#include <valgrind.h>
#include <memcheck.h>
#define VG(x) x
+#ifndef NDEBUG
#define __gen_validate_value(x) VALGRIND_CHECK_MEM_IS_DEFINED(&(x), sizeof(x))
+#endif
#else
#define VG(x)
#endif
#include "common/gen_clflush.h"
-#include "common/gen_device_info.h"
+#include "common/gen_gem.h"
+#include "dev/gen_device_info.h"
#include "blorp/blorp.h"
#include "compiler/brw_compiler.h"
#include "util/macros.h"
+#include "util/hash_table.h"
#include "util/list.h"
+#include "util/set.h"
#include "util/u_atomic.h"
#include "util/u_vector.h"
+#include "util/vma.h"
#include "vk_alloc.h"
#include "vk_debug_report.h"
#include "common/intel_log.h"
#include "wsi_common.h"
+/* anv Virtual Memory Layout
+ * =========================
+ *
+ * When the anv driver is determining the virtual graphics addresses of memory
+ * objects itself using the softpin mechanism, the following memory ranges
+ * will be used.
+ *
+ * Three special considerations to notice:
+ *
+ * (1) the dynamic state pool is located within the same 4 GiB as the low
+ * heap. This is to work around a VF cache issue described in a comment in
+ * anv_physical_device_init_heaps.
+ *
+ * (2) the binding table pool is located at lower addresses than the surface
+ * state pool, within a 4 GiB range. This allows surface state base addresses
+ * to cover both binding tables (16 bit offsets) and surface states (32 bit
+ * offsets).
+ *
+ * (3) the last 4 GiB of the address space is withheld from the high
+ * heap. Various hardware units will read past the end of an object for
+ * various reasons. This healthy margin prevents reads from wrapping around
+ * 48-bit addresses.
+ */
+#define LOW_HEAP_MIN_ADDRESS 0x000000001000ULL /* 4 KiB */
+#define LOW_HEAP_MAX_ADDRESS 0x0000bfffffffULL
+#define DYNAMIC_STATE_POOL_MIN_ADDRESS 0x0000c0000000ULL /* 3 GiB */
+#define DYNAMIC_STATE_POOL_MAX_ADDRESS 0x0000ffffffffULL
+#define BINDING_TABLE_POOL_MIN_ADDRESS 0x000100000000ULL /* 4 GiB */
+#define BINDING_TABLE_POOL_MAX_ADDRESS 0x00013fffffffULL
+#define SURFACE_STATE_POOL_MIN_ADDRESS 0x000140000000ULL /* 5 GiB */
+#define SURFACE_STATE_POOL_MAX_ADDRESS 0x00017fffffffULL
+#define INSTRUCTION_STATE_POOL_MIN_ADDRESS 0x000180000000ULL /* 6 GiB */
+#define INSTRUCTION_STATE_POOL_MAX_ADDRESS 0x0001bfffffffULL
+#define HIGH_HEAP_MIN_ADDRESS 0x0001c0000000ULL /* 7 GiB */
+#define HIGH_HEAP_MAX_ADDRESS 0xfffeffffffffULL
+
+#define LOW_HEAP_SIZE \
+ (LOW_HEAP_MAX_ADDRESS - LOW_HEAP_MIN_ADDRESS + 1)
+#define HIGH_HEAP_SIZE \
+ (HIGH_HEAP_MAX_ADDRESS - HIGH_HEAP_MIN_ADDRESS + 1)
+#define DYNAMIC_STATE_POOL_SIZE \
+ (DYNAMIC_STATE_POOL_MAX_ADDRESS - DYNAMIC_STATE_POOL_MIN_ADDRESS + 1)
+#define BINDING_TABLE_POOL_SIZE \
+ (BINDING_TABLE_POOL_MAX_ADDRESS - BINDING_TABLE_POOL_MIN_ADDRESS + 1)
+#define SURFACE_STATE_POOL_SIZE \
+ (SURFACE_STATE_POOL_MAX_ADDRESS - SURFACE_STATE_POOL_MIN_ADDRESS + 1)
+#define INSTRUCTION_STATE_POOL_SIZE \
+ (INSTRUCTION_STATE_POOL_MAX_ADDRESS - INSTRUCTION_STATE_POOL_MIN_ADDRESS + 1)
+
/* Allowing different clear colors requires us to perform a depth resolve at
* the end of certain render passes. This is because while slow clears store
* the clear color in the HiZ buffer, fast clears (without a resolve) don't.
