[Intel-gfx] [PATCH i-g-t] benchmarks/gem_wsim: Perturb static_vcs selection across clients
Tvrtko Ursulin
tvrtko.ursulin at linux.intel.com
Tue May 14 10:51:54 UTC 2019
On 14/05/2019 11:05, Chris Wilson wrote:
> Use the client id to alternate the static_vcs balancer (-b context)
> across clients - otherwise all clients end up on vcs0 and do not match
> the context balancing employed by media-driver.
>
> This may want to be behind the -R flag, but I felt it was a fundamental
> property of static context balancing that to keep it disabled by default
> causes unfair comparisons and poor workload scheduling, defeating the
> purpose of testing.
I see your reasoning but it also completely matches the design of other
balancers to keep it under control of -R switch. It can also already be
achieved with the -G switch. Which is perhaps a bit confusing.. Having
both would still make sense I think. (-G gives out engines rr to
contexts sequentially across all clients, -R start each client contexts
by rr.)
But I wouldn't enable it unconditionally. Because consider another
balancer like rr and a two same workload instances of a long context
followed by short second context batch. If suffers the same problem of
poor scheduling until -R is added.
So I think we want to have the two balancers compatible in behaviour in
this respect.
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> ---
> benchmarks/gem_wsim.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/benchmarks/gem_wsim.c b/benchmarks/gem_wsim.c
> index afb9644dd..8c7e30eb4 100644
> --- a/benchmarks/gem_wsim.c
> +++ b/benchmarks/gem_wsim.c
> @@ -939,7 +939,7 @@ alloc_step_batch(struct workload *wrk, struct w_step *w, unsigned int flags)
> static void
> prepare_workload(unsigned int id, struct workload *wrk, unsigned int flags)
> {
> - unsigned int ctx_vcs = 0;
> + unsigned int ctx_vcs = id & 1;
Therefore I think "ctx_vcs = (flags & INITVCSRR) ? id & 1 : 0" here.
> int max_ctx = -1;
> struct w_step *w;
> int i;
>
Regards,
Tvrtko
More information about the Intel-gfx
mailing list