[Mlir-commits] [llvm] [mlir] [mlir][X86] Fold VNNI operand unpacking into contractions (PR #232800)

[Adam Siemieniuk]

================
@@ -115,6 +113,12 @@ void populateVectorContractToAMXDotProductPatterns(RewritePatternSet &patterns);
 void populateMoveAccumulatorForContractLoopPatterns(
     RewritePatternSet &patterns);
 
+// Adds patterns that fold RHS unpacking into vector contractions inside
+// scf.for loops, so the contractions use VNNI-packed operands directly.
+// The patterns require constant innermost read indices; other indices may
+// be computed.
+void populatePrepareVectorContractLoopsPatterns(RewritePatternSet &patterns);
----------------
adam-smnk wrote:

I'd either generalize the comment or make the `populate*Patterns` and the source file name more specific

https://github.com/llvm/llvm-project/pull/232800