Re: [Intel-gfx] [PATCH 3/4] drm/i915: Use wm0 only during async flips for DG2

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Fri, Dec 03, 2021 at 11:40:40AM +0200, Stanislav Lisovskiy wrote:
> This optimization allows to achieve higher perfomance
> during async flips.
> For the first async flip we have to still temporarily
> switch to sync flip, in order to reprogram plane
> watermarks, so this requires taking into account
> old plane state's do_async_flip flag.
> 
> Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
> ---
>  drivers/gpu/drm/i915/display/intel_display.c | 13 ++++++++++++-
>  drivers/gpu/drm/i915/intel_pm.c              |  5 ++++-
>  2 files changed, 16 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> index b8fe40050463..872bbb965992 100644
> --- a/drivers/gpu/drm/i915/display/intel_display.c
> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> @@ -5439,6 +5439,16 @@ static bool needs_scaling(const struct intel_plane_state *state)
>  	return (src_w != dst_w || src_h != dst_h);
>  }
>  
> +static bool needs_async_flip_wm_override(struct intel_plane *plane,
> +					 const struct intel_plane_state *new_plane_state,
> +					 const struct intel_plane_state *old_plane_state)
> +{
> +	struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
> +
> +	return (DISPLAY_VER(dev_priv) >= 13) && !old_plane_state->do_async_flip &&
> +	       new_plane_state->do_async_flip;
> +}
> +
>  int intel_plane_atomic_calc_changes(const struct intel_crtc_state *old_crtc_state,
>  				    struct intel_crtc_state *new_crtc_state,
>  				    const struct intel_plane_state *old_plane_state,
> @@ -5558,7 +5568,8 @@ int intel_plane_atomic_calc_changes(const struct intel_crtc_state *old_crtc_stat
>  			 needs_scaling(new_plane_state))))
>  		new_crtc_state->disable_lp_wm = true;
>  
> -	if (new_crtc_state->uapi.async_flip && plane->async_flip)
> +	if (new_crtc_state->uapi.async_flip && new_plane_state->do_async_flip &&
> +	    !needs_async_flip_wm_override(plane, new_plane_state, old_plane_state))
>  		new_plane_state->do_async_flip = true;

That code looks super confused.

>  
>  	return 0;
> diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> index c5c1b6bef9a2..0b45d1d61d0f 100644
> --- a/drivers/gpu/drm/i915/intel_pm.c
> +++ b/drivers/gpu/drm/i915/intel_pm.c
> @@ -5542,8 +5542,11 @@ static void skl_compute_plane_wm(const struct intel_crtc_state *crtc_state,
>  	uint_fixed_16_16_t method1, method2;
>  	uint_fixed_16_16_t selected_result;
>  	u32 blocks, lines, min_ddb_alloc = 0;
> +	bool dg2_async_flip_optimization = (DISPLAY_VER(dev_priv) >= 13) &&
> +					   crtc_state->uapi.async_flip &&
> +					   plane_id != PLANE_CURSOR;

Function please. Hmm. And rather than checking the plane_id we should
probably just check plane->async_flip since we don't want to minimize
the watermarks for any other plane than the one doing the async flips.

>  
> -	if (latency == 0) {
> +	if (latency == 0 || (dg2_async_flip_optimization && (level > 0))) {
>  		/* reject it */
>  		result->min_ddb_alloc = U16_MAX;
>  		return;
> -- 
> 2.24.1.485.gad05a3d8e5

-- 
Ville Syrjälä
Intel