[PATCH] D46992: [AMDGPU] Add perf hints to functions

[Marek Olšák via Phabricator via llvm-commits]

mareko added a comment.

How can UMDs disable this optimization?

Are there cases where this decreases performance?



================
Comment at: lib/Target/AMDGPU/AMDGPUPerfHint.cpp:13
+/// functions and amdgpu-hint-wave-limiter on kernels which may benefit from
+/// limiting number of waves to reduce cache conflicts.
+///
----------------
Did you mean "cache thrashing"?


https://reviews.llvm.org/D46992