PR #22594 opened by James Almer (jamrial) URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/22594 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/22594.patch
>From bf3cae994fd932ad082e39631c28844b77433424 Mon Sep 17 00:00:00 2001 From: Anton Khirnov <[email protected]> Date: Wed, 12 Nov 2025 09:56:49 +0100 Subject: [PATCH 1/2] lavfi/vf_tiltandshift: stop (ab)using AVFrame.opaque This filter uses AVFrame.opaque to build a linked list of AVFrames. This is very wrong, as AVFrame.opaque is intended to store caller's private data and may not be touched by filters. What's worse, the filter leaks the opaque values to the outside. Use an AVFifo instead of a linked list to implement the same logic. --- libavfilter/vf_tiltandshift.c | 95 +++++++++++++++++++---------------- 1 file changed, 52 insertions(+), 43 deletions(-) diff --git a/libavfilter/vf_tiltandshift.c b/libavfilter/vf_tiltandshift.c index 05fda4433b..f22f993bc5 100644 --- a/libavfilter/vf_tiltandshift.c +++ b/libavfilter/vf_tiltandshift.c @@ -25,6 +25,8 @@ #include <string.h> +#include "libavutil/avassert.h" +#include "libavutil/fifo.h" #include "libavutil/imgutils.h" #include "libavutil/mem.h" #include "libavutil/opt.h" @@ -63,38 +65,13 @@ typedef struct TiltandshiftContext { uint8_t *black_buffers[4]; int black_linesizes[4]; - /* list containing all input frames */ - size_t input_size; - AVFrame *input; + /* FIFO containing all input frames */ + AVFifo *input; AVFrame *prev; const AVPixFmtDescriptor *desc; } TiltandshiftContext; -static int list_add_frame(TiltandshiftContext *s, AVFrame *frame) -{ - if (s->input == NULL) { - s->input = frame; - } else { - AVFrame *head = s->input; - while (head->opaque) - head = head->opaque; - head->opaque = frame; - } - s->input_size++; - return 0; -} - -static void list_remove_head(TiltandshiftContext *s) -{ - AVFrame *head = s->input; - if (head) { - s->input = head->opaque; - av_frame_free(&head); - } - s->input_size--; -} - static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV410P, @@ -103,11 +80,30 @@ static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_NONE }; +static av_cold int init(AVFilterContext *ctx) +{ + TiltandshiftContext *s = ctx->priv; + + s->input = av_fifo_alloc2(32, sizeof(AVFrame*), AV_FIFO_FLAG_AUTO_GROW); + if (!s->input) + return AVERROR(ENOMEM); + + return 0; +} + static av_cold void uninit(AVFilterContext *ctx) { TiltandshiftContext *s = ctx->priv; - while (s->input) - list_remove_head(s); + + if (s->input) { + AVFrame *frame; + + while (av_fifo_read(s->input, &frame, 1) >= 0) + av_frame_free(&frame); + + av_fifo_freep2(&s->input); + } + av_freep(&s->black_buffers); } @@ -192,6 +188,7 @@ static int output_frame(AVFilterLink *outlink) TiltandshiftContext *s = outlink->src->priv; AVFrame *head; int ret; + int nb_buffered; int ncol = 0; AVFrame *dst = ff_get_video_buffer(outlink, outlink->w, outlink->h); @@ -206,20 +203,23 @@ static int output_frame(AVFilterLink *outlink) ncol, 0); } - head = s->input; + nb_buffered = av_fifo_can_read(s->input); + av_assert0(nb_buffered); // copy a column from each input frame - for ( ; ncol < s->input_size; ncol++) { - AVFrame *src = head; + for (int in_idx = 0; ncol < nb_buffered; ncol++) { + AVFrame *src; + + av_fifo_peek(s->input, &src, 1, in_idx); copy_column(outlink, dst->data, dst->linesize, (const uint8_t **)src->data, src->linesize, ncol, s->tilt); // keep track of the last known frame in case we need it below - s->prev = head; + s->prev = src; // advance to the next frame unless we have to hold it if (s->hold <= ncol) - head = head->opaque; + in_idx++; } // pad any remaining space with black or last frame @@ -236,14 +236,14 @@ static int output_frame(AVFilterLink *outlink) } // set correct timestamps and props as long as there is proper input - ret = av_frame_copy_props(dst, s->input); + av_fifo_read(s->input, &head, 1); + ret = av_frame_copy_props(dst, head); + av_frame_free(&head); if (ret < 0) { av_frame_free(&dst); return ret; } - // discard frame at the top of the list since it has been fully processed - list_remove_head(s); // and