For this bug, whatever we do with a mir test, it is not going to be reliable in failing if the bug is present. Maybe there is a unit testing framework for LiveRangeCalc tests that I could add a test to.
Thu, Jun 13
Wed, Jun 12
V2: Lit test fix.
V2: Addressed review comments re test.
Tue, Jun 11
Thu, May 30
V2: Target-specific attribute instead of target feature, as suggested by Stas.
Wed, May 29
Tue, May 21
LGTM, but I don't think I know the legalization code well enough to approve this.
May 16 2019
Is someone now able to approve this? Eli?
May 14 2019
V4: update_mir_test_checks the test.
May 8 2019
Ping: Could someone approve this please? Thanks.
Hi Samuel. Sorry for the delay; I kind of lost track of this change.
Apr 17 2019
V3: Further reduced test case.
V2: Addressed review comments.
V2: Addressed review comment: moved tests.
Apr 15 2019
Bug for adding remaining test cases is https://bugs.llvm.org/show_bug.cgi?id=41500
V3: Addressed review comments.
PS I'm only speaking for LLPC. I don't know if Mesa uses it.
I think the only obstacle is getting round to doing the LLPC changes. So this is hopefully a short term fix until we can get around to it.
I have cut down the test a bit more and put it into fneg-combines.ll. I did not manage to repro any problems with the other cases that I added fixes for.
V2: Cut down test a bit more and put it in fneg-combines.ll.
Apr 13 2019
I did not add a test because I believe that we do not currently set the source modifiers in codegen, so this code never gets exercised.
Apr 12 2019
That's the best bugpoint could do with the test.
Apr 9 2019
Hopefully fixed by D60457.
Apr 1 2019
Mar 29 2019
LGTM modulo the wrong license on the new file.
Mar 27 2019
OK thanks Matt.
Mar 22 2019
Mar 21 2019
Mar 20 2019
Thanks for the better test Matt. But I'll abandon this one in favor of Michael's improved fix D59608.
Thanks Michael; that is much better than my fix in D59556. Let's go with this one and I'll abandon my one.
The test is already reduced as much as I can. Removing anything in there makes the problem disappear. Constructing a new test case using llvm.uadd.with.overflow does not show the problem. Can we go with this test case?
V4: Generate vendor=AMDGPU, type=32 for the metadata note record, as now expected by PAL.
Mar 19 2019
V3: Rebased, and addressed "single char in single quotes" review comment from the other change.
V3: Fixed review comments.