Re: [PATCH v6] drm/i915: Add soft-pinning API for execbuffer

From: Chris Wilson <chris@chris-wilson.co.uk>
To: "Goel, Akash" <akash.goel@intel.com>
Cc: intel-gfx@lists.freedesktop.org,
	"Kristian Høgsberg" <hoegsberg@gmail.com>,
	"Vinay Belgaumkar" <vinay.belgaumkar@intel.com>
Subject: Re: [PATCH v6] drm/i915: Add soft-pinning API for execbuffer
Date: Fri, 16 Oct 2015 15:15:06 +0100	[thread overview]
Message-ID: <20151016141506.GC5238@nuc-i3427.alporthouse.com> (raw)
In-Reply-To: <56210510.1010006@intel.com>

On Fri, Oct 16, 2015 at 07:39:20PM +0530, Goel, Akash wrote:
> 
> Discussed sometime back about this patch with Chris. He mainly has 2
> concerns with it.
> 1. The linear walk used by the patch to detect the overlapping
> objects would be expensive.
> 2. Restriction to disallow !RCS submissions for non-default
> contexts, which could lead to lot of conflicts for the placements,
> once multiple engines like media/blit/vebox are used
> 
> Both of them can be addressed by the subsequent patches.
> Chris already has the patch ready to reduce the validation overhead
> with the use of rbtree and there should be no implications of
> allowing the non-default contexts for !RCS submissions in
> execbuffer.
> 
> In the standalone form, the patch looks good, so
> Reviewed-by: "Akash Goel <akash.goel@intel.com>"

I am sorry, but this patch does not reflect the final version that I
wrote, in particular does not provide the support I actually use inside
the kernel.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx