anv/image: Support color aspects in layout_to_aux_usage
[mesa.git] / src / intel / vulkan / anv_image.c
index 1328f4cc27b12d8c6d579fda9610942e71736ac4..583e085cf94116306b90c35b7adf59eb4e91202b 100644 (file)
@@ -27,6 +27,7 @@
 #include <unistd.h>
 #include <fcntl.h>
 #include <sys/mman.h>
+#include <drm_fourcc.h>
 
 #include "anv_private.h"
 #include "util/debug.h"
 
 #include "vk_format_info.h"
 
-/**
- * Exactly one bit must be set in \a aspect.
- */
 static isl_surf_usage_flags_t
 choose_isl_surf_usage(VkImageCreateFlags vk_create_flags,
                       VkImageUsageFlags vk_usage,
-                      VkImageAspectFlags aspect)
+                      isl_surf_usage_flags_t isl_extra_usage,
+                      VkImageAspectFlagBits aspect)
 {
-   isl_surf_usage_flags_t isl_usage = 0;
+   isl_surf_usage_flags_t isl_usage = isl_extra_usage;
 
    if (vk_usage & VK_IMAGE_USAGE_SAMPLED_BIT)
       isl_usage |= ISL_SURF_USAGE_TEXTURE_BIT;
@@ -68,6 +67,9 @@ choose_isl_surf_usage(VkImageCreateFlags vk_create_flags,
       isl_usage |= ISL_SURF_USAGE_STENCIL_BIT;
       break;
    case VK_IMAGE_ASPECT_COLOR_BIT:
+   case VK_IMAGE_ASPECT_PLANE_0_BIT_KHR:
+   case VK_IMAGE_ASPECT_PLANE_1_BIT_KHR:
+   case VK_IMAGE_ASPECT_PLANE_2_BIT_KHR:
       break;
    default:
       unreachable("bad VkImageAspect");
@@ -89,32 +91,63 @@ choose_isl_surf_usage(VkImageCreateFlags vk_create_flags,
    return isl_usage;
 }
 
-/**
- * Exactly one bit must be set in \a aspect.
- */
-static struct anv_surface *
-get_surface(struct anv_image *image, VkImageAspectFlags aspect)
+static isl_tiling_flags_t
+choose_isl_tiling_flags(const struct anv_image_create_info *anv_info,
+                        const struct isl_drm_modifier_info *isl_mod_info)
 {
-   switch (aspect) {
+   const VkImageCreateInfo *base_info = anv_info->vk_info;
+   isl_tiling_flags_t flags = 0;
+
+   switch (base_info->tiling) {
    default:
-      unreachable("bad VkImageAspect");
-   case VK_IMAGE_ASPECT_COLOR_BIT:
-      return &image->color_surface;
-   case VK_IMAGE_ASPECT_DEPTH_BIT:
-      return &image->depth_surface;
-   case VK_IMAGE_ASPECT_STENCIL_BIT:
-      return &image->stencil_surface;
+      unreachable("bad VkImageTiling");
+   case VK_IMAGE_TILING_OPTIMAL:
+      flags = ISL_TILING_ANY_MASK;
+      break;
+   case VK_IMAGE_TILING_LINEAR:
+      flags = ISL_TILING_LINEAR_BIT;
+      break;
    }
+
+   if (anv_info->isl_tiling_flags)
+      flags &= anv_info->isl_tiling_flags;
+
+   if (isl_mod_info)
+      flags &= 1 << isl_mod_info->tiling;
+
+   assert(flags);
+
+   return flags;
+}
+
+static struct anv_surface *
+get_surface(struct anv_image *image, VkImageAspectFlagBits aspect)
+{
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+   return &image->planes[plane].surface;
 }
 
 static void
-add_surface(struct anv_image *image, struct anv_surface *surf)
+add_surface(struct anv_image *image, struct anv_surface *surf, uint32_t plane)
 {
    assert(surf->isl.size > 0); /* isl surface must be initialized */
 
-   surf->offset = align_u32(image->size, surf->isl.alignment);
+   if (image->disjoint) {
+      surf->offset = align_u32(image->planes[plane].size, surf->isl.alignment);
+      /* Plane offset is always 0 when it's disjoint. */
+   } else {
+      surf->offset = align_u32(image->size, surf->isl.alignment);
+      /* Determine plane's offset only once when the first surface is added. */
+      if (image->planes[plane].size == 0)
+         image->planes[plane].offset = image->size;
+   }
+
    image->size = surf->offset + surf->isl.size;
+   image->planes[plane].size = (surf->offset + surf->isl.size) - image->planes[plane].offset;
+
    image->alignment = MAX2(image->alignment, surf->isl.alignment);
+   image->planes[plane].alignment = MAX2(image->planes[plane].alignment,
+                                         surf->isl.alignment);
 }
 
 
@@ -194,11 +227,13 @@ all_formats_ccs_e_compatible(const struct gen_device_info *devinfo,
  */
 static void
 add_fast_clear_state_buffer(struct anv_image *image,
+                            VkImageAspectFlagBits aspect,
+                            uint32_t plane,
                             const struct anv_device *device)
 {
    assert(image && device);
-   assert(image->aux_surface.isl.size > 0 &&
-          image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
+   assert(image->planes[plane].aux_surface.isl.size > 0 &&
+          image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
 
    /* The offset to the buffer of clear values must be dword-aligned for GPU
     * memcpy operations. It is located immediately after the auxiliary surface.
@@ -212,31 +247,43 @@ add_fast_clear_state_buffer(struct anv_image *image,
    /* Auxiliary buffers should be a multiple of 4K, so the start of the clear
     * values buffer should already be dword-aligned.
     */
-   assert(image->aux_surface.isl.size % 4 == 0);
+   assert((image->planes[plane].offset + image->planes[plane].size) % 4 == 0);
 
-   /* This buffer should be at the very end of the image. */
-   assert(image->size ==
-          image->aux_surface.offset + image->aux_surface.isl.size);
+   /* This buffer should be at the very end of the plane. */
+   if (image->disjoint) {
+      assert(image->planes[plane].size ==
+             (image->planes[plane].offset + image->planes[plane].size));
+   } else {
+      assert(image->size ==
+             (image->planes[plane].offset + image->planes[plane].size));
+   }
 
    const unsigned entry_size = anv_fast_clear_state_entry_size(device);
    /* There's no padding between entries, so ensure that they're always a
     * multiple of 32 bits in order to enable GPU memcpy operations.
     */
    assert(entry_size % 4 == 0);
-   image->size += entry_size * anv_image_aux_levels(image);
+
+   const unsigned plane_state_size =
+      entry_size * anv_image_aux_levels(image, aspect);
+
+   image->planes[plane].fast_clear_state_offset =
+      image->planes[plane].offset + image->planes[plane].size;
+
+   image->planes[plane].size += plane_state_size;
+   image->size += plane_state_size;
 }
 
