AMDGPU: Assume f32 denormals are enabled by default
This will likely introduce catastrophic performance regressions on
older subtargets, but should be correct. A follow up change will
remove the old fp32-denormals subtarget features, and switch to using
the new denormal-fp-math/denormal-fp-math-f32 attributes. Frontends
should be making sure to add the denormal-fp-math-f32 attribute when
appropriate to avoid performance regressions.
diff --git a/llvm/docs/ReleaseNotes.rst b/llvm/docs/ReleaseNotes.rst
index 4f6e759..f23756b 100644
--- a/llvm/docs/ReleaseNotes.rst
+++ b/llvm/docs/ReleaseNotes.rst
@@ -100,6 +100,11 @@
Changes to the AMDGPU Target
-----------------------------
+* The backend default denormal handling mode has been switched to on
+ for all targets for all compute function types. Frontends wishing to
+ retain the old behavior should explicitly request f32 denormal
+ flushing.
+
Changes to the AVR Target
-----------------------------