archive/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2024-09-20 19:31:53 +02:00
Code Issues Projects Releases Packages Wiki Activity

Merge pull request #1922 from Tilka/vertex_loader_jit

Browse source 
VertexLoader: fix position offset bug
This commit is contained in:
Markus Wick 2015-01-19 17:53:50 +01:00
commit 7454297820
2 changed files with 2 additions and 49 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
Source/Core/VideoCommon/VertexLoader.cpp
View file

@ -145,7 +145,6 @@ void VertexLoader::CompileVertexTranslator()
m_VertexSize += 1;
}
if (m_VtxDesc.Tex0MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX0; WriteCall(TexMtx_ReadDirect_UByte); }
if (m_VtxDesc.Tex1MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX1; WriteCall(TexMtx_ReadDirect_UByte); }
if (m_VtxDesc.Tex2MatIdx) {m_VertexSize += 1; components |= VB_HAS_TEXMTXIDX2; WriteCall(TexMtx_ReadDirect_UByte); }
@ -159,12 +158,12 @@ void VertexLoader::CompileVertexTranslator()
WriteCall(VertexLoader_Position::GetFunction(m_VtxDesc.Position, m_VtxAttr.PosFormat, m_VtxAttr.PosElements));
m_VertexSize += VertexLoader_Position::GetSize(m_VtxDesc.Position, m_VtxAttr.PosFormat, m_VtxAttr.PosElements);
nat_offset += 12;
m_native_vtx_decl.position.components = 3;
m_native_vtx_decl.position.enable = true;
m_native_vtx_decl.position.offset = 0;
m_native_vtx_decl.position.offset = nat_offset;
m_native_vtx_decl.position.type = VAR_FLOAT;
m_native_vtx_decl.position.integer = false;
nat_offset += 12;
// Normals
if (m_VtxDesc.Normal != NOT_PRESENT)

46
Source/Core/VideoCommon/VertexLoader_Position.cpp
View file

@ -12,52 +12,6 @@
#include "VideoCommon/VertexManagerBase.h"
#include "VideoCommon/VideoCommon.h"
// Thoughts on the implementation of a vertex loader compiler.
// g_vertex_manager_write_ptr 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);
*/
template <typename T>
float PosScale(T val, float scale)
{