From 5fcc87093ce1f2aea3a43b52399b09fdf8acda36 Mon Sep 17 00:00:00 2001 From: Yingming Fan Date: Thu, 23 Apr 2015 23:19:05 +0800 Subject: [PATCH] Add simple P frame support for ffv1. --- libavcodec/ffv1.c | 10 ++++++ libavcodec/ffv1.h | 1 + libavcodec/ffv1dec.c | 54 ++++++++++++++++++++++++-------- libavcodec/ffv1enc.c | 87 ++++++++++++++++++++++++++++++++++++++++++---------- 4 files changed, 124 insertions(+), 28 deletions(-) diff --git a/libavcodec/ffv1.c b/libavcodec/ffv1.c index 7a38bf9..7fb0a0f 100644 --- a/libavcodec/ffv1.c +++ b/libavcodec/ffv1.c @@ -143,12 +143,21 @@ av_cold int ffv1_init_slice_contexts(FFV1Context *f) av_freep(&f->slice_context[i]); goto memfail; } + + fs->ref_sample_buffer = av_malloc_array((fs->width + 6), 3 * MAX_PLANES * + sizeof(*fs->ref_sample_buffer)); + if (!fs->ref_sample_buffer) { + av_freep(&f->slice_context[i]->sample_buffer); + av_freep(&f->slice_context[i]); + goto memfail; + } } return 0; memfail: while(--i >= 0) { av_freep(&f->slice_context[i]->sample_buffer); + av_freep(&f->slice_context[i]->ref_sample_buffer); av_freep(&f->slice_context[i]); } return AVERROR(ENOMEM); @@ -219,6 +228,7 @@ av_cold int ffv1_close(AVCodecContext *avctx) av_freep(&p->vlc_state); } av_freep(&fs->sample_buffer); + av_freep(&fs->ref_sample_buffer); } av_freep(&avctx->stats_out); diff --git a/libavcodec/ffv1.h b/libavcodec/ffv1.h index bfc4d71..f5eeae6 100644 --- a/libavcodec/ffv1.h +++ b/libavcodec/ffv1.h @@ -103,6 +103,7 @@ typedef struct FFV1Context { int run_index; int colorspace; int16_t *sample_buffer; + int16_t *ref_sample_buffer; int ec; int intra; diff --git a/libavcodec/ffv1dec.c b/libavcodec/ffv1dec.c index fda3f09..1874b5b 100644 --- a/libavcodec/ffv1dec.c +++ b/libavcodec/ffv1dec.c @@ -95,7 +95,7 @@ static inline int get_vlc_symbol(GetBitContext *gb, VlcState *const state, } static av_always_inline void decode_line(FFV1Context *s, int w, - int16_t *sample[2], + int16_t *sample[2], int16_t *ref_sample[2], int plane_index, int bits) { PlaneContext *const p = &s->plane[plane_index]; @@ -172,22 +172,31 @@ static av_always_inline void decode_line(FFV1Context *s, int w, if (sign) diff = -diff; - sample[1][x] = av_mod_uintp2(predict(sample[1] + x, sample[0] + x) + diff, bits); + if (s->cur->pict_type == AV_PICTURE_TYPE_I) + sample[1][x] = av_mod_uintp2(predict(sample[1] + x, sample[0] + x) + diff, bits); + else + sample[1][x] = av_mod_uintp2(diff + ref_sample[0][x], bits); } s->run_index = run_index; } -static void decode_plane(FFV1Context *s, uint8_t *src, +static void decode_plane(FFV1Context *s, uint8_t *src, uint8_t *ref, int w, int h, int stride, int plane_index) { int x, y; int16_t *sample[2]; + int16_t *ref_sample[2]; + sample[0] = s->sample_buffer + 3; sample[1] = s->sample_buffer + w + 6 + 3; + ref_sample[0] = s->ref_sample_buffer + 3; + ref_sample[1] = s->ref_sample_buffer + w + 6 + 3; + s->run_index = 0; memset(s->sample_buffer, 0, 2 * (w + 6) * sizeof(*s->sample_buffer)); + memset(s->ref_sample_buffer, 0, 2 * (w + 6) * sizeof(*s->ref_sample_buffer)); for (y = 0; y < h; y++) { int16_t *temp = sample[0]; // FIXME: try a normal buffer @@ -200,11 +209,16 @@ static void decode_plane(FFV1Context *s, uint8_t *src, // { START_TIMER if (s->avctx->bits_per_raw_sample <= 8) { - decode_line(s, w, sample, plane_index, 8); + if (ref) { + for (x = 0; x < w; x++) { + ref_sample[0][x] = ref[x + stride * y]; + } + } + decode_line(s, w, sample, ref_sample, plane_index, 8); for (x = 0; x < w; x++) src[x + stride * y] = sample[1][x]; } else { - decode_line(s, w, sample, plane_index, s->avctx->bits_per_raw_sample); + decode_line(s, w, sample, ref_sample, plane_index, s->avctx->bits_per_raw_sample); if (s->packed_at_lsb) { for (x = 0; x < w; x++) { ((uint16_t*)(src + stride*y))[x] = sample[1][x]; @@ -246,9 +260,9 @@ static void decode_rgb_frame(FFV1Context *s, uint8_t *src[3], int w, int h, int sample[p][1][-1]= sample[p][0][0 ]; sample[p][0][ w]= sample[p][0][w-1]; if (lbd && s->slice_coding_mode == 0) - decode_line(s, w, sample[p], (p + 1)/2, 9); + decode_line(s, w, sample[p], NULL, (p + 1)/2, 9); else - decode_line(s, w, sample[p], (p + 1)/2, bits + (s->slice_coding_mode != 1)); + decode_line(s, w, sample[p], NULL, (p + 1)/2, bits + (s->slice_coding_mode != 1)); } for (x = 0; x < w; x++) { int g = sample[0][1][x]; @@ -361,6 +375,8 @@ static int decode_slice(AVCodecContext *c, void *arg) int width, height, x, y, ret; const int ps = av_pix_fmt_desc_get(c->pix_fmt)->comp[0].step_minus1 + 1; AVFrame * const p = f->cur; + AVFrame * const last_picture = f->last_picture.f; + uint8_t *ref_data[] = {NULL,NULL,NULL,NULL}; int i, si; for( si=0; fs != f->slice_context[si]; si ++) @@ -434,14 +450,22 @@ static int decode_slice(AVCodecContext *c, void *arg) const int chroma_height = FF_CEIL_RSHIFT(height, f->chroma_v_shift); const int cx = x >> f->chroma_h_shift; const int cy = y >> f->chroma_v_shift; - decode_plane(fs, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0); + + if (p->pict_type == AV_PICTURE_TYPE_P) { + ref_data[0] = last_picture->data[0] + ps*x + y*p->linesize[0]; + ref_data[1] = last_picture->data[1] + ps*cx+cy*p->linesize[1]; + ref_data[2] = last_picture->data[2] + ps*cx+cy*p->linesize[2]; + ref_data[3] = last_picture->data[3] + ps*x + y*p->linesize[3]; + } + + decode_plane(fs, p->data[0] + ps*x + y*p->linesize[0], ref_data[0], width, height, p->linesize[0], 0); if (f->chroma_planes) { - decode_plane(fs, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1); - decode_plane(fs, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1); + decode_plane(fs, p->data[1] + ps*cx+cy*p->linesize[1], ref_data[1], chroma_width, chroma_height, p->linesize[1], 1); + decode_plane(fs, p->data[2] + ps*cx+cy*p->linesize[2], ref_data[2], chroma_width, chroma_height, p->linesize[2], 1); } if (fs->transparency) - decode_plane(fs, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2); + decode_plane(fs, p->data[3] + ps*x + y*p->linesize[3], ref_data[3], width, height, p->linesize[3], 2); } else { uint8_t *planes[3] = { p->data[0] + ps * x + y * p->linesize[0], p->data[1] + ps * x + y * p->linesize[1], @@ -879,13 +903,17 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPac ff_init_range_decoder(c, buf, buf_size); ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8); - p->pict_type = AV_PICTURE_TYPE_I; //FIXME I vs. P + if (f->version < 4) + p->pict_type = AV_PICTURE_TYPE_I; + if (get_rac(c, &keystate)) { p->key_frame = 1; f->key_frame_ok = 0; if ((ret = read_header(f)) < 0) return ret; f->key_frame_ok = 1; + if (f->version >= 4) + p->pict_type = AV_PICTURE_TYPE_I; } else { if (!f->key_frame_ok) { av_log(avctx, AV_LOG_ERROR, @@ -893,6 +921,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPac return AVERROR_INVALIDDATA; } p->key_frame = 0; + if (f->version >= 4) + p->pict_type = AV_PICTURE_TYPE_P; } if ((ret = ff_thread_get_buffer(avctx, &f->picture, AV_GET_BUFFER_FLAG_REF)) < 0) diff --git a/libavcodec/ffv1enc.c b/libavcodec/ffv1enc.c index 5b0ade4..f806dc5 100644 --- a/libavcodec/ffv1enc.c +++ b/libavcodec/ffv1enc.c @@ -264,7 +264,7 @@ static inline void put_vlc_symbol(PutBitContext *pb, VlcState *const state, } static av_always_inline int encode_line(FFV1Context *s, int w, - int16_t *sample[3], + int16_t *sample[3], int16_t *ref_sample[3], int plane_index, int bits) { PlaneContext *const p = &s->plane[plane_index]; @@ -302,7 +302,11 @@ static av_always_inline int encode_line(FFV1Context *s, int w, int diff, context; context = get_context(p, sample[0] + x, sample[1] + x, sample[2] + x); - diff = sample[0][x] - predict(sample[0] + x, sample[1] + x); + + if (ref_sample[0]) + diff = sample[0][x] - ref_sample[0][x]; + else + diff = sample[0][x] - predict(sample[0] + x, sample[1] + x); if (context < 0) { context = -context; @@ -365,27 +369,37 @@ static av_always_inline int encode_line(FFV1Context *s, int w, return 0; } -static int encode_plane(FFV1Context *s, uint8_t *src, int w, int h, +static int encode_plane(FFV1Context *s, uint8_t *src, uint8_t *ref, int w, int h, int stride, int plane_index) { int x, y, i, ret; const int ring_size = s->avctx->context_model ? 3 : 2; int16_t *sample[3]; + int16_t *ref_sample[3]; s->run_index = 0; memset(s->sample_buffer, 0, ring_size * (w + 6) * sizeof(*s->sample_buffer)); + memset(s->ref_sample_buffer, 0, ring_size * (w + 6) * sizeof(*s->ref_sample_buffer)); for (y = 0; y < h; y++) { - for (i = 0; i < ring_size; i++) + for (i = 0; i < ring_size; i++) { sample[i] = s->sample_buffer + (w + 6) * ((h + i - y) % ring_size) + 3; + ref_sample[i] = s->ref_sample_buffer + (w + 6) * ((h + i - y) % ring_size) + 3; + } sample[0][-1]= sample[1][0 ]; sample[1][ w]= sample[1][w-1]; // { START_TIMER if (s->bits_per_raw_sample <= 8) { - for (x = 0; x < w; x++) + for (x = 0; x < w; x++) { + if (ref) + ref_sample[0][x] = ref[x + stride * y]; + sample[0][x] = src[x + stride * y]; - if((ret = encode_line(s, w, sample, plane_index, 8)) < 0) + } + if (ref == NULL)//FIXME: try better way to let encode_line know we are encoding P frame. + ref_sample[0] = NULL; + if((ret = encode_line(s, w, sample, ref_sample, plane_index, 8)) < 0) return ret; } else { if (s->packed_at_lsb) { @@ -397,7 +411,7 @@ static int encode_plane(FFV1Context *s, uint8_t *src, int w, int h, sample[0][x] = ((uint16_t*)(src + stride*y))[x] >> (16 - s->bits_per_raw_sample); } } - if((ret = encode_line(s, w, sample, plane_index, s->bits_per_raw_sample)) < 0) + if((ret = encode_line(s, w, sample, ref_sample, plane_index, s->bits_per_raw_sample)) < 0) return ret; } // STOP_TIMER("encode line") } @@ -457,9 +471,9 @@ static int encode_rgb_frame(FFV1Context *s, const uint8_t *src[3], sample[p][0][-1] = sample[p][1][0 ]; sample[p][1][ w] = sample[p][1][w-1]; if (lbd && s->slice_coding_mode == 0) - ret = encode_line(s, w, sample[p], (p + 1) / 2, 9); + ret = encode_line(s, w, sample[p], NULL, (p + 1) / 2, 9); else - ret = encode_line(s, w, sample[p], (p + 1) / 2, bits + (s->slice_coding_mode != 1)); + ret = encode_line(s, w, sample[p], NULL, (p + 1) / 2, bits + (s->slice_coding_mode != 1)); if (ret < 0) return ret; } @@ -689,11 +703,13 @@ static av_cold int encode_init(AVCodecContext *avctx) s->ec = (s->version >= 3); } - if ((s->version == 2 || s->version>3) && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) { + if ((s->version == 2 || s->version>4) && avctx->strict_std_compliance > FF_COMPLIANCE_EXPERIMENTAL) { av_log(avctx, AV_LOG_ERROR, "Version 2 needed for requested features but version 2 is experimental and not enabled\n"); return AVERROR_INVALIDDATA; } + av_log(avctx, AV_LOG_DEBUG, "FFv1 version is %d\n", s->version); + s->ac = avctx->coder_type > 0 ? 2 : 0; s->plane_count = 3; @@ -793,6 +809,17 @@ static av_cold int encode_init(AVCodecContext *avctx) av_log(avctx, AV_LOG_ERROR, "format not supported\n"); return AVERROR(ENOSYS); } + //FIXME: support these + if (s->version == 4) { + if (s->bits_per_raw_sample > 8) { + av_log(avctx, AV_LOG_ERROR, "Version 4 only support 8 bit currently\n"); + return AVERROR_INVALIDDATA; + } + if (s->colorspace != 0) { + av_log(avctx, AV_LOG_ERROR, "Version 4 only support yuv colorspace currently\n"); + return AVERROR_INVALIDDATA; + } + } if (s->transparency) { av_log(avctx, AV_LOG_WARNING, "Storing alpha plane, this will require a recent FFV1 decoder to playback!