mirror of
https://git.suyu.dev/suyu/dynarmic.git
synced 2026-01-07 06:58:15 +01:00
backend_x64: Use a reference to BlockOfCode instead of a pointer
This commit is contained in:
MerryMage
parent
8931ee346b
commit
68f46c8334
27 changed files with 1331 additions and 1331 deletions
|
|
@ -17,13 +17,13 @@ namespace Dynarmic::BackendX64 {
|
|||
using namespace Xbyak::util;
|
||||
|
||||
template <typename Function>
|
||||
static void EmitVectorOperation(BlockOfCode* code, EmitContext& ctx, IR::Inst* inst, Function fn) {
|
||||
static void EmitVectorOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Function fn) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
|
||||
(code->*fn)(xmm_a, xmm_b);
|
||||
(code.*fn)(xmm_a, xmm_b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
}
|
||||
|
|
@ -35,15 +35,15 @@ void EmitX64::EmitVectorGetElement8(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
|
||||
code->pextrb(dest, source, index);
|
||||
code.pextrb(dest, source, index);
|
||||
ctx.reg_alloc.DefineValue(inst, dest);
|
||||
} else {
|
||||
Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
|
||||
code->pextrw(dest, source, index / 2);
|
||||
code.pextrw(dest, source, index / 2);
|
||||
if (index % 2 == 1) {
|
||||
code->shr(dest, 8);
|
||||
code.shr(dest, 8);
|
||||
}
|
||||
ctx.reg_alloc.DefineValue(inst, dest);
|
||||
}
|
||||
|
|
@ -56,7 +56,7 @@ void EmitX64::EmitVectorGetElement16(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
Xbyak::Reg32 dest = ctx.reg_alloc.ScratchGpr().cvt32();
|
||||
code->pextrw(dest, source, index);
|
||||
code.pextrw(dest, source, index);
|
||||
ctx.reg_alloc.DefineValue(inst, dest);
|
||||
}
|
||||
|
||||
|
|
@ -69,14 +69,14 @@ void EmitX64::EmitVectorGetElement32(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
if (index == 0) {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
code->movd(dest, source);
|
||||
} else if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
code.movd(dest, source);
|
||||
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
code->pextrd(dest, source, index);
|
||||
code.pextrd(dest, source, index);
|
||||
} else {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
code->pshufd(source, source, index);
|
||||
code->movd(dest, source);
|
||||
code.pshufd(source, source, index);
|
||||
code.movd(dest, source);
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, dest);
|
||||
|
|
@ -91,14 +91,14 @@ void EmitX64::EmitVectorGetElement64(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
if (index == 0) {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
code->movq(dest, source);
|
||||
} else if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
code.movq(dest, source);
|
||||
} else if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseXmm(args[0]);
|
||||
code->pextrq(dest, source, 1);
|
||||
code.pextrq(dest, source, 1);
|
||||
} else {
|
||||
Xbyak::Xmm source = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
code->punpckhqdq(source, source);
|
||||
code->movq(dest, source);
|
||||
code.punpckhqdq(source, source);
|
||||
code.movq(dest, source);
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, dest);
|
||||
|
|
@ -109,11 +109,11 @@ void EmitX64::EmitVectorSetElement8(EmitContext& ctx, IR::Inst* inst) {
|
|||
ASSERT(args[1].IsImmediate());
|
||||
u8 index = args[1].GetImmediateU8();
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg8 source_elem = ctx.reg_alloc.UseGpr(args[2]).cvt8();
|
||||
|
||||
code->pinsrb(source_vector, source_elem.cvt32(), index);
|
||||
