[Mesa-dev] [PATCH 00/32] [v2] Renderbuffer Decompression (and GBM modifiers)

[Ben Widawsky]

On 17-01-05 12:16:45, Chad Versace wrote:
>This patch has a regressing side-effect: it disables CCS for all single-sample
>miptrees created by glRenderbufferStorage(). After stepping through some Piglit
>tests, I believe the problem is that line 161 below always returns false for
>such miptrees because mt->is_scanout is true. I don't understand why
>mt->is_scanout would be set for those miptrees.
>
>The CCS gets disabled even if app never uses EGLImage nor modifiers.
>

The bandwidth savings are definitely there - so it can't be disabling it in all
cases. I'll look a bit further. Something still isn't entirely clear to me.

>    149 bool
>    150 intel_miptree_supports_non_msrt_fast_clear(struct brw_context *brw,
>    151                                            const struct intel_mipmap_tree *mt)
>    152 {
>    153    /* MCS support does not exist prior to Gen7 */
>    154    if (brw->gen < 7)
>    155       return false;
>    156
>    157    if (mt->aux_disable & INTEL_AUX_DISABLE_MCS)
>    158       return false;
>    159
>    160    if (mt->is_scanout && mt->msaa_layout != INTEL_MSAA_LAYOUT_CMS)
>  **161       return false;
>
>That explains why 'egl_khr_gl_renderbuffer_image-clear-shared-image GL_RGBA'
>passes on this patch series.