[PATCH] D57961: [X86] Add explicit alignment to __m128/__m128i/__m128d/etc. to allow matching of MSVC behavior with #pragma pack.

[Reid Kleckner via Phabricator via cfe-commits]

rnk added a comment.
Herald added a subscriber: jdoerfert.

Hm, looks like this broke a bunch of code in chromium:
bad build: https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/ToTWin%28dbg%29/2342
good build: https://ci.chromium.org/p/chromium/builders/luci.chromium.ci/ToTWin%28dbg%29/2341

A trace from one of the crashing tests:

  [ RUN      ] AudioDecoderTestScenarios/AudioDecoderTest.DecodesFramesWithVaryingDuration/3
  Received fatal exception EXCEPTION_ACCESS_VIOLATION
  Backtrace:
  	xcorr_kernel_sse [0x00D4CEF7+199] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\x86\pitch_sse.c:54)
  	celt_pitch_xcorr_c [0x00DBC09C+156] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\pitch.c:264)
  	_celt_autocorr [0x00DBAF20+368] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\celt_lpc.c:266)
  	pitch_downsample [0x00DBBC98+536] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\pitch.c:182)
  	run_prefilter [0x00D4093F+703] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\celt_encoder.c:1144)
  	celt_encode_with_ec [0x00D3C915+3701] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\celt\celt_encoder.c:1606)
  	opus_encode_native [0x00CBBF8A+16810] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\src\opus_encoder.c:2068)
  	opus_encode [0x00CBE1C6+374] (C:\b\s\w\ir\cache\builder\src\third_party\opus\src\src\opus_encoder.c:2241)
  	media::cast::AudioDecoderTest::FeedMoreAudio [0x0017F0B8+1064] (C:\b\s\w\ir\cache\builder\src\media\cast\receiver\audio_decoder_unittest.cc:111)

The code in question is calling _mm_loadu_ps:
https://cs.chromium.org/chromium/src/third_party/opus/src/celt/x86/pitch_sse.c?type=cs&q=%22opus/src/celt/x86/pitch_sse.c%22&sq=package:chromium&g=0&l=54

  for (j = 0; j < len-3; j += 4)
  {
     __m128 x0 = _mm_loadu_ps(x+j);
     __m128 yj = _mm_loadu_ps(y+j);
     __m128 y3 = _mm_loadu_ps(y+j+3); // crashes



================
Comment at: lib/Headers/xmmintrin.h:1754
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_loadu_ps(const float *__p)
 {
----------------
I guess we just missed this.


Repository:
  rC Clang

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D57961/new/

https://reviews.llvm.org/D57961