Message ID | 20170605102619.4679-4-chris@chris-wilson.co.uk (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 05/06/2017 11:26, Chris Wilson wrote: > Originally we would enable and disable the breadcrumb interrupt > immediately on demand. This was slow enough to have a large impact > (>30%) on tasks that hopped between engines. However, by using a shadow > to keep the irq alive for an extra interrupt (see commit 67b807a89230 > ("drm/i915: Delay disabling the user interrupt for breadcrumbs")) and > by recently reducing the cost in adding ourselves to the signal tree, we > no longer need to spin-request during await_request to avoid delays in > throughput tests. Without the earlier patches to stop the wakeup when > signaling if the irq was already active, we saw no improvement in > execbuf overhead (and corresponding contention in other clients) despite > the removal of the spinner in a simple test like glxgears. This means > that will be scenarios where now we spend longer enabling the interrupt "There will be" I guess? > than we would have spent spinning, but these are not likely to have as > noticeable an impact as the high frequency test cases (where there > should not be any regression). > > Ulterior motive: generalising the engine->sync_to to handle different > types of semaphores and non-semaphores. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > Cc: Oscar Mateo <oscar.mateo@intel.com> > --- > drivers/gpu/drm/i915/i915_gem_request.c | 18 ++++++------------ > 1 file changed, 6 insertions(+), 12 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c > index 46d869e26b4d..8c59c79cbd8b 100644 > --- a/drivers/gpu/drm/i915/i915_gem_request.c > +++ b/drivers/gpu/drm/i915/i915_gem_request.c > @@ -683,7 +683,6 @@ static int > i915_gem_request_await_request(struct drm_i915_gem_request *to, > struct drm_i915_gem_request *from) > { > - u32 seqno; > int ret; > > GEM_BUG_ON(to == from); > @@ -707,18 +706,14 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to, > return ret < 0 ? ret : 0; > } > > - seqno = i915_gem_request_global_seqno(from); > - if (!seqno) > - goto await_dma_fence; > + if (to->engine->semaphore.sync_to) { > + u32 seqno; > > - if (!to->engine->semaphore.sync_to) { > - if (!__i915_gem_request_started(from, seqno)) > - goto await_dma_fence; > + GEM_BUG_ON(!from->engine->semaphore.signal); > > - if (!__i915_spin_request(from, seqno, TASK_INTERRUPTIBLE, 2)) > + seqno = i915_gem_request_global_seqno(from); > + if (!seqno) > goto await_dma_fence; > - } else { > - GEM_BUG_ON(!from->engine->semaphore.signal); > > if (seqno <= to->timeline->global_sync[from->engine->id]) > return 0; > @@ -729,10 +724,9 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to, > return ret; > > to->timeline->global_sync[from->engine->id] = seqno; > + return 0; > } > > - return 0; > - > await_dma_fence: > ret = i915_sw_fence_await_dma_fence(&to->submit, > &from->fence, 0, > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Regards, Tvrtko
On ma, 2017-06-05 at 11:26 +0100, Chris Wilson wrote: > Originally we would enable and disable the breadcrumb interrupt > immediately on demand. This was slow enough to have a large impact > (>30%) on tasks that hopped between engines. However, by using a shadow > to keep the irq alive for an extra interrupt (see commit 67b807a89230 > ("drm/i915: Delay disabling the user interrupt for breadcrumbs")) and > by recently reducing the cost in adding ourselves to the signal tree, we > no longer need to spin-request during await_request to avoid delays in > throughput tests. Without the earlier patches to stop the wakeup when > signaling if the irq was already active, we saw no improvement in > execbuf overhead (and corresponding contention in other clients) despite > the removal of the spinner in a simple test like glxgears. This means > that will be scenarios where now we spend longer enabling the interrupt ^ there ? "now where we" ? > than we would have spent spinning, but these are not likely to have as > noticeable an impact as the high frequency test cases (where there > should not be any regression). > > Ulterior motive: generalising the engine->sync_to to handle different > types of semaphores and non-semaphores. > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > Cc: Oscar Mateo <oscar.mateo@intel.com> Does what is described, so code itself is: Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Some Testcase:'s would be cool, without those it's bit handwavy. Maybe an Ack from Tvrtko. Regards, Joonas
