Re: [PATCH 08/10] drm/i915: Prelude to splitting i915_gem_do_execbuffer in two

[Dave Gordon <david.s.gordon@intel.com>]

On 10/12/14 10:58, Daniel Vetter wrote:
> On Tue, Dec 09, 2014 at 12:59:11PM +0000, John.C.Harrison@Intel.com wrote:
>> From: John Harrison <John.C.Harrison@Intel.com>
>>
>> The scheduler decouples the submission of batch buffers to the driver with their
>> submission to the hardware. This basically means splitting the execbuffer()
>> function in half. This change rearranges some code ready for the split to occur.
> 
> Now there's the curios question: Where will the split in top/bottom halves
> be? Without that I have no idea whether it makes sense and so can't review
> this patch. At least if the goal is to really prep for the scheduler and
> not just move a few lines around ;-)
> -Daniel
> 
[snip]
>>  
>> +	i915_gem_execbuffer_move_to_active(vmas, ring);
>> +
>> +	/* To be split into two functions here... */
>> +
>> +	/* Unconditionally invalidate gpu caches and ensure that we do flush
>> +	 * any residual writes from the previous batch.
>> +	 */
>> +	ret = logical_ring_invalidate_all_caches(ringbuf);

It'll be where the marker comment is above. Ahead of that point is stuff
to do with setting up software state; after that we're talking to h/w.
When the scheduler code goes it, it decouples the two by interposing at
this point. Then batches go into it with s/w state set up, but don't get
to talk to the h/w until they're selected for execution, possibly in a
different order.

.Dave.
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx