thirdparty/ppsspp
2
0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

irjit: Fix Vec4Shuffle overlap issue.

Browse source
This commit is contained in:
Unknown W. Brackets 2023-08-08 22:59:55 -07:00
parent e9431d0d1e
commit e73c203984
2 changed files with 12 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

12
Core/MIPS/IR/IRCompVFPU.cpp
View file

@ -1847,22 +1847,22 @@ namespace MIPSComp {
ir.Write(IROp::Vec4Neg, IRVTEMP_0, tregs[0]); ir.Write(IROp::Vec4Neg, IRVTEMP_0, tregs[0]);
// tmp = S[x,x,x,x] * T[w,-z,y,-x] // tmp = S[x,x,x,x] * T[w,-z,y,-x]
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_S, tregs[0], IRVTEMP_0, blendConst(1, 0, 1, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_S, shuffleImm(3, 2, 1, 0));
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(0, 0, 0, 0)); ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(0, 0, 0, 0));
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_T, tregs[0], IRVTEMP_0, blendConst(1, 0, 1, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_T, shuffleImm(3, 2, 1, 0));
ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_D, IRVTEMP_PFX_S, IRVTEMP_PFX_T); ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_D, IRVTEMP_PFX_S, IRVTEMP_PFX_T);
// tmp += S[y,y,y,y] * T[z,w,-x,-y] // tmp += S[y,y,y,y] * T[z,w,-x,-y]
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_S, tregs[0], IRVTEMP_0, blendConst(1, 1, 0, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_S, shuffleImm(2, 3, 0, 1));
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(1, 1, 1, 1)); ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(1, 1, 1, 1));
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_T, tregs[0], IRVTEMP_0, blendConst(1, 1, 0, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_T, shuffleImm(2, 3, 0, 1));
ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_S, IRVTEMP_PFX_S, IRVTEMP_PFX_T); ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_S, IRVTEMP_PFX_S, IRVTEMP_PFX_T);
ir.Write(IROp::Vec4Add, IRVTEMP_PFX_D, IRVTEMP_PFX_D, IRVTEMP_PFX_S); ir.Write(IROp::Vec4Add, IRVTEMP_PFX_D, IRVTEMP_PFX_D, IRVTEMP_PFX_S);
// tmp += S[z,z,z,z] * T[-y,x,w,-z] // tmp += S[z,z,z,z] * T[-y,x,w,-z]
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_S, tregs[0], IRVTEMP_0, blendConst(0, 1, 1, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_S, shuffleImm(1, 0, 3, 2));
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(2, 2, 2, 2)); ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_S, sregs[0], shuffleImm(2, 2, 2, 2));
ir.Write(IRInst{ IROp::Vec4Blend, IRVTEMP_PFX_T, tregs[0], IRVTEMP_0, blendConst(0, 1, 1, 0) });
ir.Write(IROp::Vec4Shuffle, IRVTEMP_PFX_T, IRVTEMP_PFX_T, shuffleImm(1, 0, 3, 2));
ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_S, IRVTEMP_PFX_S, IRVTEMP_PFX_T); ir.Write(IROp::Vec4Mul, IRVTEMP_PFX_S, IRVTEMP_PFX_S, IRVTEMP_PFX_T);
ir.Write(IROp::Vec4Add, IRVTEMP_PFX_D, IRVTEMP_PFX_D, IRVTEMP_PFX_S); ir.Write(IROp::Vec4Add, IRVTEMP_PFX_D, IRVTEMP_PFX_D, IRVTEMP_PFX_S);

7
Core/MIPS/IR/IRInterpreter.cpp
View file

@ -304,11 +304,16 @@ u32 IRInterpret(MIPSState *mips, const IRInst *inst, int count) {
} }
case IROp::Vec4Shuffle: case IROp::Vec4Shuffle:
{
// Can't use the SSE shuffle here because it takes an immediate. pshufb with a table would work though, // Can't use the SSE shuffle here because it takes an immediate. pshufb with a table would work though,
// or a big switch - there are only 256 shuffles possible (4^4) // or a big switch - there are only 256 shuffles possible (4^4)
float temp[4];
for (int i = 0; i < 4; i++) for (int i = 0; i < 4; i++)
mips->f[inst->dest + i] = mips->f[inst->src1 + ((inst->src2 >> (i * 2)) & 3)]; temp[i] = mips->f[inst->src1 + ((inst->src2 >> (i * 2)) & 3)];
for (int i = 0; i < 4; i++)
mips->f[inst->dest + i] = temp[i];
break; break;
}
case IROp::Vec4Blend: case IROp::Vec4Blend:
// Could use _mm_blendv_ps (SSE4+BMI), vbslq_f32 (ARM), __riscv_vmerge_vvm (RISC-V) // Could use _mm_blendv_ps (SSE4+BMI), vbslq_f32 (ARM), __riscv_vmerge_vvm (RISC-V)