http://ffmpeg.arrozcru.org/autobuilds/blog/2010/06/30/new-build-of-libvpx-091/ Mon, 21 May 2012 16:24:41 +0000 http://wordpress.org/?v=2.7.1 hourly 1 http://ffmpeg.arrozcru.org/autobuilds/blog/2010/06/30/new-build-of-libvpx-091/comment-page-1/#comment-140 James Thu, 01 Jul 2010 22:46:27 +0000 http://ffmpeg.arrozcru.org/autobuilds/blog/?p=666#comment-140 Thanks for all your effort. I knew it was not easy!, that is really a weird bug and one hard to trace, as I was looking at the runtime and not at the compiler / linker. Thanks for all your effort. I knew it was not easy!, that is really a weird bug and one hard to trace, as I was looking at the runtime and not at the compiler / linker.

]]> http://ffmpeg.arrozcru.org/autobuilds/blog/2010/06/30/new-build-of-libvpx-091/comment-page-1/#comment-139 ramiro Thu, 01 Jul 2010 19:17:54 +0000 http://ffmpeg.arrozcru.org/autobuilds/blog/?p=666#comment-139 I checked what was the value in that %edi and where it had come from. then traced it back to a global array, checked the declaration of the array (it would have to be 16-byte aligned, and indeed in the code it was), compiled entropy.c again with -save-temps, looked at the assembly file generated, and noticed it had ".comm" in there (as in that array could be in a common section) and no alignment information. Then I remembered of a gcc < 4.5 bug on MinGW that .comm dropped alignment information, and -fno-common could be used to counter that. That flag is set unconditionally by FFmpeg for quite some time now. I checked what was the value in that %edi and where it had come from. then traced it back to a global array, checked the declaration of the array (it would have to be 16-byte aligned, and indeed in the code it was), compiled entropy.c again with -save-temps, looked at the assembly file generated, and noticed it had “.comm” in there (as in that array could be in a common section) and no alignment information. Then I remembered of a gcc < 4.5 bug on MinGW that .comm dropped alignment information, and -fno-common could be used to counter that. That flag is set unconditionally by FFmpeg for quite some time now.

]]> http://ffmpeg.arrozcru.org/autobuilds/blog/2010/06/30/new-build-of-libvpx-091/comment-page-1/#comment-138 James Thu, 01 Jul 2010 14:18:39 +0000 http://ffmpeg.arrozcru.org/autobuilds/blog/?p=666#comment-138 If I don't ask this I won't sleep! How did you figure out that -fno-common (I know what it does) could be the workaround for this bug. If I don’t ask this I won’t sleep! How did you figure out that -fno-common (I know what it does) could be the workaround for this bug.

]]>