[Intel-gfx] [PATCH v3 3/4] drm/i915: Skip CURBASE write when nothing changed

[Chris Wilson]

On Fri, Sep 12, 2014 at 08:53:34PM +0300, ville.syrjala at linux.intel.com wrote:
> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> 
> Only write CURBASE when something about the cursor changed. Also
> eliminate the unnecessary posting read after writing CURCNTR.
> 
> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> ---
>  drivers/gpu/drm/i915/intel_display.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 82c0ad1..60c1aa4 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -8322,16 +8322,18 @@ static void i9xx_update_cursor(struct drm_crtc *crtc, u32 base)
>  			cntl |= CURSOR_PIPE_CSC_ENABLE;
>  	}
>  
> -	if (intel_crtc->cursor_cntl != cntl) {
> +	if (intel_crtc->cursor_cntl != cntl)
>  		I915_WRITE(CURCNTR(pipe), cntl);
> -		POSTING_READ(CURCNTR(pipe));
> -		intel_crtc->cursor_cntl = cntl;
> -	}
> +
> +	if (intel_crtc->cursor_cntl == cntl &&
> +	    intel_crtc->cursor_base == base)
> +		return;

I'd vote for doing this first and then

I915_WRITE(CURCNTR(pipe), cntl);

unconditionally along with the CURBASE flush.

>  	/* and commit changes on next vblank */
>  	I915_WRITE(CURBASE(pipe), base);
>  	POSTING_READ(CURBASE(pipe));
>  
> +	intel_crtc->cursor_cntl = cntl;
>  	intel_crtc->cursor_base = base;
>  }

-- 
Chris Wilson, Intel Open Source Technology Centre