Quoting Mika Kuoppala (2019-08-08 16:26:35) > Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes: > > > To actually DROP_IDLE means that we have to wait for ongoing submission, > > and any new concurrently submitted, i.e. it should only be called during > > single-threaded submission to ensure the GPU is idle before the new > > action. > > > > v2: Also include SHRINK for thrash-all-the-things, and find a dupe in > > gem_reloc_vs_gpu. > > Agreed that is rather pointless trying to guarantee an idle gpu > during a test. > > Reviewed-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx> > > But between a tests, it will serve as a flush? At the end of a _test_ (binary), we do reset and idle the GPU. The theory is that we want each test to have a clean slate, and to catch any residual problems in the test that created them. But we don't do anything between subtests by default, and we frequently run into problems there if one subtest fails and causes the next to fail and so on and so on. An unsolved dilemma. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
