Emulation/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2025-04-02 11:01:50 -04:00
Code Issues Projects Releases Packages Wiki Activity

ARM64: Some minor vertex decoder work. Hm, I think SCVTF will actually divide by 128.0, not 127.0 :/

Browse source
This commit is contained in:
Henrik Rydgard 2015-03-19 23:04:46 +01:00
parent 1a02e32ad1
commit 5496b3d3b1
3 changed files with 99 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

91
GPU/Common/VertexDecoderArm64.cpp
View file

@ -53,7 +53,7 @@ static const ARM64Reg fpUVoffsetReg = D1;
static const ARM64Reg neonScratchReg = D2; static const ARM64Reg neonScratchReg = D2;
static const ARM64Reg neonScratchReg2 = D3; static const ARM64Reg neonScratchReg2 = D3;
static const ARM64Reg neonScratchRegQ = Q1; // Overlaps with all the scratch regs static const ARM64Reg neonScratchRegQ = Q1;
// Everything above S6 is fair game for skinning // Everything above S6 is fair game for skinning
@ -85,15 +85,18 @@ static const JitLookup jitLookup[] = {
{&VertexDecoder::Step_TcU8Prescale, &VertexDecoderJitCache::Jit_TcU8Prescale}, {&VertexDecoder::Step_TcU8Prescale, &VertexDecoderJitCache::Jit_TcU8Prescale},
{&VertexDecoder::Step_TcU16Prescale, &VertexDecoderJitCache::Jit_TcU16Prescale}, {&VertexDecoder::Step_TcU16Prescale, &VertexDecoderJitCache::Jit_TcU16Prescale},
{&VertexDecoder::Step_TcFloatPrescale, &VertexDecoderJitCache::Jit_TcFloatPrescale}, {&VertexDecoder::Step_TcFloatPrescale, &VertexDecoderJitCache::Jit_TcFloatPrescale},
*/
{&VertexDecoder::Step_TcU16Through, &VertexDecoderJitCache::Jit_TcU16Through}, {&VertexDecoder::Step_TcU16Through, &VertexDecoderJitCache::Jit_TcU16Through},
/*
{&VertexDecoder::Step_TcFloatThrough, &VertexDecoderJitCache::Jit_TcFloatThrough}, {&VertexDecoder::Step_TcFloatThrough, &VertexDecoderJitCache::Jit_TcFloatThrough},
{&VertexDecoder::Step_TcU16ThroughDouble, &VertexDecoderJitCache::Jit_TcU16ThroughDouble}, {&VertexDecoder::Step_TcU16ThroughDouble, &VertexDecoderJitCache::Jit_TcU16ThroughDouble},
*/
{&VertexDecoder::Step_NormalS8, &VertexDecoderJitCache::Jit_NormalS8}, {&VertexDecoder::Step_NormalS8, &VertexDecoderJitCache::Jit_NormalS8},
{&VertexDecoder::Step_NormalS16, &VertexDecoderJitCache::Jit_NormalS16}, {&VertexDecoder::Step_NormalS16, &VertexDecoderJitCache::Jit_NormalS16},
{&VertexDecoder::Step_NormalFloat, &VertexDecoderJitCache::Jit_NormalFloat}, {&VertexDecoder::Step_NormalFloat, &VertexDecoderJitCache::Jit_NormalFloat},
/*
{&VertexDecoder::Step_NormalS8Skin, &VertexDecoderJitCache::Jit_NormalS8Skin}, {&VertexDecoder::Step_NormalS8Skin, &VertexDecoderJitCache::Jit_NormalS8Skin},
{&VertexDecoder::Step_NormalS16Skin, &VertexDecoderJitCache::Jit_NormalS16Skin}, {&VertexDecoder::Step_NormalS16Skin, &VertexDecoderJitCache::Jit_NormalS16Skin},
{&VertexDecoder::Step_NormalFloatSkin, &VertexDecoderJitCache::Jit_NormalFloatSkin}, {&VertexDecoder::Step_NormalFloatSkin, &VertexDecoderJitCache::Jit_NormalFloatSkin},
@ -105,9 +108,13 @@ static const JitLookup jitLookup[] = {
{&VertexDecoder::Step_Color5551, &VertexDecoderJitCache::Jit_Color5551}, {&VertexDecoder::Step_Color5551, &VertexDecoderJitCache::Jit_Color5551},
{&VertexDecoder::Step_PosS8Through, &VertexDecoderJitCache::Jit_PosS8Through}, {&VertexDecoder::Step_PosS8Through, &VertexDecoderJitCache::Jit_PosS8Through},
*/
{&VertexDecoder::Step_PosS16Through, &VertexDecoderJitCache::Jit_PosS16Through}, {&VertexDecoder::Step_PosS16Through, &VertexDecoderJitCache::Jit_PosS16Through},
/*
{&VertexDecoder::Step_PosFloatThrough, &VertexDecoderJitCache::Jit_PosFloat}, {&VertexDecoder::Step_PosFloatThrough, &VertexDecoderJitCache::Jit_PosFloat},
*/ */
{&VertexDecoder::Step_PosS8, &VertexDecoderJitCache::Jit_PosS8},
{&VertexDecoder::Step_PosS16, &VertexDecoderJitCache::Jit_PosS16},
{&VertexDecoder::Step_PosFloat, &VertexDecoderJitCache::Jit_PosFloat}, {&VertexDecoder::Step_PosFloat, &VertexDecoderJitCache::Jit_PosFloat},
/* /*
{&VertexDecoder::Step_PosS8Skin, &VertexDecoderJitCache::Jit_PosS8Skin}, {&VertexDecoder::Step_PosS8Skin, &VertexDecoderJitCache::Jit_PosS8Skin},
@ -240,6 +247,13 @@ void VertexDecoderJitCache::Jit_TcU16() {
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.uvoff); STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.uvoff);
} }
void VertexDecoderJitCache::Jit_TcU16Through() {
LDRH(INDEX_UNSIGNED, tempReg1, srcReg, dec_->tcoff);
