piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-23 06:09:50 -06:00
Code Issues Packages Projects Releases Wiki Activity

Vertexloader cleanup/fixing: Don't use floating point copies unless necessary, don't allow scale factors for floating point texture coordinates.

Browse Source 
git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1282 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
hrydgard
2008-11-23 23:54:52 +00:00
parent 5672367b4c
commit b19859450e
4 changed files with 91 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
Source/Plugins/Plugin_VideoOGL/Src/VertexLoader_Position.h
View File

@ -20,6 +20,52 @@
#define LOG_VTX() PRIM_LOG("vtx: %f %f %f, ", ((float*)VertexManager::s_pCurBufferPointer)[0], ((float*)VertexManager::s_pCurBufferPointer)[1], ((float*)VertexManager::s_pCurBufferPointer)[2]);
// Thoughts on the implementation of a vertex loader compiler.
// s_pCurBufferPointer should definitely be in a register.
// Could load the position scale factor in XMM7, for example.
// The pointer inside DataReadU8 in another.
// Let's check out Pos_ReadDirect_UByte(). For Byte, replace MOVZX with MOVSX.
/*
MOVZX(32, R(EAX), MOffset(ESI, 0));
MOVZX(32, R(EBX), MOffset(ESI, 1));
MOVZX(32, R(ECX), MOffset(ESI, 2));
MOVD(XMM0, R(EAX));
MOVD(XMM1, R(EBX));
MOVD(XMM2, R(ECX));
CVTDQ2PS(XMM0, XMM0);
CVTDQ2PS(XMM1, XMM1);
CVTDQ2PS(XMM2, XMM2);
MULSS(XMM0, XMM7);
MULSS(XMM1, XMM7);
MULSS(XMM2, XMM7);
MOVSS(MOffset(EDI, 0), XMM0);
MOVSS(MOffset(EDI, 4), XMM1);
MOVSS(MOffset(EDI, 8), XMM2);
Alternatively, lookup table:
MOVZX(32, R(EAX), MOffset(ESI, 0));
MOVZX(32, R(EBX), MOffset(ESI, 1));
MOVZX(32, R(ECX), MOffset(ESI, 2));
MOV(32, R(EAX), MComplex(LUTREG, EAX, 4));
MOV(32, R(EBX), MComplex(LUTREG, EBX, 4));
MOV(32, R(ECX), MComplex(LUTREG, ECX, 4));
MOV(MOffset(EDI, 0), XMM0);
MOV(MOffset(EDI, 4), XMM1);
MOV(MOffset(EDI, 8), XMM2);
SSE4:
PINSRB(XMM0, MOffset(ESI, 0), 0);
PINSRB(XMM0, MOffset(ESI, 1), 4);
PINSRB(XMM0, MOffset(ESI, 2), 8);
CVTDQ2PS(XMM0, XMM0);
<two unpacks here to sign extend>
MULPS(XMM0, XMM7);
MOVUPS(MOffset(EDI, 0), XMM0);
*/
// ==============================================================================
// Direct
// ==============================================================================
@ -73,10 +119,11 @@ void LOADERDECL Pos_ReadDirect_Short()
void LOADERDECL Pos_ReadDirect_Float()
{
((float*)VertexManager::s_pCurBufferPointer)[0] = DataReadF32();
((float*)VertexManager::s_pCurBufferPointer)[1] = DataReadF32();
// No need to use floating point here.
((u32 *)VertexManager::s_pCurBufferPointer)[0] = DataReadU32();
((u32 *)VertexManager::s_pCurBufferPointer)[1] = DataReadU32();
if (pVtxAttr->PosElements)
((float*)VertexManager::s_pCurBufferPointer)[2] = DataReadF32();
((u32 *)VertexManager::s_pCurBufferPointer)[2] = DataReadU32();
else
((float*)VertexManager::s_pCurBufferPointer)[2] = 1.0f;
LOG_VTX();