Skip to content

Conversation

@johanneslenfers
Copy link
Member

The mm example execution as part of the auto tuning tests is a heavy workload if running on a non-gpu device. Therefore, the execution is allowed to fail with an EXECUTION_ERROR. This PR adds the missing assert.

@Bastacyclop
Copy link
Member

Bastacyclop commented Feb 18, 2022

I think ideally tests should be able to discriminate between different execution errors. For example in this case I think the test should only be allowed to pass if the CL_OUT_OF_RESOURCES error is triggered?

The next step after that would be to query things like clGetDeviceInfo(device, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, ...), CL_DEVICE_MAX_WORK_ITEM_SIZES, CL_DEVICE_MAX_WORK_GROUP_SIZES, CL_DEVICE_LOCAL_MEM_SIZE, ...
And either check them against clGetKernelWorkGroupInfo data, or even better allow this info to drive autotuning (as was done in Lift I believe, at least to some extent).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants