[Mesa-stable] [PATCH] radeonsi: mark fast-cleared textures as compressed when dirtying

[Marek Olšák]

Reviewed-by: Marek Olšák <marek.olsak at amd.com>

Marek

On Thu, May 4, 2017 at 3:23 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>
> There are a bunch of piglit fast clear tests that regressed on SI, for
> example ./bin/ext_framebuffer_multisample-fast-clear single-sample.
>
> The problem is that a texture is bound as a framebuffer, cleared, and
> then rendered from in a loop that loops through different clear colors.
> The texture is never rebound during all this, so the change to
> tex->dirty_level_mask during fast clear was not taken into account
> when checking for compressed textures.
>
> I have considered simply reverting the problematic commit. However,
> I think this solution is better. It does require looping through all
> bound textures after a fast clear, but the alternative would require
> visiting more textures needless on every draw. Draws are much more
> common than clears.
>
> Note that the rendering feedback loop rules do not apply here, because
> the framebuffer binding is changed between the glClear and the draw
> that samples from the texture that was cleared.
>
> Fixes: bdd644976952 ("radeonsi: don't mark non-dirty textures with CMASK as compressed")
> Cc: 17.1 <mesa-stable at lists.freedesktop.org>
> ---
>  src/gallium/drivers/radeon/r600_texture.c | 16 ++++++++++++++--
>  1 file changed, 14 insertions(+), 2 deletions(-)
>
> diff --git a/src/gallium/drivers/radeon/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c
> index 8110ec1..57e3960 100644
> --- a/src/gallium/drivers/radeon/r600_texture.c
> +++ b/src/gallium/drivers/radeon/r600_texture.c
> @@ -2713,22 +2713,29 @@ void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
>                         if (rctx->screen->debug_flags & DBG_NO_DCC_CLEAR)
>                                 continue;
>
>                         if (!vi_get_fast_clear_parameters(fb->cbufs[i]->format,
>                                                           color, &reset_value,
>                                                           &clear_words_needed))
>                                 continue;
>
>                         vi_dcc_clear_level(rctx, tex, 0, reset_value);
>
> -                       if (clear_words_needed)
> -                               tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
> +                       unsigned level_bit = 1 << fb->cbufs[i]->u.tex.level;
> +                       if (clear_words_needed) {
> +                               bool need_compressed_update = !tex->dirty_level_mask;
> +
> +                               tex->dirty_level_mask |= level_bit;
> +
> +                               if (need_compressed_update)
> +                                       p_atomic_inc(&rctx->screen->compressed_colortex_counter);
> +                       }
>                         tex->separate_dcc_dirty = true;
>                 } else {
>                         /* 128-bit formats are unusupported */
>                         if (tex->surface.bpe > 8) {
>                                 continue;
>                         }
>
>                         /* RB+ doesn't work with CMASK fast clear on Stoney. */
>                         if (rctx->family == CHIP_STONEY)
>                                 continue;
> @@ -2737,21 +2744,26 @@ void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
>                         r600_texture_alloc_cmask_separate(rctx->screen, tex);
>                         if (tex->cmask.size == 0) {
>                                 continue;
>                         }
>
>                         /* Do the fast clear. */
>                         rctx->clear_buffer(&rctx->b, &tex->cmask_buffer->b.b,
>                                            tex->cmask.offset, tex->cmask.size, 0,
>                                            R600_COHERENCY_CB_META);
>
> +                       bool need_compressed_update = !tex->dirty_level_mask;
> +
>                         tex->dirty_level_mask |= 1 << fb->cbufs[i]->u.tex.level;
> +
> +                       if (need_compressed_update)
> +                               p_atomic_inc(&rctx->screen->compressed_colortex_counter);
>                 }
>
>                 /* We can change the micro tile mode before a full clear. */
>                 if (rctx->screen->chip_class >= SI)
>                         si_set_optimal_micro_tile_mode(rctx->screen, tex);
>
>                 evergreen_set_clear_color(tex, fb->cbufs[i]->format, color);
>
>                 if (dirty_cbufs)
>                         *dirty_cbufs |= 1 << i;
> --
> 2.9.3
>
> _______________________________________________
> mesa-stable mailing list
> mesa-stable at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-stable