code.pinsrb(source_vector, source_elem.cvt32(), index);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
} else {
|
||||
|
|
@ -121,17 +121,17 @@ void EmitX64::EmitVectorSetElement8(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Reg32 source_elem = ctx.reg_alloc.UseScratchGpr(args[2]).cvt32();
|
||||
Xbyak::Reg32 tmp = ctx.reg_alloc.ScratchGpr().cvt32();
|
||||
|
||||
code->pextrw(tmp, source_vector, index / 2);
|
||||
code.pextrw(tmp, source_vector, index / 2);
|
||||
if (index % 2 == 0) {
|
||||
code->and_(tmp, 0xFF00);
|
||||
code->and_(source_elem, 0x00FF);
|
||||
code->or_(tmp, source_elem);
|
||||
code.and_(tmp, 0xFF00);
|
||||
code.and_(source_elem, 0x00FF);
|
||||
code.or_(tmp, source_elem);
|
||||
} else {
|
||||
code->and_(tmp, 0x00FF);
|
||||
code->shl(source_elem, 8);
|
||||
code->or_(tmp, source_elem);
|
||||
code.and_(tmp, 0x00FF);
|
||||
code.shl(source_elem, 8);
|
||||
code.or_(tmp, source_elem);
|
||||
}
|
||||
code->pinsrw(source_vector, tmp, index / 2);
|
||||
code.pinsrw(source_vector, tmp, index / 2);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
}
|
||||
|
|
@ -145,7 +145,7 @@ void EmitX64::EmitVectorSetElement16(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg16 source_elem = ctx.reg_alloc.UseGpr(args[2]).cvt16();
|
||||
|
||||
code->pinsrw(source_vector, source_elem.cvt32(), index);
|
||||
code.pinsrw(source_vector, source_elem.cvt32(), index);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
}
|
||||
|
|
@ -155,20 +155,20 @@ void EmitX64::EmitVectorSetElement32(EmitContext& ctx, IR::Inst* inst) {
|
|||
ASSERT(args[1].IsImmediate());
|
||||
u8 index = args[1].GetImmediateU8();
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg32 source_elem = ctx.reg_alloc.UseGpr(args[2]).cvt32();
|
||||
|
||||
code->pinsrd(source_vector, source_elem, index);
|
||||
code.pinsrd(source_vector, source_elem, index);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
} else {
|
||||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg32 source_elem = ctx.reg_alloc.UseScratchGpr(args[2]).cvt32();
|
||||
|
||||
code->pinsrw(source_vector, source_elem, index * 2);
|
||||
code->shr(source_elem, 16);
|
||||
code->pinsrw(source_vector, source_elem, index * 2 + 1);
|
||||
code.pinsrw(source_vector, source_elem, index * 2);
|
||||
code.shr(source_elem, 16);
|
||||
code.pinsrw(source_vector, source_elem, index * 2 + 1);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
}
|
||||
|
|
@ -179,24 +179,24 @@ void EmitX64::EmitVectorSetElement64(EmitContext& ctx, IR::Inst* inst) {
|
|||
ASSERT(args[1].IsImmediate());
|
||||
u8 index = args[1].GetImmediateU8();
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg64 source_elem = ctx.reg_alloc.UseGpr(args[2]);
|
||||
|
||||
code->pinsrq(source_vector, source_elem, index);
|
||||
code.pinsrq(source_vector, source_elem, index);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
} else {
|
||||
Xbyak::Xmm source_vector = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Reg64 source_elem = ctx.reg_alloc.UseScratchGpr(args[2]);
|
||||
|
||||
code->pinsrw(source_vector, source_elem.cvt32(), index * 4);
|
||||
code->shr(source_elem, 16);
|
||||
code->pinsrw(source_vector, source_elem.cvt32(), index * 4 + 1);
|
||||
code->shr(source_elem, 16);
|
||||
code->pinsrw(source_vector, source_elem.cvt32(), index * 4 + 2);
|
||||
code->shr(source_elem, 16);
|
||||
code->pinsrw(source_vector, source_elem.cvt32(), index * 4 + 3);
|
||||
code.pinsrw(source_vector, source_elem.cvt32(), index * 4);
|
||||
code.shr(source_elem, 16);
|
||||
code.pinsrw(source_vector, source_elem.cvt32(), index * 4 + 1);
|
||||
