Max pooling not working correctly

[Doralitze]

Description
Certain models, for example this one or this one, fail to execute 1x1 max pool kernels. This problem occurs both when running with spike pk as well as running directly on bare metal. Running these models with the default onnx runtime does not yield any errors. Furthermore it is strange, that such errors seam to only happen after a max pool kernel has already been run.

The exact error message:

Using Onnxruntime C++ API
MaxPool Output Edges 2
terminate called after throwing an instance of 'Ort::Exception'
 what(): Failed to find kernel for MaxPool_nhwc(1) (node 487_nhwc). Kernel not found

Urgency
No hard deadline but research is significantly slowed down by this issue.

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 22.04
• ONNX Runtime installed from (source or binary): source / chipyard gemmini installation scripts
• ONNX Runtime version: both the default submodule version as well as current main branch
• Python version: 3.10
• Visual Studio version (if applicable): does not apply
• GCC/Compiler version (if compiling from source): ubuntu package default / esp tools shipped with chipyard
• CUDA/cuDNN version: does not apply
• GPU model and memory: does not apply

To Reproduce

• Build gemmini as described in this document.
• Follow this guide to build ort_test and run the models with it. This should fail.

Expected behavior
I expect the output of labels and bounding boxes after spike finished the execution of the program.