ShuriZma
/
pcsx2
mirror of https://github.com/PCSX2/pcsx2.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

core: purge sse2

This commit is contained in:
Gauvain 'GovanifY' Roussel-Tarbouriech 2021-03-27 09:18:02 +01:00 committed by lightningterror
parent 5509bfc003
commit 8a9ec4c706
10 changed files with 71 additions and 461 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
pcsx2/gui/AppInit.cpp
View File

@ -46,13 +46,13 @@ void Pcsx2App::DetectCpuAndUserMode()
x86caps.CountCores();
x86caps.SIMD_EstablishMXCSRmask();
if (!x86caps.hasStreamingSIMD2Extensions)
if (!x86caps.hasStreamingSIMD4Extensions)
{
// This code will probably never run if the binary was correctly compiled for SSE2
// SSE2 is required for any decent speed and is supported by more than decade old x86 CPUs
// This code will probably never run if the binary was correctly compiled for SSE4
// SSE4 is required for any decent speed and is supported by more than decade old x86 CPUs
throw Exception::HardwareDeficiency()
.SetDiagMsg(L"Critical Failure: SSE2 Extensions not available.")
.SetUserMsg(_("SSE2 extensions are not available. PCSX2 requires a cpu that supports the SSE2 instruction set."));
.SetDiagMsg(L"Critical Failure: SSE4 Extensions not available.")
.SetUserMsg(_("SSE4 extensions are not available. PCSX2 requires a cpu that supports the SSE4 instruction set."));
}
#endif

3
pcsx2/gui/Panels/PluginSelectorPanel.cpp
View File

@ -684,7 +684,6 @@ void Panels::PluginSelectorPanel::OnEnumComplete( wxCommandEvent& evt )
int index_avx2 = -1;
int index_sse4 = -1;
int index_sse2 = -1;
for( int i = 0; i < count; i++ )
{
@ -692,12 +691,10 @@ void Panels::PluginSelectorPanel::OnEnumComplete( wxCommandEvent& evt )
if( x86caps.hasAVX2 && str.Contains("AVX2") ) index_avx2 = i;
if( x86caps.hasStreamingSIMD4Extensions && str.Contains("SSE4") ) index_sse4 = i;
if( str.Contains("SSE2") ) index_sse2 = i;
}
if( index_avx2 >= 0 ) m_ComponentBoxes->Get(pid).SetSelection( index_avx2 );
else if( index_sse4 >= 0 ) m_ComponentBoxes->Get(pid).SetSelection( index_sse4 );
else if( index_sse2 >= 0 ) m_ComponentBoxes->Get(pid).SetSelection( index_sse2 );
else m_ComponentBoxes->Get(pid).SetSelection( 0 );
}
else

