[PATCH] D72675: Fix -ffast-math/-ffp-contract interaction

[Sanjay Patel via Phabricator via cfe-commits]

spatel added a comment.

In D72675#1823227 <https://reviews.llvm.org/D72675#1823227>, @mcberg2017 wrote:

> We crossed that bridge internally at Apple a while ago, meaning I have some code debt for cleaning up open source for fma formation that uses contract and reassoc differently than we do today, both together and separately, case by case.


Ah, great. I don't have a preference about the order of patches (ie, is it better to make this change first or do the backend cleanup first), so whatever works better.

But it would be better to have all of the baseline tests in place, so we know where we stand currently. So I'd prefer to have the PPC and x86 tests pre-committed to master (change test comments as needed to match current/future reality).

We could also decouple the clang part of this patch from the backend change IIUC. But do we need another change (or at least tests) to show how the driver difference translates in the clang front-end to LLVM FMF and/or function attributes?



================
Comment at: clang/test/Driver/fast-math.c:183-184
 //
+// -ffp-contract=off must disable the fast-math umbrella, and the unsafe-fp-math
+// umbrella.
+// RUN: %clang -### -ffast-math -ffp-contract=off -c %s 2>&1 \
----------------
Do we need another pair of tests for -ffp-contract=on?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D72675/new/

https://reviews.llvm.org/D72675