58 bytes, text/x-review-board-request
clang-cl complains: 11:33.48 c:/m-c/media/ffvpx/libavutil/x86/intmath.h(59,33): warning: implicit declaration of function '_tzcnt_u32' is invalid in C99 [-Wimplicit-function-declaration] 11:33.48 return ((uint32_t)v == 0) ? _tzcnt_u32((uint32_t)(v >> 32)) + 32 : _tzcnt_u32((uint32_t)v); 11:33.48 ^ which then leads to linking failures later: 11:34.10 mathematics.obj : error LNK2019: unresolved external symbol __tzcnt_u32 referenced in function _av_gcd 11:34.10 11:34.10 mozavutil.dll : fatal error LNK1120: 1 unresolved externals 11:34.10 as _tzcnt_u32 was expected to be inlined. The problem here is that clang-cl only makes intrinsics available based on the compilation flags you have selected, whereas MSVC makes all the intrinsics available all the time, cf https://llvm.org/bugs/show_bug.cgi?id=25544 , which is a WONTFIX on the clang-cl side. Chromium had to work around this too; from reading the bug report below, I think the problem was fixed upstream, and then they imported a new version of the source: http://crbug.com/556735 Can we upgrade our version of ffmpeg to include the relevant fix, or patch our version of ffmpeg?
You can clear MOZ_FFVPX in configure in the meanwhile.
We use ffmpeg 3.1 as of July 2016, which is a much later version than the one mentioned in the chromium bug
3 years ago
Priority: -- → P3
Ah, apparently ffmpeg broke things a while back: https://github.com/FFmpeg/FFmpeg/commit/f98417451291a3ff6719d739b5e904e0b7bf404b and chromium has not yet updated to a newer version. Do we have a mechanism for providing local patches, so we can make this work in clang-cl?
Well, we don't really. we've been trying to be as close to 100% identical to upstream. It would be best to have the past accepted upstream and then we resync. They are very quick at accepting new patches like this, and I can do the resync as soon you you ping me that it's done...
Do you have a patch handy for this? could submit it upstream for you.
(In reply to Jean-Yves Avenard [:jya] from comment #5) > Do you have a patch handy for this? > > could submit it upstream for you. I do, but clang-cl is also working on fixing it on their side, albeit requiring special compilation options. But the patch to fix it has been waiting more than a week with no movement on the review front...so I'm not sure what the best course of action here is.
Comment on attachment 8815868 [details] Bug 1298418 - use clang modules with ffvpx's libavutil and clang-cl; https://reviewboard.mozilla.org/r/96644/#review97266 Are you still planning to remove this if it gets fixed upstream? Can you get a followup on file if so?
Attachment #8815868 - Flags: review?(mshal) → review+
(In reply to Michael Shal [:mshal] from comment #8) > Are you still planning to remove this if it gets fixed upstream? Can you get > a followup on file if so? Yes. Bug 1321651 tracks.
Pushed by email@example.com: https://hg.mozilla.org/integration/autoland/rev/a63ed01b4c56 use clang modules with ffvpx's libavutil and clang-cl; r=mshal
You need to log in before you can comment on or make changes to this bug.