February 2026 - ffmpeg-devel - lists.ffmpeg.org

Re: [FFmpeg-devel] [PATCH 2/3] libavcodec: add NETINT Quadra HW decoders & encoders
by Desmond Liu 19 Mar '26

19 Mar '26

Hi Kieran, > Roughly one year ago I tried to buy one of these and I was asked to sign an > NDA. That was a "bug". You do not need to sign an NDA. If you have difficulty obtaining our hardware, contact me. Desmond

2 9

[PATCH] avcodec/libvpxenc: fix sRGB colorspace for non-RGB pixel formats
by Guangyu Sun 17 Mar '26

17 Mar '26

When encoding VP9 with a YUV pixel format (e.g. yuv420p) and AVCOL_SPC_RGB colorspace metadata, libvpxenc unconditionally set VPX_CS_SRGB. This produced a spec-violating bitstream: Profile 0 (4:2:0) with sRGB colorspace, which is only valid for Profile 1/3 (4:4:4). The resulting file is undecodable. RGB pixel formats (GBRP, GBRP10, GBRP12) already correctly set ctx->vpx_cs to VPX_CS_SRGB in set_pix_fmt() and bypass the colorspace switch in set_colorspace(). So reaching the AVCOL_SPC_RGB case means the pixel format is YUV with incorrect RGB metadata. Log a warning and fall back to VPX_CS_BT_709. To reproduce: # generate a bad source ffmpeg -f lavfi -i testsrc=s=64x64:d=1:r=1 -pix_fmt yuv420p -colorspace rgb bad.mp4 # transcode with default parameters ffmpeg -i bad.mp4 bad.webm # check decoding ffmpeg -i bad.webm -f null - # -> 0 frames decoded, error Signed-off-by: Guangyu Sun <gsun(a)roblox.com> --- libavcodec/libvpxenc.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/libavcodec/libvpxenc.c b/libavcodec/libvpxenc.c index 88c058c403..be731a014f 100644 --- a/libavcodec/libvpxenc.c +++ b/libavcodec/libvpxenc.c @@ -902,7 +902,17 @@ static void set_colorspace(AVCodecContext *avctx) vpx_cs = ctx->vpx_cs; } else { switch (avctx->colorspace) { - case AVCOL_SPC_RGB: vpx_cs = VPX_CS_SRGB; break; + case AVCOL_SPC_RGB: + // RGB pixel formats (GBRP, GBRP10, GBRP12) set ctx->vpx_cs + // to VPX_CS_SRGB in set_pix_fmt() and take the branch above, + // so reaching here means the pixel format is YUV with + // incorrect RGB colorspace metadata. + av_log(avctx, AV_LOG_WARNING, + "RGB colorspace is not compatible with pixel format %s, " + "using BT.709 instead.\n", + av_get_pix_fmt_name(avctx->pix_fmt)); + vpx_cs = VPX_CS_BT_709; + break; case AVCOL_SPC_BT709: vpx_cs = VPX_CS_BT_709; break; case AVCOL_SPC_UNSPECIFIED: vpx_cs = VPX_CS_UNKNOWN; break; case AVCOL_SPC_RESERVED: vpx_cs = VPX_CS_RESERVED; break; -- 2.52.0

3 9

[PATCH] avfilter/af_lv2: call lilv_instance_activate before lilv_instance_run
by Karl Mogensen 08 Mar '26

