[PATCH] D142585: AMDGPU: Try to push fneg as integer into select

[Pierre van Houtryve via Phabricator via llvm-commits]

Pierre-vh added a comment.

Do we have/want something similar in GISel?



================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:9983
+    if (CRHS && CRHS->getAPIntValue().isSignMask() &&
+        shouldFoldFNegIntoSrc(N, LHS)) {
+      SDLoc DL(N);
----------------
Where is this function defined? In a parent diff?

Also would the transform still be profitable if the SELECT has >1 user and won't be deleted after the combine runs?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D142585/new/

https://reviews.llvm.org/D142585