September 2013 - ffmpeg-cvslog

avcodec/ff_init_vlc_sparse: use a spinlock for thread sync
by git＠videolan.org 17 Jun '14

17 Jun '14

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Sat Jul 13 13:32:39 2013 +0200| [6af8326354ed6c1c68b53b3f2bba6697fb2d3bff] | committer: Michael Niedermayer avcodec/ff_init_vlc_sparse: use a spinlock for thread sync Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=6af8326354ed6c1c6… --- libavcodec/bitstream.c | 28 +++++++++++++++++++--------- libavcodec/get_bits.h | 1 + 2 files changed, 20 insertions(+), 9 deletions(-) diff --git a/libavcodec/bitstream.c b/libavcodec/bitstream.c index 3c4d176..aac32a5 100644 --- a/libavcodec/bitstream.c +++ b/libavcodec/bitstream.c @@ -28,6 +28,7 @@ * bitstream api. */ +#include "libavutil/atomic.h" #include "libavutil/avassert.h" #include "avcodec.h" #include "mathops.h" @@ -269,14 +270,17 @@ int ff_init_vlc_sparse(VLC *vlc, int nb_bits, int nb_codes, { VLCcode *buf; int i, j, ret; + void *state; vlc->bits = nb_bits; if (flags & INIT_VLC_USE_NEW_STATIC) { - if (vlc->table_size && vlc->table_size == vlc->table_allocated) { - return 0; - } else if(vlc->table_size) { - abort(); // fatal error, we are called on a partially initialized table + while (state = avpriv_atomic_ptr_cas(&vlc->init_state, NULL, vlc)) { + if (state == vlc + 1) { + av_assert0(vlc->table_size && vlc->table_size == vlc->table_allocated); + return 0; + } } + av_assert0(!vlc->table_size); } else { vlc->table = NULL; vlc->table_allocated = 0; @@ -326,12 +330,18 @@ int ff_init_vlc_sparse(VLC *vlc, int nb_bits, int nb_codes, ret = build_table(vlc, nb_bits, nb_codes, buf, flags); av_free(buf); - if (ret < 0) { - av_freep(&vlc->table); - return ret; + if (flags & INIT_VLC_USE_NEW_STATIC) { + if(vlc->table_size != vlc->table_allocated) + av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", vlc->table_size, vlc->table_allocated); + state = avpriv_atomic_ptr_cas(&vlc->init_state, vlc, vlc+1); + av_assert0(state == vlc); + av_assert0(ret >= 0); + } else { + if (ret < 0) { + av_freep(&vlc->table); + return ret; + } } - if((flags & INIT_VLC_USE_NEW_STATIC) && vlc->table_size != vlc->table_allocated) - av_log(NULL, AV_LOG_ERROR, "needed %d had %d\n", vlc->table_size, vlc->table_allocated); return 0; } diff --git a/libavcodec/get_bits.h b/libavcodec/get_bits.h index b6cc75a..3967b49 100644 --- a/libavcodec/get_bits.h +++ b/libavcodec/get_bits.h @@ -64,6 +64,7 @@ typedef struct VLC { int bits; VLC_TYPE (*table)[2]; ///< code, bits int table_size, table_allocated; + void *init_state; } VLC; typedef struct RL_VLC_ELEM {

4 5

Rewrite emu_edge functions to have separate src/dst_stride arguments.
by git＠videolan.org 23 Oct '13

23 Oct '13

ffmpeg | branch: master | Ronald S. Bultje <rsbultje(a)gmail.com> | Fri Sep 20 08:01:19 2013 -0400| [face578d56c2d1375e40d5e2a28acc122132bc55] | committer: Ronald S. Bultje Rewrite emu_edge functions to have separate src/dst_stride arguments. This allows supporting files for which the image stride is smaller than the max. block size + number of subpel mc taps, e.g. a 64x64 VP9 file or a 16x16 VP8 file with -fflags +emu_edge. > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=face578d56c2d1375… --- libavcodec/cavs.c | 7 +- libavcodec/diracdec.c | 3 +- libavcodec/h264.c | 10 +- libavcodec/mpegvideo.c | 16 +- libavcodec/mpegvideo_enc.c | 6 +- libavcodec/mpegvideo_motion.c | 74 ++-- libavcodec/rv34.c | 15 +- libavcodec/snow.c | 4 +- libavcodec/svq3.c | 6 +- libavcodec/vc1dec.c | 47 ++- libavcodec/videodsp.h | 20 +- libavcodec/videodsp_template.c | 32 +- libavcodec/vp3.c | 5 +- libavcodec/vp56.c | 2 +- libavcodec/vp8.c | 38 +- libavcodec/vp8.h | 2 +- libavcodec/wmv2.c | 16 +- libavcodec/x86/dsputil_mmx.c | 7 +- libavcodec/x86/videodsp.asm | 840 ++++++++++++++++------------------------ libavcodec/x86/videodsp_init.c | 181 +++++++-- 20 files changed, 657 insertions(+), 674 deletions(-) diff --git a/libavcodec/cavs.c b/libavcodec/cavs.c index 12af0c9..28327d7 100644 --- a/libavcodec/cavs.c +++ b/libavcodec/cavs.c @@ -405,7 +405,8 @@ static inline void mc_dir_part(AVSContext *h, AVFrame *pic, || full_my < 0-extra_height || full_mx + 16/*FIXME*/ > pic_width + extra_width || full_my + 16/*FIXME*/ > pic_height + extra_height){ - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_y - 2 - 2*h->l_stride, h->l_stride, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->l_stride, + src_y - 2 - 2*h->l_stride, h->l_stride, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height); src_y= h->edge_emu_buffer + 2 + 2*h->l_stride; emu=1; @@ -414,14 +415,14 @@ static inline void mc_dir_part(AVSContext *h, AVFrame *pic, qpix_op[luma_xy](dest_y, src_y, h->l_stride); //FIXME try variable height perhaps? if(emu){ - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->c_stride, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->c_stride, src_cb, h->c_stride, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); src_cb= h->edge_emu_buffer; } chroma_op(dest_cb, src_cb, h->c_stride, chroma_height, mx&7, my&7); if(emu){ - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->c_stride, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->c_stride, src_cr, h->c_stride, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1); src_cr= h->edge_emu_buffer; } diff --git a/libavcodec/diracdec.c b/libavcodec/diracdec.c index e5239a7..3d70f17 100644 --- a/libavcodec/diracdec.c +++ b/libavcodec/diracdec.c @@ -1414,7 +1414,8 @@ static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5], y + p->yblen > p->height+EDGE_WIDTH/2 || x < 0 || y < 0) { for (i = 0; i < nplanes; i++) { - ff_emulated_edge_mc(s->edge_emu_buffer[i], src[i], p->stride, + ff_emulated_edge_mc(s->edge_emu_buffer[i], p->stride, + src[i], p->stride, p->xblen, p->yblen, x, y, p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2); src[i] = s->edge_emu_buffer[i]; diff --git a/libavcodec/h264.c b/libavcodec/h264.c index acaab39..f2ee328 100644 --- a/libavcodec/h264.c +++ b/libavcodec/h264.c @@ -932,7 +932,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, full_my < 0 - extra_height || full_mx + 16 /*FIXME*/ > pic_width + extra_width || full_my + 16 /*FIXME*/ > pic_height + extra_height) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize, src_y - (2 << pixel_shift) - 2 * h->mb_linesize, h->mb_linesize, 16 + 5, 16 + 5 /*FIXME*/, full_mx - 2, @@ -951,7 +951,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, if (chroma_idc == 3 /* yuv444 */) { src_cb = pic->f.data[1] + offset; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize, src_cb - (2 << pixel_shift) - 2 * h->mb_linesize, h->mb_linesize, 16 + 5, 16 + 5 /*FIXME*/, @@ -965,7 +965,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, src_cr = pic->f.data[2] + offset; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_linesize, src_cr - (2 << pixel_shift) - 2 * h->mb_linesize, h->mb_linesize, 16 + 5, 16 + 5 /*FIXME*/, @@ -992,7 +992,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, (my >> ysh) * h->mb_uvlinesize; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cb, h->mb_uvlinesize, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_uvlinesize, src_cb, h->mb_uvlinesize, 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh), pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */)); src_cb = h->edge_emu_buffer; @@ -1002,7 +1002,7 @@ static av_always_inline void mc_dir_part(H264Context *h, Picture *pic, mx & 7, (my << (chroma_idc == 2 /* yuv422 */)) & 7); if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src_cr, h->mb_uvlinesize, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->mb_uvlinesize, src_cr, h->mb_uvlinesize, 9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh), pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */)); src_cr = h->edge_emu_buffer; diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c index 96afca5..32d6fa7 100644 --- a/libavcodec/mpegvideo.c +++ b/libavcodec/mpegvideo.c @@ -2204,7 +2204,8 @@ static inline int hpel_motion_lowres(MpegEncContext *s, if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w, 0) || (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, w + 1, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + src, s->linesize, w + 1, (h + 1) << field_based, src_x, src_y << field_based, h_edge_pos, @@ -2306,18 +2307,20 @@ static av_always_inline void mpeg_motion_lowres(MpegEncContext *s, if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s, 0) || uvsrc_y<0 || (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, linesize >> field_based, ptr_y, linesize >> field_based, 17, 17 + field_based, src_x, src_y << field_based, h_edge_pos, v_edge_pos); ptr_y = s->edge_emu_buffer; if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) { uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize; - s->vdsp.emulated_edge_mc(uvbuf , ptr_cb, uvlinesize >> field_based, 9, + s->vdsp.emulated_edge_mc(uvbuf, uvlinesize >> field_based, + ptr_cb, uvlinesize >> field_based, 9, 9 + field_based, uvsrc_x, uvsrc_y << field_based, h_edge_pos >> 1, v_edge_pos >> 1); - s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr, uvlinesize >> field_based, 9, + s->vdsp.emulated_edge_mc(uvbuf + 16, uvlinesize >> field_based, + ptr_cr, uvlinesize >> field_based, 9, 9 + field_based, uvsrc_x, uvsrc_y << field_based, h_edge_pos >> 1, v_edge_pos >> 1); @@ -2390,7 +2393,7 @@ static inline void chroma_4mv_motion_lowres(MpegEncContext *s, ptr = ref_picture[1] + offset; if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) || (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos); ptr = s->edge_emu_buffer; emu = 1; @@ -2401,7 +2404,8 @@ static inline void chroma_4mv_motion_lowres(MpegEncContext *s, ptr = ref_picture[2] + offset; if (emu) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, + ptr, s->uvlinesize, 9, 9, src_x, src_y, h_edge_pos, v_edge_pos); ptr = s->edge_emu_buffer; } diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c index 6f0d961..7b69aca 100644 --- a/libavcodec/mpegvideo_enc.c +++ b/libavcodec/mpegvideo_enc.c @@ -1819,14 +1819,14 @@ static av_always_inline void encode_mb_internal(MpegEncContext *s, uint8_t *ebuf = s->edge_emu_buffer + 32; int cw = (s->width + s->chroma_x_shift) >> s->chroma_x_shift; int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift; - s->vdsp.emulated_edge_mc(ebuf, ptr_y, wrap_y, 16, 16, mb_x * 16, + s->vdsp.emulated_edge_mc(ebuf, wrap_y, ptr_y, wrap_y, 16, 16, mb_x * 16, mb_y * 16, s->width, s->height); ptr_y = ebuf; - s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y, ptr_cb, wrap_c, mb_block_width, + s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y, wrap_c, ptr_cb, wrap_c, mb_block_width, mb_block_height, mb_x * mb_block_width, mb_y * mb_block_height, cw, ch); ptr_cb = ebuf + 18 * wrap_y; - s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 16, ptr_cr, wrap_c, mb_block_width, + s->vdsp.emulated_edge_mc(ebuf + 18 * wrap_y + 16, wrap_c, ptr_cr, wrap_c, mb_block_width, mb_block_height, mb_x * mb_block_width, mb_y * mb_block_height, cw, ch); ptr_cr = ebuf + 18 * wrap_y + 16; diff --git a/libavcodec/mpegvideo_motion.c b/libavcodec/mpegvideo_motion.c index e5b7e46..09061d8 100644 --- a/libavcodec/mpegvideo_motion.c +++ b/libavcodec/mpegvideo_motion.c @@ -63,7 +63,9 @@ static void gmc1_motion(MpegEncContext *s, if( (unsigned)src_x >= FFMAX(s->h_edge_pos - 17, 0) || (unsigned)src_y >= FFMAX(s->v_edge_pos - 17, 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, linesize, + ptr, linesize, 17, 17, src_x, src_y, + s->h_edge_pos, s->v_edge_pos); ptr= s->edge_emu_buffer; } @@ -100,7 +102,9 @@ static void gmc1_motion(MpegEncContext *s, ptr = ref_picture[1] + offset; if( (unsigned)src_x >= FFMAX((s->h_edge_pos>>1) - 9, 0) || (unsigned)src_y >= FFMAX((s->v_edge_pos>>1) - 9, 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, uvlinesize, + ptr, uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; emu=1; } @@ -108,7 +112,9 @@ static void gmc1_motion(MpegEncContext *s, ptr = ref_picture[2] + offset; if(emu){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, uvlinesize, + ptr, uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; } s->dsp.gmc1(dest_cr, ptr, uvlinesize, 8, motion_x&15, motion_y&15, 128 - s->no_rounding); @@ -196,8 +202,9 @@ static inline int hpel_motion(MpegEncContext *s, if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){ if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&1) - 8, 0) || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&1) - 8, 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize, 9, 9, - src_x, src_y, s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + src, s->linesize, 9, 9, + src_x, src_y, s->h_edge_pos, s->v_edge_pos); src= s->edge_emu_buffer; emu=1; } @@ -287,19 +294,19 @@ if(s->quarter_sample) "MPEG motion vector out of boundary (%d %d)\n", src_x, src_y); return; } - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, - 17, 17+field_based, - src_x, src_y<<field_based, - s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + ptr_y, s->linesize, 17, 17+field_based, + src_x, src_y<<field_based, + s->h_edge_pos, s->v_edge_pos); ptr_y = s->edge_emu_buffer; if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ uint8_t *uvbuf= s->edge_emu_buffer+18*s->linesize; - s->vdsp.emulated_edge_mc(uvbuf , + s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, ptr_cb, s->uvlinesize, 9, 9+field_based, uvsrc_x, uvsrc_y<<field_based, s->h_edge_pos>>1, s->v_edge_pos>>1); - s->vdsp.emulated_edge_mc(uvbuf+16, + s->vdsp.emulated_edge_mc(uvbuf+16, s->uvlinesize, ptr_cr, s->uvlinesize, 9, 9+field_based, uvsrc_x, uvsrc_y<<field_based, @@ -501,20 +508,23 @@ static inline void qpel_motion(MpegEncContext *s, if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 16, 0) || (unsigned)src_y > FFMAX( v_edge_pos - (motion_y&3) - h , 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y, s->linesize, - 17, 17+field_based, src_x, src_y<<field_based, - s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + ptr_y, s->linesize, + 17, 17+field_based, src_x, src_y<<field_based, + s->h_edge_pos, s->v_edge_pos); ptr_y= s->edge_emu_buffer; if(!CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){ uint8_t *uvbuf= s->edge_emu_buffer + 18*s->linesize; - s->vdsp.emulated_edge_mc(uvbuf, ptr_cb, s->uvlinesize, - 9, 9 + field_based, - uvsrc_x, uvsrc_y<<field_based, - s->h_edge_pos>>1, s->v_edge_pos>>1); - s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr, s->uvlinesize, - 9, 9 + field_based, - uvsrc_x, uvsrc_y<<field_based, - s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, + ptr_cb, s->uvlinesize, + 9, 9 + field_based, + uvsrc_x, uvsrc_y<<field_based, + s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, + ptr_cr, s->uvlinesize, + 9, 9 + field_based, + uvsrc_x, uvsrc_y<<field_based, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr_cb= uvbuf; ptr_cr= uvbuf + 16; } @@ -581,9 +591,9 @@ static void chroma_4mv_motion(MpegEncContext *s, if(s->flags&CODEC_FLAG_EMU_EDGE){ if( (unsigned)src_x > FFMAX((s->h_edge_pos>>1) - (dxy &1) - 8, 0) || (unsigned)src_y > FFMAX((s->v_edge_pos>>1) - (dxy>>1) - 8, 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, - 9, 9, src_x, src_y, - s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, + ptr, s->uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; emu=1; } @@ -592,9 +602,9 @@ static void chroma_4mv_motion(MpegEncContext *s, ptr = ref_picture[2] + offset; if(emu){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, - 9, 9, src_x, src_y, - s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, + ptr, s->uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; } pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8); @@ -761,10 +771,10 @@ static av_always_inline void MPV_motion_internal(MpegEncContext *s, if(s->flags&CODEC_FLAG_EMU_EDGE){ if( (unsigned)src_x > FFMAX(s->h_edge_pos - (motion_x&3) - 8, 0) || (unsigned)src_y > FFMAX(s->v_edge_pos - (motion_y&3) - 8, 0)){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, - s->linesize, 9, 9, - src_x, src_y, - s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + ptr, s->linesize, 9, 9, + src_x, src_y, + s->h_edge_pos, s->v_edge_pos); ptr= s->edge_emu_buffer; } } diff --git a/libavcodec/rv34.c b/libavcodec/rv34.c index 8834032..f84fb00 100644 --- a/libavcodec/rv34.c +++ b/libavcodec/rv34.c @@ -724,13 +724,16 @@ static inline void rv34_mc(RV34DecContext *r, const int block_type, uint8_t *uvbuf = s->edge_emu_buffer + 22 * s->linesize; srcY -= 2 + 2*s->linesize; - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, (width<<3)+6, (height<<3)+6, - src_x - 2, src_y - 2, s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, srcY, s->linesize, + (width<<3)+6, (height<<3)+6, src_x - 2, src_y - 2, + s->h_edge_pos, s->v_edge_pos); srcY = s->edge_emu_buffer + 2 + 2*s->linesize; - s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, (width<<2)+1, (height<<2)+1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); - s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, (width<<2)+1, (height<<2)+1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, s->v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, srcU, s->uvlinesize, + (width<<2)+1, (height<<2)+1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, s->v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, srcV, s->uvlinesize, + (width<<2)+1, (height<<2)+1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, s->v_edge_pos >> 1); srcU = uvbuf; srcV = uvbuf + 16; } diff --git a/libavcodec/snow.c b/libavcodec/snow.c index dc5c78e..b54c491 100644 --- a/libavcodec/snow.c +++ b/libavcodec/snow.c @@ -349,7 +349,9 @@ void ff_snow_pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, ptrdiff_t st src += sx + sy*stride; if( (unsigned)sx >= FFMAX(w - b_w - (HTAPS_MAX-2), 0) || (unsigned)sy >= FFMAX(h - b_h - (HTAPS_MAX-2), 0)){ - s->vdsp.emulated_edge_mc(tmp + MB_SIZE, src, stride, b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, sx, sy, w, h); + s->vdsp.emulated_edge_mc(tmp + MB_SIZE, stride, src, stride, + b_w+HTAPS_MAX-1, b_h+HTAPS_MAX-1, + sx, sy, w, h); src= tmp + MB_SIZE; } diff --git a/libavcodec/svq3.c b/libavcodec/svq3.c index 9326165..6988a1d 100644 --- a/libavcodec/svq3.c +++ b/libavcodec/svq3.c @@ -317,7 +317,8 @@ static inline void svq3_mc_dir_part(SVQ3Context *s, src = pic->f.data[0] + mx + my * h->linesize; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->linesize, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->linesize, + src, h->linesize, width + 1, height + 1, mx, my, s->h_edge_pos, s->v_edge_pos); src = h->edge_emu_buffer; @@ -343,7 +344,8 @@ static inline void svq3_mc_dir_part(SVQ3Context *s, src = pic->f.data[i] + mx + my * h->uvlinesize; if (emu) { - h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->uvlinesize, + h->vdsp.emulated_edge_mc(h->edge_emu_buffer, h->uvlinesize, + src, h->uvlinesize, width + 1, height + 1, mx, my, (s->h_edge_pos >> 1), s->v_edge_pos >> 1); diff --git a/libavcodec/vc1dec.c b/libavcodec/vc1dec.c index 439d68e..399e0d2 100644 --- a/libavcodec/vc1dec.c +++ b/libavcodec/vc1dec.c @@ -454,15 +454,18 @@ static void vc1_mc_1mv(VC1Context *v, int dir) uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize; srcY -= s->mspel * (1 + s->linesize); - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + srcY, s->linesize, 17 + s->mspel * 2, 17 + s->mspel * 2, src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, v_edge_pos); srcY = s->edge_emu_buffer; - s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); - s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, srcU, s->uvlinesize, + 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, srcV, s->uvlinesize, + 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, v_edge_pos >> 1); srcU = uvbuf; srcV = uvbuf + 16; /* if we deal with range reduction we need to scale source blocks */ @@ -697,7 +700,7 @@ static void vc1_mc_4mv_luma(VC1Context *v, int n, int dir, int avg) || (unsigned)(src_y - (s->mspel << fieldmv)) > v_edge_pos - (my & 3) - ((8 + s->mspel * 2) << fieldmv)) { srcY -= s->mspel * (1 + (s->linesize << fieldmv)); /* check emulate edge stride and offset */ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, srcY, s->linesize, 9 + s->mspel * 2, (9 + s->mspel * 2) << fieldmv, src_x - s->mspel, src_y - (s->mspel << fieldmv), s->h_edge_pos, v_edge_pos); @@ -912,11 +915,11 @@ static void vc1_mc_4mv_chroma(VC1Context *v, int dir) || s->h_edge_pos < 18 || v_edge_pos < 18 || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 9 || (unsigned)uvsrc_y > (v_edge_pos >> 1) - 9) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer , srcU, s->uvlinesize, - 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, srcU, + s->uvlinesize, 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); - s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, - 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, s->uvlinesize, srcV, + s->uvlinesize, 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); srcU = s->edge_emu_buffer; srcV = s->edge_emu_buffer + 16; @@ -1033,12 +1036,12 @@ static void vc1_mc_4mv_chroma4(VC1Context *v, int dir, int dir2, int avg) || s->h_edge_pos < 10 || v_edge_pos < (5 << fieldmv) || (unsigned)uvsrc_x > (s->h_edge_pos >> 1) - 5 || (unsigned)uvsrc_y > v_edge_pos - (5 << fieldmv)) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcU, s->uvlinesize, - 5, (5 << fieldmv), uvsrc_x, uvsrc_y, - s->h_edge_pos >> 1, v_edge_pos); - s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, srcV, s->uvlinesize, - 5, (5 << fieldmv), uvsrc_x, uvsrc_y, - s->h_edge_pos >> 1, v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, srcU, + s->uvlinesize, 5, (5 << fieldmv), uvsrc_x, + uvsrc_y, s->h_edge_pos >> 1, v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer + 16, s->uvlinesize, srcV, + s->uvlinesize, 5, (5 << fieldmv), uvsrc_x, + uvsrc_y, s->h_edge_pos >> 1, v_edge_pos); srcU = s->edge_emu_buffer; srcV = s->edge_emu_buffer + 16; @@ -1966,15 +1969,17 @@ static void vc1_interp_mc(VC1Context *v) uint8_t *uvbuf = s->edge_emu_buffer + 19 * s->linesize; srcY -= s->mspel * (1 + s->linesize); - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, srcY, s->linesize, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, srcY, s->linesize, 17 + s->mspel * 2, 17 + s->mspel * 2, src_x - s->mspel, src_y - s->mspel, s->h_edge_pos, v_edge_pos); srcY = s->edge_emu_buffer; - s->vdsp.emulated_edge_mc(uvbuf , srcU, s->uvlinesize, 8 + 1, 8 + 1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); - s->vdsp.emulated_edge_mc(uvbuf + 16, srcV, s->uvlinesize, 8 + 1, 8 + 1, - uvsrc_x, uvsrc_y, s->h_edge_pos >> 1, v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf, s->uvlinesize, srcU, s->uvlinesize, + 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, v_edge_pos >> 1); + s->vdsp.emulated_edge_mc(uvbuf + 16, s->uvlinesize, srcV, s->uvlinesize, + 8 + 1, 8 + 1, uvsrc_x, uvsrc_y, + s->h_edge_pos >> 1, v_edge_pos >> 1); srcU = uvbuf; srcV = uvbuf + 16; /* if we deal with range reduction we need to scale source blocks */ diff --git a/libavcodec/videodsp.h b/libavcodec/videodsp.h index e397720..e85dff3 100644 --- a/libavcodec/videodsp.h +++ b/libavcodec/videodsp.h @@ -30,9 +30,10 @@ #include <stdint.h> #define EMULATED_EDGE(depth) \ -void ff_emulated_edge_mc_ ## depth (uint8_t *buf, const uint8_t *src, ptrdiff_t linesize,\ - int block_w, int block_h,\ - int src_x, int src_y, int w, int h); +void ff_emulated_edge_mc_ ## depth(uint8_t *dst, ptrdiff_t dst_stride, \ + const uint8_t *src, ptrdiff_t src_stride, \ + int block_w, int block_h,\ + int src_x, int src_y, int w, int h); EMULATED_EDGE(8) EMULATED_EDGE(16) @@ -42,10 +43,12 @@ typedef struct VideoDSPContext { * Copy a rectangular area of samples to a temporary buffer and replicate * the border samples. * - * @param buf destination buffer + * @param dst destination buffer + * @param dst_stride number of bytes between 2 vertically adjacent samples + * in destination buffer * @param src source buffer - * @param linesize number of bytes between 2 vertically adjacent samples - * in both the source and destination buffers + * @param src_stride number of bytes between 2 vertically adjacent samples + * in source buffer * @param block_w width of block * @param block_h height of block * @param src_x x coordinate of the top left sample of the block in the @@ -55,8 +58,9 @@ typedef struct VideoDSPContext { * @param w width of the source buffer * @param h height of the source buffer */ - void (*emulated_edge_mc)(uint8_t *buf, const uint8_t *src, - ptrdiff_t linesize, int block_w, int block_h, + void (*emulated_edge_mc)(uint8_t *dst, ptrdiff_t dst_stride, + const uint8_t *src, ptrdiff_t src_stride, + int block_w, int block_h, int src_x, int src_y, int w, int h); /** diff --git a/libavcodec/videodsp_template.c b/libavcodec/videodsp_template.c index 39aad4d..f7e7bfd 100644 --- a/libavcodec/videodsp_template.c +++ b/libavcodec/videodsp_template.c @@ -20,10 +20,10 @@ */ #include "bit_depth_template.c" -void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src, - ptrdiff_t linesize, - int block_w, int block_h, - int src_x, int src_y, int w, int h) +void FUNC(ff_emulated_edge_mc)(uint8_t *buf, ptrdiff_t buf_stride, + const uint8_t *src, ptrdiff_t src_stride, + int block_w, int block_h, + int src_x, int src_y, int w, int h) { int x, y; int start_y, start_x, end_y, end_x; @@ -32,12 +32,12 @@ void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src, return; if (src_y >= h) { - src -= src_y * linesize; - src += (h - 1) * linesize; + src -= src_y * src_stride; + src += (h - 1) * src_stride; src_y = h - 1; } else if (src_y <= -block_h) { - src -= src_y * linesize; - src += (1 - block_h) * linesize; + src -= src_y * src_stride; + src += (1 - block_h) * src_stride; src_y = 1 - block_h; } if (src_x >= w) { @@ -56,30 +56,30 @@ void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src, av_assert2(start_x < end_x && block_w); w = end_x - start_x; - src += start_y * linesize + start_x * sizeof(pixel); + src += start_y * src_stride + start_x * sizeof(pixel); buf += start_x * sizeof(pixel); // top for (y = 0; y < start_y; y++) { memcpy(buf, src, w * sizeof(pixel)); - buf += linesize; + buf += buf_stride; } // copy existing part for (; y < end_y; y++) { memcpy(buf, src, w * sizeof(pixel)); - src += linesize; - buf += linesize; + src += src_stride; + buf += buf_stride; } // bottom - src -= linesize; + src -= src_stride; for (; y < block_h; y++) { memcpy(buf, src, w * sizeof(pixel)); - buf += linesize; + buf += buf_stride; } - buf -= block_h * linesize + start_x * sizeof(pixel); + buf -= block_h * buf_stride + start_x * sizeof(pixel); while (block_h--) { pixel *bufp = (pixel *) buf; @@ -92,6 +92,6 @@ void FUNC(ff_emulated_edge_mc)(uint8_t *buf, const uint8_t *src, for (x = end_x; x < block_w; x++) { bufp[x] = bufp[end_x - 1]; } - buf += linesize; + buf += buf_stride; } } diff --git a/libavcodec/vp3.c b/libavcodec/vp3.c index 78c094c..e2e5ab9 100644 --- a/libavcodec/vp3.c +++ b/libavcodec/vp3.c @@ -1549,7 +1549,10 @@ static void render_slice(Vp3DecodeContext *s, int slice) uint8_t *temp= s->edge_emu_buffer; if(stride<0) temp -= 8*stride; - s->vdsp.emulated_edge_mc(temp, motion_source, stride, 9, 9, src_x, src_y, plane_width, plane_height); + s->vdsp.emulated_edge_mc(temp, stride, + motion_source, stride, + 9, 9, src_x, src_y, + plane_width, plane_height); motion_source= temp; } } diff --git a/libavcodec/vp56.c b/libavcodec/vp56.c index b5d1884..d7d08f2 100644 --- a/libavcodec/vp56.c +++ b/libavcodec/vp56.c @@ -339,7 +339,7 @@ static void vp56_mc(VP56Context *s, int b, int plane, uint8_t *src, if (x<0 || x+12>=s->plane_width[plane] || y<0 || y+12>=s->plane_height[plane]) { - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, stride, src + s->block_offset[b] + (dy-2)*stride + (dx-2), stride, 12, 12, x, y, s->plane_width[plane], diff --git a/libavcodec/vp8.c b/libavcodec/vp8.c index 48432c1..157cebc 100644 --- a/libavcodec/vp8.c +++ b/libavcodec/vp8.c @@ -45,7 +45,6 @@ static void free_buffers(VP8Context *s) pthread_mutex_destroy(&s->thread_data[i].lock); #endif av_freep(&s->thread_data[i].filter_strength); - av_freep(&s->thread_data[i].edge_emu_buffer); } av_freep(&s->thread_data); av_freep(&s->macroblocks_base); @@ -1186,7 +1185,7 @@ void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, uint8_t *src = ref->f->data[0]; if (AV_RN32A(mv)) { - + int src_linesize = linesize; int mx = (mv->x << 1)&7, mx_idx = subpel_idx[0][mx]; int my = (mv->y << 1)&7, my_idx = subpel_idx[0][my]; @@ -1198,12 +1197,15 @@ void vp8_mc_luma(VP8Context *s, VP8ThreadData *td, uint8_t *dst, src += y_off * linesize + x_off; if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { - s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize, - block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], + s->vdsp.emulated_edge_mc(td->edge_emu_buffer, 32, + src - my_idx * linesize - mx_idx, linesize, + block_w + subpel_idx[1][mx], + block_h + subpel_idx[1][my], x_off - mx_idx, y_off - my_idx, width, height); - src = td->edge_emu_buffer + mx_idx + linesize * my_idx; + src = td->edge_emu_buffer + mx_idx + 32 * my_idx; + src_linesize = 32; } - mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my); + mc_func[my_idx][mx_idx](dst, linesize, src, src_linesize, block_h, mx, my); } else { ff_thread_await_progress(ref, (3 + y_off + block_h) >> 4, 0); mc_func[0][0](dst, linesize, src + y_off * linesize + x_off, linesize, block_h, 0, 0); @@ -1248,17 +1250,21 @@ void vp8_mc_chroma(VP8Context *s, VP8ThreadData *td, uint8_t *dst1, uint8_t *dst ff_thread_await_progress(ref, (3 + y_off + block_h + subpel_idx[2][my]) >> 3, 0); if (x_off < mx_idx || x_off >= width - block_w - subpel_idx[2][mx] || y_off < my_idx || y_off >= height - block_h - subpel_idx[2][my]) { - s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src1 - my_idx * linesize - mx_idx, linesize, - block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], + s->vdsp.emulated_edge_mc(td->edge_emu_buffer, 32, + src1 - my_idx * linesize - mx_idx, linesize, + block_w + subpel_idx[1][mx], + block_h + subpel_idx[1][my], x_off - mx_idx, y_off - my_idx, width, height); - src1 = td->edge_emu_buffer + mx_idx + linesize * my_idx; - mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my); + src1 = td->edge_emu_buffer + mx_idx + 32 * my_idx; + mc_func[my_idx][mx_idx](dst1, linesize, src1, 32, block_h, mx, my); - s->vdsp.emulated_edge_mc(td->edge_emu_buffer, src2 - my_idx * linesize - mx_idx, linesize, - block_w + subpel_idx[1][mx], block_h + subpel_idx[1][my], + s->vdsp.emulated_edge_mc(td->edge_emu_buffer, 32, + src2 - my_idx * linesize - mx_idx, linesize, + block_w + subpel_idx[1][mx], + block_h + subpel_idx[1][my], x_off - mx_idx, y_off - my_idx, width, height); - src2 = td->edge_emu_buffer + mx_idx + linesize * my_idx; - mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my); + src2 = td->edge_emu_buffer + mx_idx + 32 * my_idx; + mc_func[my_idx][mx_idx](dst2, linesize, src2, 32, block_h, mx, my); } else { mc_func[my_idx][mx_idx](dst1, linesize, src1, linesize, block_h, mx, my); mc_func[my_idx][mx_idx](dst2, linesize, src2, linesize, block_h, mx, my); @@ -1944,10 +1950,6 @@ int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, s->linesize = curframe->tf.f->linesize[0]; s->uvlinesize = curframe->tf.f->linesize[1]; - if (!s->thread_data[0].edge_emu_buffer) - for (i = 0; i < MAX_THREADS; i++) - s->thread_data[i].edge_emu_buffer = av_malloc(21*s->linesize); - memset(s->top_nnz, 0, s->mb_width*sizeof(*s->top_nnz)); /* Zero macroblock structures for top/top-left prediction from outside the frame. */ if (!s->mb_layout) diff --git a/libavcodec/vp8.h b/libavcodec/vp8.h index 6b56e65..c9a7906 100644 --- a/libavcodec/vp8.h +++ b/libavcodec/vp8.h @@ -122,7 +122,7 @@ typedef struct VP8ThreadData { #endif int thread_mb_pos; // (mb_y << 16) | (mb_x & 0xFFFF) int wait_mb_pos; // What the current thread is waiting on. - uint8_t *edge_emu_buffer; + DECLARE_ALIGNED(16, uint8_t, edge_emu_buffer)[21*32]; VP8FilterStrength *filter_strength; } VP8ThreadData; diff --git a/libavcodec/wmv2.c b/libavcodec/wmv2.c index eea819b..8d30259 100644 --- a/libavcodec/wmv2.c +++ b/libavcodec/wmv2.c @@ -119,8 +119,10 @@ void ff_mspel_motion(MpegEncContext *s, if(src_x<1 || src_y<1 || src_x + 17 >= s->h_edge_pos || src_y + h+1 >= v_edge_pos){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr - 1 - s->linesize, s->linesize, 19, 19, - src_x-1, src_y-1, s->h_edge_pos, s->v_edge_pos); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->linesize, + ptr - 1 - s->linesize, s->linesize, 19, 19, + src_x-1, src_y-1, + s->h_edge_pos, s->v_edge_pos); ptr= s->edge_emu_buffer + 1 + s->linesize; emu=1; } @@ -159,16 +161,18 @@ void ff_mspel_motion(MpegEncContext *s, offset = (src_y * uvlinesize) + src_x; ptr = ref_picture[1] + offset; if(emu){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, - src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, + ptr, s->uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; } pix_op[1][dxy](dest_cb, ptr, uvlinesize, h >> 1); ptr = ref_picture[2] + offset; if(emu){ - s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, - src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1); + s->vdsp.emulated_edge_mc(s->edge_emu_buffer, s->uvlinesize, + ptr, s->uvlinesize, 