 /**
  * Initialize the anv_image::*_surface selected by \a aspect. Then update the
  * image's memory requirements (that is, the image's size and alignment).
- *
- * Exactly one bit must be set in \a aspect.
  */
 static VkResult
 make_surface(const struct anv_device *dev,
              struct anv_image *image,
              const struct anv_image_create_info *anv_info,
-             VkImageAspectFlags aspect)
+             isl_tiling_flags_t tiling_flags,
+             VkImageAspectFlagBits aspect)
 {
    const VkImageCreateInfo *vk_info = anv_info->vk_info;
    bool ok UNUSED;
@@ -247,26 +294,17 @@ make_surface(const struct anv_device *dev,
       [VK_IMAGE_TYPE_3D] = ISL_SURF_DIM_3D,
    };
 
-   /* Translate the Vulkan tiling to an equivalent ISL tiling, then filter the
-    * result with an optionally provided ISL tiling argument.
-    */
-   isl_tiling_flags_t tiling_flags =
-      (vk_info->tiling == VK_IMAGE_TILING_LINEAR) ?
-      ISL_TILING_LINEAR_BIT : ISL_TILING_ANY_MASK;
-
-   if (anv_info->isl_tiling_flags)
-      tiling_flags &= anv_info->isl_tiling_flags;
-
-   assert(tiling_flags);
-
-   struct anv_surface *anv_surf = get_surface(image, aspect);
-
    image->extent = anv_sanitize_image_extent(vk_info->imageType,
                                              vk_info->extent);
 
-   enum isl_format format = anv_get_isl_format(&dev->info, vk_info->format,
-                                               aspect, vk_info->tiling);
-   assert(format != ISL_FORMAT_UNSUPPORTED);
+   const unsigned plane = anv_image_aspect_to_plane(image->aspects, aspect);
+   const  struct anv_format_plane plane_format =
+      anv_get_format_plane(&dev->info, image->vk_format, aspect, image->tiling);
+   struct anv_surface *anv_surf = &image->planes[plane].surface;
+
+   const isl_surf_usage_flags_t usage =
+      choose_isl_surf_usage(vk_info->flags, image->usage,
+                            anv_info->isl_extra_usage_flags, aspect);
 
    /* If an image is created as BLOCK_TEXEL_VIEW_COMPATIBLE, then we need to
     * fall back to linear on Broadwell and earlier because we aren't
@@ -278,31 +316,31 @@ make_surface(const struct anv_device *dev,
    if (dev->info.gen <= 8 &&
        (vk_info->flags & VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT_KHR) &&
        vk_info->tiling == VK_IMAGE_TILING_OPTIMAL) {
-      assert(isl_format_is_compressed(format));
+      assert(isl_format_is_compressed(plane_format.isl_format));
       tiling_flags = ISL_TILING_LINEAR_BIT;
       needs_shadow = true;
    }
 
    ok = isl_surf_init(&dev->isl_dev, &anv_surf->isl,
       .dim = vk_to_isl_surf_dim[vk_info->imageType],
-      .format = format,
-      .width = image->extent.width,
-      .height = image->extent.height,
+      .format = plane_format.isl_format,
+      .width = image->extent.width / plane_format.denominator_scales[0],
+      .height = image->extent.height / plane_format.denominator_scales[1],
       .depth = image->extent.depth,
       .levels = vk_info->mipLevels,
       .array_len = vk_info->arrayLayers,
       .samples = vk_info->samples,
       .min_alignment = 0,
       .row_pitch = anv_info->stride,
-      .usage = choose_isl_surf_usage(vk_info->flags, image->usage, aspect),
+      .usage = usage,
       .tiling_flags = tiling_flags);
 
-   /* isl_surf_init() will fail only if provided invalid input. Invalid input
-    * is illegal in Vulkan.
-    */
-   assert(ok);
+   if (!ok)
+      return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+
+   image->planes[plane].aux_usage = ISL_AUX_USAGE_NONE;
 
-   add_surface(image, anv_surf);
+   add_surface(image, anv_surf, plane);
 
    /* 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
@@ -312,9 +350,9 @@ make_surface(const struct anv_device *dev,
       assert(aspect == VK_IMAGE_ASPECT_COLOR_BIT);
       assert(tiling_flags == ISL_TILING_LINEAR_BIT);
 
-      ok = isl_surf_init(&dev->isl_dev, &image->shadow_surface.isl,
+      ok = isl_surf_init(&dev->isl_dev, &image->planes[plane].shadow_surface.isl,
          .dim = vk_to_isl_surf_dim[vk_info->imageType],
-         .format = format,
+         .format = plane_format.isl_format,
          .width = image->extent.width,
          .height = image->extent.height,
          .depth = image->extent.depth,
@@ -323,7 +361,7 @@ make_surface(const struct anv_device *dev,
          .samples = vk_info->samples,
          .min_alignment = 0,
          .row_pitch = anv_info->stride,
-         .usage = choose_isl_surf_usage(image->usage, image->usage, aspect),
+         .usage = usage,
          .tiling_flags = ISL_TILING_ANY_MASK);
 
       /* isl_surf_init() will fail only if provided invalid input. Invalid input
@@ -331,7 +369,7 @@ make_surface(const struct anv_device *dev,
        */
       assert(ok);
 
-      add_surface(image, &image->shadow_surface);
+      add_surface(image, &image->planes[plane].shadow_surface, plane);
    }
 
    /* Add a HiZ surface to a depth buffer that will be used for rendering.
@@ -357,24 +395,43 @@ make_surface(const struct anv_device *dev,
       } else if (dev->info.gen == 8 && vk_info->samples > 1) {
          anv_perf_warn(dev->instance, image, "Enable gen8 multisampled HiZ");
       } else if (!unlikely(INTEL_DEBUG & DEBUG_NO_HIZ)) {
-         assert(image->aux_surface.isl.size == 0);
-         ok = isl_surf_get_hiz_surf(&dev->isl_dev, &image->depth_surface.isl,
-                                    &image->aux_surface.isl);
+         assert(image->planes[plane].aux_surface.isl.size == 0);
+         ok = isl_surf_get_hiz_surf(&dev->isl_dev,
+                                    &image->planes[plane].surface.isl,
+                                    &image->planes[plane].aux_surface.isl);
          assert(ok);
-         add_surface(image, &image->aux_surface);
-         image->aux_usage = ISL_AUX_USAGE_HIZ;
+         add_surface(image, &image->planes[plane].aux_surface, plane);
+         image->planes[plane].aux_usage = ISL_AUX_USAGE_HIZ;
       }
-   } else if (aspect == VK_IMAGE_ASPECT_COLOR_BIT && vk_info->samples == 1) {
-      if (!unlikely(INTEL_DEBUG & DEBUG_NO_RBC)) {
-         assert(image->aux_surface.isl.size == 0);
-         ok = isl_surf_get_ccs_surf(&dev->isl_dev, &anv_surf->isl,
-                                    &image->aux_surface.isl, 0);
+   } else if ((aspect & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) && vk_info->samples == 1) {
+      /* TODO: Disallow compression with :
+       *
+       *     1) non multiplanar images (We appear to hit a sampler bug with
+       *        CCS & R16G16 format. Putting the clear state a page/4096bytes
+       *        further fixes the issue).
+       *
+       *     2) alias images, because they might be aliases of images
+       *        described in 1)
+       *
+       *     3) compression disabled by debug
+       */
+      const bool allow_compression =
+         image->n_planes == 1 &&
+         (vk_info->flags & VK_IMAGE_CREATE_ALIAS_BIT_KHR) == 0 &&
+         likely((INTEL_DEBUG & DEBUG_NO_RBC) == 0);
+
+      if (allow_compression) {
+         assert(image->planes[plane].aux_surface.isl.size == 0);
+         ok = isl_surf_get_ccs_surf(&dev->isl_dev,
+                                    &image->planes[plane].surface.isl,
+                                    &image->planes[plane].aux_surface.isl, 0);
          if (ok) {
 
