mplayer.hg: libswscale/swscale_template.c comparison

comparison libswscale/swscale_template.c @ 29404:efbe0c7de351

Do not misuse HAVE_ defines. Introduce COMPILE_TEMPLATE_ defines and use them instead.

author	ramiro
date	Mon, 27 Jul 2009 06:47:41 +0000
parents	6c1b9f334f97
children	45759bfb13d6

comparison

equal deleted inserted replaced

-:6c1b9f334f97
+:efbe0c7de351
 #undef MOVNTQ
 #undef PAVGB
 #undef PREFETCH
 #undef PREFETCHW
-#if HAVE_AMD3DNOW
+#if COMPILE_TEMPLATE_AMD3DNOW
 #define PREFETCH  "prefetch"
 #define PREFETCHW "prefetchw"
-#elif HAVE_MMX2
+#elif COMPILE_TEMPLATE_MMX2
 #define PREFETCH "prefetchnta"
 #define PREFETCHW "prefetcht0"
 #else
 #define PREFETCH  " # nop"
 #define PREFETCHW " # nop"
 #endif
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
-#elif HAVE_AMD3DNOW
+#elif COMPILE_TEMPLATE_AMD3DNOW
 #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
 #endif
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 #define REAL_MOVNTQ(a,b) "movntq " #a ", " #b " \n\t"
 #else
 #define REAL_MOVNTQ(a,b) "movq " #a ", " #b " \n\t"
 #endif
 #define MOVNTQ(a,b)  REAL_MOVNTQ(a,b)
-#if HAVE_ALTIVEC
+#if COMPILE_TEMPLATE_ALTIVEC
 #include "ppc/swscale_altivec_template.c"
 #endif
 #define YSCALEYUV2YV12X(x, offset, dest, width) \
 __asm__ volatile(\
 \
 "add       $8, "#index"     \n\t"\
 "cmp  "#dstw", "#index"     \n\t"\
 " jb       1b               \n\t"
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 #undef WRITEBGR24
 #define WRITEBGR24(dst, dstw, index)  WRITEBGR24MMX2(dst, dstw, index)
 #else
 #undef WRITEBGR24
 #define WRITEBGR24(dst, dstw, index)  WRITEBGR24MMX(dst, dstw, index)
 static inline void RENAME(yuv2yuvX)(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize, const int16_t **alpSrc,
 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, long dstW, long chrDstW)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 if(!(c->flags & SWS_BITEXACT)){
 if (c->flags & SWS_ACCURATE_RND){
 if (uDest){
 YSCALEYUV2YV12X_ACCURATE(   "0", CHR_MMX_FILTER_OFFSET, uDest, chrDstW)
 YSCALEYUV2YV12X_ACCURATE(AV_STRINGIFY(VOF), CHR_MMX_FILTER_OFFSET, vDest, chrDstW)
 YSCALEYUV2YV12X("0", LUM_MMX_FILTER_OFFSET, dest, dstW)
 }
 return;
 }
 #endif
-#if HAVE_ALTIVEC
+#if COMPILE_TEMPLATE_ALTIVEC
 yuv2yuvX_altivec_real(lumFilter, lumSrc, lumFilterSize,
 chrFilter, chrSrc, chrFilterSize,
 dest, uDest, vDest, dstW, chrDstW);
-#else //HAVE_ALTIVEC
+#else //COMPILE_TEMPLATE_ALTIVEC
 yuv2yuvXinC(lumFilter, lumSrc, lumFilterSize,
 chrFilter, chrSrc, chrFilterSize,
 alpSrc, dest, uDest, vDest, aDest, dstW, chrDstW);
-#endif //!HAVE_ALTIVEC
+#endif //!COMPILE_TEMPLATE_ALTIVEC
 }
 static inline void RENAME(yuv2nv12X)(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
 uint8_t *dest, uint8_t *uDest, int dstW, int chrDstW, int dstFormat)
 static inline void RENAME(yuv2yuv1)(SwsContext *c, const int16_t *lumSrc, const int16_t *chrSrc, const int16_t *alpSrc,
 uint8_t *dest, uint8_t *uDest, uint8_t *vDest, uint8_t *aDest, long dstW, long chrDstW)
 {
 int i;
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 if(!(c->flags & SWS_BITEXACT)){
 long p= 4;
 uint8_t *src[4]= {alpSrc + dstW, lumSrc + dstW, chrSrc + chrDstW, chrSrc + VOFW + chrDstW};
 uint8_t *dst[4]= {aDest, dest, uDest, vDest};
 x86_reg counter[4]= {dstW, dstW, chrDstW, chrDstW};
 */
 static inline void RENAME(yuv2packedX)(SwsContext *c, const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
 const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
 const int16_t **alpSrc, uint8_t *dest, long dstW, long dstY)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 x86_reg dummy=0;
 if(!(c->flags & SWS_BITEXACT)){
 if (c->flags & SWS_ACCURATE_RND){
 switch(c->dstFormat){
 case PIX_FMT_RGB32:
 YSCALEYUV2PACKEDX_END
 return;
 }
 }
 }
-#endif /* HAVE_MMX */
+#endif /* COMPILE_TEMPLATE_MMX */
-#if HAVE_ALTIVEC
+#if COMPILE_TEMPLATE_ALTIVEC
 /* The following list of supported dstFormat values should
 match what's found in the body of ff_yuv2packedX_altivec() */
 if (!(c->flags & SWS_BITEXACT) && !c->alpPixBuf &&
 (c->dstFormat==PIX_FMT_ABGR  || c->dstFormat==PIX_FMT_BGRA  ||
 c->dstFormat==PIX_FMT_BGR24 || c->dstFormat==PIX_FMT_RGB24 ||
 {
 int  yalpha1=4095- yalpha;
 int uvalpha1=4095-uvalpha;
 int i;
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 if(!(c->flags & SWS_BITEXACT)){
 switch(c->dstFormat)
 {
 //Note 8280 == DSTW_OFFSET but the preprocessor can't handle that there :(
 case PIX_FMT_RGB32:
 );
 return;
 default: break;
 }
 }
-#endif //HAVE_MMX
+#endif //COMPILE_TEMPLATE_MMX
 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB2_C, YSCALE_YUV_2_PACKED2_C(void,0), YSCALE_YUV_2_GRAY16_2_C, YSCALE_YUV_2_MONO2_C)
 }
 /**
 * YV12 to RGB without scaling or interpolating
 {
 c->yuv2packed2(c, buf0, buf0, uvbuf0, uvbuf1, abuf0, abuf0, dest, dstW, 0, uvalpha, y);
 return;
 }
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 if(!(flags & SWS_BITEXACT)){
 if (uvalpha < 2048) // note this is not correct (shifts chrominance by 0.5 pixels) but it is a bit faster
 {
 switch(dstFormat)
 {
 );
 return;
 }
 }
 }
-#endif /* HAVE_MMX */
+#endif /* COMPILE_TEMPLATE_MMX */
 if (uvalpha < 2048)
 {
 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB1_C, YSCALE_YUV_2_PACKED1_C(void,0), YSCALE_YUV_2_GRAY16_1_C, YSCALE_YUV_2_MONO2_C)
 }else{
 YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGB1B_C, YSCALE_YUV_2_PACKED1B_C(void,0), YSCALE_YUV_2_GRAY16_1_C, YSCALE_YUV_2_MONO2_C)
 //FIXME yuy2* can read up to 7 samples too much
 static inline void RENAME(yuy2ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "movq "MANGLE(bm01010101)", %%mm2           \n\t"
 "mov                    %0, %%"REG_a"       \n\t"
 "1:                                         \n\t"
 "movq    (%1, %%"REG_a",2), %%mm0           \n\t"
 #endif
 }
 static inline void RENAME(yuy2ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "movq "MANGLE(bm01010101)", %%mm4           \n\t"
 "mov                    %0, %%"REG_a"       \n\t"
 "1:                                         \n\t"
 "movq    (%1, %%"REG_a",4), %%mm0           \n\t"
 assert(src1 == src2);
 }
 static inline void RENAME(LEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "mov                    %0, %%"REG_a"       \n\t"
 "1:                                         \n\t"
 "movq    (%1, %%"REG_a",2), %%mm0           \n\t"
 "movq   8(%1, %%"REG_a",2), %%mm1           \n\t"
 /* This is almost identical to the previous, end exists only because
 * yuy2ToY/UV)(dst, src+1, ...) would have 100% unaligned accesses. */
 static inline void RENAME(uyvyToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "mov                  %0, %%"REG_a"         \n\t"
 "1:                                         \n\t"
 "movq  (%1, %%"REG_a",2), %%mm0             \n\t"
 "movq 8(%1, %%"REG_a",2), %%mm1             \n\t"
 #endif
 }
 static inline void RENAME(uyvyToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "movq "MANGLE(bm01010101)", %%mm4           \n\t"
 "mov                    %0, %%"REG_a"       \n\t"
 "1:                                         \n\t"
 "movq    (%1, %%"REG_a",4), %%mm0           \n\t"
 assert(src1 == src2);
 }
 static inline void RENAME(BEToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 __asm__ volatile(
 "movq "MANGLE(bm01010101)", %%mm4           \n\t"
 "mov                    %0, %%"REG_a"       \n\t"
 "1:                                         \n\t"
 "movq    (%1, %%"REG_a",2), %%mm0           \n\t"
 dstV[i]= src2[2*i];
 }
 #endif
 }
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 static inline void RENAME(bgr24ToY_mmx)(uint8_t *dst, const uint8_t *src, long width, int srcFormat)
 {
 if(srcFormat == PIX_FMT_BGR24){
 __asm__ volatile(
 }
 #endif
 static inline void RENAME(bgr24ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_BGR24);
 #else
 int i;
 for (i=0; i<width; i++)
 {
 int g= src[i*3+1];
 int r= src[i*3+2];
 dst[i]= ((RY*r + GY*g + BY*b + (33<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT);
 }
-#endif /* HAVE_MMX */
+#endif /* COMPILE_TEMPLATE_MMX */
 }
 static inline void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_BGR24);
 #else
 int i;
 for (i=0; i<width; i++)
 {
 int r= src1[3*i + 2];
 dstU[i]= (RU*r + GU*g + BU*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
 dstV[i]= (RV*r + GV*g + BV*b + (257<<(RGB2YUV_SHIFT-1)))>>RGB2YUV_SHIFT;
 }
-#endif /* HAVE_MMX */
+#endif /* COMPILE_TEMPLATE_MMX */
 assert(src1 == src2);
 }
 static inline void RENAME(bgr24ToUV_half)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
 assert(src1 == src2);
 }
 static inline void RENAME(rgb24ToY)(uint8_t *dst, const uint8_t *src, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_RGB24);
 #else
 int i;
 for (i=0; i<width; i++)
 {
 #endif
 }
 static inline void RENAME(rgb24ToUV)(uint8_t *dstU, uint8_t *dstV, const uint8_t *src1, const uint8_t *src2, long width, uint32_t *unused)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 assert(src1==src2);
 RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_RGB24);
 #else
 int i;
 assert(src1==src2);
 // bilinear / bicubic scaling
 static inline void RENAME(hScale)(int16_t *dst, int dstW, const uint8_t *src, int srcW, int xInc,
 const int16_t *filter, const int16_t *filterPos, long filterSize)
 {
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 assert(filterSize % 4 == 0 && filterSize>0);
 if (filterSize==4) // Always true for upscaling, sometimes for down, too.
 {
 x86_reg counter= -2*dstW;
 filter-= counter*2;
 "m" (src), "r" ((x86_reg)filterSize*2)
 : "%"REG_a, "%"REG_c, "%"REG_d
 );
 }
 #else
-#if HAVE_ALTIVEC
+#if COMPILE_TEMPLATE_ALTIVEC
 hScale_altivec_real(dst, dstW, src, srcW, xInc, filter, filterPos, filterSize);
 #else
 int i;
 for (i=0; i<dstW; i++)
 {
 }
 //filter += hFilterSize;
 dst[i] = FFMIN(val>>7, (1<<15)-1); // the cubic equation does overflow ...
 //dst[i] = val>>7;
 }
-#endif /* HAVE_ALTIVEC */
+#endif /* COMPILE_ALTIVEC */
-#endif /* HAVE_MMX */
+#endif /* COMPILE_MMX */
 }
 #define FAST_BILINEAR_X86 \
 "subl    %%edi, %%esi    \n\t" /*  src[xx+1] - src[xx] */                   \
 "imull   %%ecx, %%esi    \n\t" /* (src[xx+1] - src[xx])*xalpha */           \
 if (internal_func) {
 internal_func(formatConvBuffer, src, srcW, pal);
 src= formatConvBuffer;
 }
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 // Use the new MMX scaler if the MMX2 one can't be used (it is faster than the x86 ASM one).
 if (!(flags&SWS_FAST_BILINEAR) || (!canMMX2BeUsed))
 #else
 if (!(flags&SWS_FAST_BILINEAR))
 #endif
 c->hScale(dst, dstWidth, src, srcW, xInc, hLumFilter, hLumFilterPos, hLumFilterSize);
 }
 else // fast bilinear upscale / crap downscale
 {
 #if ARCH_X86 && CONFIG_GPL
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 int i;
 #if defined(PIC)
 DECLARE_ALIGNED(8, uint64_t, ebxsave);
 #endif
 if (canMMX2BeUsed)
 );
 for (i=dstWidth-1; (i*xInc)>>16 >=srcW-1; i--) dst[i] = src[srcW-1]*128;
 }
 else
 {
-#endif /* HAVE_MMX2 */
+#endif /* COMPILE_TEMPLATE_MMX2 */
 x86_reg xInc_shr16 = xInc >> 16;
 uint16_t xInc_mask = xInc & 0xffff;
 //NO MMX just normal asm ...
 __asm__ volatile(
 "xor %%"REG_a", %%"REG_a"            \n\t" // i
 :: "r" (src), "m" (dst), "m" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask)
 : "%"REG_a, "%"REG_d, "%ecx", "%"REG_D, "%esi"
 );
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 } //if MMX2 can't be used
 #endif
 #else
 c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
 #endif /* ARCH_X86 */
 c->hcscale_internal(formatConvBuffer, formatConvBuffer+VOFW, src1, src2, srcW, pal);
 src1= formatConvBuffer;
 src2= formatConvBuffer+VOFW;
 }
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 // Use the new MMX scaler if the MMX2 one can't be used (it is faster than the x86 ASM one).
 if (!(flags&SWS_FAST_BILINEAR) || (!canMMX2BeUsed))
 #else
 if (!(flags&SWS_FAST_BILINEAR))
 #endif
 c->hScale(dst+VOFW, dstWidth, src2, srcW, xInc, hChrFilter, hChrFilterPos, hChrFilterSize);
 }
 else // fast bilinear upscale / crap downscale
 {
 #if ARCH_X86 && CONFIG_GPL
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 int i;
 #if defined(PIC)
 DECLARE_ALIGNED(8, uint64_t, ebxsave);
 #endif
 if (canMMX2BeUsed)
 dst[i+VOFW] = src2[srcW-1]*128;
 }
 }
 else
 {
-#endif /* HAVE_MMX2 */
+#endif /* COMPILE_TEMPLATE_MMX2 */
 x86_reg xInc_shr16 = (x86_reg) (xInc >> 16);
 uint16_t xInc_mask = xInc & 0xffff;
 __asm__ volatile(
 "xor %%"REG_a", %%"REG_a"               \n\t" // i
 "xor %%"REG_d", %%"REG_d"               \n\t" // xx
 :: "m" (src1), "m" (dst), "m" (dstWidth), "m" (xInc_shr16), "m" (xInc_mask),
 #endif
 "r" (src2)
 : "%"REG_a, "%"REG_d, "%ecx", "%"REG_D, "%esi"
 );
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 } //if MMX2 can't be used
 #endif
 #else
 c->hcscale_fast(c, dst, dstWidth, src1, src2, srcW, xInc);
 #endif /* ARCH_X86 */
 if (lumBufIndex >= vLumBufSize) lumBufIndex-= vLumBufSize;
 if (chrBufIndex >= vChrBufSize) chrBufIndex-= vChrBufSize;
 break; //we can't output a dstY line so let's try with the next slice
 }
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 c->blueDither= ff_dither8[dstY&1];
 if (c->dstFormat == PIX_FMT_RGB555 || c->dstFormat == PIX_FMT_BGR555)
 c->greenDither= ff_dither8[dstY&1];
 else
 c->greenDither= ff_dither4[dstY&1];
 if (dstY < dstH-2)
 {
 const int16_t **lumSrcPtr= (const int16_t **) lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
 const int16_t **chrSrcPtr= (const int16_t **) chrPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
 const int16_t **alpSrcPtr= (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? (const int16_t **) alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 int i;
 if (flags & SWS_ACCURATE_RND){
 int s= APCK_SIZE / 8;
 for (i=0; i<vLumFilterSize; i+=2){
 *(void**)&lumMmxFilter[s*i              ]= lumSrcPtr[i  ];
 }
 if ((dstFormat == PIX_FMT_YUVA420P) && !alpPixBuf)
 fillPlane(dst[3], dstStride[3], dstW, dstY-lastDstY, lastDstY, 255);
-#if HAVE_MMX
+#if COMPILE_TEMPLATE_MMX
 if (flags & SWS_CPU_CAPS_MMX2 )  __asm__ volatile("sfence":::"memory");
 /* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
 if (flags & SWS_CPU_CAPS_3DNOW)  __asm__ volatile("femms" :::"memory");
 else                             __asm__ volatile("emms"  :::"memory");
 #endif

Mercurial > mplayer.hg

comparison libswscale/swscale_template.c @ 29404:efbe0c7de351