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

emit_x64_vector_saturation: AVX512 implementation of VectorUnsignedSaturated{Add,Sub}{32,64}

Browse source
This commit is contained in:
Wunkolo 2021-05-22 11:38:07 -07:00 committed by Merry
parent fa8cc1ac36
commit 332c26d432
1 changed files with 107 additions and 39 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

84
src/dynarmic/backend/x64/emit_x64_vector_saturation.cpp
View file

@ -179,11 +179,24 @@ void EmitX64::EmitVectorUnsignedSaturatedAdd32(EmitContext& ctx, IR::Inst* inst)
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm addend = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg8 overflow = ctx.reg_alloc.ScratchGpr().cvt8();
// TODO AVX2, AVX-512: vpternlog
// TODO AVX2
if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512DQ)) {
// Do a regular unsigned addition
code.vpaddd(result, result, addend);
// Test if an overflow happened
code.vpcmpud(k1, result, addend, CmpInt::LessThan);
// Write 0b1111... where overflows have happened
// This is just a quick way to do this without touching memory
code.vpternlogd(result | k1, result, result, 0xFF);
// Set ZF if an overflow happened
code.ktestb(k1, k1);
} else {
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.movaps(tmp, result);
code.movaps(xmm0, result);
@ -203,6 +216,8 @@ void EmitX64::EmitVectorUnsignedSaturatedAdd32(EmitContext& ctx, IR::Inst* inst)
code.movmskps(overflow.cvt32(), tmp);
code.test(overflow.cvt32(), overflow.cvt32());
}
}
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
@ -214,11 +229,24 @@ void EmitX64::EmitVectorUnsignedSaturatedAdd64(EmitContext& ctx, IR::Inst* inst)
const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Xmm addend = ctx.reg_alloc.UseXmm(args[1]);
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg8 overflow = ctx.reg_alloc.ScratchGpr().cvt8();
// TODO AVX2, AVX-512: vpternlog
// TODO AVX2
if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512DQ)) {
// Do a regular unsigned addition
code.vpaddq(result, result, addend);
// Test if an overflow happened
code.vpcmpuq(k1, result, addend, CmpInt::LessThan);
// Write 0b1111... where overflows have happened
// This is just a quick way to do this without touching memory
code.vpternlogq(result | k1, result, result, 0xFF);
// Set ZF if an overflow happened
code.ktestb(k1, k1);
} else {
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
code.movaps(tmp, result);
code.movaps(xmm0, result);
@ -239,6 +267,8 @@ void EmitX64::EmitVectorUnsignedSaturatedAdd64(EmitContext& ctx, IR::Inst* inst)
code.movmskpd(overflow.cvt32(), tmp);
code.test(overflow.cvt32(), overflow.cvt32());
}
}
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
@ -261,9 +291,28 @@ void EmitX64::EmitVectorUnsignedSaturatedSub32(EmitContext& ctx, IR::Inst* inst)
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg8 overflow = ctx.reg_alloc.ScratchGpr().cvt8();
// TODO AVX2, AVX-512: vpternlog
code.movaps(tmp, result);
// TODO AVX2
if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512DQ)) {
// Do a regular unsigned subtraction
code.vpsubd(result, result, subtrahend);
// Test if an underflow happened
code.vpcmpud(k1, result, tmp, CmpInt::GreaterThan);
// Write 0 where underflows have happened
code.vpxord(result | k1, result, result);
// Set ZF if an underflow happened
code.ktestb(k1, k1);
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
code.movaps(xmm0, subtrahend);
code.pxor(tmp, subtrahend);
@ -295,9 +344,28 @@ void EmitX64::EmitVectorUnsignedSaturatedSub64(EmitContext& ctx, IR::Inst* inst)
const Xbyak::Xmm tmp = ctx.reg_alloc.ScratchXmm();
const Xbyak::Reg8 overflow = ctx.reg_alloc.ScratchGpr().cvt8();
// TODO AVX2, AVX-512: vpternlog
code.movaps(tmp, result);
// TODO AVX2
if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512DQ)) {
// Do a regular unsigned subtraction
code.vpsubq(result, result, subtrahend);
// Test if an underflow happened
code.vpcmpuq(k1, result, tmp, CmpInt::GreaterThan);
// Write 0 where underflows have happened
code.vpxorq(result | k1, result, result);
// Set ZF if an underflow happened
code.ktestb(k1, k1);
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
ctx.reg_alloc.DefineValue(inst, result);
return;
}
code.movaps(xmm0, subtrahend);
code.pxor(tmp, subtrahend);