#define MAX_IMAGES 8
#define MAX_PUSH_DESCRIPTORS 32 /* Minimum requirement */
+/* The kernel relocation API has a limitation of a 32-bit delta value
+ * applied to the address before it is written which, in spite of it being
+ * unsigned, is treated as signed . Because of the way that this maps to
+ * the Vulkan API, we cannot handle an offset into a buffer that does not
+ * fit into a signed 32 bits. The only mechanism we have for dealing with
+ * this at the moment is to limit all VkDeviceMemory objects to a maximum
+ * of 2GB each. The Vulkan spec allows us to do this:
+ *
+ * "Some platforms may have a limit on the maximum size of a single
+ * allocation. For example, certain systems may fail to create
+ * allocations with a size greater than or equal to 4GB. Such a limit is
+ * implementation-dependent, and if such a failure occurs then the error
+ * VK_ERROR_OUT_OF_DEVICE_MEMORY should be returned."
+ *
+ * We don't use vk_error here because it's not an error so much as an
+ * indication to the application that the allocation is too large.
+ */
+#define MAX_MEMORY_ALLOCATION_SIZE (1ull << 31)
+
#define ANV_SVGS_VB_INDEX MAX_VBS
#define ANV_DRAWID_VB_INDEX (MAX_VBS + 1)
struct anv_bo bo;
+ /* The address where the start of the pool is pinned. The various bos that
+ * are created as the pool grows will have addresses in the range
+ * [start_address, start_address + BLOCK_POOL_MEMFD_SIZE).
+ */
+ uint64_t start_address;
+
/* The offset from the start of the bo to the "center" of the block
* pool. Pointers to allocated blocks are given by
* bo.map + center_bo_offset + offsets.
*/
VkResult anv_block_pool_init(struct anv_block_pool *pool,
struct anv_device *device,
+ uint64_t start_address,
uint32_t initial_size,
uint64_t bo_flags);
void anv_block_pool_finish(struct anv_block_pool *pool);
VkResult anv_state_pool_init(struct anv_state_pool *pool,
struct anv_device *device,
+ uint64_t start_address,
uint32_t block_size,
uint64_t bo_flags);
void anv_state_pool_finish(struct anv_state_pool *pool);
void anv_bo_cache_finish(struct anv_bo_cache *cache);
VkResult anv_bo_cache_alloc(struct anv_device *device,
struct anv_bo_cache *cache,
- uint64_t size, struct anv_bo **bo);
+ uint64_t size, uint64_t bo_flags,
+ struct anv_bo **bo);
VkResult anv_bo_cache_import(struct anv_device *device,
struct anv_bo_cache *cache,
- int fd, struct anv_bo **bo);
+ int fd, uint64_t bo_flags,
+ struct anv_bo **bo);
VkResult anv_bo_cache_export(struct anv_device *device,
struct anv_bo_cache *cache,
struct anv_bo *bo_in, int *fd_out);
struct anv_instance * instance;
uint32_t chipset_id;
+ bool no_hw;
char path[20];
const char * name;
struct gen_device_info info;
bool has_exec_fence;
bool has_syncobj;
bool has_syncobj_wait;
+ bool has_context_priority;
+ bool use_softpin;
+ bool has_context_isolation;
struct anv_device_extension_table supported_extensions;
struct anv_memory_heap heaps[VK_MAX_MEMORY_HEAPS];
} memory;
+ uint8_t driver_build_sha1[20];
uint8_t pipeline_cache_uuid[VK_UUID_SIZE];
uint8_t driver_uuid[VK_UUID_SIZE];
uint8_t device_uuid[VK_UUID_SIZE];
+ struct disk_cache * disk_cache;
+
struct wsi_device wsi_device;
int local_fd;
+ int master_fd;
};
struct anv_instance {
int physicalDeviceCount;
struct anv_physical_device physicalDevice;
+ bool pipeline_cache_enabled;
+
struct vk_debug_report_instance debug_report_callbacks;
};
struct anv_device * device;
- struct anv_state_pool * pool;
+ VkDeviceQueueCreateFlags flags;
};
struct anv_pipeline_cache {
anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
const void *key_data, uint32_t key_size,
const void *kernel_data, uint32_t kernel_size,
+ const void *constant_data,
+ uint32_t constant_data_size,
const struct brw_stage_prog_data *prog_data,
uint32_t prog_data_size,
const struct anv_pipeline_bind_map *bind_map);
+struct anv_shader_bin *
+anv_device_search_for_kernel(struct anv_device *device,
+ struct anv_pipeline_cache *cache,
+ const void *key_data, uint32_t key_size);
+
+struct anv_shader_bin *
+anv_device_upload_kernel(struct anv_device *device,
+ struct anv_pipeline_cache *cache,
+ const void *key_data, uint32_t key_size,
+ const void *kernel_data, uint32_t kernel_size,
+ const void *constant_data,
+ uint32_t constant_data_size,
+ const struct brw_stage_prog_data *prog_data,
+ uint32_t prog_data_size,
+ const struct anv_pipeline_bind_map *bind_map);
+
struct anv_device {
VK_LOADER_DATA _loader_data;
struct anv_instance * instance;
uint32_t chipset_id;
+ bool no_hw;
struct gen_device_info info;
struct isl_device isl_dev;
int context_id;
struct anv_device_extension_table enabled_extensions;
struct anv_dispatch_table dispatch;
+ pthread_mutex_t vma_mutex;
+ struct util_vma_heap vma_lo;
+ struct util_vma_heap vma_hi;
+ uint64_t vma_lo_available;
+ uint64_t vma_hi_available;
+
struct anv_bo_pool batch_bo_pool;
struct anv_bo_cache bo_cache;
struct anv_state_pool dynamic_state_pool;
struct anv_state_pool instruction_state_pool;
+ struct anv_state_pool binding_table_pool;
struct anv_state_pool surface_state_pool;
struct anv_bo workaround_bo;
struct anv_bo trivial_batch_bo;
+ struct anv_bo hiz_clear_bo;
- struct anv_pipeline_cache blorp_shader_cache;
+ struct anv_pipeline_cache default_pipeline_cache;
struct blorp_context blorp;
struct anv_state border_colors;
bool lost;
};
+static inline struct anv_state_pool *
+anv_binding_table_pool(struct anv_device *device)
+{
+ if (device->instance->physicalDevice.use_softpin)
+ return &device->binding_table_pool;
+ else
+ return &device->surface_state_pool;
+}
+
+static inline struct anv_state
+anv_binding_table_pool_alloc(struct anv_device *device) {
+ if (device->instance->physicalDevice.use_softpin)
+ return anv_state_pool_alloc(&device->binding_table_pool,
+ device->binding_table_pool.block_size, 0);
+ else
+ return anv_state_pool_alloc_back(&device->surface_state_pool);
+}
+
+static inline void
+anv_binding_table_pool_free(struct anv_device *device, struct anv_state state) {
+ anv_state_pool_free(anv_binding_table_pool(device), state);
+}
+
static void inline
anv_state_flush(struct anv_device *device, struct anv_state state)
{
int anv_gem_set_tiling(struct anv_device *device, uint32_t gem_handle,
uint32_t stride, uint32_t tiling);
int anv_gem_create_context(struct anv_device *device);
+bool anv_gem_has_context_priority(int fd);
int anv_gem_destroy_context(struct anv_device *device, int context);
+int anv_gem_set_context_param(int fd, int context, uint32_t param,
+ uint64_t value);
int anv_gem_get_context_param(int fd, int context, uint32_t param,
uint64_t *value);
int anv_gem_get_param(int fd, uint32_t param);
int anv_gem_get_tiling(struct anv_device *device, uint32_t gem_handle);
bool anv_gem_get_bit6_swizzle(int fd, uint32_t tiling);
int anv_gem_get_aperture(int fd, uint64_t *size);
-bool anv_gem_supports_48b_addresses(int fd);
int anv_gem_gpu_get_reset_stats(struct anv_device *device,
uint32_t *active, uint32_t *pending);
int anv_gem_handle_to_fd(struct anv_device *device, uint32_t gem_handle);
uint32_t *handles, uint32_t num_handles,
int64_t abs_timeout_ns, bool wait_all);
+bool anv_vma_alloc(struct anv_device *device, struct anv_bo *bo);
+void anv_vma_free(struct anv_device *device, struct anv_bo *bo);
+
VkResult anv_bo_init_new(struct anv_bo *bo, struct anv_device *device, uint64_t size);
struct anv_reloc_list {
uint32_t array_length;
struct drm_i915_gem_relocation_entry * relocs;
struct anv_bo ** reloc_bos;
+ struct set * deps;
};
VkResult anv_reloc_list_init(struct anv_reloc_list *list,
uint32_t offset;
};
+#define ANV_NULL_ADDRESS ((struct anv_address) { NULL, 0 })
+
+static inline bool
+anv_address_is_null(struct anv_address addr)
+{
+ return addr.bo == NULL && addr.offset == 0;
+}
+
+static inline uint64_t
+anv_address_physical(struct anv_address addr)
+{
+ if (addr.bo && (addr.bo->flags & EXEC_OBJECT_PINNED))
+ return gen_canonical_address(addr.bo->offset + addr.offset);
+ else
+ return gen_canonical_address(addr.offset);
+}
+
+static inline struct anv_address
+anv_address_add(struct anv_address addr, uint64_t offset)
+{
+ addr.offset += offset;
+ return addr;
+}
+
+static inline void
+write_reloc(const struct anv_device *device, void *p, uint64_t v, bool flush)
+{
+ unsigned reloc_size = 0;
+ if (device->info.gen >= 8) {
+ reloc_size = sizeof(uint64_t);
+ *(uint64_t *)p = gen_canonical_address(v);
+ } else {
+ reloc_size = sizeof(uint32_t);
+ *(uint32_t *)p = v;
+ }
+
+ if (flush && !device->info.has_llc)
+ gen_flush_range(p, reloc_size);
+}
+
static inline uint64_t
_anv_combine_address(struct anv_batch *batch, void *location,
const struct anv_address address, uint32_t delta)
.IndextoMOCSTables = 1 \
}
+/* Ice Lake MOCS defines are duplicates of Skylake MOCS defines. */
+#define GEN11_MOCS (struct GEN11_MEMORY_OBJECT_CONTROL_STATE) { \
+ /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */ \
+ .IndextoMOCSTables = 2 \
+ }
+
+#define GEN11_MOCS_PTE { \
+ /* TC=LLC/eLLC, LeCC=WB, LRUM=3, L3CC=WB */ \
+ .IndextoMOCSTables = 1 \
+ }
+
struct anv_device_memory {
struct anv_bo * bo;
struct anv_memory_type * type;
};
struct anv_descriptor_set_layout {
+ /* Descriptor set layouts can be destroyed at almost any time */
+ uint32_t ref_cnt;
+
/* Number of bindings in this descriptor set */
uint16_t binding_count;
struct anv_descriptor_set_binding_layout binding[0];
};
+static inline void
+anv_descriptor_set_layout_ref(struct anv_descriptor_set_layout *layout)
+{
+ assert(layout && layout->ref_cnt >= 1);
+ p_atomic_inc(&layout->ref_cnt);
+}
+
+static inline void
+anv_descriptor_set_layout_unref(struct anv_device *device,
+ struct anv_descriptor_set_layout *layout)
+{
+ assert(layout && layout->ref_cnt >= 1);
+ if (p_atomic_dec_zero(&layout->ref_cnt))
+ vk_free(&device->alloc, layout);
+}
+
struct anv_descriptor {
VkDescriptorType type;
};
struct anv_descriptor_set {
- const struct anv_descriptor_set_layout *layout;
+ struct anv_descriptor_set_layout *layout;
uint32_t size;
uint32_t buffer_count;
struct anv_buffer_view *buffer_views;
struct anv_buffer_view {
enum isl_format format; /**< VkBufferViewCreateInfo::format */
- struct anv_bo *bo;
- uint32_t offset; /**< Offset into bo. */
uint64_t range; /**< VkBufferViewCreateInfo::range */
+ struct anv_address address;
+
struct anv_state surface_state;
struct anv_state storage_surface_state;
struct anv_state writeonly_storage_surface_state;
struct anv_descriptor_template_entry entries[0];
};
-size_t
-anv_descriptor_set_binding_layout_get_hw_size(const struct anv_descriptor_set_binding_layout *binding);
-
size_t
anv_descriptor_set_layout_size(const struct anv_descriptor_set_layout *layout);
VkResult
anv_descriptor_set_create(struct anv_device *device,
struct anv_descriptor_pool *pool,
- const struct anv_descriptor_set_layout *layout,
+ struct anv_descriptor_set_layout *layout,
struct anv_descriptor_set **out_set);
void
struct anv_descriptor_pool *pool,
struct anv_descriptor_set *set);
+#define ANV_DESCRIPTOR_SET_SHADER_CONSTANTS (UINT8_MAX - 1)
#define ANV_DESCRIPTOR_SET_COLOR_ATTACHMENTS UINT8_MAX
struct anv_pipeline_binding {
VkBufferUsageFlags usage;
/* Set when bound */
- struct anv_bo * bo;
- VkDeviceSize offset;
+ struct anv_address address;
};
static inline uint64_t
pipe_bits |= ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;
pipe_bits |= ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;
break;
+ case VK_ACCESS_MEMORY_WRITE_BIT:
+ pipe_bits |= ANV_PIPE_FLUSH_BITS;
+ break;
default:
break; /* Nothing to do */
}
case VK_ACCESS_TRANSFER_READ_BIT:
pipe_bits |= ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
break;
+ case VK_ACCESS_MEMORY_READ_BIT:
+ pipe_bits |= ANV_PIPE_INVALIDATE_BITS;
+ break;
+ case VK_ACCESS_MEMORY_WRITE_BIT:
+ pipe_bits |= ANV_PIPE_FLUSH_BITS;
+ break;
default:
break; /* Nothing to do */
}
#define VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV ( \
VK_IMAGE_ASPECT_COLOR_BIT | \
- VK_IMAGE_ASPECT_PLANE_0_BIT_KHR | \
- VK_IMAGE_ASPECT_PLANE_1_BIT_KHR | \
- VK_IMAGE_ASPECT_PLANE_2_BIT_KHR)
+ VK_IMAGE_ASPECT_PLANE_0_BIT | \
+ VK_IMAGE_ASPECT_PLANE_1_BIT | \
+ VK_IMAGE_ASPECT_PLANE_2_BIT)
#define VK_IMAGE_ASPECT_PLANES_BITS_ANV ( \
- VK_IMAGE_ASPECT_PLANE_0_BIT_KHR | \
- VK_IMAGE_ASPECT_PLANE_1_BIT_KHR | \
- VK_IMAGE_ASPECT_PLANE_2_BIT_KHR)
+ VK_IMAGE_ASPECT_PLANE_0_BIT | \
+ VK_IMAGE_ASPECT_PLANE_1_BIT | \
+ VK_IMAGE_ASPECT_PLANE_2_BIT)
struct anv_vertex_binding {
struct anv_buffer * buffer;
/* Push constant data provided by the client through vkPushConstants */
uint8_t client_data[MAX_PUSH_CONSTANTS_SIZE];
+ /* Used for vkCmdDispatchBase */
+ uint32_t base_work_group_id[3];
+
/* Image data for image_load_store on pre-SKL */
struct brw_image_param images[MAX_IMAGES];
};
*
* This address is relative to the start of the BO.
