[Intel-gfx] [PATCH] drm/i915: Emit to ringbuffer directly

[Chris Wilson]

On Wed, Feb 08, 2017 at 01:10:56PM +0000, Tvrtko Ursulin wrote:
> @@ -2288,32 +2264,33 @@ int intel_ring_begin(struct drm_i915_gem_request *req, int num_dwords)
>  		ring->space -= remain_actual;
>  	}
>  
> +	out = (u32 *)(ring->vaddr + ring->tail);

Checkpatch will complain about unaligned brackets, but I'll whine that
this cast is unnecessary.

> -static inline void intel_ring_advance(struct intel_ring *ring)
> +static inline void
> +intel_ring_advance(struct drm_i915_gem_request *req, u32 *out)
>  {
> +	struct intel_ring *ring = req->ring;
>  	/* Dummy function.
>  	 *
>  	 * This serves as a placeholder in the code so that the reader
> @@ -519,7 +510,7 @@ static inline void intel_ring_advance(struct intel_ring *ring)
>  	 * reserved for the command packet (i.e. the value passed to
>  	 * intel_ring_begin()).
>  	 */
> -	GEM_BUG_ONLY_ON(ring->tail != ring->advance);
> +	GEM_BUG_ON((ring->vaddr + ring->tail) != out);

This will generate a warning for an unused var if !GEM_DEBUG, just use
	GEM_BUG_ON(req->ring->vaddr + req->ring->tail != out);
it'll fit, the compiler will do CSE and it's only for debug...

The transformation looked good. I'd suggest we do a trybot with
i915.mmio_flip=0 to force the MI_DISPLAY_FLIP paths to be exercised and
aside from gen2/3 that should be all intel_ring_begin() covered.

In anticipation of those extra checks performed,
Reviewed-by: Chris Wilson <chris at chris-wilson.co.uk>

A couple of future steps, we can eliminate some dwords by inserting the
MI_NOOP to align the tail in add_request. I
s/intel_ring_begin/i915_gem_request_emit/ for the thin veneer of oop.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre