if (is_depth) {
surface->flags |= RADEON_SURF_ZBUFFER;
+ if (!(pCreateInfo->usage & VK_IMAGE_USAGE_STORAGE_BIT) &&
+ !(pCreateInfo->flags & (VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT |
+ VK_IMAGE_CREATE_EXTENDED_USAGE_BIT_KHR)) &&
+ pCreateInfo->tiling != VK_IMAGE_TILING_LINEAR &&
+ pCreateInfo->mipLevels <= 1 &&
+ device->physical_device->rad_info.chip_class >= VI &&
+ ((pCreateInfo->format == VK_FORMAT_D32_SFLOAT ||
+ pCreateInfo->format == VK_FORMAT_D32_SFLOAT_S8_UINT) ||
+ (device->physical_device->rad_info.chip_class >= GFX9 &&
+ pCreateInfo->format == VK_FORMAT_D16_UNORM)))
+ surface->flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
}
if (is_stencil)
surface->flags |= RADEON_SURF_OPTIMIZE_FOR_SPACE;
- bool dcc_compatible_formats = !radv_is_colorbuffer_format_supported(pCreateInfo->format, &blendable);
+ bool dcc_compatible_formats = radv_is_colorbuffer_format_supported(pCreateInfo->format, &blendable);
if (pCreateInfo->flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) {
const struct VkImageFormatListCreateInfoKHR *format_list =
(const struct VkImageFormatListCreateInfoKHR *)
vk_find_struct_const(pCreateInfo->pNext,
IMAGE_FORMAT_LIST_CREATE_INFO_KHR);
- if (format_list) {
+
+ /* We have to ignore the existence of the list if viewFormatCount = 0 */
+ if (format_list && format_list->viewFormatCount) {
/* compatibility is transitive, so we only need to check
* one format with everything else. */
for (unsigned i = 0; i < format_list->viewFormatCount; ++i) {
if ((pCreateInfo->usage & (VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
VK_IMAGE_USAGE_STORAGE_BIT)) ||
+ (pCreateInfo->flags & VK_IMAGE_CREATE_EXTENDED_USAGE_BIT_KHR) ||
!dcc_compatible_formats ||
(pCreateInfo->tiling == VK_IMAGE_TILING_LINEAR) ||
pCreateInfo->mipLevels > 1 || pCreateInfo->arrayLayers > 1 ||
device->physical_device->rad_info.chip_class < VI ||
- create_info->scanout || (device->debug_flags & RADV_DEBUG_NO_DCC))
+ create_info->scanout || (device->instance->debug_flags & RADV_DEBUG_NO_DCC) ||
+ pCreateInfo->samples >= 2)
surface->flags |= RADEON_SURF_DISABLE_DCC;
if (create_info->scanout)
surface->flags |= RADEON_SURF_SCANOUT;
return 0;
}
-#define ATI_VENDOR_ID 0x1002
+
static uint32_t si_get_bo_metadata_word1(struct radv_device *device)
{
return (ATI_VENDOR_ID << 16) | device->physical_device->rad_info.pci_id;
if (chip_class >= VI) {
state[6] &= C_008F28_COMPRESSION_EN;
state[7] = 0;
- if (image->surface.dcc_size && first_level < image->surface.num_dcc_levels) {
+ if (radv_vi_dcc_enabled(image, first_level)) {
meta_va = gpu_address + image->dcc_offset;
if (chip_class <= VI)
meta_va += base_level_info->dcc_offset;
+ } else if(image->tc_compatible_htile && image->surface.htile_size) {
+ meta_va = gpu_address + image->htile_offset;
+ }
+
+ if (meta_va) {
state[6] |= S_008F28_COMPRESSION_EN(1);
state[7] = meta_va >> 8;
state[7] |= image->surface.tile_swizzle;
data_format = 0;
}
+ /* S8 with Z32 HTILE needs a special format. */
+ if (device->physical_device->rad_info.chip_class >= GFX9 &&
+ vk_format == VK_FORMAT_S8_UINT &&
+ image->tc_compatible_htile)
+ data_format = V_008F14_IMG_DATA_FORMAT_S8_32;
+
type = radv_tex_dim(image->type, view_type, image->info.array_size, image->info.samples,
is_storage_image, device->physical_device->rad_info.chip_class >= GFX9);
if (type == V_008F1C_SQ_RSRC_IMG_1D_ARRAY) {
}
static void
-radv_image_alloc_dcc(struct radv_device *device,
- struct radv_image *image)
+radv_image_alloc_dcc(struct radv_image *image)
{
image->dcc_offset = align64(image->size, image->surface.dcc_alignment);
/* + 16 for storing the clear values + dcc pred */
}
static void
-radv_image_alloc_htile(struct radv_device *device,
- struct radv_image *image)
+radv_image_alloc_htile(struct radv_image *image)
{
- if ((device->debug_flags & RADV_DEBUG_NO_HIZ) || image->info.levels > 1) {
- image->surface.htile_size = 0;
- return;
- }
-
image->htile_offset = align64(image->size, image->surface.htile_alignment);
/* + 8 for storing the clear values */
image->alignment = align64(image->alignment, image->surface.htile_alignment);
}
+static inline bool
+radv_image_can_enable_dcc_or_cmask(struct radv_image *image)
+{
+ if (image->info.samples <= 1 &&
+ image->info.width * image->info.height <= 512 * 512) {
+ /* Do not enable CMASK or DCC for small surfaces where the cost
+ * of the eliminate pass can be higher than the benefit of fast
+ * clear. RadeonSI does this, but the image threshold is
+ * different.
+ */
+ return false;
+ }
+
+ return image->usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT &&
+ (image->exclusive || image->queue_family_mask == 1);
+}
+
+static inline bool
+radv_image_can_enable_dcc(struct radv_image *image)
+{
+ return radv_image_can_enable_dcc_or_cmask(image) &&
+ image->surface.dcc_size;
+}
+
+static inline bool
+radv_image_can_enable_cmask(struct radv_image *image)
+{
+ if (image->surface.bpe > 8 && image->info.samples == 1) {
+ /* Do not enable CMASK for non-MSAA images (fast color clear)
+ * because 128 bit formats are not supported, but FMASK might
+ * still be used.
+ */
+ return false;
+ }
+
+ return radv_image_can_enable_dcc_or_cmask(image) &&
+ image->info.levels == 1 &&
+ image->info.depth == 1 &&
+ !image->surface.is_linear;
+}
+
+static inline bool
+radv_image_can_enable_fmask(struct radv_image *image)
+{
+ return image->info.samples > 1 && vk_format_is_color(image->vk_format);
+}
+
+static inline bool
+radv_image_can_enable_htile(struct radv_image *image)
+{
+ return image->info.levels == 1 && vk_format_is_depth(image->vk_format);
+}
+
VkResult
radv_image_create(VkDevice _device,
const struct radv_image_create_info *create_info,
RADV_FROM_HANDLE(radv_device, device, _device);
const VkImageCreateInfo *pCreateInfo = create_info->vk_info;
struct radv_image *image = NULL;
- bool can_cmask_dcc = false;
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO);
radv_assert(pCreateInfo->mipLevels > 0);
radv_assert(pCreateInfo->extent.height > 0);
radv_assert(pCreateInfo->extent.depth > 0);
- image = vk_alloc2(&device->alloc, alloc, sizeof(*image), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ image = vk_zalloc2(&device->alloc, alloc, sizeof(*image), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!image)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
- memset(image, 0, sizeof(*image));
image->type = pCreateInfo->imageType;
image->info.width = pCreateInfo->extent.width;
image->info.height = pCreateInfo->extent.height;
image->size = image->surface.surf_size;
image->alignment = image->surface.surf_alignment;
- if (image->exclusive || image->queue_family_mask == 1)
- can_cmask_dcc = true;
-
- if ((pCreateInfo->usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) &&
- image->surface.dcc_size && can_cmask_dcc)
- radv_image_alloc_dcc(device, image);
- else
+ /* Try to enable DCC first. */
+ if (radv_image_can_enable_dcc(image)) {
+ radv_image_alloc_dcc(image);
+ } else {
+ /* When DCC cannot be enabled, try CMASK. */
image->surface.dcc_size = 0;
+ if (radv_image_can_enable_cmask(image)) {
