Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
 help / color / mirror / Atom feed
From: Andreas Rheinhardt <ffmpegagent-at-gmail.com@ffmpeg.org>
To: ffmpeg-devel@ffmpeg.org
Cc: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Subject: [FFmpeg-devel] [PATCH 42/47] avcodec/mpegvideo: Allocate dc_val for each encoder slice
Date: Sun, 15 Jun 2025 16:54:33 +0000
Message-ID: <b469ddf6729f46e47985d4899656d9887c624920.1750006480.git.ffmpegagent@gmail.com> (raw)
In-Reply-To: <pull.98.ffstaging.FFmpeg.1750006478.ffmpegagent@gmail.com>

From: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

This fixes data races (which are UB) in the MPEG-4 and H.263+
encoder when predicting DC values; these encoders unconditionally
read values from the line above the current line and only check
lateron (via first_slice_line) whether said prediction can be used
at all. It will also allow to remove said checks (by setting the
entries to 1024 upon opening a new slice).

The vsynth{1,2,3,_lena}-mpeg4-thread FATE tests were affected by this:
https://fate.ffmpeg.org/report.cgi?slot=x86_64-archlinux-gcc-tsan-slices&time=20250613002615

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
---
 libavcodec/mpegvideo.c     | 15 ++++++++++++---
 libavcodec/mpegvideo_enc.c |  1 +
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index 6c76a382cc..4701267d81 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -179,6 +179,7 @@ static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
     COPY(block);
     COPY(start_mb_y);
     COPY(end_mb_y);
+    COPY(dc_val);
     COPY(ac_val);
 #undef COPY
 }
@@ -235,7 +236,7 @@ av_cold int ff_mpv_init_context_frame(MpegEncContext *s)
                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
                     s->avctx->thread_count : 1;
     BufferPoolContext *const pools = &s->buffer_pools;
-    int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
+    int y_size, c_size, yc_size, mb_array_size, mv_table_size, x, y;
     int mb_height;
 
     if (s->encoding && s->avctx->slices)
@@ -337,6 +338,9 @@ av_cold int ff_mpv_init_context_frame(MpegEncContext *s)
     }
 
     if (s->h263_pred || s->h263_aic || !s->encoding) {
+        // When encoding, each slice (and therefore each thread)
+        // gets its own ac_val and dc_val buffers in order to avoid
+        // races.
         size_t allslice_yc_size = yc_size * (s->encoding ? nb_slices : 1);
         if (s->out_format == FMT_H263) {
             /* ac values */
@@ -349,10 +353,15 @@ av_cold int ff_mpv_init_context_frame(MpegEncContext *s)
         // MN: we need these for error resilience of intra-frames
         // Allocating them unconditionally for decoders also means
         // that we don't need to reinitialize when e.g. h263_aic changes.
-        if (!FF_ALLOC_TYPED_ARRAY(s->dc_val_base, yc_size))
+
+        // y_size and therefore yc_size is always odd; allocate one element
+        // more for each encoder slice in order to be able to align each slice's
+        // dc_val to four in order to use aligned stores when cleaning dc_val.
+        allslice_yc_size += s->encoding * nb_slices;
+        if (!FF_ALLOC_TYPED_ARRAY(s->dc_val_base, allslice_yc_size))
             return AVERROR(ENOMEM);
         s->dc_val = s->dc_val_base + s->b8_stride + 1;
-        for (i = 0; i < yc_size; i++)
+        for (size_t i = 0; i < allslice_yc_size; ++i)
             s->dc_val_base[i] = 1024;
     }
 
diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c
index b4e0099567..1f581ec500 100644
--- a/libavcodec/mpegvideo_enc.c
+++ b/libavcodec/mpegvideo_enc.c
@@ -545,6 +545,7 @@ static av_cold int init_slice_buffers(MPVMainEncContext *const m)
         }
 
         if (s2->c.ac_val) {
+            s2->c.dc_val += offset + i;
             s2->c.ac_val += offset;
             offset       += yc_size;
         }
-- 
ffmpeg-codebot

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

  parent reply	other threads:[~2025-06-15 17:03 UTC|newest]

Thread overview: 48+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-06-15 16:53 [FFmpeg-devel] [PATCH 00/47] avcodec/mpeg4videodec: Don't zero blocks twice ffmpegagent
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 01/47] " Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 02/47] avcodec/rv10: Perform RV20 initialization during init Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 03/47] avcodec/rv10: Perform RV20 check only for RV20 Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 04/47] avcodec/ituh263dec: Don't process unnecessarily many coefficients Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 05/47] avcodec/ituh263dec: Remove redundant store Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 06/47] avcodec/ituh263enc: Don't use array unnecessarily Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 07/47] avcodec/ituh263enc: Fix indentation Andreas Rheinhardt
2025-06-15 16:53 ` [FFmpeg-devel] [PATCH 08/47] avcodec/mpeg12dec: Don't store block_last_index unnecessarily Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 09/47] avcodec/rv10: Avoid indirection when reading VLC codes Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 10/47] avcodec/mpeg_er: Mark ff_mpeg_er_init() as av_cold Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 11/47] avcodec/mpegvideo_enc: Remove always-true branch Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 12/47] avcodec/motion_est: Fix indentation Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 13/47] avcodec/svq1enc: Don't free scratchbuf upon error Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 14/47] avcodec/svq1enc: Allocate motion_val{8, 16} during init Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 15/47] avcodec/svq1enc: Stop calling ff_mpv_common_init() Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 16/47] avcodec/svq1enc: Remove write-only c_block_{width, height} Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 17/47] avcodec/svq1enc: Set MpegEncContext.avctx only once Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 18/47] avcodec/svq1enc: Don't initialize unneeded block_index Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 19/47] configure: Factor mpegvideoencdsp out of mpegvideoenc Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 20/47] configure: Relax svq1enc->mpegvideoenc dependency Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 21/47] configure: Relax diracdec->mpegvideoenc dependency Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 22/47] configure: Relax snowenc->mpegvideoenc dependency Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 23/47] avcodec/mpegvideo: Don't reset thread_context ptrs unnecessarily Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 24/47] avcodec/mpegvideo: Don't zero unnecessarily Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 25/47] avcodec/mpegvideo: Defer init of enc slice ctxs in ff_mpv_common_init() Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 26/47] avcodec/mpegvideo_enc: Simplify allocating non-slice buffers Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 27/47] avcodec/ituh263dec: Simplify AIC handling Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 28/47] avcodec/ituh263enc: " Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 29/47] avcodec/flvenc: Combine writing bits Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 30/47] avcodec/flvenc: Move ff_flv2_encode_ac_esc() to ituh263enc.c Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 31/47] avcodec/mpegvideo: Redo resetting intra table entry Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 32/47] avcodec/mpegvideo: Avoid {a, d}c_val[{1, 2}] Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 33/47] avcodec/mpegvideo: Provide alignment hint to compiler Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 34/47] avcodec/mpegvideo: Don't reset AC values of upper-left luma block Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 35/47] avcodec/mpegvideo: Combine stores Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 36/47] avcodec/mpeg4video: Don't use ac_val[1], ac_val[2] when cleaning buffers Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 37/47] avcodec/mpegvideo: Only use a single ac_val pointer Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 38/47] avcodec/mpegvideo: Allocate ac_val jointly Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 39/47] avcodec/mpegvideo: Zero-init mbintra_table Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 40/47] avcodec/mpegvideo: Only keep MpegEncContext.dc_val[0] Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 41/47] avcodec/vc1: Only keep mb_type[0] Andreas Rheinhardt
2025-06-15 16:54 ` Andreas Rheinhardt [this message]
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 43/47] avcodec/{ituh263, mpeg4video}enc: Simplify out-of-slice DC prediction Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 44/47] avcodec/mpeg4video: Move ff_mpeg4_pred_dc() to decoder Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 45/47] avcodec/ituh263enc: Simplify encoding umotion vectors Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 46/47] avcodec/mpegvideo: Reduce stack usage when copying MpegEncContext Andreas Rheinhardt
2025-06-15 16:54 ` [FFmpeg-devel] [PATCH 47/47] avcodec/mpeg4videoenc: Fix data race when using AC prediction Andreas Rheinhardt

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=b469ddf6729f46e47985d4899656d9887c624920.1750006480.git.ffmpegagent@gmail.com \
    --to=ffmpegagent-at-gmail.com@ffmpeg.org \
    --cc=andreas.rheinhardt@outlook.com \
    --cc=ffmpeg-devel@ffmpeg.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
		ffmpegdev@gitmailbox.com
	public-inbox-index ffmpegdev

Example config snippet for mirrors.


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git