[Mlir-commits] [mlir] [mlir][ArmSME] Add rewrite to handle unsupported SVE transposes via SME/ZA (PR #98620)

[Cullen Rhodes]

================
@@ -374,6 +376,14 @@ struct LegalizeTransferWriteOpsByDecomposition
   }
 };
 
+auto makeVscaleConstantBuilder(PatternRewriter &rewriter, Location loc) {
+  Value vscale = rewriter.create<vector::VectorScaleOp>(loc);
+  return [loc, vscale, &rewriter](int64_t multiplier) {
+    return rewriter.create<arith::MulIOp>(
+        loc, vscale, rewriter.create<arith::ConstantIndexOp>(loc, multiplier));
+  };
+}
----------------
c-rhodes wrote:

this pattern is used often, move to utils?

https://github.com/llvm/llvm-project/pull/98620