libretro/ppsspp
1
0
Fork 0
mirror of https://github.com/libretro/ppsspp.git synced 2025-01-26 03:04:20 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

arm: Jit throughmode 16-bit texcoords.

Browse Source 
It's popular, and this makes decoding such verts much faster.
This commit is contained in:
Unknown W. Brackets 2017-05-06 18:58:15 -07:00
parent 257f8dbbc6
commit 7699fa55de
2 changed files with 59 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
GPU/Common/VertexDecoderArm.cpp
View File

@ -127,7 +127,7 @@ static const JitLookup jitLookup[] = {
{&VertexDecoder::Step_TcFloatPrescale, &VertexDecoderJitCache::Jit_TcFloatPrescale},
{&VertexDecoder::Step_TcFloatThrough, &VertexDecoderJitCache::Jit_TcFloatThrough},
// {&VertexDecoder::Step_TcU16ThroughToFloat, &VertexDecoderJitCache::Jit_TcU16ThroughToFloat},
{&VertexDecoder::Step_TcU16ThroughToFloat, &VertexDecoderJitCache::Jit_TcU16ThroughToFloat},
{&VertexDecoder::Step_NormalS8, &VertexDecoderJitCache::Jit_NormalS8},
{&VertexDecoder::Step_NormalS16, &VertexDecoderJitCache::Jit_NormalS16},
@ -568,6 +568,43 @@ void VertexDecoderJitCache::Jit_TcFloat() {
STR(tempReg2, dstReg, dec_->decFmt.uvoff + 4);
}
void VertexDecoderJitCache::Jit_TcU16ThroughToFloat() {
LDRH(tempReg1, srcReg, dec_->tcoff);
LDRH(tempReg2, srcReg, dec_->tcoff + 2);
MOVP2R(scratchReg, &gstate_c.vertBounds.minU);
auto updateSide = [&](ARMReg r, CCFlags cc, u32 off) {
LDRH(tempReg3, scratchReg, off);
CMP(r, tempReg3);
SetCC(cc);
STRH(r, scratchReg, off);
SetCC(CC_AL);
};
// TODO: Can this actually be fast? Hmm, floats aren't better.
updateSide(tempReg1, CC_LT, offsetof(KnownVertexBounds, minU));
updateSide(tempReg1, CC_GT, offsetof(KnownVertexBounds, maxU));
updateSide(tempReg2, CC_LT, offsetof(KnownVertexBounds, minV));
updateSide(tempReg2, CC_GT, offsetof(KnownVertexBounds, maxV));
if (cpu_info.bNEON) {
ADD(scratchReg, srcReg, dec_->tcoff);
VLD1_lane(I_32, neonScratchReg, scratchReg, 0, false);
VMOVL(I_16 | I_UNSIGNED, neonScratchRegQ, neonScratchReg); // Widen to 32-bit
VCVT(F_32 | I_UNSIGNED, neonScratchRegQ, neonScratchRegQ);
ADD(scratchReg2, dstReg, dec_->decFmt.uvoff);
VST1(F_32, neonScratchReg, scratchReg2, 1, ALIGN_NONE);
} else {
VMOV(fpScratchReg, tempReg1);
VMOV(fpScratchReg2, tempReg2);
VCVT(fpScratchReg, fpScratchReg, TO_FLOAT);
VCVT(fpScratchReg2, fpScratchReg2, TO_FLOAT);
VSTR(fpScratchReg, dstReg, dec_->decFmt.uvoff);
VSTR(fpScratchReg2, dstReg, dec_->decFmt.uvoff + 4);
}
}
void VertexDecoderJitCache::Jit_TcFloatThrough() {
LDR(tempReg1, srcReg, dec_->tcoff);
LDR(tempReg2, srcReg, dec_->tcoff + 4);

22
GPU/Common/VertexDecoderArm64.cpp
View File

@ -101,7 +101,7 @@ static const JitLookup jitLookup[] = {
{&VertexDecoder::Step_TcFloatPrescale, &VertexDecoderJitCache::Jit_TcFloatPrescale},
{&VertexDecoder::Step_TcFloatThrough, &VertexDecoderJitCache::Jit_TcFloatThrough},
// {&VertexDecoder::Step_TcU16ThroughToFloat, &VertexDecoderJitCache::Jit_TcU16ThroughToFloat},
{&VertexDecoder::Step_TcU16ThroughToFloat, &VertexDecoderJitCache::Jit_TcU16ThroughToFloat},
{&VertexDecoder::Step_NormalS8, &VertexDecoderJitCache::Jit_NormalS8},
{&VertexDecoder::Step_NormalS16, &VertexDecoderJitCache::Jit_NormalS16},
@ -579,6 +579,26 @@ void VertexDecoderJitCache::Jit_Color5551() {
CSEL(fullAlphaReg, fullAlphaReg, WZR, CC_EQ);
}
void VertexDecoderJitCache::Jit_TcU16ThroughToFloat() {
LDRH(INDEX_UNSIGNED, tempReg1, srcReg, dec_->tcoff);
LDRH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 2);
auto updateSide = [&](ARM64Reg src, CCFlags cc, ARM64Reg dst) {
CMP(src, dst);
CSEL(dst, src, dst, cc);
};
updateSide(tempReg1, CC_LT, boundsMinUReg);
updateSide(tempReg1, CC_GT, boundsMaxUReg);
updateSide(tempReg2, CC_LT, boundsMinVReg);
updateSide(tempReg2, CC_GT, boundsMaxVReg);
fp.LDUR(32, neonScratchRegD, srcReg, dec_->tcoff);
fp.UXTL(16, neonScratchRegQ, neonScratchRegD); // Widen to 32-bit
fp.UCVTF(32, neonScratchRegD, neonScratchRegD);
fp.STUR(64, neonScratchRegD, dstReg, dec_->decFmt.uvoff);
}
void VertexDecoderJitCache::Jit_TcFloatThrough() {
LDP(INDEX_SIGNED, tempReg1, tempReg2, srcReg, dec_->tcoff);
STP(INDEX_SIGNED, tempReg1, tempReg2, dstReg, dec_->decFmt.uvoff);