tu: Enable UBWC for D24S8 with USAGE_SAMPLED and formatless border color

DXVK and VKD3D-Proton use customBorderColorWithoutFormat and have most of
D24S8 images with USAGE_SAMPLED, in such case we disable UBWC for correctness.
However, games don't use border color for depth-stencil images. So we elect
to ignore this edge case and force UBWC to be enabled.

See also https://github.com/doitsujin/dxvk/issues/4191

Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30545>

src/freedreno/vulkan/tu_device.cc

@@ -1361,6 +1361,7 @@ static const driOptionDescription tu_dri_options[] = {
       DRI_CONF_DISABLE_CONSERVATIVE_LRZ(false)
       DRI_CONF_TU_DONT_RESERVE_DESCRIPTOR_SET(false)
       DRI_CONF_TU_ALLOW_OOB_INDIRECT_UBO_LOADS(false)
+      DRI_CONF_TU_DISABLE_D24S8_BORDER_COLOR_WORKAROUND(false)
    DRI_CONF_SECTION_END
 };
 
@@ -1381,6 +1382,8 @@ tu_init_dri_options(struct tu_instance *instance)
          !driQueryOptionb(&instance->dri_options, "tu_dont_reserve_descriptor_set");
    instance->allow_oob_indirect_ubo_loads =
          driQueryOptionb(&instance->dri_options, "tu_allow_oob_indirect_ubo_loads");
+   instance->disable_d24s8_border_color_workaround =
+         driQueryOptionb(&instance->dri_options, "tu_disable_d24s8_border_color_workaround");
 }
 
 VKAPI_ATTR VkResult VKAPI_CALL
@@ -2513,7 +2516,8 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
 
    device->use_z24uint_s8uint =
       physical_device->info->a6xx.has_z24uint_s8uint &&
-      !border_color_without_format;
+      (!border_color_without_format ||
+       physical_device->instance->disable_d24s8_border_color_workaround);
    device->use_lrz = !TU_DEBUG(NOLRZ);
 
    tu_gpu_tracepoint_config_variable();

src/freedreno/vulkan/tu_device.h

@@ -186,6 +186,14 @@ struct tu_instance
     * See: https://github.com/doitsujin/dxvk/issues/3861
     */
    bool allow_oob_indirect_ubo_loads;
+
+   /* DXVK and VKD3D-Proton use customBorderColorWithoutFormat
+    * and have most of D24S8 images with USAGE_SAMPLED, in such case we
+    * disable UBWC for correctness. However, games don't use border color for
+    * depth-stencil images. So we elect to ignore this edge case and force
+    * UBWC to be enabled.
+    */
+   bool disable_d24s8_border_color_workaround;
 };
 VK_DEFINE_HANDLE_CASTS(tu_instance, vk.base, VkInstance,
                        VK_OBJECT_TYPE_INSTANCE)

src/util/00-mesa-defaults.conf

@@ -1368,5 +1368,14 @@ TODO: document the other workarounds.
             -->
             <option name="tu_allow_oob_indirect_ubo_loads" value="true" />
         </engine>
+        <engine engine_name_match="DXVK|vkd3d">
+            <!--
+                DXVK and VKD3D-Proton use customBorderColorWithoutFormat and have most of
+                D24S8 images with USAGE_SAMPLED, in such case we disable UBWC for correctness.
+                However, games don't use border color for depth-stencil images. So we elect
+                to ignore this edge case and force UBWC to be enabled.
+            -->
+            <option name="tu_disable_d24s8_border_color_workaround" value="true" />
+        </engine>
     </device>
 </driconf>

src/util/driconf.h

@@ -602,6 +602,10 @@
    DRI_CONF_OPT_B(tu_allow_oob_indirect_ubo_loads, def, \
                   "Some D3D11 games rely on out-of-bounds indirect UBO loads to return real values from underlying bound descriptor, this prevents us from lowering indirectly accessed UBOs to consts")
 
+#define DRI_CONF_TU_DISABLE_D24S8_BORDER_COLOR_WORKAROUND(def) \
+   DRI_CONF_OPT_B(tu_disable_d24s8_border_color_workaround, def, \
+                  "Use UBWC for D24S8 images with VK_IMAGE_USAGE_SAMPLED_BIT when customBorderColorWithoutFormat is enabled")
+
 /**
  * \brief venus specific configuration options
  */