ffmpeg fix for clang

[revelator]

patches out inlining for clang on x86 in cabac.h allowing it to build.

0004-clang-disable-cabac-x86-inlining.zip

[1480c1]

I guess I never sent out the patch https://raw.githubusercontent.com/m-ab-s/mabs-patches/master/ffmpeg/0001-get_cabac_inline_x86-Don-t-inline-if-32-bit-clang-on.patch

@mstorsjo I noticed you had a patch a long time ago relating to clang+32-bit on windows https://patchwork.ffmpeg.org/project/ffmpeg/patch/20200523182917.11853-1-martin@martin.st/ any comment regarding that?

[mstorsjo]

I had a plan of getting back to that patch, but I never got to it. This patch probably is better than mine, as clang can build that assembly just fine, it just fails to fulfill all the assembly constraints when inlined into a larger context.

[revelator]

also seems to have been a problem on android from what i could gather from the bug reports, cabac seems to be rather picky in that regard. Found this patch while looking for fixes for windows clang builds though i believe it was originally intended for msvc clang it also worked for ours.

[1480c1]

https://trac.ffmpeg.org/ticket/8903 found the old ticket I made for this, it seems I attached the patch there but forgot to ever git send-email it

[1480c1]

https://patchwork.ffmpeg.org/project/ffmpeg/patch/20210817153539.188484-1-ccom@randomderp.com/

sent upstream

[1480c1]

@mstorsjo do you know what would be the best way to benchmark this function? I'm not sure if decoding any h264/h265 video will use this function

[mstorsjo]

Most H264 videos would exercise it (in practice, main/high profile content, unless the file was explicitly encoded with CAVLC enabled instead of CABAC), and afaik all HEVC videos. And higher bitrate content exercises it more than lower bitrate.

But I nerdsniped myself into benchmarking it already, see the mailing list.

[1480c1]

Thanks nerdsniper, interesting results btw

[mstorsjo]

I pushed this patch now, see FFmpeg/FFmpeg@8990c58.