code.shr(source_elem, 16);
|
||||
code.pinsrw(source_vector, source_elem.cvt32(), index * 4 + 2);
|
||||
code.shr(source_elem, 16);
|
||||
code.pinsrw(source_vector, source_elem.cvt32(), index * 4 + 3);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, source_vector);
|
||||
}
|
||||
|
|
@ -227,15 +227,15 @@ void EmitX64::EmitVectorLowerBroadcast8(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pxor(tmp, tmp);
|
||||
code->pshufb(a, tmp);
|
||||
code->movq(a, a);
|
||||
code.pxor(tmp, tmp);
|
||||
code.pshufb(a, tmp);
|
||||
code.movq(a, a);
|
||||
} else {
|
||||
code->punpcklbw(a, a);
|
||||
code->pshuflw(a, a, 0);
|
||||
code.punpcklbw(a, a);
|
||||
code.pshuflw(a, a, 0);
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
|
|
@ -246,7 +246,7 @@ void EmitX64::EmitVectorLowerBroadcast16(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->pshuflw(a, a, 0);
|
||||
code.pshuflw(a, a, 0);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -256,7 +256,7 @@ void EmitX64::EmitVectorLowerBroadcast32(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->pshuflw(a, a, 0b01000100);
|
||||
code.pshuflw(a, a, 0b01000100);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -266,15 +266,15 @@ void EmitX64::EmitVectorBroadcast8(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pxor(tmp, tmp);
|
||||
code->pshufb(a, tmp);
|
||||
code.pxor(tmp, tmp);
|
||||
code.pshufb(a, tmp);
|
||||
} else {
|
||||
code->punpcklbw(a, a);
|
||||
code->pshuflw(a, a, 0);
|
||||
code->punpcklqdq(a, a);
|
||||
code.punpcklbw(a, a);
|
||||
code.pshuflw(a, a, 0);
|
||||
code.punpcklqdq(a, a);
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
|
|
@ -285,8 +285,8 @@ void EmitX64::EmitVectorBroadcast16(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->pshuflw(a, a, 0);
|
||||
code->punpcklqdq(a, a);
|
||||
code.pshuflw(a, a, 0);
|
||||
code.punpcklqdq(a, a);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -296,7 +296,7 @@ void EmitX64::EmitVectorBroadcast32(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->pshufd(a, a, 0);
|
||||
code.pshufd(a, a, 0);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -306,7 +306,7 @@ void EmitX64::EmitVectorBroadcast64(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->punpcklqdq(a, a);
|
||||
code.punpcklqdq(a, a);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -325,8 +325,8 @@ void EmitX64::EmitVectorNot(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Xmm xmm_b = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pcmpeqw(xmm_b, xmm_b);
|
||||
code->pxor(xmm_a, xmm_b);
|
||||
code.pcmpeqw(xmm_b, xmm_b);
|
||||
code.pxor(xmm_a, xmm_b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
}
|
||||
|
|
@ -344,7 +344,7 @@ void EmitX64::EmitVectorEqual32(EmitContext& ctx, IR::Inst* inst) {
|
|||
}
|
||||
|
||||
void EmitX64::EmitVectorEqual64(EmitContext& ctx, IR::Inst* inst) {
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
EmitVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::pcmpeqq);
|
||||
return;
|
||||
}
|
||||
|
|
@ -355,9 +355,9 @@ void EmitX64::EmitVectorEqual64(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pcmpeqd(xmm_a, xmm_b);
|
||||
code->pshufd(tmp, xmm_a, 0b10110001);
|
||||
code->pand(xmm_a, tmp);
|
||||
code.pcmpeqd(xmm_a, xmm_b);
|
||||
code.pshufd(tmp, xmm_a, 0b10110001);
|
||||
code.pand(xmm_a, tmp);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
}
|
||||
|
|
@ -365,14 +365,14 @@ void EmitX64::EmitVectorEqual64(EmitContext& ctx, IR::Inst* inst) {
|
|||
void EmitX64::EmitVectorEqual128(EmitContext& ctx, IR::Inst* inst) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41)) {
|
||||
Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pcmpeqq(xmm_a, xmm_b);
|
||||
code->pshufd(tmp, xmm_a, 0b01001110);
|
||||
code->pand(xmm_a, tmp);
|
||||
code.pcmpeqq(xmm_a, xmm_b);
|
||||
code.pshufd(tmp, xmm_a, 0b01001110);
|
||||
code.pand(xmm_a, tmp);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
} else {
|
||||
|
|
@ -380,11 +380,11 @@ void EmitX64::EmitVectorEqual128(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->pcmpeqd(xmm_a, xmm_b);
|
||||
code->pshufd(tmp, xmm_a, 0b10110001);
|
||||
code->pand(xmm_a, tmp);
|
||||
code->pshufd(tmp, xmm_a, 0b01001110);
|
||||
code->pand(xmm_a, tmp);
|
||||
code.pcmpeqd(xmm_a, xmm_b);
|
||||
code.pshufd(tmp, xmm_a, 0b10110001);
|
||||
code.pand(xmm_a, tmp);
|
||||
code.pshufd(tmp, xmm_a, 0b01001110);
|
||||
code.pand(xmm_a, tmp);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
}
|
||||
|
|
@ -397,13 +397,13 @@ void EmitX64::EmitVectorLowerPairedAdd8(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->punpcklqdq(xmm_a, xmm_b);
|
||||
code->movdqa(tmp, xmm_a);
|
||||
code->psllw(xmm_a, 8);
|
||||
code->paddw(xmm_a, tmp);
|
||||
code->pxor(tmp, tmp);
|
||||
code->psrlw(xmm_a, 8);
|
||||
code->packuswb(xmm_a, tmp);
|
||||
code.punpcklqdq(xmm_a, xmm_b);
|
||||
code.movdqa(tmp, xmm_a);
|
||||
code.psllw(xmm_a, 8);
|
||||
code.paddw(xmm_a, tmp);
|
||||
code.pxor(tmp, tmp);
|
||||
code.psrlw(xmm_a, 8);
|
||||
code.packuswb(xmm_a, tmp);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
}
|
||||
|
|
@ -415,17 +415,17 @@ void EmitX64::EmitVectorLowerPairedAdd16(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->punpcklqdq(xmm_a, xmm_b);
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
code->pxor(tmp, tmp);
|
||||
code->phaddw(xmm_a, tmp);
|
||||
code.punpcklqdq(xmm_a, xmm_b);
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
code.pxor(tmp, tmp);
|
||||
code.phaddw(xmm_a, tmp);
|
||||
} else {
|
||||
code->movdqa(tmp, xmm_a);
|
||||
code->pslld(xmm_a, 16);
|
||||
code->paddd(xmm_a, tmp);
|
||||
code->pxor(tmp, tmp);
|
||||
code->psrad(xmm_a, 16);
|
||||
code->packssdw(xmm_a, tmp); // Note: packusdw is SSE4.1, hence the arithmetic shift above.
|
||||
code.movdqa(tmp, xmm_a);
|
||||
code.pslld(xmm_a, 16);
|
||||
code.paddd(xmm_a, tmp);
|
||||
code.pxor(tmp, tmp);
|
||||
code.psrad(xmm_a, 16);
|
||||
code.packssdw(xmm_a, tmp); // Note: packusdw is SSE4.1, hence the arithmetic shift above.
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
|
|
@ -438,16 +438,16 @@ void EmitX64::EmitVectorLowerPairedAdd32(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm xmm_b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->punpcklqdq(xmm_a, xmm_b);
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
code->pxor(tmp, tmp);
|
||||
code->phaddd(xmm_a, tmp);
|
||||
code.punpcklqdq(xmm_a, xmm_b);
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
code.pxor(tmp, tmp);
|
||||
code.phaddd(xmm_a, tmp);
|
||||
} else {
|
||||
code->movdqa(tmp, xmm_a);
|
||||
code->psllq(xmm_a, 32);
|
||||
code->paddq(xmm_a, tmp);
|
||||
code->psrlq(xmm_a, 32);
|
||||
code->pshufd(xmm_a, xmm_a, 0b11011000);
|
||||
code.movdqa(tmp, xmm_a);
|
||||
code.psllq(xmm_a, 32);
|
||||
code.paddq(xmm_a, tmp);
|
||||
code.psrlq(xmm_a, 32);
|
||||
code.pshufd(xmm_a, xmm_a, 0b11011000);
|
||||
}
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, xmm_a);
|
||||
|
|
@ -461,15 +461,15 @@ void EmitX64::EmitVectorPairedAdd8(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm c = ctx.reg_alloc.ScratchXmm();
|
||||
Xbyak::Xmm d = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->movdqa(c, a);