08 Mar '26

Why: the change is done to comply with lilv expectations of hosts. Added call lilv_instance_activate in the config_output function to abide by lilv documentation that states it must be called before lilv_instance_run: "This MUST be called before calling lilv_instance_run()" - documentation source (https://github.com/lv2/lilv/blob/main/include/lilv/lilv.h) Added call lilv_instance_deactivate in the uninit function to abide by lv2 documentation: "If a host calls activate(), it MUST call deactivate() at some point in the future" - documentation source (https://gitlab.com/lv2/lv2/-/blob/main/include/lv2/core/lv2.h) Added instance_activated integer to LV2Context struct to track if instance was activated and only do lilv_instance_deactivate if was activated to abide by lv2 documentation: "Hosts MUST NOT call deactivate() unless activate() was previously called." - documentation source (https://gitlab.com/lv2/lv2/-/blob/main/include/lv2/core/lv2.h) Regarding the patcheck warning (possibly constant :instance_activated): This is a false positive since the struct member is zero-initialized. Fixes: trac issue #11661 (https://trac.ffmpeg.org/ticket/11661) Reported-by: Dave Flater Signed-off-by: Karl Mogensen <karlmogensen0(a)gmail.com> --- libavfilter/af_lv2.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/libavfilter/af_lv2.c b/libavfilter/af_lv2.c index e7f3f9c588..df935601d4 100644 --- a/libavfilter/af_lv2.c +++ b/libavfilter/af_lv2.c @@ -74,6 +74,7 @@ typedef struct LV2Context { float *controls; LilvInstance *instance; + int instance_activated; LilvNode *atom_AtomPort; LilvNode *atom_Sequence; @@ -387,6 +388,9 @@ static int config_output(AVFilterLink *outlink) inlink->min_samples = inlink->max_samples = 4096; } + lilv_instance_activate(s->instance); + s->instance_activated = 1; + return 0; } @@ -574,6 +578,8 @@ static av_cold void uninit(AVFilterContext *ctx) lilv_node_free(s->lv2_OutputPort); lilv_node_free(s->lv2_InputPort); uri_table_destroy(&s->uri_table); + if (s->instance_activated) + lilv_instance_deactivate(s->instance); lilv_instance_free(s->instance); lilv_world_free(s->world); av_freep(&s->mins); -- 2.52.0

3 2

FFmpeg 8.1
by Michael Niedermayer 08 Mar '26

08 Mar '26

Hi Everyone I intend to branch release/8.1 in the next weeks, and i intend to have all reported fuzzer found security issues fixed in 8.1 (likely not all non security issues but more fixes is better) and i intend to have 8.1 made in march. thx -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB While the State exists there can be no freedom; when there is freedom there will be no State. -- Vladimir Lenin

2 3

[PATCH] web/donations: Add github sponsors button
by Michael Niedermayer 07 Mar '26

07 Mar '26

--- src/donations | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/src/donations b/src/donations index 73499c2..99ca443 100644 --- a/src/donations +++ b/src/donations @@ -49,6 +49,15 @@ </p> </div>  </div>  + <div class="col-md-3"> + <div class="well"> + <h3>Github sponsors</h3> + <div class="text-center"> + <iframe src="https://github.com/sponsors/FFmpeg/button" title="Sponsor FFmpeg" height="32" width="114" style="border: 0; border-radius: 6px;"></iframe> + <br><br> + </div> + </div>  + </div>  </div>  <br><br> -- 2.51.0

1 1

[PATCH] libavfilter: rename AVColorRange to AVAlphaMode
by Weixie Cui 04 Mar '26

04 Mar '26

From: Weixie Cui <cuiweixie(a)gmail.com> --- libavfilter/buffersink.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavfilter/buffersink.c b/libavfilter/buffersink.c index cc9789f57c..75b93cee30 100644 --- a/libavfilter/buffersink.c +++ b/libavfilter/buffersink.c @@ -53,7 +53,7 @@ typedef struct BufferSinkContext { int color_spaces_size; enum AVColorRange *color_ranges; ///< list of accepted color ranges int color_ranges_size; - enum AVColorRange *alpha_modes; ///< list of accepted alpha modes + enum AVAlphaMode *alpha_modes; ///< list of accepted alpha modes int alpha_modes_size; #endif -- 2.39.5 (Apple Git-154)

4 4

[PATCH] avfilter: enhance command processing with chain propagation and direction control. (PR #20731)
by cenzhanquan1 03 Mar '26

03 Mar '26

PR #20731 opened by cenzhanquan1 URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20731 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20731.patch The existing avfilter_process_command and avfilter_graph_send_command havelimitations in command propagation: they only handle commands for individualfilters and lack the ability to propagate commands through the entire filterchain, nor do they support directional traversal of the filter graph. Thismakes it difficult to control multiple filters in a chain (e.g., adjustingvolume for all related filters or enabling/disabling a series of filters)with a single command. This patch enhances the command processing logic to address these issues: Add two new flags to control command propagation: AVFILTER_CMD_FLAG_CHAIN: Enables command propagation through the entire filter chain. After processing the current filter, it traverses all associated links (inputs/outputs based on direction) and recursively forwards the command to subsequent filters, covering the full filter topology. AVFILTER_CMD_FLAG_REVERSE: Works with AVFILTER_CMD_FLAG_CHAIN to control traversal direction. Default (forward) follows data flow (source → destination filters), while reverse traversal goes against data flow (destination → source filters). Refactor avfilter_process_command to integrate chain propagation logic: Process the command for the current filter first, then check if chain propagation is enabled. Traverse all relevant links (inputs for reverse, outputs for forward) and recursively forward the command to next-level filters. Track processing status (processed) to determine if any filter in the chain handled the command. Respect AVFILTER_CMD_FLAG_ONE to stop propagation once a filter processes the command, and propagate critical errors. Improve command handling for built-in commands ("ping" and "enable") to workseamlessly with the new propagation logic, ensuring consistent behavioracross the chain. These changes enable flexible command control over entire filter chains,supporting use cases like batch adjustment of filters, topology-wide statuschecks (via "ping"), and coordinated enable/disable operations, whilemaintaining compatibility with existing filter command implementations. Signed-off-by: cenzhanquan1 <cenzhanquan1(a)xiaomi.com> From 9dba682222bd55c952e14e71789a7084529a74f7 Mon Sep 17 00:00:00 2001 From: cenzhanquan1 <cenzhanquan1(a)xiaomi.com> Date: Tue, 21 Oct 2025 17:34:22 +0800 Subject: [PATCH] avfilter: enhance command processing with chain propagation and direction control. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The existing avfilter_process_command and avfilter_graph_send_command havelimitations in command propagation: they only handle commands for individualfilters and lack the ability to propagate commands through the entire filterchain, nor do they support directional traversal of the filter graph. Thismakes it difficult to control multiple filters in a chain (e.g., adjustingvolume for all related filters or enabling/disabling a series of filters)with a single command. This patch enhances the command processing logic to address these issues: Add two new flags to control command propagation: AVFILTER_CMD_FLAG_CHAIN: Enables command propagation through the entire filter chain. After processing the current filter, it traverses all associated links (inputs/outputs based on direction) and recursively forwards the command to subsequent filters, covering the full filter topology. AVFILTER_CMD_FLAG_REVERSE: Works with AVFILTER_CMD_FLAG_CHAIN to control traversal direction. Default (forward) follows data flow (source → destination filters), while reverse traversal goes against data flow (destination → source filters). Refactor avfilter_process_command to integrate chain propagation logic: Process the command for the current filter first, then check if chain propagation is enabled. Traverse all relevant links (inputs for reverse, outputs for forward) and recursively forward the command to next-level filters. Track processing status (processed) to determine if any filter in the chain handled the command. Respect AVFILTER_CMD_FLAG_ONE to stop propagation once a filter processes the command, and propagate critical errors. Improve command handling for built-in commands ("ping" and "enable") to workseamlessly with the new propagation logic, ensuring consistent behavioracross the chain. These changes enable flexible command control over entire filter chains,supporting use cases like batch adjustment of filters, topology-wide statuschecks (via "ping"), and coordinated enable/disable operations, whilemaintaining compatibility with existing filter command implementations. Signed-off-by: cenzhanquan1 <cenzhanquan1(a)xiaomi.com> --- libavfilter/avfilter.c | 75 +++++++++++++++++++++++++++++++++--------- libavfilter/avfilter.h | 10 ++++-- 2 files changed, 67 insertions(+), 18 deletions(-) diff --git a/libavfilter/avfilter.c b/libavfilter/avfilter.c index 169c2baa42..68ec9f6f6a 100644 --- a/libavfilter/avfilter.c +++ b/libavfilter/avfilter.c @@ -607,25 +607,68 @@ static int set_enable_expr(FFFilterContext *ctxi, const char *expr) return 0; } -int avfilter_process_command(AVFilterContext *filter, const char *cmd, const char *arg, char *res, int res_len, int flags) +int avfilter_process_command(AVFilterContext *filter, const char *cmd, const char *arg, + char *res, int res_len, int flags) { - if(!strcmp(cmd, "ping")){ - char local_res[256] = {0}; + int direction = flags & AVFILTER_CMD_FLAG_REVERSE; + int ret = AVERROR(ENOSYS); + int processed = 0; - if (!res) { - res = local_res; - res_len = sizeof(local_res); - } - av_strlcatf(res, res_len, "pong from:%s %s\n", filter->filter->name, filter->name); - if (res == local_res) - av_log(filter, AV_LOG_INFO, "%s", res); - return 0; - }else if(!strcmp(cmd, "enable")) { - return set_enable_expr(fffilterctx(filter), arg); - }else if (fffilter(filter->filter)->process_command) { - return fffilter(filter->filter)->process_command(filter, cmd, arg, res, res_len, flags); + int process_flags = flags & ~AVFILTER_CMD_FLAG_CHAIN; + if (!strcmp(cmd, "ping")) { + char local_res[256] = {0}; + char *res_buf = res ? res : local_res; + size_t buf_len = res ? res_len : sizeof(local_res); + av_strlcatf(res_buf, buf_len, "pong from:%s %s\n", + filter->filter->name, filter->name ? filter->name : "unknown"); + if (!res) + av_log(filter, AV_LOG_INFO, "%s", res_buf); + ret = 0; + } else if (!strcmp(cmd, "enable")) { + ret = set_enable_expr(fffilterctx(filter), arg); + } else if (fffilter(filter->filter)->process_command) { + ret = fffilter(filter->filter)->process_command(filter, cmd, arg, res, res_len, process_flags); + } else { + ret = AVERROR(ENOSYS); } - return AVERROR(ENOSYS); + + if (ret != AVERROR(ENOSYS)) { + processed = 1; + if ((flags & AVFILTER_CMD_FLAG_ONE) || ret < 0) { + return ret; + } + } + + if (!(flags & AVFILTER_CMD_FLAG_CHAIN)) { + return processed ? 0 : AVERROR(ENOSYS); + } + + av_log(filter, AV_LOG_DEBUG, + "cmd_chain: [%s] dir:%s -> '%s' '%s' (forwarding)\n", + filter->name ? filter->name : "unknown", + direction ? "reverse" : "forward", + cmd, arg ? arg : ""); + + unsigned nb_links = direction ? filter->nb_inputs : filter->nb_outputs; + for (int i = 0; i < nb_links; i++) { + AVFilterLink *link = direction ? filter->inputs[i] : filter->outputs[i]; + AVFilterContext *next_filter = direction ? (link ? link->src : NULL) : (link ? link->dst : NULL); + + if (!link || !next_filter) { + av_log(filter, AV_LOG_DEBUG, "Invalid %s link at pad %d\n", + direction ? "input" : "output", i); + continue; + } + + ret = avfilter_process_command(next_filter, cmd, arg, res, res_len, flags); + if (ret >= 0) { + processed = 1; + } else if (ret != AVERROR(ENOSYS)) { + return ret; + } + } + + return processed ? 0 : AVERROR(ENOSYS); } unsigned avfilter_filter_pad_count(const AVFilter *filter, int is_output) diff --git a/libavfilter/avfilter.h b/libavfilter/avfilter.h index 02b58c42c2..fde3811dc3 100644 --- a/libavfilter/avfilter.h +++ b/libavfilter/avfilter.h @@ -466,8 +466,14 @@ struct AVFilterLink { int avfilter_link(AVFilterContext *src, unsigned srcpad, AVFilterContext *dst, unsigned dstpad); -#define AVFILTER_CMD_FLAG_ONE 1 ///< Stop once a filter understood the command (for target=all for example), fast filters are favored automatically -#define AVFILTER_CMD_FLAG_FAST 2 ///< Only execute command when its fast (like a video out that supports contrast adjustment in hw) +#define AVFILTER_CMD_FLAG_ONE 1 ///< Stop once a filter understood the command (for target=all for example), fast filters are favored automatically +#define AVFILTER_CMD_FLAG_FAST 2 ///< Only execute command when its fast (like a video out that supports contrast adjustment in hw) +#define AVFILTER_CMD_FLAG_CHAIN 4 ///< Propagate the command through the entire filter chain. After processing the current filter, + /// traverse all its associated links (inputs or outputs, based on direction) and recursively + /// forward the command to subsequent filters, covering the full filter topology. +#define AVFILTER_CMD_FLAG_REVERSE 8 ///< Only effective when paired with AVFILTER_CMD_FLAG_CHAIN. Changes the command's traversal + /// direction in the chain: default (forward) follows data flow (source → destination filters), + /// while reverse traversal opposes data flow (destination → source filters). /** * Make the filter instance process a command. -- 2.49.1

