Ticket #1041 (closed defect: fixed)
Illegal instruction when opening some files with (HE-)AAC streams
| Reported by: | qyot27 | Owned by: | |
|---|---|---|---|
| Priority: | important | Component: | avcodec |
| Version: | git-master | Keywords: | aac sse |
| Cc: | Blocked By: | ||
| Blocking: | Reproduced by developer: | no | |
| Analyzed by developer: | no |
Description
Sometime between the 18th of February and the 25th of February, the AAC decoder (at least, I think it's the AAC decoder doing this) started breaking on certain files. I've observed it in native builds for both Windows (built with MinGW) and Linux. It seems to fail with files that contain AAC streams that MediaInfo? detects as being HE-AAC, as my purchases from iTunes (which report as standard LC-AAC) can open fine in the same FFmpeg builds that have issues with the streams that report as HE-AAC.
Since I also saw avconv crash with the same sample, I'm guessing this issue arose with one of the merges that occurred during the timeframe between the 18th and 25th. I tried to use git revert on specific AAC-related commits from that date range, but that didn't resolve the error.
Furthermore, even this HE-AAC thing doesn't seem to occur on other computers I have access to that are much newer and more powerful than mine. For reference, my setup has a Celeron Coppermine and 512MBs of PC133 SDRAM. Another computer that didn't exhibit this issue runs an Athlon64 Orleans with 2GBs of DDR2 SDRAM (I think DDR2-800). The actual FFmpeg binary used was the same on both computers.
The output of a working FFmpeg build:
./ffmpeg.exe -v 9 -loglevel 99 -i $HOME/sample-5.flv
ffmpeg version r37999 git-57182b9 Copyright (c) 2000-2012 the FFmpeg developers
built on Feb 18 2012 21:44:50 with gcc 4.6.1
configuration: --prefix=/home/qyot27/win32_build --cross-prefix=i686-w64-mingw32- --enable-gpl --enable-version3 --disable-w32threads --enable-memalign-hack --disable-decoder=utvideo --enable-libutvideo --enable-libxvid --disable-encoder=mpeg4 --enable-avisynth --cpu=pentium3 --extra-cflags='-I/home/qyot27/win32_build/include -march=pentium3 -mtune=pentium3 -DPTW32_STATIC_LIB' --extra-ldflags=-L/home/qyot27/win32_build/lib --target-os=mingw32 --arch=x86
libavutil 51. 39.100 / 51. 39.100
libavcodec 54. 3.100 / 54. 3.100
libavformat 54. 1.100 / 54. 1.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 62.101 / 2. 62.101
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 6.100 / 0. 6.100
libpostproc 52. 0.100 / 52. 0.100
[flv @ 00142840] Format flv probed with size=2048 and score=100
[h264 @ 00148A00] no picture
[flv @ 00142840] All info found
rfps: 23.750000 0.011777
Last message repeated 1 times
rfps: 23.833333 0.004697
rfps: 23.916667 0.000836
rfps: 24.000000 0.000195
rfps: 24.083333 0.002773
rfps: 24.166667 0.008572
Last message repeated 1 times
rfps: 24.250000 0.017590
Last message repeated 1 times
rfps: 47.666667 0.018788
rfps: 47.750000 0.009456
rfps: 47.833333 0.003344
rfps: 47.916667 0.000452
rfps: 48.000000 0.000779
rfps: 48.083333 0.004327
rfps: 48.166667 0.011094
rfps: 23.976024 0.000049
Input #0, flv, from '/home/qyot27/sample-5.flv':
Metadata:
moovPosition : 272105568
avcprofile : 77
avclevel : 31
aacaot : 2
videoframerate : 24
audiochannels : 2
Duration: 00:21:22.80, start: 0.000000, bitrate: 32 kb/s
Stream #0:0, 41, 1/1000: Video: h264 (Main), yuv420p, 1280x720 [SAR 1:1 DAR 16:9], 1001/48000, 23.98 tbr, 1k tbn, 47.95 tbc
Stream #0:1, 36, 1/1000: Audio: aac, 44100 Hz, stereo, s16
At least one output file must be specified
The output of a non-working FFmpeg:
./ffmpeg -v 9 -loglevel 99 -i $HOME/sample-5.flv ffmpeg version N-38556-g456d65a Copyright (c) 2000-2012 the FFmpeg developers built on Mar 5 2012 14:32:36 with gcc 4.6.1 configuration: --prefix=/home/qyot27/linuxffmpeg_build --enable-gpl --enable-version3 --enable-libxvid --disable-encoder=mpeg4 --cpu=pentium3 --extra-cflags='-march=pentium3 -mtune=pentium3' libavutil 51. 41.100 / 51. 41.100 libavcodec 54. 8.100 / 54. 8.100 libavformat 54. 2.100 / 54. 2.100 libavdevice 53. 4.100 / 53. 4.100 libavfilter 2. 63.100 / 2. 63.100 libswscale 2. 1.100 / 2. 1.100 libswresample 0. 7.100 / 0. 7.100 libpostproc 52. 0.100 / 52. 0.100 [flv @ 0x95eeac0] Format flv probed with size=2048 and score=100 [h264 @ 0x95f4da0] no picture Illegal instruction
Only the audio (demuxed using the working build), still errors out with newer builds:
./ffmpeg -v 9 -loglevel 99 -i $HOME/sample-5.aac ffmpeg version N-38556-g456d65a Copyright (c) 2000-2012 the FFmpeg developers built on Mar 5 2012 14:32:36 with gcc 4.6.1 configuration: --prefix=/home/qyot27/linuxffmpeg_build --enable-gpl --enable-version3 --enable-libxvid --disable-encoder=mpeg4 --cpu=pentium3 --extra-cflags='-march=pentium3 -mtune=pentium3' libavutil 51. 41.100 / 51. 41.100 libavcodec 54. 8.100 / 54. 8.100 libavformat 54. 2.100 / 54. 2.100 libavdevice 53. 4.100 / 53. 4.100 libavfilter 2. 63.100 / 2. 63.100 libswscale 2. 1.100 / 2. 1.100 libswresample 0. 7.100 / 0. 7.100 libpostproc 52. 0.100 / 52. 0.100 [aac @ 0xa2f7ac0] Format aac probed with size=2048 and score=51 Illegal instruction
Remuxing instead to MKV or MP4 produces virtually the same result, with 'Illegal instruction' after the Format probing message (and the h264 no picture message, if the video stream has been retained; otherwise it jumps directly to 'Illegal instruction').
To show that it's not all AAC files, the output of a different file (bought from iTunes):
./ffmpeg -v 9 -loglevel 99 -i /home/qyot27/Music/Songbird\ Music/Depeche\ Mode\ -\ Never\ Let\ Me\ Down\ Again\ \(Single\ Version\).m4a
ffmpeg version N-38556-g456d65a Copyright (c) 2000-2012 the FFmpeg developers
built on Mar 5 2012 14:32:36 with gcc 4.6.1
configuration: --prefix=/home/qyot27/linuxffmpeg_build --enable-gpl --enable-version3 --enable-libxvid --disable-encoder=mpeg4 --cpu=pentium3 --extra-cflags='-march=pentium3 -mtune=pentium3'
libavutil 51. 41.100 / 51. 41.100
libavcodec 54. 8.100 / 54. 8.100
libavformat 54. 2.100 / 54. 2.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 63.100 / 2. 63.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 7.100 / 0. 7.100
libpostproc 52. 0.100 / 52. 0.100
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaee3ac0] Format mov,mp4,m4a,3gp,3g2,mj2 probed with size=2048 and score=100
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaee3ac0] ISO: File Type Major Brand: M4A
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaee3ac0] All info found
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from '/home/qyot27/Music/Songbird Music/Depeche Mode - Never Let Me Down Again (Single Version).m4a':
Metadata:
major_brand : M4A
minor_version : 0
compatible_brands: M4A mp42isom
creation_time : 1941-06-18 03:43:01
title : Never Let Me Down Again (Single Version)
artist : Depeche Mode
album_artist : Depeche Mode
album : Singles Box 4
genre : Rock
track : 1/8
disc : 2/6
date : 2004-12-07T08:00:00Z
gapless_playback: 0
copyright : ℗ 2004 Warner Bros. Records. Manufactured and Marketed by Warner Strategic Marketing
media_type : 1
Duration: 00:04:22.03, start: 0.000000, bitrate: 278 kb/s
Stream #0:0(eng), 1, 1/44100: Audio: aac (mp4a / 0x6134706D), 44100 Hz, stereo, s16, 259 kb/s
Metadata:
creation_time : 1941-06-18 03:43:01
handler_name :
At least one output file must be specified
or one I'd encoded myself:
./ffmpeg -v 9 -loglevel 99 -i /home/qyot27/Music/Songbird\ Music/Third\ Eye\ Blind\ -\ Semi-Charmed\ Life.m4a
ffmpeg version N-38556-g456d65a Copyright (c) 2000-2012 the FFmpeg developers
built on Mar 5 2012 14:32:36 with gcc 4.6.1
configuration: --prefix=/home/qyot27/linuxffmpeg_build --enable-gpl --enable-version3 --enable-libxvid --disable-encoder=mpeg4 --cpu=pentium3 --extra-cflags='-march=pentium3 -mtune=pentium3'
libavutil 51. 41.100 / 51. 41.100
libavcodec 54. 8.100 / 54. 8.100
libavformat 54. 2.100 / 54. 2.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 63.100 / 2. 63.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 7.100 / 0. 7.100
libpostproc 52. 0.100 / 52. 0.100
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaba5ac0] Format mov,mp4,m4a,3gp,3g2,mj2 probed with size=2048 and score=100
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaba5ac0] ISO: File Type Major Brand: mp42
[mov,mp4,m4a,3gp,3g2,mj2 @ 0xaba5ac0] All info found
Input #0, mov,mp4,m4a,3gp,3g2,mj2, from '/home/qyot27/Music/Songbird Music/Third Eye Blind - Semi-Charmed Life.m4a':
Metadata:
major_brand : mp42
minor_version : 0
compatible_brands: mp42isom
creation_time : 2008-02-17 22:24:11
track : 5/21
genre : Soundtrack
album : ESPN Presents Jock Rock 2000
artist : Third Eye Blind
title : Semi-Charmed Life
date : 1999
Duration: 00:04:24.60, start: 0.000000, bitrate: 208 kb/s
Stream #0:0(eng), 1, 1/44100: Audio: aac (mp4a / 0x6134706D), 44100 Hz, stereo, s16, 207 kb/s
Metadata:
creation_time : 2008-02-17 22:24:11
handler_name :
Stream #0:1(eng), 0, 1/1000: Data: none (mp4s / 0x7334706D)
Metadata:
creation_time : 2008-02-17 22:24:43
handler_name :
Stream #0:2(eng), 0, 1/1000: Data: none (mp4s / 0x7334706D)
Metadata:
creation_time : 2008-02-17 22:24:43
handler_name :
At least one output file must be specified
I don't know if a sample would be helpful if this actually requires a decade-old computer to reproduce the issue, but I've attached one anyway.
Attachments
-
sample-5.aac
(278.7 KB) -
added by qyot27 16 months ago.
Change History
comment:1 Changed 16 months ago by cehoyos
- Keywords aac added; aacdec removed
Since I am unable to reproduce your problem on my Pentium-M:
Can you reproduce the problem with "./configure && make"?
Please provide a backtrace with gdb and / or please use git bisect to find the change introducing the problem.
comment:2 Changed 16 months ago by llogan
I'm unable to duplicate this on a 400 MHz Pentium II "Deschutes" with plain ./configure. I also recommend trying with a plain configure to try to narrow down the issue.
comment:3 Changed 16 months ago by qyot27
Rebuilt FFmpeg with just ./configure, the Illegal instruction still happens. The backtrace seems to point to an SSE issue.
The full output of gdb as directed on the Bug Reporting page:
$ gdb ffmpeg_g
GNU gdb (Ubuntu/Linaro 7.3-0ubuntu2) 7.3-2011.08
Copyright (C) 2011 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law. Type "show copying"
and "show warranty" for details.
This GDB was configured as "i686-linux-gnu".
For bug reporting instructions, please see:
<http://bugs.launchpad.net/gdb-linaro/>...
Reading symbols from /home/qyot27/ffmpeg_g...done.
(gdb) r -v 9 -loglevel 99 -i sample-5.aac
Starting program: /home/qyot27/ffmpeg_g -v 9 -loglevel 99 -i sample-5.aac
[Thread debugging using libthread_db enabled]
ffmpeg version N-38596-gf095391 Copyright (c) 2000-2012 the FFmpeg developers
built on Mar 6 2012 14:21:10 with gcc 4.6.1
configuration: --prefix=/home/qyot27/ffmpeg_build
libavutil 51. 41.100 / 51. 41.100
libavcodec 54. 10.100 / 54. 10.100
libavformat 54. 2.100 / 54. 2.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 63.100 / 2. 63.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 7.100 / 0. 7.100
[aac @ 0x8e85ac0] Format aac probed with size=2048 and score=51
Program received signal SIGILL, Illegal instruction.
ff_sbr_hf_g_filt_sse.loop1 () at libavcodec/x86/sbrdsp.asm:107
107 movq [r0], m2
(gdb) bt
#0 ff_sbr_hf_g_filt_sse.loop1 () at libavcodec/x86/sbrdsp.asm:107
#1 0xb7f856b0 in ?? ()
Backtrace stopped: previous frame inner to this frame (corrupt stack?)
(gdb) disass $pc-32,$pc+32
Dump of assembler code from 0x86d4f35 to 0x86d4f75:
0x086d4f35 <ff_sbr_hf_g_filt_sse.loop4+59>: adc %al,0x500c1(%ecx)
0x086d4f3b <ff_sbr_hf_g_filt_sse.loop4+65>: add %al,-0x468afb3d(%ebx)
0x086d4f41 <ff_sbr_hf_g_filt_sse.loop4+71>: and $0x3,%edi
0x086d4f44 <ff_sbr_hf_g_filt_sse.loop4+74>: je 0x86d4f68 <ff_sbr_hf_g_filt_sse.end>
0x086d4f46 <ff_sbr_hf_g_filt_sse.loop1+0>: movss (%edx),%xmm0
0x086d4f4a <ff_sbr_hf_g_filt_sse.loop1+4>: movq (%ecx),%xmm2
0x086d4f4e <ff_sbr_hf_g_filt_sse.loop1+8>: punpckldq %xmm0,%xmm0
0x086d4f52 <ff_sbr_hf_g_filt_sse.loop1+12>: mulps %xmm0,%xmm2
=> 0x086d4f55 <ff_sbr_hf_g_filt_sse.loop1+15>: movq %xmm2,(%eax)
0x086d4f59 <ff_sbr_hf_g_filt_sse.loop1+19>: add $0x8,%eax
0x086d4f5c <ff_sbr_hf_g_filt_sse.loop1+22>: add $0x4,%edx
0x086d4f5f <ff_sbr_hf_g_filt_sse.loop1+25>: add $0x140,%ecx
0x086d4f65 <ff_sbr_hf_g_filt_sse.loop1+31>: dec %edi
0x086d4f66 <ff_sbr_hf_g_filt_sse.loop1+32>: jne 0x86d4f46 <ff_sbr_hf_g_filt_sse.loop1>
0x086d4f68 <ff_sbr_hf_g_filt_sse.end+0>: pop %edi
0x086d4f69 <ff_sbr_hf_g_filt_sse.end+1>: pop %esi
0x086d4f6a <ff_sbr_hf_g_filt_sse.end+2>: pop %ebx
0x086d4f6b <ff_sbr_hf_g_filt_sse.end+3>: ret
0x086d4f6c <ff_sbr_hf_g_filt_sse.end+4>: nop
0x086d4f6d <ff_sbr_hf_g_filt_sse.end+5>: nop
---Type <return> to continue, or q <return> to quit---
0x086d4f6e <ff_sbr_hf_g_filt_sse.end+6>: nop
0x086d4f6f <ff_sbr_hf_g_filt_sse.end+7>: nop
0x086d4f70 <auto_matrix+0>: push %ebp
0x086d4f71 <auto_matrix+1>: mov %eax,%ebp
0x086d4f73 <auto_matrix+3>: push %edi
0x086d4f74 <auto_matrix+4>: push %esi
End of assembler dump.