             /* Disable CCS when it is not useful (i.e., when you can't render
              * to the image with CCS enabled).
              */
-            if (!isl_format_supports_rendering(&dev->info, format)) {
+            if (!isl_format_supports_rendering(&dev->info,
+                                               plane_format.isl_format)) {
                /* While it may be technically possible to enable CCS for this
                 * image, we currently don't have things hooked up to get it
                 * working.
@@ -382,12 +439,12 @@ make_surface(const struct anv_device *dev,
                anv_perf_warn(dev->instance, image,
                              "This image format doesn't support rendering. "
                              "Not allocating an CCS buffer.");
-               image->aux_surface.isl.size = 0;
+               image->planes[plane].aux_surface.isl.size = 0;
                return VK_SUCCESS;
             }
 
-            add_surface(image, &image->aux_surface);
-            add_fast_clear_state_buffer(image, dev);
+            add_surface(image, &image->planes[plane].aux_surface, plane);
+            add_fast_clear_state_buffer(image, aspect, plane, dev);
 
             /* For images created without MUTABLE_FORMAT_BIT set, we know that
              * they will always be used with the original format.  In
@@ -399,25 +456,58 @@ make_surface(const struct anv_device *dev,
              */
             if (!(vk_info->usage & VK_IMAGE_USAGE_STORAGE_BIT) &&
                 all_formats_ccs_e_compatible(&dev->info, vk_info)) {
-               image->aux_usage = ISL_AUX_USAGE_CCS_E;
+               image->planes[plane].aux_usage = ISL_AUX_USAGE_CCS_E;
             }
          }
       }
-   } else if (aspect == VK_IMAGE_ASPECT_COLOR_BIT && vk_info->samples > 1) {
-      assert(image->aux_surface.isl.size == 0);
+   } else if ((aspect & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) && vk_info->samples > 1) {
       assert(!(vk_info->usage & VK_IMAGE_USAGE_STORAGE_BIT));
-      ok = isl_surf_get_mcs_surf(&dev->isl_dev, &anv_surf->isl,
-                                 &image->aux_surface.isl);
+      assert(image->planes[plane].aux_surface.isl.size == 0);
+      ok = isl_surf_get_mcs_surf(&dev->isl_dev,
+                                 &image->planes[plane].surface.isl,
+                                 &image->planes[plane].aux_surface.isl);
       if (ok) {
-         add_surface(image, &image->aux_surface);
-         add_fast_clear_state_buffer(image, dev);
-         image->aux_usage = ISL_AUX_USAGE_MCS;
+         add_surface(image, &image->planes[plane].aux_surface, plane);
+         add_fast_clear_state_buffer(image, aspect, plane, dev);
+         image->planes[plane].aux_usage = ISL_AUX_USAGE_MCS;
       }
    }
 
+   assert((image->planes[plane].offset + image->planes[plane].size) == image->size);
+
+   /* Upper bound of the last surface should be smaller than the plane's
+    * size.
+    */
+   assert((MAX2(image->planes[plane].surface.offset,
+                image->planes[plane].aux_surface.offset) +
+           (image->planes[plane].aux_surface.isl.size > 0 ?
+            image->planes[plane].aux_surface.isl.size :
+            image->planes[plane].surface.isl.size)) <=
+          (image->planes[plane].offset + image->planes[plane].size));
+
+   if (image->planes[plane].aux_surface.isl.size) {
+      /* assert(image->planes[plane].fast_clear_state_offset == */
+      /*        (image->planes[plane].aux_surface.offset + image->planes[plane].aux_surface.isl.size)); */
+      assert(image->planes[plane].fast_clear_state_offset <
+             (image->planes[plane].offset + image->planes[plane].size));
+   }
+
    return VK_SUCCESS;
 }
 
+static const struct isl_drm_modifier_info *
+get_legacy_scanout_drm_format_mod(VkImageTiling tiling)
+{
+   switch (tiling) {
+   case VK_IMAGE_TILING_OPTIMAL:
+      return isl_drm_modifier_get_info(I915_FORMAT_MOD_X_TILED);
+   case VK_IMAGE_TILING_LINEAR:
+      return isl_drm_modifier_get_info(DRM_FORMAT_MOD_LINEAR);
+   default:
+      unreachable("bad VkImageTiling");
+   }
+}
+
 VkResult
 anv_image_create(VkDevice _device,
                  const struct anv_image_create_info *create_info,
@@ -426,11 +516,17 @@ anv_image_create(VkDevice _device,
 {
    ANV_FROM_HANDLE(anv_device, device, _device);
    const VkImageCreateInfo *pCreateInfo = create_info->vk_info;
+   const struct isl_drm_modifier_info *isl_mod_info = NULL;
    struct anv_image *image = NULL;
    VkResult r;
 
    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO);
 
+   const struct wsi_image_create_info *wsi_info =
+      vk_find_struct_const(pCreateInfo->pNext, WSI_IMAGE_CREATE_INFO_MESA);
+   if (wsi_info && wsi_info->scanout)
+      isl_mod_info = get_legacy_scanout_drm_format_mod(pCreateInfo->tiling);
+
    anv_assert(pCreateInfo->mipLevels > 0);
    anv_assert(pCreateInfo->arrayLayers > 0);
    anv_assert(pCreateInfo->samples > 0);
@@ -446,17 +542,29 @@ anv_image_create(VkDevice _device,
    image->type = pCreateInfo->imageType;
    image->extent = pCreateInfo->extent;
    image->vk_format = pCreateInfo->format;
+   image->format = anv_get_format(pCreateInfo->format);
    image->aspects = vk_format_aspects(image->vk_format);
    image->levels = pCreateInfo->mipLevels;
    image->array_size = pCreateInfo->arrayLayers;
    image->samples = pCreateInfo->samples;
    image->usage = pCreateInfo->usage;
    image->tiling = pCreateInfo->tiling;
-   image->aux_usage = ISL_AUX_USAGE_NONE;
+   image->disjoint = pCreateInfo->flags & VK_IMAGE_CREATE_DISJOINT_BIT_KHR;
+   image->drm_format_mod = isl_mod_info ? isl_mod_info->modifier :
+                                          DRM_FORMAT_MOD_INVALID;
+
+   const struct anv_format *format = anv_get_format(image->vk_format);
+   assert(format != NULL);
+
+   const isl_tiling_flags_t isl_tiling_flags =
+      choose_isl_tiling_flags(create_info, isl_mod_info);
+
+   image->n_planes = format->n_planes;
 