|
||||
code->movdqa(d, b);
|
||||
code->psllw(a, 8);
|
||||
code->psllw(b, 8);
|
||||
code->paddw(a, c);
|
||||
code->paddw(b, d);
|
||||
code->psrlw(a, 8);
|
||||
code->psrlw(b, 8);
|
||||
code->packuswb(a, b);
|
||||
code.movdqa(c, a);
|
||||
code.movdqa(d, b);
|
||||
code.psllw(a, 8);
|
||||
code.psllw(b, 8);
|
||||
code.paddw(a, c);
|
||||
code.paddw(b, d);
|
||||
code.psrlw(a, 8);
|
||||
code.psrlw(b, 8);
|
||||
code.packuswb(a, b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -477,11 +477,11 @@ void EmitX64::EmitVectorPairedAdd8(EmitContext& ctx, IR::Inst* inst) {
|
|||
void EmitX64::EmitVectorPairedAdd16(EmitContext& ctx, IR::Inst* inst) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
|
||||
code->phaddw(a, b);
|
||||
code.phaddw(a, b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
} else {
|
||||
|
|
@ -490,15 +490,15 @@ void EmitX64::EmitVectorPairedAdd16(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm c = ctx.reg_alloc.ScratchXmm();
|
||||
Xbyak::Xmm d = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->movdqa(c, a);
|
||||
code->movdqa(d, b);
|
||||
code->pslld(a, 16);
|
||||
code->pslld(b, 16);
|
||||
code->paddd(a, c);
|
||||
code->paddd(b, d);
|
||||
code->psrad(a, 16);
|
||||
code->psrad(b, 16);
|
||||
code->packssdw(a, b);
|
||||
code.movdqa(c, a);
|
||||
code.movdqa(d, b);
|
||||
code.pslld(a, 16);
|
||||
code.pslld(b, 16);
|
||||
code.paddd(a, c);
|
||||
code.paddd(b, d);
|
||||
code.psrad(a, 16);
|
||||
code.psrad(b, 16);
|
||||
code.packssdw(a, b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -507,11 +507,11 @@ void EmitX64::EmitVectorPairedAdd16(EmitContext& ctx, IR::Inst* inst) {
|
|||
void EmitX64::EmitVectorPairedAdd32(EmitContext& ctx, IR::Inst* inst) {
|
||||
auto args = ctx.reg_alloc.GetArgumentInfo(inst);
|
||||
|
||||
if (code->DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSSE3)) {
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
|
||||
code->phaddd(a, b);
|
||||
code.phaddd(a, b);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
} else {
|
||||
|
|
@ -520,13 +520,13 @@ void EmitX64::EmitVectorPairedAdd32(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm c = ctx.reg_alloc.ScratchXmm();
|
||||
Xbyak::Xmm d = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->movdqa(c, a);
|
||||
code->movdqa(d, b);
|
||||
code->psllq(a, 32);
|
||||
code->psllq(b, 32);
|
||||
code->paddq(a, c);
|
||||
code->paddq(b, d);
|
||||
code->shufps(a, b, 0b11011101);
|
||||
code.movdqa(c, a);
|
||||
code.movdqa(d, b);
|
||||
code.psllq(a, 32);
|
||||
code.psllq(b, 32);
|
||||
code.paddq(a, c);
|
||||
code.paddq(b, d);
|
||||
code.shufps(a, b, 0b11011101);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -539,10 +539,10 @@ void EmitX64::EmitVectorPairedAdd64(EmitContext& ctx, IR::Inst* inst) {
|
|||
Xbyak::Xmm b = ctx.reg_alloc.UseXmm(args[1]);
|
||||
Xbyak::Xmm c = ctx.reg_alloc.ScratchXmm();
|
||||
|
||||
code->movdqa(c, a);
|
||||
code->punpcklqdq(a, b);
|
||||
code->punpckhqdq(c, b);
|
||||
code->paddq(a, c);
|
||||
code.movdqa(c, a);
|
||||
code.punpcklqdq(a, b);
|
||||
code.punpckhqdq(c, b);
|
||||
code.paddq(a, c);
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
@ -552,7 +552,7 @@ void EmitX64::EmitVectorZeroUpper(EmitContext& ctx, IR::Inst* inst) {
|
|||
|
||||
Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
|
||||
|
||||
code->movq(a, a); // TODO: !IsLastUse
|
||||
code.movq(a, a); // TODO: !IsLastUse
|
||||
|
||||
ctx.reg_alloc.DefineValue(inst, a);
|
||||
}
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue