| Message ID | 20200429145113.588577-1-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [i-g-t] perf: Flush the work between rounds of gen8-unprivileged-single-ctx-counter | expand |
Oh... Is that on a low end platform? Looks alright : Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com> Thanks, -Lionel On 29/04/2020 17:51, Chris Wilson wrote: > Wait until the GPU is idle before starting a fresh round of probing > gen8-unprivileged-single-ctx-counter. This avoids building up a huge > backlog of render copies, hogging buffers and stale contexts, and > invoking the oomkiller. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> > --- > tests/perf.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/tests/perf.c b/tests/perf.c > index 74fc8fd87..2d23a02c6 100644 > --- a/tests/perf.c > +++ b/tests/perf.c > @@ -3787,6 +3787,7 @@ gen8_test_single_ctx_render_target_writes_a_counter(void) > drm_intel_gem_context_destroy(context1); > drm_intel_bufmgr_destroy(bufmgr); > __perf_close(stream_fd); > + gem_quiescent_gpu(drm_fd); > } > > child_ret = igt_wait_helper(&child);
Quoting Lionel Landwerlin (2020-04-29 15:57:41)
> Oh... Is that on a low end platform?
glk. But it was mostly that it got stuck in the rut of hitting -EAGAIN.
/* This needs to be investigated... From time
* to time, the work we kick off doesn't seem
* to happen. WTH?? */
Hmm. And that's before we release everything, so yes the leak is real
and not being slow.
-Chris
diff --git a/tests/perf.c b/tests/perf.c index 74fc8fd87..2d23a02c6 100644 --- a/tests/perf.c +++ b/tests/perf.c @@ -3787,6 +3787,7 @@ gen8_test_single_ctx_render_target_writes_a_counter(void) drm_intel_gem_context_destroy(context1); drm_intel_bufmgr_destroy(bufmgr); __perf_close(stream_fd); + gem_quiescent_gpu(drm_fd); } child_ret = igt_wait_helper(&child);
Wait until the GPU is idle before starting a fresh round of probing gen8-unprivileged-single-ctx-counter. This avoids building up a huge backlog of render copies, hogging buffers and stale contexts, and invoking the oomkiller. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com> --- tests/perf.c | 1 + 1 file changed, 1 insertion(+)