pytorch
Add sortIndex to fix ordering of roctracer GPU id tracks
Summary: Although hipGetDeviceProperties shows 8 devices enumerating from 0 to 7, when using roctracer_record_t, the record->device_id enumerates from 2 to 9. This is because roctracer considers id 0 and 1 as CPU sockets.
So we can expect up to 4 CPU sockets, and 8 GPUs, up to a total count of 12. Which would mean the enumeration will be 0 - 11. This will help put the GPU tracks in order and at the bottom of the chrome trace.
Original Bug report in ROCm/roctracer: https://github.com/ROCm/roctracer/issues/98. We can just show the device id that is given by roctracer.
Differential Revision: D57067065
Pulled By: aaronenyeshi
Test Plan:
Before:
After:
@aaronenyeshi merged this pull request in pytorch/kineto@8681ff11e1fa54da39023076c5c43eddd87b7a8a.