\n"); } @@ -1112,12 +1139,14 @@ static int encode_slice(AVCodecContext *c, void *arg) int x = fs->slice_x; int y = fs->slice_y; const AVFrame *const p = f->picture.f; + const AVFrame *const last_picture = f->last_picture.f; const int ps = av_pix_fmt_desc_get(c->pix_fmt)->comp[0].step_minus1 + 1; int ret; RangeCoder c_bak = fs->c; const uint8_t *planes[3] = {p->data[0] + ps*x + y*p->linesize[0], p->data[1] + ps*x + y*p->linesize[1], p->data[2] + ps*x + y*p->linesize[2]}; + uint8_t *ref_data[] = {NULL,NULL,NULL,NULL}; fs->slice_coding_mode = 0; if (f->version > 3) { @@ -1148,14 +1177,21 @@ retry: const int cx = x >> f->chroma_h_shift; const int cy = y >> f->chroma_v_shift; - ret = encode_plane(fs, p->data[0] + ps*x + y*p->linesize[0], width, height, p->linesize[0], 0); + if (c->coded_frame->pict_type == AV_PICTURE_TYPE_P) { + ref_data[0] = last_picture->data[0] + ps*x + y*p->linesize[0]; + ref_data[1] = last_picture->data[1] + ps*cx+cy*p->linesize[1]; + ref_data[2] = last_picture->data[2] + ps*cx+cy*p->linesize[2]; + ref_data[3] = last_picture->data[3] + ps*x + y*p->linesize[3]; + } + + ret = encode_plane(fs, p->data[0] + ps*x + y*p->linesize[0], ref_data[0], width, height, p->linesize[0], 0); if (f->chroma_planes) { - ret |= encode_plane(fs, p->data[1] + ps*cx+cy*p->linesize[1], chroma_width, chroma_height, p->linesize[1], 1); - ret |= encode_plane(fs, p->data[2] + ps*cx+cy*p->linesize[2], chroma_width, chroma_height, p->linesize[2], 1); + ret |= encode_plane(fs, p->data[1] + ps*cx+cy*p->linesize[1], ref_data[1], chroma_width, chroma_height, p->linesize[1], 1); + ret |= encode_plane(fs, p->data[2] + ps*cx+cy*p->linesize[2], ref_data[2], chroma_width, chroma_height, p->linesize[2], 1); } if (fs->transparency) - ret |= encode_plane(fs, p->data[3] + ps*x + y*p->linesize[3], width, height, p->linesize[3], 2); + ret |= encode_plane(fs, p->data[3] + ps*x + y*p->linesize[3], ref_data[3], width, height, p->linesize[3], 2); } else { ret = encode_rgb_frame(fs, planes, width, height, p->linesize); } @@ -1181,7 +1217,7 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *pkt, { FFV1Context *f = avctx->priv_data; RangeCoder *const c = &f->slice_context[0]->c; - AVFrame *const p = f->picture.f; + AVFrame *p = f->picture.f; int used_count = 0; uint8_t keystate = 128; uint8_t *buf_p; @@ -1243,19 +1279,34 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *pkt, ff_init_range_encoder(c, pkt->data, pkt->size); ff_build_rac_states(c, 0.05 * (1LL << 32), 256 - 8); + if (f->last_picture.f) + ff_thread_release_buffer(avctx, &f->last_picture); + FFSWAP(ThreadFrame, f->picture, f->last_picture); + + f->cur = p = f->picture.f; + + if ((ret = ff_thread_get_buffer(avctx, &f->picture, AV_GET_BUFFER_FLAG_REF)) < 0) + return ret; + av_frame_unref(p); if ((ret = av_frame_ref(p, pict)) < 0) return ret; - avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; + + if (f->version < 4) + avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; if (avctx->gop_size == 0 || f->picture_number % avctx->gop_size == 0) { put_rac(c, &keystate, 1); avctx->coded_frame->key_frame = 1; f->gob_count++; write_header(f); + if (f->version >= 4) + avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I; } else { put_rac(c, &keystate, 0); avctx->coded_frame->key_frame = 0; + if (f->version >= 4) + avctx->coded_frame->pict_type = AV_PICTURE_TYPE_P; } if (f->ac > 1) { @@ -1308,6 +1359,10 @@ static int encode_frame(AVCodecContext *avctx, AVPacket *pkt, if (avctx->flags & CODEC_FLAG_PASS1) avctx->stats_out[0] = '\0'; + if (f->last_picture.f) + ff_thread_release_buffer(avctx, &f->last_picture); + f->cur = NULL; + f->picture_number++; pkt->size = buf_p - pkt->data; pkt->pts = -- 2.3.2 (Apple Git-55)