Quoting Joonas Lahtinen (2017-06-08 11:02:40) > On ma, 2017-06-05 at 11:26 +0100, Chris Wilson wrote: > > Originally we would enable and disable the breadcrumb interrupt > > immediately on demand. This was slow enough to have a large impact > > (>30%) on tasks that hopped between engines. However, by using a shadow > > to keep the irq alive for an extra interrupt (see commit 67b807a89230 > > ("drm/i915: Delay disabling the user interrupt for breadcrumbs")) and > > by recently reducing the cost in adding ourselves to the signal tree, we > > no longer need to spin-request during await_request to avoid delays in > > throughput tests. Without the earlier patches to stop the wakeup when > > signaling if the irq was already active, we saw no improvement in > > execbuf overhead (and corresponding contention in other clients) despite > > the removal of the spinner in a simple test like glxgears. This means > > that will be scenarios where now we spend longer enabling the interrupt > > ^ there ? "now where we" ? > > > than we would have spent spinning, but these are not likely to have as > > noticeable an impact as the high frequency test cases (where there > > should not be any regression). > > > > Ulterior motive: generalising the engine->sync_to to handle different > > types of semaphores and non-semaphores. > > > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > > Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> > > Cc: Oscar Mateo <oscar.mateo@intel.com> > > Does what is described, so code itself is: > > Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> > > Some Testcase:'s would be cool, without those it's bit handwavy. Maybe > an Ack from Tvrtko. Testcase for what though? It doesn't fix or improve anything, it just moves the overhead of the breadcrumb from one column to another. And we are sorely lacking performance tests in igt. Correctness wise there are many that cover this await_request. -Chris
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c index 46d869e26b4d..8c59c79cbd8b 100644 --- a/drivers/gpu/drm/i915/i915_gem_request.c +++ b/drivers/gpu/drm/i915/i915_gem_request.c @@ -683,7 +683,6 @@ static int i915_gem_request_await_request(struct drm_i915_gem_request *to, struct drm_i915_gem_request *from) { - u32 seqno; int ret; GEM_BUG_ON(to == from); @@ -707,18 +706,14 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to, return ret < 0 ? ret : 0; } - seqno = i915_gem_request_global_seqno(from); - if (!seqno) - goto await_dma_fence; + if (to->engine->semaphore.sync_to) { + u32 seqno; - if (!to->engine->semaphore.sync_to) { - if (!__i915_gem_request_started(from, seqno)) - goto await_dma_fence; + GEM_BUG_ON(!from->engine->semaphore.signal); - if (!__i915_spin_request(from, seqno, TASK_INTERRUPTIBLE, 2)) + seqno = i915_gem_request_global_seqno(from); + if (!seqno) goto await_dma_fence; - } else { - GEM_BUG_ON(!from->engine->semaphore.signal); if (seqno <= to->timeline->global_sync[from->engine->id]) return 0; @@ -729,10 +724,9 @@ i915_gem_request_await_request(struct drm_i915_gem_request *to, return ret; to->timeline->global_sync[from->engine->id] = seqno; + return 0; } - return 0; - await_dma_fence: ret = i915_sw_fence_await_dma_fence(&to->submit, &from->fence, 0,
Originally we would enable and disable the breadcrumb interrupt immediately on demand. This was slow enough to have a large impact (>30%) on tasks that hopped between engines. However, by using a shadow to keep the irq alive for an extra interrupt (see commit 67b807a89230 ("drm/i915: Delay disabling the user interrupt for breadcrumbs")) and by recently reducing the cost in adding ourselves to the signal tree, we no longer need to spin-request during await_request to avoid delays in throughput tests. Without the earlier patches to stop the wakeup when signaling if the irq was already active, we saw no improvement in execbuf overhead (and corresponding contention in other clients) despite the removal of the spinner in a simple test like glxgears. This means that will be scenarios where now we spend longer enabling the interrupt than we would have spent spinning, but these are not likely to have as noticeable an impact as the high frequency test cases (where there should not be any regression). Ulterior motive: generalising the engine->sync_to to handle different types of semaphores and non-semaphores. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Cc: Oscar Mateo <oscar.mateo@intel.com> --- drivers/gpu/drm/i915/i915_gem_request.c | 18 ++++++------------ 1 file changed, 6 insertions(+), 12 deletions(-)