[Intel-gfx] [PATCH i-g-t 2/3 v3] Unify handling of slow/combinatorial tests

Chris Wilson chris at chris-wilson.co.uk
Fri Oct 30 06:52:48 PDT 2015

On Fri, Oct 30, 2015 at 03:18:30PM +0200, David Weinehall wrote:
> @@ -931,16 +930,20 @@ run_basic_modes(const struct access_mode *mode,
>  	struct buffers buffers;
>  	for (h = hangs; h->suffix; h++) {
> -		if (!all && *h->suffix)
> -			continue;
> +		unsigned int subtest_flags;
> -		for (p = all ? pipelines : pskip; p->prefix; p++) {
> +		if (*h->suffix)
> +			subtest_flags = SUBTEST_TYPE_SLOW;

They aren't all slow though. The hang tests are (because it takes a long
time for a hang to occur and we need to race many times for reasonable
coverage). Many of the tests here were being skipped because QA couldn't
handle the full set.

Now that you have flags, adding h->flags would be better than heuristic
based on h->suffix. Similary we can use p->flags, then we get
igt_subtest_flags_f(h->flags | p->flags, "foo"); And if we have more
faith in QA in being able to dtrt we really only need to mark the known
slow cases (many the hang injection) as being truly slow.

Chris Wilson, Intel Open Source Technology Centre

More information about the Intel-gfx mailing list