+ radv_image_alloc_cmask(device, image);
+ }
+ }
- if ((pCreateInfo->usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) &&
- pCreateInfo->mipLevels == 1 &&
- !image->surface.dcc_size && image->info.depth == 1 && can_cmask_dcc &&
- !image->surface.is_linear)
- radv_image_alloc_cmask(device, image);
-
- if (image->info.samples > 1 && vk_format_is_color(pCreateInfo->format)) {
+ /* Try to enable FMASK for multisampled images. */
+ if (radv_image_can_enable_fmask(image)) {
radv_image_alloc_fmask(device, image);
- } else if (vk_format_is_depth(pCreateInfo->format)) {
-
- radv_image_alloc_htile(device, image);
+ } else {
+ /* Otherwise, try to enable HTILE for depth surfaces. */
+ if (radv_image_can_enable_htile(image) &&
+ !(device->instance->debug_flags & RADV_DEBUG_NO_HIZ)) {
+ radv_image_alloc_htile(image);
+ image->tc_compatible_htile = image->surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE;
+ } else {
+ image->surface.htile_size = 0;
+ }
}
if (pCreateInfo->flags & VK_IMAGE_CREATE_SPARSE_BINDING_BIT) {
bool is_stencil = iview->aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT;
uint32_t blk_w;
uint32_t *descriptor;
- uint32_t *fmask_descriptor;
uint32_t hw_level = 0;
if (is_storage_image) {
descriptor = iview->storage_descriptor;
- fmask_descriptor = iview->storage_fmask_descriptor;
} else {
descriptor = iview->descriptor;
- fmask_descriptor = iview->fmask_descriptor;
}
assert(image->surface.blk_w % vk_format_get_blockwidth(image->vk_format) == 0);
iview->extent.height,
iview->extent.depth,
descriptor,
- fmask_descriptor);
+ descriptor + 8);
const struct legacy_surf_level *base_level_info = NULL;
if (device->physical_device->rad_info.chip_class <= GFX9) {
VkImageLayout layout,
unsigned queue_mask)
{
+ if (image->surface.htile_size && image->tc_compatible_htile)
+ return layout != VK_IMAGE_LAYOUT_GENERAL;
+
return image->surface.htile_size &&
(layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL) &&
VkImageLayout layout,
unsigned queue_mask)
{
+ if (image->surface.htile_size && image->tc_compatible_htile)
+ return layout != VK_IMAGE_LAYOUT_GENERAL;
+
return image->surface.htile_size &&
(layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL) &&
const VkAllocationCallbacks *pAllocator,
VkImage *pImage)
{
+ const struct wsi_image_create_info *wsi_info =
+ vk_find_struct_const(pCreateInfo->pNext, WSI_IMAGE_CREATE_INFO_MESA);
+ bool scanout = wsi_info && wsi_info->scanout;
+
return radv_image_create(device,
&(struct radv_image_create_info) {
.vk_info = pCreateInfo,
- .scanout = false,
- },
+ .scanout = scanout,
+ },
pAllocator,
pImage);
}
if (image->type == VK_IMAGE_TYPE_3D)
pLayout->size *= u_minify(image->info.depth, level);
} else {
- pLayout->offset = surface->u.legacy.level[level].offset + surface->u.legacy.level[level].slice_size * layer;
+ pLayout->offset = surface->u.legacy.level[level].offset + (uint64_t)surface->u.legacy.level[level].slice_size_dw * 4 * layer;
pLayout->rowPitch = surface->u.legacy.level[level].nblk_x * surface->bpe;
- pLayout->arrayPitch = surface->u.legacy.level[level].slice_size;
- pLayout->depthPitch = surface->u.legacy.level[level].slice_size;
- pLayout->size = surface->u.legacy.level[level].slice_size;
+ pLayout->arrayPitch = (uint64_t)surface->u.legacy.level[level].slice_size_dw * 4;
+ pLayout->depthPitch = (uint64_t)surface->u.legacy.level[level].slice_size_dw * 4;
+ pLayout->size = (uint64_t)surface->u.legacy.level[level].slice_size_dw * 4;
if (image->type == VK_IMAGE_TYPE_3D)
pLayout->size *= u_minify(image->info.depth, level);
}