mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

IR: Add IR instruction ZeroExtendToQuad

Browse source
This commit is contained in:
MerryMage 2018-01-24 15:54:11 +00:00
parent af848c627d
commit 28ccd85e5c
5 changed files with 21 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
src/backend_x64/emit_x64_data_processing.cpp
View file

@ -1090,6 +1090,18 @@ void EmitX64::EmitZeroExtendWordToLong(EmitContext& ctx, IR::Inst* inst) {
ctx.reg_alloc.DefineValue(inst, result); ctx.reg_alloc.DefineValue(inst, result);
} }
void EmitX64::EmitZeroExtendLongToQuad(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
if (args[0].IsInGpr()) {
// We let the register allocator automatically zero extend this when necessary
ctx.reg_alloc.DefineValue(inst, args[0]);
} else {
Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
code->movq(result, result);
ctx.reg_alloc.DefineValue(inst, result);
}
}
void EmitX64::EmitByteReverseWord(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitByteReverseWord(EmitContext& ctx, IR::Inst* inst) {
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
Xbyak::Reg32 result = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32(); Xbyak::Reg32 result = ctx.reg_alloc.UseScratchGpr(args[0]).cvt32();

9
src/backend_x64/reg_alloc.cpp
View file

@ -559,22 +559,19 @@ void RegAlloc::EmitMove(HostLoc to, HostLoc from) {
if (HostLocIsXMM(to) && HostLocIsXMM(from)) { if (HostLocIsXMM(to) && HostLocIsXMM(from)) {
code->movaps(HostLocToXmm(to), HostLocToXmm(from)); code->movaps(HostLocToXmm(to), HostLocToXmm(from));
} else if (HostLocIsGPR(to) && HostLocIsGPR(from)) { } else if (HostLocIsGPR(to) && HostLocIsGPR(from)) {
ASSERT(bit_width != 128); if (bit_width >= 64) {
if (bit_width == 64) {
code->mov(HostLocToReg64(to), HostLocToReg64(from)); code->mov(HostLocToReg64(to), HostLocToReg64(from));
} else { } else {
code->mov(HostLocToReg64(to).cvt32(), HostLocToReg64(from).cvt32()); code->mov(HostLocToReg64(to).cvt32(), HostLocToReg64(from).cvt32());
} }
} else if (HostLocIsXMM(to) && HostLocIsGPR(from)) { } else if (HostLocIsXMM(to) && HostLocIsGPR(from)) {
ASSERT(bit_width != 128); if (bit_width >= 64) {
if (bit_width == 64) {
code->movq(HostLocToXmm(to), HostLocToReg64(from)); code->movq(HostLocToXmm(to), HostLocToReg64(from));
} else { } else {
code->movd(HostLocToXmm(to), HostLocToReg64(from).cvt32()); code->movd(HostLocToXmm(to), HostLocToReg64(from).cvt32());
} }
} else if (HostLocIsGPR(to) && HostLocIsXMM(from)) { } else if (HostLocIsGPR(to) && HostLocIsXMM(from)) {
ASSERT(bit_width != 128); if (bit_width >= 64) {
if (bit_width == 64) {
code->movq(HostLocToReg64(to), HostLocToXmm(from)); code->movq(HostLocToReg64(to), HostLocToXmm(from));
} else { } else {
code->movd(HostLocToReg64(to).cvt32(), HostLocToXmm(from)); code->movd(HostLocToReg64(to).cvt32(), HostLocToXmm(from));

4
src/frontend/ir/ir_emitter.cpp
View file

@ -396,6 +396,10 @@ U32 IREmitter::ZeroExtendToWord(const UAny& a) {
} }
} }
U128 IREmitter::ZeroExtendToQuad(const UAny& a) {
return Inst<U128>(Opcode::ZeroExtendLongToQuad, ZeroExtendToLong(a));
}
U64 IREmitter::ZeroExtendWordToLong(const U32& a) { U64 IREmitter::ZeroExtendWordToLong(const U32& a) {
return Inst<U64>(Opcode::ZeroExtendWordToLong, a); return Inst<U64>(Opcode::ZeroExtendWordToLong, a);
} }

1
src/frontend/ir/ir_emitter.h
View file

@ -125,6 +125,7 @@ public:
U64 SignExtendWordToLong(const U32& a); U64 SignExtendWordToLong(const U32& a);
U32 ZeroExtendToWord(const UAny& a); U32 ZeroExtendToWord(const UAny& a);
U64 ZeroExtendToLong(const UAny& a); U64 ZeroExtendToLong(const UAny& a);
U128 ZeroExtendToQuad(const UAny& a);
U32 ZeroExtendByteToWord(const U8& a); U32 ZeroExtendByteToWord(const U8& a);
U32 ZeroExtendHalfToWord(const U16& a); U32 ZeroExtendHalfToWord(const U16& a);
U64 ZeroExtendWordToLong(const U32& a); U64 ZeroExtendWordToLong(const U32& a);

1
src/frontend/ir/opcodes.inc
View file

@ -112,6 +112,7 @@ OPCODE(ZeroExtendHalfToWord, T::U32, T::U16
OPCODE(ZeroExtendByteToLong, T::U64, T::U8 ) OPCODE(ZeroExtendByteToLong, T::U64, T::U8 )
OPCODE(ZeroExtendHalfToLong, T::U64, T::U16 ) OPCODE(ZeroExtendHalfToLong, T::U64, T::U16 )
OPCODE(ZeroExtendWordToLong, T::U64, T::U32 ) OPCODE(ZeroExtendWordToLong, T::U64, T::U32 )
OPCODE(ZeroExtendLongToQuad, T::U128, T::U64 )
OPCODE(ByteReverseWord, T::U32, T::U32 ) OPCODE(ByteReverseWord, T::U32, T::U32 )
OPCODE(ByteReverseHalf, T::U16, T::U16 ) OPCODE(ByteReverseHalf, T::U16, T::U16 )
OPCODE(ByteReverseDual, T::U64, T::U64 ) OPCODE(ByteReverseDual, T::U64, T::U64 )