#include "common/gen_clflush.h"
#include "common/gen_decoder.h"
#include "common/gen_gem.h"
+#include "common/gen_l3_config.h"
#include "dev/gen_device_info.h"
#include "blorp/blorp.h"
#include "compiler/brw_compiler.h"
#include "util/xmlconfig.h"
#include "vk_alloc.h"
#include "vk_debug_report.h"
+#include "vk_object.h"
/* Pre-declarations needed for WSI entrypoints */
struct wl_surface;
struct anv_instance;
struct gen_aux_map_context;
-struct gen_l3_config;
struct gen_perf_config;
+struct gen_perf_counter_pass;
+struct gen_perf_query_result;
#include <vulkan/vulkan.h>
#include <vulkan/vulkan_intel.h>
#define MAX_PUSH_DESCRIPTORS 32 /* Minimum requirement */
#define MAX_INLINE_UNIFORM_BLOCK_SIZE 4096
#define MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS 32
+/* We need 16 for UBO block reads to work and 32 for push UBOs. However, we
+ * use 64 here to avoid cache issues. This could most likely bring it back to
+ * 32 if we had different virtual addresses for the different views on a given
+ * GEM object.
+ */
+#define ANV_UBO_ALIGNMENT 64
+#define ANV_SSBO_BOUNDS_CHECK_ALIGNMENT 4
+#define MAX_VIEWS_FOR_PRIMITIVE_REPLICATION 16
/* From the Skylake PRM Vol. 7 "Binding Table Surface State Model":
*
*/
#define ANV_PREDICATE_RESULT_REG 0x2678 /* MI_ALU_REG15 */
+/* We reserve this MI ALU register to pass around an offset computed from
+ * VkPerformanceQuerySubmitInfoKHR::counterPassIndex VK_KHR_performance_query.
+ * Other code which uses the MI ALU should leave it alone.
+ */
+#define ANV_PERF_QUERY_OFFSET_REG 0x2670 /* MI_ALU_REG14 */
+
/* For gen12 we set the streamout buffers using 4 separate commands
* (3DSTATE_SO_BUFFER_INDEX_*) instead of 3DSTATE_SO_BUFFER. However the layout
* of the 3DSTATE_SO_BUFFER_INDEX_* commands is identical to that of
return v - (v % a);
}
+static inline uint32_t
+align_down_u32(uint32_t v, uint32_t a)
+{
+ assert(a != 0 && a == (a & -a));
+ return v & ~(a - 1);
+}
+
static inline uint32_t
align_u32(uint32_t v, uint32_t a)
{
assert(a != 0 && a == (a & -a));
- return (v + a - 1) & ~(a - 1);
+ return align_down_u32(v + a - 1, a);
}
static inline uint64_t
-align_u64(uint64_t v, uint64_t a)
+align_down_u64(uint64_t v, uint64_t a)
{
assert(a != 0 && a == (a & -a));
- return (v + a - 1) & ~(a - 1);
+ return v & ~(a - 1);
+}
+
+static inline uint64_t
+align_u64(uint64_t v, uint64_t a)
+{
+ return align_down_u64(v + a - 1, a);
}
static inline int32_t
#define vk_error(error) __vk_errorf(NULL, NULL,\
VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT,\
error, __FILE__, __LINE__, NULL)
-#define vk_errorv(instance, obj, error, format, args)\
- __vk_errorv(instance, obj, REPORT_OBJECT_TYPE(obj), error,\
- __FILE__, __LINE__, format, args)
-#define vk_errorf(instance, obj, error, format, ...)\
+#define vk_errorfi(instance, obj, error, format, ...)\
__vk_errorf(instance, obj, REPORT_OBJECT_TYPE(obj), error,\
__FILE__, __LINE__, format, ## __VA_ARGS__)
+#define vk_errorf(device, obj, error, format, ...)\
+ vk_errorfi(anv_device_instance_or_null(device),\
+ obj, error, format, ## __VA_ARGS__)
#else
#define vk_error(error) error
-#define vk_errorf(instance, obj, error, format, ...) error
+#define vk_errorfi(instance, obj, error, format, ...) error
+#define vk_errorf(device, obj, error, format, ...) error
#endif
/**
#define anv_debug_ignored_stype(sType) \
intel_logd("%s: ignored VkStructureType %u\n", __func__, (sType))
-void __anv_perf_warn(struct anv_instance *instance, const void *object,
+void __anv_perf_warn(struct anv_device *device, const void *object,
VkDebugReportObjectTypeEXT type, const char *file,
int line, const char *format, ...)
anv_printflike(6, 7);
*/
uint64_t offset;
+ /** Size of the buffer not including implicit aux */
uint64_t size;
/* Map for internally mapped BOs.
*/
void *map;
+ /** Size of the implicit CCS range at the end of the buffer
+ *
+ * On Gen12, CCS data is always a direct 1/256 scale-down. A single 64K
+ * page of main surface data maps to a 256B chunk of CCS data and that
+ * mapping is provided on TGL-LP by the AUX table which maps virtual memory
+ * addresses in the main surface to virtual memory addresses for CCS data.
+ *
+ * Because we can't change these maps around easily and because Vulkan
+ * allows two VkImages to be bound to overlapping memory regions (as long
+ * as the app is careful), it's not feasible to make this mapping part of
+ * the image. (On Gen11 and earlier, the mapping was provided via
+ * RENDER_SURFACE_STATE so each image had its own main -> CCS mapping.)
+ * Instead, we attach the CCS data directly to the buffer object and setup
+ * the AUX table mapping at BO creation time.
+ *
+ * This field is for internal tracking use by the BO allocator only and
+ * should not be touched by other parts of the code. If something wants to
+ * know if a BO has implicit CCS data, it should instead look at the
+ * has_implicit_ccs boolean below.
+ *
+ * This data is not included in maps of this buffer.
+ */
+ uint32_t _ccs_size;
+
/** Flags to pass to the kernel through drm_i915_exec_object2::flags */
uint32_t flags;
/** See also ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS */
bool has_client_visible_address:1;
+
+ /** True if this BO has implicit CCS data attached to it */
+ bool has_implicit_ccs:1;
};
static inline struct anv_bo *
struct anv_state_pool {
struct anv_block_pool block_pool;
+ /* Offset into the relevant state base address where the state pool starts
+ * allocating memory.
+ */
+ int32_t start_offset;
+
struct anv_state_table table;
/* The size of blocks which will be allocated from the block pool */
struct anv_fixed_size_state_pool buckets[ANV_STATE_BUCKETS];
};
-struct anv_state_stream_block;
+struct anv_state_reserved_pool {
+ struct anv_state_pool *pool;
+ union anv_free_list reserved_blocks;
+ uint32_t count;
+};
struct anv_state_stream {
struct anv_state_pool *state_pool;
uint32_t next;
/* List of all blocks allocated from this pool */
- struct anv_state_stream_block *block_list;
+ struct util_dynarray all_blocks;
};
/* The block_pool functions exported for testing only. The block pool should
uint32_t block_size, uint32_t *padding);
int32_t anv_block_pool_alloc_back(struct anv_block_pool *pool,
uint32_t block_size);
-void* anv_block_pool_map(struct anv_block_pool *pool, int32_t offset);
+void* anv_block_pool_map(struct anv_block_pool *pool, int32_t offset, uint32_t
+size);
VkResult anv_state_pool_init(struct anv_state_pool *pool,
struct anv_device *device,
- uint64_t start_address,
+ uint64_t base_address,
+ int32_t start_offset,
uint32_t block_size);
void anv_state_pool_finish(struct anv_state_pool *pool);
struct anv_state anv_state_pool_alloc(struct anv_state_pool *pool,
struct anv_state anv_state_stream_alloc(struct anv_state_stream *stream,
uint32_t size, uint32_t alignment);
+void anv_state_reserved_pool_init(struct anv_state_reserved_pool *pool,
+ struct anv_state_pool *parent,
+ uint32_t count, uint32_t size,
+ uint32_t alignment);
+void anv_state_reserved_pool_finish(struct anv_state_reserved_pool *pool);
+struct anv_state anv_state_reserved_pool_alloc(struct anv_state_reserved_pool *pool);
+void anv_state_reserved_pool_free(struct anv_state_reserved_pool *pool,
+ struct anv_state state);
+
VkResult anv_state_table_init(struct anv_state_table *table,
struct anv_device *device,
uint32_t initial_entries);
VkDeviceSize size;
VkMemoryHeapFlags flags;
- /* Driver-internal book-keeping */
- VkDeviceSize used;
+ /** Driver-internal book-keeping.
+ *
+ * Align it to 64 bits to make atomic operations faster on 32 bit platforms.
+ */
+ VkDeviceSize used __attribute__ ((aligned (8)));
};
struct anv_physical_device {
- VK_LOADER_DATA _loader_data;
+ struct vk_object_base base;
+
+ /* Link in anv_instance::physical_devices */
+ struct list_head link;
struct anv_instance * instance;
- uint32_t chipset_id;
bool no_hw;
char path[20];
const char * name;
bool has_context_priority;
bool has_context_isolation;
bool has_mem_available;
+ bool has_mmap_offset;
uint64_t gtt_size;
bool use_softpin;
bool always_use_bindless;
+ bool use_call_secondary;
/** True if we can access buffers using A64 messages */
bool has_a64_buffer_access;
/** True if we can use bindless access for samplers */
bool has_bindless_samplers;
+ /** True if we can read the GPU timestamp register
+ *
+ * When running in a virtual context, the timestamp register is unreadable
+ * on Gen12+.
+ */
+ bool has_reg_timestamp;
+
+ /** True if this device has implicit AUX
+ *
+ * If true, CCS is handled as an implicit attachment to the BO rather than
+ * as an explicitly bound surface.
+ */
+ bool has_implicit_ccs;
+
bool always_flush_cache;
struct anv_device_extension_table supported_extensions;
- struct anv_physical_device_dispatch_table dispatch;
uint32_t eu_total;
uint32_t subslice_total;
};
struct anv_instance {
- VK_LOADER_DATA _loader_data;
+ struct vk_object_base base;
VkAllocationCallbacks alloc;
struct anv_instance_extension_table enabled_extensions;
struct anv_instance_dispatch_table dispatch;
+ struct anv_physical_device_dispatch_table physical_device_dispatch;
struct anv_device_dispatch_table device_dispatch;
- int physicalDeviceCount;
- struct anv_physical_device physicalDevice;
+ bool physical_devices_enumerated;
+ struct list_head physical_devices;
bool pipeline_cache_enabled;
*/
uintptr_t * fence_bos;
+ int perf_query_pass;
+
const VkAllocationCallbacks * alloc;
VkSystemAllocationScope alloc_scope;
};
struct anv_queue {
- VK_LOADER_DATA _loader_data;
+ struct vk_object_base base;
struct anv_device * device;
};
struct anv_pipeline_cache {
+ struct vk_object_base base;
struct anv_device * device;
pthread_mutex_t mutex;
struct hash_table * nir_cache;
struct hash_table * cache;
+
+ bool external_sync;
};
struct nir_xfb_info;
void anv_pipeline_cache_init(struct anv_pipeline_cache *cache,
struct anv_device *device,
- bool cache_enabled);
+ bool cache_enabled,
+ bool external_sync);
void anv_pipeline_cache_finish(struct anv_pipeline_cache *cache);
struct anv_shader_bin *
const void *key, uint32_t key_size);
struct anv_shader_bin *
anv_pipeline_cache_upload_kernel(struct anv_pipeline_cache *cache,
+ gl_shader_stage stage,
const void *key_data, uint32_t key_size,
const void *kernel_data, uint32_t kernel_size,
const void *constant_data,
struct anv_shader_bin *
anv_device_upload_kernel(struct anv_device *device,
struct anv_pipeline_cache *cache,
+ gl_shader_stage stage,
const void *key_data, uint32_t key_size,
const void *kernel_data, uint32_t kernel_size,
const void *constant_data,
const struct nir_shader *nir,
unsigned char sha1_key[20]);
-struct anv_device {
- VK_LOADER_DATA _loader_data;
+struct anv_address {
+ struct anv_bo *bo;
+ uint32_t offset;
+};
- VkAllocationCallbacks alloc;
+struct anv_device {
+ struct vk_device vk;
- struct anv_instance * instance;
struct anv_physical_device * physical;
- uint32_t chipset_id;
bool no_hw;
struct gen_device_info info;
struct isl_device isl_dev;
struct anv_state_pool binding_table_pool;
struct anv_state_pool surface_state_pool;
+ struct anv_state_reserved_pool custom_border_colors;
+
+ /** BO used for various workarounds
+ *
+ * There are a number of workarounds on our hardware which require writing
+ * data somewhere and it doesn't really matter where. For that, we use
+ * this BO and just write to the first dword or so.
+ *
+ * We also need to be able to handle NULL buffers bound as pushed UBOs.
+ * For that, we use the high bytes (>= 1024) of the workaround BO.
+ */
struct anv_bo * workaround_bo;
+ struct anv_address workaround_address;
+
struct anv_bo * trivial_batch_bo;
struct anv_bo * hiz_clear_bo;
+ struct anv_state null_surface_state;
struct anv_pipeline_cache default_pipeline_cache;
struct blorp_context blorp;
uint64_t perf_metric; /* 0 if unset */
struct gen_aux_map_context *aux_map_ctx;
+
+ struct gen_debug_block_frame *debug_frame_desc;
};
+static inline struct anv_instance *
+anv_device_instance_or_null(const struct anv_device *device)
+{
+ return device ? device->physical->instance : NULL;
+}
+
static inline struct anv_state_pool *
anv_binding_table_pool(struct anv_device *device)
{
/** Has an address which is visible to the client */
ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS = (1 << 8),
+
+ /** This buffer has implicit CCS data attached to it */
+ ANV_BO_ALLOC_IMPLICIT_CCS = (1 << 9),
};
VkResult anv_device_alloc_bo(struct anv_device *device, uint64_t size,
void* anv_gem_mmap(struct anv_device *device,
uint32_t gem_handle, uint64_t offset, uint64_t size, uint32_t flags);
-void anv_gem_munmap(void *p, uint64_t size);
+void anv_gem_munmap(struct anv_device *device, void *p, uint64_t size);
uint32_t anv_gem_create(struct anv_device *device, uint64_t size);
void anv_gem_close(struct anv_device *device, uint32_t gem_handle);
uint32_t anv_gem_userptr(struct anv_device *device, void *mem, size_t size);
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);
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);
-int anv_gem_reg_read(struct anv_device *device,
- uint32_t offset, uint64_t *result);
+int anv_gem_reg_read(int fd, uint32_t offset, uint64_t *result);
uint32_t anv_gem_fd_to_handle(struct anv_device *device, int fd);
int anv_gem_set_caching(struct anv_device *device, uint32_t gem_handle, uint32_t caching);
int anv_gem_set_domain(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,
- uint64_t client_address);
-void anv_vma_free(struct anv_device *device, struct anv_bo *bo);
+uint64_t anv_vma_alloc(struct anv_device *device,
+ uint64_t size, uint64_t align,
+ enum anv_bo_alloc_flags alloc_flags,
+ uint64_t client_address);
+void anv_vma_free(struct anv_device *device,
+ uint64_t address, uint64_t size);
struct anv_reloc_list {
uint32_t num_relocs;
struct anv_batch {
const VkAllocationCallbacks * alloc;
+ struct anv_address start_addr;
+
void * start;
void * end;
void * next;
void anv_batch_emit_batch(struct anv_batch *batch, struct anv_batch *other);
uint64_t anv_batch_emit_reloc(struct anv_batch *batch,
void *location, struct anv_bo *bo, uint32_t offset);
+struct anv_address anv_batch_address(struct anv_batch *batch, void *batch_location);
+
+static inline void
+anv_batch_set_storage(struct anv_batch *batch, struct anv_address addr,
+ void *map, size_t size)
+{
+ batch->start_addr = addr;
+ batch->next = batch->start = map;
+ batch->end = map + size;
+}
static inline VkResult
anv_batch_set_error(struct anv_batch *batch, VkResult error)
return batch->status != VK_SUCCESS;
}
-struct anv_address {
- struct anv_bo *bo;
- uint32_t offset;
-};
-
#define ANV_NULL_ADDRESS ((struct anv_address) { NULL, 0 })
static inline bool
_dst = NULL; \
}))
+/* #define __gen_get_batch_dwords anv_batch_emit_dwords */
+/* #define __gen_get_batch_address anv_batch_address */
