X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libswscale%2Futils.c;h=b644ed9610fa798429e8f185d82d17a47bb1391d;hb=b4dc68803bd7fa73c2e37f55ae73b5fdb293c68d;hp=18d1227c830ceac030a9b3696bd3efe9845db438;hpb=ac0fb5934893be554a44d2a1eb7a3bc7bf39da4a;p=ffmpeg diff --git a/libswscale/utils.c b/libswscale/utils.c index 18d1227c830..b644ed9610f 100644 --- a/libswscale/utils.c +++ b/libswscale/utils.c @@ -18,6 +18,7 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#define _SVID_SOURCE //needed for MAP_ANONYMOUS #include #include #include @@ -42,6 +43,7 @@ #include "libavutil/cpu.h" #include "libavutil/avutil.h" #include "libavutil/bswap.h" +#include "libavutil/mathematics.h" #include "libavutil/opt.h" #include "libavutil/pixdesc.h" @@ -63,133 +65,99 @@ const char *swscale_license(void) #define RET 0xC3 //near return opcode for x86 -#define isSupportedIn(x) ( \ - (x)==PIX_FMT_YUV420P \ - || (x)==PIX_FMT_YUVA420P \ - || (x)==PIX_FMT_YUYV422 \ - || (x)==PIX_FMT_UYVY422 \ - || (x)==PIX_FMT_RGB48BE \ - || (x)==PIX_FMT_RGB48LE \ - || (x)==PIX_FMT_RGB32 \ - || (x)==PIX_FMT_RGB32_1 \ - || (x)==PIX_FMT_BGR48BE \ - || (x)==PIX_FMT_BGR48LE \ - || (x)==PIX_FMT_BGR24 \ - || (x)==PIX_FMT_BGR565LE \ - || (x)==PIX_FMT_BGR565BE \ - || (x)==PIX_FMT_BGR555LE \ - || (x)==PIX_FMT_BGR555BE \ - || (x)==PIX_FMT_BGR32 \ - || (x)==PIX_FMT_BGR32_1 \ - || (x)==PIX_FMT_RGB24 \ - || (x)==PIX_FMT_RGB565LE \ - || (x)==PIX_FMT_RGB565BE \ - || (x)==PIX_FMT_RGB555LE \ - || (x)==PIX_FMT_RGB555BE \ - || (x)==PIX_FMT_GRAY8 \ - || (x)==PIX_FMT_Y400A \ - || (x)==PIX_FMT_YUV410P \ - || (x)==PIX_FMT_YUV440P \ - || (x)==PIX_FMT_NV12 \ - || (x)==PIX_FMT_NV21 \ - || (x)==PIX_FMT_GRAY16BE \ - || (x)==PIX_FMT_GRAY16LE \ - || (x)==PIX_FMT_YUV444P \ - || (x)==PIX_FMT_YUV422P \ - || (x)==PIX_FMT_YUV411P \ - || (x)==PIX_FMT_YUVJ420P \ - || (x)==PIX_FMT_YUVJ422P \ - || (x)==PIX_FMT_YUVJ440P \ - || (x)==PIX_FMT_YUVJ444P \ - || (x)==PIX_FMT_PAL8 \ - || (x)==PIX_FMT_BGR8 \ - || (x)==PIX_FMT_RGB8 \ - || (x)==PIX_FMT_BGR4_BYTE \ - || (x)==PIX_FMT_RGB4_BYTE \ - || (x)==PIX_FMT_YUV440P \ - || (x)==PIX_FMT_MONOWHITE \ - || (x)==PIX_FMT_MONOBLACK \ - || (x)==PIX_FMT_YUV420P9LE \ - || (x)==PIX_FMT_YUV444P9LE \ - || (x)==PIX_FMT_YUV420P10LE \ - || (x)==PIX_FMT_YUV422P10LE \ - || (x)==PIX_FMT_YUV444P10LE \ - || (x)==PIX_FMT_YUV420P16LE \ - || (x)==PIX_FMT_YUV422P16LE \ - || (x)==PIX_FMT_YUV444P16LE \ - || (x)==PIX_FMT_YUV420P9BE \ - || (x)==PIX_FMT_YUV444P9BE \ - || (x)==PIX_FMT_YUV420P10BE \ - || (x)==PIX_FMT_YUV444P10BE \ - || (x)==PIX_FMT_YUV422P10BE \ - || (x)==PIX_FMT_YUV420P16BE \ - || (x)==PIX_FMT_YUV422P16BE \ - || (x)==PIX_FMT_YUV444P16BE \ - ) +typedef struct FormatEntry { + int is_supported_in, is_supported_out; +} FormatEntry; + +const static FormatEntry format_entries[PIX_FMT_NB] = { + [PIX_FMT_YUV420P] = { 1 , 1 }, + [PIX_FMT_YUYV422] = { 1 , 1 }, + [PIX_FMT_RGB24] = { 1 , 1 }, + [PIX_FMT_BGR24] = { 1 , 1 }, + [PIX_FMT_YUV422P] = { 1 , 1 }, + [PIX_FMT_YUV444P] = { 1 , 1 }, + [PIX_FMT_YUV410P] = { 1 , 1 }, + [PIX_FMT_YUV411P] = { 1 , 1 }, + [PIX_FMT_GRAY8] = { 1 , 1 }, + [PIX_FMT_MONOWHITE] = { 1 , 1 }, + [PIX_FMT_MONOBLACK] = { 1 , 1 }, + [PIX_FMT_PAL8] = { 1 , 0 }, + [PIX_FMT_YUVJ420P] = { 1 , 1 }, + [PIX_FMT_YUVJ422P] = { 1 , 1 }, + [PIX_FMT_YUVJ444P] = { 1 , 1 }, + [PIX_FMT_UYVY422] = { 1 , 1 }, + [PIX_FMT_UYYVYY411] = { 0 , 0 }, + [PIX_FMT_BGR8] = { 1 , 1 }, + [PIX_FMT_BGR4] = { 0 , 1 }, + [PIX_FMT_BGR4_BYTE] = { 1 , 1 }, + [PIX_FMT_RGB8] = { 1 , 1 }, + [PIX_FMT_RGB4] = { 0 , 1 }, + [PIX_FMT_RGB4_BYTE] = { 1 , 1 }, + [PIX_FMT_NV12] = { 1 , 1 }, + [PIX_FMT_NV21] = { 1 , 1 }, + [PIX_FMT_ARGB] = { 1 , 1 }, + [PIX_FMT_RGBA] = { 1 , 1 }, + [PIX_FMT_ABGR] = { 1 , 1 }, + [PIX_FMT_BGRA] = { 1 , 1 }, + [PIX_FMT_GRAY16BE] = { 1 , 1 }, + [PIX_FMT_GRAY16LE] = { 1 , 1 }, + [PIX_FMT_YUV440P] = { 1 , 1 }, + [PIX_FMT_YUVJ440P] = { 1 , 1 }, + [PIX_FMT_YUVA420P] = { 1 , 1 }, + [PIX_FMT_RGB48BE] = { 1 , 1 }, + [PIX_FMT_RGB48LE] = { 1 , 1 }, + [PIX_FMT_RGB565BE] = { 1 , 1 }, + [PIX_FMT_RGB565LE] = { 1 , 1 }, + [PIX_FMT_RGB555BE] = { 1 , 1 }, + [PIX_FMT_RGB555LE] = { 1 , 1 }, + [PIX_FMT_BGR565BE] = { 1 , 1 }, + [PIX_FMT_BGR565LE] = { 1 , 1 }, + [PIX_FMT_BGR555BE] = { 1 , 1 }, + [PIX_FMT_BGR555LE] = { 1 , 1 }, + [PIX_FMT_YUV420P16LE] = { 1 , 1 }, + [PIX_FMT_YUV420P16BE] = { 1 , 1 }, + [PIX_FMT_YUV422P16LE] = { 1 , 1 }, + [PIX_FMT_YUV422P16BE] = { 1 , 1 }, + [PIX_FMT_YUV444P16LE] = { 1 , 1 }, + [PIX_FMT_YUV444P16BE] = { 1 , 1 }, + [PIX_FMT_RGB444LE] = { 0 , 1 }, + [PIX_FMT_RGB444BE] = { 0 , 1 }, + [PIX_FMT_BGR444LE] = { 0 , 1 }, + [PIX_FMT_BGR444BE] = { 0 , 1 }, + [PIX_FMT_Y400A] = { 1 , 0 }, + [PIX_FMT_BGR48BE] = { 1 , 1 }, + [PIX_FMT_BGR48LE] = { 1 , 1 }, + [PIX_FMT_YUV420P9BE] = { 1 , 1 }, + [PIX_FMT_YUV420P9LE] = { 1 , 1 }, + [PIX_FMT_YUV420P10BE] = { 1 , 1 }, + [PIX_FMT_YUV420P10LE] = { 1 , 1 }, + [PIX_FMT_YUV422P9BE] = { 1 , 1 }, + [PIX_FMT_YUV422P9LE] = { 1 , 1 }, + [PIX_FMT_YUV422P10BE] = { 1 , 1 }, + [PIX_FMT_YUV422P10LE] = { 1 , 1 }, + [PIX_FMT_YUV444P9BE] = { 1 , 1 }, + [PIX_FMT_YUV444P9LE] = { 1 , 1 }, + [PIX_FMT_YUV444P10BE] = { 1 , 1 }, + [PIX_FMT_YUV444P10LE] = { 1 , 1 }, + [PIX_FMT_GBRP] = { 1 , 0 }, + [PIX_FMT_GBRP9LE] = { 1 , 0 }, + [PIX_FMT_GBRP9BE] = { 1 , 0 }, + [PIX_FMT_GBRP10LE] = { 1 , 0 }, + [PIX_FMT_GBRP10BE] = { 1 , 0 }, + [PIX_FMT_GBRP16LE] = { 1 , 0 }, + [PIX_FMT_GBRP16BE] = { 1 , 0 }, +}; int sws_isSupportedInput(enum PixelFormat pix_fmt) { - return isSupportedIn(pix_fmt); + return (unsigned)pix_fmt < PIX_FMT_NB ? + format_entries[pix_fmt].is_supported_in : 0; } -#define isSupportedOut(x) ( \ - (x)==PIX_FMT_YUV420P \ - || (x)==PIX_FMT_YUVA420P \ - || (x)==PIX_FMT_YUYV422 \ - || (x)==PIX_FMT_UYVY422 \ - || (x)==PIX_FMT_YUV444P \ - || (x)==PIX_FMT_YUV422P \ - || (x)==PIX_FMT_YUV411P \ - || (x)==PIX_FMT_YUVJ420P \ - || (x)==PIX_FMT_YUVJ422P \ - || (x)==PIX_FMT_YUVJ440P \ - || (x)==PIX_FMT_YUVJ444P \ - || isRGBinBytes(x) \ - || isBGRinBytes(x) \ - || (x)==PIX_FMT_RGB565LE \ - || (x)==PIX_FMT_RGB565BE \ - || (x)==PIX_FMT_RGB555LE \ - || (x)==PIX_FMT_RGB555BE \ - || (x)==PIX_FMT_RGB444LE \ - || (x)==PIX_FMT_RGB444BE \ - || (x)==PIX_FMT_BGR565LE \ - || (x)==PIX_FMT_BGR565BE \ - || (x)==PIX_FMT_BGR555LE \ - || (x)==PIX_FMT_BGR555BE \ - || (x)==PIX_FMT_BGR444LE \ - || (x)==PIX_FMT_BGR444BE \ - || (x)==PIX_FMT_RGB8 \ - || (x)==PIX_FMT_BGR8 \ - || (x)==PIX_FMT_RGB4_BYTE \ - || (x)==PIX_FMT_BGR4_BYTE \ - || (x)==PIX_FMT_RGB4 \ - || (x)==PIX_FMT_BGR4 \ - || (x)==PIX_FMT_MONOBLACK \ - || (x)==PIX_FMT_MONOWHITE \ - || (x)==PIX_FMT_NV12 \ - || (x)==PIX_FMT_NV21 \ - || (x)==PIX_FMT_GRAY16BE \ - || (x)==PIX_FMT_GRAY16LE \ - || (x)==PIX_FMT_GRAY8 \ - || (x)==PIX_FMT_YUV410P \ - || (x)==PIX_FMT_YUV440P \ - || (x)==PIX_FMT_YUV420P9LE \ - || (x)==PIX_FMT_YUV420P10LE \ - || (x)==PIX_FMT_YUV420P16LE \ - || (x)==PIX_FMT_YUV422P10LE \ - || (x)==PIX_FMT_YUV422P16LE \ - || (x)==PIX_FMT_YUV444P16LE \ - || (x)==PIX_FMT_YUV420P9BE \ - || (x)==PIX_FMT_YUV420P10BE \ - || (x)==PIX_FMT_YUV422P10BE \ - || (x)==PIX_FMT_YUV420P16BE \ - || (x)==PIX_FMT_YUV422P16BE \ - || (x)==PIX_FMT_YUV444P16BE \ - ) - int sws_isSupportedOutput(enum PixelFormat pix_fmt) { - return isSupportedOut(pix_fmt); + return (unsigned)pix_fmt < PIX_FMT_NB ? + format_entries[pix_fmt].is_supported_out : 0; } extern const int32_t ff_yuv2rgb_coeffs[8][4]; @@ -227,8 +195,8 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions) - // NOTE: the +1 is for the MMX scaler which reads over the end - FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail); + // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end + FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail); if (FFABS(xInc - 0x10000) <10) { // unscaled int i; @@ -316,15 +284,18 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi if (flags & SWS_BICUBIC) { int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24); int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24); - int64_t dd = ( d*d)>>30; - int64_t ddd= (dd*d)>>30; - if (d < 1LL<<30) - coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30); - else if (d < 1LL<<31) - coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30); - else - coeff=0.0; + if (d >= 1LL<<31) { + coeff = 0.0; + } else { + int64_t dd = (d * d) >> 30; + int64_t ddd = (dd * d) >> 30; + + if (d < 1LL<<30) + coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30); + else + coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30); + } coeff *= fone>>(30+24); } /* else if (flags & SWS_X) { @@ -514,7 +485,7 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi // Note the +1 is for the MMX scaler which reads over the end /* align at 16 for AltiVec (needed by hScale_altivec_real) */ - FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail); + FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail); /* normalize & store in outFilter */ for (i=0; iscalingBpp = 1 + FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1, - av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1); - if (c->scalingBpp <= 8) - c->scalingBpp = 8; - if (c->scalingBpp == 16) + c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1; + if (c->srcBpc < 8) + c->srcBpc = 8; + c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1; + if (c->dstBpc < 8) + c->dstBpc = 8; + if (c->dstBpc == 16) dst_stride <<= 1; FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, - FFALIGN(srcW, 16) * 2 * FFALIGN(c->scalingBpp, 8) >> 3, + (FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3) + 16, fail); - if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) { + if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) { c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0; if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) { if (flags&SWS_PRINT_INFO) @@ -994,7 +971,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter) /* precalculate vertical scaler filter coefficients */ { const int filterAlign= - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 : + (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 2 : (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 : 1; @@ -1058,8 +1035,8 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter) FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail); c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize]; } - // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate) - c->uv_off_px = dst_stride_px + 64 / (c->scalingBpp &~ 7); + // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate) + c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7); c->uv_off_byte = dst_stride + 16; for (i=0; ivChrBufSize; i++) { FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail); @@ -1109,72 +1086,6 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter) else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n"); else av_log(c, AV_LOG_INFO, "using C\n"); - if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) { - if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR)) - av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n"); - else { - if (c->hLumFilterSize==4) - av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n"); - else if (c->hLumFilterSize==8) - av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n"); - else - av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n"); - - if (c->hChrFilterSize==4) - av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n"); - else if (c->hChrFilterSize==8) - av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n"); - else - av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n"); - } - } else { -#if HAVE_MMX - av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n"); -#else - if (flags & SWS_FAST_BILINEAR) - av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n"); - else - av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n"); -#endif - } - if (isPlanarYUV(dstFormat)) { - if (c->vLumFilterSize==1) - av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else - av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - } else { - if (c->vLumFilterSize==1 && c->vChrFilterSize==2) - av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n" - " 2-tap scaler for vertical chrominance scaling (BGR)\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else if (c->vLumFilterSize==2 && c->vChrFilterSize==2) - av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else - av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - } - - if (dstFormat==PIX_FMT_BGR24) - av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n", - (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" : - ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C")); - else if (dstFormat==PIX_FMT_RGB32) - av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else if (dstFormat==PIX_FMT_BGR565) - av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else if (dstFormat==PIX_FMT_BGR555) - av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE || - dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE) - av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", - (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"); - av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH); av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n", c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);