Intel warp shuffle dose not support width != warp_size , while compiling the MXNet ResNet_18 model

Question

while compiling the MXNet ResNet_18 model for the Intel integrated graphics, using the below command:

 target = tvm.target.intel_graphics() 
 relay.build_module.build(mod, target, params=params)

getting error:

File "./source/intrin_rule_opencl.cc", line 80
TVMError: 
---------------------------------------------------------------
An internal invariant was violated during the execution of TVM.
Please read TVM's error reporting guidelines.
More details can be found here: https://discuss.tvm.ai/t/error-reporting/7793.
---------------------------------------------------------------
  Check failed: (analyzer.CanProve(call->args[3] == call->args[4])) is false: Intel warp shuffle dose not support width != warp_size

I am using MXNet ResNet_18 model from its model zoo.

Answer 1

The issue seems to be related to TVM, not to MXNet. Also, please note that MXNet does not support “Intel integrated graphics” . Please open an issue on https://github.com/apache/tvm/issues or maybe in https://github.com/dmlc/gluon-cv/issues