[PATCH] D149773: [AMDGPU][IGLP] Add iglp_opt(1) strategy for single wave gemms

[Austin Kerbow via Phabricator via llvm-commits]

kerbowa added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp:85
 
-typedef DenseMap<SUnit *, SmallVector<int, 4>> SUnitsToCandidateSGsMap;
-
 typedef function_ref<bool(const SUnit *, const ArrayRef<SUnit *>,
                           const SIInstrInfo *, SmallVectorImpl<SchedGroup> &,
----------------
Put this in the class so that it can access the cache?


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp:88
+                          unsigned, SmallVector<SUnit *, 4> *)>
+    InstructionClassifier;
+
----------------
What if this was a member function in InstructionRule, that way rules could be reused. Or do you think that the rules will always be so specific that it is not needed and that common functionality should just be in some functions somewhere?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D149773/new/

https://reviews.llvm.org/D149773