MachineSink creates zombie defines

[llvmbot]

Bugzilla Link	10814
Resolution	FIXED
Resolved on	Oct 14, 2011 12:27
Version	trunk
OS	All
Attachments	Fix for the bug
Reporter	LLVM Bugzilla Contributor
CC	@asl,@efriedma-quic,@stoklund

Extended Description

After a critical edge split the flags of the successor were not reflected in the new block, which allowed later instructions to sink even though they write the EFLAGS.
The example below shows the problem. Unfortunately there is a test that relies on this bug, although the problem is that the EFLAGS are live-in when they should not be. The patch disables this test for now.

I attached a fix for the bug, however it was not accepted yet since it makes the test XFAIL, and I (still ) do not have time to fix it, but I don't want the issue ot be fogotten, so I created the bug report instead.

See:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20110725/125216.html

[stoklund]

Jan, I would like this patch to go in, it is an important fix.

The sinking is not essential to the uint64-to-float.ll test case. It is really about the instruction sequence selected. Please just rewrite the CHECKs to be independent of the code sinking instead of XFAILing the test.

Please don't add trailing whitespace.

Don't test livein_empty(). The for loop does that already.

[llvmbot]

Jan, I would like this patch to go in, it is an important fix.

The sinking is not essential to the uint64-to-float.ll test case. It is really
about the instruction sequence selected. Please just rewrite the CHECKs to be
independent of the code sinking instead of XFAILing the test.

Please don't add trailing whitespace.

Don't test livein_empty(). The for loop does that already.

Ok, I will try and get this done quickly.

[llvmbot]

Jan,

Has there been any further progress on this fix?

[llvmbot]

Jan,

Has there been any further progress on this fix?

No, I am totally swamped right now. I wanted to get it in before the 3.0 branch. It is really just the test case that needs to be fixed.

[llvmbot]

Jan, I would like this patch to go in, it is an important fix.

The sinking is not essential to the uint64-to-float.ll test case. It is really
about the instruction sequence selected. Please just rewrite the CHECKs to be
independent of the code sinking instead of XFAILing the test.

Please don't add trailing whitespace.

Don't test livein_empty(). The for loop does that already.

I posted an updated patch on llvm-commits.

[stoklund]

Jan's patch applied in r141960.