[Intel-gfx] [PATCH i-g-t] tests/i915/perf_pmu: Subtest to measure sampling error for 100% busy

Chris Wilson chris at chris-wilson.co.uk
Tue Feb 16 12:47:00 UTC 2021


Quoting Tvrtko Ursulin (2021-02-16 10:50:50)
> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> 
> Test that periodic reads of engine busyness against a constant 100% load
> are within the 5000ppm tolerance when comparing perf timestamp versus
> counter values.
> 
> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> ---
>  tests/i915/perf_pmu.c | 46 ++++++++++++++++++++++++++++++++++++++-----
>  1 file changed, 41 insertions(+), 5 deletions(-)
> 
> diff --git a/tests/i915/perf_pmu.c b/tests/i915/perf_pmu.c
> index 50b5c82bc472..728312be5293 100644
> --- a/tests/i915/perf_pmu.c
> +++ b/tests/i915/perf_pmu.c
> @@ -26,6 +26,7 @@
>  #include <stdio.h>
>  #include <string.h>
>  #include <fcntl.h>
> +#include <float.h>
>  #include <inttypes.h>
>  #include <errno.h>
>  #include <signal.h>
> @@ -46,6 +47,7 @@
>  #include "igt_perf.h"
>  #include "igt_sysfs.h"
>  #include "igt_pm.h"
> +#include "igt_stats.h"
>  #include "sw_sync.h"
>  
>  IGT_TEST_DESCRIPTION("Test the i915 pmu perf interface");
> @@ -278,8 +280,11 @@ static void end_spin(int fd, igt_spin_t *spin, unsigned int flags)
>  static void
>  single(int gem_fd, const struct intel_execution_engine2 *e, unsigned int flags)
>  {
> +       unsigned int loops = flags & FLAG_LONG ? 20 : 1;
> +       double err_min = DBL_MAX, err_max = -DBL_MAX;
>         unsigned long slept;
>         igt_spin_t *spin;
> +       igt_stats_t s;
>         uint64_t val;
>         int fd;

Something to record is that TEST_TRAILING_IDLE and TEST_LONG are mutually
exclusive.

So assert(igt_hweight(flags & (TEST_TRAILING_IDLE | TEST_LONG)) <= 1); ?
-Chris


More information about the Intel-gfx mailing list