    uint32_t b;
    for_each_bit(b, image->aspects) {
-      r = make_surface(device, image, create_info, (1 << b));
+      r = make_surface(device, image, create_info, isl_tiling_flags,
+                       (1 << b));
       if (r != VK_SUCCESS)
          goto fail;
    }
@@ -478,6 +586,15 @@ anv_CreateImage(VkDevice device,
                 const VkAllocationCallbacks *pAllocator,
                 VkImage *pImage)
 {
+#ifdef ANDROID
+   const VkNativeBufferANDROID *gralloc_info =
+      vk_find_struct_const(pCreateInfo->pNext, NATIVE_BUFFER_ANDROID);
+
+   if (gralloc_info)
+      return anv_image_from_gralloc(device, pCreateInfo, gralloc_info,
+                                    pAllocator, pImage);
+#endif
+
    return anv_image_create(device,
       &(struct anv_image_create_info) {
          .vk_info = pCreateInfo,
@@ -496,61 +613,106 @@ anv_DestroyImage(VkDevice _device, VkImage _image,
    if (!image)
       return;
 
+   for (uint32_t p = 0; p < image->n_planes; ++p) {
+      if (image->planes[p].bo_is_owned) {
+         assert(image->planes[p].bo != NULL);
+         anv_bo_cache_release(device, &device->bo_cache, image->planes[p].bo);
+      }
+   }
+
    vk_free2(&device->alloc, pAllocator, image);
 }
 
-static void
-anv_bind_image_memory(const VkBindImageMemoryInfoKHR *pBindInfo)
+static void anv_image_bind_memory_plane(struct anv_device *device,
+                                        struct anv_image *image,
+                                        uint32_t plane,
+                                        struct anv_device_memory *memory,
+                                        uint32_t memory_offset)
 {
-   ANV_FROM_HANDLE(anv_device_memory, mem, pBindInfo->memory);
-   ANV_FROM_HANDLE(anv_image, image, pBindInfo->image);
-
-   assert(pBindInfo->sType == VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO_KHR);
+   assert(!image->planes[plane].bo_is_owned);
 
-   if (mem == NULL) {
-      image->bo = NULL;
-      image->offset = 0;
+   if (!memory) {
+      image->planes[plane].bo = NULL;
+      image->planes[plane].bo_offset = 0;
       return;
    }
 
-   image->bo = mem->bo;
-   image->offset = pBindInfo->memoryOffset;
+   image->planes[plane].bo = memory->bo;
+   image->planes[plane].bo_offset = memory_offset;
 }
 
 VkResult anv_BindImageMemory(
-    VkDevice                                    device,
-    VkImage                                     image,
-    VkDeviceMemory                              memory,
+    VkDevice                                    _device,
+    VkImage                                     _image,
+    VkDeviceMemory                              _memory,
     VkDeviceSize                                memoryOffset)
 {
-   anv_bind_image_memory(
-      &(VkBindImageMemoryInfoKHR) {
-         .sType         = VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO_KHR,
-         .image         = image,
-         .memory        = memory,
-         .memoryOffset  = memoryOffset,
-      });
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
+   ANV_FROM_HANDLE(anv_image, image, _image);
+
+   uint32_t aspect_bit;
+   anv_foreach_image_aspect_bit(aspect_bit, image, image->aspects) {
+      uint32_t plane =
+         anv_image_aspect_to_plane(image->aspects, 1UL << aspect_bit);
+      anv_image_bind_memory_plane(device, image, plane, mem, memoryOffset);
+   }
 
    return VK_SUCCESS;
 }
 
 VkResult anv_BindImageMemory2KHR(
-    VkDevice                                    device,
+    VkDevice                                    _device,
     uint32_t                                    bindInfoCount,
     const VkBindImageMemoryInfoKHR*             pBindInfos)
 {
-   for (uint32_t i = 0; i < bindInfoCount; i++)
-      anv_bind_image_memory(&pBindInfos[i]);
+   ANV_FROM_HANDLE(anv_device, device, _device);
+
+   for (uint32_t i = 0; i < bindInfoCount; i++) {
+      const VkBindImageMemoryInfoKHR *bind_info = &pBindInfos[i];
+      ANV_FROM_HANDLE(anv_device_memory, mem, bind_info->memory);
+      ANV_FROM_HANDLE(anv_image, image, bind_info->image);
+      VkImageAspectFlags aspects = image->aspects;
+
+      vk_foreach_struct_const(s, bind_info->pNext) {
+         switch (s->sType) {
+         case VK_STRUCTURE_TYPE_BIND_IMAGE_PLANE_MEMORY_INFO_KHR: {
+            const VkBindImagePlaneMemoryInfoKHR *plane_info =
+               (const VkBindImagePlaneMemoryInfoKHR *) s;
+
+            aspects = plane_info->planeAspect;
+            break;
+         }
+         default:
+            anv_debug_ignored_stype(s->sType);
+            break;
+         }
+      }
+
+      uint32_t aspect_bit;
+      anv_foreach_image_aspect_bit(aspect_bit, image, aspects) {
+         uint32_t plane =
+            anv_image_aspect_to_plane(image->aspects, 1UL << aspect_bit);
+         anv_image_bind_memory_plane(device, image, plane,
+                                     mem, bind_info->memoryOffset);
+      }
+   }
 
    return VK_SUCCESS;
 }
 
-static void
-anv_surface_get_subresource_layout(struct anv_image *image,
-                                   struct anv_surface *surface,
-                                   const VkImageSubresource *subresource,
-                                   VkSubresourceLayout *layout)
+void anv_GetImageSubresourceLayout(
+    VkDevice                                    device,
+    VkImage                                     _image,
+    const VkImageSubresource*                   subresource,
+    VkSubresourceLayout*                        layout)
 {
+   ANV_FROM_HANDLE(anv_image, image, _image);
+   const struct anv_surface *surface =
+      get_surface(image, subresource->aspectMask);
+
+   assert(__builtin_popcount(subresource->aspectMask) == 1);
+
    /* If we are on a non-zero mip level or array slice, we need to
     * calculate a real offset.
     */
@@ -564,34 +726,6 @@ anv_surface_get_subresource_layout(struct anv_image *image,
    layout->size = surface->isl.size;
 }
 
