The-Homebrew-Cloud/dolphin
2
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-02-13 15:59:23 +01:00
Code Issues Packages Projects Releases Wiki Activity

Some fixes to JIT IL's quantized stores

Browse Source 
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@3310 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
LinesPrower 2009-06-03 04:43:34 +00:00
parent c391e9659a
commit 3c953e485c
1 changed files with 40 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

46
Source/Core/Core/Src/PowerPC/Jit64IL/JitAsm.cpp
View File

@ -211,6 +211,15 @@ const float m_dequantizeTableS[] =
float psTemp[2]; float psTemp[2];
const float m_65535 = 65535.0f;
#define QUANTIZE_OVERFLOW_SAFE
// according to Intel Docs CVTPS2DQ writes 0x80000000 if the source floating point value is out of int32 range
// while it's OK for large negatives, it isn't for positives
// I don't know whether the overflow actually happens in any games
// but it potentially can cause problems, so we need some clamping
void AsmRoutineManager::GenQuantizedStores() { void AsmRoutineManager::GenQuantizedStores() {
const u8* storePairedIllegal = AlignCode4(); const u8* storePairedIllegal = AlignCode4();
UD2(); UD2();
@ -253,6 +262,11 @@ void AsmRoutineManager::GenQuantizedStores() {
MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS)); MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS));
PUNPCKLDQ(XMM1, R(XMM1)); PUNPCKLDQ(XMM1, R(XMM1));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE
MOVSS(XMM1, M((void *)&m_65535));
PUNPCKLDQ(XMM1, R(XMM1));
MINPS(XMM0, R(XMM1));
#endif
CVTPS2DQ(XMM0, R(XMM0)); CVTPS2DQ(XMM0, R(XMM0));
PACKSSDW(XMM0, R(XMM0)); PACKSSDW(XMM0, R(XMM0));
PACKUSWB(XMM0, R(XMM0)); PACKUSWB(XMM0, R(XMM0));
@ -270,6 +284,11 @@ void AsmRoutineManager::GenQuantizedStores() {
MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS)); MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS));
PUNPCKLDQ(XMM1, R(XMM1)); PUNPCKLDQ(XMM1, R(XMM1));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE
MOVSS(XMM1, M((void *)&m_65535));
PUNPCKLDQ(XMM1, R(XMM1));
MINPS(XMM0, R(XMM1));
#endif
CVTPS2DQ(XMM0, R(XMM0)); CVTPS2DQ(XMM0, R(XMM0));
PACKSSDW(XMM0, R(XMM0)); PACKSSDW(XMM0, R(XMM0));
PACKSSWB(XMM0, R(XMM0)); PACKSSWB(XMM0, R(XMM0));
@ -287,14 +306,24 @@ void AsmRoutineManager::GenQuantizedStores() {
MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS)); MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS));
PUNPCKLDQ(XMM1, R(XMM1)); PUNPCKLDQ(XMM1, R(XMM1));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
CVTPS2DQ(XMM0, R(XMM0));
// PACKUSDW is available only in SSE4
PXOR(XMM1, R(XMM1)); PXOR(XMM1, R(XMM1));
PCMPGTD(XMM1, R(XMM0)); MAXPS(XMM0, R(XMM1));
PANDN(XMM0, R(XMM1)); MOVSS(XMM1, M((void *)&m_65535));
PACKSSDW(XMM0, R(XMM0)); //PACKUSDW(XMM0, R(XMM0)); // FIXME: Wrong! PUNPCKLDQ(XMM1, R(XMM1));
MOVD_xmm(R(EAX), XMM0); MINPS(XMM0, R(XMM1));
CVTPS2DQ(XMM0, R(XMM0));
MOVQ_xmm(M(psTemp), XMM0);
// place ps[0] into the higher word, ps[1] into the lower
// so no need in ROL after BSWAP
MOVZX(32, 16, EAX, M((char*)psTemp + 0));
SHL(32, R(EAX), Imm8(16));
MOV(16, R(AX), M((char*)psTemp + 4));
BSWAP(32, EAX); BSWAP(32, EAX);
ROL(32, R(EAX), Imm8(16)); //ROL(32, R(EAX), Imm8(16));
#ifdef _M_X64 #ifdef _M_X64
MOV(32, MComplex(RBX, RCX, 1, 0), R(EAX)); MOV(32, MComplex(RBX, RCX, 1, 0), R(EAX));
#else #else
@ -308,6 +337,11 @@ void AsmRoutineManager::GenQuantizedStores() {
MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS)); MOVSS(XMM1, MDisp(EAX, (u32)(u64)m_quantizeTableS));
PUNPCKLDQ(XMM1, R(XMM1)); PUNPCKLDQ(XMM1, R(XMM1));
MULPS(XMM0, R(XMM1)); MULPS(XMM0, R(XMM1));
#ifdef QUANTIZE_OVERFLOW_SAFE
MOVSS(XMM1, M((void *)&m_65535));
PUNPCKLDQ(XMM1, R(XMM1));
MINPS(XMM0, R(XMM1));
#endif
CVTPS2DQ(XMM0, R(XMM0)); CVTPS2DQ(XMM0, R(XMM0));
PACKSSDW(XMM0, R(XMM0)); PACKSSDW(XMM0, R(XMM0));
MOVD_xmm(R(EAX), XMM0); MOVD_xmm(R(EAX), XMM0);