(gdb) info all-registers
eax 0xb7f91de0 -1208410656
ecx 0xb7fcc8c0 -1208170304
edx 0xb7f96880 -1208391552
ebx 0x0 0
esp 0xbfffd020 0xbfffd020
ebp 0xb7f91d20 0xb7f91d20
esi 0x2 2
edi 0x3 3
eip 0x86d4f55 0x86d4f55 <ff_sbr_hf_g_filt_sse.loop1+15>
eflags 0x210206 [ PF IF RF ID ]
cs 0x73 115
ss 0x7b 123
ds 0x7b 123
es 0x7b 123
fs 0x0 0
gs 0x33 51
st0 -inf (raw 0xffff0000000000000000)
st1 -inf (raw 0xffff0000000000000000)
st2 -0 (raw 0x80000000000000000000)
st3 0.9921586513519287109375 (raw 0x3ffefdfe1c0000000000)
st4 0.98437878945247803130769170820713043 (raw 0x3ffefc003f9393100000)
st5 1.1920928955078125e-07 (raw 0x3fe88000000000000000)
---Type <return> to continue, or q <return> to quit---
st6 0 (raw 0x00000000000000000000)
st7 1.58489322662353515625 (raw 0x3fffcaddc80000000000)
fctrl 0x37f 895
fstat 0x461 1121
ftag 0x101a 4122
fiseg 0x73 115
fioff 0x863674c 140732236
foseg 0x7b 123
fooff 0xb7fd8568 -1208122008
fop 0x5d8 1496
xmm0 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x0, 0x0, 0xc0, 0xff, 0x0 <repeats 12 times>}, v8_int16 = {0x0,
0xffc0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int32 = {0xffc00000, 0x0, 0x0,
0x0}, v2_int64 = {0xffc00000, 0x0},
uint128 = 0x000000000000000000000000ffc00000}
xmm1 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x0 <repeats 16 times>}, v8_int16 = {0x0, 0x0, 0x0, 0x0, 0x0,
0x0, 0x0, 0x0}, v4_int32 = {0x0, 0x0, 0x0, 0x0}, v2_int64 = {0x0, 0x0},
uint128 = 0x00000000000000000000000000000000}
xmm2 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x0, 0x0, 0xc0, 0xff, 0x0 <repeats 12 times>}, v8_int16 = {0x0,
0xffc0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}, v4_int32 = {0xffc00000, 0x0, 0x0,
0x0}, v2_int64 = {0xffc00000, 0x0},
---Type <return> to continue, or q <return> to quit---
uint128 = 0x000000000000000000000000ffc00000}
xmm3 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x86, 0xe5, 0x4b, 0x2d, 0x4f, 0xd5, 0x54, 0x2d, 0x0, 0x0, 0x0,
0x0, 0x0, 0x0, 0x0, 0x0}, v8_int16 = {0xe586, 0x2d4b, 0xd54f, 0x2d54, 0x0,
0x0, 0x0, 0x0}, v4_int32 = {0x2d4be586, 0x2d54d54f, 0x0, 0x0}, v2_int64 = {
0x2d54d54f2d4be586, 0x0}, uint128 = 0x00000000000000002d54d54f2d4be586}
xmm4 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0xb5, 0xb7, 0xd8, 0x2a, 0x8a, 0xdf, 0xc6, 0x2a, 0x3a, 0x62,
0x1d, 0x2c, 0x6f, 0x90, 0xe3, 0x2a}, v8_int16 = {0xb7b5, 0x2ad8, 0xdf8a,
0x2ac6, 0x623a, 0x2c1d, 0x906f, 0x2ae3}, v4_int32 = {0x2ad8b7b5,
0x2ac6df8a, 0x2c1d623a, 0x2ae3906f}, v2_int64 = {0x2ac6df8a2ad8b7b5,
0x2ae3906f2c1d623a}, uint128 = 0x2ae3906f2c1d623a2ac6df8a2ad8b7b5}
xmm5 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x7c, 0xf9, 0xfa, 0x2c, 0xc9, 0x25, 0x5c, 0x2d, 0xfe, 0xbe,
0x5b, 0x29, 0x60, 0x85, 0x21, 0x2a}, v8_int16 = {0xf97c, 0x2cfa, 0x25c9,
0x2d5c, 0xbefe, 0x295b, 0x8560, 0x2a21}, v4_int32 = {0x2cfaf97c,
0x2d5c25c9, 0x295bbefe, 0x2a218560}, v2_int64 = {0x2d5c25c92cfaf97c,
0x2a218560295bbefe}, uint128 = 0x2a218560295bbefe2d5c25c92cfaf97c}
xmm6 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x28, 0xe4, 0x7b, 0x37, 0xa0, 0x37, 0x82, 0x37, 0x2c, 0x59,
0x70, 0x37, 0xbd, 0x65, 0x6b, 0x37}, v8_int16 = {0xe428, 0x377b, 0x37a0,
0x3782, 0x592c, 0x3770, 0x65bd, 0x376b}, v4_int32 = {0x377be428,
0x378237a0, 0x3770592c, 0x376b65bd}, v2_int64 = {0x378237a0377be428,
---Type <return> to continue, or q <return> to quit---
0x376b65bd3770592c}, uint128 = 0x376b65bd3770592c378237a0377be428}
xmm7 {v4_float = {0x0, 0x0, 0x0, 0x0}, v2_double = {0x0, 0x0},
v16_int8 = {0x80, 0xe1, 0x10, 0x35, 0xe3, 0x2a, 0xf, 0x36, 0x2f, 0x5a, 0x25,
0x36, 0xd4, 0xff, 0xa9, 0x36}, v8_int16 = {0xe180, 0x3510, 0x2ae3, 0x360f,
0x5a2f, 0x3625, 0xffd4, 0x36a9}, v4_int32 = {0x3510e180, 0x360f2ae3,
0x36255a2f, 0x36a9ffd4}, v2_int64 = {0x360f2ae33510e180,
0x36a9ffd436255a2f}, uint128 = 0x36a9ffd436255a2f360f2ae33510e180}
mxcsr 0x1fb2 [ DE UE PE IM DM ZM OM UM PM ]
mm0 {uint64 = 0x0, v2_int32 = {0x0, 0x0}, v4_int16 = {0x0, 0x0,
0x0, 0x0}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}}
mm1 {uint64 = 0x0, v2_int32 = {0x0, 0x0}, v4_int16 = {0x0, 0x0,
0x0, 0x0}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}}
mm2 {uint64 = 0x0, v2_int32 = {0x0, 0x0}, v4_int16 = {0x0, 0x0,
0x0, 0x0}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}}
mm3 {uint64 = 0xfdfe1c0000000000, v2_int32 = {0x0, 0xfdfe1c00},
v4_int16 = {0x0, 0x0, 0x1c00, 0xfdfe}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0,
0x1c, 0xfe, 0xfd}}
mm4 {uint64 = 0xfc003f9393100000, v2_int32 = {0x93100000,
0xfc003f93}, v4_int16 = {0x0, 0x9310, 0x3f93, 0xfc00}, v8_int8 = {0x0,
0x0, 0x10, 0x93, 0x93, 0x3f, 0x0, 0xfc}}
mm5 {uint64 = 0x8000000000000000, v2_int32 = {0x0, 0x80000000},
v4_int16 = {0x0, 0x0, 0x0, 0x8000}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
0x0, 0x80}}
---Type <return> to continue, or q <return> to quit---
mm6 {uint64 = 0x0, v2_int32 = {0x0, 0x0}, v4_int16 = {0x0, 0x0,
0x0, 0x0}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}}
mm7 {uint64 = 0xcaddc80000000000, v2_int32 = {0x0, 0xcaddc800},
v4_int16 = {0x0, 0x0, 0xc800, 0xcadd}, v8_int8 = {0x0, 0x0, 0x0, 0x0, 0x0,
0xc8, 0xdd, 0xca}}
(gdb)
I'll also go about trying to find the particular change. Looking at the commit log, it may have been this one, since it makes reference to sbr_hf_g_filt.
comment:4 Changed 16 months ago by qyot27
Update: Yes, apparently it is that commit.
git revert 2784d187919b48022a89633fb3b5a99ca97cf869 seems to resolve it.
comment:5 Changed 16 months ago by qyot27
Update 2
Reverting 2784d187919b48022a89633fb3b5a99ca97cf869 only allows it to use -acodec copy and not error out, but trying to decode HE-AAC results in a silent track with clicks. For it to decode like before, 34454c761f01275d4adaf40df6d70a59011c4a6c also needs to be reverted.
comment:6 Changed 16 months ago by cehoyos
- Priority changed from normal to important
- Status changed from new to open
- Component changed from undetermined to avcodec
comment:7 Changed 16 months ago by cehoyos
@qyot27:
Could you test this patch?
http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/141806
comment:8 Changed 16 months ago by qyot27
I actually drew that patch out of Libav and tested it yesterday before it got merged in, and unless I screwed something up (which is very possible, as I can't quite remember if I'd been messing with git revert on that test), it only eliminated the error; decoding resulted in a silent track.
I'll rebuild with a fresh clone and test.