-void anv_GetImageSubresourceLayout(
-    VkDevice                                    device,
-    VkImage                                     _image,
-    const VkImageSubresource*                   pSubresource,
-    VkSubresourceLayout*                        pLayout)
-{
-   ANV_FROM_HANDLE(anv_image, image, _image);
-
-   assert(__builtin_popcount(pSubresource->aspectMask) == 1);
-
-   switch (pSubresource->aspectMask) {
-   case VK_IMAGE_ASPECT_COLOR_BIT:
-      anv_surface_get_subresource_layout(image, &image->color_surface,
-                                         pSubresource, pLayout);
-      break;
-   case VK_IMAGE_ASPECT_DEPTH_BIT:
-      anv_surface_get_subresource_layout(image, &image->depth_surface,
-                                         pSubresource, pLayout);
-      break;
-   case VK_IMAGE_ASPECT_STENCIL_BIT:
-      anv_surface_get_subresource_layout(image, &image->stencil_surface,
-                                         pSubresource, pLayout);
-      break;
-   default:
-      assert(!"Invalid image aspect");
-   }
-}
-
 /**
  * This function determines the optimal buffer to use for a given
  * VkImageLayout and other pieces of information needed to make that
@@ -600,7 +734,7 @@ void anv_GetImageSubresourceLayout(
  *
  * @param devinfo The device information of the Intel GPU.
  * @param image The image that may contain a collection of buffers.
- * @param aspects The aspect(s) of the image to be accessed.
+ * @param aspect The aspect of the image to be accessed.
  * @param layout The current layout of the image aspect(s).
  *
  * @return The primary buffer that should be used for the given layout.
@@ -624,26 +758,19 @@ anv_layout_to_aux_usage(const struct gen_device_info * const devinfo,
 
    /* Determine the optimal buffer. */
 
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+
    /* If there is no auxiliary surface allocated, we must use the one and only
     * main buffer.
     */
-   if (image->aux_surface.isl.size == 0)
+   if (image->planes[plane].aux_surface.isl.size == 0)
       return ISL_AUX_USAGE_NONE;
 
    /* All images that use an auxiliary surface are required to be tiled. */
    assert(image->tiling == VK_IMAGE_TILING_OPTIMAL);
 
    /* Stencil has no aux */
-   if (aspect == VK_IMAGE_ASPECT_STENCIL_BIT)
-      return ISL_AUX_USAGE_NONE;
-
-   const bool color_aspect = aspect == VK_IMAGE_ASPECT_COLOR_BIT;
-
-   /* The following switch currently only handles depth stencil aspects.
-    * TODO: Handle the color aspect.
-    */
-   if (color_aspect)
-      return image->aux_usage;
+   assert(aspect != VK_IMAGE_ASPECT_STENCIL_BIT);
 
    switch (layout) {
 
@@ -664,31 +791,41 @@ anv_layout_to_aux_usage(const struct gen_device_info * const devinfo,
 
 
    /* Transfer Layouts
-    *
-    * This buffer could be a depth buffer used in a transfer operation. BLORP
-    * currently doesn't use HiZ for transfer operations so we must use the main
-    * buffer for this layout. TODO: Enable HiZ in BLORP.
     */
    case VK_IMAGE_LAYOUT_GENERAL:
    case VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL:
    case VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL:
-      return ISL_AUX_USAGE_NONE;
+      if (aspect == VK_IMAGE_ASPECT_DEPTH_BIT) {
+         /* This buffer could be a depth buffer used in a transfer operation.
+          * BLORP currently doesn't use HiZ for transfer operations so we must
+          * use the main buffer for this layout. TODO: Enable HiZ in BLORP.
+          */
+         assert(image->planes[plane].aux_usage == ISL_AUX_USAGE_HIZ);
+         return ISL_AUX_USAGE_NONE;
+      } else {
+         assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
+         return image->planes[plane].aux_usage;
+      }
 
 
    /* Sampling Layouts */
    case VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL:
-      assert(!color_aspect);
+   case VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL_KHR:
+      assert((image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) == 0);
       /* Fall-through */
    case VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL:
-   case VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL_KHR:
-      assert(aspect == VK_IMAGE_ASPECT_DEPTH_BIT);
-      if (anv_can_sample_with_hiz(devinfo, image))
-         return ISL_AUX_USAGE_HIZ;
-      else
-         return ISL_AUX_USAGE_NONE;
+      if (aspect == VK_IMAGE_ASPECT_DEPTH_BIT) {
+         if (anv_can_sample_with_hiz(devinfo, image))
+            return ISL_AUX_USAGE_HIZ;
+         else
+            return ISL_AUX_USAGE_NONE;
+      } else {
+         return image->planes[plane].aux_usage;
+      }
+
 
    case VK_IMAGE_LAYOUT_PRESENT_SRC_KHR:
-      assert(color_aspect);
+      assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
 
       /* On SKL+, the render buffer can be decompressed by the presentation
        * engine. Support for this feature has not yet landed in the wider
@@ -710,8 +847,14 @@ anv_layout_to_aux_usage(const struct gen_device_info * const devinfo,
 
    /* Rendering Layouts */
    case VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL:
-      assert(color_aspect);
-      unreachable("Color images are not yet supported.");
+      assert(aspect & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
+      if (image->planes[plane].aux_usage == ISL_AUX_USAGE_NONE) {
+         assert(image->samples == 1);
+         return ISL_AUX_USAGE_CCS_D;
+      } else {
+         assert(image->planes[plane].aux_usage != ISL_AUX_USAGE_CCS_D);
+         return image->planes[plane].aux_usage;
+      }
 
    case VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL:
    case VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL_KHR:
@@ -728,6 +871,73 @@ anv_layout_to_aux_usage(const struct gen_device_info * const devinfo,
    unreachable("layout is not a VkImageLayout enumeration member.");
 }
 
+/**
+ * This function returns the level of unresolved fast-clear support of the
+ * given image in the given VkImageLayout.
+ *
+ * @param devinfo The device information of the Intel GPU.
+ * @param image The image that may contain a collection of buffers.
+ * @param aspect The aspect of the image to be accessed.
+ * @param layout The current layout of the image aspect(s).
+ */
+enum anv_fast_clear_type
+anv_layout_to_fast_clear_type(const struct gen_device_info * const devinfo,
+                              const struct anv_image * const image,
+                              const VkImageAspectFlagBits aspect,
+                              const VkImageLayout layout)
+{
+   /* The aspect must be exactly one of the image aspects. */
+   assert(_mesa_bitcount(aspect) == 1 && (aspect & image->aspects));
+
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+
+   /* If there is no auxiliary surface allocated, there are no fast-clears */
+   if (image->planes[plane].aux_surface.isl.size == 0)
+      return ANV_FAST_CLEAR_NONE;
+
+   /* All images that use an auxiliary surface are required to be tiled. */
+   assert(image->tiling == VK_IMAGE_TILING_OPTIMAL);
+
+   /* Stencil has no aux */
+   assert(aspect != VK_IMAGE_ASPECT_STENCIL_BIT);
+
+   if (aspect == VK_IMAGE_ASPECT_DEPTH_BIT) {
+      /* For depth images (with HiZ), the layout supports fast-clears if and
+       * only if it supports HiZ.  However, we only support fast-clears to the
+       * default depth value.
+       */
+      enum isl_aux_usage aux_usage =
+         anv_layout_to_aux_usage(devinfo, image, aspect, layout);
+      return aux_usage == ISL_AUX_USAGE_HIZ ?
+             ANV_FAST_CLEAR_DEFAULT_VALUE : ANV_FAST_CLEAR_NONE;
+   }
+
+   assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
+
+   /* Multisample fast-clear is not yet supported. */
+   if (image->samples > 1)
+      return ANV_FAST_CLEAR_NONE;
+
+   switch (layout) {
+   case VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL:
+      return ANV_FAST_CLEAR_ANY;
+
+   case VK_IMAGE_LAYOUT_PRESENT_SRC_KHR:
+      return ANV_FAST_CLEAR_NONE;
+
+   default:
+      /* If the image has CCS_E enabled all the time then we can use
+       * fast-clear as long as the clear color is the default value of zero
+       * since this is the default value we program into every surface state
+       * used for texturing.
+       */
+      if (image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E)
+         return ANV_FAST_CLEAR_DEFAULT_VALUE;
+      else
+         return ANV_FAST_CLEAR_NONE;
+   }
+}
+
 
 static struct anv_state
 alloc_surface_state(struct anv_device *device)
@@ -766,8 +976,10 @@ anv_image_fill_surface_state(struct anv_device *device,
                              struct anv_surface_state *state_inout,
                              struct brw_image_param *image_param_out)
 {
-   const struct anv_surface *surface =
-      anv_image_get_surface_for_aspect_mask(image, aspect);
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+
+   const struct anv_surface *surface = &image->planes[plane].surface,
+      *aux_surface = &image->planes[plane].aux_surface;
 
    struct isl_view view = *view_in;
    view.usage |= view_usage;
@@ -777,13 +989,13 @@ anv_image_fill_surface_state(struct anv_device *device,
     * the primary surface.  The shadow surface will be tiled, unlike the main
     * surface, so it should get significantly better performance.
     */
-   if (image->shadow_surface.isl.size > 0 &&
+   if (image->planes[plane].shadow_surface.isl.size > 0 &&
        isl_format_is_compressed(view.format) &&
        (flags & ANV_IMAGE_VIEW_STATE_TEXTURE_OPTIMAL)) {
       assert(isl_format_is_compressed(surface->isl.format));
       assert(surface->isl.tiling == ISL_TILING_LINEAR);
-      assert(image->shadow_surface.isl.tiling != ISL_TILING_LINEAR);
-      surface = &image->shadow_surface;
+      assert(image->planes[plane].shadow_surface.isl.tiling != ISL_TILING_LINEAR);
+      surface = &image->planes[plane].shadow_surface;
    }
 
    if (view_usage == ISL_SURF_USAGE_RENDER_TARGET_BIT)
@@ -798,9 +1010,9 @@ anv_image_fill_surface_state(struct anv_device *device,
    if (!clear_color)
       clear_color = &default_clear_color;
 
-   const uint64_t address = image->offset + surface->offset;
-   const uint64_t aux_address = (aux_usage == ISL_AUX_USAGE_NONE) ? 0 :
-                                image->offset + image->aux_surface.offset;
+   const uint64_t address = image->planes[plane].bo_offset + surface->offset;
+   const uint64_t aux_address = aux_usage == ISL_AUX_USAGE_NONE ?
+      0 : (image->planes[plane].bo_offset + aux_surface->offset);
 
    if (view_usage == ISL_SURF_USAGE_STORAGE_BIT &&
        !(flags & ANV_IMAGE_VIEW_STATE_STORAGE_WRITE_ONLY) &&
@@ -889,7 +1101,7 @@ anv_image_fill_surface_state(struct anv_device *device,
                           .view = &view,
                           .address = address + offset_B,
                           .clear_color = *clear_color,
-                          .aux_surf = &image->aux_surface.isl,
+                          .aux_surf = &aux_surface->isl,
                           .aux_usage = aux_usage,
                           .aux_address = aux_address,
                           .mocs = device->default_mocs,
@@ -920,6 +1132,22 @@ anv_image_fill_surface_state(struct anv_device *device,
    }
 }
 
+static VkImageAspectFlags
+remap_aspect_flags(VkImageAspectFlags view_aspects)
+{
+   if (view_aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
+      if (_mesa_bitcount(view_aspects) == 1)
+         return VK_IMAGE_ASPECT_COLOR_BIT;
+
+      VkImageAspectFlags color_aspects = 0;
+      for (uint32_t i = 0; i < _mesa_bitcount(view_aspects); i++)
+         color_aspects |= VK_IMAGE_ASPECT_PLANE_0_BIT_KHR << i;
+      return color_aspects;
+   }
+   /* No special remapping needed for depth & stencil aspects. */
+   return view_aspects;
+}
+
 VkResult
 anv_CreateImageView(VkDevice _device,
                     const VkImageViewCreateInfo *pCreateInfo,
@@ -964,106 +1192,130 @@ anv_CreateImageView(VkDevice _device,
       break;
    }
 
+   /* First expand aspects to the image's ones (for example
+    * VK_IMAGE_ASPECT_COLOR_BIT will be converted to
+    * VK_IMAGE_ASPECT_PLANE_0_BIT_KHR | VK_IMAGE_ASPECT_PLANE_1_BIT_KHR |
+    * VK_IMAGE_ASPECT_PLANE_2_BIT_KHR for an image of format
+    * VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM_KHR.
+    */
+   VkImageAspectFlags expanded_aspects =
+      anv_image_expand_aspects(image, range->aspectMask);
+
    iview->image = image;
 
-   iview->aspect_mask = pCreateInfo->subresourceRange.aspectMask;
+   /* Remap the expanded aspects for the image view. For example if only
+    * VK_IMAGE_ASPECT_PLANE_1_BIT_KHR was given in range->aspectMask, we will
+    * convert it to VK_IMAGE_ASPECT_COLOR_BIT since from the point of view of
+    * the image view, it only has a single plane.
+    */
+   iview->aspect_mask = remap_aspect_flags(expanded_aspects);
+   iview->n_planes = anv_image_aspect_get_planes(iview->aspect_mask);
    iview->vk_format = pCreateInfo->format;
 
-   struct anv_format_plane format =
-      anv_get_format_plane(&device->info, pCreateInfo->format,
-                           range->aspectMask == (VK_IMAGE_ASPECT_DEPTH_BIT |
-                                                 VK_IMAGE_ASPECT_STENCIL_BIT) ?
-                           VK_IMAGE_ASPECT_DEPTH_BIT : range->aspectMask,
-                           image->tiling);
-
-   iview->isl = (struct isl_view) {
-      .format = format.isl_format,
-      .base_level = range->baseMipLevel,
-      .levels = anv_get_levelCount(image, range),
-      .base_array_layer = range->baseArrayLayer,
-      .array_len = anv_get_layerCount(image, range),
-      .swizzle = {
-         .r = remap_swizzle(pCreateInfo->components.r,
-                            VK_COMPONENT_SWIZZLE_R, format.swizzle),
-         .g = remap_swizzle(pCreateInfo->components.g,
-                            VK_COMPONENT_SWIZZLE_G, format.swizzle),
-         .b = remap_swizzle(pCreateInfo->components.b,
-                            VK_COMPONENT_SWIZZLE_B, format.swizzle),
-         .a = remap_swizzle(pCreateInfo->components.a,
-                            VK_COMPONENT_SWIZZLE_A, format.swizzle),
-      },
-   };
-
    iview->extent = (VkExtent3D) {
       .width  = anv_minify(image->extent.width , range->baseMipLevel),
       .height = anv_minify(image->extent.height, range->baseMipLevel),
       .depth  = anv_minify(image->extent.depth , range->baseMipLevel),
    };
 
-   if (pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_3D) {
-      iview->isl.base_array_layer = 0;
-      iview->isl.array_len = iview->extent.depth;
-   }
+   /* Now go through the underlying image selected planes (computed in
+    * expanded_aspects) and map them to planes in the image view.
+    */
+   uint32_t iaspect_bit, vplane = 0;
+   anv_foreach_image_aspect_bit(iaspect_bit, image, expanded_aspects) {
+      uint32_t iplane =
+         anv_image_aspect_to_plane(expanded_aspects, 1UL << iaspect_bit);
+      VkImageAspectFlags vplane_aspect =
+         anv_plane_to_aspect(iview->aspect_mask, vplane);
+      struct anv_format_plane format =
+         anv_get_format_plane(&device->info, pCreateInfo->format,
+                              vplane_aspect, image->tiling);
+
+      iview->planes[vplane].image_plane = iplane;
+
+      iview->planes[vplane].isl = (struct isl_view) {
+         .format = format.isl_format,
+         .base_level = range->baseMipLevel,
+         .levels = anv_get_levelCount(image, range),
+         .base_array_layer = range->baseArrayLayer,
+         .array_len = anv_get_layerCount(image, range),
+         .swizzle = {
+            .r = remap_swizzle(pCreateInfo->components.r,
+                               VK_COMPONENT_SWIZZLE_R, format.swizzle),
+            .g = remap_swizzle(pCreateInfo->components.g,
+                               VK_COMPONENT_SWIZZLE_G, format.swizzle),
+            .b = remap_swizzle(pCreateInfo->components.b,
+                               VK_COMPONENT_SWIZZLE_B, format.swizzle),
+            .a = remap_swizzle(pCreateInfo->components.a,
+                               VK_COMPONENT_SWIZZLE_A, format.swizzle),
+         },
+      };
+
+      if (pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_3D) {
+         iview->planes[vplane].isl.base_array_layer = 0;
+         iview->planes[vplane].isl.array_len = iview->extent.depth;
+      }
 
-   if (pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_CUBE ||
-       pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY) {
-      iview->isl.usage = ISL_SURF_USAGE_CUBE_BIT;
-   } else {
-      iview->isl.usage = 0;
-   }
+      if (pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_CUBE ||
+          pCreateInfo->viewType == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY) {
+         iview->planes[vplane].isl.usage = ISL_SURF_USAGE_CUBE_BIT;
+      } else {
+         iview->planes[vplane].isl.usage = 0;
+      }
 
-   /* Input attachment surfaces for color are allocated and filled
-    * out at BeginRenderPass time because they need compression information.
-    * Compression is not yet enabled for depth textures and stencil doesn't
-    * allow compression so we can just use the texture surface state from the
-    * view.
-    */
-   if (view_usage & VK_IMAGE_USAGE_SAMPLED_BIT ||
-       (view_usage & VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT &&
-        !(iview->aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT))) {
-      iview->optimal_sampler_surface_state.state = alloc_surface_state(device);
-      iview->general_sampler_surface_state.state = alloc_surface_state(device);
-
-      enum isl_aux_usage general_aux_usage =
-         anv_layout_to_aux_usage(&device->info, image, iview->aspect_mask,
-                                 VK_IMAGE_LAYOUT_GENERAL);
-      enum isl_aux_usage optimal_aux_usage =
-         anv_layout_to_aux_usage(&device->info, image, iview->aspect_mask,
-                                 VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
-
-      anv_image_fill_surface_state(device, image, iview->aspect_mask,
-                                   &iview->isl, ISL_SURF_USAGE_TEXTURE_BIT,
-                                   optimal_aux_usage, NULL,
-                                   ANV_IMAGE_VIEW_STATE_TEXTURE_OPTIMAL,
-                                   &iview->optimal_sampler_surface_state,
-                                   NULL);
-
-      anv_image_fill_surface_state(device, image, iview->aspect_mask,
-                                   &iview->isl, ISL_SURF_USAGE_TEXTURE_BIT,
-                                   general_aux_usage, NULL,
-                                   0,
-                                   &iview->general_sampler_surface_state,
-                                   NULL);
-   }
+      if (view_usage & VK_IMAGE_USAGE_SAMPLED_BIT ||
+          (view_usage & VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT &&
+           !(iview->aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT))) {
+         iview->planes[vplane].optimal_sampler_surface_state.state = alloc_surface_state(device);
+         iview->planes[vplane].general_sampler_surface_state.state = alloc_surface_state(device);
+
+         enum isl_aux_usage general_aux_usage =
+            anv_layout_to_aux_usage(&device->info, image, 1UL << iaspect_bit,
+                                    VK_IMAGE_LAYOUT_GENERAL);
+         enum isl_aux_usage optimal_aux_usage =
+            anv_layout_to_aux_usage(&device->info, image, 1UL << iaspect_bit,
+                                    VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
+
+         anv_image_fill_surface_state(device, image, 1ULL << iaspect_bit,
+                                      &iview->planes[vplane].isl,
+                                      ISL_SURF_USAGE_TEXTURE_BIT,
+                                      optimal_aux_usage, NULL,
+                                      ANV_IMAGE_VIEW_STATE_TEXTURE_OPTIMAL,
+                                      &iview->planes[vplane].optimal_sampler_surface_state,
+                                      NULL);
+
+         anv_image_fill_surface_state(device, image, 1ULL << iaspect_bit,
+                                      &iview->planes[vplane].isl,
+                                      ISL_SURF_USAGE_TEXTURE_BIT,
+                                      general_aux_usage, NULL,
+                                      0,
+                                      &iview->planes[vplane].general_sampler_surface_state,
+                                      NULL);
+      }
 
