Repro coming in a second
repro https://bugs.chromium.org/p/chromium/issues/detail?id=609976 (too big for here)
This should have been fixed in r268787; I'm still working on refining the assertion and writing a small enough testcase.
Looks like one of the bots recovered: http://lab.llvm.org:8080/green/job/llvm-stage2-cmake-RgLTO_build/7119/
No, this is still broken at r268792
I'll revert for now.
c:\src\llvm-rw>svn commit -m "Revert r268760, it caused PR27670." Sending lib\Target\X86\X86FixupBWInsts.cpp Sending test\CodeGen\X86\2011-06-14-PreschedRegalias.ll Sending test\CodeGen\X86\anyext.ll Sending test\CodeGen\X86\avx512-calling-conv.ll Sending test\CodeGen\X86\avx512-mask-op.ll Sending test\CodeGen\X86\avx512-select.ll Sending test\CodeGen\X86\avx512dq-mask-op.ll Sending test\CodeGen\X86\cmovcmov.ll Deleting test\CodeGen\X86\fixup-bw-copy.ll Sending test\CodeGen\X86\float-conv-elim.ll Sending test\CodeGen\X86\machine-combiner-int.ll Sending test\CodeGen\X86\opt-ext-uses.ll Sending test\CodeGen\X86\pr23664.ll Sending test\CodeGen\X86\vector-bitreverse.ll Sending test\CodeGen\X86\vector-idiv-sdiv-512.ll Sending test\CodeGen\X86\vector-idiv-udiv-512.ll Sending test\CodeGen\X86\xaluo.ll Transmitting file data ................ Committed revision 268794.
Interesting, thanks! I'll try your testcase.
Bots look green; resolving.