[Intel-gfx] [igt-dev] [PATCH i-g-t 1/2] i915/gem_exec_latency: Normalize results into ns

[Chris Wilson]

Quoting Antonio Argenziano (2019-01-29 17:55:45)
> 
> 
> On 29/01/19 01:55, Chris Wilson wrote:
> > Present the latency results in nanoseconds not RCS cycles.
> > 
> > Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> > ---
> >   tests/i915/gem_exec_latency.c | 38 +++++++++++++++++++++++++++++++----
> >   1 file changed, 34 insertions(+), 4 deletions(-)
> > 
> > diff --git a/tests/i915/gem_exec_latency.c b/tests/i915/gem_exec_latency.c
> > index de16322a6..ea44adc14 100644
> > --- a/tests/i915/gem_exec_latency.c
> > +++ b/tests/i915/gem_exec_latency.c
> > @@ -59,6 +59,7 @@
> >   #define PREEMPT 0x2
> >   
> >   static unsigned int ring_size;
> > +static double rcs_clock;
> >   
> >   static void
> >   poll_ring(int fd, unsigned ring, const char *name)
> > @@ -207,7 +208,7 @@ static void latency_on_ring(int fd,
> >               igt_cork_unplug(&c);
> >   
> >       gem_set_domain(fd, obj[1].handle, I915_GEM_DOMAIN_GTT, 0);
> > -     gpu_latency = (results[repeats-1] - results[0]) / (double)(repeats-1);
> > +     gpu_latency = (results[repeats-1] - results[1]) / (double)(repeats-2);
> 
> How come you don't like the value at 0? Maybe adding a comment would 
> make it clearer.

I was thinking of trying to reduce some context warmup latency, but
it doesn't matter and the spinner in the second patch is much more
effective overall.
-Chris