9, 9, src_x, src_y, + s->h_edge_pos>>1, s->v_edge_pos>>1); ptr= s->edge_emu_buffer; } pix_op[1][dxy](dest_cr, ptr, uvlinesize, h >> 1); diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c index 421a6c2..9cbf58a 100644 --- a/libavcodec/x86/dsputil_mmx.c +++ b/libavcodec/x86/dsputil_mmx.c @@ -405,8 +405,9 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, } } -typedef void emulated_edge_mc_func(uint8_t *dst, const uint8_t *src, - ptrdiff_t linesize, int block_w, int block_h, +typedef void emulated_edge_mc_func(uint8_t *dst, ptrdiff_t dst_stride, + const uint8_t *src, ptrdiff_t src_linesize, + int block_w, int block_h, int src_x, int src_y, int w, int h); static av_always_inline void gmc(uint8_t *dst, uint8_t *src, @@ -454,7 +455,7 @@ static av_always_inline void gmc(uint8_t *dst, uint8_t *src, src += ix + iy * stride; if (need_emu) { - emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height); + emu_edge_fn(edge_buf, stride, src, stride, w + 1, h + 1, ix, iy, width, height); src = edge_buf; } diff --git a/libavcodec/x86/videodsp.asm b/libavcodec/x86/videodsp.asm index 3f2f01a..cc2105d 100644 --- a/libavcodec/x86/videodsp.asm +++ b/libavcodec/x86/videodsp.asm @@ -23,576 +23,394 @@ SECTION .text -; void ff_emu_edge_core(uint8_t *buf, const uint8_t *src, x86_reg linesize, -; x86_reg start_y, x86_reg end_y, x86_reg block_h, -; x86_reg start_x, x86_reg end_x, x86_reg block_w); -; -; The actual function itself is below. It basically wraps a very simple -; w = end_x - start_x -; if (w) { -; if (w > 22) { -; jump to the slow loop functions -; } else { -; jump to the fast loop functions -; } -; } -; -; ... and then the same for left/right extend also. See below for loop -; function implementations. Fast are fixed-width, slow is variable-width - -%macro EMU_EDGE_FUNC 0 -%if ARCH_X86_64 -%define w_reg r7 -cglobal emu_edge_core, 6, 9, 1 - mov r8, r5 ; save block_h -%else -%define w_reg r6 -cglobal emu_edge_core, 2, 7, 0 - mov r4, r4m ; end_y - mov r5, r5m ; block_h -%endif - - ; start with vertical extend (top/bottom) and body pixel copy - mov w_reg, r7m - sub w_reg, r6m ; w = start_x - end_x - sub r5, r4 -%if ARCH_X86_64 - sub r4, r3 -%else - sub r4, dword r3m -%endif - cmp w_reg, 22 - jg .slow_v_extend_loop -%if ARCH_X86_32 - mov r2, r2m ; linesize -%endif - sal w_reg, 7 ; w * 128 -%ifdef PIC - lea rax, [.emuedge_v_extend_1 - (.emuedge_v_extend_2 - .emuedge_v_extend_1)] - add w_reg, rax -%else - lea w_reg, [.emuedge_v_extend_1 - (.emuedge_v_extend_2 - .emuedge_v_extend_1)+w_reg] -%endif - call w_reg ; fast top extend, body copy and bottom extend -.v_extend_end: +; slow vertical extension loop function. Works with variable-width, and +; does per-line reading/writing of source data + +%macro V_COPY_ROW 2 ; type (top/body/bottom), h +.%1_y_loop: ; do { + mov wq, r7mp ; initialize w (r7mp = wmp) +.%1_x_loop: ; do { + movu m0, [srcq+wq] ; m0 = read($mmsize) + movu [dstq+wq], m0 ; write(m0, $mmsize) + add wq, mmsize ; w -= $mmsize + cmp wq, -mmsize ; } while (w > $mmsize); + jl .%1_x_loop + movu m0, [srcq-mmsize] ; m0 = read($mmsize) + movu [dstq-mmsize], m0 ; write(m0, $mmsize) +%ifidn %1, body ; if ($type == body) { + add srcq, src_strideq ; src += src_stride +%endif ; } + add dstq, dst_strideq ; dst += dst_stride + dec %2 ; } while (--$h); + jnz .%1_y_loop +%endmacro - ; horizontal extend (left/right) - mov w_reg, r6m ; start_x - sub r0, w_reg +%macro vvar_fn 0 +; .----. <- zero +; | | <- top is copied from first line in body of source +; |----| <- start_y +; | | <- body is copied verbatim (line-by-line) from source +; |----| <- end_y +; | | <- bottom is copied from last line in body of source +; '----' <- bh %if ARCH_X86_64 - mov r3, r0 ; backup of buf+block_h*linesize - mov r5, r8 -%else - mov r0m, r0 ; backup of buf+block_h*linesize - mov r5, r5m +cglobal emu_edge_vvar, 7, 8, 1, dst, dst_stride, src, src_stride, \ + start_y, end_y, bh, w +%else ; x86-32 +cglobal emu_edge_vvar, 1, 6, 1, dst, src, start_y, end_y, bh, w +%define src_strideq r3mp +%define dst_strideq r1mp + mov srcq, r2mp + mov start_yq, r4mp + mov end_yq, r5mp + mov bhq, r6mp %endif - test w_reg, w_reg - jz .right_extend - cmp w_reg, 22 - jg .slow_left_extend_loop - mov r1, w_reg - dec w_reg - ; FIXME we can do a if size == 1 here if that makes any speed difference, test me - sar w_reg, 1 - sal w_reg, 6 - ; r0=buf+block_h*linesize,r7(64)/r6(32)=start_x offset for funcs - ; r6(rax)/r3(ebx)=val,r2=linesize,r1=start_x,r5=block_h -%ifdef PIC - lea rax, [.emuedge_extend_left_2] - add w_reg, rax -%else - lea w_reg, [.emuedge_extend_left_2+w_reg] -%endif - call w_reg + sub bhq, end_yq ; bh -= end_q + sub end_yq, start_yq ; end_q -= start_q + add srcq, r7mp ; (r7mp = wmp) + add dstq, r7mp ; (r7mp = wmp) + neg r7mp ; (r7mp = wmp) + test start_yq, start_yq ; if (start_q) { + jz .body + V_COPY_ROW top, start_yq ; v_copy_row(top, start_yq) +.body: ; } + V_COPY_ROW body, end_yq ; v_copy_row(body, end_yq) + test bhq, bhq ; if (bh) { + jz .end + sub srcq, src_strideq ; src -= src_stride + V_COPY_ROW bottom, bhq ; v_copy_row(bottom, bh) +.end: ; } + RET +%endmacro - ; now r3(64)/r0(32)=buf,r2=linesize,r8/r5=block_h,r6/r3=val, r7/r6=end_x, r1=block_w -.right_extend: %if ARCH_X86_32 - mov r0, r0m - mov r5, r5m +INIT_MMX mmx +vvar_fn %endif - mov w_reg, r7m ; end_x - mov r1, r8m ; block_w - mov r4, r1 - sub r1, w_reg - jz .h_extend_end ; if (end_x == block_w) goto h_extend_end - cmp r1, 22 - jg .slow_right_extend_loop - dec r1 - ; FIXME we can do a if size == 1 here if that makes any speed difference, test me - sar r1, 1 - sal r1, 6 -%ifdef PIC - lea rax, [.emuedge_extend_right_2] - add r1, rax -%else - lea r1, [.emuedge_extend_right_2+r1] -%endif - call r1 -.h_extend_end: + +INIT_XMM sse +vvar_fn + +%macro hvar_fn 0 +cglobal emu_edge_hvar, 5, 6, 1, dst, dst_stride, start_x, n_words, h, w + lea dstq, [dstq+n_wordsq*2] + neg n_wordsq + lea start_xq, [start_xq+n_wordsq*2] +.y_loop: ; do { + ; FIXME also write a ssse3 version using pshufb + movzx wd, byte [dstq+start_xq] ; w = read(1) + imul wd, 0x01010101 ; w *= 0x01010101 + movd m0, wd + mov wq, n_wordsq ; initialize w +%if cpuflag(sse) + shufps m0, m0, q0000 ; splat +%else ; mmx + punpckldq m0, m0 ; splat +%endif ; mmx/sse +.x_loop: ; do { + movu [dstq+wq*2], m0 ; write($reg, $mmsize) + add wq, mmsize/2 ; w -= $mmsize/2 + cmp wq, -mmsize/2 ; } while (w > $mmsize/2) + jl .x_loop + movu [dstq-mmsize], m0 ; write($reg, $mmsize) + add dstq, dst_strideq ; dst += dst_stride + dec hq ; } while (h--) + jnz .y_loop RET +%endmacro -%if ARCH_X86_64 -%define vall al -%define valh ah -%define valw ax -%define valw2 r7w -%define valw3 r3w -%if WIN64 -%define valw4 r7w -%else ; unix64 -%define valw4 r3w -%endif -%define vald eax -%else -%define vall bl -%define valh bh -%define valw bx -%define valw2 r6w -%define valw3 valw2 -%define valw4 valw3 -%define vald ebx -%define stack_offset 0x14 +%if ARCH_X86_32 +INIT_MMX mmx +hvar_fn %endif -%endmacro +INIT_XMM sse +hvar_fn ; macro to read/write a horizontal number of pixels (%2) to/from registers -; on x86-64, - fills xmm0-15 for consecutive sets of 16 pixels -; - if (%2 & 15 == 8) fills the last 8 bytes into rax -; - else if (%2 & 8) fills 8 bytes into mm0 -; - if (%2 & 7 == 4) fills the last 4 bytes into rax -; - else if (%2 & 4) fills 4 bytes into mm0-1 -; - if (%2 & 3 == 3) fills 2 bytes into r7/r3, and 1 into eax -; (note that we're using r3 for body/bottom because it's a shorter -; opcode, and then the loop fits in 128 bytes) -; - else fills remaining bytes into rax -; on x86-32, - fills mm0-7 for consecutive sets of 8 pixels -; - if (%2 & 7 == 4) fills 4 bytes into ebx -; - else if (%2 & 4) fills 4 bytes into mm0-7 -; - if (%2 & 3 == 3) fills 2 bytes into r6, and 1 into ebx -; - else fills remaining bytes into ebx +; on sse, - fills xmm0-15 for consecutive sets of 16 pixels +; - if (%2 & 8) fills 8 bytes into xmm$next +; - if (%2 & 4) fills 4 bytes into xmm$next +; - if (%2 & 3) fills 1, 2 or 4 bytes in eax +; on mmx, - fills mm0-7 for consecutive sets of 8 pixels +; - if (%2 & 4) fills 4 bytes into mm$next +; - if (%2 & 3) fills 1, 2 or 4 bytes in eax ; writing data out is in the same way %macro READ_NUM_BYTES 2 -%assign %%src_off 0 ; offset in source buffer -%assign %%smidx 0 ; mmx register idx -%assign %%sxidx 0 ; xmm register idx - -%if cpuflag(sse) -%rep %2/16 - movups xmm %+ %%sxidx, [r1+%%src_off] -%assign %%src_off %%src_off+16 -%assign %%sxidx %%sxidx+1 -%endrep ; %2/16 +%assign %%off 0 ; offset in source buffer +%assign %%idx 0 ; mmx/xmm register index + +%rep %2/mmsize + movu m %+ %%idx, [srcq+%%off] +%assign %%off %%off+mmsize +%assign %%idx %%idx+1 +%endrep ; %2/mmsize + +%if mmsize == 16 +%if (%2-%%off) >= 8 +%if %2 > 16 && (%2-%%off) > 8 + movu m %+ %%idx, [srcq+%2-16] +%assign %%off %2 +%else + movq m %+ %%idx, [srcq+%%off] +%assign %%off %%off+8 +%endif +%assign %%idx %%idx+1 +%endif ; (%2-%%off) >= 8 %endif -%if ARCH_X86_64 -%if (%2-%%src_off) == 8 - mov rax, [r1+%%src_off] -%assign %%src_off %%src_off+8 -%endif ; (%2-%%src_off) == 8 -%endif ; x86-64 - -%rep (%2-%%src_off)/8 - movq mm %+ %%smidx, [r1+%%src_off] -%assign %%src_off %%src_off+8 -%assign %%smidx %%smidx+1 -%endrep ; (%2-%%dst_off)/8 - -%if (%2-%%src_off) == 4 - mov vald, [r1+%%src_off] -%elif (%2-%%src_off) & 4 - movd mm %+ %%smidx, [r1+%%src_off] -%assign %%src_off %%src_off+4 -%endif ; (%2-%%src_off) ==/& 4 - -%if (%2-%%src_off) == 1 - mov vall, [r1+%%src_off] -%elif (%2-%%src_off) == 2 - mov valw, [r1+%%src_off] -%elif (%2-%%src_off) == 3 -%ifidn %1, top - mov valw2, [r1+%%src_off] +%if (%2-%%off) >= 4 +%if %2 > 8 && (%2-%%off) > 4 + movq m %+ %%idx, [srcq+%2-8] +%assign %%off %2 +%else + movd m %+ %%idx, [srcq+%%off] +%assign %%off %%off+4 +%endif +%assign %%idx %%idx+1 +%endif ; (%2-%%off) >= 4 + +%if (%2-%%off) >= 1 +%if %2 >= 4 + movd m %+ %%idx, [srcq+%2-4] +%elif (%2-%%off) == 1 + mov valb, [srcq+%2-1] +%elif (%2-%%off) == 2 + mov valw, [srcq+%2-2] %elifidn %1, body - mov valw3, [r1+%%src_off] -%elifidn %1, bottom - mov valw4, [r1+%%src_off] -%endif ; %1 ==/!= top - mov vall, [r1+%%src_off+2] -%endif ; (%2-%%src_off) == 1/2/3 + mov vald, [srcq+%2-3] +%else + movd m %+ %%idx, [srcq+%2-3] +%endif +%endif ; (%2-%%off) >= 1 %endmacro ; READ_NUM_BYTES %macro WRITE_NUM_BYTES 2 -%assign %%dst_off 0 ; offset in destination buffer -%assign %%dmidx 0 ; mmx register idx -%assign %%dxidx 0 ; xmm register idx - -%if cpuflag(sse) -%rep %2/16 - movups [r0+%%dst_off], xmm %+ %%dxidx -%assign %%dst_off %%dst_off+16 -%assign %%dxidx %%dxidx+1 -%endrep ; %2/16 +%assign %%off 0 ; offset in destination buffer +%assign %%idx 0 ; mmx/xmm register index + +%rep %2/mmsize + movu [dstq+%%off], m %+ %%idx +%assign %%off %%off+mmsize +%assign %%idx %%idx+1 +%endrep ; %2/mmsize + +%if mmsize == 16 +%if (%2-%%off) >= 8 +%if %2 > 16 && (%2-%%off) > 8 + movu [dstq+%2-16], m %+ %%idx +%assign %%off %2 +%else + movq [dstq+%%off], m %+ %%idx +%assign %%off %%off+8 +%endif +%assign %%idx %%idx+1 +%endif ; (%2-%%off) >= 8 %endif -%if ARCH_X86_64 -%if (%2-%%dst_off) == 8 - mov [r0+%%dst_off], rax -%assign %%dst_off %%dst_off+8 -%endif ; (%2-%%dst_off) == 8 -%endif ; x86-64 - -%rep (%2-%%dst_off)/8 - movq [r0+%%dst_off], mm %+ %%dmidx -%assign %%dst_off %%dst_off+8 -%assign %%dmidx %%dmidx+1 -%endrep ; (%2-%%dst_off)/8 - -%if (%2-%%dst_off) == 4 - mov [r0+%%dst_off], vald -%elif (%2-%%dst_off) & 4 - movd [r0+%%dst_off], mm %+ %%dmidx -%assign %%dst_off %%dst_off+4 -%endif ; (%2-%%dst_off) ==/& 4 - -%if (%2-%%dst_off) == 1 - mov [r0+%%dst_off], vall -%elif (%2-%%dst_off) == 2 - mov [r0+%%dst_off], valw -%elif (%2-%%dst_off) == 3 -%ifidn %1, top - mov [r0+%%dst_off], valw2 +%if (%2-%%off) >= 4 +%if %2 > 8 && (%2-%%off) > 4 + movq [dstq+%2-8], m %+ %%idx +%assign %%off %2 +%else + movd [dstq+%%off], m %+ %%idx +%assign %%off %%off+4 +%endif +%assign %%idx %%idx+1 +%endif ; (%2-%%off) >= 4 + +%if (%2-%%off) >= 1 +%if %2 >= 4 + movd [dstq+%2-4], m %+ %%idx +%elif (%2-%%off) == 1 + mov [dstq+%2-1], valb +%elif (%2-%%off) == 2 + mov [dstq+%2-2], valw %elifidn %1, body - mov [r0+%%dst_off], valw3 -%elifidn %1, bottom - mov [r0+%%dst_off], valw4 -%endif ; %1 ==/!= top - mov [r0+%%dst_off+2], vall -%endif ; (%2-%%dst_off) == 1/2/3 + mov [dstq+%2-3], valw + shr vald, 16 + mov [dstq+%2-1], valb +%else + movd vald, m %+ %%idx + mov [dstq+%2-3], valw + shr vald, 16 + mov [dstq+%2-1], valb +%endif +%endif ; (%2-%%off) >= 1 %endmacro ; WRITE_NUM_BYTES ; vertical top/bottom extend and body copy fast loops ; these are function pointers to set-width line copy functions, i.e. ; they read a fixed number of pixels into set registers, and write ; those out into the destination buffer -; r0=buf,r1=src,r2=linesize,r3(64)/r3m(32)=start_x,r4=end_y,r5=block_h -; r6(eax/64)/r3(ebx/32)=val_reg -%macro VERTICAL_EXTEND 0 -%assign %%n 1 -%rep 22 -ALIGN 128 -.emuedge_v_extend_ %+ %%n: - ; extend pixels above body +%macro VERTICAL_EXTEND 2 +%assign %%n %1 +%rep 1+%2-%1 +%if %%n <= 3 %if ARCH_X86_64 - test r3 , r3 ; if (!start_y) - jz .emuedge_copy_body_ %+ %%n %+ _loop ; goto body -%else ; ARCH_X86_32 - cmp dword r3m, 0 - je .emuedge_copy_body_ %+ %%n %+ _loop -%endif ; ARCH_X86_64/32 - READ_NUM_BYTES top, %%n ; read bytes -.emuedge_extend_top_ %+ %%n %+ _loop: ; do { - WRITE_NUM_BYTES top, %%n ; write bytes - add r0 , r2 ; dst += linesize +cglobal emu_edge_vfix %+ %%n, 6, 8, 0, dst, dst_stride, src, src_stride, \ + start_y, end_y, val, bh + mov bhq, r6mp ; r6mp = bhmp +%else ; x86-32 +cglobal emu_edge_vfix %+ %%n, 0, 6, 0, val, dst, src, start_y, end_y, bh + mov dstq, r0mp + mov srcq, r2mp + mov start_yq, r4mp + mov end_yq, r5mp + mov bhq, r6mp +%define dst_strideq r1mp +%define src_strideq r3mp +%endif ; x86-64/32 +%else %if ARCH_X86_64 - dec r3d -%else ; ARCH_X86_32 - dec dword r3m -%endif ; ARCH_X86_64/32 - jnz .emuedge_extend_top_ %+ %%n %+ _loop ; } while (--start_y) +cglobal emu_edge_vfix %+ %%n, 7, 7, 1, dst, dst_stride, src, src_stride, \ + start_y, end_y, bh +%else ; x86-32 +cglobal emu_edge_vfix %+ %%n, 1, 5, 1, dst, src, start_y, end_y, bh + mov srcq, r2mp + mov start_yq, r4mp + mov end_yq, r5mp + mov bhq, r6mp +%define dst_strideq r1mp +%define src_strideq r3mp +%endif ; x86-64/32 +%endif + ; FIXME move this to c wrapper? + sub bhq, end_yq ; bh -= end_y + sub end_yq, start_yq ; end_y -= start_y + + ; extend pixels above body + test start_yq, start_yq ; if (start_y) { + jz .body_loop + READ_NUM_BYTES top, %%n ; $variable_regs = read($n) +.top_loop: ; do { + WRITE_NUM_BYTES top, %%n ; write($variable_regs, $n) + add dstq, dst_strideq ; dst += linesize + dec start_yq ; } while (--start_y) + jnz .top_loop ; } ; copy body pixels -.emuedge_copy_body_ %+ %%n %+ _loop: ; do { - READ_NUM_BYTES body, %%n ; read bytes - WRITE_NUM_BYTES body, %%n ; write bytes - add r0 , r2 ; dst += linesize - add r1 , r2 ; src += linesize - dec r4d - jnz .emuedge_copy_body_ %+ %%n %+ _loop ; } while (--end_y) +.body_loop: ; do { + READ_NUM_BYTES body, %%n ; $variable_regs = read($n) + WRITE_NUM_BYTES body, %%n ; write($variable_regs, $n) + add dstq, dst_strideq ; dst += dst_stride + add srcq, src_strideq ; src += src_stride + dec end_yq ; } while (--end_y) + jnz .body_loop ; copy bottom pixels - test r5 , r5 ; if (!block_h) - jz .emuedge_v_extend_end_ %+ %%n ; goto end - sub r1 , r2 ; src -= linesize - READ_NUM_BYTES bottom, %%n ; read bytes -.emuedge_extend_bottom_ %+ %%n %+ _loop: ; do { - WRITE_NUM_BYTES bottom, %%n ; write bytes - add r0 , r2 ; dst += linesize - dec r5d - jnz .emuedge_extend_bottom_ %+ %%n %+ _loop ; } while (--block_h) - -.emuedge_v_extend_end_ %+ %%n: -%if ARCH_X86_64 - ret -%else ; ARCH_X86_32 - rep ret -%endif ; ARCH_X86_64/32 + test bhq, bhq ; if (block_h) { + jz .end + sub srcq, src_strideq ; src -= linesize + READ_NUM_BYTES bottom, %%n ; $variable_regs = read($n) +.bottom_loop: ; do { + WRITE_NUM_BYTES bottom, %%n ; write($variable_regs, $n) + add dstq, dst_strideq ; dst += linesize + dec bhq ; } while (--bh) + jnz .bottom_loop ; } + +.end: + RET %assign %%n %%n+1 -%endrep -%endmacro VERTICAL_EXTEND +%endrep ; 1+%2-%1 +%endmacro ; VERTICAL_EXTEND + +INIT_MMX mmx +VERTICAL_EXTEND 1, 15 +%if ARCH_X86_32 +VERTICAL_EXTEND 16, 22 +%endif + +INIT_XMM sse +VERTICAL_EXTEND 16, 22 ; left/right (horizontal) fast extend functions ; these are essentially identical to the vertical extend ones above, ; just left/right separated because number of pixels to extend is ; obviously not the same on both sides. -; for reading, pixels are placed in eax (x86-64) or ebx (x86-64) in the -; lowest two bytes of the register (so val*0x0101), and are splatted -; into each byte of mm0 as well if n_pixels >= 8 %macro READ_V_PIXEL 2 - mov vall, %2 - mov valh, vall -%if %1 >= 8 - movd mm0, vald -%if cpuflag(mmxext) - pshufw mm0, mm0, 0 -%else ; mmx - punpcklwd mm0, mm0 - punpckldq mm0, mm0 -%endif ; sse -%endif ; %1 >= 8 -%endmacro - -%macro WRITE_V_PIXEL 2 -%assign %%dst_off 0 -%rep %1/8 - movq [%2+%%dst_off], mm0 -%assign %%dst_off %%dst_off+8 -%endrep -%if %1 & 4 +%if %1 == 2 + movzx valw, byte %2 + imul valw, 0x0101 +%else + movzx vald, byte %2 + imul vald, 0x01010101 %if %1 >= 8 - movd [%2+%%dst_off], mm0 -%else ; %1 < 8 - mov [%2+%%dst_off] , valw - mov [%2+%%dst_off+2], valw -%endif ; %1 >=/< 8 -%assign %%dst_off %%dst_off+4 -%endif ; %1 & 4 -%if %1&2 - mov [%2+%%dst_off], valw -%endif ; %1 & 2 -%endmacro - -; r0=buf+block_h*linesize, r1=start_x, r2=linesize, r5=block_h, r6/r3=val -%macro LEFT_EXTEND 0 -%assign %%n 2 -%rep 11 -ALIGN 64 -.emuedge_extend_left_ %+ %%n: ; do { - sub r0, r2 ; dst -= linesize - READ_V_PIXEL %%n, [r0+r1] ; read pixels - WRITE_V_PIXEL %%n, r0 ; write pixels - dec r5 - jnz .emuedge_extend_left_ %+ %%n ; } while (--block_h) -%if ARCH_X86_64 - ret -%else ; ARCH_X86_32 - rep ret -%endif ; ARCH_X86_64/32 -%assign %%n %%n+2 -%endrep -%endmacro ; LEFT_EXTEND - -; r3/r0=buf+block_h*linesize, r2=linesize, r8/r5=block_h, r0/r6=end_x, r6/r3=val -%macro RIGHT_EXTEND 0 -%assign %%n 2 -%rep 11 -ALIGN 64 -.emuedge_extend_right_ %+ %%n: ; do { -%if ARCH_X86_64 - sub r3, r2 ; dst -= linesize - READ_V_PIXEL %%n, [r3+w_reg-1] ; read pixels - WRITE_V_PIXEL %%n, r3+r4-%%n ; write pixels - dec r8 -%else ; ARCH_X86_32 - sub r0, r2 ; dst -= linesize - READ_V_PIXEL %%n, [r0+w_reg-1] ; read pixels - WRITE_V_PIXEL %%n, r0+r4-%%n ; write pixels - dec r5 -%endif ; ARCH_X86_64/32 - jnz .emuedge_extend_right_ %+ %%n ; } while (--block_h) -%if ARCH_X86_64 - ret -%else ; ARCH_X86_32 - rep ret -%endif ; ARCH_X86_64/32 -%assign %%n %%n+2 -%endrep - -%if ARCH_X86_32 -%define stack_offset 0x10 -%endif -%endmacro ; RIGHT_EXTEND - -; below follow the "slow" copy/extend functions, these act on a non-fixed -; width specified in a register, and run a loop to copy the full amount -; of bytes. They are optimized for copying of large amounts of pixels per -; line, so they unconditionally splat data into mm registers to copy 8 -; bytes per loop iteration. It could be considered to use xmm for x86-64 -; also, but I haven't optimized this as much (i.e. FIXME) -%macro V_COPY_NPX 4-5 -%if %0 == 4 - test w_reg, %4 - jz .%1_skip_%4_px -%else ; %0 == 5 -.%1_%4_px_loop: + movd m0, vald +%if mmsize == 16 + shufps m0, m0, q0000 +%else + punpckldq m0, m0 %endif - %3 %2, [r1+cnt_reg] - %3 [r0+cnt_reg], %2 - add cnt_reg, %4 -%if %0 == 5 - sub w_reg, %4 - test w_reg, %5 - jnz .%1_%4_px_loop +%endif ; %1 >= 8 %endif -.%1_skip_%4_px: -%endmacro +%endmacro ; READ_V_PIXEL -%macro V_COPY_ROW 2 -%ifidn %1, bottom - sub r1, linesize +%macro WRITE_V_PIXEL 2 +%assign %%off 0 +%rep %1/mmsize + movu [%2+%%off], m0 +%assign %%off %%off+mmsize +%endrep ; %1/mmsize + +%if mmsize == 16 +%if %1-%%off >= 8 +%if %1 > 16 && %1-%%off > 8 + movu [%2+%1-16], m0 +%assign %%off %1 +%else + movq [%2+%%off], m0 +%assign %%off %%off+8 %endif -.%1_copy_loop: - xor cnt_reg, cnt_reg -%if notcpuflag(sse) -%define linesize r2m - V_COPY_NPX %1, mm0, movq, 8, 0xFFFFFFF8 -%else ; sse - V_COPY_NPX %1, xmm0, movups, 16, 0xFFFFFFF0 -%if ARCH_X86_64 -%define linesize r2 - V_COPY_NPX %1, rax , mov, 8 -%else ; ARCH_X86_32 -%define linesize r2m - V_COPY_NPX %1, mm0, movq, 8 -%endif ; ARCH_X86_64/32 -%endif ; sse - V_COPY_NPX %1, vald, mov, 4 - V_COPY_NPX %1, valw, mov, 2 - V_COPY_NPX %1, vall, mov, 1 - mov w_reg, cnt_reg -%ifidn %1, body - add r1, linesize +%endif ; %1-%%off >= 8 %endif - add r0, linesize - dec %2 - jnz .%1_copy_loop -%endmacro -%macro SLOW_V_EXTEND 0 -.slow_v_extend_loop: -; r0=buf,r1=src,r2(64)/r2m(32)=linesize,r3(64)/r3m(32)=start_x,r4=end_y,r5=block_h -; r8(64)/r3(later-64)/r2(32)=cnt_reg,r6(64)/r3(32)=val_reg,r7(64)/r6(32)=w=end_x-start_x -%if ARCH_X86_64 - push r8 ; save old value of block_h - test r3, r3 -%define cnt_reg r8 - jz .do_body_copy ; if (!start_y) goto do_body_copy - V_COPY_ROW top, r3 +%if %1-%%off >= 4 +%if %1 > 8 %% %1-%%off > 4 + movq [%2+%1-8], m0 +%assign %%off %1 +%elif %1 >= 8 && %1-%%off >= 4 + movd [%2+%%off], m0 +%assign %%off %%off+4 %else - cmp dword r3m, 0 -%define cnt_reg r2 - je .do_body_copy ; if (!start_y) goto do_body_copy - V_COPY_ROW top, dword r3m + mov [%2+%%off], vald +%assign %%off %%off+4 %endif +%endif ; %1-%%off >= 4 -.do_body_copy: - V_COPY_ROW body, r4 - -%if ARCH_X86_64 - pop r8 ; restore old value of block_h -%define cnt_reg r3 -%endif - test r5, r5 -%if ARCH_X86_64 - jz .v_extend_end +%if %1-%%off >= 2 +%if %1 >= 8 + movd [%2+%1-4], m0 %else - jz .skip_bottom_extend + mov [%2+%%off], valw %endif - V_COPY_ROW bottom, r5 -%if ARCH_X86_32 -.skip_bottom_extend: - mov r2, r2m -%endif - jmp .v_extend_end -%endmacro +%endif ; (%1-%%off)/2 +%endmacro ; WRITE_V_PIXEL + +%macro H_EXTEND 2 +%assign %%n %1 +%rep 1+(%2-%1)/2 +cglobal emu_edge_hfix %+ %%n, 4, 5, 1, dst, dst_stride, start_x, bh, val +.loop_y: ; do { + READ_V_PIXEL %%n, [dstq+start_xq] ; $variable_regs = read($n) + WRITE_V_PIXEL %%n, dstq ; write($variable_regs, $n) + add dstq, dst_strideq ; dst += dst_stride + dec bhq ; } while (--bh) + jnz .loop_y + RET +%assign %%n %%n+2 +%endrep ; 1+(%2-%1)/2 +%endmacro ; H_EXTEND -%macro SLOW_LEFT_EXTEND 0 -.slow_left_extend_loop: -; r0=buf+block_h*linesize,r2=linesize,r6(64)/r3(32)=val,r5=block_h,r4=cntr,r7/r6=start_x - mov r4, 8 - sub r0, linesize - READ_V_PIXEL 8, [r0+w_reg] -.left_extend_8px_loop: - movq [r0+r4-8], mm0 - add r4, 8 - cmp r4, w_reg - jle .left_extend_8px_loop - sub r4, 8 - cmp r4, w_reg - jge .left_extend_loop_end -.left_extend_2px_loop: - mov [r0+r4], valw - add r4, 2 - cmp r4, w_reg - jl .left_extend_2px_loop -.left_extend_loop_end: - dec r5 - jnz .slow_left_extend_loop +INIT_MMX mmx +H_EXTEND 2, 14 %if ARCH_X86_32 - mov r2, r2m -%endif - jmp .right_extend -%endmacro - -%macro SLOW_RIGHT_EXTEND 0 -.slow_right_extend_loop: -; r3(64)/r0(32)=buf+block_h*linesize,r2=linesize,r4=block_w,r8(64)/r5(32)=block_h, -; r7(64)/r6(32)=end_x,r6/r3=val,r1=cntr -%if ARCH_X86_64 -%define buf_reg r3 -%define bh_reg r8 -%else -%define buf_reg r0 -%define bh_reg r5 +H_EXTEND 16, 22 %endif - lea r1, [r4-8] - sub buf_reg, linesize - READ_V_PIXEL 8, [buf_reg+w_reg-1] -.right_extend_8px_loop: - movq [buf_reg+r1], mm0 - sub r1, 8 - cmp r1, w_reg - jge .right_extend_8px_loop - add r1, 8 - cmp r1, w_reg - je .right_extend_loop_end -.right_extend_2px_loop: - sub r1, 2 - mov [buf_reg+r1], valw - cmp r1, w_reg - jg .right_extend_2px_loop -.right_extend_loop_end: - dec bh_reg - jnz .slow_right_extend_loop - jmp .h_extend_end -%endmacro - -%macro emu_edge 1 -INIT_XMM %1 -EMU_EDGE_FUNC -VERTICAL_EXTEND -LEFT_EXTEND -RIGHT_EXTEND -SLOW_V_EXTEND -SLOW_LEFT_EXTEND -SLOW_RIGHT_EXTEND -%endmacro -emu_edge sse -%if ARCH_X86_32 -emu_edge mmx -%endif +INIT_XMM sse +H_EXTEND 16, 22 %macro PREFETCH_FN 1 cglobal prefetch, 3, 3, 0, buf, stride, h diff --git a/libavcodec/x86/videodsp_init.c b/libavcodec/x86/videodsp_init.c index 17a9011..a709d03 100644 --- a/libavcodec/x86/videodsp_init.c +++ b/libavcodec/x86/videodsp_init.c @@ -30,32 +30,126 @@ #include "libavcodec/videodsp.h" #if HAVE_YASM -typedef void emu_edge_core_func(uint8_t *buf, const uint8_t *src, - x86_reg linesize, x86_reg start_y, - x86_reg end_y, x86_reg block_h, - x86_reg start_x, x86_reg end_x, - x86_reg block_w); -extern emu_edge_core_func ff_emu_edge_core_mmx; -extern emu_edge_core_func ff_emu_edge_core_sse; - -static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src, - ptrdiff_t linesize, - int block_w, int block_h, - int src_x, int src_y, - int w, int h, - emu_edge_core_func *core_fn) +typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride, + const uint8_t *src, x86_reg src_stride, + x86_reg start_y, x86_reg end_y, x86_reg bh); +typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride, + const uint8_t *src, x86_reg src_stride, + x86_reg start_y, x86_reg end_y, x86_reg bh, + x86_reg w); + +extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx; +#if ARCH_X86_32 +static emu_edge_vfix_func *vfixtbl_mmx[22] = { + &ff_emu_edge_vfix1_mmx, &ff_emu_edge_vfix2_mmx, &ff_emu_edge_vfix3_mmx, + &ff_emu_edge_vfix4_mmx, &ff_emu_edge_vfix5_mmx, &ff_emu_edge_vfix6_mmx, + &ff_emu_edge_vfix7_mmx, &ff_emu_edge_vfix8_mmx, &ff_emu_edge_vfix9_mmx, + &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx, + &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx, + &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx, + &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx, + &ff_emu_edge_vfix22_mmx +}; +#endif +extern emu_edge_vvar_func ff_emu_edge_vvar_mmx; +extern emu_edge_vfix_func ff_emu_edge_vfix16_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix17_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix18_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix19_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix20_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix21_sse; +extern emu_edge_vfix_func ff_emu_edge_vfix22_sse; +static emu_edge_vfix_func *vfixtbl_sse[22] = { + ff_emu_edge_vfix1_mmx, ff_emu_edge_vfix2_mmx, ff_emu_edge_vfix3_mmx, + ff_emu_edge_vfix4_mmx, ff_emu_edge_vfix5_mmx, ff_emu_edge_vfix6_mmx, + ff_emu_edge_vfix7_mmx, ff_emu_edge_vfix8_mmx, ff_emu_edge_vfix9_mmx, + ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx, + ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx, + ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse, + ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse, + ff_emu_edge_vfix22_sse +}; +extern emu_edge_vvar_func ff_emu_edge_vvar_sse; + +typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride, + x86_reg start_x, x86_reg bh); +typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride, + x86_reg start_x, x86_reg n_words, x86_reg bh); + +extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx; +#if ARCH_X86_32 +static emu_edge_hfix_func *hfixtbl_mmx[11] = { + ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, + ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx, + ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx, + ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx +}; +#endif +extern emu_edge_hvar_func ff_emu_edge_hvar_mmx; +extern emu_edge_hfix_func ff_emu_edge_hfix16_sse; +extern emu_edge_hfix_func ff_emu_edge_hfix18_sse; +extern emu_edge_hfix_func ff_emu_edge_hfix20_sse; +extern emu_edge_hfix_func ff_emu_edge_hfix22_sse; +static emu_edge_hfix_func *hfixtbl_sse[11] = { + ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, + ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx, + ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse, ff_emu_edge_hfix18_sse, + ff_emu_edge_hfix20_sse, ff_emu_edge_hfix22_sse +}; +extern emu_edge_hvar_func ff_emu_edge_hvar_sse; + +static av_always_inline void emulated_edge_mc(uint8_t *dst, ptrdiff_t dst_stride, + const uint8_t *src, ptrdiff_t src_stride, + x86_reg block_w, x86_reg block_h, + x86_reg src_x, x86_reg src_y, + x86_reg w, x86_reg h, + emu_edge_vfix_func **vfix_tbl, + emu_edge_vvar_func *v_extend_var, + emu_edge_hfix_func **hfix_tbl, + emu_edge_hvar_func *h_extend_var) { - int start_y, start_x, end_y, end_x, src_y_add = 0; + x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p; if(!w || !h) return; if (src_y >= h) { - src -= src_y*linesize; + src -= src_y*src_stride; src_y_add = h - 1; src_y = h - 1; } else if (src_y <= -block_h) { - src -= src_y*linesize; + src -= src_y*src_stride; src_y_add = 1 - block_h; src_y = 1 - block_h; } @@ -75,30 +169,59 @@ static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src, av_assert2(start_y < end_y && block_h > 0); // fill in the to-be-copied part plus all above/below - src += (src_y_add + start_y) * linesize + start_x; - buf += start_x; - core_fn(buf, src, linesize, start_y, end_y, - block_h, start_x, end_x, block_w); + src += (src_y_add + start_y) * src_stride + start_x; + w = end_x - start_x; + if (w <= 22) { + vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride, + start_y, end_y, block_h); + } else { + v_extend_var(dst + start_x, dst_stride, src, src_stride, + start_y, end_y, block_h, w); + } + + // fill left + if (start_x) { + if (start_x <= 22) { + hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h); + } else { + h_extend_var(dst, dst_stride, + start_x, (start_x + 1) >> 1, block_h); + } + } + + // fill right + p = block_w - end_x; + if (p) { + if (p <= 22) { + hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride, + -!(p & 1), block_h); + } else { + h_extend_var(dst + end_x - (p & 1), dst_stride, + -!(p & 1), (p + 1) >> 1, block_h); + } + } } #if ARCH_X86_32 -static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src, - ptrdiff_t linesize, +static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, ptrdiff_t buf_stride, + const uint8_t *src, ptrdiff_t src_stride, int block_w, int block_h, int src_x, int src_y, int w, int h) { - emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y, - w, h, &ff_emu_edge_core_mmx); + emulated_edge_mc(buf, buf_stride, src, src_stride, block_w, block_h, + src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx, + hfixtbl_mmx, &ff_emu_edge_hvar_mmx); } #endif -static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src, - ptrdiff_t linesize, +static av_noinline void emulated_edge_mc_sse(uint8_t *buf, ptrdiff_t buf_stride, + const uint8_t *src, ptrdiff_t src_stride, int block_w, int block_h, int src_x, int src_y, int w, int h) { - emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y, - w, h, &ff_emu_edge_core_sse); + emulated_edge_mc(buf, buf_stride, src, src_stride, block_w, block_h, + src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, + hfixtbl_sse, &ff_emu_edge_hvar_sse); } #endif /* HAVE_YASM */

