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

Add commented-out code to save XMM4/XMM5.

Browse source 
According to all calling convention manuals I can find, we don't really
need to preserve them. If they become problematic as mentioned, we can
activate this.
This commit is contained in:
Henrik Rydgard 2013-11-07 09:54:58 +01:00
parent b203da05e9
commit c4e3dd14fd
1 changed files with 21 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

30
GPU/GLES/VertexDecoder.cpp
View file

@ -1235,6 +1235,7 @@ static const X64Reg srcReg = ESI;
static const X64Reg dstReg = EDI;
static const X64Reg counterReg = ECX;
#endif
// XMM0-XMM5 are volatile on Windows X64
// XMM0-XMM7 are arguments (and thus volatile) on System V ABI (other x64 platforms)
static const X64Reg fpUscaleReg = XMM0;
@ -1244,7 +1245,6 @@ static const X64Reg fpVoffsetReg = XMM3;
static const X64Reg fpScratchReg = XMM4;
static const X64Reg fpScratchReg2 = XMM5;
// To debug, just comment them out one at a time until it works. We fall back
// on the interpreter if the compiler fails.
@ -1282,6 +1282,13 @@ static const JitLookup jitLookup[] = {
{&VertexDecoder::Step_PosFloat, &VertexDecoderJitCache::Jit_PosFloat},
};
// TODO: This should probably be global...
#ifdef _M_X64
#define PTRBITS 64
#else
#define PTRBITS 32
#endif
JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec) {
dec_ = &dec;
const u8 *start = this->GetCodePtr();
@ -1298,9 +1305,15 @@ JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec) {
MOV(32, R(srcReg), MDisp(ESP, 16 + offset + 0));
MOV(32, R(dstReg), MDisp(ESP, 16 + offset + 4));
MOV(32, R(counterReg), MDisp(ESP, 16 + offset + 8));
#endif
// Save XMM4/XMM5 which apparently can be problematic?
// Actually, if they are, it must be a compiler bug because they SHOULD be ok.
// So I won't bother.
// SUB(PTRBITS, R(ESP), Imm8(32));
// MOVUPS(MDisp(ESP, 0), XMM4);
// MOVUPS(MDisp(ESP, 16), XMM5);
bool prescaleStep = false;
// Look for prescaled texcoord steps
for (int i = 0; i < dec.numSteps_; i++) {
@ -1341,16 +1354,15 @@ JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec) {
}
}
#ifdef _M_X64
ADD(64, R(srcReg), Imm32(dec.VertexSize()));
ADD(64, R(dstReg), Imm32(dec.decFmt.stride));
#else
ADD(32, R(srcReg), Imm32(dec.VertexSize()));
ADD(32, R(dstReg), Imm32(dec.decFmt.stride));
#endif
ADD(PTRBITS, R(srcReg), Imm32(dec.VertexSize()));
ADD(PTRBITS, R(dstReg), Imm32(dec.decFmt.stride));
SUB(32, R(counterReg), Imm8(1));
J_CC(CC_NZ, loopStart, true);
// MOVUPS(XMM4, MDisp(ESP, 0));
// MOVUPS(XMM5, MDisp(ESP, 16));
// ADD(PTRBITS, R(ESP), Imm8(32));
#ifdef _M_IX86
// Restore register values
POP(EBP);