search for: syncobj_wait

...lso test new fd (implicit default context) igt/gem_workarounds: Reduce manual list to combinatorial loops igt/gem_workarounds: Add hibernation coverage igt/gem_exec_scheduler: Add small priority sorting smoketest igt/gem_fence_thrash: Use streaming reads for verify igt/syncobj_wait: Replace open-coded calls to __syncobj_wait() igt/syncobj_wait: Close the sw_sync timeline after the test igt/drv_hangman: Convert from local recursive batch to igt_spin_t benchmark/gem_busy: Compare polling with syncobj_wait igt/gem_eio: Check hang/eio recovery during suspe...

...> > The test, roughly, does the following: > ?1. Allocates and binds 1000 BOs > ?2. Constructs a pushbuf that executes a no-op compute shader. > ?3. Does a single EXEC/wait combo to warm up the kernel > ?4. Loops 10,000 times, doing SYNCOBJ_RESET (fast), EXEC, and then > SYNCOBJ_WAIT and times the loop > > Of course, there's a bit of userspace driver overhead but that's > negledgable. > > If you drop the top patch which allocates 1k buffers, the submit time on > the old uAPI is 54 us/exec vs. 66 us/exec on the new UAPI. This includes > the time...

...ughly, does the following: > > 1. Allocates and binds 1000 BOs > > 2. Constructs a pushbuf that executes a no-op compute shader. > > 3. Does a single EXEC/wait combo to warm up the kernel > > 4. Loops 10,000 times, doing SYNCOBJ_RESET (fast), EXEC, and then > > SYNCOBJ_WAIT and times the loop > > > > Of course, there's a bit of userspace driver overhead but that's > > negledgable. > > > > If you drop the top patch which allocates 1k buffers, the submit time on > > the old uAPI is 54 us/exec vs. 66 us/exec on the new UAPI. Th...

...: > >? ?1. Allocates and binds 1000 BOs > >? ?2. Constructs a pushbuf that executes a no-op compute shader. > >? ?3. Does a single EXEC/wait combo to warm up the kernel > >? ?4. Loops 10,000 times, doing SYNCOBJ_RESET (fast), EXEC, and then > > SYNCOBJ_WAIT and times the loop > > > > Of course, there's a bit of userspace driver overhead but that's > > negledgable. > > > > If you drop the top patch which allocates 1k buffers, the submit > time on > > the old uAPI is 54 us/...