5 8

[PATCH] avcodec/dstdec: fix arithmetic coder read past end of stream
by Alexander Wichers 28 Feb '26

28 Feb '26

The ac_get() renormalization reads n bits from the bitstream to refill the arithmetic coder state, but does not check whether enough bits remain. When decoding near the end of the arithmetic coded data section, this causes reads past the AC data boundary into trailing frame data, producing corrupted output in the last bytes of the decoded DSD frame. Add a bounds check using get_bits_left() before reading. When fewer than n bits remain, shift in zeros for the missing bits, matching the behavior of the ISO/IEC 14496-3 reference implementation. Signed-off-by: Alexander Wichers <sander.wichers(a)gmail.com> --- libavcodec/dstdec.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) diff --git a/libavcodec/dstdec.c b/libavcodec/dstdec.c index cfb34b7b3c..e37d830ae4 100644 --- a/libavcodec/dstdec.c +++ b/libavcodec/dstdec.c @@ -205,8 +205,15 @@ static av_always_inline void ac_get(ArithCoder *ac, GetBitContext *gb, int p, in if (ac->a < 2048) { int n = 11 - av_log2(ac->a); + int left = get_bits_left(gb); ac->a <<= n; - ac->c = (ac->c << n) | get_bits(gb, n); + if (left >= n) { + ac->c = (ac->c << n) | get_bits(gb, n); + } else { + ac->c <<= n; + if (left > 0) + ac->c |= get_bits(gb, left) << (n - left); + } } } -- 2.34.1

2 1

[PATCH] avformat/dashdec: fix calculation and usage of cur_seq_no
by nicolas.dato＠gmail.com 28 Feb '26

28 Feb '26

From: Nicolas Jorge Dato <nicolas.dato(a)gmail.com> In live dash streams with some startNumber, dashdec was always repositioning itself 60 seconds before the end each time the manifest reloaded. This was more notorious when the manifest reloaded slower than the segments duration, resulting in a new manifest with more than one new segment, and by repositioning 60 seconds before the end some fragments where skipped. It happened to me with a multi-period dash stream, and a startNumber > 2000. Functions like calc_cur_seg_no, calc_min_seg_no, and calc_max_seg_no calculated the segment number taking into account the first_seq_no. However, functions like get_segment_start_time_based_on_timeline and calc_cur_seg_no didn't take first_seq_no into account. This made dashdec believe that the cur_seq_no was always less than min_seq_no, logging 'old fragment' and calling calc_cur_seq_no. Signed-off-by: Nicolas Jorge Dato <nicolas.dato(a)gmail.com> --- libavformat/dashdec.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/libavformat/dashdec.c b/libavformat/dashdec.c index 9a2ea78149..2068a827f3 100644 --- a/libavformat/dashdec.c +++ b/libavformat/dashdec.c @@ -260,6 +260,7 @@ static int64_t get_segment_start_time_based_on_timeline(struct representation *p int64_t j = 0; int64_t num = 0; + cur_seq_no -= pls->first_seq_no; if (pls->n_timelines) { for (i = 0; i < pls->n_timelines; i++) { if (pls->timelines[i]->starttime > 0) { @@ -312,10 +313,10 @@ static int64_t calc_next_seg_no_from_timelines(struct representation *pls, int64 num++; } - return -1; + return pls->first_seq_no; finish: - return num; + return num + pls->first_seq_no; } static void free_fragment(struct fragment **seg) @@ -1417,10 +1418,6 @@ static int64_t calc_cur_seg_no(AVFormatContext *s, struct representation *pls) av_log(s, AV_LOG_TRACE, "in n_timelines mode\n"); start_time_offset = get_segment_start_time_based_on_timeline(pls, 0xFFFFFFFF) - 60 * pls->fragment_timescale; // 60 seconds before end num = calc_next_seg_no_from_timelines(pls, start_time_offset); - if (num == -1) - num = pls->first_seq_no; - else - num += pls->first_seq_no; } else if (pls->fragment_duration){ av_log(s, AV_LOG_TRACE, "in fragment_duration mode fragment_timescale = %"PRId64", presentation_timeoffset = %"PRId64"\n", pls->fragment_timescale, pls->presentation_timeoffset); if (pls->presentation_timeoffset) { -- 2.53.0

1 0

[PATCH] avformat/dashdec: fix calculation and usage of cur_seq_no
by nicolas.dato＠gmail.com 28 Feb '26

28 Feb '26

From: Nicolas Jorge Dato <nicolas.dato(a)gmail.com> In live dash streams with some startNumber, dashdec was always repositioning itself 60 seconds before the end each time the manifest reloaded. This was more notorious when the manifest reloaded slower than the segments duration, resulting in a new manifest with more than one new segment, and by repositioning 60 seconds before the end some fragments where skipped. It happened to me with a multi-period dash stream, and a startNumber > 2000. Functions like calc_cur_seg_no, calc_min_seg_no, and calc_max_seg_no calculated the segment number taking into account the first_seq_no. However, functions like get_segment_start_time_based_on_timeline and calc_cur_seg_no didn't take first_seq_no into account. This made dashdec believe that the cur_seq_no was always less than min_seq_no, logging 'old fragment' and calling calc_cur_seq_no. --- libavformat/dashdec.c | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/libavformat/dashdec.c b/libavformat/dashdec.c index 9a2ea78149..2068a827f3 100644 --- a/libavformat/dashdec.c +++ b/libavformat/dashdec.c @@ -260,6 +260,7 @@ static int64_t get_segment_start_time_based_on_timeline(struct representation *p int64_t j = 0; int64_t num = 0; + cur_seq_no -= pls->first_seq_no; if (pls->n_timelines) { for (i = 0; i < pls->n_timelines; i++) { if (pls->timelines[i]->starttime > 0) { @@ -312,10 +313,10 @@ static int64_t calc_next_seg_no_from_timelines(struct representation *pls, int64 num++; } - return -1; + return pls->first_seq_no; finish: - return num; + return num + pls->first_seq_no; } static void free_fragment(struct fragment **seg) @@ -1417,10 +1418,6 @@ static int64_t calc_cur_seg_no(AVFormatContext *s, struct representation *pls) av_log(s, AV_LOG_TRACE, "in n_timelines mode\n"); start_time_offset = get_segment_start_time_based_on_timeline(pls, 0xFFFFFFFF) - 60 * pls->fragment_timescale; // 60 seconds before end num = calc_next_seg_no_from_timelines(pls, start_time_offset); - if (num == -1) - num = pls->first_seq_no; - else - num += pls->first_seq_no; } else if (pls->fragment_duration){ av_log(s, AV_LOG_TRACE, "in fragment_duration mode fragment_timescale = %"PRId64", presentation_timeoffset = %"PRId64"\n", pls->fragment_timescale, pls->presentation_timeoffset); if (pls->presentation_timeoffset) { -- 2.53.0

1 0