OU
/
CWE-ou-minkata
mirror of https://foundry.openuru.org/gitblit/r/CWE-ou-minkata.git
2
1
Fork 0
Code Releases Activity
Browse Source

Remove x87 hack for typical (avatar) use cases

Adam Johnson 12 years ago
parent
342dd5fe14
commit
4a189faf36
1 changed files with 34 additions and 29 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 57
      Sources/Plasma/PubUtilLib/plPipeline/plDXPipeline.cpp

57
Sources/Plasma/PubUtilLib/plPipeline/plDXPipeline.cpp
Unescape View File

@ -203,10 +203,10 @@ void plReleaseObject(IUnknown* x)
//// Local Static Stuff /////////////////////////////////////////////////////// //// Local Static Stuff ///////////////////////////////////////////////////////
/// Macros for getting/setting data in a D3D vertex buffer /// Macros for getting/setting data in a D3D vertex buffer
inline uint8_t* inlStuffPoint( uint8_t* ptr, const hsScalarTriple& point ) inline uint8_t* inlStuffPoint( uint8_t* ptr, const hsScalarTriple* point )
{ {
register float* dst = (float*)ptr; register float* dst = (float*)ptr;
register const float* src = (float*)&point.fX; register const float* src = (float*)&point->fX;
*dst++ = *src++; *dst++ = *src++;
*dst++ = *src++; *dst++ = *src++;
*dst++ = *src++; *dst++ = *src++;
@ -10617,8 +10617,7 @@ inline void inlTESTPOINT(const hsPoint3& destP,
// SSE3 version // SSE3 version
#ifdef HS_SSE3 #ifdef HS_SSE3
#define MATRIXMULTBEGIN_SSE3(xfm, wgt) \ #define MATRIXMULTBEGIN_SSE3(xfm, wgt) \
__m128 mc0, mc1, mc2, mwt, msr, _x, _y, _z, hbuf1, hbuf2; \ __m128 mc0, mc1, mc2, mwt, msr, _x, _y, _z, hbuf1, hbuf2, _dst; \
ALIGN(16) float hack[4]; \
mc0 = _mm_load_ps(xfm.fMap[0]); \ mc0 = _mm_load_ps(xfm.fMap[0]); \
mc1 = _mm_load_ps(xfm.fMap[1]); \ mc1 = _mm_load_ps(xfm.fMap[1]); \
mc2 = _mm_load_ps(xfm.fMap[2]); \ mc2 = _mm_load_ps(xfm.fMap[2]); \
@ -10632,10 +10631,9 @@ inline void inlTESTPOINT(const hsPoint3& destP,
hbuf1 = _mm_hadd_ps(_x, _y); \ hbuf1 = _mm_hadd_ps(_x, _y); \
hbuf2 = _mm_hadd_ps(_z, _z); \ hbuf2 = _mm_hadd_ps(_z, _z); \
hbuf1 = _mm_hadd_ps(hbuf1, hbuf2); \ hbuf1 = _mm_hadd_ps(hbuf1, hbuf2); \
_mm_store_ps(hack, hbuf1); \ _dst = _mm_load_ps(dst); \
dst.fX += hack[0]; \ _dst = _mm_add_ps(_dst, hbuf1); \
dst.fY += hack[1]; \ _mm_store_ps(dst, _dst);
dst.fZ += hack[2];
#define MATRIXMULTVECTORADD_SSE3(dst, src) \ #define MATRIXMULTVECTORADD_SSE3(dst, src) \
msr = _mm_set_ps(0.f, src.fZ, src.fY, src.fX); \ msr = _mm_set_ps(0.f, src.fZ, src.fY, src.fX); \
_x = _mm_mul_ps(_mm_mul_ps(mc0, msr), mwt); \ _x = _mm_mul_ps(_mm_mul_ps(mc0, msr), mwt); \
@ -10645,10 +10643,13 @@ inline void inlTESTPOINT(const hsPoint3& destP,
hbuf1 = _mm_hadd_ps(_x, _y); \ hbuf1 = _mm_hadd_ps(_x, _y); \
hbuf2 = _mm_hadd_ps(_z, _z); \ hbuf2 = _mm_hadd_ps(_z, _z); \
hbuf1 = _mm_hadd_ps(hbuf1, hbuf2); \ hbuf1 = _mm_hadd_ps(hbuf1, hbuf2); \
{ \
ALIGN(16) float hack[4]; \
_mm_store_ps(hack, hbuf1); \ _mm_store_ps(hack, hbuf1); \
dst.fX += hack[0]; \ dst.fX += hack[0]; \
dst.fY += hack[1]; \ dst.fY += hack[1]; \
dst.fZ += hack[2]; dst.fZ += hack[2]; \
}
#endif #endif
// CPU-optimized functions requiring dispatch // CPU-optimized functions requiring dispatch
@ -10656,15 +10657,17 @@ hsFunctionDispatcher<plDXPipeline::blend_vert_buffer_ptr> plDXPipeline::blend_ve
// Temporary macros for IBlendVertsIntoBuffer dispatch code de-duplication // Temporary macros for IBlendVertsIntoBuffer dispatch code de-duplication
#define BLENDVERTSTART \ #define BLENDVERTSTART \
uint8_t numUVs, numWeights; \
uint32_t i, j, indices, color, specColor, uvChanSize; \
float weights[ 4 ], weightSum; \
ALIGN(16) float pt_buf[] = { 0.f, 0.f, 0.f, 1.f }; \ ALIGN(16) float pt_buf[] = { 0.f, 0.f, 0.f, 1.f }; \
ALIGN(16) float vec_buf[] = { 0.f, 0.f, 0.f, 0.f }; \ ALIGN(16) float vec_buf[] = { 0.f, 0.f, 0.f, 0.f }; \
ALIGN(16) float destPt_buf[4], destNorm_buf[4]; \
hsPoint3* pt = reinterpret_cast<hsPoint3*>(pt_buf); \ hsPoint3* pt = reinterpret_cast<hsPoint3*>(pt_buf); \
hsPoint3* destPt = reinterpret_cast<hsPoint3*>(destPt_buf); \
hsVector3* vec = reinterpret_cast<hsVector3*>(vec_buf); \ hsVector3* vec = reinterpret_cast<hsVector3*>(vec_buf); \
hsPoint3 destPt; \ hsVector3* destNorm = reinterpret_cast<hsVector3*>(destNorm_buf); \
hsVector3 destNorm; \ \
uint8_t numUVs, numWeights; \
uint32_t i, j, indices, color, specColor, uvChanSize; \
float weights[ 4 ], weightSum; \
\ \
/* Get some counts */\ /* Get some counts */\
switch( format & plGBufferGroup::kSkinWeightMask ) \ switch( format & plGBufferGroup::kSkinWeightMask ) \
@ -10710,8 +10713,9 @@ hsFunctionDispatcher<plDXPipeline::blend_vert_buffer_ptr> plDXPipeline::blend_ve
src = inlExtractUInt32( src, specColor ); \ src = inlExtractUInt32( src, specColor ); \
\ \
/* Blend */\ /* Blend */\
destPt.Set( 0, 0, 0 ); \ destPt->Set(0.f, 0.f, 0.f); \
destNorm.Set( 0, 0, 0 ); \ destPt_buf[3] = 1.f; \
destNorm->Set(0.f, 0.f, 0.f); \
for( j = 0; j < numWeights + 1; j++ ) \ for( j = 0; j < numWeights + 1; j++ ) \
{ \ { \
if( weights[ j ] ) \ if( weights[ j ] ) \
