[PATCH] D84779: [AMDGPU] Add amdgpu specific loop threshold metadata

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp:123
+  if (MDNode *AmdgpuThreshold =
+          findOptionMDForLoop(L, "amdgpu.loop.unroll.threshold")) {
+    if (AmdgpuThreshold->getNumOperands() == 2) {
----------------
I have 2 concerns about this. First there's nothing AMDGPU specific about this. Second, the unroll threshold interpretation isn't really a stable concept. What happens if we adjust the cost model and now all the programs using this metadata change?

Why is directly setting the threshold this way useful compared to setting a target unroll factor?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84779/new/

https://reviews.llvm.org/D84779