Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-12-14 17:09:33 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

samplerjit: Tiny dependency optimizations.

Browse Source 
This had a small but measureable impact (~0.3%.)
This commit is contained in:
Unknown W. Brackets 2021-12-31 08:11:57 -08:00
parent fc3688d273
commit 8ea67b571b
1 changed files with 8 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
GPU/Software/SamplerX86.cpp
View File

@ -1908,17 +1908,19 @@ bool SamplerJitCache::Jit_GetTexelCoordsQuad(const SamplerID &id) {
SUB(32, R(xReg), Imm8(128 - 12)); SUB(32, R(xReg), Imm8(128 - 12));
NEG(32, R(yReg)); NEG(32, R(yReg));
SUB(32, R(yReg), Imm8(128 - 12)); SUB(32, R(yReg), Imm8(128 - 12));
SHL(64, R(yReg), Imm8(32));
OR(64, R(xReg), R(yReg));
// Add them in. We do this in the SSE because we have more to do there... // Add them in. We do this in the SSE because we have more to do there...
X64Reg tempXYReg = regCache_.Alloc(RegCache::VEC_TEMP0); X64Reg tempXYReg = regCache_.Alloc(RegCache::VEC_TEMP0);
MOVD_xmm(tempXYReg, R(xReg)); MOVQ_xmm(tempXYReg, R(xReg));
PADDD(sReg, R(tempXYReg)); PADDD(sReg, R(tempXYReg));
if (regCache_.Has(RegCache::VEC_U1)) { if (regCache_.Has(RegCache::VEC_U1)) {
X64Reg u1Reg = regCache_.Find(RegCache::VEC_U1); X64Reg u1Reg = regCache_.Find(RegCache::VEC_U1);
PADDD(u1Reg, R(tempXYReg)); PADDD(u1Reg, R(tempXYReg));
regCache_.Unlock(u1Reg, RegCache::VEC_U1); regCache_.Unlock(u1Reg, RegCache::VEC_U1);
} }
MOVD_xmm(tempXYReg, R(yReg)); PSHUFD(tempXYReg, R(tempXYReg), _MM_SHUFFLE(1, 1, 1, 1));
PADDD(tReg, R(tempXYReg)); PADDD(tReg, R(tempXYReg));
if (regCache_.Has(RegCache::VEC_V1)) { if (regCache_.Has(RegCache::VEC_V1)) {
X64Reg v1Reg = regCache_.Find(RegCache::VEC_V1); X64Reg v1Reg = regCache_.Find(RegCache::VEC_V1);
@ -1962,10 +1964,10 @@ bool SamplerJitCache::Jit_GetTexelCoordsQuad(const SamplerID &id) {
regCache_.Unlock(fracVReg, RegCache::GEN_ARG_FRAC_V); regCache_.Unlock(fracVReg, RegCache::GEN_ARG_FRAC_V);
// Get rid of the fractional bits, and spread out. // Get rid of the fractional bits, and spread out.
PSRAD(sReg, 8);
PSRAD(tReg, 8);
PSHUFD(sReg, R(sReg), _MM_SHUFFLE(0, 0, 0, 0)); PSHUFD(sReg, R(sReg), _MM_SHUFFLE(0, 0, 0, 0));
PSHUFD(tReg, R(tReg), _MM_SHUFFLE(0, 0, 0, 0)); PSHUFD(tReg, R(tReg), _MM_SHUFFLE(0, 0, 0, 0));
PSRAD(sReg, 8);
PSRAD(tReg, 8);
// Add U/V values for the next coords. // Add U/V values for the next coords.
PADDD(sReg, M(constUNext_)); PADDD(sReg, M(constUNext_));
PADDD(tReg, M(constVNext_)); PADDD(tReg, M(constVNext_));
@ -1973,10 +1975,10 @@ bool SamplerJitCache::Jit_GetTexelCoordsQuad(const SamplerID &id) {
if (regCache_.Has(RegCache::VEC_U1)) { if (regCache_.Has(RegCache::VEC_U1)) {
X64Reg u1Reg = regCache_.Find(RegCache::VEC_U1); X64Reg u1Reg = regCache_.Find(RegCache::VEC_U1);
X64Reg v1Reg = regCache_.Find(RegCache::VEC_V1); X64Reg v1Reg = regCache_.Find(RegCache::VEC_V1);
PSRAD(u1Reg, 8);
PSRAD(v1Reg, 8);
PSHUFD(u1Reg, R(u1Reg), _MM_SHUFFLE(0, 0, 0, 0)); PSHUFD(u1Reg, R(u1Reg), _MM_SHUFFLE(0, 0, 0, 0));
PSHUFD(v1Reg, R(v1Reg), _MM_SHUFFLE(0, 0, 0, 0)); PSHUFD(v1Reg, R(v1Reg), _MM_SHUFFLE(0, 0, 0, 0));
PSRAD(u1Reg, 8);
PSRAD(v1Reg, 8);
PADDD(u1Reg, M(constUNext_)); PADDD(u1Reg, M(constUNext_));
PADDD(v1Reg, M(constVNext_)); PADDD(v1Reg, M(constVNext_));
regCache_.Unlock(u1Reg, RegCache::VEC_U1); regCache_.Unlock(u1Reg, RegCache::VEC_U1);