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest64(ResultCondition cond, RegisterID reg, TrustedImm32 mask = TrustedImm32(-1)) + { + // if we are only interested in the low seven bits, this can be tested with a testb + if (mask.m_value == -1) + m_assembler.testq_rr(reg, reg); + else if ((mask.m_value & ~0x7f) == 0) + m_assembler.testb_i8r(mask.m_value, reg); + else + m_assembler.testq_i32r(mask.m_value, reg); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + void test64(ResultCondition cond, RegisterID reg, TrustedImm32 mask, RegisterID dest) + { + if (mask.m_value == -1) + m_assembler.testq_rr(reg, reg); + else if ((mask.m_value & ~0x7f) == 0) + m_assembler.testb_i8r(mask.m_value, reg); + else + m_assembler.testq_i32r(mask.m_value, reg); + set32(x86Condition(cond), dest); + } + + void test64(ResultCondition cond, RegisterID reg, RegisterID mask, RegisterID dest) + { + m_assembler.testq_rr(reg, mask); + set32(x86Condition(cond), dest); + } + + Jump branchTest64(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + load64(address.m_ptr, scratchRegister); + return branchTest64(cond, scratchRegister, mask); + } + + Jump branchTest64(ResultCondition cond, Address address, TrustedImm32 mask = TrustedImm32(-1)) + { + if (mask.m_value == -1) + m_assembler.cmpq_im(0, address.offset, address.base); + else + m_assembler.testq_i32m(mask.m_value, address.offset, address.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest64(ResultCondition cond, Address address, RegisterID reg) + { + m_assembler.testq_rm(reg, address.offset, address.base); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchTest64(ResultCondition cond, BaseIndex address, TrustedImm32 mask = TrustedImm32(-1)) + { + if (mask.m_value == -1) + m_assembler.cmpq_im(0, address.offset, address.base, address.index, address.scale); + else + m_assembler.testq_i32m(mask.m_value, address.offset, address.base, address.index, address.scale); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + + Jump branchAdd64(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + add64(imm, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchAdd64(ResultCondition cond, RegisterID src, RegisterID dest) + { + add64(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub64(ResultCondition cond, TrustedImm32 imm, RegisterID dest) + { + sub64(imm, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub64(ResultCondition cond, RegisterID src, RegisterID dest) + { + sub64(src, dest); + return Jump(m_assembler.jCC(x86Condition(cond))); + } + + Jump branchSub64(ResultCondition cond, RegisterID src1, TrustedImm32 src2, RegisterID dest) + { + move(src1, dest); + return branchSub64(cond, src2, dest); + } + + ConvertibleLoadLabel convertibleLoadPtr(Address address, RegisterID dest) + { + ConvertibleLoadLabel result = ConvertibleLoadLabel(this); + m_assembler.movq_mr(address.offset, address.base, dest); + return result; + } + + DataLabelPtr moveWithPatch(TrustedImmPtr initialValue, RegisterID dest) + { + padBeforePatch(); + m_assembler.movq_i64r(initialValue.asIntptr(), dest); + return DataLabelPtr(this); + } + + Jump branchPtrWithPatch(RelationalCondition cond, RegisterID left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + dataLabel = moveWithPatch(initialRightValue, scratchRegister); + return branch64(cond, left, scratchRegister); + } + + Jump branchPtrWithPatch(RelationalCondition cond, Address left, DataLabelPtr& dataLabel, TrustedImmPtr initialRightValue = TrustedImmPtr(0)) + { + dataLabel = moveWithPatch(initialRightValue, scratchRegister); + return branch64(cond, left, scratchRegister); + } + + DataLabelPtr storePtrWithPatch(TrustedImmPtr initialValue, ImplicitAddress address) + { + DataLabelPtr label = moveWithPatch(initialValue, scratchRegister); + store64(scratchRegister, address); + return label; + } + + using MacroAssemblerX86Common::branchTest8; + Jump branchTest8(ResultCondition cond, ExtendedAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + TrustedImmPtr addr(reinterpret_cast<void*>(address.offset)); + MacroAssemblerX86Common::move(addr, scratchRegister); + return MacroAssemblerX86Common::branchTest8(cond, BaseIndex(scratchRegister, address.base, TimesOne), mask); + } + + Jump branchTest8(ResultCondition cond, AbsoluteAddress address, TrustedImm32 mask = TrustedImm32(-1)) + { + MacroAssemblerX86Common::move(TrustedImmPtr(address.m_ptr), scratchRegister); + return MacroAssemblerX86Common::branchTest8(cond, Address(scratchRegister), mask); + } + + static bool supportsFloatingPoint() { return true; } + // See comment on MacroAssemblerARMv7::supportsFloatingPointTruncate() + static bool supportsFloatingPointTruncate() { return true; } + static bool supportsFloatingPointSqrt() { return true; } + static bool supportsFloatingPointAbs() { return true; } + + static FunctionPtr readCallTarget(CodeLocationCall call) + { + return FunctionPtr(X86Assembler::readPointer(call.dataLabelPtrAtOffset(-REPTACH_OFFSET_CALL_R11).dataLocation())); + } + + static RegisterID scratchRegisterForBlinding() { return scratchRegister; } + + static bool canJumpReplacePatchableBranchPtrWithPatch() { return true; } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + const int rexBytes = 1; + const int opcodeBytes = 1; + const int immediateBytes = 8; + const int totalBytes = rexBytes + opcodeBytes + immediateBytes; + ASSERT(totalBytes >= maxJumpReplacementSize()); + return label.labelAtOffset(-totalBytes); + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr label) + { + return startOfBranchPtrWithPatchOnRegister(label); + } + + static void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel instructionStart, Address, void* initialValue) + { + X86Assembler::revertJumpTo_movq_i64r(instructionStart.executableAddress(), reinterpret_cast<intptr_t>(initialValue), scratchRegister); + } + + static void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, RegisterID, void* initialValue) + { + X86Assembler::revertJumpTo_movq_i64r(instructionStart.executableAddress(), reinterpret_cast<intptr_t>(initialValue), scratchRegister); + } + +private: + friend class LinkBuffer; + friend class RepatchBuffer; + + static void linkCall(void* code, Call call, FunctionPtr function) + { + if (!call.isFlagSet(Call::Near)) + X86Assembler::linkPointer(code, call.m_label.labelAtOffset(-REPTACH_OFFSET_CALL_R11), function.value()); + else + X86Assembler::linkCall(code, call.m_label, function.value()); + } + + static void repatchCall(CodeLocationCall call, CodeLocationLabel destination) + { + X86Assembler::repatchPointer(call.dataLabelPtrAtOffset(-REPTACH_OFFSET_CALL_R11).dataLocation(), destination.executableAddress()); + } + + static void repatchCall(CodeLocationCall call, FunctionPtr destination) + { + X86Assembler::repatchPointer(call.dataLabelPtrAtOffset(-REPTACH_OFFSET_CALL_R11).dataLocation(), destination.executableAddress()); + } + +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // MacroAssemblerX86_64_h diff --git a/src/3rdparty/masm/assembler/RepatchBuffer.h b/src/3rdparty/masm/assembler/RepatchBuffer.h new file mode 100644 index 0000000000..dbb56f9ad5 --- /dev/null +++ b/src/3rdparty/masm/assembler/RepatchBuffer.h @@ -0,0 +1,181 @@ +/* + * Copyright (C) 2009 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef RepatchBuffer_h +#define RepatchBuffer_h + +#if ENABLE(JIT) + +#include "CodeBlock.h" +#include <MacroAssembler.h> +#include <wtf/Noncopyable.h> + +namespace JSC { + +// RepatchBuffer: +// +// This class is used to modify code after code generation has been completed, +// and after the code has potentially already been executed. This mechanism is +// used to apply optimizations to the code. +// +class RepatchBuffer { + typedef MacroAssemblerCodePtr CodePtr; + +public: + RepatchBuffer(CodeBlock* codeBlock) + { + JITCode& code = codeBlock->getJITCode(); + m_start = code.start(); + m_size = code.size(); + + ExecutableAllocator::makeWritable(m_start, m_size); + } + + ~RepatchBuffer() + { + ExecutableAllocator::makeExecutable(m_start, m_size); + } + + void relink(CodeLocationJump jump, CodeLocationLabel destination) + { + MacroAssembler::repatchJump(jump, destination); + } + + void relink(CodeLocationCall call, CodeLocationLabel destination) + { + MacroAssembler::repatchCall(call, destination); + } + + void relink(CodeLocationCall call, FunctionPtr destination) + { + MacroAssembler::repatchCall(call, destination); + } + + void relink(CodeLocationNearCall nearCall, CodePtr destination) + { + MacroAssembler::repatchNearCall(nearCall, CodeLocationLabel(destination)); + } + + void relink(CodeLocationNearCall nearCall, CodeLocationLabel destination) + { + MacroAssembler::repatchNearCall(nearCall, destination); + } + + void repatch(CodeLocationDataLabel32 dataLabel32, int32_t value) + { + MacroAssembler::repatchInt32(dataLabel32, value); + } + + void repatch(CodeLocationDataLabelCompact dataLabelCompact, int32_t value) + { + MacroAssembler::repatchCompact(dataLabelCompact, value); + } + + void repatch(CodeLocationDataLabelPtr dataLabelPtr, void* value) + { + MacroAssembler::repatchPointer(dataLabelPtr, value); + } + + void relinkCallerToTrampoline(ReturnAddressPtr returnAddress, CodeLocationLabel label) + { + relink(CodeLocationCall(CodePtr(returnAddress)), label); + } + + void relinkCallerToTrampoline(ReturnAddressPtr returnAddress, CodePtr newCalleeFunction) + { + relinkCallerToTrampoline(returnAddress, CodeLocationLabel(newCalleeFunction)); + } + + void relinkCallerToFunction(ReturnAddressPtr returnAddress, FunctionPtr function) + { + relink(CodeLocationCall(CodePtr(returnAddress)), function); + } + + void relinkNearCallerToTrampoline(ReturnAddressPtr returnAddress, CodeLocationLabel label) + { + relink(CodeLocationNearCall(CodePtr(returnAddress)), label); + } + + void relinkNearCallerToTrampoline(ReturnAddressPtr returnAddress, CodePtr newCalleeFunction) + { + relinkNearCallerToTrampoline(returnAddress, CodeLocationLabel(newCalleeFunction)); + } + + void replaceWithLoad(CodeLocationConvertibleLoad label) + { + MacroAssembler::replaceWithLoad(label); + } + + void replaceWithAddressComputation(CodeLocationConvertibleLoad label) + { + MacroAssembler::replaceWithAddressComputation(label); + } + + void setLoadInstructionIsActive(CodeLocationConvertibleLoad label, bool isActive) + { + if (isActive) + replaceWithLoad(label); + else + replaceWithAddressComputation(label); + } + + static CodeLocationLabel startOfBranchPtrWithPatchOnRegister(CodeLocationDataLabelPtr label) + { + return MacroAssembler::startOfBranchPtrWithPatchOnRegister(label); + } + + static CodeLocationLabel startOfPatchableBranchPtrWithPatchOnAddress(CodeLocationDataLabelPtr label) + { + return MacroAssembler::startOfPatchableBranchPtrWithPatchOnAddress(label); + } + + void replaceWithJump(CodeLocationLabel instructionStart, CodeLocationLabel destination) + { + MacroAssembler::replaceWithJump(instructionStart, destination); + } + + // This is a *bit* of a silly API, since we currently always also repatch the + // immediate after calling this. But I'm fine with that, since this just feels + // less yucky. + void revertJumpReplacementToBranchPtrWithPatch(CodeLocationLabel instructionStart, MacroAssembler::RegisterID reg, void* value) + { + MacroAssembler::revertJumpReplacementToBranchPtrWithPatch(instructionStart, reg, value); + } + + void revertJumpReplacementToPatchableBranchPtrWithPatch(CodeLocationLabel instructionStart, MacroAssembler::Address address, void* value) + { + MacroAssembler::revertJumpReplacementToPatchableBranchPtrWithPatch(instructionStart, address, value); + } + +private: + void* m_start; + size_t m_size; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) + +#endif // RepatchBuffer_h diff --git a/src/3rdparty/masm/assembler/SH4Assembler.h b/src/3rdparty/masm/assembler/SH4Assembler.h new file mode 100644 index 0000000000..b7a166ea99 --- /dev/null +++ b/src/3rdparty/masm/assembler/SH4Assembler.h @@ -0,0 +1,2152 @@ +/* + * Copyright (C) 2009-2011 STMicroelectronics. All rights reserved. + * Copyright (C) 2008 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef SH4Assembler_h +#define SH4Assembler_h + +#if ENABLE(ASSEMBLER) && CPU(SH4) + +#include "AssemblerBuffer.h" +#include "AssemblerBufferWithConstantPool.h" +#include "JITCompilationEffort.h" +#include <stdarg.h> +#include <stdint.h> +#include <stdio.h> +#include <wtf/Assertions.h> +#include <wtf/DataLog.h> +#include <wtf/Vector.h> + +#ifndef NDEBUG +#define SH4_ASSEMBLER_TRACING +#endif + +namespace JSC { +typedef uint16_t SH4Word; + +enum { + INVALID_OPCODE = 0xffff, + ADD_OPCODE = 0x300c, + ADDIMM_OPCODE = 0x7000, + ADDC_OPCODE = 0x300e, + ADDV_OPCODE = 0x300f, + AND_OPCODE = 0x2009, + ANDIMM_OPCODE = 0xc900, + DIV0_OPCODE = 0x2007, + DIV1_OPCODE = 0x3004, + BF_OPCODE = 0x8b00, + BFS_OPCODE = 0x8f00, + BRA_OPCODE = 0xa000, + BRAF_OPCODE = 0x0023, + NOP_OPCODE = 0x0009, + BSR_OPCODE = 0xb000, + RTS_OPCODE = 0x000b, + BT_OPCODE = 0x8900, + BTS_OPCODE = 0x8d00, + BSRF_OPCODE = 0x0003, + BRK_OPCODE = 0x003b, + FTRC_OPCODE = 0xf03d, + CMPEQ_OPCODE = 0x3000, + CMPEQIMM_OPCODE = 0x8800, + CMPGE_OPCODE = 0x3003, + CMPGT_OPCODE = 0x3007, + CMPHI_OPCODE = 0x3006, + CMPHS_OPCODE = 0x3002, + CMPPL_OPCODE = 0x4015, + CMPPZ_OPCODE = 0x4011, + CMPSTR_OPCODE = 0x200c, + DT_OPCODE = 0x4010, + FCMPEQ_OPCODE = 0xf004, + FCMPGT_OPCODE = 0xf005, + FMOV_OPCODE = 0xf00c, + FADD_OPCODE = 0xf000, + FMUL_OPCODE = 0xf002, + FSUB_OPCODE = 0xf001, + FDIV_OPCODE = 0xf003, + FNEG_OPCODE = 0xf04d, + JMP_OPCODE = 0x402b, + JSR_OPCODE = 0x400b, + LDSPR_OPCODE = 0x402a, + LDSLPR_OPCODE = 0x4026, + MOV_OPCODE = 0x6003, + MOVIMM_OPCODE = 0xe000, + MOVB_WRITE_RN_OPCODE = 0x2000, + MOVB_WRITE_RNDEC_OPCODE = 0x2004, + MOVB_WRITE_R0RN_OPCODE = 0x0004, + MOVB_WRITE_OFFGBR_OPCODE = 0xc000, + MOVB_WRITE_OFFRN_OPCODE = 0x8000, + MOVB_READ_RM_OPCODE = 0x6000, + MOVB_READ_RMINC_OPCODE = 0x6004, + MOVB_READ_R0RM_OPCODE = 0x000c, + MOVB_READ_OFFGBR_OPCODE = 0xc400, + MOVB_READ_OFFRM_OPCODE = 0x8400, + MOVL_WRITE_RN_OPCODE = 0x2002, + MOVL_WRITE_RNDEC_OPCODE = 0x2006, + MOVL_WRITE_R0RN_OPCODE = 0x0006, + MOVL_WRITE_OFFGBR_OPCODE = 0xc200, + MOVL_WRITE_OFFRN_OPCODE = 0x1000, + MOVL_READ_RM_OPCODE = 0x6002, + MOVL_READ_RMINC_OPCODE = 0x6006, + MOVL_READ_R0RM_OPCODE = 0x000e, + MOVL_READ_OFFGBR_OPCODE = 0xc600, + MOVL_READ_OFFPC_OPCODE = 0xd000, + MOVL_READ_OFFRM_OPCODE = 0x5000, + MOVW_WRITE_RN_OPCODE = 0x2001, + MOVW_READ_RM_OPCODE = 0x6001, + MOVW_READ_R0RM_OPCODE = 0x000d, + MOVW_READ_OFFRM_OPCODE = 0x8500, + MOVW_READ_OFFPC_OPCODE = 0x9000, + MOVA_READ_OFFPC_OPCODE = 0xc700, + MOVT_OPCODE = 0x0029, + MULL_OPCODE = 0x0007, + DMULL_L_OPCODE = 0x3005, + STSMACL_OPCODE = 0x001a, + STSMACH_OPCODE = 0x000a, + DMULSL_OPCODE = 0x300d, + NEG_OPCODE = 0x600b, + NEGC_OPCODE = 0x600a, + NOT_OPCODE = 0x6007, + OR_OPCODE = 0x200b, + ORIMM_OPCODE = 0xcb00, + ORBIMM_OPCODE = 0xcf00, + SETS_OPCODE = 0x0058, + SETT_OPCODE = 0x0018, + SHAD_OPCODE = 0x400c, + SHAL_OPCODE = 0x4020, + SHAR_OPCODE = 0x4021, + SHLD_OPCODE = 0x400d, + SHLL_OPCODE = 0x4000, + SHLL2_OPCODE = 0x4008, + SHLL8_OPCODE = 0x4018, + SHLL16_OPCODE = 0x4028, + SHLR_OPCODE = 0x4001, + SHLR2_OPCODE = 0x4009, + SHLR8_OPCODE = 0x4019, + SHLR16_OPCODE = 0x4029, + STSPR_OPCODE = 0x002a, + STSLPR_OPCODE = 0x4022, + FLOAT_OPCODE = 0xf02d, + SUB_OPCODE = 0x3008, + SUBC_OPCODE = 0x300a, + SUBV_OPCODE = 0x300b, + TST_OPCODE = 0x2008, + TSTIMM_OPCODE = 0xc800, + TSTB_OPCODE = 0xcc00, + EXTUB_OPCODE = 0x600c, + EXTUW_OPCODE = 0x600d, + XOR_OPCODE = 0x200a, + XORIMM_OPCODE = 0xca00, + XORB_OPCODE = 0xce00, + FMOVS_READ_RM_INC_OPCODE = 0xf009, + FMOVS_READ_RM_OPCODE = 0xf008, + FMOVS_READ_R0RM_OPCODE = 0xf006, + FMOVS_WRITE_RN_OPCODE = 0xf00a, + FMOVS_WRITE_RN_DEC_OPCODE = 0xf00b, + FMOVS_WRITE_R0RN_OPCODE = 0xf007, + FCNVDS_DRM_FPUL_OPCODE = 0xf0bd, + FCNVSD_FPUL_DRN_OPCODE = 0xf0ad, + LDS_RM_FPUL_OPCODE = 0x405a, + FLDS_FRM_FPUL_OPCODE = 0xf01d, + STS_FPUL_RN_OPCODE = 0x005a, + FSTS_FPUL_FRN_OPCODE = 0xF00d, + LDSFPSCR_OPCODE = 0x406a, + STSFPSCR_OPCODE = 0x006a, + LDSRMFPUL_OPCODE = 0x405a, + FSTSFPULFRN_OPCODE = 0xf00d, + FSQRT_OPCODE = 0xf06d, + FSCHG_OPCODE = 0xf3fd, + CLRT_OPCODE = 8, +}; + +namespace SH4Registers { +typedef enum { + r0, + r1, + r2, + r3, + r4, + r5, + r6, + r7, + r8, + r9, + r10, + r11, + r12, + r13, + r14, fp = r14, + r15, sp = r15, + pc, + pr, +} RegisterID; + +typedef enum { + fr0, dr0 = fr0, + fr1, + fr2, dr2 = fr2, + fr3, + fr4, dr4 = fr4, + fr5, + fr6, dr6 = fr6, + fr7, + fr8, dr8 = fr8, + fr9, + fr10, dr10 = fr10, + fr11, + fr12, dr12 = fr12, + fr13, + fr14, dr14 = fr14, + fr15, +} FPRegisterID; +} + +inline uint16_t getOpcodeGroup1(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0xf) << 8) | ((rn & 0xf) << 4)); +} + +inline uint16_t getOpcodeGroup2(uint16_t opc, int rm) +{ + return (opc | ((rm & 0xf) << 8)); +} + +inline uint16_t getOpcodeGroup3(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0xf) << 8) | (rn & 0xff)); +} + +inline uint16_t getOpcodeGroup4(uint16_t opc, int rm, int rn, int offset) +{ + return (opc | ((rm & 0xf) << 8) | ((rn & 0xf) << 4) | (offset & 0xf)); +} + +inline uint16_t getOpcodeGroup5(uint16_t opc, int rm) +{ + return (opc | (rm & 0xff)); +} + +inline uint16_t getOpcodeGroup6(uint16_t opc, int rm) +{ + return (opc | (rm & 0xfff)); +} + +inline uint16_t getOpcodeGroup7(uint16_t opc, int rm) +{ + return (opc | ((rm & 0x7) << 9)); +} + +inline uint16_t getOpcodeGroup8(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0x7) << 9) | ((rn & 0x7) << 5)); +} + +inline uint16_t getOpcodeGroup9(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0xf) << 8) | ((rn & 0x7) << 5)); +} + +inline uint16_t getOpcodeGroup10(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0x7) << 9) | ((rn & 0xf) << 4)); +} + +inline uint16_t getOpcodeGroup11(uint16_t opc, int rm, int rn) +{ + return (opc | ((rm & 0xf) << 4) | (rn & 0xf)); +} + +inline uint16_t getRn(uint16_t x) +{ + return ((x & 0xf00) >> 8); +} + +inline uint16_t getRm(uint16_t x) +{ + return ((x & 0xf0) >> 4); +} + +inline uint16_t getDisp(uint16_t x) +{ + return (x & 0xf); +} + +inline uint16_t getImm8(uint16_t x) +{ + return (x & 0xff); +} + +inline uint16_t getImm12(uint16_t x) +{ + return (x & 0xfff); +} + +inline uint16_t getDRn(uint16_t x) +{ + return ((x & 0xe00) >> 9); +} + +inline uint16_t getDRm(uint16_t x) +{ + return ((x & 0xe0) >> 5); +} + +class SH4Assembler { +public: + typedef SH4Registers::RegisterID RegisterID; + typedef SH4Registers::FPRegisterID FPRegisterID; + typedef AssemblerBufferWithConstantPool<512, 4, 2, SH4Assembler> SH4Buffer; + static const RegisterID scratchReg1 = SH4Registers::r3; + static const RegisterID scratchReg2 = SH4Registers::r11; + static const uint32_t maxInstructionSize = 16; + + enum { + padForAlign8 = 0x00, + padForAlign16 = 0x0009, + padForAlign32 = 0x00090009, + }; + + enum JumpType { + JumpFar, + JumpNear + }; + + SH4Assembler() + { + m_claimscratchReg = 0x0; + } + + // SH4 condition codes + typedef enum { + EQ = 0x0, // Equal + NE = 0x1, // Not Equal + HS = 0x2, // Unsigend Greater Than equal + HI = 0x3, // Unsigend Greater Than + LS = 0x4, // Unsigend Lower or Same + LI = 0x5, // Unsigend Lower + GE = 0x6, // Greater or Equal + LT = 0x7, // Less Than + GT = 0x8, // Greater Than + LE = 0x9, // Less or Equal + OF = 0xa, // OverFlow + SI = 0xb, // Signed + EQU= 0xc, // Equal or unordered(NaN) + NEU= 0xd, + GTU= 0xe, + GEU= 0xf, + LTU= 0x10, + LEU= 0x11, + } Condition; + + // Opaque label types +public: + bool isImmediate(int constant) + { + return ((constant <= 127) && (constant >= -128)); + } + + RegisterID claimScratch() + { + ASSERT((m_claimscratchReg != 0x3)); + + if (!(m_claimscratchReg & 0x1)) { + m_claimscratchReg = (m_claimscratchReg | 0x1); + return scratchReg1; + } + + m_claimscratchReg = (m_claimscratchReg | 0x2); + return scratchReg2; + } + + void releaseScratch(RegisterID scratchR) + { + if (scratchR == scratchReg1) + m_claimscratchReg = (m_claimscratchReg & 0x2); + else + m_claimscratchReg = (m_claimscratchReg & 0x1); + } + + // Stack operations + + void pushReg(RegisterID reg) + { + if (reg == SH4Registers::pr) { + oneShortOp(getOpcodeGroup2(STSLPR_OPCODE, SH4Registers::sp)); + return; + } + + oneShortOp(getOpcodeGroup1(MOVL_WRITE_RNDEC_OPCODE, SH4Registers::sp, reg)); + } + + void popReg(RegisterID reg) + { + if (reg == SH4Registers::pr) { + oneShortOp(getOpcodeGroup2(LDSLPR_OPCODE, SH4Registers::sp)); + return; + } + + oneShortOp(getOpcodeGroup1(MOVL_READ_RMINC_OPCODE, reg, SH4Registers::sp)); + } + + void movt(RegisterID dst) + { + uint16_t opc = getOpcodeGroup2(MOVT_OPCODE, dst); + oneShortOp(opc); + } + + // Arithmetic operations + + void addlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(ADD_OPCODE, dst, src); + oneShortOp(opc); + } + + void addclRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(ADDC_OPCODE, dst, src); + oneShortOp(opc); + } + + void addvlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(ADDV_OPCODE, dst, src); + oneShortOp(opc); + } + + void addlImm8r(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 127) && (imm8 >= -128)); + + uint16_t opc = getOpcodeGroup3(ADDIMM_OPCODE, dst, imm8); + oneShortOp(opc); + } + + void andlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(AND_OPCODE, dst, src); + oneShortOp(opc); + } + + void andlImm8r(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 255) && (imm8 >= 0)); + ASSERT(dst == SH4Registers::r0); + + uint16_t opc = getOpcodeGroup5(ANDIMM_OPCODE, imm8); + oneShortOp(opc); + } + + void div1lRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(DIV1_OPCODE, dst, src); + oneShortOp(opc); + } + + void div0lRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(DIV0_OPCODE, dst, src); + oneShortOp(opc); + } + + void notlReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(NOT_OPCODE, dst, src); + oneShortOp(opc); + } + + void orlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(OR_OPCODE, dst, src); + oneShortOp(opc); + } + + void orlImm8r(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 255) && (imm8 >= 0)); + ASSERT(dst == SH4Registers::r0); + + uint16_t opc = getOpcodeGroup5(ORIMM_OPCODE, imm8); + oneShortOp(opc); + } + + void sublRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(SUB_OPCODE, dst, src); + oneShortOp(opc); + } + + void subvlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(SUBV_OPCODE, dst, src); + oneShortOp(opc); + } + + void xorlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(XOR_OPCODE, dst, src); + oneShortOp(opc); + } + + void xorlImm8r(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 255) && (imm8 >= 0)); + ASSERT(dst == SH4Registers::r0); + + uint16_t opc = getOpcodeGroup5(XORIMM_OPCODE, imm8); + oneShortOp(opc); + } + + void shllImm8r(int imm, RegisterID dst) + { + switch (imm) { + case 1: + oneShortOp(getOpcodeGroup2(SHLL_OPCODE, dst)); + break; + case 2: + oneShortOp(getOpcodeGroup2(SHLL2_OPCODE, dst)); + break; + case 8: + oneShortOp(getOpcodeGroup2(SHLL8_OPCODE, dst)); + break; + case 16: + oneShortOp(getOpcodeGroup2(SHLL16_OPCODE, dst)); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + void neg(RegisterID dst, RegisterID src) + { + uint16_t opc = getOpcodeGroup1(NEG_OPCODE, dst, src); + oneShortOp(opc); + } + + void shllRegReg(RegisterID dst, RegisterID rShift) + { + uint16_t opc = getOpcodeGroup1(SHLD_OPCODE, dst, rShift); + oneShortOp(opc); + } + + void shlrRegReg(RegisterID dst, RegisterID rShift) + { + neg(rShift, rShift); + shllRegReg(dst, rShift); + } + + void sharRegReg(RegisterID dst, RegisterID rShift) + { + neg(rShift, rShift); + shaRegReg(dst, rShift); + } + + void shaRegReg(RegisterID dst, RegisterID rShift) + { + uint16_t opc = getOpcodeGroup1(SHAD_OPCODE, dst, rShift); + oneShortOp(opc); + } + + void shlrImm8r(int imm, RegisterID dst) + { + switch (imm) { + case 1: + oneShortOp(getOpcodeGroup2(SHLR_OPCODE, dst)); + break; + case 2: + oneShortOp(getOpcodeGroup2(SHLR2_OPCODE, dst)); + break; + case 8: + oneShortOp(getOpcodeGroup2(SHLR8_OPCODE, dst)); + break; + case 16: + oneShortOp(getOpcodeGroup2(SHLR16_OPCODE, dst)); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + void imullRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MULL_OPCODE, dst, src); + oneShortOp(opc); + } + + void dmullRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(DMULL_L_OPCODE, dst, src); + oneShortOp(opc); + } + + void dmulslRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(DMULSL_OPCODE, dst, src); + oneShortOp(opc); + } + + void stsmacl(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(STSMACL_OPCODE, reg); + oneShortOp(opc); + } + + void stsmach(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(STSMACH_OPCODE, reg); + oneShortOp(opc); + } + + // Comparisons + + void cmplRegReg(RegisterID left, RegisterID right, Condition cond) + { + switch (cond) { + case NE: + oneShortOp(getOpcodeGroup1(CMPEQ_OPCODE, right, left)); + break; + case GT: + oneShortOp(getOpcodeGroup1(CMPGT_OPCODE, right, left)); + break; + case EQ: + oneShortOp(getOpcodeGroup1(CMPEQ_OPCODE, right, left)); + break; + case GE: + oneShortOp(getOpcodeGroup1(CMPGE_OPCODE, right, left)); + break; + case HS: + oneShortOp(getOpcodeGroup1(CMPHS_OPCODE, right, left)); + break; + case HI: + oneShortOp(getOpcodeGroup1(CMPHI_OPCODE, right, left)); + break; + case LI: + oneShortOp(getOpcodeGroup1(CMPHI_OPCODE, left, right)); + break; + case LS: + oneShortOp(getOpcodeGroup1(CMPHS_OPCODE, left, right)); + break; + case LE: + oneShortOp(getOpcodeGroup1(CMPGE_OPCODE, left, right)); + break; + case LT: + oneShortOp(getOpcodeGroup1(CMPGT_OPCODE, left, right)); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + void cmppl(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(CMPPL_OPCODE, reg); + oneShortOp(opc); + } + + void cmppz(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(CMPPZ_OPCODE, reg); + oneShortOp(opc); + } + + void cmpEqImmR0(int imm, RegisterID dst) + { + uint16_t opc = getOpcodeGroup5(CMPEQIMM_OPCODE, imm); + oneShortOp(opc); + } + + void testlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(TST_OPCODE, dst, src); + oneShortOp(opc); + } + + void testlImm8r(int imm, RegisterID dst) + { + ASSERT((dst == SH4Registers::r0) && (imm <= 255) && (imm >= 0)); + + uint16_t opc = getOpcodeGroup5(TSTIMM_OPCODE, imm); + oneShortOp(opc); + } + + void nop() + { + oneShortOp(NOP_OPCODE, false); + } + + void sett() + { + oneShortOp(SETT_OPCODE); + } + + void clrt() + { + oneShortOp(CLRT_OPCODE); + } + + void fschg() + { + oneShortOp(FSCHG_OPCODE); + } + + void bkpt() + { + oneShortOp(BRK_OPCODE, false); + } + + void branch(uint16_t opc, int label) + { + switch (opc) { + case BT_OPCODE: + ASSERT((label <= 127) && (label >= -128)); + oneShortOp(getOpcodeGroup5(BT_OPCODE, label)); + break; + case BRA_OPCODE: + ASSERT((label <= 2047) && (label >= -2048)); + oneShortOp(getOpcodeGroup6(BRA_OPCODE, label)); + break; + case BF_OPCODE: + ASSERT((label <= 127) && (label >= -128)); + oneShortOp(getOpcodeGroup5(BF_OPCODE, label)); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + void branch(uint16_t opc, RegisterID reg) + { + switch (opc) { + case BRAF_OPCODE: + oneShortOp(getOpcodeGroup2(BRAF_OPCODE, reg)); + break; + case JMP_OPCODE: + oneShortOp(getOpcodeGroup2(JMP_OPCODE, reg)); + break; + case JSR_OPCODE: + oneShortOp(getOpcodeGroup2(JSR_OPCODE, reg)); + break; + case BSRF_OPCODE: + oneShortOp(getOpcodeGroup2(BSRF_OPCODE, reg)); + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + void ldspr(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(LDSPR_OPCODE, reg); + oneShortOp(opc); + } + + void stspr(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(STSPR_OPCODE, reg); + oneShortOp(opc); + } + + void extub(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(EXTUB_OPCODE, dst, src); + oneShortOp(opc); + } + + void extuw(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(EXTUW_OPCODE, dst, src); + oneShortOp(opc); + } + + // float operations + + void ldsrmfpul(RegisterID src) + { + uint16_t opc = getOpcodeGroup2(LDS_RM_FPUL_OPCODE, src); + oneShortOp(opc); + } + + void fneg(FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup2(FNEG_OPCODE, dst); + oneShortOp(opc, true, false); + } + + void fsqrt(FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup2(FSQRT_OPCODE, dst); + oneShortOp(opc, true, false); + } + + void stsfpulReg(RegisterID src) + { + uint16_t opc = getOpcodeGroup2(STS_FPUL_RN_OPCODE, src); + oneShortOp(opc); + } + + void floatfpulfrn(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup2(FLOAT_OPCODE, src); + oneShortOp(opc, true, false); + } + + void fmull(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMUL_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsReadrm(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_READ_RM_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsWriterm(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_WRITE_RN_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsWriter0r(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_WRITE_R0RN_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsReadr0r(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_READ_R0RM_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsReadrminc(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_READ_RM_INC_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void fmovsWriterndec(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(FMOVS_WRITE_RN_DEC_OPCODE, dst, src); + oneShortOp(opc, true, false); + } + + void ftrcRegfpul(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup2(FTRC_OPCODE, src); + oneShortOp(opc, true, false); + } + + void fldsfpul(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup2(FLDS_FRM_FPUL_OPCODE, src); + oneShortOp(opc); + } + + void fstsfpul(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup2(FSTS_FPUL_FRN_OPCODE, src); + oneShortOp(opc); + } + + void ldsfpscr(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(LDSFPSCR_OPCODE, reg); + oneShortOp(opc); + } + + void stsfpscr(RegisterID reg) + { + uint16_t opc = getOpcodeGroup2(STSFPSCR_OPCODE, reg); + oneShortOp(opc); + } + + // double operations + + void dcnvds(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup7(FCNVDS_DRM_FPUL_OPCODE, src >> 1); + oneShortOp(opc); + } + + void dcnvsd(FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup7(FCNVSD_FPUL_DRN_OPCODE, dst >> 1); + oneShortOp(opc); + } + + void dcmppeq(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FCMPEQ_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void dcmppgt(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FCMPGT_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void dmulRegReg(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FMUL_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void dsubRegReg(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FSUB_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void daddRegReg(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FADD_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void dmovRegReg(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FMOV_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void ddivRegReg(FPRegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup8(FDIV_OPCODE, dst >> 1, src >> 1); + oneShortOp(opc); + } + + void dsqrt(FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup7(FSQRT_OPCODE, dst >> 1); + oneShortOp(opc); + } + + void dneg(FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup7(FNEG_OPCODE, dst >> 1); + oneShortOp(opc); + } + + void fmovReadrm(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup10(FMOVS_READ_RM_OPCODE, dst >> 1, src); + oneShortOp(opc); + } + + void fmovWriterm(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup9(FMOVS_WRITE_RN_OPCODE, dst, src >> 1); + oneShortOp(opc); + } + + void fmovWriter0r(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup9(FMOVS_WRITE_R0RN_OPCODE, dst, src >> 1); + oneShortOp(opc); + } + + void fmovReadr0r(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup10(FMOVS_READ_R0RM_OPCODE, dst >> 1, src); + oneShortOp(opc); + } + + void fmovReadrminc(RegisterID src, FPRegisterID dst) + { + uint16_t opc = getOpcodeGroup10(FMOVS_READ_RM_INC_OPCODE, dst >> 1, src); + oneShortOp(opc); + } + + void fmovWriterndec(FPRegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup9(FMOVS_WRITE_RN_DEC_OPCODE, dst, src >> 1); + oneShortOp(opc); + } + + void floatfpulDreg(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup7(FLOAT_OPCODE, src >> 1); + oneShortOp(opc); + } + + void ftrcdrmfpul(FPRegisterID src) + { + uint16_t opc = getOpcodeGroup7(FTRC_OPCODE, src >> 1); + oneShortOp(opc); + } + + // Various move ops + + void movImm8(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 127) && (imm8 >= -128)); + + uint16_t opc = getOpcodeGroup3(MOVIMM_OPCODE, dst, imm8); + oneShortOp(opc); + } + + void movlRegReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOV_OPCODE, dst, src); + oneShortOp(opc); + } + + void movwRegMem(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVW_WRITE_RN_OPCODE, dst, src); + oneShortOp(opc); + } + + void movwMemReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVW_READ_RM_OPCODE, dst, src); + oneShortOp(opc); + } + + void movwPCReg(int offset, RegisterID base, RegisterID dst) + { + ASSERT(base == SH4Registers::pc); + ASSERT((offset <= 255) && (offset >= 0)); + + uint16_t opc = getOpcodeGroup3(MOVW_READ_OFFPC_OPCODE, dst, offset); + oneShortOp(opc); + } + + void movwMemReg(int offset, RegisterID base, RegisterID dst) + { + ASSERT(dst == SH4Registers::r0); + + uint16_t opc = getOpcodeGroup11(MOVW_READ_OFFRM_OPCODE, base, offset); + oneShortOp(opc); + } + + void movwR0mr(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVW_READ_R0RM_OPCODE, dst, src); + oneShortOp(opc); + } + + void movlRegMem(RegisterID src, int offset, RegisterID base) + { + ASSERT((offset <= 15) && (offset >= 0)); + + if (!offset) { + oneShortOp(getOpcodeGroup1(MOVL_WRITE_RN_OPCODE, base, src)); + return; + } + + oneShortOp(getOpcodeGroup4(MOVL_WRITE_OFFRN_OPCODE, base, src, offset)); + } + + void movlRegMem(RegisterID src, RegisterID base) + { + uint16_t opc = getOpcodeGroup1(MOVL_WRITE_RN_OPCODE, base, src); + oneShortOp(opc); + } + + void movlMemReg(int offset, RegisterID base, RegisterID dst) + { + if (base == SH4Registers::pc) { + ASSERT((offset <= 255) && (offset >= 0)); + oneShortOp(getOpcodeGroup3(MOVL_READ_OFFPC_OPCODE, dst, offset)); + return; + } + + ASSERT((offset <= 15) && (offset >= 0)); + if (!offset) { + oneShortOp(getOpcodeGroup1(MOVL_READ_RM_OPCODE, dst, base)); + return; + } + + oneShortOp(getOpcodeGroup4(MOVL_READ_OFFRM_OPCODE, dst, base, offset)); + } + + void movlMemRegCompact(int offset, RegisterID base, RegisterID dst) + { + oneShortOp(getOpcodeGroup4(MOVL_READ_OFFRM_OPCODE, dst, base, offset)); + } + + void movbRegMem(RegisterID src, RegisterID base) + { + uint16_t opc = getOpcodeGroup1(MOVB_WRITE_RN_OPCODE, base, src); + oneShortOp(opc); + } + + void movbMemReg(int offset, RegisterID base, RegisterID dst) + { + ASSERT(dst == SH4Registers::r0); + + uint16_t opc = getOpcodeGroup11(MOVB_READ_OFFRM_OPCODE, base, offset); + oneShortOp(opc); + } + + void movbR0mr(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVB_READ_R0RM_OPCODE, dst, src); + oneShortOp(opc); + } + + void movbMemReg(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVB_READ_RM_OPCODE, dst, src); + oneShortOp(opc); + } + + void movlMemReg(RegisterID base, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVL_READ_RM_OPCODE, dst, base); + oneShortOp(opc); + } + + void movlMemRegIn(RegisterID base, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVL_READ_RMINC_OPCODE, dst, base); + oneShortOp(opc); + } + + void movlR0mr(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVL_READ_R0RM_OPCODE, dst, src); + oneShortOp(opc); + } + + void movlRegMemr0(RegisterID src, RegisterID dst) + { + uint16_t opc = getOpcodeGroup1(MOVL_WRITE_R0RN_OPCODE, dst, src); + oneShortOp(opc); + } + + void movlImm8r(int imm8, RegisterID dst) + { + ASSERT((imm8 <= 127) && (imm8 >= -128)); + + uint16_t opc = getOpcodeGroup3(MOVIMM_OPCODE, dst, imm8); + oneShortOp(opc); + } + + void loadConstant(uint32_t constant, RegisterID dst) + { + if (((int)constant <= 0x7f) && ((int)constant >= -0x80)) { + movImm8(constant, dst); + return; + } + + uint16_t opc = getOpcodeGroup3(MOVIMM_OPCODE, dst, 0); + + m_buffer.ensureSpace(maxInstructionSize, sizeof(uint32_t)); + printInstr(getOpcodeGroup3(MOVIMM_OPCODE, dst, constant), m_buffer.codeSize()); + m_buffer.putShortWithConstantInt(opc, constant, true); + } + + void loadConstantUnReusable(uint32_t constant, RegisterID dst, bool ensureSpace = false) + { + uint16_t opc = getOpcodeGroup3(MOVIMM_OPCODE, dst, 0); + + if (ensureSpace) + m_buffer.ensureSpace(maxInstructionSize, sizeof(uint32_t)); + + printInstr(getOpcodeGroup3(MOVIMM_OPCODE, dst, constant), m_buffer.codeSize()); + m_buffer.putShortWithConstantInt(opc, constant); + } + + // Flow control + + AssemblerLabel call() + { + RegisterID scr = claimScratch(); + m_buffer.ensureSpace(maxInstructionSize + 4, sizeof(uint32_t)); + loadConstantUnReusable(0x0, scr); + branch(JSR_OPCODE, scr); + nop(); + releaseScratch(scr); + return m_buffer.label(); + } + + AssemblerLabel call(RegisterID dst) + { + m_buffer.ensureSpace(maxInstructionSize + 2); + branch(JSR_OPCODE, dst); + nop(); + return m_buffer.label(); + } + + AssemblerLabel jmp() + { + RegisterID scr = claimScratch(); + m_buffer.ensureSpace(maxInstructionSize + 4, sizeof(uint32_t)); + AssemblerLabel label = m_buffer.label(); + loadConstantUnReusable(0x0, scr); + branch(BRAF_OPCODE, scr); + nop(); + releaseScratch(scr); + return label; + } + + void extraInstrForBranch(RegisterID dst) + { + loadConstantUnReusable(0x0, dst); + nop(); + nop(); + } + + AssemblerLabel jmp(RegisterID dst) + { + jmpReg(dst); + return m_buffer.label(); + } + + void jmpReg(RegisterID dst) + { + m_buffer.ensureSpace(maxInstructionSize + 2); + branch(JMP_OPCODE, dst); + nop(); + } + + AssemblerLabel jne() + { + AssemblerLabel label = m_buffer.label(); + branch(BF_OPCODE, 0); + return label; + } + + AssemblerLabel je() + { + AssemblerLabel label = m_buffer.label(); + branch(BT_OPCODE, 0); + return label; + } + + AssemblerLabel bra() + { + AssemblerLabel label = m_buffer.label(); + branch(BRA_OPCODE, 0); + return label; + } + + void ret() + { + m_buffer.ensureSpace(maxInstructionSize + 2); + oneShortOp(RTS_OPCODE, false); + } + + AssemblerLabel labelIgnoringWatchpoints() + { + m_buffer.ensureSpaceForAnyInstruction(); + return m_buffer.label(); + } + + AssemblerLabel label() + { + m_buffer.ensureSpaceForAnyInstruction(); + return m_buffer.label(); + } + + int sizeOfConstantPool() + { + return m_buffer.sizeOfConstantPool(); + } + + AssemblerLabel align(int alignment) + { + m_buffer.ensureSpace(maxInstructionSize + 2); + while (!m_buffer.isAligned(alignment)) { + nop(); + m_buffer.ensureSpace(maxInstructionSize + 2); + } + return label(); + } + + static void changePCrelativeAddress(int offset, uint16_t* instructionPtr, uint32_t newAddress) + { + uint32_t address = (offset << 2) + ((reinterpret_cast<uint32_t>(instructionPtr) + 4) &(~0x3)); + *reinterpret_cast<uint32_t*>(address) = newAddress; + } + + static uint32_t readPCrelativeAddress(int offset, uint16_t* instructionPtr) + { + uint32_t address = (offset << 2) + ((reinterpret_cast<uint32_t>(instructionPtr) + 4) &(~0x3)); + return *reinterpret_cast<uint32_t*>(address); + } + + static uint16_t* getInstructionPtr(void* code, int offset) + { + return reinterpret_cast<uint16_t*> (reinterpret_cast<uint32_t>(code) + offset); + } + + static void linkJump(void* code, AssemblerLabel from, void* to) + { + ASSERT(from.isSet()); + + uint16_t* instructionPtr = getInstructionPtr(code, from.m_offset); + uint16_t instruction = *instructionPtr; + int offsetBits = (reinterpret_cast<uint32_t>(to) - reinterpret_cast<uint32_t>(code)) - from.m_offset; + + if (((instruction & 0xff00) == BT_OPCODE) || ((instruction & 0xff00) == BF_OPCODE)) { + /* BT label ==> BF 2 + nop LDR reg + nop braf @reg + nop nop + */ + offsetBits -= 8; + instruction ^= 0x0202; + *instructionPtr++ = instruction; + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits); + instruction = (BRAF_OPCODE | (*instructionPtr++ & 0xf00)); + *instructionPtr = instruction; + printBlockInstr(instructionPtr - 2, from.m_offset, 3); + return; + } + + /* MOV #imm, reg => LDR reg + braf @reg braf @reg + nop nop + */ + ASSERT((*(instructionPtr + 1) & BRAF_OPCODE) == BRAF_OPCODE); + + offsetBits -= 4; + if (offsetBits >= -4096 && offsetBits <= 4094) { + *instructionPtr = getOpcodeGroup6(BRA_OPCODE, offsetBits >> 1); + *(++instructionPtr) = NOP_OPCODE; + printBlockInstr(instructionPtr - 1, from.m_offset, 2); + return; + } + + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits - 2); + printInstr(*instructionPtr, from.m_offset + 2); + } + + static void linkCall(void* code, AssemblerLabel from, void* to) + { + uint16_t* instructionPtr = getInstructionPtr(code, from.m_offset); + instructionPtr -= 3; + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, reinterpret_cast<uint32_t>(to)); + } + + static void linkPointer(void* code, AssemblerLabel where, void* value) + { + uint16_t* instructionPtr = getInstructionPtr(code, where.m_offset); + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, reinterpret_cast<uint32_t>(value)); + } + + static unsigned getCallReturnOffset(AssemblerLabel call) + { + ASSERT(call.isSet()); + return call.m_offset; + } + + static uint32_t* getLdrImmAddressOnPool(SH4Word* insn, uint32_t* constPool) + { + return (constPool + (*insn & 0xff)); + } + + static SH4Word patchConstantPoolLoad(SH4Word load, int value) + { + return ((load & ~0xff) | value); + } + + static SH4Buffer::TwoShorts placeConstantPoolBarrier(int offset) + { + ASSERT(((offset >> 1) <=2047) && ((offset >> 1) >= -2048)); + + SH4Buffer::TwoShorts m_barrier; + m_barrier.high = (BRA_OPCODE | (offset >> 1)); + m_barrier.low = NOP_OPCODE; + printInstr(((BRA_OPCODE | (offset >> 1))), 0); + printInstr(NOP_OPCODE, 0); + return m_barrier; + } + + static void patchConstantPoolLoad(void* loadAddr, void* constPoolAddr) + { + SH4Word* instructionPtr = reinterpret_cast<SH4Word*>(loadAddr); + SH4Word instruction = *instructionPtr; + SH4Word index = instruction & 0xff; + + if ((instruction & 0xf000) != MOVIMM_OPCODE) + return; + + ASSERT((((reinterpret_cast<uint32_t>(constPoolAddr) - reinterpret_cast<uint32_t>(loadAddr)) + index * 4)) < 1024); + + int offset = reinterpret_cast<uint32_t>(constPoolAddr) + (index * 4) - ((reinterpret_cast<uint32_t>(instructionPtr) & ~0x03) + 4); + instruction &=0xf00; + instruction |= 0xd000; + offset &= 0x03ff; + instruction |= (offset >> 2); + *instructionPtr = instruction; + printInstr(instruction, reinterpret_cast<uint32_t>(loadAddr)); + } + + static void repatchPointer(void* where, void* value) + { + patchPointer(where, value); + } + + static void* readPointer(void* code) + { + return reinterpret_cast<void*>(readInt32(code)); + } + + static void repatchInt32(void* where, int32_t value) + { + uint16_t* instructionPtr = reinterpret_cast<uint16_t*>(where); + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, value); + } + + static void repatchCompact(void* where, int32_t value) + { + ASSERT(value >= 0); + ASSERT(value <= 60); + *reinterpret_cast<uint16_t*>(where) = ((*reinterpret_cast<uint16_t*>(where) & 0xfff0) | (value >> 2)); + cacheFlush(reinterpret_cast<uint16_t*>(where), sizeof(uint16_t)); + } + + static void relinkCall(void* from, void* to) + { + uint16_t* instructionPtr = reinterpret_cast<uint16_t*>(from); + instructionPtr -= 3; + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, reinterpret_cast<uint32_t>(to)); + } + + static void relinkJump(void* from, void* to) + { + uint16_t* instructionPtr = reinterpret_cast<uint16_t*> (from); + uint16_t instruction = *instructionPtr; + int32_t offsetBits = (reinterpret_cast<uint32_t>(to) - reinterpret_cast<uint32_t>(from)); + + if (((*instructionPtr & 0xff00) == BT_OPCODE) || ((*instructionPtr & 0xff00) == BF_OPCODE)) { + offsetBits -= 8; + instructionPtr++; + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits); + instruction = (BRAF_OPCODE | (*instructionPtr++ & 0xf00)); + *instructionPtr = instruction; + printBlockInstr(instructionPtr, reinterpret_cast<uint32_t>(from) + 1, 3); + return; + } + + ASSERT((*(instructionPtr + 1) & BRAF_OPCODE) == BRAF_OPCODE); + offsetBits -= 4; + if (offsetBits >= -4096 && offsetBits <= 4094) { + *instructionPtr = getOpcodeGroup6(BRA_OPCODE, offsetBits >> 1); + *(++instructionPtr) = NOP_OPCODE; + printBlockInstr(instructionPtr - 2, reinterpret_cast<uint32_t>(from), 2); + return; + } + + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits - 2); + printInstr(*instructionPtr, reinterpret_cast<uint32_t>(from)); + } + + // Linking & patching + + static void revertJump(void* instructionStart, SH4Word imm) + { + SH4Word *insn = reinterpret_cast<SH4Word*>(instructionStart); + SH4Word disp; + + ASSERT((insn[0] & 0xf000) == MOVL_READ_OFFPC_OPCODE); + + disp = insn[0] & 0x00ff; + insn += 2 + (disp << 1); // PC += 4 + (disp*4) + insn = (SH4Word *) ((unsigned) insn & (~3)); + insn[0] = imm; + cacheFlush(insn, sizeof(SH4Word)); + } + + void linkJump(AssemblerLabel from, AssemblerLabel to, JumpType type = JumpFar) + { + ASSERT(to.isSet()); + ASSERT(from.isSet()); + + uint16_t* instructionPtr = getInstructionPtr(data(), from.m_offset); + uint16_t instruction = *instructionPtr; + int offsetBits; + + if (type == JumpNear) { + ASSERT((instruction == BT_OPCODE) || (instruction == BF_OPCODE) || (instruction == BRA_OPCODE)); + int offset = (codeSize() - from.m_offset) - 4; + *instructionPtr++ = instruction | (offset >> 1); + printInstr(*instructionPtr, from.m_offset + 2); + return; + } + + if (((instruction & 0xff00) == BT_OPCODE) || ((instruction & 0xff00) == BF_OPCODE)) { + /* BT label => BF 2 + nop LDR reg + nop braf @reg + nop nop + */ + offsetBits = (to.m_offset - from.m_offset) - 8; + instruction ^= 0x0202; + *instructionPtr++ = instruction; + if ((*instructionPtr & 0xf000) == 0xe000) { + uint32_t* addr = getLdrImmAddressOnPool(instructionPtr, m_buffer.poolAddress()); + *addr = offsetBits; + } else + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits); + instruction = (BRAF_OPCODE | (*instructionPtr++ & 0xf00)); + *instructionPtr = instruction; + printBlockInstr(instructionPtr - 2, from.m_offset, 3); + return; + } + + /* MOV # imm, reg => LDR reg + braf @reg braf @reg + nop nop + */ + ASSERT((*(instructionPtr + 1) & BRAF_OPCODE) == BRAF_OPCODE); + offsetBits = (to.m_offset - from.m_offset) - 4; + if (offsetBits >= -4096 && offsetBits <= 4094) { + *instructionPtr = getOpcodeGroup6(BRA_OPCODE, offsetBits >> 1); + *(++instructionPtr) = NOP_OPCODE; + printBlockInstr(instructionPtr - 1, from.m_offset, 2); + return; + } + + instruction = *instructionPtr; + if ((instruction & 0xf000) == 0xe000) { + uint32_t* addr = getLdrImmAddressOnPool(instructionPtr, m_buffer.poolAddress()); + *addr = offsetBits - 2; + printInstr(*instructionPtr, from.m_offset + 2); + return; + } + + changePCrelativeAddress((*instructionPtr & 0xff), instructionPtr, offsetBits - 2); + printInstr(*instructionPtr, from.m_offset + 2); + } + + static void* getRelocatedAddress(void* code, AssemblerLabel label) + { + return reinterpret_cast<void*>(reinterpret_cast<char*>(code) + label.m_offset); + } + + static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b) + { + return b.m_offset - a.m_offset; + } + + static void patchPointer(void* code, AssemblerLabel where, void* value) + { + patchPointer(reinterpret_cast<uint32_t*>(code) + where.m_offset, value); + } + + static void patchPointer(void* code, void* value) + { + patchInt32(code, reinterpret_cast<uint32_t>(value)); + } + + static void patchInt32(void* code, uint32_t value) + { + changePCrelativeAddress((*(reinterpret_cast<uint16_t*>(code)) & 0xff), reinterpret_cast<uint16_t*>(code), value); + } + + static uint32_t readInt32(void* code) + { + return readPCrelativeAddress((*(reinterpret_cast<uint16_t*>(code)) & 0xff), reinterpret_cast<uint16_t*>(code)); + } + + static void* readCallTarget(void* from) + { + uint16_t* instructionPtr = static_cast<uint16_t*>(from); + instructionPtr -= 3; + return reinterpret_cast<void*>(readPCrelativeAddress((*instructionPtr & 0xff), instructionPtr)); + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + return m_buffer.executableCopy(globalData, ownerUID, effort); + } + + static void cacheFlush(void* code, size_t size) + { +#if !OS(LINUX) +#error "The cacheFlush support is missing on this platform." +#elif defined CACHEFLUSH_D_L2 + syscall(__NR_cacheflush, reinterpret_cast<unsigned>(code), size, CACHEFLUSH_D_WB | CACHEFLUSH_I | CACHEFLUSH_D_L2); +#else + syscall(__NR_cacheflush, reinterpret_cast<unsigned>(code), size, CACHEFLUSH_D_WB | CACHEFLUSH_I); +#endif + } + + void prefix(uint16_t pre) + { + m_buffer.putByte(pre); + } + + void oneShortOp(uint16_t opcode, bool checksize = true, bool isDouble = true) + { + printInstr(opcode, m_buffer.codeSize(), isDouble); + if (checksize) + m_buffer.ensureSpace(maxInstructionSize); + m_buffer.putShortUnchecked(opcode); + } + + void ensureSpace(int space) + { + m_buffer.ensureSpace(space); + } + + void ensureSpace(int insnSpace, int constSpace) + { + m_buffer.ensureSpace(insnSpace, constSpace); + } + + // Administrative methods + + void* data() const { return m_buffer.data(); } + size_t codeSize() const { return m_buffer.codeSize(); } + +#ifdef SH4_ASSEMBLER_TRACING + static void printInstr(uint16_t opc, unsigned size, bool isdoubleInst = true) + { + if (!getenv("JavaScriptCoreDumpJIT")) + return; + + const char *format = 0; + printfStdoutInstr("offset: 0x%8.8x\t", size); + switch (opc) { + case BRK_OPCODE: + format = " BRK\n"; + break; + case NOP_OPCODE: + format = " NOP\n"; + break; + case RTS_OPCODE: + format =" *RTS\n"; + break; + case SETS_OPCODE: + format = " SETS\n"; + break; + case SETT_OPCODE: + format = " SETT\n"; + break; + case CLRT_OPCODE: + format = " CLRT\n"; + break; + case FSCHG_OPCODE: + format = " FSCHG\n"; + break; + } + if (format) { + printfStdoutInstr(format); + return; + } + switch (opc & 0xf0ff) { + case BRAF_OPCODE: + format = " *BRAF R%d\n"; + break; + case DT_OPCODE: + format = " DT R%d\n"; + break; + case CMPPL_OPCODE: + format = " CMP/PL R%d\n"; + break; + case CMPPZ_OPCODE: + format = " CMP/PZ R%d\n"; + break; + case JMP_OPCODE: + format = " *JMP @R%d\n"; + break; + case JSR_OPCODE: + format = " *JSR @R%d\n"; + break; + case LDSPR_OPCODE: + format = " LDS R%d, PR\n"; + break; + case LDSLPR_OPCODE: + format = " LDS.L @R%d+, PR\n"; + break; + case MOVT_OPCODE: + format = " MOVT R%d\n"; + break; + case SHAL_OPCODE: + format = " SHAL R%d\n"; + break; + case SHAR_OPCODE: + format = " SHAR R%d\n"; + break; + case SHLL_OPCODE: + format = " SHLL R%d\n"; + break; + case SHLL2_OPCODE: + format = " SHLL2 R%d\n"; + break; + case SHLL8_OPCODE: + format = " SHLL8 R%d\n"; + break; + case SHLL16_OPCODE: + format = " SHLL16 R%d\n"; + break; + case SHLR_OPCODE: + format = " SHLR R%d\n"; + break; + case SHLR2_OPCODE: + format = " SHLR2 R%d\n"; + break; + case SHLR8_OPCODE: + format = " SHLR8 R%d\n"; + break; + case SHLR16_OPCODE: + format = " SHLR16 R%d\n"; + break; + case STSPR_OPCODE: + format = " STS PR, R%d\n"; + break; + case STSLPR_OPCODE: + format = " STS.L PR, @-R%d\n"; + break; + case LDS_RM_FPUL_OPCODE: + format = " LDS R%d, FPUL\n"; + break; + case STS_FPUL_RN_OPCODE: + format = " STS FPUL, R%d \n"; + break; + case FLDS_FRM_FPUL_OPCODE: + format = " FLDS FR%d, FPUL\n"; + break; + case FSTS_FPUL_FRN_OPCODE: + format = " FSTS FPUL, R%d \n"; + break; + case LDSFPSCR_OPCODE: + format = " LDS R%d, FPSCR \n"; + break; + case STSFPSCR_OPCODE: + format = " STS FPSCR, R%d \n"; + break; + case STSMACL_OPCODE: + format = " STS MACL, R%d \n"; + break; + case STSMACH_OPCODE: + format = " STS MACH, R%d \n"; + break; + case BSRF_OPCODE: + format = " *BSRF R%d"; + break; + case FTRC_OPCODE: + format = " FTRC FR%d, FPUL\n"; + break; + } + if (format) { + printfStdoutInstr(format, getRn(opc)); + return; + } + switch (opc & 0xf0ff) { + case FNEG_OPCODE: + format = " FNEG DR%d\n"; + break; + case FLOAT_OPCODE: + format = " FLOAT DR%d\n"; + break; + case FTRC_OPCODE: + format = " FTRC FR%d, FPUL\n"; + break; + case FSQRT_OPCODE: + format = " FSQRT FR%d\n"; + break; + case FCNVDS_DRM_FPUL_OPCODE: + format = " FCNVDS FR%d, FPUL\n"; + break; + case FCNVSD_FPUL_DRN_OPCODE: + format = " FCNVSD FPUL, FR%d\n"; + break; + } + if (format) { + if (isdoubleInst) + printfStdoutInstr(format, getDRn(opc) << 1); + else + printfStdoutInstr(format, getRn(opc)); + return; + } + switch (opc & 0xf00f) { + case ADD_OPCODE: + format = " ADD R%d, R%d\n"; + break; + case ADDC_OPCODE: + format = " ADDC R%d, R%d\n"; + break; + case ADDV_OPCODE: + format = " ADDV R%d, R%d\n"; + break; + case AND_OPCODE: + format = " AND R%d, R%d\n"; + break; + case DIV1_OPCODE: + format = " DIV1 R%d, R%d\n"; + break; + case CMPEQ_OPCODE: + format = " CMP/EQ R%d, R%d\n"; + break; + case CMPGE_OPCODE: + format = " CMP/GE R%d, R%d\n"; + break; + case CMPGT_OPCODE: + format = " CMP/GT R%d, R%d\n"; + break; + case CMPHI_OPCODE: + format = " CMP/HI R%d, R%d\n"; + break; + case CMPHS_OPCODE: + format = " CMP/HS R%d, R%d\n"; + break; + case MOV_OPCODE: + format = " MOV R%d, R%d\n"; + break; + case MOVB_WRITE_RN_OPCODE: + format = " MOV.B R%d, @R%d\n"; + break; + case MOVB_WRITE_RNDEC_OPCODE: + format = " MOV.B R%d, @-R%d\n"; + break; + case MOVB_WRITE_R0RN_OPCODE: + format = " MOV.B R%d, @(R0, R%d)\n"; + break; + case MOVB_READ_RM_OPCODE: + format = " MOV.B @R%d, R%d\n"; + break; + case MOVB_READ_RMINC_OPCODE: + format = " MOV.B @R%d+, R%d\n"; + break; + case MOVB_READ_R0RM_OPCODE: + format = " MOV.B @(R0, R%d), R%d\n"; + break; + case MOVL_WRITE_RN_OPCODE: + format = " MOV.L R%d, @R%d\n"; + break; + case MOVL_WRITE_RNDEC_OPCODE: + format = " MOV.L R%d, @-R%d\n"; + break; + case MOVL_WRITE_R0RN_OPCODE: + format = " MOV.L R%d, @(R0, R%d)\n"; + break; + case MOVL_READ_RM_OPCODE: + format = " MOV.L @R%d, R%d\n"; + break; + case MOVL_READ_RMINC_OPCODE: + format = " MOV.L @R%d+, R%d\n"; + break; + case MOVL_READ_R0RM_OPCODE: + format = " MOV.L @(R0, R%d), R%d\n"; + break; + case MULL_OPCODE: + format = " MUL.L R%d, R%d\n"; + break; + case DMULL_L_OPCODE: + format = " DMULU.L R%d, R%d\n"; + break; + case DMULSL_OPCODE: + format = " DMULS.L R%d, R%d\n"; + break; + case NEG_OPCODE: + format = " NEG R%d, R%d\n"; + break; + case NEGC_OPCODE: + format = " NEGC R%d, R%d\n"; + break; + case NOT_OPCODE: + format = " NOT R%d, R%d\n"; + break; + case OR_OPCODE: + format = " OR R%d, R%d\n"; + break; + case SHAD_OPCODE: + format = " SHAD R%d, R%d\n"; + break; + case SHLD_OPCODE: + format = " SHLD R%d, R%d\n"; + break; + case SUB_OPCODE: + format = " SUB R%d, R%d\n"; + break; + case SUBC_OPCODE: + format = " SUBC R%d, R%d\n"; + break; + case SUBV_OPCODE: + format = " SUBV R%d, R%d\n"; + break; + case TST_OPCODE: + format = " TST R%d, R%d\n"; + break; + case XOR_OPCODE: + format = " XOR R%d, R%d\n";break; + case MOVW_WRITE_RN_OPCODE: + format = " MOV.W R%d, @R%d\n"; + break; + case MOVW_READ_RM_OPCODE: + format = " MOV.W @R%d, R%d\n"; + break; + case MOVW_READ_R0RM_OPCODE: + format = " MOV.W @(R0, R%d), R%d\n"; + break; + case EXTUB_OPCODE: + format = " EXTU.B R%d, R%d\n"; + break; + case EXTUW_OPCODE: + format = " EXTU.W R%d, R%d\n"; + break; + } + if (format) { + printfStdoutInstr(format, getRm(opc), getRn(opc)); + return; + } + switch (opc & 0xf00f) { + case FSUB_OPCODE: + format = " FSUB FR%d, FR%d\n"; + break; + case FADD_OPCODE: + format = " FADD FR%d, FR%d\n"; + break; + case FDIV_OPCODE: + format = " FDIV FR%d, FR%d\n"; + break; + case FMUL_OPCODE: + format = " DMULL FR%d, FR%d\n"; + break; + case FMOV_OPCODE: + format = " FMOV FR%d, FR%d\n"; + break; + case FCMPEQ_OPCODE: + format = " FCMP/EQ FR%d, FR%d\n"; + break; + case FCMPGT_OPCODE: + format = " FCMP/GT FR%d, FR%d\n"; + break; + } + if (format) { + if (isdoubleInst) + printfStdoutInstr(format, getDRm(opc) << 1, getDRn(opc) << 1); + else + printfStdoutInstr(format, getRm(opc), getRn(opc)); + return; + } + switch (opc & 0xf00f) { + case FMOVS_WRITE_RN_DEC_OPCODE: + format = " %s FR%d, @-R%d\n"; + break; + case FMOVS_WRITE_RN_OPCODE: + format = " %s FR%d, @R%d\n"; + break; + case FMOVS_WRITE_R0RN_OPCODE: + format = " %s FR%d, @(R0, R%d)\n"; + break; + } + if (format) { + if (isdoubleInst) + printfStdoutInstr(format, "FMOV", getDRm(opc) << 1, getDRn(opc)); + else + printfStdoutInstr(format, "FMOV.S", getRm(opc), getRn(opc)); + return; + } + switch (opc & 0xf00f) { + case FMOVS_READ_RM_OPCODE: + format = " %s @R%d, FR%d\n"; + break; + case FMOVS_READ_RM_INC_OPCODE: + format = " %s @R%d+, FR%d\n"; + break; + case FMOVS_READ_R0RM_OPCODE: + format = " %s @(R0, R%d), FR%d\n"; + break; + } + if (format) { + if (isdoubleInst) + printfStdoutInstr(format, "FMOV", getDRm(opc), getDRn(opc) << 1); + else + printfStdoutInstr(format, "FMOV.S", getRm(opc), getRn(opc)); + return; + } + switch (opc & 0xff00) { + case BF_OPCODE: + format = " BF %d\n"; + break; + case BFS_OPCODE: + format = " *BF/S %d\n"; + break; + case ANDIMM_OPCODE: + format = " AND #%d, R0\n"; + break; + case BT_OPCODE: + format = " BT %d\n"; + break; + case BTS_OPCODE: + format = " *BT/S %d\n"; + break; + case CMPEQIMM_OPCODE: + format = " CMP/EQ #%d, R0\n"; + break; + case MOVB_WRITE_OFFGBR_OPCODE: + format = " MOV.B R0, @(%d, GBR)\n"; + break; + case MOVB_READ_OFFGBR_OPCODE: + format = " MOV.B @(%d, GBR), R0\n"; + break; + case MOVL_WRITE_OFFGBR_OPCODE: + format = " MOV.L R0, @(%d, GBR)\n"; + break; + case MOVL_READ_OFFGBR_OPCODE: + format = " MOV.L @(%d, GBR), R0\n"; + break; + case MOVA_READ_OFFPC_OPCODE: + format = " MOVA @(%d, PC), R0\n"; + break; + case ORIMM_OPCODE: + format = " OR #%d, R0\n"; + break; + case ORBIMM_OPCODE: + format = " OR.B #%d, @(R0, GBR)\n"; + break; + case TSTIMM_OPCODE: + format = " TST #%d, R0\n"; + break; + case TSTB_OPCODE: + format = " TST.B %d, @(R0, GBR)\n"; + break; + case XORIMM_OPCODE: + format = " XOR #%d, R0\n"; + break; + case XORB_OPCODE: + format = " XOR.B %d, @(R0, GBR)\n"; + break; + } + if (format) { + printfStdoutInstr(format, getImm8(opc)); + return; + } + switch (opc & 0xff00) { + case MOVB_WRITE_OFFRN_OPCODE: + format = " MOV.B R0, @(%d, R%d)\n"; + break; + case MOVB_READ_OFFRM_OPCODE: + format = " MOV.B @(%d, R%d), R0\n"; + break; + } + if (format) { + printfStdoutInstr(format, getDisp(opc), getRm(opc)); + return; + } + switch (opc & 0xf000) { + case BRA_OPCODE: + format = " *BRA %d\n"; + break; + case BSR_OPCODE: + format = " *BSR %d\n"; + break; + } + if (format) { + printfStdoutInstr(format, getImm12(opc)); + return; + } + switch (opc & 0xf000) { + case MOVL_READ_OFFPC_OPCODE: + format = " MOV.L @(%d, PC), R%d\n"; + break; + case ADDIMM_OPCODE: + format = " ADD #%d, R%d\n"; + break; + case MOVIMM_OPCODE: + format = " MOV #%d, R%d\n"; + break; + case MOVW_READ_OFFPC_OPCODE: + format = " MOV.W @(%d, PC), R%d\n"; + break; + } + if (format) { + printfStdoutInstr(format, getImm8(opc), getRn(opc)); + return; + } + switch (opc & 0xf000) { + case MOVL_WRITE_OFFRN_OPCODE: + format = " MOV.L R%d, @(%d, R%d)\n"; + printfStdoutInstr(format, getRm(opc), getDisp(opc), getRn(opc)); + break; + case MOVL_READ_OFFRM_OPCODE: + format = " MOV.L @(%d, R%d), R%d\n"; + printfStdoutInstr(format, getDisp(opc), getRm(opc), getRn(opc)); + break; + } + } + + static void printfStdoutInstr(const char* format, ...) + { + if (getenv("JavaScriptCoreDumpJIT")) { + va_list args; + va_start(args, format); + vprintfStdoutInstr(format, args); + va_end(args); + } + } + + static void vprintfStdoutInstr(const char* format, va_list args) + { + if (getenv("JavaScriptCoreDumpJIT")) + WTF::dataLogFV(format, args); + } + + static void printBlockInstr(uint16_t* first, unsigned offset, int nbInstr) + { + printfStdoutInstr(">> repatch instructions after link\n"); + for (int i = 0; i <= nbInstr; i++) + printInstr(*(first + i), offset + i); + printfStdoutInstr(">> end repatch\n"); + } +#else + static void printInstr(uint16_t opc, unsigned size, bool isdoubleInst = true) { }; + static void printBlockInstr(uint16_t* first, unsigned offset, int nbInstr) { }; +#endif + + static void replaceWithLoad(void* instructionStart) + { + SH4Word* insPtr = reinterpret_cast<SH4Word*>(instructionStart); + + insPtr += 2; // skip MOV and ADD opcodes + + if (((*insPtr) & 0xf00f) != MOVL_READ_RM_OPCODE) { + *insPtr = MOVL_READ_RM_OPCODE | (*insPtr & 0x0ff0); + cacheFlush(insPtr, sizeof(SH4Word)); + } + } + + static void replaceWithAddressComputation(void* instructionStart) + { + SH4Word* insPtr = reinterpret_cast<SH4Word*>(instructionStart); + + insPtr += 2; // skip MOV and ADD opcodes + + if (((*insPtr) & 0xf00f) != MOV_OPCODE) { + *insPtr = MOV_OPCODE | (*insPtr & 0x0ff0); + cacheFlush(insPtr, sizeof(SH4Word)); + } + } + +private: + SH4Buffer m_buffer; + int m_claimscratchReg; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(SH4) + +#endif // SH4Assembler_h diff --git a/src/3rdparty/masm/assembler/X86Assembler.h b/src/3rdparty/masm/assembler/X86Assembler.h new file mode 100644 index 0000000000..092e775ab5 --- /dev/null +++ b/src/3rdparty/masm/assembler/X86Assembler.h @@ -0,0 +1,2540 @@ +/* + * Copyright (C) 2008, 2012 Apple Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR + * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR + * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY + * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef X86Assembler_h +#define X86Assembler_h + +#if ENABLE(ASSEMBLER) && (CPU(X86) || CPU(X86_64)) + +#include "AssemblerBuffer.h" +#include "JITCompilationEffort.h" +#include <stdint.h> +#include <wtf/Assertions.h> +#include <wtf/Vector.h> + +namespace JSC { + +inline bool CAN_SIGN_EXTEND_8_32(int32_t value) { return value == (int32_t)(signed char)value; } + +namespace X86Registers { + typedef enum { + eax, + ecx, + edx, + ebx, + esp, + ebp, + esi, + edi, + +#if CPU(X86_64) + r8, + r9, + r10, + r11, + r12, + r13, + r14, + r15, +#endif + } RegisterID; + + typedef enum { + xmm0, + xmm1, + xmm2, + xmm3, + xmm4, + xmm5, + xmm6, + xmm7, + } XMMRegisterID; +} + +class X86Assembler { +public: + typedef X86Registers::RegisterID RegisterID; + typedef X86Registers::XMMRegisterID XMMRegisterID; + typedef XMMRegisterID FPRegisterID; + + typedef enum { + ConditionO, + ConditionNO, + ConditionB, + ConditionAE, + ConditionE, + ConditionNE, + ConditionBE, + ConditionA, + ConditionS, + ConditionNS, + ConditionP, + ConditionNP, + ConditionL, + ConditionGE, + ConditionLE, + ConditionG, + + ConditionC = ConditionB, + ConditionNC = ConditionAE, + } Condition; + +private: + typedef enum { + OP_ADD_EvGv = 0x01, + OP_ADD_GvEv = 0x03, + OP_OR_EvGv = 0x09, + OP_OR_GvEv = 0x0B, + OP_2BYTE_ESCAPE = 0x0F, + OP_AND_EvGv = 0x21, + OP_AND_GvEv = 0x23, + OP_SUB_EvGv = 0x29, + OP_SUB_GvEv = 0x2B, + PRE_PREDICT_BRANCH_NOT_TAKEN = 0x2E, + OP_XOR_EvGv = 0x31, + OP_XOR_GvEv = 0x33, + OP_CMP_EvGv = 0x39, + OP_CMP_GvEv = 0x3B, +#if CPU(X86_64) + PRE_REX = 0x40, +#endif + OP_PUSH_EAX = 0x50, + OP_POP_EAX = 0x58, +#if CPU(X86_64) + OP_MOVSXD_GvEv = 0x63, +#endif + PRE_OPERAND_SIZE = 0x66, + PRE_SSE_66 = 0x66, + OP_PUSH_Iz = 0x68, + OP_IMUL_GvEvIz = 0x69, + OP_GROUP1_EbIb = 0x80, + OP_GROUP1_EvIz = 0x81, + OP_GROUP1_EvIb = 0x83, + OP_TEST_EbGb = 0x84, + OP_TEST_EvGv = 0x85, + OP_XCHG_EvGv = 0x87, + OP_MOV_EbGb = 0x88, + OP_MOV_EvGv = 0x89, + OP_MOV_GvEv = 0x8B, + OP_LEA = 0x8D, + OP_GROUP1A_Ev = 0x8F, + OP_NOP = 0x90, + OP_CDQ = 0x99, + OP_MOV_EAXOv = 0xA1, + OP_MOV_OvEAX = 0xA3, + OP_MOV_EAXIv = 0xB8, + OP_GROUP2_EvIb = 0xC1, + OP_RET = 0xC3, + OP_GROUP11_EvIb = 0xC6, + OP_GROUP11_EvIz = 0xC7, + OP_INT3 = 0xCC, + OP_GROUP2_Ev1 = 0xD1, + OP_GROUP2_EvCL = 0xD3, + OP_ESCAPE_DD = 0xDD, + OP_CALL_rel32 = 0xE8, + OP_JMP_rel32 = 0xE9, + PRE_SSE_F2 = 0xF2, + PRE_SSE_F3 = 0xF3, + OP_HLT = 0xF4, + OP_GROUP3_EbIb = 0xF6, + OP_GROUP3_Ev = 0xF7, + OP_GROUP3_EvIz = 0xF7, // OP_GROUP3_Ev has an immediate, when instruction is a test. + OP_GROUP5_Ev = 0xFF, + } OneByteOpcodeID; + + typedef enum { + OP2_MOVSD_VsdWsd = 0x10, + OP2_MOVSD_WsdVsd = 0x11, + OP2_MOVSS_VsdWsd = 0x10, + OP2_MOVSS_WsdVsd = 0x11, + OP2_CVTSI2SD_VsdEd = 0x2A, + OP2_CVTTSD2SI_GdWsd = 0x2C, + OP2_UCOMISD_VsdWsd = 0x2E, + OP2_ADDSD_VsdWsd = 0x58, + OP2_MULSD_VsdWsd = 0x59, + OP2_CVTSD2SS_VsdWsd = 0x5A, + OP2_CVTSS2SD_VsdWsd = 0x5A, + OP2_SUBSD_VsdWsd = 0x5C, + OP2_DIVSD_VsdWsd = 0x5E, + OP2_SQRTSD_VsdWsd = 0x51, + OP2_ANDNPD_VpdWpd = 0x55, + OP2_XORPD_VpdWpd = 0x57, + OP2_MOVD_VdEd = 0x6E, + OP2_MOVD_EdVd = 0x7E, + OP2_JCC_rel32 = 0x80, + OP_SETCC = 0x90, + OP2_IMUL_GvEv = 0xAF, + OP2_MOVZX_GvEb = 0xB6, + OP2_MOVSX_GvEb = 0xBE, + OP2_MOVZX_GvEw = 0xB7, + OP2_MOVSX_GvEw = 0xBF, + OP2_PEXTRW_GdUdIb = 0xC5, + OP2_PSLLQ_UdqIb = 0x73, + OP2_PSRLQ_UdqIb = 0x73, + OP2_POR_VdqWdq = 0XEB, + } TwoByteOpcodeID; + + TwoByteOpcodeID jccRel32(Condition cond) + { + return (TwoByteOpcodeID)(OP2_JCC_rel32 + cond); + } + + TwoByteOpcodeID setccOpcode(Condition cond) + { + return (TwoByteOpcodeID)(OP_SETCC + cond); + } + + typedef enum { + GROUP1_OP_ADD = 0, + GROUP1_OP_OR = 1, + GROUP1_OP_ADC = 2, + GROUP1_OP_AND = 4, + GROUP1_OP_SUB = 5, + GROUP1_OP_XOR = 6, + GROUP1_OP_CMP = 7, + + GROUP1A_OP_POP = 0, + + GROUP2_OP_ROL = 0, + GROUP2_OP_ROR = 1, + GROUP2_OP_RCL = 2, + GROUP2_OP_RCR = 3, + + GROUP2_OP_SHL = 4, + GROUP2_OP_SHR = 5, + GROUP2_OP_SAR = 7, + + GROUP3_OP_TEST = 0, + GROUP3_OP_NOT = 2, + GROUP3_OP_NEG = 3, + GROUP3_OP_IDIV = 7, + + GROUP5_OP_CALLN = 2, + GROUP5_OP_JMPN = 4, + GROUP5_OP_PUSH = 6, + + GROUP11_MOV = 0, + + GROUP14_OP_PSLLQ = 6, + GROUP14_OP_PSRLQ = 2, + + ESCAPE_DD_FSTP_doubleReal = 3, + } GroupOpcodeID; + + class X86InstructionFormatter; +public: + + X86Assembler() + : m_indexOfLastWatchpoint(INT_MIN) + , m_indexOfTailOfLastWatchpoint(INT_MIN) + { + } + + // Stack operations: + + void push_r(RegisterID reg) + { + m_formatter.oneByteOp(OP_PUSH_EAX, reg); + } + + void pop_r(RegisterID reg) + { + m_formatter.oneByteOp(OP_POP_EAX, reg); + } + + void push_i32(int imm) + { + m_formatter.oneByteOp(OP_PUSH_Iz); + m_formatter.immediate32(imm); + } + + void push_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_PUSH, base, offset); + } + + void pop_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP1A_Ev, GROUP1A_OP_POP, base, offset); + } + + // Arithmetic operations: + +#if !CPU(X86_64) + void adcl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_ADC, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_ADC, addr); + m_formatter.immediate32(imm); + } + } +#endif + + void addl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_ADD_EvGv, src, dst); + } + + void addl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_ADD_GvEv, dst, base, offset); + } + +#if !CPU(X86_64) + void addl_mr(const void* addr, RegisterID dst) + { + m_formatter.oneByteOp(OP_ADD_GvEv, dst, addr); + } +#endif + + void addl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_ADD_EvGv, src, base, offset); + } + + void addl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_ADD, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_ADD, dst); + m_formatter.immediate32(imm); + } + } + + void addl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_ADD, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_ADD, base, offset); + m_formatter.immediate32(imm); + } + } + +#if CPU(X86_64) + void addq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_ADD_EvGv, src, dst); + } + + void addq_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp64(OP_ADD_GvEv, dst, base, offset); + } + + void addq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_ADD, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_ADD, dst); + m_formatter.immediate32(imm); + } + } + + void addq_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_ADD, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_ADD, base, offset); + m_formatter.immediate32(imm); + } + } +#else + void addl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_ADD, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_ADD, addr); + m_formatter.immediate32(imm); + } + } +#endif + + void andl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_AND_EvGv, src, dst); + } + + void andl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_AND_GvEv, dst, base, offset); + } + + void andl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_AND_EvGv, src, base, offset); + } + + void andl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_AND, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_AND, dst); + m_formatter.immediate32(imm); + } + } + + void andl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_AND, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_AND, base, offset); + m_formatter.immediate32(imm); + } + } + +#if CPU(X86_64) + void andq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_AND_EvGv, src, dst); + } + + void andq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_AND, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_AND, dst); + m_formatter.immediate32(imm); + } + } +#else + void andl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_AND, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_AND, addr); + m_formatter.immediate32(imm); + } + } +#endif + + void negl_r(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP3_Ev, GROUP3_OP_NEG, dst); + } + +#if CPU(X86_64) + void negq_r(RegisterID dst) + { + m_formatter.oneByteOp64(OP_GROUP3_Ev, GROUP3_OP_NEG, dst); + } +#endif + + void negl_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP3_Ev, GROUP3_OP_NEG, base, offset); + } + + void notl_r(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP3_Ev, GROUP3_OP_NOT, dst); + } + + void notl_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP3_Ev, GROUP3_OP_NOT, base, offset); + } + + void orl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_OR_EvGv, src, dst); + } + + void orl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_OR_GvEv, dst, base, offset); + } + + void orl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_OR_EvGv, src, base, offset); + } + + void orl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_OR, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_OR, dst); + m_formatter.immediate32(imm); + } + } + + void orl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_OR, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_OR, base, offset); + m_formatter.immediate32(imm); + } + } + +#if CPU(X86_64) + void orq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_OR_EvGv, src, dst); + } + + void orq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_OR, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_OR, dst); + m_formatter.immediate32(imm); + } + } +#else + void orl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_OR, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_OR, addr); + m_formatter.immediate32(imm); + } + } + + void orl_rm(RegisterID src, const void* addr) + { + m_formatter.oneByteOp(OP_OR_EvGv, src, addr); + } +#endif + + void subl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_SUB_EvGv, src, dst); + } + + void subl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_SUB_GvEv, dst, base, offset); + } + + void subl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_SUB_EvGv, src, base, offset); + } + + void subl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_SUB, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_SUB, dst); + m_formatter.immediate32(imm); + } + } + + void subl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_SUB, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_SUB, base, offset); + m_formatter.immediate32(imm); + } + } + +#if CPU(X86_64) + void subq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_SUB_EvGv, src, dst); + } + + void subq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_SUB, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_SUB, dst); + m_formatter.immediate32(imm); + } + } +#else + void subl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_SUB, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_SUB, addr); + m_formatter.immediate32(imm); + } + } +#endif + + void xorl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_XOR_EvGv, src, dst); + } + + void xorl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_XOR_GvEv, dst, base, offset); + } + + void xorl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_XOR_EvGv, src, base, offset); + } + + void xorl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_XOR, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_XOR, base, offset); + m_formatter.immediate32(imm); + } + } + + void xorl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_XOR, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_XOR, dst); + m_formatter.immediate32(imm); + } + } + +#if CPU(X86_64) + void xorq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_XOR_EvGv, src, dst); + } + + void xorq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_XOR, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_XOR, dst); + m_formatter.immediate32(imm); + } + } + + void xorq_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_XOR_EvGv, src, base, offset); + } + + void rorq_i8r(int imm, RegisterID dst) + { + if (imm == 1) + m_formatter.oneByteOp64(OP_GROUP2_Ev1, GROUP2_OP_ROR, dst); + else { + m_formatter.oneByteOp64(OP_GROUP2_EvIb, GROUP2_OP_ROR, dst); + m_formatter.immediate8(imm); + } + } + +#endif + + void sarl_i8r(int imm, RegisterID dst) + { + if (imm == 1) + m_formatter.oneByteOp(OP_GROUP2_Ev1, GROUP2_OP_SAR, dst); + else { + m_formatter.oneByteOp(OP_GROUP2_EvIb, GROUP2_OP_SAR, dst); + m_formatter.immediate8(imm); + } + } + + void sarl_CLr(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP2_EvCL, GROUP2_OP_SAR, dst); + } + + void shrl_i8r(int imm, RegisterID dst) + { + if (imm == 1) + m_formatter.oneByteOp(OP_GROUP2_Ev1, GROUP2_OP_SHR, dst); + else { + m_formatter.oneByteOp(OP_GROUP2_EvIb, GROUP2_OP_SHR, dst); + m_formatter.immediate8(imm); + } + } + + void shrl_CLr(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP2_EvCL, GROUP2_OP_SHR, dst); + } + + void shll_i8r(int imm, RegisterID dst) + { + if (imm == 1) + m_formatter.oneByteOp(OP_GROUP2_Ev1, GROUP2_OP_SHL, dst); + else { + m_formatter.oneByteOp(OP_GROUP2_EvIb, GROUP2_OP_SHL, dst); + m_formatter.immediate8(imm); + } + } + + void shll_CLr(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP2_EvCL, GROUP2_OP_SHL, dst); + } + +#if CPU(X86_64) + void sarq_CLr(RegisterID dst) + { + m_formatter.oneByteOp64(OP_GROUP2_EvCL, GROUP2_OP_SAR, dst); + } + + void sarq_i8r(int imm, RegisterID dst) + { + if (imm == 1) + m_formatter.oneByteOp64(OP_GROUP2_Ev1, GROUP2_OP_SAR, dst); + else { + m_formatter.oneByteOp64(OP_GROUP2_EvIb, GROUP2_OP_SAR, dst); + m_formatter.immediate8(imm); + } + } +#endif + + void imull_rr(RegisterID src, RegisterID dst) + { + m_formatter.twoByteOp(OP2_IMUL_GvEv, dst, src); + } + + void imull_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.twoByteOp(OP2_IMUL_GvEv, dst, base, offset); + } + + void imull_i32r(RegisterID src, int32_t value, RegisterID dst) + { + m_formatter.oneByteOp(OP_IMUL_GvEvIz, dst, src); + m_formatter.immediate32(value); + } + + void idivl_r(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP3_Ev, GROUP3_OP_IDIV, dst); + } + + // Comparisons: + + void cmpl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_CMP_EvGv, src, dst); + } + + void cmpl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_CMP_EvGv, src, base, offset); + } + + void cmpl_mr(int offset, RegisterID base, RegisterID src) + { + m_formatter.oneByteOp(OP_CMP_GvEv, src, base, offset); + } + + void cmpl_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, dst); + m_formatter.immediate32(imm); + } + } + + void cmpl_ir_force32(int imm, RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, dst); + m_formatter.immediate32(imm); + } + + void cmpl_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, offset); + m_formatter.immediate32(imm); + } + } + + void cmpb_im(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP1_EbIb, GROUP1_OP_CMP, base, offset); + m_formatter.immediate8(imm); + } + + void cmpb_im(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp(OP_GROUP1_EbIb, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate8(imm); + } + +#if CPU(X86) + void cmpb_im(int imm, const void* addr) + { + m_formatter.oneByteOp(OP_GROUP1_EbIb, GROUP1_OP_CMP, addr); + m_formatter.immediate8(imm); + } +#endif + + void cmpl_im(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate32(imm); + } + } + + void cmpl_im_force32(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, offset); + m_formatter.immediate32(imm); + } + +#if CPU(X86_64) + void cmpq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_CMP_EvGv, src, dst); + } + + void cmpq_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_CMP_EvGv, src, base, offset); + } + + void cmpq_mr(int offset, RegisterID base, RegisterID src) + { + m_formatter.oneByteOp64(OP_CMP_GvEv, src, base, offset); + } + + void cmpq_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_CMP, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_CMP, dst); + m_formatter.immediate32(imm); + } + } + + void cmpq_im(int imm, int offset, RegisterID base) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_CMP, base, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, offset); + m_formatter.immediate32(imm); + } + } + + void cmpq_im(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp64(OP_GROUP1_EvIb, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp64(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate32(imm); + } + } +#else + void cmpl_rm(RegisterID reg, const void* addr) + { + m_formatter.oneByteOp(OP_CMP_EvGv, reg, addr); + } + + void cmpl_im(int imm, const void* addr) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, addr); + m_formatter.immediate8(imm); + } else { + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, addr); + m_formatter.immediate32(imm); + } + } +#endif + + void cmpw_ir(int imm, RegisterID dst) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, dst); + m_formatter.immediate8(imm); + } else { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, dst); + m_formatter.immediate16(imm); + } + } + + void cmpw_rm(RegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_CMP_EvGv, src, base, index, scale, offset); + } + + void cmpw_im(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + if (CAN_SIGN_EXTEND_8_32(imm)) { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_GROUP1_EvIb, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate8(imm); + } else { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_GROUP1_EvIz, GROUP1_OP_CMP, base, index, scale, offset); + m_formatter.immediate16(imm); + } + } + + void testl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_TEST_EvGv, src, dst); + } + + void testl_i32r(int imm, RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP3_EvIz, GROUP3_OP_TEST, dst); + m_formatter.immediate32(imm); + } + + void testl_i32m(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP3_EvIz, GROUP3_OP_TEST, base, offset); + m_formatter.immediate32(imm); + } + + void testb_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp8(OP_TEST_EbGb, src, dst); + } + + void testb_im(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP3_EbIb, GROUP3_OP_TEST, base, offset); + m_formatter.immediate8(imm); + } + + void testb_im(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp(OP_GROUP3_EbIb, GROUP3_OP_TEST, base, index, scale, offset); + m_formatter.immediate8(imm); + } + +#if CPU(X86) + void testb_im(int imm, const void* addr) + { + m_formatter.oneByteOp(OP_GROUP3_EbIb, GROUP3_OP_TEST, addr); + m_formatter.immediate8(imm); + } +#endif + + void testl_i32m(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp(OP_GROUP3_EvIz, GROUP3_OP_TEST, base, index, scale, offset); + m_formatter.immediate32(imm); + } + +#if CPU(X86_64) + void testq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_TEST_EvGv, src, dst); + } + + void testq_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_TEST_EvGv, src, base, offset); + } + + void testq_i32r(int imm, RegisterID dst) + { + m_formatter.oneByteOp64(OP_GROUP3_EvIz, GROUP3_OP_TEST, dst); + m_formatter.immediate32(imm); + } + + void testq_i32m(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_GROUP3_EvIz, GROUP3_OP_TEST, base, offset); + m_formatter.immediate32(imm); + } + + void testq_i32m(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp64(OP_GROUP3_EvIz, GROUP3_OP_TEST, base, index, scale, offset); + m_formatter.immediate32(imm); + } +#endif + + void testw_rr(RegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp(OP_TEST_EvGv, src, dst); + } + + void testb_i8r(int imm, RegisterID dst) + { + m_formatter.oneByteOp8(OP_GROUP3_EbIb, GROUP3_OP_TEST, dst); + m_formatter.immediate8(imm); + } + + void setCC_r(Condition cond, RegisterID dst) + { + m_formatter.twoByteOp8(setccOpcode(cond), (GroupOpcodeID)0, dst); + } + + void sete_r(RegisterID dst) + { + m_formatter.twoByteOp8(setccOpcode(ConditionE), (GroupOpcodeID)0, dst); + } + + void setz_r(RegisterID dst) + { + sete_r(dst); + } + + void setne_r(RegisterID dst) + { + m_formatter.twoByteOp8(setccOpcode(ConditionNE), (GroupOpcodeID)0, dst); + } + + void setnz_r(RegisterID dst) + { + setne_r(dst); + } + + // Various move ops: + + void cdq() + { + m_formatter.oneByteOp(OP_CDQ); + } + + void fstpl(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_ESCAPE_DD, ESCAPE_DD_FSTP_doubleReal, base, offset); + } + + void xchgl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_XCHG_EvGv, src, dst); + } + +#if CPU(X86_64) + void xchgq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_XCHG_EvGv, src, dst); + } +#endif + + void movl_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp(OP_MOV_EvGv, src, dst); + } + + void movl_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_MOV_EvGv, src, base, offset); + } + + void movl_rm_disp32(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp_disp32(OP_MOV_EvGv, src, base, offset); + } + + void movl_rm(RegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp(OP_MOV_EvGv, src, base, index, scale, offset); + } + + void movl_mEAX(const void* addr) + { + m_formatter.oneByteOp(OP_MOV_EAXOv); +#if CPU(X86_64) + m_formatter.immediate64(reinterpret_cast<int64_t>(addr)); +#else + m_formatter.immediate32(reinterpret_cast<int>(addr)); +#endif + } + + void movl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_MOV_GvEv, dst, base, offset); + } + + void movl_mr_disp32(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp_disp32(OP_MOV_GvEv, dst, base, offset); + } + + void movl_mr_disp8(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp_disp8(OP_MOV_GvEv, dst, base, offset); + } + + void movl_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.oneByteOp(OP_MOV_GvEv, dst, base, index, scale, offset); + } + + void movl_i32r(int imm, RegisterID dst) + { + m_formatter.oneByteOp(OP_MOV_EAXIv, dst); + m_formatter.immediate32(imm); + } + + void movl_i32m(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP11_EvIz, GROUP11_MOV, base, offset); + m_formatter.immediate32(imm); + } + + void movl_i32m(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp(OP_GROUP11_EvIz, GROUP11_MOV, base, index, scale, offset); + m_formatter.immediate32(imm); + } + +#if !CPU(X86_64) + void movb_i8m(int imm, const void* addr) + { + ASSERT(-128 <= imm && imm < 128); + m_formatter.oneByteOp(OP_GROUP11_EvIb, GROUP11_MOV, addr); + m_formatter.immediate8(imm); + } +#endif + + void movb_i8m(int imm, int offset, RegisterID base) + { + ASSERT(-128 <= imm && imm < 128); + m_formatter.oneByteOp(OP_GROUP11_EvIb, GROUP11_MOV, base, offset); + m_formatter.immediate8(imm); + } + + void movb_i8m(int imm, int offset, RegisterID base, RegisterID index, int scale) + { + ASSERT(-128 <= imm && imm < 128); + m_formatter.oneByteOp(OP_GROUP11_EvIb, GROUP11_MOV, base, index, scale, offset); + m_formatter.immediate8(imm); + } + + void movb_rm(RegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp8(OP_MOV_EbGb, src, base, index, scale, offset); + } + + void movw_rm(RegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.prefix(PRE_OPERAND_SIZE); + m_formatter.oneByteOp8(OP_MOV_EvGv, src, base, index, scale, offset); + } + + void movl_EAXm(const void* addr) + { + m_formatter.oneByteOp(OP_MOV_OvEAX); +#if CPU(X86_64) + m_formatter.immediate64(reinterpret_cast<int64_t>(addr)); +#else + m_formatter.immediate32(reinterpret_cast<int>(addr)); +#endif + } + +#if CPU(X86_64) + void movq_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_MOV_EvGv, src, dst); + } + + void movq_rm(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_MOV_EvGv, src, base, offset); + } + + void movq_rm_disp32(RegisterID src, int offset, RegisterID base) + { + m_formatter.oneByteOp64_disp32(OP_MOV_EvGv, src, base, offset); + } + + void movq_rm(RegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.oneByteOp64(OP_MOV_EvGv, src, base, index, scale, offset); + } + + void movq_mEAX(const void* addr) + { + m_formatter.oneByteOp64(OP_MOV_EAXOv); + m_formatter.immediate64(reinterpret_cast<int64_t>(addr)); + } + + void movq_EAXm(const void* addr) + { + m_formatter.oneByteOp64(OP_MOV_OvEAX); + m_formatter.immediate64(reinterpret_cast<int64_t>(addr)); + } + + void movq_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp64(OP_MOV_GvEv, dst, base, offset); + } + + void movq_mr_disp32(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp64_disp32(OP_MOV_GvEv, dst, base, offset); + } + + void movq_mr_disp8(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp64_disp8(OP_MOV_GvEv, dst, base, offset); + } + + void movq_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.oneByteOp64(OP_MOV_GvEv, dst, base, index, scale, offset); + } + + void movq_i32m(int imm, int offset, RegisterID base) + { + m_formatter.oneByteOp64(OP_GROUP11_EvIz, GROUP11_MOV, base, offset); + m_formatter.immediate32(imm); + } + + void movq_i64r(int64_t imm, RegisterID dst) + { + m_formatter.oneByteOp64(OP_MOV_EAXIv, dst); + m_formatter.immediate64(imm); + } + + void movsxd_rr(RegisterID src, RegisterID dst) + { + m_formatter.oneByteOp64(OP_MOVSXD_GvEv, dst, src); + } + + +#else + void movl_rm(RegisterID src, const void* addr) + { + if (src == X86Registers::eax) + movl_EAXm(addr); + else + m_formatter.oneByteOp(OP_MOV_EvGv, src, addr); + } + + void movl_mr(const void* addr, RegisterID dst) + { + if (dst == X86Registers::eax) + movl_mEAX(addr); + else + m_formatter.oneByteOp(OP_MOV_GvEv, dst, addr); + } + + void movl_i32m(int imm, const void* addr) + { + m_formatter.oneByteOp(OP_GROUP11_EvIz, GROUP11_MOV, addr); + m_formatter.immediate32(imm); + } +#endif + + void movzwl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVZX_GvEw, dst, base, offset); + } + + void movzwl_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVZX_GvEw, dst, base, index, scale, offset); + } + + void movswl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVSX_GvEw, dst, base, offset); + } + + void movswl_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVSX_GvEw, dst, base, index, scale, offset); + } + + void movzbl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVZX_GvEb, dst, base, offset); + } + + void movzbl_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVZX_GvEb, dst, base, index, scale, offset); + } + + void movsbl_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVSX_GvEb, dst, base, offset); + } + + void movsbl_mr(int offset, RegisterID base, RegisterID index, int scale, RegisterID dst) + { + m_formatter.twoByteOp(OP2_MOVSX_GvEb, dst, base, index, scale, offset); + } + + void movzbl_rr(RegisterID src, RegisterID dst) + { + // In 64-bit, this may cause an unnecessary REX to be planted (if the dst register + // is in the range ESP-EDI, and the src would not have required a REX). Unneeded + // REX prefixes are defined to be silently ignored by the processor. + m_formatter.twoByteOp8(OP2_MOVZX_GvEb, dst, src); + } + + void leal_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp(OP_LEA, dst, base, offset); + } +#if CPU(X86_64) + void leaq_mr(int offset, RegisterID base, RegisterID dst) + { + m_formatter.oneByteOp64(OP_LEA, dst, base, offset); + } +#endif + + // Flow control: + + AssemblerLabel call() + { + m_formatter.oneByteOp(OP_CALL_rel32); + return m_formatter.immediateRel32(); + } + + AssemblerLabel call(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_CALLN, dst); + return m_formatter.label(); + } + + void call_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_CALLN, base, offset); + } + + AssemblerLabel jmp() + { + m_formatter.oneByteOp(OP_JMP_rel32); + return m_formatter.immediateRel32(); + } + + // Return a AssemblerLabel so we have a label to the jump, so we can use this + // To make a tail recursive call on x86-64. The MacroAssembler + // really shouldn't wrap this as a Jump, since it can't be linked. :-/ + AssemblerLabel jmp_r(RegisterID dst) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_JMPN, dst); + return m_formatter.label(); + } + + void jmp_m(int offset, RegisterID base) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_JMPN, base, offset); + } + +#if !CPU(X86_64) + void jmp_m(const void* address) + { + m_formatter.oneByteOp(OP_GROUP5_Ev, GROUP5_OP_JMPN, address); + } +#endif + + AssemblerLabel jne() + { + m_formatter.twoByteOp(jccRel32(ConditionNE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jnz() + { + return jne(); + } + + AssemblerLabel je() + { + m_formatter.twoByteOp(jccRel32(ConditionE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jz() + { + return je(); + } + + AssemblerLabel jl() + { + m_formatter.twoByteOp(jccRel32(ConditionL)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jb() + { + m_formatter.twoByteOp(jccRel32(ConditionB)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jle() + { + m_formatter.twoByteOp(jccRel32(ConditionLE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jbe() + { + m_formatter.twoByteOp(jccRel32(ConditionBE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jge() + { + m_formatter.twoByteOp(jccRel32(ConditionGE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jg() + { + m_formatter.twoByteOp(jccRel32(ConditionG)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel ja() + { + m_formatter.twoByteOp(jccRel32(ConditionA)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jae() + { + m_formatter.twoByteOp(jccRel32(ConditionAE)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jo() + { + m_formatter.twoByteOp(jccRel32(ConditionO)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jnp() + { + m_formatter.twoByteOp(jccRel32(ConditionNP)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jp() + { + m_formatter.twoByteOp(jccRel32(ConditionP)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel js() + { + m_formatter.twoByteOp(jccRel32(ConditionS)); + return m_formatter.immediateRel32(); + } + + AssemblerLabel jCC(Condition cond) + { + m_formatter.twoByteOp(jccRel32(cond)); + return m_formatter.immediateRel32(); + } + + // SSE operations: + + void addsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_ADDSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void addsd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_ADDSD_VsdWsd, (RegisterID)dst, base, offset); + } + +#if !CPU(X86_64) + void addsd_mr(const void* address, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_ADDSD_VsdWsd, (RegisterID)dst, address); + } +#endif + + void cvtsi2sd_rr(RegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_CVTSI2SD_VsdEd, (RegisterID)dst, src); + } + + void cvtsi2sd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_CVTSI2SD_VsdEd, (RegisterID)dst, base, offset); + } + +#if !CPU(X86_64) + void cvtsi2sd_mr(const void* address, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_CVTSI2SD_VsdEd, (RegisterID)dst, address); + } +#endif + + void cvttsd2si_rr(XMMRegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_CVTTSD2SI_GdWsd, dst, (RegisterID)src); + } + + void cvtsd2ss_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_CVTSD2SS_VsdWsd, dst, (RegisterID)src); + } + + void cvtss2sd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F3); + m_formatter.twoByteOp(OP2_CVTSS2SD_VsdWsd, dst, (RegisterID)src); + } + +#if CPU(X86_64) + void cvttsd2siq_rr(XMMRegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp64(OP2_CVTTSD2SI_GdWsd, dst, (RegisterID)src); + } +#endif + + void movd_rr(XMMRegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_MOVD_EdVd, (RegisterID)src, dst); + } + + void movd_rr(RegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_MOVD_VdEd, (RegisterID)dst, src); + } + +#if CPU(X86_64) + void movq_rr(XMMRegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp64(OP2_MOVD_EdVd, (RegisterID)src, dst); + } + + void movq_rr(RegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp64(OP2_MOVD_VdEd, (RegisterID)dst, src); + } +#endif + + void movsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void movsd_rm(XMMRegisterID src, int offset, RegisterID base) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_WsdVsd, (RegisterID)src, base, offset); + } + + void movsd_rm(XMMRegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_WsdVsd, (RegisterID)src, base, index, scale, offset); + } + + void movss_rm(XMMRegisterID src, int offset, RegisterID base, RegisterID index, int scale) + { + m_formatter.prefix(PRE_SSE_F3); + m_formatter.twoByteOp(OP2_MOVSD_WsdVsd, (RegisterID)src, base, index, scale, offset); + } + + void movsd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, (RegisterID)dst, base, offset); + } + + void movsd_mr(int offset, RegisterID base, RegisterID index, int scale, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, dst, base, index, scale, offset); + } + + void movss_mr(int offset, RegisterID base, RegisterID index, int scale, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F3); + m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, dst, base, index, scale, offset); + } + +#if !CPU(X86_64) + void movsd_mr(const void* address, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, (RegisterID)dst, address); + } + void movsd_rm(XMMRegisterID src, const void* address) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MOVSD_WsdVsd, (RegisterID)src, address); + } +#endif + + void mulsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MULSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void mulsd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_MULSD_VsdWsd, (RegisterID)dst, base, offset); + } + + void pextrw_irr(int whichWord, XMMRegisterID src, RegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_PEXTRW_GdUdIb, (RegisterID)dst, (RegisterID)src); + m_formatter.immediate8(whichWord); + } + + void psllq_i8r(int imm, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp8(OP2_PSLLQ_UdqIb, GROUP14_OP_PSLLQ, (RegisterID)dst); + m_formatter.immediate8(imm); + } + + void psrlq_i8r(int imm, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp8(OP2_PSRLQ_UdqIb, GROUP14_OP_PSRLQ, (RegisterID)dst); + m_formatter.immediate8(imm); + } + + void por_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_POR_VdqWdq, (RegisterID)dst, (RegisterID)src); + } + + void subsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_SUBSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void subsd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_SUBSD_VsdWsd, (RegisterID)dst, base, offset); + } + + void ucomisd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_UCOMISD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void ucomisd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_UCOMISD_VsdWsd, (RegisterID)dst, base, offset); + } + + void divsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_DIVSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + void divsd_mr(int offset, RegisterID base, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_DIVSD_VsdWsd, (RegisterID)dst, base, offset); + } + + void xorpd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_XORPD_VpdWpd, (RegisterID)dst, (RegisterID)src); + } + + void andnpd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_66); + m_formatter.twoByteOp(OP2_ANDNPD_VpdWpd, (RegisterID)dst, (RegisterID)src); + } + + void sqrtsd_rr(XMMRegisterID src, XMMRegisterID dst) + { + m_formatter.prefix(PRE_SSE_F2); + m_formatter.twoByteOp(OP2_SQRTSD_VsdWsd, (RegisterID)dst, (RegisterID)src); + } + + // Misc instructions: + + void int3() + { + m_formatter.oneByteOp(OP_INT3); + } + + void ret() + { + m_formatter.oneByteOp(OP_RET); + } + + void predictNotTaken() + { + m_formatter.prefix(PRE_PREDICT_BRANCH_NOT_TAKEN); + } + + // Assembler admin methods: + + size_t codeSize() const + { + return m_formatter.codeSize(); + } + + AssemblerLabel labelForWatchpoint() + { + AssemblerLabel result = m_formatter.label(); + if (static_cast<int>(result.m_offset) != m_indexOfLastWatchpoint) + result = label(); + m_indexOfLastWatchpoint = result.m_offset; + m_indexOfTailOfLastWatchpoint = result.m_offset + maxJumpReplacementSize(); + return result; + } + + AssemblerLabel labelIgnoringWatchpoints() + { + return m_formatter.label(); + } + + AssemblerLabel label() + { + AssemblerLabel result = m_formatter.label(); + while (UNLIKELY(static_cast<int>(result.m_offset) < m_indexOfTailOfLastWatchpoint)) { + nop(); + result = m_formatter.label(); + } + return result; + } + + AssemblerLabel align(int alignment) + { + while (!m_formatter.isAligned(alignment)) + m_formatter.oneByteOp(OP_HLT); + + return label(); + } + + // Linking & patching: + // + // 'link' and 'patch' methods are for use on unprotected code - such as the code + // within the AssemblerBuffer, and code being patched by the patch buffer. Once + // code has been finalized it is (platform support permitting) within a non- + // writable region of memory; to modify the code in an execute-only execuable + // pool the 'repatch' and 'relink' methods should be used. + + void linkJump(AssemblerLabel from, AssemblerLabel to) + { + ASSERT(from.isSet()); + ASSERT(to.isSet()); + + char* code = reinterpret_cast<char*>(m_formatter.data()); + ASSERT(!reinterpret_cast<int32_t*>(code + from.m_offset)[-1]); + setRel32(code + from.m_offset, code + to.m_offset); + } + + static void linkJump(void* code, AssemblerLabel from, void* to) + { + ASSERT(from.isSet()); + + setRel32(reinterpret_cast<char*>(code) + from.m_offset, to); + } + + static void linkCall(void* code, AssemblerLabel from, void* to) + { + ASSERT(from.isSet()); + + setRel32(reinterpret_cast<char*>(code) + from.m_offset, to); + } + + static void linkPointer(void* code, AssemblerLabel where, void* value) + { + ASSERT(where.isSet()); + + setPointer(reinterpret_cast<char*>(code) + where.m_offset, value); + } + + static void relinkJump(void* from, void* to) + { + setRel32(from, to); + } + + static void relinkCall(void* from, void* to) + { + setRel32(from, to); + } + + static void repatchCompact(void* where, int32_t value) + { + ASSERT(value >= std::numeric_limits<int8_t>::min()); + ASSERT(value <= std::numeric_limits<int8_t>::max()); + setInt8(where, value); + } + + static void repatchInt32(void* where, int32_t value) + { + setInt32(where, value); + } + + static void repatchPointer(void* where, void* value) + { + setPointer(where, value); + } + + static void* readPointer(void* where) + { + return reinterpret_cast<void**>(where)[-1]; + } + + static void replaceWithJump(void* instructionStart, void* to) + { + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); + uint8_t* dstPtr = reinterpret_cast<uint8_t*>(to); + intptr_t distance = (intptr_t)(dstPtr - (ptr + 5)); + ptr[0] = static_cast<uint8_t>(OP_JMP_rel32); + *reinterpret_cast<int32_t*>(ptr + 1) = static_cast<int32_t>(distance); + } + + static ptrdiff_t maxJumpReplacementSize() + { + return 5; + } + +#if CPU(X86_64) + static void revertJumpTo_movq_i64r(void* instructionStart, int64_t imm, RegisterID dst) + { + const int rexBytes = 1; + const int opcodeBytes = 1; + ASSERT(rexBytes + opcodeBytes <= maxJumpReplacementSize()); + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); + ptr[0] = PRE_REX | (1 << 3) | (dst >> 3); + ptr[1] = OP_MOV_EAXIv | (dst & 7); + + union { + uint64_t asWord; + uint8_t asBytes[8]; + } u; + u.asWord = imm; + for (unsigned i = rexBytes + opcodeBytes; i < static_cast<unsigned>(maxJumpReplacementSize()); ++i) + ptr[i] = u.asBytes[i - rexBytes - opcodeBytes]; + } +#endif + + static void revertJumpTo_cmpl_ir_force32(void* instructionStart, int32_t imm, RegisterID dst) + { + const int opcodeBytes = 1; + const int modRMBytes = 1; + ASSERT(opcodeBytes + modRMBytes <= maxJumpReplacementSize()); + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); + ptr[0] = OP_GROUP1_EvIz; + ptr[1] = (X86InstructionFormatter::ModRmRegister << 6) | (GROUP1_OP_CMP << 3) | dst; + union { + uint32_t asWord; + uint8_t asBytes[4]; + } u; + u.asWord = imm; + for (unsigned i = opcodeBytes + modRMBytes; i < static_cast<unsigned>(maxJumpReplacementSize()); ++i) + ptr[i] = u.asBytes[i - opcodeBytes - modRMBytes]; + } + + static void revertJumpTo_cmpl_im_force32(void* instructionStart, int32_t imm, int offset, RegisterID dst) + { + ASSERT_UNUSED(offset, !offset); + const int opcodeBytes = 1; + const int modRMBytes = 1; + ASSERT(opcodeBytes + modRMBytes <= maxJumpReplacementSize()); + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); + ptr[0] = OP_GROUP1_EvIz; + ptr[1] = (X86InstructionFormatter::ModRmMemoryNoDisp << 6) | (GROUP1_OP_CMP << 3) | dst; + union { + uint32_t asWord; + uint8_t asBytes[4]; + } u; + u.asWord = imm; + for (unsigned i = opcodeBytes + modRMBytes; i < static_cast<unsigned>(maxJumpReplacementSize()); ++i) + ptr[i] = u.asBytes[i - opcodeBytes - modRMBytes]; + } + + static void replaceWithLoad(void* instructionStart) + { + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); +#if CPU(X86_64) + if ((*ptr & ~15) == PRE_REX) + ptr++; +#endif + switch (*ptr) { + case OP_MOV_GvEv: + break; + case OP_LEA: + *ptr = OP_MOV_GvEv; + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + static void replaceWithAddressComputation(void* instructionStart) + { + uint8_t* ptr = reinterpret_cast<uint8_t*>(instructionStart); +#if CPU(X86_64) + if ((*ptr & ~15) == PRE_REX) + ptr++; +#endif + switch (*ptr) { + case OP_MOV_GvEv: + *ptr = OP_LEA; + break; + case OP_LEA: + break; + default: + RELEASE_ASSERT_NOT_REACHED(); + } + } + + static unsigned getCallReturnOffset(AssemblerLabel call) + { + ASSERT(call.isSet()); + return call.m_offset; + } + + static void* getRelocatedAddress(void* code, AssemblerLabel label) + { + ASSERT(label.isSet()); + return reinterpret_cast<void*>(reinterpret_cast<ptrdiff_t>(code) + label.m_offset); + } + + static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b) + { + return b.m_offset - a.m_offset; + } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + return m_formatter.executableCopy(globalData, ownerUID, effort); + } + + unsigned debugOffset() { return m_formatter.debugOffset(); } + + void nop() + { + m_formatter.oneByteOp(OP_NOP); + } + + // This is a no-op on x86 + ALWAYS_INLINE static void cacheFlush(void*, size_t) { } + +private: + + static void setPointer(void* where, void* value) + { + reinterpret_cast<void**>(where)[-1] = value; + } + + static void setInt32(void* where, int32_t value) + { + reinterpret_cast<int32_t*>(where)[-1] = value; + } + + static void setInt8(void* where, int8_t value) + { + reinterpret_cast<int8_t*>(where)[-1] = value; + } + + static void setRel32(void* from, void* to) + { + intptr_t offset = reinterpret_cast<intptr_t>(to) - reinterpret_cast<intptr_t>(from); + ASSERT(offset == static_cast<int32_t>(offset)); + + setInt32(from, offset); + } + + class X86InstructionFormatter { + + static const int maxInstructionSize = 16; + + public: + + enum ModRmMode { + ModRmMemoryNoDisp, + ModRmMemoryDisp8, + ModRmMemoryDisp32, + ModRmRegister, + }; + + // Legacy prefix bytes: + // + // These are emmitted prior to the instruction. + + void prefix(OneByteOpcodeID pre) + { + m_buffer.putByte(pre); + } + + // Word-sized operands / no operand instruction formatters. + // + // In addition to the opcode, the following operand permutations are supported: + // * None - instruction takes no operands. + // * One register - the low three bits of the RegisterID are added into the opcode. + // * Two registers - encode a register form ModRm (for all ModRm formats, the reg field is passed first, and a GroupOpcodeID may be passed in its place). + // * Three argument ModRM - a register, and a register and an offset describing a memory operand. + // * Five argument ModRM - a register, and a base register, an index, scale, and offset describing a memory operand. + // + // For 32-bit x86 targets, the address operand may also be provided as a void*. + // On 64-bit targets REX prefixes will be planted as necessary, where high numbered registers are used. + // + // The twoByteOp methods plant two-byte Intel instructions sequences (first opcode byte 0x0F). + + void oneByteOp(OneByteOpcodeID opcode) + { + m_buffer.ensureSpace(maxInstructionSize); + m_buffer.putByteUnchecked(opcode); + } + + void oneByteOp(OneByteOpcodeID opcode, RegisterID reg) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(0, 0, reg); + m_buffer.putByteUnchecked(opcode + (reg & 7)); + } + + void oneByteOp(OneByteOpcodeID opcode, int reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, rm); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } + + void oneByteOp(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, offset); + } + + void oneByteOp_disp32(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM_disp32(reg, base, offset); + } + + void oneByteOp_disp8(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM_disp8(reg, base, offset); + } + + void oneByteOp(OneByteOpcodeID opcode, int reg, RegisterID base, RegisterID index, int scale, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, index, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, index, scale, offset); + } + +#if !CPU(X86_64) + void oneByteOp(OneByteOpcodeID opcode, int reg, const void* address) + { + m_buffer.ensureSpace(maxInstructionSize); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, address); + } +#endif + + void twoByteOp(TwoByteOpcodeID opcode) + { + m_buffer.ensureSpace(maxInstructionSize); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + } + + void twoByteOp(TwoByteOpcodeID opcode, int reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, rm); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } + + void twoByteOp(TwoByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, 0, base); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, offset); + } + + void twoByteOp(TwoByteOpcodeID opcode, int reg, RegisterID base, RegisterID index, int scale, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIfNeeded(reg, index, base); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, index, scale, offset); + } + +#if !CPU(X86_64) + void twoByteOp(TwoByteOpcodeID opcode, int reg, const void* address) + { + m_buffer.ensureSpace(maxInstructionSize); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, address); + } +#endif + +#if CPU(X86_64) + // Quad-word-sized operands: + // + // Used to format 64-bit operantions, planting a REX.w prefix. + // When planting d64 or f64 instructions, not requiring a REX.w prefix, + // the normal (non-'64'-postfixed) formatters should be used. + + void oneByteOp64(OneByteOpcodeID opcode) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(0, 0, 0); + m_buffer.putByteUnchecked(opcode); + } + + void oneByteOp64(OneByteOpcodeID opcode, RegisterID reg) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(0, 0, reg); + m_buffer.putByteUnchecked(opcode + (reg & 7)); + } + + void oneByteOp64(OneByteOpcodeID opcode, int reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, 0, rm); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } + + void oneByteOp64(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, offset); + } + + void oneByteOp64_disp32(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM_disp32(reg, base, offset); + } + + void oneByteOp64_disp8(OneByteOpcodeID opcode, int reg, RegisterID base, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, 0, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM_disp8(reg, base, offset); + } + + void oneByteOp64(OneByteOpcodeID opcode, int reg, RegisterID base, RegisterID index, int scale, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, index, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, index, scale, offset); + } + + void twoByteOp64(TwoByteOpcodeID opcode, int reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexW(reg, 0, rm); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } +#endif + + // Byte-operands: + // + // These methods format byte operations. Byte operations differ from the normal + // formatters in the circumstances under which they will decide to emit REX prefixes. + // These should be used where any register operand signifies a byte register. + // + // The disctinction is due to the handling of register numbers in the range 4..7 on + // x86-64. These register numbers may either represent the second byte of the first + // four registers (ah..bh) or the first byte of the second four registers (spl..dil). + // + // Since ah..bh cannot be used in all permutations of operands (specifically cannot + // be accessed where a REX prefix is present), these are likely best treated as + // deprecated. In order to ensure the correct registers spl..dil are selected a + // REX prefix will be emitted for any byte register operand in the range 4..15. + // + // These formatters may be used in instructions where a mix of operand sizes, in which + // case an unnecessary REX will be emitted, for example: + // movzbl %al, %edi + // In this case a REX will be planted since edi is 7 (and were this a byte operand + // a REX would be required to specify dil instead of bh). Unneeded REX prefixes will + // be silently ignored by the processor. + // + // Address operands should still be checked using regRequiresRex(), while byteRegRequiresRex() + // is provided to check byte register operands. + + void oneByteOp8(OneByteOpcodeID opcode, GroupOpcodeID groupOp, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIf(byteRegRequiresRex(rm), 0, 0, rm); + m_buffer.putByteUnchecked(opcode); + registerModRM(groupOp, rm); + } + + void oneByteOp8(OneByteOpcodeID opcode, int reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIf(byteRegRequiresRex(reg) || byteRegRequiresRex(rm), reg, 0, rm); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } + + void oneByteOp8(OneByteOpcodeID opcode, int reg, RegisterID base, RegisterID index, int scale, int offset) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIf(byteRegRequiresRex(reg) || regRequiresRex(index) || regRequiresRex(base), reg, index, base); + m_buffer.putByteUnchecked(opcode); + memoryModRM(reg, base, index, scale, offset); + } + + void twoByteOp8(TwoByteOpcodeID opcode, RegisterID reg, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIf(byteRegRequiresRex(reg)|byteRegRequiresRex(rm), reg, 0, rm); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + registerModRM(reg, rm); + } + + void twoByteOp8(TwoByteOpcodeID opcode, GroupOpcodeID groupOp, RegisterID rm) + { + m_buffer.ensureSpace(maxInstructionSize); + emitRexIf(byteRegRequiresRex(rm), 0, 0, rm); + m_buffer.putByteUnchecked(OP_2BYTE_ESCAPE); + m_buffer.putByteUnchecked(opcode); + registerModRM(groupOp, rm); + } + + // Immediates: + // + // An immedaite should be appended where appropriate after an op has been emitted. + // The writes are unchecked since the opcode formatters above will have ensured space. + + void immediate8(int imm) + { + m_buffer.putByteUnchecked(imm); + } + + void immediate16(int imm) + { + m_buffer.putShortUnchecked(imm); + } + + void immediate32(int imm) + { + m_buffer.putIntUnchecked(imm); + } + + void immediate64(int64_t imm) + { + m_buffer.putInt64Unchecked(imm); + } + + AssemblerLabel immediateRel32() + { + m_buffer.putIntUnchecked(0); + return label(); + } + + // Administrative methods: + + size_t codeSize() const { return m_buffer.codeSize(); } + AssemblerLabel label() const { return m_buffer.label(); } + bool isAligned(int alignment) const { return m_buffer.isAligned(alignment); } + void* data() const { return m_buffer.data(); } + + PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData& globalData, void* ownerUID, JITCompilationEffort effort) + { + return m_buffer.executableCopy(globalData, ownerUID, effort); + } + + unsigned debugOffset() { return m_buffer.debugOffset(); } + + private: + + // Internals; ModRm and REX formatters. + + static const RegisterID noBase = X86Registers::ebp; + static const RegisterID hasSib = X86Registers::esp; + static const RegisterID noIndex = X86Registers::esp; +#if CPU(X86_64) + static const RegisterID noBase2 = X86Registers::r13; + static const RegisterID hasSib2 = X86Registers::r12; + + // Registers r8 & above require a REX prefixe. + inline bool regRequiresRex(int reg) + { + return (reg >= X86Registers::r8); + } + + // Byte operand register spl & above require a REX prefix (to prevent the 'H' registers be accessed). + inline bool byteRegRequiresRex(int reg) + { + return (reg >= X86Registers::esp); + } + + // Format a REX prefix byte. + inline void emitRex(bool w, int r, int x, int b) + { + ASSERT(r >= 0); + ASSERT(x >= 0); + ASSERT(b >= 0); + m_buffer.putByteUnchecked(PRE_REX | ((int)w << 3) | ((r>>3)<<2) | ((x>>3)<<1) | (b>>3)); + } + + // Used to plant a REX byte with REX.w set (for 64-bit operations). + inline void emitRexW(int r, int x, int b) + { + emitRex(true, r, x, b); + } + + // Used for operations with byte operands - use byteRegRequiresRex() to check register operands, + // regRequiresRex() to check other registers (i.e. address base & index). + inline void emitRexIf(bool condition, int r, int x, int b) + { + if (condition) emitRex(false, r, x, b); + } + + // Used for word sized operations, will plant a REX prefix if necessary (if any register is r8 or above). + inline void emitRexIfNeeded(int r, int x, int b) + { + emitRexIf(regRequiresRex(r) || regRequiresRex(x) || regRequiresRex(b), r, x, b); + } +#else + // No REX prefix bytes on 32-bit x86. + inline bool regRequiresRex(int) { return false; } + inline bool byteRegRequiresRex(int) { return false; } + inline void emitRexIf(bool, int, int, int) {} + inline void emitRexIfNeeded(int, int, int) {} +#endif + + void putModRm(ModRmMode mode, int reg, RegisterID rm) + { + m_buffer.putByteUnchecked((mode << 6) | ((reg & 7) << 3) | (rm & 7)); + } + + void putModRmSib(ModRmMode mode, int reg, RegisterID base, RegisterID index, int scale) + { + ASSERT(mode != ModRmRegister); + + putModRm(mode, reg, hasSib); + m_buffer.putByteUnchecked((scale << 6) | ((index & 7) << 3) | (base & 7)); + } + + void registerModRM(int reg, RegisterID rm) + { + putModRm(ModRmRegister, reg, rm); + } + + void memoryModRM(int reg, RegisterID base, int offset) + { + // A base of esp or r12 would be interpreted as a sib, so force a sib with no index & put the base in there. +#if CPU(X86_64) + if ((base == hasSib) || (base == hasSib2)) { +#else + if (base == hasSib) { +#endif + if (!offset) // No need to check if the base is noBase, since we know it is hasSib! + putModRmSib(ModRmMemoryNoDisp, reg, base, noIndex, 0); + else if (CAN_SIGN_EXTEND_8_32(offset)) { + putModRmSib(ModRmMemoryDisp8, reg, base, noIndex, 0); + m_buffer.putByteUnchecked(offset); + } else { + putModRmSib(ModRmMemoryDisp32, reg, base, noIndex, 0); + m_buffer.putIntUnchecked(offset); + } + } else { +#if CPU(X86_64) + if (!offset && (base != noBase) && (base != noBase2)) +#else + if (!offset && (base != noBase)) +#endif + putModRm(ModRmMemoryNoDisp, reg, base); + else if (CAN_SIGN_EXTEND_8_32(offset)) { + putModRm(ModRmMemoryDisp8, reg, base); + m_buffer.putByteUnchecked(offset); + } else { + putModRm(ModRmMemoryDisp32, reg, base); + m_buffer.putIntUnchecked(offset); + } + } + } + + void memoryModRM_disp8(int reg, RegisterID base, int offset) + { + // A base of esp or r12 would be interpreted as a sib, so force a sib with no index & put the base in there. + ASSERT(CAN_SIGN_EXTEND_8_32(offset)); +#if CPU(X86_64) + if ((base == hasSib) || (base == hasSib2)) { +#else + if (base == hasSib) { +#endif + putModRmSib(ModRmMemoryDisp8, reg, base, noIndex, 0); + m_buffer.putByteUnchecked(offset); + } else { + putModRm(ModRmMemoryDisp8, reg, base); + m_buffer.putByteUnchecked(offset); + } + } + + void memoryModRM_disp32(int reg, RegisterID base, int offset) + { + // A base of esp or r12 would be interpreted as a sib, so force a sib with no index & put the base in there. +#if CPU(X86_64) + if ((base == hasSib) || (base == hasSib2)) { +#else + if (base == hasSib) { +#endif + putModRmSib(ModRmMemoryDisp32, reg, base, noIndex, 0); + m_buffer.putIntUnchecked(offset); + } else { + putModRm(ModRmMemoryDisp32, reg, base); + m_buffer.putIntUnchecked(offset); + } + } + + void memoryModRM(int reg, RegisterID base, RegisterID index, int scale, int offset) + { + ASSERT(index != noIndex); + +#if CPU(X86_64) + if (!offset && (base != noBase) && (base != noBase2)) +#else + if (!offset && (base != noBase)) +#endif + putModRmSib(ModRmMemoryNoDisp, reg, base, index, scale); + else if (CAN_SIGN_EXTEND_8_32(offset)) { + putModRmSib(ModRmMemoryDisp8, reg, base, index, scale); + m_buffer.putByteUnchecked(offset); + } else { + putModRmSib(ModRmMemoryDisp32, reg, base, index, scale); + m_buffer.putIntUnchecked(offset); + } + } + +#if !CPU(X86_64) + void memoryModRM(int reg, const void* address) + { + // noBase + ModRmMemoryNoDisp means noBase + ModRmMemoryDisp32! + putModRm(ModRmMemoryNoDisp, reg, noBase); + m_buffer.putIntUnchecked(reinterpret_cast<int32_t>(address)); + } +#endif + + AssemblerBuffer m_buffer; + } m_formatter; + int m_indexOfLastWatchpoint; + int m_indexOfTailOfLastWatchpoint; +}; + +} // namespace JSC + +#endif // ENABLE(ASSEMBLER) && CPU(X86) + +#endif // X86Assembler_h |