I am trying to compile the MXNet ResNet_18 model (from its model zoo) for the Intel integrated graphics, using:
target = tvm.target.intel_graphics()
relay.build_module.build(mod, target, params=params)
however, this results in:
File "/mnt/c/Users/<user>/WSL/tvm/src/target/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 tested with some random PyTorch network and achieved the same error. I am unsure, if this could be a cause of this problem, but I am using TVM inside Windows Subsystem for Linux (V1), which usually worked fine.
Is the Target still supported or have I just made a mistake?