Any clues what broke the ROCm build?

The following worked on February 11th:

cd /src/tvm/apps/benchmark python3 gpu_imagenet_bench.py --target=rocm --network=resnet-18

Since then something broke and I first receive a set of spurious messages:

/src/tvm/src/target/parsers/aprofile.cc:97: Warning: Cannot parse target features. LLVM was not compiled with support for Arm(R)-based targets.

Neither my LLVM instance nor the command line options I have passed in expect this to be an ARM system. I get a target triple of “arm64-apple-macos” which is rather strange since I am on Linux with an x86 host and amdgpu target.

I am in process of narrowing this, but also posting here to see where to look for what changed in last two months to break the ROCm build by turning it into an ARM target…

i had a fpllowup comment here [Target] Use LLVM target parser for determining Arm(R) A-Profile Architecture features by lhutton1 · Pull Request #16425 · apache/tvm · GitHub

Hi @mvermeulen, thanks for reporting this issue. I have created the PR [Target] Don't register AArch64 target tags without LLVM compiler support by lhutton1 · Pull Request #16897 · apache/tvm · GitHub which I hope fixes the issue. Would you be able to confirm?

I built and tested ROCm with PR 16897.

I no longer see the warning messages I reported above.

Thank you.