+/* #define __gen_address_value anv_address_physical */
+/* #define __gen_address_offset anv_address_add */
+
struct anv_device_memory {
+ struct vk_object_base base;
+
struct list_head link;
struct anv_bo * bo;
bool sampler);
struct anv_descriptor_set_layout {
+ struct vk_object_base base;
+
/* Descriptor set layouts can be destroyed at almost any time */
uint32_t ref_cnt;
struct anv_descriptor_set_binding_layout binding[0];
};
+void anv_descriptor_set_layout_destroy(struct anv_device *device,
+ struct anv_descriptor_set_layout *layout);
+
static inline void
anv_descriptor_set_layout_ref(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);
+ anv_descriptor_set_layout_destroy(device, layout);
}
struct anv_descriptor {
};
struct anv_descriptor_set {
+ struct vk_object_base base;
+
struct anv_descriptor_pool *pool;
struct anv_descriptor_set_layout *layout;
+
+ /* Amount of space occupied in the the pool by this descriptor set. It can
+ * be larger than the size of the descriptor set.
+ */
uint32_t size;
/* State relative to anv_descriptor_pool::bo */
};
struct anv_buffer_view {
+ struct vk_object_base base;
+
enum isl_format format; /**< VkBufferViewCreateInfo::format */
uint64_t range; /**< VkBufferViewCreateInfo::range */
};
struct anv_descriptor_pool {
+ struct vk_object_base base;
+
uint32_t size;
uint32_t next;
uint32_t free_list;
};
struct anv_descriptor_update_template {
+ struct vk_object_base base;
+
VkPipelineBindPoint bind_point;
/* The descriptor set this template corresponds to. This value is only
};
struct anv_pipeline_layout {
+ struct vk_object_base base;
+
struct {
struct anv_descriptor_set_layout *layout;
uint32_t dynamic_offset_start;
};
struct anv_buffer {
+ struct vk_object_base base;
+
struct anv_device * device;
VkDeviceSize size;
}
enum anv_cmd_dirty_bits {
- ANV_CMD_DIRTY_DYNAMIC_VIEWPORT = 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
- ANV_CMD_DIRTY_DYNAMIC_SCISSOR = 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
- ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH = 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
- ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS = 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
- ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS = 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
- ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS = 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
- ANV_CMD_DIRTY_PIPELINE = 1 << 9,
- ANV_CMD_DIRTY_INDEX_BUFFER = 1 << 10,
- ANV_CMD_DIRTY_RENDER_TARGETS = 1 << 11,
- ANV_CMD_DIRTY_XFB_ENABLE = 1 << 12,
- ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE = 1 << 13, /* VK_DYNAMIC_STATE_LINE_STIPPLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_VIEWPORT = 1 << 0, /* VK_DYNAMIC_STATE_VIEWPORT */
+ ANV_CMD_DIRTY_DYNAMIC_SCISSOR = 1 << 1, /* VK_DYNAMIC_STATE_SCISSOR */
+ ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH = 1 << 2, /* VK_DYNAMIC_STATE_LINE_WIDTH */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS = 1 << 3, /* VK_DYNAMIC_STATE_DEPTH_BIAS */
+ ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS = 1 << 4, /* VK_DYNAMIC_STATE_BLEND_CONSTANTS */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS = 1 << 5, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS */
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK = 1 << 6, /* VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK */
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7, /* VK_DYNAMIC_STATE_STENCIL_WRITE_MASK */
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1 << 8, /* VK_DYNAMIC_STATE_STENCIL_REFERENCE */
+ ANV_CMD_DIRTY_PIPELINE = 1 << 9,
+ ANV_CMD_DIRTY_INDEX_BUFFER = 1 << 10,
+ ANV_CMD_DIRTY_RENDER_TARGETS = 1 << 11,
+ ANV_CMD_DIRTY_XFB_ENABLE = 1 << 12,
+ ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE = 1 << 13, /* VK_DYNAMIC_STATE_LINE_STIPPLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_CULL_MODE = 1 << 14, /* VK_DYNAMIC_STATE_CULL_MODE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_FRONT_FACE = 1 << 15, /* VK_DYNAMIC_STATE_FRONT_FACE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY = 1 << 16, /* VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE = 1 << 17, /* VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE = 1 << 18, /* VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE = 1 << 19, /* VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP = 1 << 20, /* VK_DYNAMIC_STATE_DEPTH_COMPARE_OP_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE = 1 << 21, /* VK_DYNAMIC_STATE_DEPTH_BOUNDS_TEST_ENABLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE = 1 << 22, /* VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE_EXT */
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_OP = 1 << 23, /* VK_DYNAMIC_STATE_STENCIL_OP_EXT */
};
typedef uint32_t anv_cmd_dirty_mask_t;
-#define ANV_CMD_DIRTY_DYNAMIC_ALL \
- (ANV_CMD_DIRTY_DYNAMIC_VIEWPORT | \
- ANV_CMD_DIRTY_DYNAMIC_SCISSOR | \
- ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH | \
- ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS | \
- ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS | \
- ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS | \
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | \
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | \
- ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE | \
- ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE)
+#define ANV_CMD_DIRTY_DYNAMIC_ALL \
+ (ANV_CMD_DIRTY_DYNAMIC_VIEWPORT | \
+ ANV_CMD_DIRTY_DYNAMIC_SCISSOR | \
+ ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS | \
+ ANV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS | \
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK | \
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK | \
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE | \
+ ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE | \
+ ANV_CMD_DIRTY_DYNAMIC_CULL_MODE | \
+ ANV_CMD_DIRTY_DYNAMIC_FRONT_FACE | \
+ ANV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY | \
+ ANV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP | \
+ ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE | \
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE | \
+ ANV_CMD_DIRTY_DYNAMIC_STENCIL_OP)
static inline enum anv_cmd_dirty_bits
anv_cmd_dirty_bit_for_vk_dynamic_state(VkDynamicState vk_state)
{
switch (vk_state) {
case VK_DYNAMIC_STATE_VIEWPORT:
+ case VK_DYNAMIC_STATE_VIEWPORT_WITH_COUNT_EXT:
return ANV_CMD_DIRTY_DYNAMIC_VIEWPORT;
case VK_DYNAMIC_STATE_SCISSOR:
+ case VK_DYNAMIC_STATE_SCISSOR_WITH_COUNT_EXT:
return ANV_CMD_DIRTY_DYNAMIC_SCISSOR;
case VK_DYNAMIC_STATE_LINE_WIDTH:
return ANV_CMD_DIRTY_DYNAMIC_LINE_WIDTH;
return ANV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE;
case VK_DYNAMIC_STATE_LINE_STIPPLE_EXT:
return ANV_CMD_DIRTY_DYNAMIC_LINE_STIPPLE;
+ case VK_DYNAMIC_STATE_CULL_MODE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_CULL_MODE;
+ case VK_DYNAMIC_STATE_FRONT_FACE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_FRONT_FACE;
+ case VK_DYNAMIC_STATE_PRIMITIVE_TOPOLOGY_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_PRIMITIVE_TOPOLOGY;
+ case VK_DYNAMIC_STATE_VERTEX_INPUT_BINDING_STRIDE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_VERTEX_INPUT_BINDING_STRIDE;
+ case VK_DYNAMIC_STATE_DEPTH_TEST_ENABLE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_DEPTH_TEST_ENABLE;
+ case VK_DYNAMIC_STATE_DEPTH_WRITE_ENABLE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_DEPTH_WRITE_ENABLE;
+ case VK_DYNAMIC_STATE_DEPTH_COMPARE_OP_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_DEPTH_COMPARE_OP;
+ case VK_DYNAMIC_STATE_DEPTH_BOUNDS_TEST_ENABLE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS_TEST_ENABLE;
+ case VK_DYNAMIC_STATE_STENCIL_TEST_ENABLE_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_STENCIL_TEST_ENABLE;
+ case VK_DYNAMIC_STATE_STENCIL_OP_EXT:
+ return ANV_CMD_DIRTY_DYNAMIC_STENCIL_OP;
default:
assert(!"Unsupported dynamic state");
return 0;
ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT = (1 << 12),
ANV_PIPE_DEPTH_STALL_BIT = (1 << 13),
ANV_PIPE_CS_STALL_BIT = (1 << 20),
+ ANV_PIPE_END_OF_PIPE_SYNC_BIT = (1 << 21),
/* This bit does not exist directly in PIPE_CONTROL. Instead it means that
* a flush has happened but not a CS stall. The next time we do any sort
* of invalidation we need to insert a CS stall at that time. Otherwise,
* we would have to CS stall on every flush which could be bad.
*/
- ANV_PIPE_NEEDS_CS_STALL_BIT = (1 << 21),
+ ANV_PIPE_NEEDS_END_OF_PIPE_SYNC_BIT = (1 << 22),
/* This bit does not exist directly in PIPE_CONTROL. It means that render
* target operations related to transfer commands with VkBuffer as
* streamer might need to be aware of this to trigger the appropriate stall
* before they can proceed with the copy.
*/
- ANV_PIPE_RENDER_TARGET_BUFFER_WRITES = (1 << 22),
+ ANV_PIPE_RENDER_TARGET_BUFFER_WRITES = (1 << 23),
+
+ /* This bit does not exist directly in PIPE_CONTROL. It means that Gen12
+ * AUX-TT data has changed and we need to invalidate AUX-TT data. This is
+ * done by writing the AUX-TT register.
+ */
+ ANV_PIPE_AUX_TABLE_INVALIDATE_BIT = (1 << 24),
+
+ /* This bit does not exist directly in PIPE_CONTROL. It means that a
+ * PIPE_CONTROL with a post-sync operation will follow. This is used to
+ * implement a workaround for Gen9.
+ */
+ ANV_PIPE_POST_SYNC_BIT = (1 << 25),
};
#define ANV_PIPE_FLUSH_BITS ( \
ANV_PIPE_VF_CACHE_INVALIDATE_BIT | \
ANV_PIPE_DATA_CACHE_FLUSH_BIT | \
ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT | \
- ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT)
+ ANV_PIPE_INSTRUCTION_CACHE_INVALIDATE_BIT | \
+ ANV_PIPE_AUX_TABLE_INVALIDATE_BIT)
static inline enum anv_pipe_bits
anv_pipe_flush_bits_for_access_flags(VkAccessFlags flags)
struct anv_vertex_binding {
struct anv_buffer * buffer;
VkDeviceSize offset;
+ VkDeviceSize stride;
+ VkDeviceSize size;
};
struct anv_xfb_binding {
/** Dynamic offsets for dynamic UBOs and SSBOs */
uint32_t dynamic_offsets[MAX_DYNAMIC_BUFFERS];
+ uint64_t push_reg_mask;
+
+ /** Pad out to a multiple of 32 bytes */
+ uint32_t pad[2];
+
struct {
/** Base workgroup ID
*
* uploading the push constants for compute shaders.
