radeonsi: work around an MSAA fast stencil clear problem
authorNicolai Hähnle <nicolai.haehnle@amd.com>
Fri, 22 Apr 2016 22:28:46 +0000 (17:28 -0500)
committerNicolai Hähnle <nicolai.haehnle@amd.com>
Wed, 27 Apr 2016 16:16:40 +0000 (11:16 -0500)
A piglit test (arb_texture_multisample-stencil-clear) has been sent.
This problem was discovered analyzing

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93767
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
src/gallium/drivers/radeonsi/si_state.c

index 6017383a675d5e271921c669589cb7ab4a51321d..06798f2ac600aef5c7d8ee9001efa8a5e5821107 100644 (file)
@@ -2383,9 +2383,21 @@ static void si_init_depth_surface(struct si_context *sctx,
                z_info |= S_028040_TILE_SURFACE_ENABLE(1) |
                          S_028040_ALLOW_EXPCLEAR(1);
 
-               if (rtex->surface.flags & RADEON_SURF_SBUFFER)
-                       s_info |= S_028044_ALLOW_EXPCLEAR(1);
-               else
+               if (rtex->surface.flags & RADEON_SURF_SBUFFER) {
+                       /* Workaround: For a not yet understood reason, the
+                        * combination of MSAA, fast stencil clear and stencil
+                        * decompress messes with subsequent stencil buffer
+                        * uses. Problem was reproduced on Verde, Bonaire,
+                        * Tonga, and Carrizo.
+                        *
+                        * Disabling EXPCLEAR works around the problem.
+                        *
+                        * Check piglit's arb_texture_multisample-stencil-clear
+                        * test if you want to try changing this.
+                        */
+                       if (rtex->resource.b.b.nr_samples <= 1)
+                               s_info |= S_028044_ALLOW_EXPCLEAR(1);
+               } else
                        /* Use all of the htile_buffer for depth if there's no stencil. */
                        s_info |= S_028044_TILE_STENCIL_DISABLE(1);