#include "vk_util.h"
#include "radv_radeon_winsys.h"
#include "sid.h"
-#include "gfx9d.h"
#include "util/debug.h"
#include "util/u_atomic.h"
+
static unsigned
radv_choose_tiling(struct radv_device *device,
const struct radv_image_create_info *create_info)
if (!vk_format_is_compressed(pCreateInfo->format) &&
!vk_format_is_depth_or_stencil(pCreateInfo->format)
- && device->physical_device->rad_info.chip_class <= VI) {
+ && device->physical_device->rad_info.chip_class <= GFX8) {
/* this causes hangs in some VK CTS tests on GFX9. */
/* Textures with a very small height are recommended to be linear. */
if (pCreateInfo->imageType == VK_IMAGE_TYPE_1D ||
return RADEON_SURF_MODE_2D;
}
+
+static bool
+radv_use_tc_compat_htile_for_image(struct radv_device *device,
+ const VkImageCreateInfo *pCreateInfo)
+{
+ /* TC-compat HTILE is only available for GFX8+. */
+ if (device->physical_device->rad_info.chip_class < GFX8)
+ return false;
+
+ if ((pCreateInfo->usage & VK_IMAGE_USAGE_STORAGE_BIT) ||
+ (pCreateInfo->flags & VK_IMAGE_CREATE_EXTENDED_USAGE_BIT))
+ return false;
+
+ /* TODO: Implement layout transitions with variable sample locations
+ * before enabling HTILE for depth/stencil images created with this
+ * flags because the depth decompress pass needs to know them.
+ */
+ if (pCreateInfo->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT)
+ return false;
+
+ if (pCreateInfo->tiling == VK_IMAGE_TILING_LINEAR)
+ return false;
+
+ if (pCreateInfo->mipLevels > 1)
+ return false;
+
+ /* FIXME: for some reason TC compat with 2/4/8 samples breaks some cts
+ * tests - disable for now */
+ if (pCreateInfo->samples >= 2 &&
+ pCreateInfo->format == VK_FORMAT_D32_SFLOAT_S8_UINT)
+ return false;
+
+ /* GFX9 supports both 32-bit and 16-bit depth surfaces, while GFX8 only
+ * supports 32-bit. Though, it's possible to enable TC-compat for
+ * 16-bit depth surfaces if no Z planes are compressed.
+ */
+ if (pCreateInfo->format != VK_FORMAT_D32_SFLOAT_S8_UINT &&
+ pCreateInfo->format != VK_FORMAT_D32_SFLOAT &&
+ pCreateInfo->format != VK_FORMAT_D16_UNORM)
+ return false;
+
+ 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);
+
+ /* 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->format != format_list->pViewFormats[i])
+ return false;
+ }
+ } else {
+ return false;
+ }
+ }
+
+ return true;
+}
+
+static bool
+radv_surface_has_scanout(struct radv_device *device, const struct radv_image_create_info *info)
+{
+ if (info->scanout)
+ return true;
+
+ if (!info->bo_metadata)
+ return false;
+
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ return info->bo_metadata->u.gfx9.swizzle_mode == 0 || info->bo_metadata->u.gfx9.swizzle_mode % 4 == 2;
+ } else {
+ return info->bo_metadata->u.legacy.scanout;
+ }
+}
+
+static bool
+radv_use_dcc_for_image(struct radv_device *device,
+ const struct radv_image *image,
+ const struct radv_image_create_info *create_info,
+ const VkImageCreateInfo *pCreateInfo)
+{
+ bool dcc_compatible_formats;
+ bool blendable;
+
+ /* DCC (Delta Color Compression) is only available for GFX8+. */
+ if (device->physical_device->rad_info.chip_class < GFX8)
+ return false;
+
+ if (device->instance->debug_flags & RADV_DEBUG_NO_DCC)
+ return false;
+
+ /* FIXME: DCC is broken for shareable images starting with GFX9 */
+ if (device->physical_device->rad_info.chip_class >= GFX9 &&
+ image->shareable)
+ return false;
+
+ /* TODO: Enable DCC for storage images. */
+ if ((pCreateInfo->usage & VK_IMAGE_USAGE_STORAGE_BIT) ||
+ (pCreateInfo->flags & VK_IMAGE_CREATE_EXTENDED_USAGE_BIT))
+ return false;
+
+ if (pCreateInfo->tiling == VK_IMAGE_TILING_LINEAR)
+ return false;
+
+ if (vk_format_is_subsampled(pCreateInfo->format) ||
+ vk_format_get_plane_count(pCreateInfo->format) > 1)
+ return false;
+
+ /* TODO: Enable DCC for mipmaps and array layers. */
+ if (pCreateInfo->mipLevels > 1 || pCreateInfo->arrayLayers > 1)
+ return false;
+
+ if (radv_surface_has_scanout(device, create_info))
+ return false;
+
+ /* FIXME: DCC for MSAA with 4x and 8x samples doesn't work yet, while
+ * 2x can be enabled with an option.
+ */
+ if (pCreateInfo->samples > 2 ||
+ (pCreateInfo->samples == 2 &&
+ !device->physical_device->dcc_msaa_allowed))
+ return false;
+
+ /* Determine if the formats are DCC compatible. */
+ 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);
+
+ /* 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 (!radv_dcc_formats_compatible(pCreateInfo->format,
+ format_list->pViewFormats[i]))
+ dcc_compatible_formats = false;
+ }
+ } else {
+ dcc_compatible_formats = false;
+ }
+ }
+
+ if (!dcc_compatible_formats)
+ return false;
+
+ return true;
+}
+
+static void
+radv_prefill_surface_from_metadata(struct radv_device *device,
+ struct radeon_surf *surface,
+ const struct radv_image_create_info *create_info)
+{
+ const struct radeon_bo_metadata *md = create_info->bo_metadata;
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ if (md->u.gfx9.swizzle_mode > 0)
+ surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
+ else
+ surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE);
+
+ surface->u.gfx9.surf.swizzle_mode = md->u.gfx9.swizzle_mode;
+ } else {
+ surface->u.legacy.pipe_config = md->u.legacy.pipe_config;
+ surface->u.legacy.bankw = md->u.legacy.bankw;
+ surface->u.legacy.bankh = md->u.legacy.bankh;
+ surface->u.legacy.tile_split = md->u.legacy.tile_split;
+ surface->u.legacy.mtilea = md->u.legacy.mtilea;
+ surface->u.legacy.num_banks = md->u.legacy.num_banks;
+
+ if (md->u.legacy.macrotile == RADEON_LAYOUT_TILED)
+ surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
+ else if (md->u.legacy.microtile == RADEON_LAYOUT_TILED)
+ surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_1D, MODE);
+ else
+ surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE);
+
+ }
+}
+
static int
radv_init_surface(struct radv_device *device,
+ const struct radv_image *image,
struct radeon_surf *surface,
+ unsigned plane_id,
const struct radv_image_create_info *create_info)
{
const VkImageCreateInfo *pCreateInfo = create_info->vk_info;
unsigned array_mode = radv_choose_tiling(device, create_info);
- const struct vk_format_description *desc =
- vk_format_description(pCreateInfo->format);
- bool is_depth, is_stencil, blendable;
+ VkFormat format = vk_format_get_plane_format(pCreateInfo->format, plane_id);
+ const struct vk_format_description *desc = vk_format_description(format);
+ bool is_depth, is_stencil;
is_depth = vk_format_has_depth(desc);
is_stencil = vk_format_has_stencil(desc);
- surface->blk_w = vk_format_get_blockwidth(pCreateInfo->format);
- surface->blk_h = vk_format_get_blockheight(pCreateInfo->format);
+ surface->blk_w = vk_format_get_blockwidth(format);
+ surface->blk_h = vk_format_get_blockheight(format);
- surface->bpe = vk_format_get_blocksize(vk_format_depth_only(pCreateInfo->format));
+ surface->bpe = vk_format_get_blocksize(vk_format_depth_only(format));
/* align byte per element on dword */
if (surface->bpe == 3) {
surface->bpe = 4;
}
- surface->flags = RADEON_SURF_SET(array_mode, MODE);
+ if (create_info->bo_metadata) {
+ radv_prefill_surface_from_metadata(device, surface, create_info);
+ } else {
+ surface->flags = RADEON_SURF_SET(array_mode, MODE);
+ }
switch (pCreateInfo->imageType){
case VK_IMAGE_TYPE_1D:
if (is_depth) {
surface->flags |= RADEON_SURF_ZBUFFER;
+ if (radv_use_tc_compat_htile_for_image(device, pCreateInfo))
+ surface->flags |= RADEON_SURF_TC_COMPATIBLE_HTILE;
}
if (is_stencil)
surface->flags |= RADEON_SURF_SBUFFER;
+ if (device->physical_device->rad_info.chip_class >= GFX9 &&
+ pCreateInfo->imageType == VK_IMAGE_TYPE_3D &&
+ vk_format_get_blocksizebits(pCreateInfo->format) == 128 &&
+ vk_format_is_compressed(pCreateInfo->format))
+ surface->flags |= RADEON_SURF_NO_RENDER_TARGET;
+
surface->flags |= RADEON_SURF_OPTIMIZE_FOR_SPACE;
- if ((pCreateInfo->usage & (VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
- VK_IMAGE_USAGE_STORAGE_BIT)) ||
- (pCreateInfo->flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) ||
- (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) ||
- !radv_is_colorbuffer_format_supported(pCreateInfo->format, &blendable))
+ if (!radv_use_dcc_for_image(device, image, create_info, pCreateInfo))
surface->flags |= RADEON_SURF_DISABLE_DCC;
- if (create_info->scanout)
+
+ if (radv_surface_has_scanout(device, create_info))
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;
}
static inline unsigned
-si_tile_mode_index(const struct radv_image *image, unsigned level, bool stencil)
+si_tile_mode_index(const struct radv_image_plane *plane, unsigned level, bool stencil)
{
if (stencil)
- return image->surface.u.legacy.stencil_tiling_index[level];
+ return plane->surface.u.legacy.stencil_tiling_index[level];
else
- return image->surface.u.legacy.tiling_index[level];
+ return plane->surface.u.legacy.tiling_index[level];
}
static unsigned radv_map_swizzle(unsigned swizzle)
{
const struct vk_format_description *desc;
unsigned stride;
- uint64_t gpu_address = device->ws->buffer_get_va(buffer->bo);
+ uint64_t gpu_address = radv_buffer_get_va(buffer->bo);
uint64_t va = gpu_address + buffer->offset;
unsigned num_format, data_format;
int first_non_void;
state[1] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
S_008F04_STRIDE(stride);
