[CI] Set USE_CMSISNN and USE_ETHOSU off in task_config_build_cpu.sh

[ekalda]

The dependencies for these have moved into ci_cortexm Docker
image, so there is not much point in building them for ci_cpu as we
can't run the associated tests.

cc @Mousius @areusch @driazati @gigiblender

[ekalda]

cc @leandron

[leandron]

Yes, the flags here are not tested in a context of ci_cpu, and are covered by the https://github.com/apache/tvm/blob/main/tests/scripts/task_config_build_cortexm.sh setup, where they are properly tested.

LGTM

[areusch]

thanks @ekalda ! looks like we are still running one of those tests in CPU somehow, maybe need to make sure that one got moved to arm and then remove it from CPU.

[ekalda]

Hi @areusch , sorry I've had very busy last few days, but I'm on it now - I looked into this and the problem is that we still install Vela in ci_cpu (https://github.com/apache/tvm/blob/main/docker/Dockerfile.ci_cpu#L129) despite moving Ethos-U into ci_cortexm and we use the presence of Vela as a proxy when deciding whether to run a test or not if it is outside of test_ethosu folder. As a result of this, there are some Ethos-U based tests in tvmc testing which are failing because pytest.importorskip("ethosu.vela") can't do it's job (https://github.com/apache/tvm/blob/main/tests/python/driver/tvmc/test_compiler.py#L456) .

So I think I'll park this change for a bit and propose a docker image change and once that is done, we should be able to merge this twoliner patch :)

[ekalda]

@tvm-bot rerun

[ekalda]

@driazati @areusch @leandron it's passing the CI now after the upgrade to the Docker image, so should be fine to merge.