comment:9 Changed 16 months ago by cehoyos
New patch to test:
http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/141844
comment:10 Changed 16 months ago by qyot27
The new patch still results in silent decoding. I also tried with the unpatched git master since the most recent merge brought that change in, and it has the silence issue too. The output of a decoding operation:
./ffmpeg_g -v 9 -loglevel 99 -i sample-5.aac -acodec pcm_s16le testaac.wav
ffmpeg version N-38662-gbf807a5 Copyright (c) 2000-2012 the FFmpeg developers
built on Mar 7 2012 22:19:10 with gcc 4.6.1
configuration: --prefix=/home/qyot27/ffmpeg_build --extra-cflags='-m32 -DARCH_X86_32' --extra-ldflags=-m32
libavutil 51. 42.100 / 51. 42.100
libavcodec 54. 10.100 / 54. 10.100
libavformat 54. 2.100 / 54. 2.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 63.100 / 2. 63.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 7.100 / 0. 7.100
[aac @ 0x9133ac0] Format aac probed with size=2048 and score=51
[aac @ 0x9133ac0] max_analyze_duration 5000000 reached at 5015510
[aac @ 0x9133ac0] Estimating duration from bitrate, this may be inaccurate
Input #0, aac, from 'sample-5.aac':
Duration: 00:00:23.79, bitrate: 95 kb/s
Stream #0:0, 111, 1/28224000: Audio: aac, 44100 Hz, stereo, s16, 95 kb/s
Output #0, wav, to 'testaac.wav':
Metadata:
encoder : Lavf54.2.100
Stream #0:0, 0, 1/44100: Audio: pcm_s16le ([1][0][0][0] / 0x0001), 44100 Hz, stereo, s16, 1411 kb/s
Stream mapping:
Stream #0:0 -> #0:0 (aac -> pcm_s16le)
Press [q] to stop, [?] for help
size= 4096kB time=00:00:23.77 bitrate=1411.2kbits/s
video:0kB audio:4096kB global headers:0kB muxing overhead 0.001097%
comment:11 Changed 16 months ago by cehoyos
Does decoding work if you compile with "./configure --disable-asm && make" (or --disable-sse)?
comment:12 Changed 16 months ago by qyot27
Yes, decoding is fine if built with --disable-sse.
comment:13 Changed 16 months ago by cehoyos
Does decoding work if you compile with "./configure --disable-yasm && make" ?
If it does work, you could look at dsputil_init_sse() in libavcodec/x86/dsputil_mmx.c (line 2700) and test if removing the line with "scalarproduct_float" or "butterflies_float_interleave" fixes the problem.
(Sorry, but I cannot test myself atm.)
comment:14 Changed 16 months ago by qyot27
Building with --disable-yasm = decoding works
Commenting out scalarproduct_float = still silent
Commenting out butterflies_float_interleave = still silent
Commenting out both lines = still silent
Forcing SSE2 by changing line 35 of libavcodec/x86/sbrdsp_init.c from
if (mm_flags & AV_CPU_FLAG_SSE) {
to
if (mm_flags & AV_CPU_FLAG_SSE2) {
makes decoding work, but I don't know if that's the 'ideal' solution here.
comment:15 Changed 16 months ago by cehoyos
Is the responsible function ff_sbr_sum_square_sse() or ff_sbr_hf_g_filt_sse() (or both)?
comment:16 Changed 16 months ago by qyot27
ff_sbr_sum_square_sse() seems to be the one causing the issue. Commenting out those two lines in sbrdsp_init.c makes decoding work.
comment:17 Changed 16 months ago by cehoyos
Commenting out only "s->sum_square = ff_sbr_sum_square_sse;" does not help?
comment:18 Changed 16 months ago by qyot27
I didn't know which one would need to be removed, so I commented out both of the lines that referenced ff_sbr_sum_square_sse just to be overly cautious.
comment:19 Changed 16 months ago by qyot27
Now that I've had the time to test again, it is only the "s->sum_square = ff_sbr_sum_square_sse;" that causes the issue, not both lines.
comment:20 follow-up: ↓ 21 Changed 15 months ago by ami_stuff
Probably related to ticket #213. My own build (compiled with --disable-yasm) works correctly.
comment:21 in reply to: ↑ 20 Changed 15 months ago by qyot27
Replying to ami_stuff:
Probably related to ticket #213. My own build (compiled with --disable-yasm) works correctly.
Except that this is caused by a line in libavcodec/x86/sbrdsp_init.c, a source file which didn't even exist until a little less than a month ago. Prior to which, HE-AAC decoding worked fine, and still works fine if the problematic line is commented out (as worked out from comments #14-19). And even with this issue, LC-AAC decoding still works.