-   /* NOTE: This one needs to go last since it may stomp isl_view.format */
-   if (view_usage & VK_IMAGE_USAGE_STORAGE_BIT) {
-      iview->storage_surface_state.state = alloc_surface_state(device);
-      iview->writeonly_storage_surface_state.state = alloc_surface_state(device);
-
-      anv_image_fill_surface_state(device, image, iview->aspect_mask,
-                                   &iview->isl, ISL_SURF_USAGE_STORAGE_BIT,
-                                   ISL_AUX_USAGE_NONE, NULL,
-                                   0,
-                                   &iview->storage_surface_state,
-                                   &iview->storage_image_param);
-
-      anv_image_fill_surface_state(device, image, iview->aspect_mask,
-                                   &iview->isl, ISL_SURF_USAGE_STORAGE_BIT,
-                                   ISL_AUX_USAGE_NONE, NULL,
-                                   ANV_IMAGE_VIEW_STATE_STORAGE_WRITE_ONLY,
-                                   &iview->writeonly_storage_surface_state,
-                                   NULL);
+      /* NOTE: This one needs to go last since it may stomp isl_view.format */
+      if (view_usage & VK_IMAGE_USAGE_STORAGE_BIT) {
+         iview->planes[vplane].storage_surface_state.state = alloc_surface_state(device);
+         iview->planes[vplane].writeonly_storage_surface_state.state = alloc_surface_state(device);
+
+         anv_image_fill_surface_state(device, image, 1ULL << iaspect_bit,
+                                      &iview->planes[vplane].isl,
+                                      ISL_SURF_USAGE_STORAGE_BIT,
+                                      ISL_AUX_USAGE_NONE, NULL,
+                                      0,
+                                      &iview->planes[vplane].storage_surface_state,
+                                      &iview->planes[vplane].storage_image_param);
+
+         anv_image_fill_surface_state(device, image, 1ULL << iaspect_bit,
+                                      &iview->planes[vplane].isl,
+                                      ISL_SURF_USAGE_STORAGE_BIT,
+                                      ISL_AUX_USAGE_NONE, NULL,
+                                      ANV_IMAGE_VIEW_STATE_STORAGE_WRITE_ONLY,
+                                      &iview->planes[vplane].writeonly_storage_surface_state,
+                                      NULL);
+      }
+
+      vplane++;
    }
 
    *pView = anv_image_view_to_handle(iview);
@@ -1081,24 +1333,26 @@ anv_DestroyImageView(VkDevice _device, VkImageView _iview,
    if (!iview)
       return;
 
-   if (iview->optimal_sampler_surface_state.state.alloc_size > 0) {
-      anv_state_pool_free(&device->surface_state_pool,
-                          iview->optimal_sampler_surface_state.state);
-   }
+   for (uint32_t plane = 0; plane < iview->n_planes; plane++) {
+      if (iview->planes[plane].optimal_sampler_surface_state.state.alloc_size > 0) {
+         anv_state_pool_free(&device->surface_state_pool,
+                             iview->planes[plane].optimal_sampler_surface_state.state);
+      }
 
-   if (iview->general_sampler_surface_state.state.alloc_size > 0) {
-      anv_state_pool_free(&device->surface_state_pool,
-                          iview->general_sampler_surface_state.state);
-   }
+      if (iview->planes[plane].general_sampler_surface_state.state.alloc_size > 0) {
+         anv_state_pool_free(&device->surface_state_pool,
+                             iview->planes[plane].general_sampler_surface_state.state);
+      }
 
-   if (iview->storage_surface_state.state.alloc_size > 0) {
-      anv_state_pool_free(&device->surface_state_pool,
-                          iview->storage_surface_state.state);
-   }
+      if (iview->planes[plane].storage_surface_state.state.alloc_size > 0) {
+         anv_state_pool_free(&device->surface_state_pool,
+                             iview->planes[plane].storage_surface_state.state);
+      }
 
-   if (iview->writeonly_storage_surface_state.state.alloc_size > 0) {
-      anv_state_pool_free(&device->surface_state_pool,
-                          iview->writeonly_storage_surface_state.state);
+      if (iview->planes[plane].writeonly_storage_surface_state.state.alloc_size > 0) {
+         anv_state_pool_free(&device->surface_state_pool,
+                             iview->planes[plane].writeonly_storage_surface_state.state);
+      }
    }
 
    vk_free2(&device->alloc, pAllocator, iview);
@@ -1206,16 +1460,21 @@ const struct anv_surface *
 anv_image_get_surface_for_aspect_mask(const struct anv_image *image,
                                       VkImageAspectFlags aspect_mask)
 {
+   VkImageAspectFlags sanitized_mask;
+
    switch (aspect_mask) {
    case VK_IMAGE_ASPECT_COLOR_BIT:
       assert(image->aspects == VK_IMAGE_ASPECT_COLOR_BIT);
-      return &image->color_surface;
+      sanitized_mask = VK_IMAGE_ASPECT_COLOR_BIT;
+      break;
    case VK_IMAGE_ASPECT_DEPTH_BIT:
       assert(image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT);
-      return &image->depth_surface;
+      sanitized_mask = VK_IMAGE_ASPECT_DEPTH_BIT;
+      break;
    case VK_IMAGE_ASPECT_STENCIL_BIT:
       assert(image->aspects & VK_IMAGE_ASPECT_STENCIL_BIT);
-      return &image->stencil_surface;
+      sanitized_mask = VK_IMAGE_ASPECT_STENCIL_BIT;
+      break;
    case VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT:
       /* FINISHME: The Vulkan spec (git a511ba2) requires support for
        * combined depth stencil formats. Specifically, it states:
@@ -1229,13 +1488,29 @@ anv_image_get_surface_for_aspect_mask(const struct anv_image *image,
        * stencil surfaces from the underlying surface.
        */
       if (image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT) {
-         return &image->depth_surface;
+         sanitized_mask = VK_IMAGE_ASPECT_DEPTH_BIT;
       } else {
          assert(image->aspects == VK_IMAGE_ASPECT_STENCIL_BIT);
-         return &image->stencil_surface;
+         sanitized_mask = VK_IMAGE_ASPECT_STENCIL_BIT;
       }
-    default:
+      break;
+   case VK_IMAGE_ASPECT_PLANE_0_BIT_KHR:
+      assert((image->aspects & ~VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) == 0);
+      sanitized_mask = VK_IMAGE_ASPECT_PLANE_0_BIT_KHR;
+      break;
+   case VK_IMAGE_ASPECT_PLANE_1_BIT_KHR:
+      assert((image->aspects & ~VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) == 0);
+      sanitized_mask = VK_IMAGE_ASPECT_PLANE_1_BIT_KHR;
+      break;
+   case VK_IMAGE_ASPECT_PLANE_2_BIT_KHR:
+      assert((image->aspects & ~VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) == 0);
+      sanitized_mask = VK_IMAGE_ASPECT_PLANE_2_BIT_KHR;
+      break;
+   default:
        unreachable("image does not have aspect");
        return NULL;
    }
+
+   uint32_t plane = anv_image_aspect_to_plane(image->aspects, sanitized_mask);
+   return &image->planes[plane].surface;
 }