LDRH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 2);
ORR(tempReg1, tempReg1, tempReg2, ArithOption(tempReg2, ST_LSL, 16));
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.uvoff);
}
void VertexDecoderJitCache::Jit_TcFloat() { void VertexDecoderJitCache::Jit_TcFloat() {
LDR(INDEX_UNSIGNED, tempReg1, srcReg, dec_->tcoff); LDR(INDEX_UNSIGNED, tempReg1, srcReg, dec_->tcoff);
LDR(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 4); LDR(INDEX_UNSIGNED, tempReg2, srcReg, dec_->tcoff + 4);
@ -247,6 +261,20 @@ void VertexDecoderJitCache::Jit_TcFloat() {
STR(INDEX_UNSIGNED, tempReg2, dstReg, dec_->decFmt.uvoff + 4); STR(INDEX_UNSIGNED, tempReg2, dstReg, dec_->decFmt.uvoff + 4);
} }
void VertexDecoderJitCache::Jit_PosS8() {
Jit_AnyS8ToFloat(dec_->posoff);
STR(INDEX_UNSIGNED, src[0], dstReg, dec_->decFmt.posoff);
STR(INDEX_UNSIGNED, src[1], dstReg, dec_->decFmt.posoff + 4);
STR(INDEX_UNSIGNED, src[2], dstReg, dec_->decFmt.posoff + 8);
}
void VertexDecoderJitCache::Jit_PosS16() {
Jit_AnyS16ToFloat(dec_->posoff);
STR(INDEX_UNSIGNED, src[0], dstReg, dec_->decFmt.posoff);
STR(INDEX_UNSIGNED, src[1], dstReg, dec_->decFmt.posoff + 4);
STR(INDEX_UNSIGNED, src[2], dstReg, dec_->decFmt.posoff + 8);
}
// Just copy 12 bytes. // Just copy 12 bytes.
void VertexDecoderJitCache::Jit_PosFloat() { void VertexDecoderJitCache::Jit_PosFloat() {
LDR(INDEX_UNSIGNED, tempReg1, srcReg, dec_->posoff); LDR(INDEX_UNSIGNED, tempReg1, srcReg, dec_->posoff);
@ -256,3 +284,62 @@ void VertexDecoderJitCache::Jit_PosFloat() {
STR(INDEX_UNSIGNED, tempReg2, dstReg, dec_->decFmt.posoff + 4); STR(INDEX_UNSIGNED, tempReg2, dstReg, dec_->decFmt.posoff + 4);
STR(INDEX_UNSIGNED, tempReg3, dstReg, dec_->decFmt.posoff + 8); STR(INDEX_UNSIGNED, tempReg3, dstReg, dec_->decFmt.posoff + 8);
} }
void VertexDecoderJitCache::Jit_PosS16Through() {
LDRSH(INDEX_UNSIGNED, tempReg1, srcReg, dec_->posoff);
LDRSH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->posoff + 2);
LDRH(INDEX_UNSIGNED, tempReg3, srcReg, dec_->posoff + 4);
fp.SCVTF(fpScratchReg, tempReg1);
fp.SCVTF(fpScratchReg2, tempReg2);
fp.SCVTF(fpScratchReg3, tempReg3);
STR(INDEX_UNSIGNED, fpScratchReg, dstReg, dec_->decFmt.posoff);
STR(INDEX_UNSIGNED, fpScratchReg2, dstReg, dec_->decFmt.posoff + 4);
STR(INDEX_UNSIGNED, fpScratchReg3, dstReg, dec_->decFmt.posoff + 8);
}
void VertexDecoderJitCache::Jit_NormalS8() {
LDRB(INDEX_UNSIGNED, tempReg1, srcReg, dec_->nrmoff);
LDRB(INDEX_UNSIGNED, tempReg2, srcReg, dec_->nrmoff + 1);
LDRB(INDEX_UNSIGNED, tempReg3, srcReg, dec_->nrmoff + 2);
ORR(tempReg1, tempReg1, tempReg2, ArithOption(tempReg2, ST_LSL, 8));
ORR(tempReg1, tempReg1, tempReg3, ArithOption(tempReg3, ST_LSL, 16));
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.nrmoff);
}
// Copy 6 bytes and then 2 zeroes.
void VertexDecoderJitCache::Jit_NormalS16() {
LDRH(INDEX_UNSIGNED, tempReg1, srcReg, dec_->nrmoff);
LDRH(INDEX_UNSIGNED, tempReg2, srcReg, dec_->nrmoff + 2);
LDRH(INDEX_UNSIGNED, tempReg3, srcReg, dec_->nrmoff + 4);
ORR(tempReg1, tempReg1, tempReg2, ArithOption(tempReg2, ST_LSL, 16));
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.nrmoff);
STR(INDEX_UNSIGNED, tempReg3, dstReg, dec_->decFmt.nrmoff + 4);
}
void VertexDecoderJitCache::Jit_NormalFloat() {
LDR(INDEX_UNSIGNED, tempReg1, srcReg, dec_->nrmoff);
LDR(INDEX_UNSIGNED, tempReg2, srcReg, dec_->nrmoff + 4);
LDR(INDEX_UNSIGNED, tempReg3, srcReg, dec_->nrmoff + 8);
STR(INDEX_UNSIGNED, tempReg1, dstReg, dec_->decFmt.nrmoff);
STR(INDEX_UNSIGNED, tempReg2, dstReg, dec_->decFmt.nrmoff + 4);
STR(INDEX_UNSIGNED, tempReg3, dstReg, dec_->decFmt.nrmoff + 8);
}
void VertexDecoderJitCache::Jit_AnyS8ToFloat(int srcoff) {
// TODO: NEONize. In that case we'll leave all three floats in one register instead, so callers must change too.
LDRSB(INDEX_UNSIGNED, tempReg1, srcReg, srcoff);
LDRSB(INDEX_UNSIGNED, tempReg2, srcReg, srcoff + 1);
LDRSB(INDEX_UNSIGNED, tempReg3, srcReg, srcoff + 2);
fp.SCVTF(src[0], tempReg1, 7);
fp.SCVTF(src[1], tempReg2, 7);
fp.SCVTF(src[2], tempReg3, 7);
}
void VertexDecoderJitCache::Jit_AnyS16ToFloat(int srcoff) {
LDRSH(INDEX_UNSIGNED, tempReg1, srcReg, srcoff);
LDRSH(INDEX_UNSIGNED, tempReg2, srcReg, srcoff + 2);
LDRSH(INDEX_UNSIGNED, tempReg3, srcReg, srcoff + 4);
fp.SCVTF(src[0], tempReg1, 15);
fp.SCVTF(src[1], tempReg2, 15);
fp.SCVTF(src[2], tempReg3, 15);
}

