[issue2082] Crash in FFT-SSE code when playing mmst URI
Michael Niedermayer michae...@gmx.at added the comment: On Wed, Jul 14, 2010 at 11:07:09PM +, Ronald S. Bultje wrote: A make clean fixed it. -- status: new - closed substatus: new - invalid Do we consider bugs in the compilation dependancies to be invalid? [...] FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082
[issue2082] Crash in FFT-SSE code when playing mmst URI
Mans Rullgard m...@mansr.com added the comment: Michael Niedermayer iss...@roundup.ffmpeg.org writes: On Wed, Jul 14, 2010 at 11:07:09PM +, Ronald S. Bultje wrote: A make clean fixed it. -- status: new - closed substatus: new - invalid Do we consider bugs in the compilation dependancies to be invalid? Unless someone provides me with exact steps to reproduce the problem, I will not look at it. FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082
[issue2082] Crash in FFT-SSE code when playing mmst URI
Ronald S. Bultje rsbul...@gmail.com added the comment: It crashes (also in the C version) in this piece of code: __asm__ volatile( 1: \n CMUL(%0, %%xmm0, %%xmm1) CMUL(%1, %%xmm4, %%xmm5) Particularly, the second mulps in CMUL. That's a multiply with variables in s-tsin/s-tcos, suggesting these aren't properly assigned. (gdb) print tcos $1 = value temporarily unavailable, due to optimizations (gdb) print s $2 = (FFTContext *) 0x2de2490 (gdb) print *s $3 = { nbits = 10, inverse = 1, revtab = 0x1842200, tmp_buf = 0x183ca00, mdct_size = 0, mdct_bits = 0, tcos = 0x1000, tsin = 0xc, fft_permute = 0x182a800, fft_calc = 0x182b800, imdct_calc = 0x45b6f0 ff_imdct_calc_sse, imdct_half = 0x45b3f0 ff_imdct_half_sse, mdct_calc = 0x225730 ff_mdct_calc_c, permutation = 0 } (gdb) print *s-tcos $4 = -1.58164644e+38 (gdb) print *s-tsin Cannot access memory at address 0xc (gdb) Where are these allocated, and by who? It looks like they're broke here. FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082
[issue2082] Crash in FFT-SSE code when playing mmst URI
Ronald S. Bultje rsbul...@gmail.com added the comment: I also think wmadec.c never calls ff_imdct_init(), which is probably related to all this? FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082
[issue2082] Crash in FFT-SSE code when playing mmst URI
Ronald S. Bultje rsbul...@gmail.com added the comment: Valgrind suggests something else. First error: ==70074== Invalid write of size 4 ==70074==at 0x2F285D: ff_rdft_init (in ./ffplay) ==70074== Address 0x13f6ee28 is 8 bytes after a block of size 80 alloc'd ==70074==at 0xB5042A: memalign (vg_replace_malloc.c:532) ==70074==by 0xB50475: posix_memalign (vg_replace_malloc.c:660) ==70074==by 0x4B2510: av_malloc (in ./ffplay) Indeed, that code writes into a struct RDFTContext, sizeof() which is 80, at position 0x58 (i.e. 88-92): 0x002f2856 ff_rdft_init+326: inc%esi 0x002f2857 ff_rdft_init+327: cmp0x20(%esp),%esi 0x002f285b ff_rdft_init+331: jne0x2f2830 ff_rdft_init+288 0x002f285d ff_rdft_init+333: movl $0x2f2870,0x58(%ebp) 0x002f2864 ff_rdft_init+340: xor%eax,%eax 0x002f2866 ff_rdft_init+342: jmp0x2f2793 ff_rdft_init+131 I have absolutely no idea what this means but something seems miscompiled? FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082
[issue2082] Crash in FFT-SSE code when playing mmst URI
Ronald S. Bultje rsbul...@gmail.com added the comment: A make clean fixed it. -- status: new - closed substatus: new - invalid FFmpeg issue tracker iss...@roundup.ffmpeg.org https://roundup.ffmpeg.org/issue2082