[Mlir-commits] [mlir] [mlir][AMDGPU] Remove an old bf16 workaround (PR #108409)

[Jakub Kuderski]

================
@@ -671,18 +671,25 @@ struct WMMAOpLowering : public ConvertOpToLLVMPattern<WMMAOp> {
   matchAndRewrite(WMMAOp op, WMMAOpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override {
     Location loc = op.getLoc();
-    Type outType = typeConverter->convertType(op.getDestD().getType());
+    auto outType =
+        cast<VectorType>(typeConverter->convertType(op.getDestD().getType()));
----------------
kuhar wrote:

```suggestion
        typeConverter->convertType<VectorType>(op.getDestD().getType());
```

and check if the conversion succeeded (non-null). Even if it passes now, it could start failing with a different type converter.

https://github.com/llvm/llvm-project/pull/108409