162
pcsx2/x86/iMMI.cpp
View File

@ -215,17 +215,9 @@ void recPMTHL()
int info = eeRecompileCodeXMM( XMMINFO_READS|XMMINFO_READLO|XMMINFO_READHI|XMMINFO_WRITELO|XMMINFO_WRITEHI );
if ( x86caps.hasStreamingSIMD4Extensions ) {
xBLEND.PS(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_S), 0x5);
xSHUF.PS(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_S), 0xdd);
xSHUF.PS(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0x72);
}
else {
xSHUF.PS(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_S), 0x8d);
xSHUF.PS(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_S), 0xdd);
xSHUF.PS(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO), 0x72);
xSHUF.PS(xRegisterSSE(EEREC_HI), xRegisterSSE(EEREC_HI), 0x72);
}
_clearNeededXMMregs();
}
@ -400,7 +392,6 @@ void recPMAXW()
EE::Profiler.EmitOp(eeOpcode::PMAXW);
int info = eeRecompileCodeXMM( XMMINFO_READS|XMMINFO_READT|XMMINFO_WRITED );
if ( x86caps.hasStreamingSIMD4Extensions ) {
if( EEREC_S == EEREC_T ) xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
else if( EEREC_D == EEREC_S ) xPMAX.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
else if ( EEREC_D == EEREC_T ) xPMAX.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
@ -408,40 +399,6 @@ void recPMAXW()
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPMAX.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
}
}
else {
int t0reg;
if( EEREC_S == EEREC_T ) {
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
}
else {
t0reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_S));
xPCMP.GTD(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
if( EEREC_D == EEREC_S ) {
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
}
else if( EEREC_D == EEREC_T ) {
int t1reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t1reg), xRegisterSSE(EEREC_T));
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(t1reg));
_freeXMMreg(t1reg);
}
else {
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
}
xPOR(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
}
_clearNeededXMMregs();
}
@ -1173,18 +1130,7 @@ void recPABSW() //needs clamping
xPCMP.EQD(xRegisterSSE(t0reg), xRegisterSSE(t0reg));
xPSLL.D(xRegisterSSE(t0reg), 31);
xPCMP.EQD(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T)); //0xffffffff if equal to 0x80000000
if( x86caps.hasSupplementalStreamingSIMD3Extensions ) {
xPABS.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T)); //0x80000000 -> 0x80000000
}
else {
int t1reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t1reg), xRegisterSSE(EEREC_T));
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
xPSRA.D(xRegisterSSE(t1reg), 31);
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xPSUB.D(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg)); //0x80000000 -> 0x80000000
_freeXMMreg(t1reg);
}
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg)); //0x80000000 -> 0x7fffffff
_freeXMMreg(t0reg);
_clearNeededXMMregs();
@ -1203,18 +1149,7 @@ void recPABSH()
xPCMP.EQW(xRegisterSSE(t0reg), xRegisterSSE(t0reg));
xPSLL.W(xRegisterSSE(t0reg), 15);
xPCMP.EQW(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T)); //0xffff if equal to 0x8000
if( x86caps.hasSupplementalStreamingSIMD3Extensions ) {
xPABS.W(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T)); //0x8000 -> 0x8000
}
else {
int t1reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t1reg), xRegisterSSE(EEREC_T));
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
xPSRA.W(xRegisterSSE(t1reg), 15);
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xPSUB.W(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg)); //0x8000 -> 0x8000
_freeXMMreg(t1reg);
}
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg)); //0x8000 -> 0x7fff
_freeXMMreg(t0reg);
_clearNeededXMMregs();
@ -1228,7 +1163,6 @@ void recPMINW()
EE::Profiler.EmitOp(eeOpcode::PMINW);
int info = eeRecompileCodeXMM( XMMINFO_READS|XMMINFO_READT|XMMINFO_WRITED );
if ( x86caps.hasStreamingSIMD4Extensions ) {
if( EEREC_S == EEREC_T ) xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
else if( EEREC_D == EEREC_S ) xPMIN.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
else if ( EEREC_D == EEREC_T ) xPMIN.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
@ -1236,40 +1170,6 @@ void recPMINW()
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPMIN.SD(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T));
}
}
else {
int t0reg;
if( EEREC_S == EEREC_T ) {
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