it is safe to reduce the hold value (even if unused) s->hold--; @@ -257,16 +257,21 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *frame) AVFilterLink *outlink = inlink->dst->outputs[0]; AVFilterContext *ctx = outlink->src; TiltandshiftContext *s = inlink->dst->priv; + int ret; - int ret = list_add_frame(s, frame); + ret = av_fifo_write(s->input, &frame, 1); if (ret < 0) { + av_frame_free(&frame); return ret; } // load up enough frames to fill a frame and keep the queue filled on subsequent // calls, until we receive EOF, and then we either pad or end - if (!s->eof_recv && s->input_size < outlink->w - s->pad) { - av_log(ctx, AV_LOG_DEBUG, "Not enough frames in the list (%zu/%d), waiting for more.\n", s->input_size, outlink->w - s->pad); + if (!s->eof_recv && + av_fifo_can_read(s->input) < outlink->w - s->pad) { + av_log(ctx, AV_LOG_DEBUG, + "Not enough frames in the list (%zu/%d), waiting for more.\n", + av_fifo_can_read(s->input), outlink->w - s->pad); return 0; } @@ -277,11 +282,13 @@ static int request_frame(AVFilterLink *outlink) { AVFilterContext *ctx = outlink->src; TiltandshiftContext *s = ctx->priv; + size_t nb_buffered = av_fifo_can_read(s->input); int ret; // signal job finished when list is empty or when padding is either // limited or disabled and eof was received - if ((s->input_size <= 0 || s->input_size == outlink->w - s->pad || s->end == TILT_NONE) && s->eof_recv) { + if ((nb_buffered <= 0 || nb_buffered == outlink->w - s->pad || + s->end == TILT_NONE) && s->eof_recv) { return AVERROR_EOF; } @@ -293,8 +300,9 @@ static int request_frame(AVFilterLink *outlink) } if (s->eof_recv) { - while (s->input_size) { - av_log(ctx, AV_LOG_DEBUG, "Emptying buffers (%zu/%d).\n", s->input_size, outlink->w - s->pad); + while (av_fifo_can_read(s->input)) { + av_log(ctx, AV_LOG_DEBUG, "Emptying buffers (%zu/%d).\n", + av_fifo_can_read(s->input), outlink->w - s->pad); ret = output_frame(outlink); if (ret < 0) { return ret; @@ -361,6 +369,7 @@ const FFFilter ff_vf_tiltandshift = { .p.description = NULL_IF_CONFIG_SMALL("Generate a tilt-and-shift'd video."), .p.priv_class = &tiltandshift_class, .priv_size = sizeof(TiltandshiftContext), + .init = init, .uninit = uninit, FILTER_INPUTS(tiltandshift_inputs), FILTER_OUTPUTS(tiltandshift_outputs), -- 2.52.0 >From f61c916df5b9d00e0b46fdf648b88b3354937d6e Mon Sep 17 00:00:00 2001 From: Anton Khirnov <[email protected]> Date: Mon, 5 Jan 2026 12:23:55 +0100 Subject: [PATCH 2/2] lavc/decode: change sw format selection logic in avcodec_default_get_format() Choose the first non-hwaccel format rather than the last one. This matches the logic in ffmpeg CLI and selects YUVA rather than YUV for HEVC with alpha. --- libavcodec/decode.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/libavcodec/decode.c b/libavcodec/decode.c index 5de5a84038..4ce05ea820 100644 --- a/libavcodec/decode.c +++ b/libavcodec/decode.c @@ -987,7 +987,6 @@ int avcodec_decode_subtitle2(AVCodecContext *avctx, AVSubtitle *sub, enum AVPixelFormat avcodec_default_get_format(struct AVCodecContext *avctx, const enum AVPixelFormat *fmt) { - const AVPixFmtDescriptor *desc; const AVCodecHWConfig *config; int i, n; @@ -1014,12 +1013,13 @@ enum AVPixelFormat avcodec_default_get_format(struct AVCodecContext *avctx, // No device or other setup, so we have to choose from things which // don't any other external information. - // If the last element of the list is a software format, choose it + // Choose the first software format // (this should be best software format if any exist). - for (n = 0; fmt[n] != AV_PIX_FMT_NONE; n++); - desc = av_pix_fmt_desc_get(fmt[n - 1]); - if (!(desc->flags & AV_PIX_FMT_FLAG_HWACCEL)) - return fmt[n - 1]; + for (n = 0; fmt[n] != AV_PIX_FMT_NONE; n++) { + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(fmt[n]); + if (!(desc->flags & AV_PIX_FMT_FLAG_HWACCEL)) + return fmt[n]; + } // Finally, traverse the list in order and choose the first entry // with no external dependencies (if there is no hardware configuration -- 2.52.0 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
