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

VertexLoader: remove non-JIT SSE code

Browse Source
This commit is contained in:
Tillmann Karras
2015-01-19 18:33:08 +01:00
parent 38cb9bbfae
commit 873902b4a3
6 changed files with 8 additions and 349 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
Source/Core/VideoCommon/VertexLoader.h
View File

@ -18,13 +18,6 @@
#include "VideoCommon/VertexLoaderBase.h"
#include "VideoCommon/VertexLoaderUtils.h"
#if _M_SSE >= 0x401
#include <smmintrin.h>
#include <emmintrin.h>
#elif _M_SSE >= 0x301 && !(defined __GNUC__ && !defined __SSSE3__)
#include <tmmintrin.h>
#endif
#ifdef WIN32
#define LOADERDECL __cdecl
#else
@ -37,11 +30,6 @@ typedef void (LOADERDECL *TPipelineFunction)(VertexLoader* loader);
class VertexLoader : public VertexLoaderBase
{
public:
// This class need a 16 byte alignment. As this is broken on
// MSVC right now (Dec 2014), we use custom allocation.
void* operator new (size_t size);
void operator delete (void *p);
VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr);
int RunVertices(int primitive, int count, DataReader src, DataReader dst) override;
@ -49,9 +37,8 @@ public:
bool IsInitialized() override { return true; } // This vertex loader supports all formats
// They are used for the communication with the loader functions
// Duplicated (4x and 2x respectively) and used in SSE code in the vertex loader JIT
GC_ALIGNED128(float m_posScale[4]);
GC_ALIGNED64(float m_tcScale[8][2]);
float m_posScale;
float m_tcScale[8];
int m_tcIndex;
int m_colIndex;
int m_colElements[2];
@ -73,61 +60,3 @@ private:
void WriteCall(TPipelineFunction);
};
#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*)g_vertex_manager_write_ptr, coords);
else
_mm_storel_epi64((__m128i*)g_vertex_manager_write_ptr, 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*)g_vertex_manager_write_ptr, out);
else
_mm_storel_pi((__m64*)g_vertex_manager_write_ptr, out);
}
g_vertex_manager_write_ptr += sizeof(float) * (2 + threeOut);
}
#endif