[Mlir-commits] [mlir] [mlir][x86] Decouple Accumulator from ADD-Based Vector Contractions (over loops) (PR #204327)

Adam Siemieniuk llvmlistbot at llvm.org
Tue Jul 28 02:34:46 PDT 2026


================
@@ -59,9 +63,9 @@ LogicalResult shuffleAfterReadLikeOp(PatternRewriter &rewriter, Operation *opA,
 
 // Shuffles vectors produced by vector.contraction ops into a flat layout
 // before they are written to memory.
-LogicalResult shuffleBeforeWriteLikeOp(PatternRewriter &rewriter,
-                                       Operation *opA, Operation *opB,
-                                       int64_t nonUnitDimAcc, VectorType accTy);
+LogicalResult shuffleBeforeWriteLikeOp(PatternRewriter &rewriter, Value opA,
+                                       Value opB, int64_t nonUnitDimAcc,
----------------
adam-smnk wrote:

I'd still suggest more expressive name to help understand which values of A and B are expected. I assume it's not an arbitrary operand?

https://github.com/llvm/llvm-project/pull/204327


More information about the Mlir-commits mailing list