[LLVMdev] LICM promoting memory to scalar

[Robin Morisset]

I forgot to add that we already reported something like that to the gcc
developers who fixed it, but the fix is gated behind some obscure command
line option: "--param allow-store-data-race=0"​.
(The issue was found by an automated tool described in our previous paper
http://www.di.ens.fr/~zappa/readings/pldi13.pdf).
Could you retry compiling your code with that parameter ? If the
hoisting/sinking disappears then it is definitely unsound in the way I
described, and should not be done by LLVM.

Thanks,
Robin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20140902/f498f5f5/attachment.html>