[Intel-gfx] [PATCH] drm/i915: push commit_output_state past the crtc/encoder preparing

[Jesse Barnes]

On Wed, 5 Sep 2012 21:48:52 +0200
Daniel Vetter <daniel.vetter at ffwll.ch> wrote:

> On Wed, Sep 5, 2012 at 8:28 PM, Jesse Barnes <jbarnes at virtuousgeek.org> wrote:
> >
> > The variables have me confused a little... I would have expected
> > update_state to take modeset_pipes rather than prepare_pipes.  Could
> > you use either?  Or will that not catch cases where we updated a pipe
> > that was already on?
> 
> The abstract idea for these masks was the following: Any pipe that
> changes anything goes into prepare_pipes. For any pipe that also
> changes the mode, it goes in addition into the modeset_pipes mask, so
> the later is a subset of prepare pipes. The idea here was to avoid the
> modeset step where not necessary (e.g. when disabling the 2nd output
> of a cloned crtc we only need to disable/enable, not change anything
> with the mode or clocks). But after some in-depth discussion with
> Paulo Zanoni I think we'll move large parts of the mode_set step into
> the enable function (at least for hsw due to funky ordering
> requirements), so I think this disdinction doesn't make sense.
> 
> The disable mask just contains those pipes that get fully disable (and
> which then also get removed from the prepares/modset masks).
> 
> Hence I pass the prepares mask into update_states, not just the modeset mask.

Ok, that makes some sense.  Hopefully we can preserve the full mode set
vs simple update behavior even after the refactoring for HSW.

-- 
Jesse Barnes, Intel Open Source Technology Center