piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-07-23 22:29:39 -06:00
Code Issues Packages Projects Releases Wiki Activity

Vertex Loader: SSE implementations of more position/texcoord/normal formats

Browse Source 
~35-45% faster NFS:HP2, possibly other vertex-bound games.
This commit is contained in:
Fiora
2014-11-11 01:48:38 -08:00
parent da962a3d2b
commit 3ddf82a318
7 changed files with 299 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
Source/Core/VideoCommon/VertexLoader.h
View File

@ -19,6 +19,13 @@
#include "VideoCommon/DataReader.h"
#include "VideoCommon/NativeVertexFormat.h"
#if _M_SSE >= 0x401
#include <smmintrin.h>
#include <emmintrin.h>
#elif _M_SSE >= 0x301 && !(defined __GNUC__ && !defined __SSSE3__)
#include <tmmintrin.h>
#endif
#ifdef _M_X86
#define USE_VERTEX_LOADER_JIT
#endif
@ -27,8 +34,8 @@
extern int tcIndex;
extern int colIndex;
extern int colElements[2];
extern float posScale;
extern float tcScale[8];
GC_ALIGNED128(extern float posScale[4]);
GC_ALIGNED64(extern float tcScale[8][2]);
class VertexLoaderUID
{
@ -155,3 +162,61 @@ private:
void WriteSetVariable(int bits, void *address, Gen::OpArg dest);
#endif
};
#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);
static const __m128i kMaskSwap16to32l_3 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFF0405L, 0xFFFF0203L, 0xFFFF0001L);
static const __m128i kMaskSwap16to32l_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFF0203L, 0xFFFF0001L);
static const __m128i kMaskSwap16to32h_3 = _mm_set_epi32(0xFFFFFFFFL, 0x0405FFFFL, 0x0203FFFFL, 0x0001FFFFL);
static const __m128i kMaskSwap16to32h_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x0203FFFFL, 0x0001FFFFL);
static const __m128i kMask8to32l_3 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFF02L, 0xFFFFFF01L, 0xFFFFFF00L);
static const __m128i kMask8to32l_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0xFFFFFF01L, 0xFFFFFF00L);
static const __m128i kMask8to32h_3 = _mm_set_epi32(0xFFFFFFFFL, 0x02FFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL);
static const __m128i kMask8to32h_2 = _mm_set_epi32(0xFFFFFFFFL, 0xFFFFFFFFL, 0x01FFFFFFL, 0x00FFFFFFL);
template <typename T, bool threeIn, bool threeOut>
__forceinline void Vertex_Read_SSSE3(const T* pData, __m128 scale)
{
__m128i coords, mask;
int loadBytes = sizeof(T) * (2 + threeIn);
if (loadBytes > 8)
coords = _mm_loadu_si128((__m128i*)pData);
else if (loadBytes > 4)
coords = _mm_loadl_epi64((__m128i*)pData);
else
coords = _mm_cvtsi32_si128(*(u32*)pData);
// Float case (no scaling)
if (sizeof(T) == 4)
{
coords = _mm_shuffle_epi8(coords, threeIn ? kMaskSwap32_3 : kMaskSwap32_2);
if (threeOut)
_mm_storeu_si128((__m128i*)VertexManager::s_pCurBufferPointer, coords);
else
_mm_storel_epi64((__m128i*)VertexManager::s_pCurBufferPointer, coords);
}
else
{
// Byte swap, unpack, and move to high bytes for sign extend.
if (std::is_unsigned<T>::value)
mask = sizeof(T) == 2 ? (threeIn ? kMaskSwap16to32l_3 : kMaskSwap16to32l_2) : (threeIn ? kMask8to32l_3 : kMask8to32l_2);
else
mask = sizeof(T) == 2 ? (threeIn ? kMaskSwap16to32h_3 : kMaskSwap16to32h_2) : (threeIn ? kMask8to32h_3 : kMask8to32h_2);
coords = _mm_shuffle_epi8(coords, mask);
// Sign extend
if (std::is_signed<T>::value)
coords = _mm_srai_epi32(coords, 32 - sizeof(T) * 8);
__m128 out = _mm_mul_ps(_mm_cvtepi32_ps(coords), scale);
if (threeOut)
_mm_storeu_ps((float*)VertexManager::s_pCurBufferPointer, out);
else
_mm_storel_pi((__m64*)VertexManager::s_pCurBufferPointer, out);
}
VertexManager::s_pCurBufferPointer += sizeof(float) * (2 + threeOut);
}
#endif