@ -10783,8 +10787,9 @@ hsFunctionDispatcher<plDXPipeline::blend_vert_buffer_ptr> plDXPipeline::blend_ve
dstUVWs[hiChan].Set(0,0,0); \ dstUVWs[hiChan].Set(0,0,0); \
\ \
/* Blend */\ /* Blend */\
destPt.Set( 0, 0, 0 ); \ destPt->Set(0.f, 0.f, 0.f); \
destNorm.Set( 0, 0, 0 ); \ destPt_buf[3] = 1.f; \
destNorm->Set(0.f, 0.f, 0.f); \
for( j = 0; j < numWeights + 1; j++ ) \ for( j = 0; j < numWeights + 1; j++ ) \
{ \ { \
if( weights[ j ] ) \ if( weights[ j ] ) \
@ -10827,13 +10832,13 @@ void plDXPipeline::blend_vert_buffer_fpu( plSpan* span,
BLENDVERTSTART BLENDVERTSTART
MATRIXMULTBEGIN_FPU(matrixPalette[indices & 0xff], weights[j]); MATRIXMULTBEGIN_FPU(matrixPalette[indices & 0xff], weights[j]);
MATRIXMULTPOINTADD_FPU(destPt, (*pt)); MATRIXMULTPOINTADD_FPU((*destPt), (*pt));
MATRIXMULTVECTORADD_FPU(destNorm, (*vec)); MATRIXMULTVECTORADD_FPU((*destNorm), (*vec));
BLENDVERTMID BLENDVERTMID
MATRIXMULTBEGIN_FPU(matrixPalette[indices & 0xff], weights[j]); MATRIXMULTBEGIN_FPU(matrixPalette[indices & 0xff], weights[j]);
MATRIXMULTPOINTADD_FPU(destPt, (*pt)); MATRIXMULTPOINTADD_FPU((*destPt), (*pt));
MATRIXMULTVECTORADD_FPU(destNorm, (*vec)); MATRIXMULTVECTORADD_FPU((*destNorm), (*vec));
MATRIXMULTVECTORADD_FPU(dstUVWs[loChan], srcUVWs[loChan]); MATRIXMULTVECTORADD_FPU(dstUVWs[loChan], srcUVWs[loChan]);
MATRIXMULTVECTORADD_FPU(dstUVWs[hiChan], srcUVWs[hiChan]); MATRIXMULTVECTORADD_FPU(dstUVWs[hiChan], srcUVWs[hiChan]);
@ -10850,13 +10855,13 @@ void plDXPipeline::blend_vert_buffer_sse3( plSpan* span,
BLENDVERTSTART BLENDVERTSTART
MATRIXMULTBEGIN_SSE3(matrixPalette[indices & 0xff], weights[j]); MATRIXMULTBEGIN_SSE3(matrixPalette[indices & 0xff], weights[j]);
MATRIXMULTBUFADD_SSE3(destPt, pt_buf); MATRIXMULTBUFADD_SSE3(destPt_buf, pt_buf);
MATRIXMULTBUFADD_SSE3(destNorm, vec_buf); MATRIXMULTBUFADD_SSE3(destNorm_buf, vec_buf);
BLENDVERTMID BLENDVERTMID
MATRIXMULTBEGIN_SSE3(matrixPalette[indices & 0xff], weights[j]); MATRIXMULTBEGIN_SSE3(matrixPalette[indices & 0xff], weights[j]);
MATRIXMULTBUFADD_SSE3(destPt, pt_buf); MATRIXMULTBUFADD_SSE3(destPt_buf, pt_buf);
MATRIXMULTBUFADD_SSE3(destNorm, vec_buf); MATRIXMULTBUFADD_SSE3(destNorm_buf, vec_buf);
MATRIXMULTVECTORADD_SSE3(dstUVWs[loChan], srcUVWs[loChan]); MATRIXMULTVECTORADD_SSE3(dstUVWs[loChan], srcUVWs[loChan]);
MATRIXMULTVECTORADD_SSE3(dstUVWs[hiChan], srcUVWs[hiChan]); MATRIXMULTVECTORADD_SSE3(dstUVWs[hiChan], srcUVWs[hiChan]);
BLENDVERTEND BLENDVERTEND

Write Preview
Loading…
Cancel
Save