[PATCH] drm/i915: Prefer to pageflip on the active ring on IvyBridge

[PATCH] drm/i915: Prefer to pageflip on the active ring on IvyBridge

[Chris Wilson]

IvyBridge has the ability to pageflip from either the Render or Blitter
rings. Scheduling the pageflip on the active ring (if the new scanout is
being rendered on either pipeline) saves inserting a semaphore (if
available) and forcing a stall on both pipelines. We presume that the
current scanout is inactive.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/intel_display.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
index 984834a..ead2822 100644
--- a/drivers/gpu/drm/i915/intel_display.c
+++ b/drivers/gpu/drm/i915/intel_display.c
@@ -7908,9 +7908,13 @@ static int intel_gen7_queue_flip(struct drm_device *dev,
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-	struct intel_ring_buffer *ring = &dev_priv->ring[BCS];
+	struct intel_ring_buffer *ring;
 	int ret;
 
+	ring = obj->ring;
+	if (ring == NULL || ring == &dev_priv->ring[RCS])
+		ring = &dev_priv->ring[BCS];
+
 	ret = intel_pin_and_fence_fb_obj(dev, obj, ring);
 	if (ret)
 		goto out;
-- 
1.7.10

Re: [PATCH] drm/i915: Prefer to pageflip on the active ring on IvyBridge

[Chris Wilson]

On Mon, 16 Apr 2012 15:29:41 +0100, Chris Wilson <chris@chris-wilson.co.uk> wrote:
> +	ring = obj->ring;
> +	if (ring == NULL || ring == &dev_priv->ring[RCS])
should have been: ring == &dev_priv->ring[VCS]

This does require testing on IVB, since the BCS was choosen for a
reason, I'm sure.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre

Re: [PATCH] drm/i915: Prefer to pageflip on the active ring on IvyBridge

[Chris Wilson]

On Mon, 16 Apr 2012 15:29:41 +0100, Chris Wilson <chris@chris-wilson.co.uk> wrote:
> IvyBridge has the ability to pageflip from either the Render or Blitter
> rings. Scheduling the pageflip on the active ring (if the new scanout is
> being rendered on either pipeline) saves inserting a semaphore (if
> available) and forcing a stall on both pipelines. We presume that the
> current scanout is inactive.

As a follow-up to my own suggestion; don't even think of pageflipping from
the render ring, it is full of disappointment.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre