V2: Better fix that does not accidentally allow pow() transforms.

@foad pointed out that this fix is wrong. TLI saying pow() is not supported means if we find a call to a function called pow() then we don't know its semantics. So I will push a revised fix.

In D67680#1673252, @arsenm wrote:

LGTM. Can you also add a test for the mad case

V2: Disable all library functions, not just tan.

In D67003#1654339, @cwabbott wrote:

I just noticed that this already came up in D65813 and it does the right thing, it's just waiting review.

Possibly, but our majority use case is to not need the disassembly, and our thinking is that we don't want to change the tool flow between the majority case and the minority case. So can we get this fix in please?

I don't think anyone is a fan of dumpcode. But we're still in the position that the proper disassembler does not support gfx6 or gfx7, and we need to get this particular problem fixed in the short term.

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.

V2: Lit test fix.

V2: Addressed review comments re test.

V2: Target-specific attribute instead of target feature, as suggested by Stas.

LGTM, but I don't think I know the legalization code well enough to approve this.

Is someone now able to approve this? Eli?

V4: update_mir_test_checks the test.

Ping: Could someone approve this please? Thanks.

Hi Samuel. Sorry for the delay; I kind of lost track of this change.

V3: Further reduced test case.

V2: Addressed review comments.

V2: Addressed review comment: moved tests.

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.

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.

That's the best bugpoint could do with the test.

Hopefully fixed by D60457.

Hi Samuel

LGTM modulo the wrong license on the new file.

OK thanks Matt.