[MINOR] Type inference for conv2d() forward pass

This change adds type inference to the conv2d operation. It merely passes the input type of the matrix to the output type and sets the dimensions to the index/size type.

src/ir/daphneir/DaphneInferTypesOpInterface.cpp

@@ -544,6 +544,14 @@ std::vector<Type> daphne::RecodeOp::inferTypes() {
     return {resTy, dictTy};
 }
 
+std::vector<Type> daphne::Conv2DForwardOp::inferTypes() {
+    MLIRContext * ctx = getContext();
+    Type srcTy = getInput().getType().dyn_cast<daphne::MatrixType>();
+    Builder builder(ctx);
+
+    // output matrix of same type as input, height/width dimensions as size/index type
+    return {srcTy, builder.getIndexType(), builder.getIndexType()};
+}
 // ****************************************************************************
 // Type inference function
 // ****************************************************************************

src/ir/daphneir/DaphneOps.td

@@ -820,7 +820,8 @@ def Daphne_BiasAddForwardOp : Daphne_Op<"biasAddForward", [ CUDASupport ]> {
 // Convolution
 // ----------------------------------------------------------------------------
 
-def Daphne_Conv2DForwardOp : Daphne_Op<"Convolution_Forward", [ CUDASupport ]> {
+def Daphne_Conv2DForwardOp : Daphne_Op<"Convolution_Forward",
+    [ DeclareOpInterfaceMethods<InferTypesOpInterface>, CUDASupport ]> {
     let arguments = (ins
         MatrixOf<[FloatScalar]>:$input, MatrixOf<[FloatScalar]>:$filter, MatrixOf<[FloatScalar]>:$bias,
         // input shape