anv: Add stencil texturing support for gen7

Intel hardware didn't get support for sampling from W-tiled (required
for stencil) images until Broadwell so we can't directly sample from
stencil.  Instead, if we want to support stencil texturing on gen7
hardware, we have to keep a texture-capable shadow copy around and use
BLORP to update when stencil changes.  The one thing this commit does
not implement is self-dependencies with stencil input attachments.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99493
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

diff --git a/src/intel/vulkan/anv_formats.c b/src/intel/vulkan/anv_formats.c
index 70543e0cfdfc7137d1bbeb09afa62cb7b4aa8f16..d46da754ba3593147c9e67c3b5756532b5a9eb59 100644 (file)
--- a/src/intel/vulkan/anv_formats.c
+++ b/src/intel/vulkan/anv_formats.c
@@ -514,14 +514,12 @@ anv_get_image_format_features(const struct gen_device_info *devinfo,
          return 0;
 
       flags |= VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT |
+               VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
                VK_FORMAT_FEATURE_BLIT_SRC_BIT |
                VK_FORMAT_FEATURE_BLIT_DST_BIT |
                VK_FORMAT_FEATURE_TRANSFER_SRC_BIT |
                VK_FORMAT_FEATURE_TRANSFER_DST_BIT;
 
-      if (aspects == VK_IMAGE_ASPECT_DEPTH_BIT || devinfo->gen >= 8)
-         flags |= VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT;
-
       if ((aspects & VK_IMAGE_ASPECT_DEPTH_BIT) && devinfo->gen >= 9)
          flags |= VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_MINMAX_BIT_EXT;
 

diff --git a/src/intel/vulkan/anv_image.c b/src/intel/vulkan/anv_image.c
index 038a01f8a396017cfa867c82b71970454870ef93..f405aa8067f8467f757e2db133390ffcd7639bf5 100644 (file)
--- a/src/intel/vulkan/anv_image.c
+++ b/src/intel/vulkan/anv_image.c
@@ -336,6 +336,12 @@ make_surface(const struct anv_device *dev,
       needs_shadow = true;
    }
 
+   if (dev->info.gen <= 7 &&
+       aspect == VK_IMAGE_ASPECT_STENCIL_BIT &&
+       (image->usage & VK_IMAGE_USAGE_SAMPLED_BIT)) {
+      needs_shadow = true;
+   }
+
    ok = isl_surf_init(&dev->isl_dev, &anv_surf->isl,
       .dim = vk_to_isl_surf_dim[image->type],
       .format = plane_format.isl_format,
@@ -359,12 +365,11 @@ make_surface(const struct anv_device *dev,
 
    /* If an image is created as BLOCK_TEXEL_VIEW_COMPATIBLE, then we need to
     * create an identical tiled shadow surface for use while texturing so we
-    * don't get garbage performance.
+    * don't get garbage performance.  If we're on gen7 and the image contains
+    * stencil, then we need to maintain a shadow because we can't texture from
+    * W-tiled images.
     */
    if (needs_shadow) {
-      assert(aspect == VK_IMAGE_ASPECT_COLOR_BIT);
-      assert(tiling_flags == ISL_TILING_LINEAR_BIT);
-
       ok = isl_surf_init(&dev->isl_dev, &image->planes[plane].shadow_surface.isl,
          .dim = vk_to_isl_surf_dim[image->type],
          .format = plane_format.isl_format,
@@ -1275,6 +1280,16 @@ anv_image_fill_surface_state(struct anv_device *device,
       surface = &image->planes[plane].shadow_surface;
    }
 
+   /* For texturing from stencil on gen7, we have to sample from a shadow
+    * surface because we don't support W-tiling in the sampler.
+    */
+   if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
+       aspect == VK_IMAGE_ASPECT_STENCIL_BIT) {
+      assert(device->info.gen == 7);
+      assert(view_usage & ISL_SURF_USAGE_TEXTURE_BIT);
+      surface = &image->planes[plane].shadow_surface;
+   }
+
    if (view_usage == ISL_SURF_USAGE_RENDER_TARGET_BIT)
       view.swizzle = anv_swizzle_for_render(view.swizzle);
 

diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c
index 2ee06618d07392e1c171efa931c137a8c40843f5..a580fda6c2c3914d838a222ddf2243d293e248e9 100644 (file)
--- a/src/intel/vulkan/genX_cmd_buffer.c
+++ b/src/intel/vulkan/genX_cmd_buffer.c
@@ -474,6 +474,13 @@ transition_depth_buffer(struct anv_cmd_buffer *cmd_buffer,
                        0, 0, 1, hiz_op);
 }
 
+static inline bool
+vk_image_layout_stencil_write_optimal(VkImageLayout layout)
+{
+   return layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
+          layout == VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL;
+}
+
 /* Transitions a HiZ-enabled depth buffer from one layout to another. Unless
  * the initial layout is undefined, the HiZ buffer and depth buffer will
  * represent the same data at the end of this operation.
@@ -486,6 +493,34 @@ transition_stencil_buffer(struct anv_cmd_buffer *cmd_buffer,
                           VkImageLayout initial_layout,
                           VkImageLayout final_layout)
 {
+#if GEN_GEN == 7
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects,
+                                              VK_IMAGE_ASPECT_STENCIL_BIT);
+
+   /* On gen7, we have to store a texturable version of the stencil buffer in
+    * a shadow whenever VK_IMAGE_USAGE_SAMPLED_BIT is set and copy back and
+    * forth at strategic points.  Stencil writes are only allowed in three
+    * layouts:
+    *
+    *  - VK_IMAGE_LAYOUT_GENERAL
+    *  - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
+    *  - VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
+    *  - VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL
+    *
+    * For general, we have no nice opportunity to transition so we do the copy
+    * to the shadow unconditionally at the end of the subpass.  For transfer
+    * destinations, we can update it as part of the transfer op.  For the
+    * other two, we delay the copy until a transition into some other layout.
+    */
+   if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
+       vk_image_layout_stencil_write_optimal(initial_layout) &&
+       !vk_image_layout_stencil_write_optimal(final_layout)) {
+      anv_image_copy_to_shadow(cmd_buffer, image,
+                               VK_IMAGE_ASPECT_STENCIL_BIT,
+                               base_level, level_count,
+                               base_layer, layer_count);
+   }
+#endif /* GEN_GEN == 7 */
 }
 
 #define MI_PREDICATE_SRC0    0x2400
@@ -4503,6 +4538,47 @@ cmd_buffer_end_subpass(struct anv_cmd_buffer *cmd_buffer)
       }
    }
 
+#if GEN_GEN == 7
+   /* On gen7, we have to store a texturable version of the stencil buffer in
+    * a shadow whenever VK_IMAGE_USAGE_SAMPLED_BIT is set and copy back and
+    * forth at strategic points.  Stencil writes are only allowed in three
+    * layouts:
+    *
+    *  - VK_IMAGE_LAYOUT_GENERAL
+    *  - VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL
+    *  - VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL
+    *  - VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL
+    *
+    * For general, we have no nice opportunity to transition so we do the copy
+    * to the shadow unconditionally at the end of the subpass.  For transfer
+    * destinations, we can update it as part of the transfer op.  For the
+    * other two, we delay the copy until a transition into some other layout.
+    */
+   if (subpass->depth_stencil_attachment) {
+      uint32_t a = subpass->depth_stencil_attachment->attachment;
+      assert(a != VK_ATTACHMENT_UNUSED);
+
+      struct anv_attachment_state *att_state = &cmd_state->attachments[a];
+      struct anv_image_view *iview = fb->attachments[a];
+      const struct anv_image *image = iview->image;
+
+      if (image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
+         uint32_t plane = anv_image_aspect_to_plane(image->aspects,
+                                                    VK_IMAGE_ASPECT_STENCIL_BIT);
+
+         if (image->planes[plane].shadow_surface.isl.size_B > 0 &&
+             att_state->current_layout == VK_IMAGE_LAYOUT_GENERAL) {
+            assert(image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT);
+            anv_image_copy_to_shadow(cmd_buffer, image,
+                                     VK_IMAGE_ASPECT_STENCIL_BIT,
+                                     iview->planes[plane].isl.base_level, 1,
+                                     iview->planes[plane].isl.base_array_layer,
+                                     fb->layers);
+         }
+      }
+   }
+#endif /* GEN_GEN == 7 */
+
    for (uint32_t i = 0; i < subpass->attachment_count; ++i) {
       const uint32_t a = subpass->attachments[i].attachment;
       if (a == VK_ATTACHMENT_UNUSED)