*/
- uint64_t address;
+ struct anv_address address;
/* Address of the aux surface, if any
*
- * This field is 0 if and only if no aux surface exists.
+ * This field is ANV_NULL_ADDRESS if and only if no aux surface exists.
*
- * This address is relative to the start of the BO. On gen7, the bottom 12
- * bits of this address include extra aux information.
+ * With the exception of gen8, the bottom 12 bits of this address' offset
+ * include extra aux information.
*/
- uint64_t aux_address;
+ struct anv_address aux_address;
+ /* Address of the clear color, if any
+ *
+ * This address is relative to the start of the BO.
+ */
+ struct anv_address clear_address;
};
/**
VkImageLayout current_layout;
VkImageAspectFlags pending_clear_aspects;
+ VkImageAspectFlags pending_load_aspects;
bool fast_clear;
VkClearValue clear_value;
bool clear_color_is_zero_one;
bool clear_color_is_zero;
+
+ /* When multiview is active, attachments with a renderpass clear
+ * operation have their respective layers cleared on the first
+ * subpass that uses them, and only in that subpass. We keep track
+ * of this using a bitfield to indicate which layers of an attachment
+ * have not been cleared yet when multiview is active.
+ */
+ uint32_t pending_clear_views;
};
/** State tracking for particular pipeline bind point
*/
struct anv_cmd_pipeline_state {
struct anv_pipeline *pipeline;
+ struct anv_pipeline_layout *layout;
struct anv_descriptor_set *descriptors[MAX_SETS];
uint32_t dynamic_offsets[MAX_DYNAMIC_BUFFERS];
uint32_t vb_dirty;
struct anv_dynamic_state dynamic;
+
+ struct {
+ struct anv_buffer *index_buffer;
+ uint32_t index_type; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */
+ uint32_t index_offset;
+ } gen7;
};
/** State tracking for compute pipeline
struct anv_cmd_pipeline_state base;
bool pipeline_dirty;
+
+ struct anv_address num_workgroups;
};
/** State required while building cmd buffer */
struct anv_cmd_compute_state compute;
enum anv_pipe_bits pending_pipe_bits;
- uint32_t num_workgroups_offset;
- struct anv_bo *num_workgroups_bo;
VkShaderStageFlags descriptors_dirty;
VkShaderStageFlags push_constants_dirty;
* is one of the states in render_pass_states.
*/
struct anv_state null_surface_state;
-
- struct {
- struct anv_buffer * index_buffer;
- uint32_t index_type; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */
- uint32_t index_offset;
- } gen7;
};
struct anv_cmd_pool {
struct anv_state
anv_cmd_buffer_cs_push_constants(struct anv_cmd_buffer *cmd_buffer);
-void anv_cmd_buffer_clear_subpass(struct anv_cmd_buffer *cmd_buffer);
void anv_cmd_buffer_resolve_subpass(struct anv_cmd_buffer *cmd_buffer);
const struct anv_image_view *
ANV_FENCE_TYPE_NONE = 0,
ANV_FENCE_TYPE_BO,
ANV_FENCE_TYPE_SYNCOBJ,
+ ANV_FENCE_TYPE_WSI,
};
enum anv_bo_fence_state {
/** DRM syncobj handle for syncobj-based fences */
uint32_t syncobj;
+
+ /** WSI fence */
+ struct wsi_fence *fence_wsi;
};
};
struct anv_state kernel;
uint32_t kernel_size;
+ struct anv_state constant_data;
+ uint32_t constant_data_size;
+
const struct brw_stage_prog_data *prog_data;
uint32_t prog_data_size;
anv_shader_bin_create(struct anv_device *device,
const void *key, uint32_t key_size,
const void *kernel, uint32_t kernel_size,
+ const void *constant_data, uint32_t constant_data_size,
const struct brw_stage_prog_data *prog_data,
uint32_t prog_data_size, const void *prog_data_param,
const struct anv_pipeline_bind_map *bind_map);
struct anv_dynamic_state dynamic_state;
struct anv_subpass * subpass;
- struct anv_pipeline_layout * layout;
bool needs_data_cache;
struct anv_state blend_state;
uint32_t vb_used;
- uint32_t binding_stride[MAX_VBS];
- bool instancing_enable[MAX_VBS];
+ struct anv_pipeline_vertex_binding {
+ uint32_t stride;
+ bool instanced;
+ uint32_t instance_divisor;
+ } vb[MAX_VBS];
+
bool primitive_restart;
uint32_t topology;
anv_pipeline_compile_cs(struct anv_pipeline *pipeline,
struct anv_pipeline_cache *cache,
const VkComputePipelineCreateInfo *info,
- struct anv_shader_module *module,
+ const struct anv_shader_module *module,
const char *entrypoint,
const VkSpecializationInfo *spec_info);
switch (aspect_mask) {
case VK_IMAGE_ASPECT_COLOR_BIT:
case VK_IMAGE_ASPECT_DEPTH_BIT:
- case VK_IMAGE_ASPECT_PLANE_0_BIT_KHR:
+ case VK_IMAGE_ASPECT_PLANE_0_BIT:
return 0;
case VK_IMAGE_ASPECT_STENCIL_BIT:
if ((image_aspects & VK_IMAGE_ASPECT_DEPTH_BIT) == 0)
return 0;
/* Fall-through */
- case VK_IMAGE_ASPECT_PLANE_1_BIT_KHR:
+ case VK_IMAGE_ASPECT_PLANE_1_BIT:
return 1;
- case VK_IMAGE_ASPECT_PLANE_2_BIT_KHR:
+ case VK_IMAGE_ASPECT_PLANE_2_BIT:
return 2;
default:
/* Purposefully assert with depth/stencil aspects. */
if (aspect_mask & (VK_IMAGE_ASPECT_COLOR_BIT |
VK_IMAGE_ASPECT_DEPTH_BIT |
VK_IMAGE_ASPECT_STENCIL_BIT |
- VK_IMAGE_ASPECT_PLANE_0_BIT_KHR))
+ VK_IMAGE_ASPECT_PLANE_0_BIT))
planes++;
- if (aspect_mask & VK_IMAGE_ASPECT_PLANE_1_BIT_KHR)
+ if (aspect_mask & VK_IMAGE_ASPECT_PLANE_1_BIT)
planes++;
- if (aspect_mask & VK_IMAGE_ASPECT_PLANE_2_BIT_KHR)
+ if (aspect_mask & VK_IMAGE_ASPECT_PLANE_2_BIT)
+ planes++;
+
+ if ((aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0 &&
+ (aspect_mask & VK_IMAGE_ASPECT_STENCIL_BIT) != 0)
planes++;
return planes;
{
if (image_aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV) {
if (_mesa_bitcount(image_aspects) > 1)
- return VK_IMAGE_ASPECT_PLANE_0_BIT_KHR << plane;
+ return VK_IMAGE_ASPECT_PLANE_0_BIT << plane;
return VK_IMAGE_ASPECT_COLOR_BIT;
}
if (image_aspects & VK_IMAGE_ASPECT_DEPTH_BIT)
VkImageUsageFlags usage; /**< Superset of VkImageCreateInfo::usage. */
VkImageTiling tiling; /** VkImageCreateInfo::tiling */
+ /** True if this is needs to be bound to an appropriately tiled BO.
+ *
+ * When not using modifiers, consumers such as X11, Wayland, and KMS need
+ * the tiling passed via I915_GEM_SET_TILING. When exporting these buffers
+ * we require a dedicated allocation so that we can know to allocate a
+ * tiled buffer.
+ */
+ bool needs_set_tiling;
+
/**
- * DRM format modifier for this image or DRM_FORMAT_MOD_INVALID.
+ * Must be DRM_FORMAT_MOD_INVALID unless tiling is
+ * VK_IMAGE_TILING_DRM_FORMAT_MODIFIER_EXT.
*/
uint64_t drm_format_mod;
/**
* BO associated with this plane, set when bound.
*/
- struct anv_bo *bo;
- VkDeviceSize bo_offset;
+ struct anv_address address;
/**
* When destroying the image, also free the bo.
} planes[3];
};
+/* The ordering of this enum is important */
+enum anv_fast_clear_type {
+ /** Image does not have/support any fast-clear blocks */
+ ANV_FAST_CLEAR_NONE = 0,
+ /** Image has/supports fast-clear but only to the default value */
+ ANV_FAST_CLEAR_DEFAULT_VALUE = 1,
+ /** Image has/supports fast-clear with an arbitrary fast-clear value */
+ ANV_FAST_CLEAR_ANY = 2,
+};
+
/* Returns the number of auxiliary buffer levels attached to an image. */
static inline uint8_t
anv_image_aux_levels(const struct anv_image * const image,
}
}
-static inline unsigned
-anv_fast_clear_state_entry_size(const struct anv_device *device)
+static inline struct anv_address
+anv_image_get_clear_color_addr(const struct anv_device *device,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect)
{
- assert(device);
- /* Entry contents:
- * +--------------------------------------------+
- * | clear value dword(s) | needs resolve dword |
- * +--------------------------------------------+
- */
+ assert(image->aspects & VK_IMAGE_ASPECT_ANY_COLOR_BIT_ANV);
- /* Ensure that the needs resolve dword is in fact dword-aligned to enable
- * GPU memcpy operations.
- */
- assert(device->isl_dev.ss.clear_value_size % 4 == 0);
- return device->isl_dev.ss.clear_value_size + 4;
+ uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+ return anv_address_add(image->planes[plane].address,
+ image->planes[plane].fast_clear_state_offset);
}
static inline struct anv_address
-anv_image_get_clear_color_addr(const struct anv_device *device,
- const struct anv_image *image,
- VkImageAspectFlagBits aspect,
- unsigned level)
+anv_image_get_fast_clear_type_addr(const struct anv_device *device,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect)
{
- uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
- return (struct anv_address) {
- .bo = image->planes[plane].bo,
- .offset = image->planes[plane].bo_offset +
- image->planes[plane].fast_clear_state_offset +
- anv_fast_clear_state_entry_size(device) * level,
- };
+ struct anv_address addr =
+ anv_image_get_clear_color_addr(device, image, aspect);
+
+ const unsigned clear_color_state_size = device->info.gen >= 10 ?
+ device->isl_dev.ss.clear_color_state_size :
+ device->isl_dev.ss.clear_value_size;
+ addr.offset += clear_color_state_size;
+ return addr;
}
static inline struct anv_address
-anv_image_get_needs_resolve_addr(const struct anv_device *device,
- const struct anv_image *image,
- VkImageAspectFlagBits aspect,
- unsigned level)
+anv_image_get_compression_state_addr(const struct anv_device *device,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect,
+ uint32_t level, uint32_t array_layer)
{
+ assert(level < anv_image_aux_levels(image, aspect));
+ assert(array_layer < anv_image_aux_layers(image, aspect, level));
+ UNUSED uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+ assert(image->planes[plane].aux_usage == ISL_AUX_USAGE_CCS_E);
+
struct anv_address addr =
- anv_image_get_clear_color_addr(device, image, aspect, level);
- addr.offset += device->isl_dev.ss.clear_value_size;
+ anv_image_get_fast_clear_type_addr(device, image, aspect);
+ addr.offset += 4; /* Go past the fast clear type */
+
+ if (image->type == VK_IMAGE_TYPE_3D) {
+ for (uint32_t l = 0; l < level; l++)
+ addr.offset += anv_minify(image->extent.depth, l) * 4;
+ } else {
+ addr.offset += level * image->array_size * 4;
+ }
+ addr.offset += array_layer * 4;
+
return addr;
}
}
void
-anv_gen8_hiz_op_resolve(struct anv_cmd_buffer *cmd_buffer,
- const struct anv_image *image,
- enum blorp_hiz_op op);
-void
-anv_ccs_resolve(struct anv_cmd_buffer * const cmd_buffer,
- const struct anv_image * const image,
- VkImageAspectFlagBits aspect,
- const uint8_t level,
- const uint32_t start_layer, const uint32_t layer_count,
- const enum blorp_fast_clear_op op);
+anv_cmd_buffer_mark_image_written(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect,
+ enum isl_aux_usage aux_usage,
+ uint32_t level,
+ uint32_t base_layer,
+ uint32_t layer_count);
void
-anv_image_fast_clear(struct anv_cmd_buffer *cmd_buffer,
- const struct anv_image *image,
- VkImageAspectFlagBits aspect,
- const uint32_t base_level, const uint32_t level_count,
- const uint32_t base_layer, uint32_t layer_count);
+anv_image_clear_color(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect,
+ enum isl_aux_usage aux_usage,
+ enum isl_format format, struct isl_swizzle swizzle,
+ uint32_t level, uint32_t base_layer, uint32_t layer_count,
+ VkRect2D area, union isl_color_value clear_color);
+void
+anv_image_clear_depth_stencil(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlags aspects,
+ enum isl_aux_usage depth_aux_usage,
+ uint32_t level,
+ uint32_t base_layer, uint32_t layer_count,
+ VkRect2D area,
+ float depth_value, uint8_t stencil_value);
+void
+anv_image_hiz_op(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect, uint32_t level,
+ uint32_t base_layer, uint32_t layer_count,
+ enum isl_aux_op hiz_op);
+void
+anv_image_hiz_clear(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlags aspects,
+ uint32_t level,
+ uint32_t base_layer, uint32_t layer_count,
+ VkRect2D area, uint8_t stencil_value);
+void
+anv_image_mcs_op(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect,
+ uint32_t base_layer, uint32_t layer_count,
+ enum isl_aux_op mcs_op, union isl_color_value *clear_value,
+ bool predicate);
+void
+anv_image_ccs_op(struct anv_cmd_buffer *cmd_buffer,
+ const struct anv_image *image,
+ VkImageAspectFlagBits aspect, uint32_t level,
+ uint32_t base_layer, uint32_t layer_count,
+ enum isl_aux_op ccs_op, union isl_color_value *clear_value,
+ bool predicate);
void
anv_image_copy_to_shadow(struct anv_cmd_buffer *cmd_buffer,
const VkImageAspectFlagBits aspect,
const VkImageLayout layout);
+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);
+
/* This is defined as a macro so that it works for both
* VkImageSubresourceRange and VkImageSubresourceLayers
*/
void anv_fill_buffer_surface_state(struct anv_device *device,
struct anv_state state,
enum isl_format format,
- uint32_t offset, uint32_t range,
- uint32_t stride);
+ struct anv_address address,
+ uint32_t range, uint32_t stride);
+
+static inline void
+anv_clear_color_from_att_state(union isl_color_value *clear_color,
+ const struct anv_attachment_state *att_state,
+ const struct anv_image_view *iview)
+{
+ const struct isl_format_layout *view_fmtl =
+ isl_format_get_layout(iview->planes[0].isl.format);
+
+#define COPY_CLEAR_COLOR_CHANNEL(c, i) \
+ if (view_fmtl->channels.c.bits) \
+ clear_color->u32[i] = att_state->clear_value.color.uint32[i]
+
+ COPY_CLEAR_COLOR_CHANNEL(r, 0);
+ COPY_CLEAR_COLOR_CHANNEL(g, 1);
+ COPY_CLEAR_COLOR_CHANNEL(b, 2);
+ COPY_CLEAR_COLOR_CHANNEL(a, 3);
+
+#undef COPY_CLEAR_COLOR_CHANNEL
+}
struct anv_ycbcr_conversion {
const struct anv_format * format;
- VkSamplerYcbcrModelConversionKHR ycbcr_model;
- VkSamplerYcbcrRangeKHR ycbcr_range;
+ VkSamplerYcbcrModelConversion ycbcr_model;
+ VkSamplerYcbcrRange ycbcr_range;
VkComponentSwizzle mapping[4];
- VkChromaLocationKHR chroma_offsets[2];
+ VkChromaLocation chroma_offsets[2];
VkFilter chroma_filter;
bool chroma_reconstruction;
};
struct anv_image_view * attachments[0];
};
+struct anv_subpass_attachment {
+ VkImageUsageFlagBits usage;
+ uint32_t attachment;
+ VkImageLayout layout;
+};
+
struct anv_subpass {
uint32_t attachment_count;
* A pointer to all attachment references used in this subpass.
* Only valid if ::attachment_count > 0.
*/
- VkAttachmentReference * attachments;
+ struct anv_subpass_attachment * attachments;
uint32_t input_count;
- VkAttachmentReference * input_attachments;
+ struct anv_subpass_attachment * input_attachments;
uint32_t color_count;
- VkAttachmentReference * color_attachments;
- VkAttachmentReference * resolve_attachments;
+ struct anv_subpass_attachment * color_attachments;
+ struct anv_subpass_attachment * resolve_attachments;
- VkAttachmentReference depth_stencil_attachment;
+ struct anv_subpass_attachment * depth_stencil_attachment;
uint32_t view_mask;
ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_semaphore, VkSemaphore)
ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_shader_module, VkShaderModule)
ANV_DEFINE_NONDISP_HANDLE_CASTS(vk_debug_report_callback, VkDebugReportCallbackEXT)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_ycbcr_conversion, VkSamplerYcbcrConversionKHR)
+ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_ycbcr_conversion, VkSamplerYcbcrConversion)
/* Gen-specific function declarations */
#ifdef genX
# define genX(x) gen10_##x
# include "anv_genX.h"
# undef genX
+# define genX(x) gen11_##x
+# include "anv_genX.h"
+# undef genX
#endif
#endif /* ANV_PRIVATE_H */