| Message ID | 20170920143705.11277-9-mika.kuoppala@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Quoting Mika Kuoppala (2017-09-20 15:37:05) > -static void i915_guc_submit(struct intel_engine_cs *engine) > +static void i915_guc_submit(struct intel_engine_cs *engine, > + const unsigned int first) > { > struct drm_i915_private *dev_priv = engine->i915; > struct intel_guc *guc = &dev_priv->guc; > @@ -498,7 +500,7 @@ static void i915_guc_submit(struct intel_engine_cs *engine) > const unsigned int engine_id = engine->id; > unsigned int n; > > - for (n = 0; n < execlist_active_ports(el); n++) { > + for (n = first; n < execlist_active_ports(el); n++) { > struct execlist_port *port; > struct drm_i915_gem_request *rq; > unsigned int count; > @@ -506,21 +508,22 @@ static void i915_guc_submit(struct intel_engine_cs *engine) > port = execlist_port_index(el, n); > > rq = port_unpack(port, &count); > - if (rq && count == 0) { > - port_set(port, port_pack(rq, ++count)); > + GEM_BUG_ON(!rq); > + GEM_BUG_ON(count); > > - if (i915_vma_is_map_and_fenceable(rq->ring->vma)) > - POSTING_READ_FW(GUC_STATUS); > + port_set(port, port_pack(rq, ++count)); Ok, with this method we don't need count anymore. Seems sensible. -Chris
On Wed, Sep 20, 2017 at 05:37:05PM +0300, Mika Kuoppala wrote: > Now that we can keep track of what ports we have > dequeued, coalesce only those ports instead of iterating > through all ports. s/coalesce/submit. By coalescing I meant that we're no longer have a 1:1 relationship between a request and GuC workitem. But we're doing that in guc_dequeue by keeping the request-to-be-turned-into-workitem in port. > > Cc: Michał Winiarski <michal.winiarski@intel.com> > Cc: Chris Wilson <chris@chris-wilson.co.uk> > Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com> > --- > drivers/gpu/drm/i915/i915_guc_submission.c | 31 +++++++++++++++++------------- > drivers/gpu/drm/i915/intel_ringbuffer.h | 9 +++++++++ > 2 files changed, 27 insertions(+), 13 deletions(-) > > diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c > index 359f57a59cba..1057a0fb9f27 100644 > --- a/drivers/gpu/drm/i915/i915_guc_submission.c > +++ b/drivers/gpu/drm/i915/i915_guc_submission.c > @@ -485,11 +485,13 @@ static void guc_ring_doorbell(struct i915_guc_client *client) > /** > * i915_guc_submit() - Submit commands through GuC > * @engine: engine associated with the commands > + * @first: index of first execlist port to start coalescing from s/coalescing/submitting Reviewed-by: Michał Winiarski <michal.winiarski@intel.com> -Michał > * > * The only error here arises if the doorbell hardware isn't functioning > * as expected, which really shouldn't happen. > */ > -static void i915_guc_submit(struct intel_engine_cs *engine) > +static void i915_guc_submit(struct intel_engine_cs *engine, > + const unsigned int first) > { > struct drm_i915_private *dev_priv = engine->i915; > struct intel_guc *guc = &dev_priv->guc; > @@ -498,7 +500,7 @@ static void i915_guc_submit(struct intel_engine_cs *engine) > const unsigned int engine_id = engine->id; > unsigned int n; > > - for (n = 0; n < execlist_active_ports(el); n++) { > + for (n = first; n < execlist_active_ports(el); n++) { > struct execlist_port *port; > struct drm_i915_gem_request *rq; > unsigned int count; > @@ -506,21 +508,22 @@ static void i915_guc_submit(struct intel_engine_cs *engine) > port = execlist_port_index(el, n); > > rq = port_unpack(port, &count); > - if (rq && count == 0) { > - port_set(port, port_pack(rq, ++count)); > + GEM_BUG_ON(!rq); > + GEM_BUG_ON(count); > > - if (i915_vma_is_map_and_fenceable(rq->ring->vma)) > - POSTING_READ_FW(GUC_STATUS); > + port_set(port, port_pack(rq, ++count)); > > - spin_lock(&client->wq_lock); > + if (i915_vma_is_map_and_fenceable(rq->ring->vma)) > + POSTING_READ_FW(GUC_STATUS); > > - guc_wq_item_append(client, rq); > - guc_ring_doorbell(client); > + spin_lock(&client->wq_lock); > > - client->submissions[engine_id] += 1; > + guc_wq_item_append(client, rq); > + guc_ring_doorbell(client); > > - spin_unlock(&client->wq_lock); > - } > + client->submissions[engine_id] += 1; > + > + spin_unlock(&client->wq_lock); > } > } > > @@ -566,6 +569,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) > struct drm_i915_gem_request *last = NULL; > bool submit = false; > struct rb_node *rb; > + unsigned int first_idx; > > spin_lock_irq(&engine->timeline->lock); > rb = el->first; > @@ -575,6 +579,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) > goto done; > > port = execlist_request_port(el); > + first_idx = execlist_get_port_index(el, port); > > do { > struct i915_priolist *p = rb_entry(rb, typeof(*p), node); > @@ -614,7 +619,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) > el->first = rb; > if (submit) { > port_assign(port, last); > - i915_guc_submit(engine); > + i915_guc_submit(engine, first_idx); > } > spin_unlock_irq(&engine->timeline->lock); > } > diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h > index efa5a8ea1ecb..f2eb32539300 100644 > --- a/drivers/gpu/drm/i915/intel_ringbuffer.h > +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h > @@ -556,6 +556,15 @@ execlist_port_index(struct intel_engine_execlist * const el, > return &el->port[__port_idx(el->port_head, n, el->port_mask)]; > } > > +static inline unsigned int > +execlist_get_port_index(const struct intel_engine_execlist * const el, > + const struct execlist_port * const port) > +{ > + const unsigned int n = port_index(port, el); > + > + return __port_idx(n, -el->port_head, el->port_mask); > +} > + > static inline struct execlist_port * > execlist_port_head(struct intel_engine_execlist * const el) > { > -- > 2.11.0 >
diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c index 359f57a59cba..1057a0fb9f27 100644 --- a/drivers/gpu/drm/i915/i915_guc_submission.c +++ b/drivers/gpu/drm/i915/i915_guc_submission.c @@ -485,11 +485,13 @@ static void guc_ring_doorbell(struct i915_guc_client *client) /** * i915_guc_submit() - Submit commands through GuC * @engine: engine associated with the commands + * @first: index of first execlist port to start coalescing from * * The only error here arises if the doorbell hardware isn't functioning * as expected, which really shouldn't happen. */ -static void i915_guc_submit(struct intel_engine_cs *engine) +static void i915_guc_submit(struct intel_engine_cs *engine, + const unsigned int first) { struct drm_i915_private *dev_priv = engine->i915; struct intel_guc *guc = &dev_priv->guc; @@ -498,7 +500,7 @@ static void i915_guc_submit(struct intel_engine_cs *engine) const unsigned int engine_id = engine->id; unsigned int n; - for (n = 0; n < execlist_active_ports(el); n++) { + for (n = first; n < execlist_active_ports(el); n++) { struct execlist_port *port; struct drm_i915_gem_request *rq; unsigned int count; @@ -506,21 +508,22 @@ static void i915_guc_submit(struct intel_engine_cs *engine) port = execlist_port_index(el, n); rq = port_unpack(port, &count); - if (rq && count == 0) { - port_set(port, port_pack(rq, ++count)); + GEM_BUG_ON(!rq); + GEM_BUG_ON(count); - if (i915_vma_is_map_and_fenceable(rq->ring->vma)) - POSTING_READ_FW(GUC_STATUS); + port_set(port, port_pack(rq, ++count)); - spin_lock(&client->wq_lock); + if (i915_vma_is_map_and_fenceable(rq->ring->vma)) + POSTING_READ_FW(GUC_STATUS); - guc_wq_item_append(client, rq); - guc_ring_doorbell(client); + spin_lock(&client->wq_lock); - client->submissions[engine_id] += 1; + guc_wq_item_append(client, rq); + guc_ring_doorbell(client); - spin_unlock(&client->wq_lock); - } + client->submissions[engine_id] += 1; + + spin_unlock(&client->wq_lock); } } @@ -566,6 +569,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) struct drm_i915_gem_request *last = NULL; bool submit = false; struct rb_node *rb; + unsigned int first_idx; spin_lock_irq(&engine->timeline->lock); rb = el->first; @@ -575,6 +579,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) goto done; port = execlist_request_port(el); + first_idx = execlist_get_port_index(el, port); do { struct i915_priolist *p = rb_entry(rb, typeof(*p), node); @@ -614,7 +619,7 @@ static void i915_guc_dequeue(struct intel_engine_cs *engine) el->first = rb; if (submit) { port_assign(port, last); - i915_guc_submit(engine); + i915_guc_submit(engine, first_idx); } spin_unlock_irq(&engine->timeline->lock); } diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h index efa5a8ea1ecb..f2eb32539300 100644 --- a/drivers/gpu/drm/i915/intel_ringbuffer.h +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h @@ -556,6 +556,15 @@ execlist_port_index(struct intel_engine_execlist * const el, return &el->port[__port_idx(el->port_head, n, el->port_mask)]; } +static inline unsigned int +execlist_get_port_index(const struct intel_engine_execlist * const el, + const struct execlist_port * const port) +{ + const unsigned int n = port_index(port, el); + + return __port_idx(n, -el->port_head, el->port_mask); +} + static inline struct execlist_port * execlist_port_head(struct intel_engine_execlist * const el) {
Now that we can keep track of what ports we have dequeued, coalesce only those ports instead of iterating through all ports. Cc: Michał Winiarski <michal.winiarski@intel.com> Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com> --- drivers/gpu/drm/i915/i915_guc_submission.c | 31 +++++++++++++++++------------- drivers/gpu/drm/i915/intel_ringbuffer.h | 9 +++++++++ 2 files changed, 27 insertions(+), 13 deletions(-)