About custom mlir dialect lowing?

MLIR
1

Hi, I am a mlir newbie, thanks.
I define a custom dialect and custom op like

def SumOp : Flow_Op<"sum"> {
    let summary = "sum operation";
    let description = [{ "flow.sum" }];
    let arguments = (ins F64Tensor:$in);
    let results = (outs F64:$out);

    let hasCustomAssemblyFormat = 1;
    let builders = [
        OpBuilder<(ins "Value":$in)>
     ];
    let assemblyFormat = "$in attr-dict `:` type($in) `to` type($out)";
}

and I want to lowing this op to affine dialect, but I do not how to write this lowering,

  struct SumOpLowering : public ConversionPattern {
    SumOpLowering(MLIRContext *context)
        : ConversionPattern(flow::SumOp::getOperationName(), 1, context) {}

    LogicalResult matchAndRewrite(Operation *op, ArrayRef<Value> operands,
                                  ConversionPatternRewriter &rewriter) const final {
      llvm::errs() << "sum op lowering\n";
      auto loc = op->getLoc();
      /// %0 = flow.constant dense<[1.0, 2.0, 3.0, 4.0]> : tensor<4xf64>
      /// %2 = flow.sum %0: tensor<4xf64> to f64

      auto input = operands[0];
      input.dump();
      auto inputType = input.getType().cast<mlir::ShapedType>();// memref<fxf64>
      inputType.dump();
      auto outputType = op->getResult(0).getType();// f64
      outputType.dump();

      auto shape = inputType.getShape();// [4]
      assert(shape.size() == 1 && "expected 1D tensor");
      auto rank = shape.size();

     ? 
      rewriter.eraseOp(op);
      return success();
    }
  };

where can I reference related code , and how can I implenment mlir knowledge, thanks

2

Have you taken a look at the toy tutorial? Chapter 5 may be a useful example here:

3

Something that would help is if you would write an example of the IR before as well as the expect results here.