9
GPU/Common/VertexDecoderCommon.cpp
View file

@ -134,7 +134,8 @@ void PrintDecodedVertex(VertexReader &vtx) {
printf("P: %f %f %f\n", pos[0], pos[1], pos[2]); printf("P: %f %f %f\n", pos[0], pos[1], pos[2]);
} }
VertexDecoder::VertexDecoder() : jitted_(0), decoded_(nullptr), ptr_(nullptr) { VertexDecoder::VertexDecoder() : jitted_(0), decoded_(nullptr), ptr_(nullptr)
{
} }
void VertexDecoder::Step_WeightsU8() const void VertexDecoder::Step_WeightsU8() const
@ -1093,7 +1094,11 @@ int VertexDecoder::ToString(char *output) const {
return output - start; return output - start;
} }
VertexDecoderJitCache::VertexDecoderJitCache() { VertexDecoderJitCache::VertexDecoderJitCache()
#ifdef ARM64
: fp(this)
#endif
{
// 256k should be enough. // 256k should be enough.
AllocCodeSpace(1024 * 64 * 4); AllocCodeSpace(1024 * 64 * 4);

3
GPU/Common/VertexDecoderCommon.h
View file

@ -683,4 +683,7 @@ private:
void Jit_AnyFloatMorph(int srcoff, int dstoff); void Jit_AnyFloatMorph(int srcoff, int dstoff);
const VertexDecoder *dec_; const VertexDecoder *dec_;
#ifdef ARM64
Arm64Gen::ARM64FloatEmitter fp;
#endif
}; };