# HG changeset patch # User michael # Date 1221143952 0 # Node ID 78da4c8b22936eabcfa03f817535dd9ee0250aa8 # Parent a917f63f61d3b8dddff5ce8ae3b35785a36e0356 Implement full horizontal chroma for rgb/bgr24/32 output. diff -r a917f63f61d3 -r 78da4c8b2293 libswscale/swscale.c --- a/libswscale/swscale.c Thu Sep 11 03:22:39 2008 +0000 +++ b/libswscale/swscale.c Thu Sep 11 14:39:12 2008 +0000 @@ -473,6 +473,43 @@ else if (V<0) V=0; \ } +#define YSCALE_YUV_2_PACKEDX_FULL_C \ + for (i=0; i>=10;\ + U >>=10;\ + V >>=10;\ + +#define YSCALE_YUV_2_RGBX_FULL_C(rnd) \ + YSCALE_YUV_2_PACKEDX_FULL_C\ + Y-= c->oy;\ + Y*= c->cy;\ + Y+= rnd;\ + R= Y + V*c->cvr;\ + G= Y + V*c->cvg + U*c->cug;\ + B= Y + U*c->cub;\ + if ((R|G|B)&(0xC0000000)){\ + if (R>=(256<<22)) R=(256<<22)-1; \ + else if (R<0)R=0; \ + if (G>=(256<<22)) G=(256<<22)-1; \ + else if (G<0)G=0; \ + if (B>=(256<<22)) B=(256<<22)-1; \ + else if (B<0)B=0; \ + }\ + + #define YSCALE_YUV_2_GRAY16_C \ for (i=0; i<(dstW>>1); i++){\ int j;\ @@ -756,6 +793,42 @@ YSCALE_YUV_2_ANYRGB_C(YSCALE_YUV_2_RGBX_C, YSCALE_YUV_2_PACKEDX_C(void), YSCALE_YUV_2_GRAY16_C, YSCALE_YUV_2_MONOBLACKX_C) } +static inline void yuv2rgbXinC_full(SwsContext *c, int16_t *lumFilter, int16_t **lumSrc, int lumFilterSize, + int16_t *chrFilter, int16_t **chrSrc, int chrFilterSize, + uint8_t *dest, int dstW, int y) +{ + int i; + int step= fmt_depth(c->dstFormat)/8; + + switch(c->dstFormat){ + case PIX_FMT_ARGB: + dest++; + case PIX_FMT_RGB24: + case PIX_FMT_RGBA: + YSCALE_YUV_2_RGBX_FULL_C(1<<21) + dest[0]= R>>22; + dest[1]= G>>22; + dest[2]= B>>22; + dest[3]= 0; + dest+= step; + } + break; + case PIX_FMT_ABGR: + dest++; + case PIX_FMT_BGR24: + case PIX_FMT_BGRA: + YSCALE_YUV_2_RGBX_FULL_C(1<<21) + dest[0]= B>>22; + dest[1]= G>>22; + dest[2]= R>>22; + dest[3]= 0; + dest+= step; + } + break; + default: + assert(0); + } +} //Note: we have C, X86, MMX, MMX2, 3DNOW version therse no 3DNOW+MMX2 one //Plain C versions @@ -1880,6 +1953,13 @@ c->ugCoeff= roundToInt16(cgu*8192) * 0x0001000100010001ULL; c->yOffset= roundToInt16(oy * 8) * 0x0001000100010001ULL; + c->cy = (int16_t)roundToInt16(cy <<13); + c->oy = (int16_t)roundToInt16(oy <<9); + c->cvr= (int16_t)roundToInt16(crv<<13); + c->cvg= (int16_t)roundToInt16(cgv<<13); + c->cug= (int16_t)roundToInt16(cgu<<13); + c->cub= (int16_t)roundToInt16(cbu<<13); + yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation); //FIXME factorize @@ -1993,7 +2073,13 @@ av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be choosen\n"); return NULL; } - +if( dstFormat != PIX_FMT_RGB32 //HACK + && dstFormat != PIX_FMT_RGB32_1 + && dstFormat != PIX_FMT_RGB24 + && dstFormat != PIX_FMT_BGR24 + && dstFormat != PIX_FMT_BGR32 + && dstFormat != PIX_FMT_BGR32_1) + flags &= ~SWS_FULL_CHR_H_INT; /* sanity check */ if (srcW<4 || srcH<1 || dstW<8 || dstH<1) //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code diff -r a917f63f61d3 -r 78da4c8b2293 libswscale/swscale_template.c --- a/libswscale/swscale_template.c Thu Sep 11 03:22:39 2008 +0000 +++ b/libswscale/swscale_template.c Thu Sep 11 14:39:12 2008 +0000 @@ -3132,8 +3132,15 @@ if (vLumFilterSize == 1 && vChrFilterSize == 2) //unscaled RGB { int chrAlpha= vChrFilter[2*dstY+1]; + if(flags & SWS_FULL_CHR_H_INT){ + yuv2rgbXinC_full(c, //FIXME write a packed1_full function + vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, + vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, + dest, dstW, dstY); + }else{ RENAME(yuv2packed1)(c, *lumSrcPtr, *chrSrcPtr, *(chrSrcPtr+1), dest, dstW, chrAlpha, dstFormat, flags, dstY); + } } else if (vLumFilterSize == 2 && vChrFilterSize == 2) //bilinear upscale RGB { @@ -3143,15 +3150,29 @@ lumMmxFilter[3]= vLumFilter[2*dstY ]*0x10001; chrMmxFilter[2]= chrMmxFilter[3]= vChrFilter[2*chrDstY]*0x10001; + if(flags & SWS_FULL_CHR_H_INT){ + yuv2rgbXinC_full(c, //FIXME write a packed2_full function + vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, + vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, + dest, dstW, dstY); + }else{ RENAME(yuv2packed2)(c, *lumSrcPtr, *(lumSrcPtr+1), *chrSrcPtr, *(chrSrcPtr+1), dest, dstW, lumAlpha, chrAlpha, dstY); + } } else //general RGB { + if(flags & SWS_FULL_CHR_H_INT){ + yuv2rgbXinC_full(c, + vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, + vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, + dest, dstW, dstY); + }else{ RENAME(yuv2packedX)(c, vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, dest, dstW, dstY); + } } } } @@ -3180,10 +3201,17 @@ { assert(lumSrcPtr + vLumFilterSize - 1 < lumPixBuf + vLumBufSize*2); assert(chrSrcPtr + vChrFilterSize - 1 < chrPixBuf + vChrBufSize*2); + if(flags & SWS_FULL_CHR_H_INT){ + yuv2rgbXinC_full(c, + vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, + vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, + dest, dstW, dstY); + }else{ yuv2packedXinC(c, vLumFilter+dstY*vLumFilterSize, lumSrcPtr, vLumFilterSize, vChrFilter+dstY*vChrFilterSize, chrSrcPtr, vChrFilterSize, dest, dstW, dstY); + } } } }