Soren Lassen

@Connor-XY I had a look at this one, ran ``` Debug/bin/onnx-mlir --EmitONNXIR demo.onnx --mlir-print-ir-before-all ``` and saw ``` // -----// IR Dump Before (anonymous namespace)::ConstPropONNXToONNXPass //----- // func.func @main_graph() ->...

> If `ends` is not a constant we cannot come to the code in ConstProp.cpp yes but I'm guessing that it wasn't a constant yet during the preceding shape inference...

> I changed from tensor to tensor, but still could observe the error: > [demo.onnx.zip](https://github.com/onnx/onnx-mlir/files/9428622/demo.onnx.zip) @Connor-XY after you changed the tensor shape I suggest you try the hack to run...

Actually, please ignore my inferShapes hack. It doesn't make sense. ONNXSliceOpShapeHelper is called later in the method and, if anything, we could get the static output shape (no unknown dim...

> @chentong319 is the person on our end that did a lot of the sequences. I really would like him to review it, he is currently away for another 2...

@chentong319 thank you for the review. Please merge it for me.

> @sorenlassen Please resolve the conflict. I will try to merge this PR first once it passes the test. @chentong319: done

> To keep the code simple, I prefer not to change the index for the 'then' and 'else' subgraph. To me, the index is anyway implementation dependent and the string...

@Connor-XY ^

with the recent removal compiler warnings since the upgrade to c++17 will it work to treat compiler warnings as errors from now on?