comment:22 Changed 15 months ago by ami_stuff
This is what I get on Athlon XP:
http://www.datafilehost.com/download-97c1ba24.html
C:\>ffmpeg -i sample-5.aac out.wav
ffmpeg version N-38938-ge01f478 Copyright (c) 2000-2012 the FFmpeg developers
built on Mar 20 2012 02:36:44 with gcc 4.5.0 20100414 (Fedora MinGW 4.5.0-1.fc
14)
configuration: --prefix=/var/www/users/research/ffmpeg/snapshots/build --arch=
x86 --target-os=mingw32 --cross-prefix=i686-pc-mingw32- --cc='ccache i686-pc-min
gw32-gcc' --enable-w32threads --enable-memalign-hack --enable-runtime-cpudetect
--enable-cross-compile --enable-static --disable-shared --extra-libs='-lws2_32 -
lwinmm' --extra-cflags='--static -I/var/www/users/research/ffmpeg/snapshots/buil
d/include' --extra-ldflags='-static -L/var/www/users/research/ffmpeg/snapshots/b
uild/lib' --enable-bzlib --enable-zlib --enable-gpl --enable-version3 --enable-n
onfree --enable-libx264 --enable-libspeex --enable-libtheora --enable-libvorbis
--enable-libfaac --enable-libxvid --enable-libopencore-amrnb --enable-libopencor
e-amrwb --enable-libmp3lame --enable-libvpx --disable-decoder=libvpx
libavutil 51. 42.100 / 51. 42.100
libavcodec 54. 12.100 / 54. 12.100
libavformat 54. 2.100 / 54. 2.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 65.101 / 2. 65.101
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 7.100 / 0. 7.100
libpostproc 52. 0.100 / 52. 0.100
[aac @ 0x1da9be0] max_analyze_duration 5000000 reached at 5015510
[aac @ 0x1da9be0] Estimating duration from bitrate, this may be inaccurate
Input #0, aac, from 'sample-5.aac':
Duration: 00:00:23.79, bitrate: 95 kb/s
Stream #0:0: Audio: aac, 44100 Hz, stereo, s16, 95 kb/s
File 'out.wav' already exists. Overwrite ? [y/N] y
Output #0, wav, to 'out.wav':
Metadata:
encoder : Lavf54.2.100
Stream #0:0: Audio: pcm_s16le ([1][0][0][0] / 0x0001), 44100 Hz, stereo, s16
, 1411 kb/s
Stream mapping:
Stream #0:0 -> #0:0 (aac -> pcm_s16le)
Press [q] to stop, [?] for help
size= 4096kB time=00:00:23.77 bitrate=1411.2kbits/s
video:0kB audio:4096kB global headers:0kB muxing overhead 0.001097%
comment:23 Changed 15 months ago by qyot27
The issue is (as far as I can tell), that the assembly for function ff_sbr_sum_square_sse is somehow not fully correct SSE, perhaps - as was the case with ff_sbr_hf_g_filt_sse before it got fixed - mixed with SSE2 instructions. Processors that lack SSE2 support, like the Pentium III and Athlon XP, will generate silent output when attempting to process SBR-laced data because that assembly isn't strictly SSE and can't correctly be used on such processors. Comment out that specific line which calls it, and decoding works properly without having to disable all assembly optimizations. This was already known.
As evidence of this, the below patch is roughly what I've been using for the last 10 days since it was pinned down (sans the git-am metadata):
http://pastebin.com/P7nDj6vH
comment:25 follow-up: ↓ 26 Changed 15 months ago by reimar
Could you please try below change? I always build for 64 bit, so not even compilation is tested from my side...
diff --git a/libavcodec/x86/sbrdsp.asm b/libavcodec/x86/sbrdsp.asm
index 31a1c8b..682531d 100644
--- a/libavcodec/x86/sbrdsp.asm
+++ b/libavcodec/x86/sbrdsp.asm
@@ -68,7 +68,7 @@ cglobal sbr_sum_square, 2, 3, 6
shufps m0, m0, 1
addss m0, m1
%if ARCH_X86_64 == 0
- movd r0m, m0
+ movss r0m, m0
fld dword r0m
%endif
RET
comment:26 in reply to: ↑ 25 Changed 15 months ago by qyot27
Replying to reimar:
Could you please try below change? I always build for 64 bit, so not even compilation is tested from my side...
diff --git a/libavcodec/x86/sbrdsp.asm b/libavcodec/x86/sbrdsp.asm index 31a1c8b..682531d 100644 --- a/libavcodec/x86/sbrdsp.asm +++ b/libavcodec/x86/sbrdsp.asm @@ -68,7 +68,7 @@ cglobal sbr_sum_square, 2, 3, 6 shufps m0, m0, 1 addss m0, m1 %if ARCH_X86_64 == 0 - movd r0m, m0 + movss r0m, m0 fld dword r0m %endif RET
Tested. Decoding works correctly with this patch.
comment:27 Changed 15 months ago by reimar
- Status changed from open to closed
- Resolution set to fixed
Thanks, pushed and thus closing.
comment:28 Changed 15 months ago by ami_stuff
Fixed for me as well.