- if (device->physical_device->rad_info.chip_class != VI && stride) {
+ if (device->physical_device->rad_info.chip_class != GFX8 && stride) {
range /= stride;
}
si_set_mutable_tex_desc_fields(struct radv_device *device,
struct radv_image *image,
const struct legacy_surf_level *base_level_info,
+ unsigned plane_id,
unsigned base_level, unsigned first_level,
unsigned block_width, bool is_stencil,
- uint32_t *state)
+ bool is_storage_image, uint32_t *state)
{
- uint64_t gpu_address = image->bo ? device->ws->buffer_get_va(image->bo) + image->offset : 0;
- uint64_t va = gpu_address;
+ struct radv_image_plane *plane = &image->planes[plane_id];
+ uint64_t gpu_address = image->bo ? radv_buffer_get_va(image->bo) + image->offset : 0;
+ uint64_t va = gpu_address + plane->offset;
enum chip_class chip_class = device->physical_device->rad_info.chip_class;
uint64_t meta_va = 0;
if (chip_class >= GFX9) {
if (is_stencil)
- va += image->surface.u.gfx9.stencil_offset;
+ va += plane->surface.u.gfx9.stencil_offset;
else
- va += image->surface.u.gfx9.surf_offset;
+ va += plane->surface.u.gfx9.surf_offset;
} else
va += base_level_info->offset;
state[0] = va >> 8;
if (chip_class >= GFX9 ||
base_level_info->mode == RADEON_SURF_MODE_2D)
- state[0] |= image->surface.tile_swizzle;
+ state[0] |= plane->surface.tile_swizzle;
state[1] &= C_008F14_BASE_ADDRESS_HI;
state[1] |= S_008F14_BASE_ADDRESS_HI(va >> 40);
- if (chip_class >= VI) {
+ if (chip_class >= GFX8) {
state[6] &= C_008F28_COMPRESSION_EN;
state[7] = 0;
- if (image->surface.dcc_size && first_level < image->surface.num_dcc_levels) {
+ if (!is_storage_image && radv_dcc_enabled(image, first_level)) {
meta_va = gpu_address + image->dcc_offset;
- if (chip_class <= VI)
+ if (chip_class <= GFX8)
meta_va += base_level_info->dcc_offset;
+ } else if (!is_storage_image &&
+ radv_image_is_tc_compat_htile(image)) {
+ 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;
+ state[7] |= plane->surface.tile_swizzle;
}
}
if (chip_class >= GFX9) {
state[3] &= C_008F1C_SW_MODE;
- state[4] &= C_008F20_PITCH_GFX9;
+ state[4] &= C_008F20_PITCH;
if (is_stencil) {
- state[3] |= S_008F1C_SW_MODE(image->surface.u.gfx9.stencil.swizzle_mode);
- state[4] |= S_008F20_PITCH_GFX9(image->surface.u.gfx9.stencil.epitch);
+ state[3] |= S_008F1C_SW_MODE(plane->surface.u.gfx9.stencil.swizzle_mode);
+ state[4] |= S_008F20_PITCH(plane->surface.u.gfx9.stencil.epitch);
} else {
- state[3] |= S_008F1C_SW_MODE(image->surface.u.gfx9.surf.swizzle_mode);
- state[4] |= S_008F20_PITCH_GFX9(image->surface.u.gfx9.surf.epitch);
+ state[3] |= S_008F1C_SW_MODE(plane->surface.u.gfx9.surf.swizzle_mode);
+ state[4] |= S_008F20_PITCH(plane->surface.u.gfx9.surf.epitch);
}
state[5] &= C_008F24_META_DATA_ADDRESS &
struct gfx9_surf_meta_flags meta;
if (image->dcc_offset)
- meta = image->surface.u.gfx9.dcc;
+ meta = plane->surface.u.gfx9.dcc;
else
- meta = image->surface.u.gfx9.htile;
+ meta = plane->surface.u.gfx9.htile;
state[5] |= S_008F24_META_DATA_ADDRESS(meta_va >> 40) |
S_008F24_META_PIPE_ALIGNED(meta.pipe_aligned) |
S_008F24_META_RB_ALIGNED(meta.rb_aligned);
}
} else {
- /* SI-CI-VI */
+ /* GFX6-GFX8 */
unsigned pitch = base_level_info->nblk_x * block_width;
- unsigned index = si_tile_mode_index(image, base_level, is_stencil);
+ unsigned index = si_tile_mode_index(plane, base_level, is_stencil);
state[3] &= C_008F1C_TILING_INDEX;
state[3] |= S_008F1C_TILING_INDEX(index);
- state[4] &= C_008F20_PITCH_GFX6;
- state[4] |= S_008F20_PITCH_GFX6(pitch - 1);
+ state[4] &= C_008F20_PITCH;
+ state[4] |= S_008F20_PITCH(pitch - 1);
}
}
else
return V_008F1C_SQ_RSRC_IMG_2D_ARRAY;
default:
- unreachable("illegale image type");
+ unreachable("illegal image type");
}
}
-static unsigned gfx9_border_color_swizzle(const unsigned char swizzle[4])
+static unsigned gfx9_border_color_swizzle(const enum vk_swizzle swizzle[4])
{
unsigned bc_swizzle = V_008F20_BC_SWIZZLE_XYZW;
data_format = 0;
}
+ /* S8 with either Z16 or Z32 HTILE need a special format. */
+ if (device->physical_device->rad_info.chip_class >= GFX9 &&
+ vk_format == VK_FORMAT_S8_UINT &&
+ radv_image_is_tc_compat_htile(image)) {
+ if (image->vk_format == VK_FORMAT_D32_SFLOAT_S8_UINT)
+ data_format = V_008F14_IMG_DATA_FORMAT_S8_32;
+ else if (image->vk_format == VK_FORMAT_D16_UNORM_S8_UINT)
+ data_format = V_008F14_IMG_DATA_FORMAT_S8_16;
+ }
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) {
depth = image->info.array_size / 6;
state[0] = 0;
- state[1] = (S_008F14_DATA_FORMAT_GFX6(data_format) |
- S_008F14_NUM_FORMAT_GFX6(num_format));
+ state[1] = (S_008F14_DATA_FORMAT(data_format) |
+ S_008F14_NUM_FORMAT(num_format));
state[2] = (S_008F18_WIDTH(width - 1) |
S_008F18_HEIGHT(height - 1) |
S_008F18_PERF_MOD(4));
state[7] = 0;
if (device->physical_device->rad_info.chip_class >= GFX9) {
- unsigned bc_swizzle = gfx9_border_color_swizzle(desc->swizzle);
+ unsigned bc_swizzle = gfx9_border_color_swizzle(swizzle);
- /* Depth is the the last accessible layer on Gfx9.
+ /* Depth is the last accessible layer on Gfx9.
* The hw doesn't need to know the total number of layers.
*/
if (type == V_008F1C_SQ_RSRC_IMG_3D)
/* The last dword is unused by hw. The shader uses it to clear
* bits in the first dword of sampler state.
*/
- if (device->physical_device->rad_info.chip_class <= CIK && image->info.samples <= 1) {
+ if (device->physical_device->rad_info.chip_class <= GFX7 && image->info.samples <= 1) {
if (first_level == last_level)
state[7] = C_008F30_MAX_ANISO_RATIO;
else
}
/* Initialize the sampler view for FMASK. */
- if (image->fmask.size) {
+ if (radv_image_has_fmask(image)) {
uint32_t fmask_format, num_format;
- uint64_t gpu_address = device->ws->buffer_get_va(image->bo);
+ uint64_t gpu_address = radv_buffer_get_va(image->bo);
uint64_t va;
+ assert(image->plane_count == 1);
+
va = gpu_address + image->offset + image->fmask.offset;
if (device->physical_device->rad_info.chip_class >= GFX9) {
fmask_state[0] = va >> 8;
fmask_state[0] |= image->fmask.tile_swizzle;
fmask_state[1] = S_008F14_BASE_ADDRESS_HI(va >> 40) |
- S_008F14_DATA_FORMAT_GFX6(fmask_format) |
- S_008F14_NUM_FORMAT_GFX6(num_format);
+ S_008F14_DATA_FORMAT(fmask_format) |
+ S_008F14_NUM_FORMAT(num_format);
fmask_state[2] = S_008F18_WIDTH(width - 1) |
S_008F18_HEIGHT(height - 1);
fmask_state[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X) |
S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X) |
S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X) |
S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X) |
- S_008F1C_TYPE(radv_tex_dim(image->type, view_type, 1, 0, false, false));
+ S_008F1C_TYPE(radv_tex_dim(image->type, view_type, image->info.array_size, 0, false, false));
fmask_state[4] = 0;
fmask_state[5] = S_008F24_BASE_ARRAY(first_layer);
fmask_state[6] = 0;
fmask_state[7] = 0;
if (device->physical_device->rad_info.chip_class >= GFX9) {
- fmask_state[3] |= S_008F1C_SW_MODE(image->surface.u.gfx9.fmask.swizzle_mode);
+ fmask_state[3] |= S_008F1C_SW_MODE(image->planes[0].surface.u.gfx9.fmask.swizzle_mode);
fmask_state[4] |= S_008F20_DEPTH(last_layer) |
- S_008F20_PITCH_GFX9(image->surface.u.gfx9.fmask.epitch);
- fmask_state[5] |= S_008F24_META_PIPE_ALIGNED(image->surface.u.gfx9.cmask.pipe_aligned) |
- S_008F24_META_RB_ALIGNED(image->surface.u.gfx9.cmask.rb_aligned);
+ S_008F20_PITCH(image->planes[0].surface.u.gfx9.fmask.epitch);
+ fmask_state[5] |= S_008F24_META_PIPE_ALIGNED(image->planes[0].surface.u.gfx9.cmask.pipe_aligned) |
+ S_008F24_META_RB_ALIGNED(image->planes[0].surface.u.gfx9.cmask.rb_aligned);
} else {
fmask_state[3] |= S_008F1C_TILING_INDEX(image->fmask.tile_mode_index);
fmask_state[4] |= S_008F20_DEPTH(depth - 1) |
- S_008F20_PITCH_GFX6(image->fmask.pitch_in_pixels - 1);
+ S_008F20_PITCH(image->fmask.pitch_in_pixels - 1);
fmask_state[5] |= S_008F24_LAST_ARRAY(last_layer);
}
} else if (fmask_state)
static const VkComponentMapping fixedmapping;
uint32_t desc[8], i;
+ assert(image->plane_count == 1);
+
/* Metadata image format format version 1:
* [0] = 1 (metadata format identifier)
* [1] = (VENDOR_ID << 16) | PCI_ID
si_make_texture_descriptor(device, image, false,
(VkImageViewType)image->type, image->vk_format,
&fixedmapping, 0, image->info.levels - 1, 0,
- image->info.array_size,
+ image->info.array_size - 1,
image->info.width, image->info.height,
image->info.depth,
desc, NULL);
- si_set_mutable_tex_desc_fields(device, image, &image->surface.u.legacy.level[0], 0, 0,
- image->surface.blk_w, false, desc);
+ si_set_mutable_tex_desc_fields(device, image, &image->planes[0].surface.u.legacy.level[0], 0, 0, 0,
+ image->planes[0].surface.blk_w, false, false, desc);
/* Clear the base address and set the relative DCC offset. */
desc[0] = 0;
memcpy(&md->metadata[2], desc, sizeof(desc));
/* Dwords [10:..] contain the mipmap level offsets. */
- if (device->physical_device->rad_info.chip_class <= VI) {
+ if (device->physical_device->rad_info.chip_class <= GFX8) {
for (i = 0; i <= image->info.levels - 1; i++)
- md->metadata[10+i] = image->surface.u.legacy.level[i].offset >> 8;
+ md->metadata[10+i] = image->planes[0].surface.u.legacy.level[i].offset >> 8;
md->size_metadata = (11 + image->info.levels - 1) * 4;
}
}
struct radv_image *image,
struct radeon_bo_metadata *metadata)
{
- struct radeon_surf *surface = &image->surface;
+ struct radeon_surf *surface = &image->planes[0].surface;
memset(metadata, 0, sizeof(*metadata));
radv_query_opaque_metadata(device, image, metadata);
}
+void
+radv_image_override_offset_stride(struct radv_device *device,
+ struct radv_image *image,
+ uint64_t offset, uint32_t stride)
+{
+ struct radeon_surf *surface = &image->planes[0].surface;
+ unsigned bpe = vk_format_get_blocksizebits(image->vk_format) / 8;
+
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ if (stride) {
+ surface->u.gfx9.surf_pitch = stride;
+ surface->u.gfx9.surf_slice_size =
+ (uint64_t)stride * surface->u.gfx9.surf_height * bpe;
+ }
+ surface->u.gfx9.surf_offset = offset;
+ } else {
+ surface->u.legacy.level[0].nblk_x = stride;
+ surface->u.legacy.level[0].slice_size_dw =
+ ((uint64_t)stride * surface->u.legacy.level[0].nblk_y * bpe) / 4;
+
+ if (offset) {
+ for (unsigned i = 0; i < ARRAY_SIZE(surface->u.legacy.level); ++i)
+ surface->u.legacy.level[i].offset += offset;
+ }
+
+ }
+}
+
/* The number of samples can be specified independently of the texture. */
static void
radv_image_get_fmask_info(struct radv_device *device,
unsigned nr_samples,
struct radv_fmask_info *out)
{
- /* FMASK is allocated like an ordinary texture. */
- struct radeon_surf fmask = {};
- struct ac_surf_info info = image->info;
- memset(out, 0, sizeof(*out));
-
if (device->physical_device->rad_info.chip_class >= GFX9) {
- out->alignment = image->surface.u.gfx9.fmask_alignment;
- out->size = image->surface.u.gfx9.fmask_size;
+ out->alignment = image->planes[0].surface.fmask_alignment;
+ out->size = image->planes[0].surface.fmask_size;
+ out->tile_swizzle = image->planes[0].surface.fmask_tile_swizzle;
return;
}
- fmask.blk_w = image->surface.blk_w;
- fmask.blk_h = image->surface.blk_h;
- info.samples = 1;
- fmask.flags = image->surface.flags | RADEON_SURF_FMASK;
-
- if (!image->shareable)
- info.surf_index = &device->fmask_mrt_offset_counter;
-
- /* Force 2D tiling if it wasn't set. This may occur when creating
- * FMASK for MSAA resolve on R6xx. On R6xx, the single-sample
- * destination buffer must have an FMASK too. */
- fmask.flags = RADEON_SURF_CLR(fmask.flags, MODE);
- fmask.flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE);
-
- switch (nr_samples) {
- case 2:
- case 4:
- fmask.bpe = 1;
- break;
- case 8:
- fmask.bpe = 4;
- break;
- default:
- return;
- }
-
- device->ws->surface_init(device->ws, &info, &fmask);
- assert(fmask.u.legacy.level[0].mode == RADEON_SURF_MODE_2D);
-
- out->slice_tile_max = (fmask.u.legacy.level[0].nblk_x * fmask.u.legacy.level[0].nblk_y) / 64;
- if (out->slice_tile_max)
- out->slice_tile_max -= 1;
-
- out->tile_mode_index = fmask.u.legacy.tiling_index[0];
- out->pitch_in_pixels = fmask.u.legacy.level[0].nblk_x;
- out->bank_height = fmask.u.legacy.bankh;
- out->tile_swizzle = fmask.tile_swizzle;
- out->alignment = MAX2(256, fmask.surf_alignment);
- out->size = fmask.surf_size;
+ out->slice_tile_max = image->planes[0].surface.u.legacy.fmask.slice_tile_max;
+ out->tile_mode_index = image->planes[0].surface.u.legacy.fmask.tiling_index;
+ out->pitch_in_pixels = image->planes[0].surface.u.legacy.fmask.pitch_in_pixels;
+ out->bank_height = image->planes[0].surface.u.legacy.fmask.bankh;
+ out->tile_swizzle = image->planes[0].surface.fmask_tile_swizzle;
+ out->alignment = image->planes[0].surface.fmask_alignment;
+ out->size = image->planes[0].surface.fmask_size;
assert(!out->tile_swizzle || !image->shareable);
}
unsigned num_pipes = device->physical_device->rad_info.num_tile_pipes;
unsigned cl_width, cl_height;
+ assert(image->plane_count == 1);
+
if (device->physical_device->rad_info.chip_class >= GFX9) {
- out->alignment = image->surface.u.gfx9.cmask_alignment;
- out->size = image->surface.u.gfx9.cmask_size;
+ out->alignment = image->planes[0].surface.cmask_alignment;
+ out->size = image->planes[0].surface.cmask_size;
return;
}
unsigned base_align = num_pipes * pipe_interleave_bytes;
- unsigned width = align(image->info.width, cl_width*8);
- unsigned height = align(image->info.height, cl_height*8);
+ unsigned width = align(image->planes[0].surface.u.legacy.level[0].nblk_x, cl_width*8);
+ unsigned height = align(image->planes[0].surface.u.legacy.level[0].nblk_y, cl_height*8);
unsigned slice_elements = (width * height) / (8*8);
/* Each element of CMASK is a nibble. */
}
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);
+ assert(image->plane_count == 1);
+
+ image->dcc_offset = align64(image->size, image->planes[0].surface.dcc_alignment);
/* + 16 for storing the clear values + dcc pred */
- image->clear_value_offset = image->dcc_offset + image->surface.dcc_size;
- image->dcc_pred_offset = image->clear_value_offset + 8;
- image->size = image->dcc_offset + image->surface.dcc_size + 16;
- image->alignment = MAX2(image->alignment, image->surface.dcc_alignment);
+ image->clear_value_offset = image->dcc_offset + image->planes[0].surface.dcc_size;
+ image->fce_pred_offset = image->clear_value_offset + 8;
+ image->dcc_pred_offset = image->clear_value_offset + 16;
+ image->size = image->dcc_offset + image->planes[0].surface.dcc_size + 24;
+ image->alignment = MAX2(image->alignment, image->planes[0].surface.dcc_alignment);
}
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);
+ image->htile_offset = align64(image->size, image->planes[0].surface.htile_alignment);
/* + 8 for storing the clear values */
- image->clear_value_offset = image->htile_offset + image->surface.htile_size;
+ image->clear_value_offset = image->htile_offset + image->planes[0].surface.htile_size;
image->size = image->clear_value_offset + 8;
- image->alignment = align64(image->alignment, image->surface.htile_alignment);
+ if (radv_image_is_tc_compat_htile(image)) {
+ /* Metadata for the TC-compatible HTILE hardware bug which
+ * have to be fixed by updating ZRANGE_PRECISION when doing
+ * fast depth clears to 0.0f.
+ */
+ image->tc_compat_zrange_offset = image->clear_value_offset + 8;
+ image->size = image->clear_value_offset + 16;
+ }
+ image->alignment = align64(image->alignment, image->planes[0].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) &&
+ radv_image_has_dcc(image);
+}
+
+static inline bool
+radv_image_can_enable_cmask(struct radv_image *image)
+{
+ if (image->planes[0].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->planes[0].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 radv_image_has_htile(image) &&
+ image->info.levels == 1 &&
+ image->info.width * image->info.height >= 8 * 8;
+}
+
+static void radv_image_disable_dcc(struct radv_image *image)
+{
+ for (unsigned i = 0; i < image->plane_count; ++i)
+ image->planes[i].surface.dcc_size = 0;
+}
+
+static void radv_image_disable_htile(struct radv_image *image)
+{
+ for (unsigned i = 0; i < image->plane_count; ++i)
+ image->planes[i].surface.htile_size = 0;
}
VkResult
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);
+ const unsigned plane_count = vk_format_get_plane_count(pCreateInfo->format);
+ const size_t image_struct_size = sizeof(*image) + sizeof(struct radv_image_plane) * plane_count;
+
radv_assert(pCreateInfo->mipLevels > 0);
radv_assert(pCreateInfo->arrayLayers > 0);
radv_assert(pCreateInfo->samples > 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, image_struct_size, 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!image)
- return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ return vk_error(device->instance, 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->info.depth = pCreateInfo->extent.depth;
image->info.samples = pCreateInfo->samples;
+ image->info.storage_samples = pCreateInfo->samples;
image->info.array_size = pCreateInfo->arrayLayers;
image->info.levels = pCreateInfo->mipLevels;
+ image->info.num_channels = vk_format_get_nr_components(pCreateInfo->format);
image->vk_format = pCreateInfo->format;
image->tiling = pCreateInfo->tiling;
image->exclusive = pCreateInfo->sharingMode == VK_SHARING_MODE_EXCLUSIVE;
if (pCreateInfo->sharingMode == VK_SHARING_MODE_CONCURRENT) {
for (uint32_t i = 0; i < pCreateInfo->queueFamilyIndexCount; ++i)
- if (pCreateInfo->pQueueFamilyIndices[i] == VK_QUEUE_FAMILY_EXTERNAL_KHR)
+ if (pCreateInfo->pQueueFamilyIndices[i] == VK_QUEUE_FAMILY_EXTERNAL)
image->queue_family_mask |= (1u << RADV_MAX_QUEUE_FAMILIES) - 1u;
else
image->queue_family_mask |= 1u << pCreateInfo->pQueueFamilyIndices[i];
}
image->shareable = vk_find_struct_const(pCreateInfo->pNext,
- EXTERNAL_MEMORY_IMAGE_CREATE_INFO_KHR) != NULL;
- if (!vk_format_is_depth(pCreateInfo->format) && !create_info->scanout && !image->shareable) {
+ EXTERNAL_MEMORY_IMAGE_CREATE_INFO) != NULL;
+ if (!vk_format_is_depth_or_stencil(pCreateInfo->format) &&
+ !radv_surface_has_scanout(device, create_info) && !image->shareable) {
image->info.surf_index = &device->image_mrt_offset_counter;
}
- radv_init_surface(device, &image->surface, create_info);
+ image->plane_count = plane_count;
+ image->size = 0;
+ image->alignment = 1;
+ for (unsigned plane = 0; plane < plane_count; ++plane) {
+ struct ac_surf_info info = image->info;
+ radv_init_surface(device, image, &image->planes[plane].surface, plane, create_info);
- device->ws->surface_init(device->ws, &image->info, &image->surface);
+ if (plane) {
+ const struct vk_format_description *desc = vk_format_description(pCreateInfo->format);
+ assert(info.width % desc->width_divisor == 0);
+ assert(info.height % desc->height_divisor == 0);
- image->size = image->surface.surf_size;
- image->alignment = image->surface.surf_alignment;
+ info.width /= desc->width_divisor;
+ info.height /= desc->height_divisor;
+ }
- if (image->exclusive || image->queue_family_mask == 1)
- can_cmask_dcc = true;
+ device->ws->surface_init(device->ws, &info, &image->planes[plane].surface);
- if ((pCreateInfo->usage & VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT) &&
- image->surface.dcc_size && can_cmask_dcc)
- radv_image_alloc_dcc(device, image);
- else
- image->surface.dcc_size = 0;
+ image->planes[plane].offset = align(image->size, image->planes[plane].surface.surf_alignment);
+ image->size = image->planes[plane].offset + image->planes[plane].surface.surf_size;
+ image->alignment = image->planes[plane].surface.surf_alignment;
- 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);
+ image->planes[plane].format = vk_format_get_plane_format(image->vk_format, plane);
+ }
- if (image->info.samples > 1 && vk_format_is_color(pCreateInfo->format)) {
- radv_image_alloc_fmask(device, image);
- } else if (vk_format_is_depth(pCreateInfo->format)) {
+ if (!create_info->no_metadata_planes) {
+ /* Try to enable DCC first. */
+ if (radv_image_can_enable_dcc(image)) {
+ radv_image_alloc_dcc(image);
+ if (image->info.samples > 1) {
+ /* CMASK should be enabled because DCC fast
+ * clear with MSAA needs it.
+ */
+ assert(radv_image_can_enable_cmask(image));
+ radv_image_alloc_cmask(device, image);
+ }
+ } else {
+ /* When DCC cannot be enabled, try CMASK. */
+ radv_image_disable_dcc(image);
+ if (radv_image_can_enable_cmask(image)) {
+ radv_image_alloc_cmask(device, image);
+ }
+ }
- radv_image_alloc_htile(device, image);
+ /* Try to enable FMASK for multisampled images. */
+ if (radv_image_can_enable_fmask(image)) {
+ radv_image_alloc_fmask(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)) {
+ image->tc_compatible_htile = image->planes[0].surface.flags & RADEON_SURF_TC_COMPATIBLE_HTILE;
+ radv_image_alloc_htile(image);
+ } else {
+ radv_image_disable_htile(image);
+ }
+ }
+ } else {
+ radv_image_disable_dcc(image);
+ radv_image_disable_htile(image);
}
if (pCreateInfo->flags & VK_IMAGE_CREATE_SPARSE_BINDING_BIT) {
image->offset = 0;
image->bo = device->ws->buffer_create(device->ws, image->size, image->alignment,
- 0, RADEON_FLAG_VIRTUAL);
+ 0, RADEON_FLAG_VIRTUAL, RADV_BO_PRIORITY_VIRTUAL);
if (!image->bo) {
vk_free2(&device->alloc, alloc, image);
- return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
}
}
static void
radv_image_view_make_descriptor(struct radv_image_view *iview,
struct radv_device *device,
+ VkFormat vk_format,
const VkComponentMapping *components,
- bool is_storage_image)
+ bool is_storage_image, unsigned plane_id,
+ unsigned descriptor_plane_id)
{
struct radv_image *image = iview->image;
+ struct radv_image_plane *plane = &image->planes[plane_id];
+ const struct vk_format_description *format_desc = vk_format_description(image->vk_format);
bool is_stencil = iview->aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT;
uint32_t blk_w;
- uint32_t *descriptor;
- uint32_t *fmask_descriptor;
+ union radv_descriptor *descriptor;
uint32_t hw_level = 0;
if (is_storage_image) {
- descriptor = iview->storage_descriptor;
- fmask_descriptor = iview->storage_fmask_descriptor;
+ descriptor = &iview->storage_descriptor;
} else {
- descriptor = iview->descriptor;
- fmask_descriptor = iview->fmask_descriptor;
+ descriptor = &iview->descriptor;
}
- assert(image->surface.blk_w % vk_format_get_blockwidth(image->vk_format) == 0);
- blk_w = image->surface.blk_w / vk_format_get_blockwidth(image->vk_format) * vk_format_get_blockwidth(iview->vk_format);
+ assert(vk_format_get_plane_count(vk_format) == 1);
+ assert(plane->surface.blk_w % vk_format_get_blockwidth(plane->format) == 0);
+ blk_w = plane->surface.blk_w / vk_format_get_blockwidth(plane->format) * vk_format_get_blockwidth(vk_format);
if (device->physical_device->rad_info.chip_class >= GFX9)
hw_level = iview->base_mip;
si_make_texture_descriptor(device, image, is_storage_image,
iview->type,
- iview->vk_format,
+ vk_format,
components,
hw_level, hw_level + iview->level_count - 1,
iview->base_layer,
iview->base_layer + iview->layer_count - 1,
- iview->extent.width,
- iview->extent.height,
+ iview->extent.width / (plane_id ? format_desc->width_divisor : 1),
+ iview->extent.height / (plane_id ? format_desc->height_divisor : 1),
iview->extent.depth,
- descriptor,
- fmask_descriptor);
+ descriptor->plane_descriptors[descriptor_plane_id],
+ descriptor_plane_id ? NULL : descriptor->fmask_descriptor);
const struct legacy_surf_level *base_level_info = NULL;
if (device->physical_device->rad_info.chip_class <= GFX9) {
if (is_stencil)
- base_level_info = &image->surface.u.legacy.stencil_level[iview->base_mip];
+ base_level_info = &plane->surface.u.legacy.stencil_level[iview->base_mip];
else
- base_level_info = &image->surface.u.legacy.level[iview->base_mip];
+ base_level_info = &plane->surface.u.legacy.level[iview->base_mip];
}
si_set_mutable_tex_desc_fields(device, image,
base_level_info,
+ plane_id,
iview->base_mip,
iview->base_mip,
- blk_w, is_stencil, descriptor);
+ blk_w, is_stencil, is_storage_image, descriptor->plane_descriptors[descriptor_plane_id]);
+}
+
+static unsigned
+radv_plane_from_aspect(VkImageAspectFlags mask)
+{
+ switch(mask) {
+ case VK_IMAGE_ASPECT_PLANE_1_BIT:
+ return 1;
+ case VK_IMAGE_ASPECT_PLANE_2_BIT:
+ return 2;
+ default:
+ return 0;
+ }
+}
+
+VkFormat
+radv_get_aspect_format(struct radv_image *image, VkImageAspectFlags mask)
+{
+ switch(mask) {
+ case VK_IMAGE_ASPECT_PLANE_0_BIT:
+ return image->planes[0].format;
+ case VK_IMAGE_ASPECT_PLANE_1_BIT:
+ return image->planes[1].format;
+ case VK_IMAGE_ASPECT_PLANE_2_BIT:
+ return image->planes[2].format;
+ case VK_IMAGE_ASPECT_STENCIL_BIT:
+ return vk_format_stencil_only(image->vk_format);
+ case VK_IMAGE_ASPECT_DEPTH_BIT:
+ return vk_format_depth_only(image->vk_format);
+ case VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT:
+ return vk_format_depth_only(image->vk_format);
+ default:
+ return image->vk_format;
+ }
}
void
iview->image = image;
iview->bo = image->bo;
iview->type = pCreateInfo->viewType;
- iview->vk_format = pCreateInfo->format;
+ iview->plane_id = radv_plane_from_aspect(pCreateInfo->subresourceRange.aspectMask);
iview->aspect_mask = pCreateInfo->subresourceRange.aspectMask;
+ iview->multiple_planes = vk_format_get_plane_count(image->vk_format) > 1 && iview->aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT;
+ iview->vk_format = pCreateInfo->format;
if (iview->aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT) {
iview->vk_format = vk_format_stencil_only(iview->vk_format);
};
}
- if (iview->vk_format != image->vk_format) {
- iview->extent.width = round_up_u32(iview->extent.width * vk_format_get_blockwidth(iview->vk_format),
- vk_format_get_blockwidth(image->vk_format));
- iview->extent.height = round_up_u32(iview->extent.height * vk_format_get_blockheight(iview->vk_format),
- vk_format_get_blockheight(image->vk_format));
+ if (iview->vk_format != image->planes[iview->plane_id].format) {
+ unsigned view_bw = vk_format_get_blockwidth(iview->vk_format);
+ unsigned view_bh = vk_format_get_blockheight(iview->vk_format);
+ unsigned img_bw = vk_format_get_blockwidth(image->vk_format);
+ unsigned img_bh = vk_format_get_blockheight(image->vk_format);
+
+ iview->extent.width = round_up_u32(iview->extent.width * view_bw, img_bw);
+ iview->extent.height = round_up_u32(iview->extent.height * view_bh, img_bh);
+
+ /* Comment ported from amdvlk -
+ * If we have the following image:
+ * Uncompressed pixels Compressed block sizes (4x4)
+ * mip0: 22 x 22 6 x 6
+ * mip1: 11 x 11 3 x 3
+ * mip2: 5 x 5 2 x 2
+ * mip3: 2 x 2 1 x 1
+ * mip4: 1 x 1 1 x 1
+ *
+ * On GFX9 the descriptor is always programmed with the WIDTH and HEIGHT of the base level and the HW is
+ * calculating the degradation of the block sizes down the mip-chain as follows (straight-up
+ * divide-by-two integer math):
+ * mip0: 6x6
+ * mip1: 3x3
+ * mip2: 1x1
+ * mip3: 1x1
+ *
+ * This means that mip2 will be missing texels.
+ *
+ * Fix this by calculating the base mip's width and height, then convert that, and round it
+ * back up to get the level 0 size.
+ * Clamp the converted size between the original values, and next power of two, which
+ * means we don't oversize the image.
+ */
+ if (device->physical_device->rad_info.chip_class >= GFX9 &&
+ vk_format_is_compressed(image->vk_format) &&
+ !vk_format_is_compressed(iview->vk_format)) {
+ unsigned lvl_width = radv_minify(image->info.width , range->baseMipLevel);
+ unsigned lvl_height = radv_minify(image->info.height, range->baseMipLevel);
+
+ lvl_width = round_up_u32(lvl_width * view_bw, img_bw);
+ lvl_height = round_up_u32(lvl_height * view_bh, img_bh);
+
+ lvl_width <<= range->baseMipLevel;
+ lvl_height <<= range->baseMipLevel;
+
+ iview->extent.width = CLAMP(lvl_width, iview->extent.width, iview->image->planes[0].surface.u.gfx9.surf_pitch);
+ iview->extent.height = CLAMP(lvl_height, iview->extent.height, iview->image->planes[0].surface.u.gfx9.surf_height);
+ }
}
iview->base_layer = range->baseArrayLayer;
iview->base_mip = range->baseMipLevel;
iview->level_count = radv_get_levelCount(image, range);
- radv_image_view_make_descriptor(iview, device, &pCreateInfo->components, false);
- radv_image_view_make_descriptor(iview, device, &pCreateInfo->components, true);
+ for (unsigned i = 0; i < (iview->multiple_planes ? vk_format_get_plane_count(image->vk_format) : 1); ++i) {
+ VkFormat format = vk_format_get_plane_format(iview->vk_format, i);
+ radv_image_view_make_descriptor(iview, device, format, &pCreateInfo->components, false, iview->plane_id + i, i);
+ radv_image_view_make_descriptor(iview, device, format, &pCreateInfo->components, true, iview->plane_id + i, i);
+ }
}
bool radv_layout_has_htile(const struct radv_image *image,
VkImageLayout layout,
unsigned queue_mask)
{
- return image->surface.htile_size &&
+ if (radv_image_is_tc_compat_htile(image))
+ return layout != VK_IMAGE_LAYOUT_GENERAL;
+
+ return radv_image_has_htile(image) &&
(layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
- layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL) &&
- queue_mask == (1u << RADV_QUEUE_GENERAL);
+ (layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL &&
+ queue_mask == (1u << RADV_QUEUE_GENERAL)));
}
bool radv_layout_is_htile_compressed(const struct radv_image *image,
VkImageLayout layout,
unsigned queue_mask)
{
- return image->surface.htile_size &&
+ if (radv_image_is_tc_compat_htile(image))
+ return layout != VK_IMAGE_LAYOUT_GENERAL;
+
+ return radv_image_has_htile(image) &&
(layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL ||
- layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL) &&
- queue_mask == (1u << RADV_QUEUE_GENERAL);
+ (layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL &&
+ queue_mask == (1u << RADV_QUEUE_GENERAL)));
}
bool radv_layout_can_fast_clear(const struct radv_image *image,
VkImageLayout layout,
unsigned queue_mask)
{
- return layout == VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL &&
- queue_mask == (1u << RADV_QUEUE_GENERAL);
+ return layout == VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
+}
+
+bool radv_layout_dcc_compressed(const struct radv_image *image,
+ VkImageLayout layout,
+ unsigned queue_mask)
+{
+ /* Don't compress compute transfer dst, as image stores are not supported. */
+ if (layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL &&
+ (queue_mask & (1u << RADV_QUEUE_COMPUTE)))
+ return false;
+
+ return radv_image_has_dcc(image) && layout != VK_IMAGE_LAYOUT_GENERAL;
}
{
if (!image->exclusive)
return image->queue_family_mask;
- if (family == VK_QUEUE_FAMILY_EXTERNAL_KHR)
+ if (family == VK_QUEUE_FAMILY_EXTERNAL)
return (1u << RADV_MAX_QUEUE_FAMILIES) - 1u;
if (family == VK_QUEUE_FAMILY_IGNORED)
return 1u << queue_family;
const VkAllocationCallbacks *pAllocator,
VkImage *pImage)
{
+#ifdef ANDROID
+ const VkNativeBufferANDROID *gralloc_info =
+ vk_find_struct_const(pCreateInfo->pNext, NATIVE_BUFFER_ANDROID);
+
+ if (gralloc_info)
+ return radv_image_from_gralloc(device, pCreateInfo, gralloc_info,
+ pAllocator, pImage);
+#endif
+
+ 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->flags & VK_IMAGE_CREATE_SPARSE_BINDING_BIT)
device->ws->buffer_destroy(image->bo);
+ if (image->owned_memory != VK_NULL_HANDLE)
+ radv_FreeMemory(_device, image->owned_memory, pAllocator);
+
vk_free2(&device->alloc, pAllocator, image);
}
RADV_FROM_HANDLE(radv_device, device, _device);
int level = pSubresource->mipLevel;
int layer = pSubresource->arrayLayer;
- struct radeon_surf *surface = &image->surface;
+
+ unsigned plane_id = radv_plane_from_aspect(pSubresource->aspectMask);
+
+ struct radv_image_plane *plane = &image->planes[plane_id];
+ struct radeon_surf *surface = &plane->surface;
if (device->physical_device->rad_info.chip_class >= GFX9) {
- pLayout->offset = surface->u.gfx9.offset[level] + surface->u.gfx9.surf_slice_size * layer;
- pLayout->rowPitch = surface->u.gfx9.surf_pitch * surface->bpe;
+ pLayout->offset = plane->offset + surface->u.gfx9.offset[level] + surface->u.gfx9.surf_slice_size * layer;
+ if (image->vk_format == VK_FORMAT_R32G32B32_UINT ||
+ image->vk_format == VK_FORMAT_R32G32B32_SINT ||
+ image->vk_format == VK_FORMAT_R32G32B32_SFLOAT) {
+ /* Adjust the number of bytes between each row because
+ * the pitch is actually the number of components per
+ * row.
+ */
+ pLayout->rowPitch = surface->u.gfx9.surf_pitch * surface->bpe / 3;
+ } else {
+ assert(util_is_power_of_two_nonzero(surface->bpe));
+ pLayout->rowPitch = surface->u.gfx9.surf_pitch * surface->bpe;
+ }
+
pLayout->arrayPitch = surface->u.gfx9.surf_slice_size;
pLayout->depthPitch = surface->u.gfx9.surf_slice_size;
pLayout->size = surface->u.gfx9.surf_slice_size;
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 = plane->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);
}
view = vk_alloc2(&device->alloc, pAllocator, sizeof(*view), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (view == NULL)
- return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
radv_image_view_init(view, device, pCreateInfo);
void radv_buffer_view_init(struct radv_buffer_view *view,
struct radv_device *device,
- const VkBufferViewCreateInfo* pCreateInfo,
- struct radv_cmd_buffer *cmd_buffer)
+ const VkBufferViewCreateInfo* pCreateInfo)
{
RADV_FROM_HANDLE(radv_buffer, buffer, pCreateInfo->buffer);
view = vk_alloc2(&device->alloc, pAllocator, sizeof(*view), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!view)
- return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
- radv_buffer_view_init(view, device, pCreateInfo, NULL);
+ radv_buffer_view_init(view, device, pCreateInfo);
*pView = radv_buffer_view_to_handle(view);