}
else {
t0reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
xPCMP.GTD(xRegisterSSE(t0reg), xRegisterSSE(EEREC_S));
if( EEREC_D == EEREC_S ) {
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
}
else if( EEREC_D == EEREC_T ) {
int t1reg = _allocTempXMMreg(XMMT_INT, -1);
xMOVDQA(xRegisterSSE(t1reg), xRegisterSSE(EEREC_T));
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(t1reg));
_freeXMMreg(t1reg);
}
else {
xMOVDQA(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_S));
xPAND(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
xPANDN(xRegisterSSE(t0reg), xRegisterSSE(EEREC_T));
}
xPOR(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
}
_clearNeededXMMregs();
}
@ -1718,12 +1618,6 @@ void recPMADDW()
{
EE::Profiler.EmitOp(eeOpcode::PMADDW);
if( !x86caps.hasStreamingSIMD4Extensions ) {
_deleteEEreg(_Rd_, 0);
recCall(Interp::PMADDW);
return;
}
int info = eeRecompileCodeXMM( (((_Rs_)&&(_Rt_))?XMMINFO_READS:0)|(((_Rs_)&&(_Rt_))?XMMINFO_READT:0)|(_Rd_?XMMINFO_WRITED:0)|XMMINFO_WRITELO|XMMINFO_WRITEHI|XMMINFO_READLO|XMMINFO_READHI );
xSHUF.PS(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88);
xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO), 0xd8); // LO = {LO[0], HI[0], LO[2], HI[2]}
@ -1775,19 +1669,9 @@ void recPSLLVW()
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
if ( x86caps.hasStreamingSIMD4Extensions ) {
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
int t0reg = _allocTempXMMreg(XMMT_INT, -1);
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31);
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
}
}
else if( _Rt_ == 0 ) {
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
@ -1813,16 +1697,8 @@ void recPSLLVW()
xPSLL.D(xRegisterSSE(t1reg), xRegisterSSE(t0reg));
// merge & sign extend
if ( x86caps.hasStreamingSIMD4Extensions ) {
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31); // get the signs
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
}
_freeXMMreg(t0reg);
_freeXMMreg(t1reg);
@ -1843,19 +1719,9 @@ void recPSRLVW()
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
if ( x86caps.hasStreamingSIMD4Extensions ) {
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
int t0reg = _allocTempXMMreg(XMMT_INT, -1);
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31);
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
}
}
else if( _Rt_ == 0 ) {
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
@ -1881,16 +1747,8 @@ void recPSRLVW()
xPSRL.D(xRegisterSSE(t1reg), xRegisterSSE(t0reg));
// merge & sign extend
if ( x86caps.hasStreamingSIMD4Extensions ) {
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t1reg));
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31); // get the signs
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
}
_freeXMMreg(t0reg);
_freeXMMreg(t1reg);
@ -1903,11 +1761,6 @@ void recPMSUBW()
{
EE::Profiler.EmitOp(eeOpcode::PMSUBW);
if( !x86caps.hasStreamingSIMD4Extensions ) {
_deleteEEreg(_Rd_, 0);
recCall(Interp::PMSUBW);
return;
}
int info = eeRecompileCodeXMM( (((_Rs_)&&(_Rt_))?XMMINFO_READS:0)|(((_Rs_)&&(_Rt_))?XMMINFO_READT:0)|(_Rd_?XMMINFO_WRITED:0)|XMMINFO_WRITELO|XMMINFO_WRITEHI|XMMINFO_READLO|XMMINFO_READHI );
xSHUF.PS(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_HI), 0x88);
xPSHUF.D(xRegisterSSE(EEREC_LO), xRegisterSSE(EEREC_LO), 0xd8); // LO = {LO[0], HI[0], LO[2], HI[2]}
@ -1957,11 +1810,6 @@ void recPMULTW()
{
EE::Profiler.EmitOp(eeOpcode::PMULTW);
if( !x86caps.hasStreamingSIMD4Extensions ) {
_deleteEEreg(_Rd_, 0);
recCall(Interp::PMULTW);
return;
}
int info = eeRecompileCodeXMM( (((_Rs_)&&(_Rt_))?XMMINFO_READS:0)|(((_Rs_)&&(_Rt_))?XMMINFO_READT:0)|(_Rd_?XMMINFO_WRITED:0)|XMMINFO_WRITELO|XMMINFO_WRITEHI );
if( !_Rs_ || !_Rt_ ) {
if( _Rd_ ) xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
@ -2455,19 +2303,9 @@ void recPSRAVW()
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
if ( x86caps.hasStreamingSIMD4Extensions ) {
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xPMOVSX.DQ(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));
}
else {
int t0reg = _allocTempXMMreg(XMMT_INT, -1);
xPSHUF.D(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_T), 0x88);
xMOVDQA(xRegisterSSE(t0reg), xRegisterSSE(EEREC_D));
xPSRA.D(xRegisterSSE(t0reg), 31);
xPUNPCK.LDQ(xRegisterSSE(EEREC_D), xRegisterSSE(t0reg));
_freeXMMreg(t0reg);
}
}
}
else if( _Rt_ == 0 ) {
xPXOR(xRegisterSSE(EEREC_D), xRegisterSSE(EEREC_D));

4
pcsx2/x86/ix86-32/iR5900-32.cpp
View File

@ -508,8 +508,8 @@ static void recReserve()
{
// Hardware Requirements Check...
if ( !x86caps.hasStreamingSIMD2Extensions )
recThrowHardwareDeficiency( L"SSE2" );
if ( !x86caps.hasStreamingSIMD4Extensions )
recThrowHardwareDeficiency( L"SSE4" );
recReserveCache();
}

2
pcsx2/x86/microVU.cpp
View File

@ -47,7 +47,7 @@ void mVUreserveCache(microVU& mVU) {
// Only run this once per VU! ;)
void mVUinit(microVU& mVU, uint vuIndex) {
if(!x86caps.hasStreamingSIMD2Extensions) mVUthrowHardwareDeficiency( L"SSE2", vuIndex );
if(!x86caps.hasStreamingSIMD4Extensions) mVUthrowHardwareDeficiency( L"SSE4", vuIndex );
memzero(mVU.prog);

10
pcsx2/x86/microVU_Alloc.inl
View File

@ -166,13 +166,7 @@ __fi void getQreg(const xmm& reg, int qInstance)
__ri void writeQreg(const xmm& reg, int qInstance)
{
if (qInstance) {
if (!x86caps.hasStreamingSIMD4Extensions) {
xPSHUF.D(xmmPQ, xmmPQ, 0xe1);
xMOVSS(xmmPQ, reg);
xPSHUF.D(xmmPQ, xmmPQ, 0xe1);
}
else xINSERTPS(xmmPQ, reg, _MM_MK_INSERTPS_NDX(0, 1, 0));
}
if (qInstance)
xINSERTPS(xmmPQ, reg, _MM_MK_INSERTPS_NDX(0, 1, 0));
else xMOVSS(xmmPQ, reg);
}

23
pcsx2/x86/microVU_Clamp.inl
View File

@ -56,34 +56,11 @@ void mVUclamp1(const xmm& reg, const xmm& regT1, int xyzw, bool bClampE = 0) {
// so we just use a temporary mem location for our backup for now... (non-sse4 version only)
void mVUclamp2(microVU& mVU, const xmm& reg, const xmm& regT1in, int xyzw, bool bClampE = 0) {
if ((!clampE && CHECK_VU_SIGN_OVERFLOW) || (clampE && bClampE && CHECK_VU_SIGN_OVERFLOW)) {
if (x86caps.hasStreamingSIMD4Extensions) {
int i = (xyzw==1||xyzw==2||xyzw==4||xyzw==8) ? 0: 1;
xPMIN.SD(reg, ptr128[&sse4_maxvals[i][0]]);
xPMIN.UD(reg, ptr128[&sse4_minvals[i][0]]);
return;
}
//const xmm& regT1 = regT1b ? mVU.regAlloc->allocReg() : regT1in;
const xmm& regT1 = regT1in.IsEmpty() ? xmm((reg.Id + 1) % 8) : regT1in;
if (regT1 != regT1in) xMOVAPS(ptr128[mVU.xmmCTemp], regT1);
switch (xyzw) {
case 1: case 2: case 4: case 8:
xMOVAPS(regT1, reg);
xAND.PS(regT1, ptr128[mVUglob.signbit]);
xMIN.SS(reg, ptr128[mVUglob.maxvals]);
xMAX.SS(reg, ptr128[mVUglob.minvals]);
xOR.PS (reg, regT1);
break;
default:
xMOVAPS(regT1, reg);
xAND.PS(regT1, ptr128[mVUglob.signbit]);
xMIN.PS(reg, ptr128[mVUglob.maxvals]);
xMAX.PS(reg, ptr128[mVUglob.minvals]);
xOR.PS (reg, regT1);
break;
}
//if (regT1 != regT1in) mVU.regAlloc->clearNeeded(regT1);
if (regT1 != regT1in) xMOVAPS(regT1, ptr128[mVU.xmmCTemp]);
}
else mVUclamp1(reg, regT1in, xyzw, bClampE);
}

16
pcsx2/x86/microVU_Lower.inl
View File

@ -28,11 +28,7 @@ static __fi void testZero(const xmm& xmmReg, const xmm& xmmTemp, const x32& gprT
{
xXOR.PS(xmmTemp, xmmTemp);
xCMPEQ.SS(xmmTemp, xmmReg);
if (!x86caps.hasStreamingSIMD4Extensions) {
xMOVMSKPS(gprTemp, xmmTemp);
xTEST(gprTemp, 1);
}
else xPTEST(xmmTemp, xmmTemp);
xPTEST(xmmTemp, xmmTemp);
}
// Test if Vector is Negative (Set Flags and Makes Positive)
@ -298,19 +294,9 @@ mVUop(mVU_EEXP) {
// sumXYZ(): PQ.x = x ^ 2 + y ^ 2 + z ^ 2
static __fi void mVU_sumXYZ(mV, const xmm& PQ, const xmm& Fs) {
if (x86caps.hasStreamingSIMD4Extensions) {
xDP.PS(Fs, Fs, 0x71);
xMOVSS(PQ, Fs);
}
else {
SSE_MULPS(mVU, Fs, Fs); // wzyx ^ 2
xMOVSS (PQ, Fs); // x ^ 2
xPSHUF.D (Fs, Fs, 0xe1); // wzyx -> wzxy
SSE_ADDSS(mVU, PQ, Fs); // x ^ 2 + y ^ 2
xPSHUF.D (Fs, Fs, 0xd2); // wzxy -> wxyz
SSE_ADDSS(mVU, PQ, Fs); // x ^ 2 + y ^ 2 + z ^ 2
}
}
mVUop(mVU_ELENG) {
pass1 { mVUanalyzeEFU2(mVU, _Fs_, 18); }

