On 25/09/2018 10:00, Chris Wilson wrote:
Quoting Tvrtko Ursulin (2018-09-24 12:29:45)
On 19/09/2018 20:55, Chris Wilson wrote:
+#define I915_PRIORITY_WAIT ((u8)BIT(1))
#define I915_PRIORITY_NEWCLIENT ((u8)BIT(0))
Put a comment here explaining the priority order is reversed in the
internal range.
With new client protection against being repeatedly pre-empted added in
a respective previous patch, I am okay that we give this a go.
Hmm, actually it is not reversed. So you would prefer new clients to
have a small priority boost over the stalling clients (which are in
effect under control of the user). Ok.
I thought it was reversed:
+ /* buckets sorted from highest [in slot 0] to lowest priority */
+ idx = I915_PRIORITY_MASK - (prio & I915_PRIORITY_MASK);
+ prio >>= I915_USER_PRIORITY_SHIFT;
But my two comments are not related - I just wanted a comment next to
internal level definitions in case they are reversed. If they are not
reversed it is fine without a comment. I guess the comment only applies
to bucket organization, not the priority levels, that was my confusion..
It made sense to me that new clients would be more important than
something stuck on a potentially long wait.. but I don't know, it's all
guesswork.
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