4 5

avfilter: port pullup filter from libmpcodecs
by git＠videolan.org 20 Oct '13

20 Oct '13

ffmpeg | branch: master | Paul B Mahol <onemda(a)gmail.com> | Mon Jul 8 12:42:53 2013 +0000| [9c774459a95833014163ebfdf216860bc7fa14b0] | committer: Paul B Mahol avfilter: port pullup filter from libmpcodecs Signed-off-by: Paul B Mahol <onemda(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=9c774459a95833014… --- Changelog | 1 + LICENSE | 1 + configure | 1 + doc/filters.texi | 53 +++ libavfilter/Makefile | 1 + libavfilter/allfilters.c | 1 + libavfilter/version.h | 2 +- libavfilter/vf_pullup.c | 762 ++++++++++++++++++++++++++++++++++++++ libavfilter/vf_pullup.h | 71 ++++ libavfilter/x86/Makefile | 2 + libavfilter/x86/vf_pullup.asm | 178 +++++++++ libavfilter/x86/vf_pullup_init.c | 41 ++ 12 files changed, 1113 insertions(+), 1 deletion(-) diff --git a/Changelog b/Changelog index 904c36d..3e4653b 100644 --- a/Changelog +++ b/Changelog @@ -24,6 +24,7 @@ version <next> - changed DTS stream id in lavf mpeg ps muxer from 0x8a to 0x88, to be more consistent with other muxers. - adelay filter +- pullup filter ported from libmpcodecs version 2.0: diff --git a/LICENSE b/LICENSE index 12b08c2..6e56e3b 100644 --- a/LICENSE +++ b/LICENSE @@ -41,6 +41,7 @@ Specifically, the GPL parts of FFmpeg are - vf_perspective.c - vf_phase.c - vf_pp.c + - vf_pullup.c - vf_sab.c - vf_smartblur.c - vf_spp.c diff --git a/configure b/configure index 5401dc2..e0a6073 100755 --- a/configure +++ b/configure @@ -2221,6 +2221,7 @@ owdenoise_filter_deps="gpl" pan_filter_deps="swresample" phase_filter_deps="gpl" pp_filter_deps="gpl postproc" +pullup_filter_deps="gpl" removelogo_filter_deps="avcodec avformat swscale" sab_filter_deps="gpl swscale" scale_filter_deps="swscale" diff --git a/doc/filters.texi b/doc/filters.texi index 3404f8b..664012f 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -6184,6 +6184,59 @@ On this example the input file being processed is compared with the reference file @file{ref_movie.mpg}. The PSNR of each individual frame is stored in @file{stats.log}. +@section pullup + +Pulldown reversal (inverse telecine) filter, capable of handling mixed +hard-telecine, 24000/1001 fps progressive, and 30000/1001 fps progressive +content. + +The pullup filter is designed to take advantage of future context in making +its decisions. This filter is stateless in the sense that it does not lock +onto a pattern to follow, but it instead looks forward to the following +fields in order to identify matches and rebuild progressive frames. + +The filter accepts the following options: + +@table @option +@item jl +@item jr +@item jt +@item jb +These options set the amount of "junk" to ignore at the left, right, top, and +bottom of the image, respectively. Left and right are in units of 8 pixels, +while top and bottom are in units of 2 lines. +The default is 8 pixels on each side. + +@item sb +Set the strict breaks. Setting this option to 1 will reduce the chances of +filter generating an occasional mismatched frame, but it may also cause an +excessive number of frames to be dropped during high motion sequences. +Conversely, setting it to -1 will make filter match fields more easily. +This may help processing of video where there is slight blurring between +the fields, but may also cause there to be interlaced frames in the output. +Default value is @code{0}. + +@item mp +Set the metric plane to use. It accepts the following values: +@table @samp +@item l +Use luma plane. + +@item u +Use chroma blue plane. + +@item v +Use chroma red plane. +@end table + +This option may be set to use chroma plane instead of the default luma plane +for doing filter's computations. This may improve accuracy on very clean +source material, but more likely will decrease accuracy, especially if there +is chroma noise (rainbow effect) or any grayscale video. +The main purpose of setting @option{mp} to a chroma plane is to reduce CPU +load and make pullup usable in realtime on slow machines. +@end table + @section removelogo Suppress a TV station logo, using an image file to determine which diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 5a82c84..78b5f17 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -173,6 +173,7 @@ OBJS-$(CONFIG_PHASE_FILTER) += vf_phase.o OBJS-$(CONFIG_PIXDESCTEST_FILTER) += vf_pixdesctest.o OBJS-$(CONFIG_PP_FILTER) += vf_pp.o OBJS-$(CONFIG_PSNR_FILTER) += vf_psnr.o dualinput.o +OBJS-$(CONFIG_PULLUP_FILTER) += vf_pullup.o OBJS-$(CONFIG_REMOVELOGO_FILTER) += bbox.o lswsutils.o lavfutils.o vf_removelogo.o OBJS-$(CONFIG_ROTATE_FILTER) += vf_rotate.o OBJS-$(CONFIG_SEPARATEFIELDS_FILTER) += vf_separatefields.o diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index f7e4342..2825304 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -168,6 +168,7 @@ void avfilter_register_all(void) REGISTER_FILTER(PIXDESCTEST, pixdesctest, vf); REGISTER_FILTER(PP, pp, vf); REGISTER_FILTER(PSNR, psnr, vf); + REGISTER_FILTER(PULLUP, pullup, vf); REGISTER_FILTER(REMOVELOGO, removelogo, vf); REGISTER_FILTER(ROTATE, rotate, vf); REGISTER_FILTER(SAB, sab, vf); diff --git a/libavfilter/version.h b/libavfilter/version.h index f0d4952..20c15f6 100644 --- a/libavfilter/version.h +++ b/libavfilter/version.h @@ -30,7 +30,7 @@ #include "libavutil/avutil.h" #define LIBAVFILTER_VERSION_MAJOR 3 -#define LIBAVFILTER_VERSION_MINOR 85 +#define LIBAVFILTER_VERSION_MINOR 86 #define LIBAVFILTER_VERSION_MICRO 100 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \ diff --git a/libavfilter/vf_pullup.c b/libavfilter/vf_pullup.c new file mode 100644 index 0000000..b1e0b49 --- /dev/null +++ b/libavfilter/vf_pullup.c @@ -0,0 +1,762 @@ +/* + * Copyright (c) 2003 Rich Felker + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/avassert.h" +#include "libavutil/imgutils.h" +#include "libavutil/opt.h" +#include "libavutil/pixdesc.h" +#include "avfilter.h" +#include "formats.h" +#include "internal.h" +#include "video.h" +#include "vf_pullup.h" + +#define F_HAVE_BREAKS 1 +#define F_HAVE_AFFINITY 2 + +#define BREAK_LEFT 1 +#define BREAK_RIGHT 2 + +#define OFFSET(x) offsetof(PullupContext, x) +#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM + +static const AVOption pullup_options[] = { + { "jl", "set left junk size", OFFSET(junk_left), AV_OPT_TYPE_INT, {.i64=1}, 0, INT_MAX, FLAGS }, + { "jr", "set right junk size", OFFSET(junk_right), AV_OPT_TYPE_INT, {.i64=1}, 0, INT_MAX, FLAGS }, + { "jt", "set top junk size", OFFSET(junk_top), AV_OPT_TYPE_INT, {.i64=4}, 1, INT_MAX, FLAGS }, + { "jd", "set down junk size", OFFSET(junk_down), AV_OPT_TYPE_INT, {.i64=4}, 1, INT_MAX, FLAGS }, + { "sb", "set strict breaks", OFFSET(strict_breaks), AV_OPT_TYPE_INT, {.i64=0},-1, 1, FLAGS }, + { "mp", "set metric plane", OFFSET(metric_plane), AV_OPT_TYPE_INT, {.i64=0}, 0, 2, FLAGS, "mp" }, + { "y", "luma", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "mp" }, + { "u", "chroma blue", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "mp" }, + { "v", "chroma red", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, FLAGS, "mp" }, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(pullup); + +static int query_formats(AVFilterContext *ctx) +{ + static const enum AVPixelFormat pix_fmts[] = { + AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, + AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P, + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P, + AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, + AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P, + AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_GRAY8, + AV_PIX_FMT_NONE + }; + ff_set_common_formats(ctx, ff_make_format_list(pix_fmts)); + return 0; +} + +#define ABS(a) (((a) ^ ((a) >> 31)) - ((a) >> 31)) + +static int diff_c(const uint8_t *a, const uint8_t *b, int s) +{ + int i, j, diff = 0; + + for (i = 0; i < 4; i++) { + for (j = 0; j < 8; j++) + diff += ABS(a[j] - b[j]); + a += s; + b += s; + } + + return diff; +} + +static int comb_c(const uint8_t *a, const uint8_t *b, int s) +{ + int i, j, comb = 0; + + for (i = 0; i < 4; i++) { + for (j = 0; j < 8; j++) + comb += ABS((a[j] << 1) - b[j - s] - b[j ]) + + ABS((b[j] << 1) - a[j ] - a[j + s]); + a += s; + b += s; + } + + return comb; +} + +static int var_c(const uint8_t *a, const uint8_t *b, int s) +{ + int i, j, var = 0; + + for (i = 0; i < 3; i++) { + for (j = 0; j < 8; j++) + var += ABS(a[j] - a[j + s]); + a += s; + } + + return 4 * var; /* match comb scaling */ +} + +static int alloc_metrics(PullupContext *s, PullupField *f) +{ + f->diffs = av_calloc(FFALIGN(s->metric_length, 16), sizeof(*f->diffs)); + f->combs = av_calloc(FFALIGN(s->metric_length, 16), sizeof(*f->combs)); + f->vars = av_calloc(FFALIGN(s->metric_length, 16), sizeof(*f->vars)); + + if (!f->diffs || !f->combs || !f->vars) { + av_freep(&f->diffs); + av_freep(&f->combs); + av_freep(&f->vars); + return AVERROR(ENOMEM); + } + return 0; +} + +static PullupField *make_field_queue(PullupContext *s, int len) +{ + PullupField *head, *f; + + f = head = av_mallocz(sizeof(*head)); + if (!f) + return NULL; + + if (alloc_metrics(s, f) < 0) { + av_free(f); + return NULL; + } + + for (; len > 0; len--) { + f->next = av_mallocz(sizeof(*f->next)); + if (!f->next) + return NULL; + + f->next->prev = f; + f = f->next; + if (alloc_metrics(s, f) < 0) + return NULL; + } + + f->next = head; + head->prev = f; + + return head; +} + +static int config_input(AVFilterLink *inlink) +{ + AVFilterContext *ctx = inlink->dst; + PullupContext *s = ctx->priv; + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); + const int mp = s->metric_plane; + + s->planeheight[1] = s->planeheight[2] = FF_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h); + s->planeheight[0] = s->planeheight[3] = inlink->h; + s->planewidth[1] = s->planewidth[2] = FF_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w); + s->planewidth[0] = s->planewidth[3] = inlink->w; + s->nb_planes = av_pix_fmt_count_planes(inlink->format); + + s->metric_w = (s->planewidth[mp] - ((s->junk_left + s->junk_right) << 3)) >> 3; + s->metric_h = (s->planeheight[mp] - ((s->junk_top + s->junk_bottom) << 1)) >> 3; + s->metric_offset = (s->junk_left << 3) + (s->junk_top << 1) * s->planewidth[mp]; + s->metric_length = s->metric_w * s->metric_h; + + av_log(ctx, AV_LOG_DEBUG, "w: %d h: %d\n", s->metric_w, s->metric_h); + av_log(ctx, AV_LOG_DEBUG, "offset: %d length: %d\n", s->metric_offset, s->metric_length); + + s->head = make_field_queue(s, 8); + if (!s->head) + return AVERROR(ENOMEM); + + s->diff = diff_c; + s->comb = comb_c; + s->var = var_c; + + if (ARCH_X86) + ff_pullup_init_x86(s); + return 0; +} + +static int config_output(AVFilterLink *outlink) +{ + outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP; + return 0; +} + +static PullupBuffer *pullup_lock_buffer(PullupBuffer *b, int parity) +{ + if (!b) + return NULL; + + if ((parity + 1) & 1) + b->lock[0]++; + if ((parity + 1) & 2) + b->lock[1]++; + + return b; +} + +static void pullup_release_buffer(PullupBuffer *b, int parity) +{ + if (!b) + return; + + if ((parity + 1) & 1) + b->lock[0]--; + if ((parity + 1) & 2) + b->lock[1]--; +} + +static int alloc_buffer(PullupContext *s, PullupBuffer *b) +{ + int i; + + if (b->planes[0]) + return 0; + for (i = 0; i < s->nb_planes; i++) { + b->planes[i] = av_malloc(s->planeheight[i] * s->planewidth[i]); + } + + return 0; +} + +static PullupBuffer *pullup_get_buffer(PullupContext *s, int parity) +{ + int i; + + /* Try first to get the sister buffer for the previous field */ + if (parity < 2 && s->last && parity != s->last->parity + && !s->last->buffer->lock[parity]) { + alloc_buffer(s, s->last->buffer); + return pullup_lock_buffer(s->last->buffer, parity); + } + + /* Prefer a buffer with both fields open */ + for (i = 0; i < FF_ARRAY_ELEMS(s->buffers); i++) { + if (s->buffers[i].lock[0]) + continue; + if (s->buffers[i].lock[1]) + continue; + alloc_buffer(s, &s->buffers[i]); + return pullup_lock_buffer(&s->buffers[i], parity); + } + + if (parity == 2) + return 0; + + /* Search for any half-free buffer */ + for (i = 0; i < FF_ARRAY_ELEMS(s->buffers); i++) { + if (((parity + 1) & 1) && s->buffers[i].lock[0]) + continue; + if (((parity + 1) & 2) && s->buffers[i].lock[1]) + continue; + alloc_buffer(s, &s->buffers[i]); + return pullup_lock_buffer(&s->buffers[i], parity); + } + + return NULL; +} + +static int queue_length(PullupField *begin, PullupField *end) +{ + PullupField *f; + int count = 1; + + if (!begin || !end) + return 0; + + for (f = begin; f != end; f = f->next) + count++; + + return count; +} + +static int find_first_break(PullupField *f, int max) +{ + int i; + + for (i = 0; i < max; i++) { + if (f->breaks & BREAK_RIGHT || f->next->breaks & BREAK_LEFT) + return i + 1; + f = f->next; + } + + return 0; +} + +static void compute_breaks(PullupContext *s, PullupField *f0) +{ + PullupField *f1 = f0->next; + PullupField *f2 = f1->next; + PullupField *f3 = f2->next; + int i, l, max_l = 0, max_r = 0; + + if (f0->flags & F_HAVE_BREAKS) + return; + + f0->flags |= F_HAVE_BREAKS; + + /* Special case when fields are 100% identical */ + if (f0->buffer == f2->buffer && f1->buffer != f3->buffer) { + f2->breaks |= BREAK_RIGHT; + return; + } + + if (f0->buffer != f2->buffer && f1->buffer == f3->buffer) { + f1->breaks |= BREAK_LEFT; + return; + } + + for (i = 0; i < s->metric_length; i++) { + l = f2->diffs[i] - f3->diffs[i]; + + if ( l > max_l) + max_l = l; + if (-l > max_r) + max_r = -l; + } + + /* Don't get tripped up when differences are mostly quant error */ + if (max_l + max_r < 128) + return; + if (max_l > 4 * max_r) + f1->breaks |= BREAK_LEFT; + if (max_r > 4 * max_l) + f2->breaks |= BREAK_RIGHT; +} + +static void compute_affinity(PullupContext *s, PullupField *f) +{ + int i, max_l = 0, max_r = 0, l; + + if (f->flags & F_HAVE_AFFINITY) + return; + + f->flags |= F_HAVE_AFFINITY; + + if (f->buffer == f->next->next->buffer) { + f->affinity = 1; + f->next->affinity = 0; + f->next->next->affinity = -1; + f->next->flags |= F_HAVE_AFFINITY; + f->next->next->flags |= F_HAVE_AFFINITY; + return; + } + + for (i = 0; i < s->metric_length; i++) { + int v = f->vars[i]; + int lv = f->prev->vars[i]; + int rv = f->next->vars[i]; + int lc = f->combs[i] - (v + lv) + ABS(v - lv); + int rc = f->next->combs[i] - (v + rv) + ABS(v - rv); + + lc = FFMAX(lc, 0); + rc = FFMAX(rc, 0); + l = lc - rc; + + if ( l > max_l) + max_l = l; + if (-l > max_r) + max_r = -l; + } + + if (max_l + max_r < 64) + return; + + if (max_r > 6 * max_l) + f->affinity = -1; + else if (max_l > 6 * max_r) + f->affinity = 1; +} + +static int decide_frame_length(PullupContext *s) +{ + PullupField *f0 = s->first; + PullupField *f1 = f0->next; + PullupField *f2 = f1->next; + PullupField *f; + int i, l, n; + + if (queue_length(s->first, s->last) < 4) + return 0; + + f = s->first; + n = queue_length(f, s->last); + for (i = 0; i < n - 1; i++) { + if (i < n - 3) + compute_breaks(s, f); + + compute_affinity(s, f); + + f = f->next; + } + + if (f0->affinity == -1) + return 1; + + l = find_first_break(f0, 3); + + if (l == 1 && s->strict_breaks < 0) + l = 0; + + switch (l) { + case 1: + return 1 + (s->strict_breaks < 1 && f0->affinity == 1 && f1->affinity == -1); + case 2: + /* FIXME: strictly speaking, f0->prev is no longer valid... :) */ + if (s->strict_pairs + && (f0->prev->breaks & BREAK_RIGHT) && (f2->breaks & BREAK_LEFT) + && (f0->affinity != 1 || f1->affinity != -1) ) + return 1; + return 1 + (f1->affinity != 1); + case 3: + return 2 + (f2->affinity != 1); + default: + /* 9 possibilities covered before switch */ + if (f1->affinity == 1) + return 1; /* covers 6 */ + else if (f1->affinity == -1) + return 2; /* covers 6 */ + else if (f2->affinity == -1) { /* covers 2 */ + return (f0->affinity == 1) ? 3 : 1; + } else { + return 2; /* the remaining 6 */ + } + } +} + +static PullupFrame *pullup_get_frame(PullupContext *s) +{ + PullupFrame *fr = &s->frame; + int i, n = decide_frame_length(s); + int aff = s->first->next->affinity; + + av_assert1(n < FF_ARRAY_ELEMS(fr->ifields)); + if (!n || fr->lock) + return NULL; + + fr->lock++; + fr->length = n; + fr->parity = s->first->parity; + fr->buffer = 0; + + for (i = 0; i < n; i++) { + /* We cheat and steal the buffer without release+relock */ + fr->ifields[i] = s->first->buffer; + s->first->buffer = 0; + s->first = s->first->next; + } + + if (n == 1) { + fr->ofields[fr->parity ] = fr->ifields[0]; + fr->ofields[fr->parity ^ 1] = 0; + } else if (n == 2) { + fr->ofields[fr->parity ] = fr->ifields[0]; + fr->ofields[fr->parity ^ 1] = fr->ifields[1]; + } else if (n == 3) { + if (!aff) + aff = (fr->ifields[0] == fr->ifields[1]) ? -1 : 1; + fr->ofields[fr->parity ] = fr->ifields[1 + aff]; + fr->ofields[fr->parity ^ 1] = fr->ifields[1 ]; + } + + pullup_lock_buffer(fr->ofields[0], 0); + pullup_lock_buffer(fr->ofields[1], 1); + + if (fr->ofields[0] == fr->ofields[1]) { + fr->buffer = fr->ofields[0]; + pullup_lock_buffer(fr->buffer, 2); + return fr; + } + + return fr; +} + +static void pullup_release_frame(PullupFrame *f) +{ + int i; + + for (i = 0; i < f->length; i++) + pullup_release_buffer(f->ifields[i], f->parity ^ (i & 1)); + + pullup_release_buffer(f->ofields[0], 0); + pullup_release_buffer(f->ofields[1], 1); + + if (f->buffer) + pullup_release_buffer(f->buffer, 2); + f->lock--; +} + +static void compute_metric(PullupContext *s, int *dest, + PullupField *fa, int pa, PullupField *fb, int pb, + int (*func)(const uint8_t *, const uint8_t *, int)) +{ + int mp = s->metric_plane; + int xstep = 8; + int ystep = s->planewidth[mp] << 3; + int stride = s->planewidth[mp] << 1; /* field stride */ + int w = s->metric_w * xstep; + uint8_t *a, *b; + int x, y; + + if (!fa->buffer || !fb->buffer) + return; + + /* Shortcut for duplicate fields (e.g. from RFF flag) */ + if (fa->buffer == fb->buffer && pa == pb) { + memset(dest, 0, s->metric_length * sizeof(*dest)); + return; + } + + a = fa->buffer->planes[mp] + pa * s->planewidth[mp] + s->metric_offset; + b = fb->buffer->planes[mp] + pb * s->planewidth[mp] + s->metric_offset; + + for (y = 0; y < s->metric_h; y++) { + for (x = 0; x < w; x += xstep) + *dest++ = func(a + x, b + x, stride); + a += ystep; b += ystep; + } +} + +static int check_field_queue(PullupContext *s) +{ + int ret; + + if (s->head->next == s->first) { + PullupField *f = av_mallocz(sizeof(*f)); + + if (!f) + return AVERROR(ENOMEM); + + if ((ret = alloc_metrics(s, f)) < 0) { + av_free(f); + return ret; + } + + f->prev = s->head; + f->next = s->first; + s->head->next = f; + s->first->prev = f; + } + + return 0; +} + +static void pullup_submit_field(PullupContext *s, PullupBuffer *b, int parity) +{ + PullupField *f; + + /* Grow the circular list if needed */ + if (check_field_queue(s) < 0) + return; + + /* Cannot have two fields of same parity in a row; drop the new one */ + if (s->last && s->last->parity == parity) + return; + + f = s->head; + f->parity = parity; + f->buffer = pullup_lock_buffer(b, parity); + f->flags = 0; + f->breaks = 0; + f->affinity = 0; + + compute_metric(s, f->diffs, f, parity, f->prev->prev, parity, s->diff); + compute_metric(s, f->combs, parity ? f->prev : f, 0, parity ? f : f->prev, 1, s->comb); + compute_metric(s, f->vars, f, parity, f, -1, s->var); + emms_c(); + + /* Advance the circular list */ + if (!s->first) + s->first = s->head; + + s->last = s->head; + s->head = s->head->next; +} + +static void copy_field(PullupContext *s, + PullupBuffer *dst, PullupBuffer *src, int parity) +{ + uint8_t *dd, *ss; + int i; + + for (i = 0; i < s->nb_planes; i++) { + ss = src->planes[i] + parity * s->planewidth[i]; + dd = dst->planes[i] + parity * s->planewidth[i]; + + av_image_copy_plane(dd, s->planewidth[i] << 1, + ss, s->planewidth[i] << 1, + s->planewidth[i], s->planeheight[i] >> 1); + } +} + +static void pullup_pack_frame(PullupContext *s, PullupFrame *fr) +{ + int i; + + if (fr->buffer) + return; + + if (fr->length < 2) + return; /* FIXME: deal with this */ + + for (i = 0; i < 2; i++) { + if (fr->ofields[i]->lock[i^1]) + continue; + + fr->buffer = fr->ofields[i]; + pullup_lock_buffer(fr->buffer, 2); + copy_field(s, fr->buffer, fr->ofields[i^1], i^1); + return; + } + + fr->buffer = pullup_get_buffer(s, 2); + + copy_field(s, fr->buffer, fr->ofields[0], 0); + copy_field(s, fr->buffer, fr->ofields[1], 1); +} + +static int filter_frame(AVFilterLink *inlink, AVFrame *in) +{ + AVFilterContext *ctx = inlink->dst; + AVFilterLink *outlink = ctx->outputs[0]; + PullupContext *s = ctx->priv; + PullupBuffer *b; + PullupFrame *f; + AVFrame *out; + int p, ret = 0; + + b = pullup_get_buffer(s, 2); + if (!b) { + av_log(ctx, AV_LOG_WARNING, "Could not get buffer!\n"); + f = pullup_get_frame(s); + pullup_release_frame(f); + goto end; + } + + av_image_copy(b->planes, s->planewidth, + (const uint8_t**)in->data, in->linesize, + inlink->format, inlink->w, inlink->h); + + p = !!in->interlaced_frame; + pullup_submit_field(s, b, p ); + pullup_submit_field(s, b, p^1); + + if (in->repeat_pict) + pullup_submit_field(s, b, p); + + pullup_release_buffer(b, 2); + + f = pullup_get_frame(s); + if (!f) + goto end; + + if (f->length < 2) { + pullup_release_frame(f); + f = pullup_get_frame(s); + if (!f) + goto end; + if (f->length < 2) { + pullup_release_frame(f); + if (!in->repeat_pict) + goto end; + f = pullup_get_frame(s); + if (!f) + goto end; + if (f->length < 2) { + pullup_release_frame(f); + goto end; + } + } + } + + /* If the frame isn't already exportable... */ + if (!f->buffer) + pullup_pack_frame(s, f); + + out = ff_get_video_buffer(outlink, outlink->w, outlink->h); + if (!out) { + ret = AVERROR(ENOMEM); + goto end; + } + av_frame_copy_props(out, in); + + av_image_copy(out->data, out->linesize, + (const uint8_t**)f->buffer->planes, s->planewidth, + inlink->format, inlink->w, inlink->h); + + ret = ff_filter_frame(outlink, out); + pullup_release_frame(f); +end: + av_frame_free(&in); + return ret; +} + +static av_cold void uninit(AVFilterContext *ctx) +{ + PullupContext *s = ctx->priv; + PullupField *f; + int i; + + f = s->head; + do { + if (!f) + break; + + av_free(f->diffs); + av_free(f->combs); + av_free(f->vars); + f = f->next; + av_freep(&f->prev); + } while (f != s->last); + av_freep(&s->last); + + for (i = 0; i < FF_ARRAY_ELEMS(s->buffers); i++) { + av_freep(&s->buffers[i].planes[0]); + av_freep(&s->buffers[i].planes[1]); + av_freep(&s->buffers[i].planes[2]); + } +} + +static const AVFilterPad pullup_inputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .filter_frame = filter_frame, + .config_props = config_input, + }, + { NULL } +}; + +static const AVFilterPad pullup_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_VIDEO, + .config_props = config_output, + }, + { NULL } +}; + +AVFilter avfilter_vf_pullup = { + .name = "pullup", + .description = NULL_IF_CONFIG_SMALL("Pullup from field sequence to frames."), + .priv_size = sizeof(PullupContext), + .priv_class = &pullup_class, + .uninit = uninit, + .query_formats = query_formats, + .inputs = pullup_inputs, + .outputs = pullup_outputs, +}; diff --git a/libavfilter/vf_pullup.h b/libavfilter/vf_pullup.h new file mode 100644 index 0000000..6a31e36 --- /dev/null +++ b/libavfilter/vf_pullup.h @@ -0,0 +1,71 @@ +/* + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef AVFILTER_PULLUP_H +#define AVFILTER_PULLUP_H + +#include "avfilter.h" + +typedef struct PullupBuffer { + int lock[2]; + uint8_t *planes[4]; +} PullupBuffer; + +typedef struct PullupField { + int parity; + PullupBuffer *buffer; + unsigned flags; + int breaks; + int affinity; + int *diffs; + int *combs; + int *vars; + struct PullupField *prev, *next; +} PullupField; + +typedef struct PullupFrame { + int lock; + int length; + int parity; + PullupBuffer *ifields[4], *ofields[2]; + PullupBuffer *buffer; +} PullupFrame; + +typedef struct PullupContext { + const AVClass *class; + int junk_left, junk_right, junk_top, junk_down, junk_bottom; + int metric_plane; + int strict_breaks; + int strict_pairs; + int metric_w, metric_h, metric_length; + int metric_offset; + int nb_planes; + int planewidth[4]; + int planeheight[4]; + PullupField *first, *last, *head; + PullupBuffer buffers[10]; + PullupFrame frame; + + int (*diff)(const uint8_t *a, const uint8_t *b, int s); + int (*comb)(const uint8_t *a, const uint8_t *b, int s); + int (*var )(const uint8_t *a, const uint8_t *b, int s); +} PullupContext; + +void ff_pullup_init_x86(PullupContext *s); + +#endif /* AVFILTER_PULLUP_H */ diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile index 0b19c4b..b15c562 100644 --- a/libavfilter/x86/Makefile +++ b/libavfilter/x86/Makefile @@ -1,9 +1,11 @@ OBJS-$(CONFIG_GRADFUN_FILTER) += x86/vf_gradfun.o OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d_init.o +OBJS-$(CONFIG_PULLUP_FILTER) += x86/vf_pullup_init.o OBJS-$(CONFIG_SPP_FILTER) += x86/vf_spp.o OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume_init.o OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif_init.o YASM-OBJS-$(CONFIG_HQDN3D_FILTER) += x86/vf_hqdn3d.o +YASM-OBJS-$(CONFIG_PULLUP_FILTER) += x86/vf_pullup.o YASM-OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume.o YASM-OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif.o x86/yadif-16.o x86/yadif-10.o diff --git a/libavfilter/x86/vf_pullup.asm b/libavfilter/x86/vf_pullup.asm new file mode 100644 index 0000000..ca0806a --- /dev/null +++ b/libavfilter/x86/vf_pullup.asm @@ -0,0 +1,178 @@ +;***************************************************************************** +;* x86-optimized functions for pullup filter +;* +;* This file is part of FFmpeg. +;* +;* FFmpeg is free software; you can redistribute it and/or modify +;* it under the terms of the GNU General Public License as published by +;* the Free Software Foundation; either version 2 of the License, or +;* (at your option) any later version. +;* +;* FFmpeg is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +;* GNU General Public License for more details. +;* +;* You should have received a copy of the GNU General Public License along +;* with FFmpeg; if not, write to the Free Software Foundation, Inc., +;* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. +;****************************************************************************** + +%include "libavutil/x86/x86util.asm" + +SECTION_TEXT + +INIT_MMX mmx +cglobal pullup_filter_diff, 3, 5, 8, first, second, size + mov r3, 4 + pxor m4, m4 + pxor m7, m7 + +.loop: + movq m0, [firstq] + movq m2, [firstq] + add firstq, sizeq + movq m1, [secondq] + add secondq, sizeq + psubusb m2, m1 + psubusb m1, m0 + movq m0, m2 + movq m3, m1 + punpcklbw m0, m7 + punpcklbw m1, m7 + punpckhbw m2, m7 + punpckhbw m3, m7 + paddw m4, m0 + paddw m4, m1 + paddw m4, m2 + paddw m4, m3 + + dec r3 + jnz .loop + + movq m3, m4 + punpcklwd m4, m7 + punpckhwd m3, m7 + paddd m3, m4 + movd eax, m3 + psrlq m3, 32 + movd r4, m3 + add eax, r4 + RET + +INIT_MMX mmx +cglobal pullup_filter_comb, 3, 5, 8, first, second, size + mov r3, 4 + pxor m6, m6 + pxor m7, m7 + sub secondq, sizeq + +.loop: + movq m0, [secondq] + movq m1, [secondq] + punpcklbw m0, m7 + movq m2, [secondq+sizeq] + punpcklbw m1, m7 + punpcklbw m2, m7 + paddw m0, m0 + paddw m1, m2 + movq m2, m0 + psubusw m0, m1 + psubusw m1, m2 + paddw m6, m0 + paddw m6, m1 + + movq m0, [firstq] + movq m1, [secondq] + punpckhbw m0, m7 + movq m2, [secondq+sizeq] + punpckhbw m1, m7 + punpckhbw m2, m7 + paddw m0, m0 + paddw m1, m2 + movq m2, m0 + psubusw m0, m1 + psubusw m1, m2 + paddw m6, m0 + paddw m6, m1 + + movq m0, [secondq+sizeq] + movq m1, [firstq] + punpcklbw m0, m7 + movq m2, [firstq+sizeq] + punpcklbw m1, m7 + punpcklbw m2, m7 + paddw m0, m0 + paddw m1, m2 + movq m2, m0 + psubusw m0, m1 + psubusw m1, m2 + paddw m6, m0 + paddw m6, m1 + + movq m0, [secondq+sizeq] + movq m1, [firstq] + punpckhbw m0, m7 + movq m2, [firstq+sizeq] + punpckhbw m1, m7 + punpckhbw m2, m7 + paddw m0, m0 + paddw m1, m2 + movq m2, m0 + psubusw m0, m1 + psubusw m1, m2 + paddw m6, m0 + paddw m6, m1 + + add firstq, sizeq + add secondq, sizeq + dec r3 + jnz .loop + + movq m5, m6 + punpcklwd m6, m7 + punpckhwd m5, m7 + paddd m5, m6 + movd eax, m5 + psrlq m5, 32 + movd r4, m5 + add eax, r4 + RET + +INIT_MMX mmx +cglobal pullup_filter_var, 3, 5, 8, first, second, size + mov r3, 3 + pxor m4, m4 + pxor m7, m7 + +.loop: + movq m0, [firstq] + movq m2, [firstq] + movq m1, [firstq+sizeq] + add firstq, sizeq + psubusb m2, m1 + psubusb m1, m0 + movq m0, m2 + movq m3, m1 + punpcklbw m0, m7 + punpcklbw m1, m7 + punpckhbw m2, m7 + punpckhbw m3, m7 + paddw m4, m0 + paddw m4, m1 + paddw m4, m2 + paddw m4, m3 + + dec r3 + jnz .loop + + movq m3, m4 + punpcklwd m4, m7 + punpckhwd m3, m7 + paddd m3, m4 + movd eax, m3 + psrlq m3, 32 + movd r4, m3 + add eax, r4 + shl eax, 2 + RET diff --git a/libavfilter/x86/vf_pullup_init.c b/libavfilter/x86/vf_pullup_init.c new file mode 100644 index 0000000..9948abf --- /dev/null +++ b/libavfilter/x86/vf_pullup_init.c @@ -0,0 +1,41 @@ +/* + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public + * License as published by the Free Software Foundation; either + * version 2 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +*/ + +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/mem.h" +#include "libavutil/x86/asm.h" +#include "libavutil/x86/cpu.h" +#include "libavfilter/vf_pullup.h" + +int ff_pullup_filter_diff_mmx(const uint8_t *a, const uint8_t *b, int s); +int ff_pullup_filter_comb_mmx(const uint8_t *a, const uint8_t *b, int s); +int ff_pullup_filter_var_mmx (const uint8_t *a, const uint8_t *b, int s); + +av_cold void ff_pullup_init_x86(PullupContext *s) +{ +#if HAVE_YASM + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_MMX(cpu_flags)) { + s->diff = ff_pullup_filter_diff_mmx; + s->comb = ff_pullup_filter_comb_mmx; + s->var = ff_pullup_filter_var_mmx; + } +#endif +}

2 1

avformat/mux: set avoid_negative_ts based on flags if not overridden by user or muxer
by git＠videolan.org 17 Oct '13

17 Oct '13

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Fri Apr 26 11:52:51 2013 +0200| [eba33396f007dc373678adea6c1a4b1a50373263] | committer: Michael Niedermayer avformat/mux: set avoid_negative_ts based on flags if not overridden by user or muxer This changes the default to avoid negative timestamps. Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=eba33396f007dc373… --- libavformat/avformat.h | 5 ++++- libavformat/mux.c | 7 ++++++ tests/ref/lavf/ts | 2 +- tests/ref/lavf/wtv | 2 +- tests/ref/seek/lavf-ts | 54 +++++++++++++++++++++++------------------------ tests/ref/seek/lavf-wtv | 44 +++++++++++++++++++------------------- 6 files changed, 62 insertions(+), 52 deletions(-) diff --git a/libavformat/avformat.h b/libavformat/avformat.h index d26964a..c8f4912 100644 --- a/libavformat/avformat.h +++ b/libavformat/avformat.h @@ -370,7 +370,10 @@ typedef struct AVProbeData { timestamps. If not set the timestamp will be shifted in av_write_frame and av_interleaved_write_frame so they - start from 0. */ + start from 0. + The user or muxer can override this through + AVFormatContext.avoid_negative_ts + */ #define AVFMT_SEEK_TO_PTS 0x4000000 /**< Seeking is based on PTS */ diff --git a/libavformat/mux.c b/libavformat/mux.c index 4781a3e..eba4ec3 100644 --- a/libavformat/mux.c +++ b/libavformat/mux.c @@ -398,6 +398,13 @@ int avformat_write_header(AVFormatContext *s, AVDictionary **options) if ((ret = init_pts(s)) < 0) return ret; + if (s->avoid_negative_ts < 0) { + if (s->oformat->flags & (AVFMT_TS_NEGATIVE | AVFMT_NOTIMESTAMPS)) { + s->avoid_negative_ts = 0; + } else + s->avoid_negative_ts = 1; + } + return 0; } diff --git a/tests/ref/lavf/ts b/tests/ref/lavf/ts index 0cab3dd..d795c27 100644 --- a/tests/ref/lavf/ts +++ b/tests/ref/lavf/ts @@ -1,3 +1,3 @@ -4531d0f4a80d5a2ee2b08a8d4ba3b442 *./tests/data/lavf/lavf.ts +04ba219e6f56f7b0fa4a47765323d204 *./tests/data/lavf/lavf.ts 407020 ./tests/data/lavf/lavf.ts ./tests/data/lavf/lavf.ts CRC=0xcc2dc628 diff --git a/tests/ref/lavf/wtv b/tests/ref/lavf/wtv index 8f2c2ac..f54f862 100644 --- a/tests/ref/lavf/wtv +++ b/tests/ref/lavf/wtv @@ -1,3 +1,3 @@ -345516d3a03fd239c62e5e7257c9f4a2 *./tests/data/lavf/lavf.wtv +237211f6aa3df9da2dfe0f7d5b8ba85c *./tests/data/lavf/lavf.wtv 413696 ./tests/data/lavf/lavf.wtv ./tests/data/lavf/lavf.wtv CRC=0xcc2dc628 diff --git a/tests/ref/seek/lavf-ts b/tests/ref/seek/lavf-ts index 407ba59..e57651e 100644 --- a/tests/ref/seek/lavf-ts +++ b/tests/ref/seek/lavf-ts @@ -1,53 +1,53 @@ -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st:-1 flags:0 ts:-1.000000 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st:-1 flags:1 ts: 1.894167 -ret: 0 st: 0 flags:0 dts: 1.880000 pts: 1.920000 pos: 216388 size: 17440 +ret: 0 st: 0 flags:1 dts: 1.880000 pts: 1.920000 pos: 189692 size: 24786 ret: 0 st: 0 flags:0 ts: 0.788333 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 0 flags:1 ts:-0.317500 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 1 flags:0 ts: 2.576667 -ret: 0 st: 1 flags:1 dts: 2.120522 pts: 2.120522 pos: 404576 size: 209 +ret: 0 st: 1 flags:1 dts: 2.160522 pts: 2.160522 pos: 404576 size: 209 ret: 0 st: 1 flags:1 ts: 1.470833 -ret: 0 st: 1 flags:1 dts: 1.389089 pts: 1.389089 pos: 159988 size: 208 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st:-1 flags:0 ts: 0.365002 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st:-1 flags:1 ts:-0.740831 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 0 flags:0 ts: 2.153333 -ret: 0 st: 0 flags:0 dts: 2.160000 pts: 2.200000 pos: 325992 size: 12692 +ret: 0 st: 1 flags:1 dts: 1.794811 pts: 1.794811 pos: 322608 size: 209 ret: 0 st: 0 flags:1 ts: 1.047500 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 1 flags:0 ts:-0.058333 -ret: 0 st: 1 flags:1 dts: 1.389089 pts: 1.389089 pos: 159988 size: 208 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st: 1 flags:1 ts: 2.835833 -ret: 0 st: 1 flags:1 dts: 2.120522 pts: 2.120522 pos: 404576 size: 209 +ret: 0 st: 1 flags:1 dts: 2.160522 pts: 2.160522 pos: 404576 size: 209 ret: 0 st:-1 flags:0 ts: 1.730004 -ret: 0 st: 0 flags:0 dts: 1.760000 pts: 1.800000 pos: 162996 size: 12135 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st:-1 flags:1 ts: 0.624171 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 0 flags:0 ts:-0.481667 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 0 flags:1 ts: 2.412500 -ret: 0 st: 1 flags:1 dts: 2.120522 pts: 2.120522 pos: 404576 size: 209 +ret: 0 st: 1 flags:1 dts: 2.160522 pts: 2.160522 pos: 404576 size: 209 ret: 0 st: 1 flags:0 ts: 1.306667 -ret: 0 st: 1 flags:1 dts: 1.389089 pts: 1.389089 pos: 159988 size: 208 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st: 1 flags:1 ts: 0.200844 -ret: 0 st: 1 flags:1 dts: 1.389089 pts: 1.389089 pos: 159988 size: 208 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st:-1 flags:0 ts:-0.904994 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st:-1 flags:1 ts: 1.989173 -ret: 0 st: 0 flags:0 dts: 1.960000 pts: 2.000000 pos: 251356 size: 13449 +ret: 0 st: 0 flags:0 dts: 1.960000 pts: 2.000000 pos: 235000 size: 15019 ret: 0 st: 0 flags:0 ts: 0.883344 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 0 flags:1 ts:-0.222489 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st: 1 flags:0 ts: 2.671678 -ret: 0 st: 1 flags:1 dts: 2.120522 pts: 2.120522 pos: 404576 size: 209 +ret: 0 st: 1 flags:1 dts: 2.160522 pts: 2.160522 pos: 404576 size: 209 ret: 0 st: 1 flags:1 ts: 1.565844 -ret: 0 st: 1 flags:1 dts: 1.389089 pts: 1.389089 pos: 159988 size: 208 +ret: 0 st: 1 flags:1 dts: 1.429089 pts: 1.429089 pos: 159988 size: 208 ret: 0 st:-1 flags:0 ts: 0.460008 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 ret: 0 st:-1 flags:1 ts:-0.645825 -ret: 0 st: 0 flags:1 dts: 1.360000 pts: 1.400000 pos: 564 size: 24801 +ret: 0 st: 0 flags:1 dts: 1.400000 pts: 1.440000 pos: 564 size: 24801 diff --git a/tests/ref/seek/lavf-wtv b/tests/ref/seek/lavf-wtv index 67def8a..71703c5 100644 --- a/tests/ref/seek/lavf-wtv +++ b/tests/ref/seek/lavf-wtv @@ -1,48 +1,48 @@ -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st:-1 flags:0 ts:-1.000000 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st:-1 flags:1 ts: 1.894167 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st: 0 flags:0 ts: 0.788334 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st: 0 flags:1 ts:-0.317499 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret:-1 st: 1 flags:0 ts: 2.576668 ret: 0 st: 1 flags:1 ts: 1.470835 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st:-1 flags:0 ts: 0.365002 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st:-1 flags:1 ts:-0.740831 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret:-1 st: 0 flags:0 ts: 2.153336 ret: 0 st: 0 flags:1 ts: 1.047503 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st: 1 flags:0 ts:-0.058330 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st: 1 flags:1 ts: 2.835837 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret:-1 st:-1 flags:0 ts: 1.730004 ret: 0 st:-1 flags:1 ts: 0.624171 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st: 0 flags:0 ts:-0.481662 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st: 0 flags:1 ts: 2.412505 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret:-1 st: 1 flags:0 ts: 1.306672 ret: 0 st: 1 flags:1 ts: 0.200839 -ret: 0 st: 1 flags:1 dts: 0.224195 pts: 0.224195 pos: 112904 size: 209 +ret: 0 st: 1 flags:1 dts: 0.211950 pts: 0.211950 pos: 99352 size: 209 ret: 0 st:-1 flags:0 ts:-0.904994 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st:-1 flags:1 ts: 1.989173 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st: 0 flags:0 ts: 0.883340 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st: 0 flags:1 ts:-0.222493 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret:-1 st: 1 flags:0 ts: 2.671674 ret: 0 st: 1 flags:1 ts: 1.565841 -ret: 0 st: 1 flags:1 dts: 0.694399 pts: 0.694399 pos: 294744 size: 209 +ret: 0 st: 1 flags:1 dts: 0.734399 pts: 0.734399 pos: 294744 size: 209 ret: 0 st:-1 flags:0 ts: 0.460008 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208 ret: 0 st:-1 flags:1 ts:-0.645825 -ret: 0 st: 1 flags:1 dts:-0.010907 pts:-0.010907 pos: 26344 size: 208 +ret: 0 st: 1 flags:1 dts: 0.029093 pts: 0.029093 pos: 26344 size: 208

3 2

swscale/utils: dont divide by zero with zero vectors
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | Michael Niedermayer <michaelni(a)gmx.at> | Tue Oct 1 00:27:28 2013 +0200| [2e2a2d8801b045b3dd58a4e49e8e040b559bc84a] | committer: Michael Niedermayer swscale/utils: dont divide by zero with zero vectors Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=2e2a2d8801b045b3d… --- libswscale/utils.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/libswscale/utils.c b/libswscale/utils.c index 05ab0c6..a2e3ce1 100644 --- a/libswscale/utils.c +++ b/libswscale/utils.c @@ -622,6 +622,10 @@ static av_cold int initFilter(int16_t **outFilter, int32_t **filterPos, sum += filter[i * filterSize + j]; } sum = (sum + one / 2) / one; + if (!sum) { + av_log(NULL, AV_LOG_WARNING, "SwScaler: zero vector in scaling\n"); + sum = 1; + } for (j = 0; j < *outFilterSize; j++) { int64_t v = filter[i * filterSize + j] + error; int intV = ROUNDED_DIV(v, sum);

1 0

swscale: make bilinear scaling the default
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | wm4 <nfxjfg(a)googlemail.com> | Sun Sep 29 19:52:37 2013 +0200| [01e3340fb6022465ab23cf4a67a89c869b64476e] | committer: Michael Niedermayer swscale: make bilinear scaling the default Before this commit, sws_init_context() failed with an error if no scaler was explicitly set. Defaulting to something reasonable is better behavior. Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=01e3340fb6022465a… --- libswscale/options.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libswscale/options.c b/libswscale/options.c index 8985e6b..2b3147b 100644 --- a/libswscale/options.c +++ b/libswscale/options.c @@ -34,7 +34,7 @@ static const char *sws_context_to_name(void *ptr) #define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM static const AVOption swscale_options[] = { - { "sws_flags", "scaler flags", OFFSET(flags), AV_OPT_TYPE_FLAGS, { .i64 = DEFAULT }, 0, UINT_MAX, VE, "sws_flags" }, + { "sws_flags", "scaler flags", OFFSET(flags), AV_OPT_TYPE_FLAGS, { .i64 = SWS_BILINEAR }, 0, UINT_MAX, VE, "sws_flags" }, { "fast_bilinear", "fast bilinear", 0, AV_OPT_TYPE_CONST, { .i64 = SWS_FAST_BILINEAR }, INT_MIN, INT_MAX, VE, "sws_flags" }, { "bilinear", "bilinear", 0, AV_OPT_TYPE_CONST, { .i64 = SWS_BILINEAR }, INT_MIN, INT_MAX, VE, "sws_flags" }, { "bicubic", "bicubic", 0, AV_OPT_TYPE_CONST, { .i64 = SWS_BICUBIC }, INT_MIN, INT_MAX, VE, "sws_flags" },

1 0

fate: add fieldorder test
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | Paul B Mahol <onemda(a)gmail.com> | Mon Sep 30 14:06:02 2013 +0000| [fc0d8cf185921ed193566411699e0d6cf67c2096] | committer: Paul B Mahol fate: add fieldorder test Signed-off-by: Paul B Mahol <onemda(a)gmail.com> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=fc0d8cf185921ed19… --- tests/fate/filter-video.mak | 3 + tests/ref/fate/filter-pixfmts-fieldorder | 89 ++++++++++++++++++++++++++++++ 2 files changed, 92 insertions(+) diff --git a/tests/fate/filter-video.mak b/tests/fate/filter-video.mak index 9b9acf7..1ba3beb 100644 --- a/tests/fate/filter-video.mak +++ b/tests/fate/filter-video.mak @@ -244,6 +244,9 @@ fate-filter-pixfmts-crop: CMD = pixfmts "100:100:100:100" FATE_FILTER_PIXFMTS-$(CONFIG_FIELD_FILTER) += fate-filter-pixfmts-field fate-filter-pixfmts-field: CMD = pixfmts "bottom" +FATE_FILTER_PIXFMTS-$(CONFIG_FIELDORDER_FILTER) += fate-filter-pixfmts-fieldorder +fate-filter-pixfmts-fieldorder: CMD = pixfmts "tff" "setfield=bff," + FATE_FILTER_PIXFMTS-$(CONFIG_HFLIP_FILTER) += fate-filter-pixfmts-hflip fate-filter-pixfmts-hflip: CMD = pixfmts diff --git a/tests/ref/fate/filter-pixfmts-fieldorder b/tests/ref/fate/filter-pixfmts-fieldorder new file mode 100644 index 0000000..d743caa --- /dev/null +++ b/tests/ref/fate/filter-pixfmts-fieldorder @@ -0,0 +1,89 @@ +0bgr a8559fc1395908cda27a27bf79819d61 +0rgb 704924e65927896b84bd645ca59fb7f2 +abgr f479df06a4b73d348b3af004923d7eb9 +argb c16c9af56802c3b821b30c613c041622 +bgr0 b7ee10fec20508b64dfca22d61533a88 +bgr24 a1c59acf60652578b802e7f6639d1ac5 +bgr444be d41639928d05bdee67fa2a76c7f74533 +bgr444le 45f616b1ec9014f8d618878c4db850c1 +bgr48be 7015ba40d142742e5c6a39113fee58af +bgr48le fd33d4a60a789d170b871784d0859089 +bgr4_byte d1ff07684bc1be194d26cdf598385b8e +bgr555be 2295d642e02f9dcb49c20142bfdd9c1a +bgr555le a95883301ab991a3cf8bb063b7912410 +bgr565be c37ced8fa4951da20831edf9bd46b35c +bgr565le da0ee7f773efa07fdacc62b7deac452b +bgr8 94a043f6d5e01de077a92a9be6f77582 +bgra 02db5e046ced11d3f09d901cae205e2f +gbrap 5317a66b677c943203b2a5e93607dc68 +gbrp b1ffeba90a60e9a23c9f35466c135069 +gbrp10be 07b2842fdf9b92f6cd8e8cf3a8abdfac +gbrp10le 7fa5429bb4bc6f0e1f3c190bc8064b38 +gbrp12be 5ea6cb457c88b4b3d48f6f63b6208828 +gbrp12le f969a3b017a0f46008f81453174f4ecd +gbrp14be b2879b691564f66b88e63efd268de8c5 +gbrp14le 7baa94cd296e6ec8e41446bca95151e4 +gbrp9be a6eb7cde03f19a25bf13627d731a2c9a +gbrp9le 1b6d228b97a370ec76707ed2dcc15a66 +gray 3258910ef8c6f0c4a331368e7af87507 +gray16be 50cc8a29e8e81e174676542347804a4f +gray16le fb93e8aa2deed734dfd1ca6a5c48cf18 +rgb0 5f8f8076e2b9a2422cac25a4f0459d79 +rgb24 66ca89ced3ade4c239ad6c5a79a6b8cd +rgb444be 7197a09137ab5630a26226396bb7e313 +rgb444le 2d10d7d2ef4527152a9ba8b2020c8d41 +rgb48be 7aee07b85b137365eb04a7ea4c10dcfe +rgb48le debaa00dfed0234efd1441aea8195026 +rgb4_byte 12d6a5a7ca40c531a4d480546ed766eb +rgb555be 5d36ead35aa8820daab7d84e1ffbb4ac +rgb555le 10f43b4c9b06742abc114f0ff1864111 +rgb565be 998ca430a1173f194d78a92695e097f9 +rgb565le 99be7760861aa9c1d3d743a1ea311689 +rgb8 6303c75883c1e2e7e13644fb5dcd11a7 +rgba af33be882959a2fc963d6e5984d24b6d +rgba64be 9ae73ceaa9d95e7981b4df559b0d3b6d +rgba64le 86997e62f62d2a5351b5ca9c9c93c2d1 +uyvy422 a07829fe630b59855c80d87d0e8f0dbb +xyz12be 5785613045b051cee508625dde3357df +xyz12le 3645601dc8bfbdff68cc823a0fc85b27 +yuv411p 3d3d427999863945aa5ec03edf98c19a +yuv422p 961d8a22bfe8868eb3c743b4241a3c99 +yuv422p10be 845e80406c880f33a384b5ebdd21796d +yuv422p10le 0dbf5d1eb94f10cc4413976fa175fa90 +yuv422p12be 9e96f9324d0a4eac4b45161b02c14209 +yuv422p12le db17368e8cd5409826a088ee26ad2a1b +yuv422p14be e20b379fb17e50f279a5ecd9c895537f +yuv422p14le 9440b0c46dadddc19154054978fce20c +yuv422p16be 09d09b64aa4b6ae8479a833622bf27e8 +yuv422p16le 03dbbf3394ba66cda54ef53772258dae +yuv422p9be 7c59e15cb5e6eac9cd090c6ecab4c6b8 +yuv422p9le 98b6b1fbacdfd4c25dbda89b675b660b +yuv444p 709ffd0e95a9438b14d9ba59c0237a61 +yuv444p10be 5ae7571651faebac6dca2ee4fd24610c +yuv444p10le eb3acb92448d8b0e6f145e9d8af43378 +yuv444p12be df1e9bb90364d874dbac41a8237d6068 +yuv444p12le 7e21d00cb42404df0780317dc135f03d +yuv444p14be 6abf70b4b7ff55552327230e08e3dd52 +yuv444p14le 85597559a748aa5aeed2d7dc49d7f66e +yuv444p16be 2a19a8ca44198d0e90c02fd4db24d0c6 +yuv444p16le 5cb12916afef8c82d1796fdc39b97174 +yuv444p9be fc731b5b0afe0a9de6bd46179c692efa +yuv444p9le 7449c543e528067afed5d3c96f7b8da2 +yuva422p d07e0ed0cb20eee67007b154f02cf37c +yuva422p10be 56bfd0af50b9450f72061c605f6537dd +yuva422p10le bfe1df3aae3e02f0a2ddbd404199e6ac +yuva422p16be db80a2e1ddc93126206ad4864f67b50c +yuva422p16le 847e541568b4bce64f916d42173c93c7 +yuva422p9be 5487d986b52f5744a8adac77e46c08f1 +yuva422p9le 6faa61b431b2e0bd4064e5fcecb7130a +yuva444p 377f65cb10dd8013011e185b6aa66697 +yuva444p10be 5e42b4e4b6b1bfd12aa5869750cd2cce +yuva444p10le c1fddd5f488fd4f8d1e07288f8a49182 +yuva444p16be ba2b267e0cc0c36041f3db4f4072ee9e +yuva444p16le c2cfda36aa173be3a4a5c9a8fd66d8db +yuva444p9be e047805a8f9552a6e0b6d4f772126808 +yuva444p9le 28336905569a55e935d708527a3c9d9d +yuvj411p 09f79c56109a13eefb68ee729d9a624b +yuvj422p 942043a34ac7d0f65edced1f6361259c +yuvj444p 7e4758df8eb9b18ad60e1b69a913f8c8 +yuyv422 6b0c70d5ebf1685857b65456c547ea1c

1 0

doc: move sws_dither option description to scaler.texi
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | Stefano Sabatini <stefasab(a)gmail.com> | Mon Sep 30 16:54:46 2013 +0200| [8fafaf18342696d37e69876f340fc0d080e48409] | committer: Stefano Sabatini doc: move sws_dither option description to scaler.texi That's the more proper place where the scaler options should be documented. > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=8fafaf18342696d37… --- doc/filters.texi | 17 ----------------- doc/scaler.texi | 18 ++++++++++++++++++ 2 files changed, 18 insertions(+), 17 deletions(-) diff --git a/doc/filters.texi b/doc/filters.texi index 9e3fe5c..93cdad8 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -6669,23 +6669,6 @@ Set full range (0-255 in case of 8-bit luma). Set "MPEG" range (16-235 in case of 8-bit luma). @end table -@item sws_dither -Set the dithering algorithm - -@table @samp -@item auto -Choose automatically. - -@item none -No dithering - -@item bayer -bayer dither - -@item ed -error diffusion dither -@end table - @item force_original_aspect_ratio Enable decreasing or increasing output video width or height if necessary to keep the original aspect ratio. Possible values: diff --git a/doc/scaler.texi b/doc/scaler.texi index 0921551..08d90bc 100644 --- a/doc/scaler.texi +++ b/doc/scaler.texi @@ -96,6 +96,24 @@ Set scaling algorithm parameters. The specified values are specific of some scaling algorithms and ignored by others. The specified values are floating point number values. +@item sws_dither +Set the dithering algorithm. Accepts one of the following +values. Default value is @samp{auto}. + +@table @samp +@item auto +automatic choice + +@item none +no dithering + +@item bayer +bayer dither + +@item ed +error diffusion dither +@end table + @end table @c man end SCALER OPTIONS

1 0

doc/filters/scale: mention and link scaler options
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | Stefano Sabatini <stefasab(a)gmail.com> | Mon Sep 30 16:50:44 2013 +0200| [d17aaad6308150828bbae2c78ad789401753b22d] | committer: Stefano Sabatini doc/filters/scale: mention and link scaler options > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=d17aaad6308150828… --- doc/filters.texi | 6 +++++- doc/scaler.texi | 1 + 2 files changed, 6 insertions(+), 1 deletion(-) diff --git a/doc/filters.texi b/doc/filters.texi index 5f22331..9e3fe5c 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -6559,7 +6559,11 @@ the next filter, the scale filter will convert the input to the requested format. @subsection Options -The filter accepts the following options: +The filter accepts the following options, or any of the options +supported by the libswscale scaler. + +See @ref{scaler_options,,the ffmpeg-scaler manual,ffmpeg-scaler} for +the complete list of scaler options. @table @option @item width, w diff --git a/doc/scaler.texi b/doc/scaler.texi index 0c61f41..0921551 100644 --- a/doc/scaler.texi +++ b/doc/scaler.texi @@ -1,3 +1,4 @@ +@anchor{scaler_options} @chapter Scaler Options @c man begin SCALER OPTIONS

1 0

doc/filters/scale: fix MAN rendering of @ref arguments
by git＠videolan.org 30 Sep '13

30 Sep '13

ffmpeg | branch: master | Stefano Sabatini <stefasab(a)gmail.com> | Mon Sep 30 16:43:33 2013 +0200| [fd0a1647f26a98f5265c837f6ba323374aab4520] | committer: Stefano Sabatini doc/filters/scale: fix MAN rendering of @ref arguments > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=fd0a1647f26a98f52… --- doc/filters.texi | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/doc/filters.texi b/doc/filters.texi index 5bc4ce5..5f22331 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -6597,7 +6597,8 @@ Default value is @samp{0}. @item flags Set libswscale scaling flags. If not explictly specified the filter applies a bilinear scaling algorithm. -See the @ref{sws_flags,,,ffmpeg-scaler.texi} for the values. +See @ref{sws_flags,,the ffmpeg-scaler manual,ffmpeg-scaler} for +the complete list of values. @item size, s Set the video size, the value must be a valid abbreviation or in the

2 1