Smaller cleanup

Cleanup the code to generate a swapValues instruction
to only require one code path for 32/64 bit. In
addition, this seems to be slightly faster even on
x86-64.

Change-Id: I0584c3eb5249606ca7541abfbce227e5cb44711f
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>

2 files changed, 7 insertions, 15 deletions

diff --git a/src/qml/jit/qv4assembler_p.h b/src/qml/jit/qv4assembler_p.h
index dd1209a026..bc379386d4 100644
--- a/src/qml/jit/qv4assembler_p.h
+++ b/src/qml/jit/qv4assembler_p.h
@@ -145,6 +145,7 @@ public:
     static const RegisterID ReturnValueRegister = JSC::X86Registers::eax;
     static const RegisterID ScratchRegister = JSC::X86Registers::ecx;
     static const FPRegisterID FPGpr0 = JSC::X86Registers::xmm0;
+    static const FPRegisterID FPGpr1 = JSC::X86Registers::xmm1;
 
     static const int RegisterSize = 4;
 
@@ -227,6 +228,7 @@ public:
     static const RegisterID ReturnValueRegister = JSC::ARMRegisters::r0;
     static const RegisterID ScratchRegister = JSC::ARMRegisters::r6;
     static const FPRegisterID FPGpr0 = JSC::ARMRegisters::d0;
+    static const FPRegisterID FPGpr1 = JSC::ARMRegisters::d1;
 
     static const int RegisterSize = 4;
 
diff --git a/src/qml/jit/qv4isel_masm.cpp b/src/qml/jit/qv4isel_masm.cpp
index 035d0fec76..706130e995 100644
--- a/src/qml/jit/qv4isel_masm.cpp
+++ b/src/qml/jit/qv4isel_masm.cpp
@@ -818,23 +818,13 @@ void InstructionSelection::swapValues(IR::Temp *sourceTemp, IR::Temp *targetTemp
     } else if (sourceTemp->kind == IR::Temp::StackSlot) {
         if (targetTemp->kind == IR::Temp::StackSlot) {
             // Note: a swap for two stack-slots can involve different types.
-#if CPU(X86_64)
-            _as->load64(_as->stackSlotPointer(targetTemp), Assembler::ReturnValueRegister);
-            _as->load64(_as->stackSlotPointer(sourceTemp), Assembler::ScratchRegister);
-            _as->store64(Assembler::ScratchRegister, _as->stackSlotPointer(targetTemp));
-            _as->store64(Assembler::ReturnValueRegister, _as->stackSlotPointer(sourceTemp));
-#else
-            Assembler::FPRegisterID tReg = _as->toDoubleRegister(targetTemp);
             Assembler::Pointer sAddr = _as->stackSlotPointer(sourceTemp);
             Assembler::Pointer tAddr = _as->stackSlotPointer(targetTemp);
-            _as->load32(sAddr, Assembler::ScratchRegister);
-            _as->store32(Assembler::ScratchRegister, tAddr);
-            sAddr.offset += 4;
-            tAddr.offset += 4;
-            _as->load32(sAddr, Assembler::ScratchRegister);
-            _as->store32(Assembler::ScratchRegister, tAddr);
-            _as->storeDouble(tReg, _as->stackSlotPointer(sourceTemp));
-#endif
+            // use the implementation in JSC::MacroAssembler, as it doesn't do bit swizzling
+            _as->JSC::MacroAssembler::loadDouble(sAddr, Assembler::FPGpr0);
+            _as->JSC::MacroAssembler::loadDouble(tAddr, Assembler::FPGpr1);
+            _as->JSC::MacroAssembler::storeDouble(Assembler::FPGpr1, sAddr);
+            _as->JSC::MacroAssembler::storeDouble(Assembler::FPGpr0, tAddr);
             return;
         }
     }