Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-21 Thread Teres Alexis, Alan Previn
On Fri, 2022-09-16 at 08:36 -0700, Ceraolo Spurio, Daniele wrote: > > On 9/16/2022 1:58 AM, Tvrtko Ursulin wrote: > > On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: > > > On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: > > > > On 15/09/2022 03:12, Alan Previn wrote: > > > > > > >

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-21 Thread Teres Alexis, Alan Previn
On Fri, 2022-09-16 at 09:58 +0100, Tvrtko Ursulin wrote: > On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: > > On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: > > > On 15/09/2022 03:12, Alan Previn wrote: > > > > From: Matthew Brost > > > > > > > > +static void guc_flush_all_delay

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-21 Thread Teres Alexis, Alan Previn
On Fri, 2022-09-16 at 08:36 -0700, Ceraolo Spurio, Daniele wrote: > > alan: [snip] > > > > > +/* > > > > > + * If the context gets closed while the execbuf is ongoing, > > > > > the context > > > > > + * close code will race with the below code to cancel the > > > > > delayed work.

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-20 Thread Tvrtko Ursulin
On 16/09/2022 16:36, Ceraolo Spurio, Daniele wrote: On 9/16/2022 1:58 AM, Tvrtko Ursulin wrote: On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: [snip] +    /* + * If the context gets closed while the execbuf is ongoing, the context + * close code will race with the bel

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-19 Thread Teres Alexis, Alan Previn
On Fri, 2022-09-16 at 08:36 -0700, Ceraolo Spurio, Daniele wrote: > > On 9/16/2022 1:58 AM, Tvrtko Ursulin wrote: > > On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: > > > On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: > > > > On 15/09/2022 03:12, Alan Previn wrote: > > > > > From

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-16 Thread Ceraolo Spurio, Daniele
On 9/16/2022 1:58 AM, Tvrtko Ursulin wrote: On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: On 15/09/2022 03:12, Alan Previn wrote: From: Matthew Brost Add a delay, configurable via debugfs (default 34ms), to disable --- a/d

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-16 Thread Tvrtko Ursulin
On 16/09/2022 08:53, Teres Alexis, Alan Previn wrote: On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: On 15/09/2022 03:12, Alan Previn wrote: From: Matthew Brost Add a delay, configurable via debugfs (default 34ms), to disable --- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h +++ b/dri

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-16 Thread Teres Alexis, Alan Previn
On Thu, 2022-09-15 at 09:48 +0100, Tvrtko Ursulin wrote: > On 15/09/2022 03:12, Alan Previn wrote: > > From: Matthew Brost > > > > Add a delay, configurable via debugfs (default 34ms), to disable > > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h > > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h >

Re: [Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-15 Thread Tvrtko Ursulin
On 15/09/2022 03:12, Alan Previn wrote: From: Matthew Brost Add a delay, configurable via debugfs (default 34ms), to disable scheduling of a context after the pin count goes to zero. Disable scheduling is a costly operation as it requires synchronizing with the GuC. So the idea is that a dela

[Intel-gfx] [PATCH 1/1] drm/i915/guc: Delay disabling guc_id scheduling for better hysteresis

2022-09-14 Thread Alan Previn
From: Matthew Brost Add a delay, configurable via debugfs (default 34ms), to disable scheduling of a context after the pin count goes to zero. Disable scheduling is a costly operation as it requires synchronizing with the GuC. So the idea is that a delay allows the user to resubmit something befo