Re: Testing HIP with the amdgpu driver

[Cordell Bloor <cgmb-deb@slerp.xyz>]

Hello,

I'm seeing much better results running the HIP tests with Linux 5.18. 
Running the tests with -j32 does still result in many test failures, but 
it no longer results in any graphics problems. Adding --no-parallel to 
the dh_auto_test results in 99% tests passing with the current HIP 
package on a Debian Sid userland and kernel.

Kernel:
Linux scorbunny 5.18.0-1-amd64 #1 SMP PREEMPT_DYNAMIC Debian 5.18.2-1 
(2022-06-06) x86_64 GNU/Linux

Results:
99% tests passed, 4 tests failed out of 408

Total Test time (real) = 3666.45 sec

The following tests FAILED:
         99 - directed_tests/hiprtc/hiprtcGetLoweredName.tst (SEGFAULT)
        100 - directed_tests/hiprtc/saxpy.tst (SEGFAULT)
        102 - directed_tests/ipc/hipMultiProcIpcMem.tst (Timeout)
        197 - directed_tests/runtimeApi/memory/hipIpcMemAccessTest.tst 
(Timeout)

The CTest timeout is set to 1500 sec, so that's really 6 minutes for all 
the passing tests and 3000 secs for hipMultiProcIpcMem and 
hipIpcMemAccessTest. Those tests should probably just be suppressed. 
It's possible those are real failures, but I don't think they're 
critical. The HIP repo Jenkinsfile [1] has included all but 
hipMultiProcIpcMem in its exclude-list at one point or other in the past 
few versions.

Sincerely,
Cory Bloor

[1]: 
https://github.com/ROCm-Developer-Tools/HIP/blob/develop/.jenkins/Jenkinsfile#L61