Re: [PATCH 5/6] drm/i915: Split batch pool into size buckets

From: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
To: Chris Wilson <chris@chris-wilson.co.uk>, intel-gfx@lists.freedesktop.org
Subject: Re: [PATCH 5/6] drm/i915: Split batch pool into size buckets
Date: Fri, 20 Mar 2015 09:25:31 +0000	[thread overview]
Message-ID: <550BE78B.9010706@linux.intel.com> (raw)
In-Reply-To: <20150319210937.GA10812@nuc-i3427.alporthouse.com>

On 03/19/2015 09:09 PM, Chris Wilson wrote:
> On Thu, Mar 19, 2015 at 05:35:35PM +0000, Tvrtko Ursulin wrote:
>>> -	list_for_each_entry_safe(tmp, next,
>>> -				 &pool->cache_list, batch_pool_list) {
>>> +	n = fls(size >> PAGE_SHIFT) - 1;
>>
>> Is this better in some way that size / PAGE_SIZE - 1? Or I went from
>> trusting compiler smartness to little to too much? :)
>
> The month has 31 days in it, we use >> PAGE_SHIFT, otherwise /
> PAGE_SIZE. Note the fls() puts these into power-of-two buckets not pure
> page-count buckets.

Yeah silly me.. was thinking about 1-2 page buckets form the commit 
message only. Perhaps add a comment about bucket sizes above n = fls 
line if you feel like it. Shouldn't harm.

Regards,

Tvrtko

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