diff options
Diffstat (limited to 'src/3rdparty/pcre/sljit/sljitNativeX86_common.c')
| -rw-r--r-- | src/3rdparty/pcre/sljit/sljitNativeX86_common.c | 151 |
1 files changed, 94 insertions, 57 deletions
diff --git a/src/3rdparty/pcre/sljit/sljitNativeX86_common.c b/src/3rdparty/pcre/sljit/sljitNativeX86_common.c index ab98a03d2c..ceb3d675b7 100644 --- a/src/3rdparty/pcre/sljit/sljitNativeX86_common.c +++ b/src/3rdparty/pcre/sljit/sljitNativeX86_common.c @@ -206,6 +206,7 @@ static SLJIT_CONST sljit_ub reg_lmap[SLJIT_NO_REGISTERS + 4] = { #define OR_r_rm 0x0b #define OR_EAX_i32 0x0d #define OR_rm_r 0x09 +#define OR_rm8_r8 0x08 #define POP_r 0x58 #define POP_rm 0x8f #define POPF 0x9d @@ -267,75 +268,54 @@ static sljit_si cpu_has_sse2 = -1; #endif static sljit_si cpu_has_cmov = -1; -#if defined(_MSC_VER) && (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) -#if _MSC_VER >= 1400 +#if defined(_MSC_VER) && _MSC_VER >= 1400 #include <intrin.h> -#else -#error "MSVC does not support inline assembly in 64 bit mode" #endif -#endif /* _MSC_VER && SLJIT_CONFIG_X86_64 */ static void get_cpu_features(void) { sljit_ui features; -#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) +#if defined(_MSC_VER) && _MSC_VER >= 1400 + + int CPUInfo[4]; + __cpuid(CPUInfo, 1); + features = (sljit_ui)CPUInfo[3]; + +#elif defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) -#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) /* AT&T syntax. */ __asm__ ( - "pushl %%ebx\n" "movl $0x1, %%eax\n" +#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) + /* On x86-32, there is no red zone, so this + should work (no need for a local variable). */ + "push %%ebx\n" +#endif "cpuid\n" - "popl %%ebx\n" +#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) + "pop %%ebx\n" +#endif "movl %%edx, %0\n" : "=g" (features) : +#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) : "%eax", "%ecx", "%edx" - ); -#elif defined(_MSC_VER) || defined(__BORLANDC__) - /* Intel syntax. */ - __asm { - mov eax, 1 - push ebx - cpuid - pop ebx - mov features, edx - } #else -# error "SLJIT_DETECT_SSE2 is not implemented for this C compiler" + : "%rax", "%rbx", "%rcx", "%rdx" #endif - -#else /* SLJIT_CONFIG_X86_32 */ - -#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) - /* AT&T syntax. */ - __asm__ ( - "pushq %%rbx\n" - "movl $0x1, %%eax\n" - "cpuid\n" - "popq %%rbx\n" - "movl %%edx, %0\n" - : "=g" (features) - : - : "%rax", "%rcx", "%rdx" ); -#elif defined(_MSC_VER) && _MSC_VER >= 1400 - int CPUInfo[4]; - __cpuid(CPUInfo, 1); - features = (sljit_ui)CPUInfo[3]; -#else +#else /* _MSC_VER && _MSC_VER >= 1400 */ + + /* Intel syntax. */ __asm { mov eax, 1 - push rbx cpuid - pop rbx mov features, edx } -#endif -#endif /* SLJIT_CONFIG_X86_32 */ +#endif /* _MSC_VER && _MSC_VER >= 1400 */ #if (defined SLJIT_SSE2 && SLJIT_SSE2) && (defined SLJIT_DETECT_SSE2 && SLJIT_DETECT_SSE2) cpu_has_sse2 = (features >> 26) & 0x1; @@ -570,7 +550,7 @@ SLJIT_API_FUNC_ATTRIBUTE void* sljit_generate_code(struct sljit_compiler *compil /* Maybe we waste some space because of short jumps. */ SLJIT_ASSERT(code_ptr <= code + compiler->size); compiler->error = SLJIT_ERR_COMPILED; - compiler->executable_size = compiler->size; + compiler->executable_size = code_ptr - code; return (void*)code; } @@ -650,9 +630,10 @@ static void SLJIT_CALL sljit_grow_stack(sljit_sw local_size) This function touches all 4k pages belongs to the requested stack space, which size is passed in local_size. This is necessary on Windows where the stack can only grow in 4k steps. However, this function just burn - CPU cycles if the stack is large enough, but you don't know it in advance. - I think this is a bad design even if it has some reasons. */ - alloca(local_size); + CPU cycles if the stack is large enough. However, you don't know it in + advance, so it must always be called. I think this is a bad design in + general even if it has some reasons. */ + *(sljit_si*)alloca(local_size) = 0; } #endif @@ -1785,7 +1766,7 @@ static sljit_si emit_mul(struct sljit_compiler *compiler, return SLJIT_SUCCESS; } -static sljit_si emit_lea_binary(struct sljit_compiler *compiler, +static sljit_si emit_lea_binary(struct sljit_compiler *compiler, sljit_si keep_flags, sljit_si dst, sljit_sw dstw, sljit_si src1, sljit_sw src1w, sljit_si src2, sljit_sw src2w) @@ -1794,10 +1775,12 @@ static sljit_si emit_lea_binary(struct sljit_compiler *compiler, sljit_si dst_r, done = 0; /* These cases better be left to handled by normal way. */ - if (dst == src1 && dstw == src1w) - return SLJIT_ERR_UNSUPPORTED; - if (dst == src2 && dstw == src2w) - return SLJIT_ERR_UNSUPPORTED; + if (!keep_flags) { + if (dst == src1 && dstw == src1w) + return SLJIT_ERR_UNSUPPORTED; + if (dst == src2 && dstw == src2w) + return SLJIT_ERR_UNSUPPORTED; + } dst_r = (dst <= TMP_REGISTER) ? dst : TMP_REGISTER; @@ -2153,7 +2136,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_op2(struct sljit_compiler *compiler switch (GET_OPCODE(op)) { case SLJIT_ADD: if (!GET_FLAGS(op)) { - if (emit_lea_binary(compiler, dst, dstw, src1, src1w, src2, src2w) != SLJIT_ERR_UNSUPPORTED) + if (emit_lea_binary(compiler, op & SLJIT_KEEP_FLAGS, dst, dstw, src1, src1w, src2, src2w) != SLJIT_ERR_UNSUPPORTED) return compiler->error; } else @@ -2173,7 +2156,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_op2(struct sljit_compiler *compiler dst, dstw, src1, src1w, src2, src2w); case SLJIT_SUB: if (!GET_FLAGS(op)) { - if ((src2 & SLJIT_IMM) && emit_lea_binary(compiler, dst, dstw, src1, src1w, SLJIT_IMM, -src2w) != SLJIT_ERR_UNSUPPORTED) + if ((src2 & SLJIT_IMM) && emit_lea_binary(compiler, op & SLJIT_KEEP_FLAGS, dst, dstw, src1, src1w, SLJIT_IMM, -src2w) != SLJIT_ERR_UNSUPPORTED) return compiler->error; } else @@ -2231,6 +2214,12 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_get_register_index(sljit_si reg) return reg_map[reg]; } +SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_get_float_register_index(sljit_si reg) +{ + check_sljit_get_float_register_index(reg); + return reg; +} + SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_op_custom(struct sljit_compiler *compiler, void *instruction, sljit_si size) { @@ -2637,6 +2626,21 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_op_flags(struct sljit_compiler *com cond_set = get_jump_code(type) + 0x10; #if (defined SLJIT_CONFIG_X86_64 && SLJIT_CONFIG_X86_64) + if (GET_OPCODE(op) == SLJIT_OR && !GET_ALL_FLAGS(op) && dst <= TMP_REGISTER && dst == src) { + inst = (sljit_ub*)ensure_buf(compiler, 1 + 4 + 3); + FAIL_IF(!inst); + INC_SIZE(4 + 3); + /* Set low register to conditional flag. */ + *inst++ = (reg_map[TMP_REGISTER] <= 7) ? REX : REX_B; + *inst++ = GROUP_0F; + *inst++ = cond_set; + *inst++ = MOD_REG | reg_lmap[TMP_REGISTER]; + *inst++ = REX | (reg_map[TMP_REGISTER] <= 7 ? 0 : REX_R) | (reg_map[dst] <= 7 ? 0 : REX_B); + *inst++ = OR_rm8_r8; + *inst++ = MOD_REG | (reg_lmap[TMP_REGISTER] << 3) | reg_lmap[dst]; + return SLJIT_SUCCESS; + } + reg = (op == SLJIT_MOV && dst <= TMP_REGISTER) ? dst : TMP_REGISTER; inst = (sljit_ub*)ensure_buf(compiler, 1 + 4 + 4); @@ -2717,6 +2721,39 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_op_flags(struct sljit_compiler *com return SLJIT_SUCCESS; } + if (GET_OPCODE(op) == SLJIT_OR && !GET_ALL_FLAGS(op) && dst <= TMP_REGISTER && dst == src && reg_map[dst] <= 4) { + SLJIT_COMPILE_ASSERT(reg_map[SLJIT_SCRATCH_REG1] == 0, scratch_reg1_must_be_eax); + if (dst != SLJIT_SCRATCH_REG1) { + inst = (sljit_ub*)ensure_buf(compiler, 1 + 1 + 3 + 2 + 1); + FAIL_IF(!inst); + INC_SIZE(1 + 3 + 2 + 1); + /* Set low register to conditional flag. */ + *inst++ = XCHG_EAX_r + reg_map[TMP_REGISTER]; + *inst++ = GROUP_0F; + *inst++ = cond_set; + *inst++ = MOD_REG | 0 /* eax */; + *inst++ = OR_rm8_r8; + *inst++ = MOD_REG | (0 /* eax */ << 3) | reg_map[dst]; + *inst++ = XCHG_EAX_r + reg_map[TMP_REGISTER]; + } + else { + inst = (sljit_ub*)ensure_buf(compiler, 1 + 2 + 3 + 2 + 2); + FAIL_IF(!inst); + INC_SIZE(2 + 3 + 2 + 2); + /* Set low register to conditional flag. */ + *inst++ = XCHG_r_rm; + *inst++ = MOD_REG | (1 /* ecx */ << 3) | reg_map[TMP_REGISTER]; + *inst++ = GROUP_0F; + *inst++ = cond_set; + *inst++ = MOD_REG | 1 /* ecx */; + *inst++ = OR_rm8_r8; + *inst++ = MOD_REG | (1 /* ecx */ << 3) | 0 /* eax */; + *inst++ = XCHG_r_rm; + *inst++ = MOD_REG | (1 /* ecx */ << 3) | reg_map[TMP_REGISTER]; + } + return SLJIT_SUCCESS; + } + /* Set TMP_REGISTER to the bit. */ inst = (sljit_ub*)ensure_buf(compiler, 1 + 1 + 3 + 3 + 1); FAIL_IF(!inst); @@ -2761,16 +2798,16 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_get_local_base(struct sljit_compiler *co if (NOT_HALFWORD(offset)) { FAIL_IF(emit_load_imm64(compiler, TMP_REGISTER, offset)); #if (defined SLJIT_DEBUG && SLJIT_DEBUG) - SLJIT_ASSERT(emit_lea_binary(compiler, dst, dstw, SLJIT_LOCALS_REG, 0, TMP_REGISTER, 0) != SLJIT_ERR_UNSUPPORTED); + SLJIT_ASSERT(emit_lea_binary(compiler, SLJIT_KEEP_FLAGS, dst, dstw, SLJIT_LOCALS_REG, 0, TMP_REGISTER, 0) != SLJIT_ERR_UNSUPPORTED); return compiler->error; #else - return emit_lea_binary(compiler, dst, dstw, SLJIT_LOCALS_REG, 0, TMP_REGISTER, 0); + return emit_lea_binary(compiler, SLJIT_KEEP_FLAGS, dst, dstw, SLJIT_LOCALS_REG, 0, TMP_REGISTER, 0); #endif } #endif if (offset != 0) - return emit_lea_binary(compiler, dst, dstw, SLJIT_LOCALS_REG, 0, SLJIT_IMM, offset); + return emit_lea_binary(compiler, SLJIT_KEEP_FLAGS, dst, dstw, SLJIT_LOCALS_REG, 0, SLJIT_IMM, offset); return emit_mov(compiler, dst, dstw, SLJIT_LOCALS_REG, 0); } |