[Mlir-commits] [mlir] [mlir][amdgpu] implement amdgpu.sparse_mfma wrapper for smfmac instructions (PR #171968)

[Eric Feng]

================
@@ -661,6 +661,30 @@ static Value convertMFMAVectorOperand(ConversionPatternRewriter &rewriter,
   return input;
 }
 
+/// Converts sparse MFMA (smfmac) operands to the expected ROCDL types.
+static Value convertSparseMFMAVectorOperand(ConversionPatternRewriter &rewriter,
+                                            Location loc, Value input,
+                                            bool allowBf16 = true) {
+  Type inputType = input.getType();
+  if (auto vectorType = dyn_cast<VectorType>(inputType)) {
----------------
efric wrote:

just switched to `cast` since it's checked in the verifier anyway

https://github.com/llvm/llvm-project/pull/171968