PR #20376 opened by damitha URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20376 Patch URL: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/20376.patch
Have noticed that escape coefficient on spectral data can be less than 16 on rare occasions. The failure reproduces deterministically on macOS (ARM) and x86-64, at the same frame for the same source. Example | field | Header | |---------|---------| | fabsf(in[i+j]) | 275295.156 | | Q | 0.00014516688 | | ROUNDING | 0.105400003 | | coef_float | 15.99999932 | | coef_int | 15 | The casting to int ends up being 15. This breaks the `put_bits(pb, len - 4 + 1, (1 << (len - 4 + 1)) - 2);` to be noop. Then `put_sbits(pb, len, coef);` becomes 000. This cause the decoder to have over read here. Eventually the fails at the sections_data() read on the next channel. (At least on the samples I have noticed). So forcing the coefficient minimum to be 16. Issue Link: https://code.ffmpeg.org/FFmpeg/FFmpeg/issues/20373 >From 6d4dbf148b8f272c95669636e0a6b10f1797552e Mon Sep 17 00:00:00 2001 From: damitha <damitha.gunaward...@gmail.com> Date: Sun, 31 Aug 2025 06:50:48 +1000 Subject: [PATCH] add clamping for escpase coeff --- libavcodec/aaccoder.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libavcodec/aaccoder.c b/libavcodec/aaccoder.c index 96915c9731..7f1c4cdcc1 100644 --- a/libavcodec/aaccoder.c +++ b/libavcodec/aaccoder.c @@ -173,7 +173,7 @@ static av_always_inline float quantize_and_encode_band_cost_template( if (BT_ESC) { for (int j = 0; j < 2; j++) { if (ff_aac_codebook_vectors[cb-1][curidx*2+j] == 64.0f) { - int coef = av_clip_uintp2(quant(fabsf(in[i+j]), Q, ROUNDING), 13); + int coef = av_clip(quant(fabsf(in[i+j]), Q, ROUNDING), 16, (1 << 13) - 1); int len = av_log2(coef); put_bits(pb, len - 4 + 1, (1 << (len - 4 + 1)) - 2); -- 2.49.1 _______________________________________________ ffmpeg-devel mailing list -- ffmpeg-devel@ffmpeg.org To unsubscribe send an email to ffmpeg-devel-le...@ffmpeg.org
