Add a regression test for the bug exposed by r158087, which has been
authorChandler Carruth <chandlerc@gmail.com>
Mon, 18 Jun 2012 09:15:04 +0000 (09:15 +0000)
committerChandler Carruth <chandlerc@gmail.com>
Mon, 18 Jun 2012 09:15:04 +0000 (09:15 +0000)
commit457dfbac8ac1e502f5ce2b86daa80b3c7db5bd08
tree2835e5586f76749c379fce51c80969cc450ea23a
parent43369249e7c2be91535e3ef900673e6116a953de
Add a regression test for the bug exposed by r158087, which has been
temporarily reverted.

This test is annoyingly overspecified, but I don't know of another way
to thoroughly test the saving and restoring of the registers. While this
will have to be adjusted even with the issue fixed in order to re-apply
r158087, those adjustments should very clearly indicate that it is still
correct (%esp getting restored prior to pops), whereas without it, this
case can easily slip under the radar.

Still, any suggestions for improvements are very welcome.

All credit to Matt Beaumont-Gay for reducing this out of an insane
Address Sanitizer crash to a reasonably small seg-faulting C program
when built with -mstackrealign. I just reduced it to IR, which was much
simpler. =]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@158656 91177308-0d34-0410-b5e6-96231b3b80d8
test/CodeGen/X86/force-align-stack-alloca.ll [new file with mode: 0644]