115
pcsx2/x86/microVU_Misc.inl
View File

@ -59,72 +59,29 @@ void mVUsaveReg(const xmm& reg, xAddressVoid ptr, int xyzw, bool modXYZW)
return;*/
switch ( xyzw ) {
case 5: if (x86caps.hasStreamingSIMD4Extensions) {
xEXTRACTPS(ptr32[ptr+4], reg, 1);
case 5: xEXTRACTPS(ptr32[ptr+4], reg, 1);
xEXTRACTPS(ptr32[ptr+12], reg, 3);
}
else {
xPSHUF.D(reg, reg, 0xe1); //WZXY
xMOVSS(ptr32[ptr+4], reg);
xPSHUF.D(reg, reg, 0xff); //WWWW
xMOVSS(ptr32[ptr+12], reg);
}
break; // YW
case 6: xPSHUF.D(reg, reg, 0xc9);
xMOVL.PS(ptr64[ptr+4], reg);
break; // YZ
case 7: if (x86caps.hasStreamingSIMD4Extensions) {
xMOVH.PS(ptr64[ptr+8], reg);
case 7: xMOVH.PS(ptr64[ptr+8], reg);
xEXTRACTPS(ptr32[ptr+4], reg, 1);
}
else {
xPSHUF.D(reg, reg, 0x93); //ZYXW
xMOVH.PS(ptr64[ptr+4], reg);
xMOVSS(ptr32[ptr+12], reg);
}
break; // YZW
case 9: if (x86caps.hasStreamingSIMD4Extensions) {
xMOVSS(ptr32[ptr], reg);
case 9: xMOVSS(ptr32[ptr], reg);
xEXTRACTPS(ptr32[ptr+12], reg, 3);
}
else {
xMOVSS(ptr32[ptr], reg);
xPSHUF.D(reg, reg, 0xff); //WWWW
xMOVSS(ptr32[ptr+12], reg);
}
break; // XW
case 10: if (x86caps.hasStreamingSIMD4Extensions) {
xMOVSS(ptr32[ptr], reg);
case 10: xMOVSS(ptr32[ptr], reg);
xEXTRACTPS(ptr32[ptr+8], reg, 2);
}
else {
xMOVSS(ptr32[ptr], reg);
xMOVHL.PS(reg, reg);
xMOVSS(ptr32[ptr+8], reg);
}
break; //XZ
case 11: xMOVSS(ptr32[ptr], reg);
xMOVH.PS(ptr64[ptr+8], reg);
break; //XZW
case 13: if (x86caps.hasStreamingSIMD4Extensions) {
xMOVL.PS(ptr64[ptr], reg);
case 13: xMOVL.PS(ptr64[ptr], reg);
xEXTRACTPS(ptr32[ptr+12], reg, 3);
}
else {
xPSHUF.D(reg, reg, 0x4b); //YXZW
xMOVH.PS(ptr64[ptr], reg);
xMOVSS(ptr32[ptr+12], reg);
}
break; // XYW
case 14: if (x86caps.hasStreamingSIMD4Extensions) {
xMOVL.PS(ptr64[ptr], reg);
case 14: xMOVL.PS(ptr64[ptr], reg);
xEXTRACTPS(ptr32[ptr+8], reg, 2);
}
else {
xMOVL.PS(ptr64[ptr], reg);
xMOVHL.PS(reg, reg);
xMOVSS(ptr32[ptr+8], reg);
}
break; // XYZ
case 4: if (!modXYZW) mVUunpack_xyzw(reg, reg, 1);
