On Wed, Mar 25, 2015 at 04:53:38PM +0000, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
>
> New kernels add the ability to pad objects to specified size at execbuf time.
>
> Add the drm_intel_bo_pad_to_size API via which this padded size can be set.
Looks good, exactly what I had in mind.
> static int
> +drm_intel_gem_bo_pad_to_size(drm_intel_bo *bo, uint64_t pad_to_size)
> +{
> + drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
> +
> + bo_gem->pad_to_size = bo->size < pad_size ? pad_size : 0;
Probably best to return -EINVAL here rather than silently drop the
request. Knowing our callers, they will ignore the error anyway.
One thing I was considering inside the kernel was to allow the GTT
allocation to be smaller than the object. I don't have a good usecase
though, because with a subobject you are likely to want to use all of it
at some point, so you will likely find it all bound anyway. (And I don't
think the penalty for the occasional rebind is that onerous.)
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
