[PATCH] D76266: [mlir][Vector] Mostly-NFC - Restructure options for lowering to LLVM Matrix Intrinsics

[Mehdi AMINI via Phabricator via llvm-commits]

mehdi_amini added inline comments.


================
Comment at: mlir/lib/Dialect/VectorOps/VectorTransforms.cpp:955
+
+      // Avoid all LLVM scalarization issues.
+      // TODO(ntv): finer-grained selection, atm if any dim is 1 bail out.
----------------
Can you expand the comment to describe the problem with LLVM?
Also is it a fundamental problem is it just a temporary bug/limitation in LLVM that should be lifted there? (what is the tracking bug? Is there a test-case?)


================
Comment at: mlir/test/lib/Transforms/TestVectorTransforms.cpp:28
+    llvm::cl::desc("Lower vector.contract to llvm.intr.matrix.multiply"),
+    llvm::cl::init(false), llvm::cl::cat(clOptionsCategory));
+
----------------
Can you use pass options instead?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76266/new/

https://reviews.llvm.org/D76266