xMOVSS(ptr32[ptr+4], reg);
@ -146,8 +103,14 @@ void mVUsaveReg(const xmm& reg, xAddressVoid ptr, int xyzw, bool modXYZW)
void mVUmergeRegs(const xmm& dest, const xmm& src, int xyzw, bool modXYZW)
{
xyzw &= 0xf;
if ( (dest != src) && (xyzw != 0) ) {
if (x86caps.hasStreamingSIMD4Extensions && (xyzw != 0x8) && (xyzw != 0xf)) {
if ( (dest != src) && (xyzw != 0) )
{
if (xyzw == 0x8)
xMOVSS(dest, src);
else if (xyzw == 0xf)
xMOVAPS(dest, src);
else
{
if (modXYZW) {
if (xyzw == 1) { xINSERTPS(dest, src, _MM_MK_INSERTPS_NDX(0, 3, 0)); return; }
else if (xyzw == 2) { xINSERTPS(dest, src, _MM_MK_INSERTPS_NDX(0, 2, 0)); return; }
@ -156,56 +119,6 @@ void mVUmergeRegs(const xmm& dest, const xmm& src, int xyzw, bool modXYZW)
xyzw = ((xyzw & 1) << 3) | ((xyzw & 2) << 1) | ((xyzw & 4) >> 1) | ((xyzw & 8) >> 3);
xBLEND.PS(dest, src, xyzw);
}
else {
switch (xyzw) {
case 1: if (modXYZW) mVUunpack_xyzw(src, src, 0);
xMOVHL.PS(src, dest); // src = Sw Sz Dw Dz
xSHUF.PS(dest, src, 0xc4); // 11 00 01 00
break;
case 2: if (modXYZW) mVUunpack_xyzw(src, src, 0);
xMOVHL.PS(src, dest);
xSHUF.PS(dest, src, 0x64);
break;
case 3: xSHUF.PS(dest, src, 0xe4);
break;
case 4: if (modXYZW) mVUunpack_xyzw(src, src, 0);
xMOVSS(src, dest);
xMOVSD(dest, src);
break;
case 5: xSHUF.PS(dest, src, 0xd8);
xPSHUF.D(dest, dest, 0xd8);
break;
case 6: xSHUF.PS(dest, src, 0x9c);
xPSHUF.D(dest, dest, 0x78);
break;
case 7: xMOVSS(src, dest);
xMOVAPS(dest, src);
break;
case 8: xMOVSS(dest, src);
break;
case 9: xSHUF.PS(dest, src, 0xc9);
xPSHUF.D(dest, dest, 0xd2);
break;
case 10: xSHUF.PS(dest, src, 0x8d);
xPSHUF.D(dest, dest, 0x72);
break;
case 11: xMOVSS(dest, src);
xSHUF.PS(dest, src, 0xe4);
break;
case 12: xMOVSD(dest, src);
break;
case 13: xMOVHL.PS(dest, src);
xSHUF.PS(src, dest, 0x64);
xMOVAPS(dest, src);
break;
case 14: xMOVHL.PS(dest, src);
xSHUF.PS(src, dest, 0xc4);
xMOVAPS(dest, src);
break;
default: xMOVAPS(dest, src);
break;
}
}
}
}

97
pcsx2/x86/newVif_UnpackSSE.cpp
View File

@ -35,24 +35,8 @@ static RecompiledCodeReserve* nVifUpkExec = NULL;
// Merges xmm vectors without modifying source reg
void mergeVectors(xRegisterSSE dest, xRegisterSSE src, xRegisterSSE temp, int xyzw) {
if (x86caps.hasStreamingSIMD4Extensions || (xyzw==15)
|| (xyzw==12) || (xyzw==11) || (xyzw==8) || (xyzw==3)) {
mVUmergeRegs(dest, src, xyzw);
}
else
{
if(temp != src) xMOVAPS(temp, src); //Sometimes we don't care if the source is modified and is temp reg.
if(dest == temp)
{
//VIF can sent the temp directory as the source and destination, just need to clear the ones we dont want in which case.
if(!(xyzw & 0x1)) xAND.PS( dest, ptr128[SSEXYZWMask[0]]);
if(!(xyzw & 0x2)) xAND.PS( dest, ptr128[SSEXYZWMask[1]]);
if(!(xyzw & 0x4)) xAND.PS( dest, ptr128[SSEXYZWMask[2]]);
if(!(xyzw & 0x8)) xAND.PS( dest, ptr128[SSEXYZWMask[3]]);
}
else mVUmergeRegs(dest, temp, xyzw);
}
}
// =====================================================================================================
// VifUnpackSSE_Base Section
@ -113,16 +97,6 @@ void VifUnpackSSE_Base::xUPK_S_32() const {
void VifUnpackSSE_Base::xUPK_S_16() const {
if (!x86caps.hasStreamingSIMD4Extensions)
{
xMOV16 (workReg, ptr32[srcIndirect]);
xPUNPCK.LWD(workReg, workReg);
xShiftR (workReg, 16);
xPSHUF.D (destReg, workReg, _v0);
return;
}
switch(UnpkLoopIteration)
{
case 0:
@ -144,17 +118,6 @@ void VifUnpackSSE_Base::xUPK_S_16() const {
void VifUnpackSSE_Base::xUPK_S_8() const {
if (!x86caps.hasStreamingSIMD4Extensions)
{
xMOV8 (workReg, ptr32[srcIndirect]);
xPUNPCK.LBW(workReg, workReg);
xPUNPCK.LWD(workReg, workReg);
xShiftR (workReg, 24);
xPSHUF.D (destReg, workReg, _v0);
return;
}
switch(UnpkLoopIteration)
{
case 0:
@ -199,18 +162,8 @@ void VifUnpackSSE_Base::xUPK_V2_32() const {
void VifUnpackSSE_Base::xUPK_V2_16() const {
if(UnpkLoopIteration == 0)
{
if (x86caps.hasStreamingSIMD4Extensions)
{
xPMOVXX16 (workReg);
}
else
{
xMOV64 (workReg, ptr64[srcIndirect]);
xPUNPCK.LWD(workReg, workReg);
xShiftR (workReg, 16);
}
xPSHUF.D (destReg, workReg, 0x44); //v1v0v1v0
}
else
@ -223,19 +176,9 @@ void VifUnpackSSE_Base::xUPK_V2_16() const {
void VifUnpackSSE_Base::xUPK_V2_8() const {
if(UnpkLoopIteration == 0 || !x86caps.hasStreamingSIMD4Extensions)
{
if (x86caps.hasStreamingSIMD4Extensions)
if(UnpkLoopIteration == 0)
{
xPMOVXX8 (workReg);
}
else
{
xMOV16 (workReg, ptr32[srcIndirect]);
xPUNPCK.LBW(workReg, workReg);
xPUNPCK.LWD(workReg, workReg);
xShiftR (workReg, 24);
}
xPSHUF.D (destReg, workReg, 0x44); //v1v0v1v0
}
else
@ -254,16 +197,7 @@ void VifUnpackSSE_Base::xUPK_V3_32() const {
void VifUnpackSSE_Base::xUPK_V3_16() const {
if (x86caps.hasStreamingSIMD4Extensions)
{
xPMOVXX16 (destReg);
}
else
{
xMOV64 (destReg, ptr32[srcIndirect]);
xPUNPCK.LWD(destReg, destReg);
xShiftR (destReg, 16);
}
//With V3-16, it takes the first vector from the next position as the W vector
//However - IF the end of this iteration of the unpack falls on a quadword boundary, W becomes 0
@ -278,17 +212,7 @@ void VifUnpackSSE_Base::xUPK_V3_16() const {
void VifUnpackSSE_Base::xUPK_V3_8() const {
if (x86caps.hasStreamingSIMD4Extensions)
{
xPMOVXX8 (destReg);
}
else
{
xMOV32 (destReg, ptr32[srcIndirect]);
xPUNPCK.LBW(destReg, destReg);
xPUNPCK.LWD(destReg, destReg);
xShiftR (destReg, 24);
}
if (UnpkLoopIteration != IsAligned)
xAND.PS(destReg, ptr128[SSEXYZWMask[0]]);
}
@ -300,32 +224,13 @@ void VifUnpackSSE_Base::xUPK_V4_32() const {
void VifUnpackSSE_Base::xUPK_V4_16() const {
if (x86caps.hasStreamingSIMD4Extensions)
{
xPMOVXX16 (destReg);
}
else
{
xMOV64 (destReg, ptr32[srcIndirect]);
xPUNPCK.LWD(destReg, destReg);
xShiftR (destReg, 16);
}
}
void VifUnpackSSE_Base::xUPK_V4_8() const {
if (x86caps.hasStreamingSIMD4Extensions)
{
xPMOVXX8 (destReg);
}
else
{
xMOV32 (destReg, ptr32[srcIndirect]);
xPUNPCK.LBW(destReg, destReg);
xPUNPCK.LWD(destReg, destReg);
xShiftR (destReg, 24);
}
}
void VifUnpackSSE_Base::xUPK_V4_5() const {