[Intel-gfx] [PATCH 1/2] drm/i915: Re-order some checks to do the unlikely one first

[Chris Wilson]

On Tue, Feb 17, 2015 at 11:48:25AM +0000, Damien Lespiau wrote:
> On Mon, Feb 16, 2015 at 09:03:51PM +0000, Chris Wilson wrote:
> > On Mon, Feb 16, 2015 at 06:25:10PM +0000, Damien Lespiau wrote:
> > > instpm_mode != relative_constants_mode is quite unlikely to happen, so
> > > we can test it first to use C's && short-circuiting and not test on
> > > 'ring'.
> > > 
> > > I know, probably a useless micro-optimisation in the big scheme of
> > > things, but I'm going to add another test here, so might as well do it.
> > 
> > If you want to get pedantic, we want to move this to per-context :)
> 
> Do we? the API is per execbuf call, so theoretically application can
> change that during the context life time (it'd be silly, but they can).
> Or am I missing the point?

It was that this is a context specific register and more about handling
the transaction unwind if the execbuffer were to fail later i.e. an
issue about to arise with new features modifying the code. But in theory
we could do fewer LRI for that one application that used it...
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre