[mlir][tosa] Use typeConverter->convertType<T> (#150578)

Since `resultTy` might be nullptr, we should use `dyn_cast` instead of
`cast`. Additionally, `typeConverter->convertType<T>` is more
appropriate in this context.

Author: CoTinker

mlir/lib/Conversion/TosaToLinalg/TosaToLinalgNamed.cpp

@@ -735,7 +735,7 @@ class MaxPool2dConverter : public OpConversionPattern<tosa::MaxPool2dOp> {
 
     bool isUnsigned = op.getType().getElementType().isUnsignedInteger();
     ShapedType resultTy =
-        cast<ShapedType>(getTypeConverter()->convertType(op.getType()));
+        getTypeConverter()->convertType<ShapedType>(op.getType());
     if (!resultTy)
       return rewriter.notifyMatchFailure(op, "failed to convert type");
     Type resultETy = inputTy.getElementType();

mlir/lib/Conversion/TosaToTensor/TosaToTensor.cpp

@@ -229,8 +229,8 @@ class ReshapeConverter : public OpConversionPattern<tosa::ReshapeOp> {
   matchAndRewrite(tosa::ReshapeOp reshape, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const final {
     auto loc = reshape.getLoc();
-    auto resultType = cast_if_present<ShapedType>(
-        getTypeConverter()->convertType(reshape.getType()));
+    auto resultType =
+        getTypeConverter()->convertType<ShapedType>(reshape.getType());
     if (!resultType) {
       return rewriter.notifyMatchFailure(reshape.getLoc(),
                                          "could not convert result type");