dolphin/Source/Core/VideoCommon/Src/VertexLoader_Position.cpp

204 lines
5.8 KiB
C++

// Copyright (C) 2003 Dolphin Project.
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 2.0.
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License 2.0 for more details.
// A copy of the GPL 2.0 should have been included with the program.
// If not, see http://www.gnu.org/licenses/
// Official SVN repository and contact information can be found at
// http://code.google.com/p/dolphin-emu/
#include <limits>
#include "Common.h"
#include "VideoCommon.h"
#include "VertexLoader.h"
#include "VertexLoader_Position.h"
#include "VertexManagerBase.h"
#include "CPUDetect.h"
extern float posScale;
extern TVtxAttr *pVtxAttr;
// 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);
*/
template <typename T>
float PosScale(T val)
{
return val * posScale;
}
template <>
float PosScale(float val)
{ return val; }
template <typename T, int N>
void LOADERDECL Pos_ReadDirect()
{
static_assert(N <= 3, "N > 3 is not sane!");
for (int i = 0; i < 3; ++i)
DataWrite(i<N ? PosScale(DataRead<T>()) : 0.f);
LOG_VTX();
}
template <typename I, typename T, int N>
void LOADERDECL Pos_ReadIndex()
{
static_assert(!std::numeric_limits<I>::is_signed, "Only unsigned I is sane!");
static_assert(N <= 3, "N > 3 is not sane!");
auto const index = DataRead<I>();
if (index < std::numeric_limits<I>::max())
{
auto const data = reinterpret_cast<const T*>(cached_arraybases[ARRAY_POSITION] + (index * arraystrides[ARRAY_POSITION]));
for (int i = 0; i < 3; ++i)
DataWrite(i<N ? PosScale(Common::FromBigEndian(data[i])) : 0.f);
LOG_VTX();
}
}
#if _M_SSE >= 0x301
static const __m128i kMaskSwap32_3 = _mm_set_epi32(0xFFFFFFFFL, 0x08090A0BL, 0x04050607L, 0x00010203L);
static const __m128i kMaskSwap32_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x04050607L, 0x00010203L);
template <typename I, bool three>
void LOADERDECL Pos_ReadIndex_Float_SSSE3()
{
auto const index = DataRead<I>();
if (index < std::numeric_limits<I>::max())
{
const u32* pData = (const u32 *)(cached_arraybases[ARRAY_POSITION] + (index * arraystrides[ARRAY_POSITION]));
GC_ALIGNED128(const __m128i a = _mm_loadu_si128((__m128i*)pData));
GC_ALIGNED128(__m128i b = _mm_shuffle_epi8(a, three ? kMaskSwap32_3 : kMaskSwap32_2));
_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, b);
VertexManager::s_pCurBufferPointer += sizeof(float) * 3;
LOG_VTX();
}
}
#endif
static TPipelineFunction tableReadPosition[4][8][2] = {
{
{NULL, NULL,},
{NULL, NULL,},
{NULL, NULL,},
{NULL, NULL,},
{NULL, NULL,},
},
{
{Pos_ReadDirect<u8, 2>, Pos_ReadDirect<u8, 3>,},
{Pos_ReadDirect<s8, 2>, Pos_ReadDirect<s8, 3>,},
{Pos_ReadDirect<u16, 2>, Pos_ReadDirect<u16, 3>,},
{Pos_ReadDirect<s16, 2>, Pos_ReadDirect<s16, 3>,},
{Pos_ReadDirect<float, 2>, Pos_ReadDirect<float, 3>,},
},
{
{Pos_ReadIndex<u8, u8, 2>, Pos_ReadIndex<u8, u8, 3>,},
{Pos_ReadIndex<u8, s8, 2>, Pos_ReadIndex<u8, s8, 3>,},
{Pos_ReadIndex<u8, u16, 2>, Pos_ReadIndex<u8, u16, 3>,},
{Pos_ReadIndex<u8, s16, 2>, Pos_ReadIndex<u8, s16, 3>,},
{Pos_ReadIndex<u8, float, 2>, Pos_ReadIndex<u8, float, 3>,},
},
{
{Pos_ReadIndex<u16, u8, 2>, Pos_ReadIndex<u16, u8, 3>,},
{Pos_ReadIndex<u16, s8, 2>, Pos_ReadIndex<u16, s8, 3>,},
{Pos_ReadIndex<u16, u16, 2>, Pos_ReadIndex<u16, u16, 3>,},
{Pos_ReadIndex<u16, s16, 2>, Pos_ReadIndex<u16, s16, 3>,},
{Pos_ReadIndex<u16, float, 2>, Pos_ReadIndex<u16, float, 3>,},
},
};
static int tableReadPositionVertexSize[4][8][2] = {
{
{0, 0,}, {0, 0,}, {0, 0,}, {0, 0,}, {0, 0,},
},
{
{2, 3,}, {2, 3,}, {4, 6,}, {4, 6,}, {8, 12,},
},
{
{1, 1,}, {1, 1,}, {1, 1,}, {1, 1,}, {1, 1,},
},
{
{2, 2,}, {2, 2,}, {2, 2,}, {2, 2,}, {2, 2,},
},
};
void VertexLoader_Position::Init(void) {
#if _M_SSE >= 0x301
if (cpu_info.bSSSE3) {
tableReadPosition[2][4][0] = Pos_ReadIndex_Float_SSSE3<u8, false>;
tableReadPosition[2][4][1] = Pos_ReadIndex_Float_SSSE3<u8, true>;
tableReadPosition[3][4][0] = Pos_ReadIndex_Float_SSSE3<u16, false>;
tableReadPosition[3][4][1] = Pos_ReadIndex_Float_SSSE3<u16, true>;
}
#endif
}
unsigned int VertexLoader_Position::GetSize(unsigned int _type, unsigned int _format, unsigned int _elements) {
return tableReadPositionVertexSize[_type][_format][_elements];
}
TPipelineFunction VertexLoader_Position::GetFunction(unsigned int _type, unsigned int _format, unsigned int _elements) {
return tableReadPosition[_type][_format][_elements];
}