*/
uint32_t subgroup_id;
-
- /** Pad out to a multiple of 32 bytes */
- uint32_t pad[4];
} cs;
};
uint32_t back;
} stencil_reference;
+ struct {
+ struct {
+ VkStencilOp fail_op;
+ VkStencilOp pass_op;
+ VkStencilOp depth_fail_op;
+ VkCompareOp compare_op;
+ } front;
+ struct {
+ VkStencilOp fail_op;
+ VkStencilOp pass_op;
+ VkStencilOp depth_fail_op;
+ VkCompareOp compare_op;
+ } back;
+ } stencil_op;
+
struct {
uint32_t factor;
uint16_t pattern;
} line_stipple;
+
+ VkCullModeFlags cull_mode;
+ VkFrontFace front_face;
+ VkPrimitiveTopology primitive_topology;
+ bool depth_test_enable;
+ bool depth_write_enable;
+ VkCompareOp depth_compare_op;
+ bool depth_bounds_test_enable;
+ bool stencil_test_enable;
+ bool dyn_vbo_stride;
+ bool dyn_vbo_size;
};
extern const struct anv_dynamic_state default_dynamic_state;
*/
struct anv_attachment_state {
enum isl_aux_usage aux_usage;
- enum isl_aux_usage input_aux_usage;
struct anv_surface_state color;
struct anv_surface_state input;
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
* per-stage array in anv_cmd_state.
*/
struct anv_cmd_pipeline_state {
- struct anv_pipeline *pipeline;
-
struct anv_descriptor_set *descriptors[MAX_SETS];
struct anv_push_descriptor_set *push_descriptors[MAX_SETS];
};
struct anv_cmd_graphics_state {
struct anv_cmd_pipeline_state base;
+ struct anv_graphics_pipeline *pipeline;
+
anv_cmd_dirty_mask_t dirty;
uint32_t vb_dirty;
struct anv_dynamic_state dynamic;
+ uint32_t primitive_topology;
+
struct {
struct anv_buffer *index_buffer;
uint32_t index_type; /**< 3DSTATE_INDEX_BUFFER.IndexFormat */
struct anv_cmd_compute_state {
struct anv_cmd_pipeline_state base;
+ struct anv_compute_pipeline *pipeline;
+
bool pipeline_dirty;
struct anv_address num_workgroups;
* flat array. For depth-stencil attachments, the surface state is simply
* left blank.
*/
- struct anv_state render_pass_states;
+ struct anv_state attachment_states;
/**
* A null surface state of the right size to match the framebuffer. This
- * is one of the states in render_pass_states.
+ * is one of the states in attachment_states.
*/
struct anv_state null_surface_state;
};
struct anv_cmd_pool {
+ struct vk_object_base base;
VkAllocationCallbacks alloc;
struct list_head cmd_buffers;
};
ANV_CMD_BUFFER_EXEC_MODE_GROW_AND_EMIT,
ANV_CMD_BUFFER_EXEC_MODE_CHAIN,
ANV_CMD_BUFFER_EXEC_MODE_COPY_AND_CHAIN,
+ ANV_CMD_BUFFER_EXEC_MODE_CALL_AND_RETURN,
};
struct anv_cmd_buffer {
- VK_LOADER_DATA _loader_data;
+ struct vk_object_base base;
struct anv_device * device;
VkCommandBufferUsageFlags usage_flags;
VkCommandBufferLevel level;
+ struct anv_query_pool *perf_query_pool;
+
struct anv_cmd_state state;
+ struct anv_address return_addr;
+
/* Set by SetPerformanceMarkerINTEL, written into queries by CmdBeginQuery */
uint64_t intel_perf_marker;
};
const VkSemaphore *out_semaphores,
const uint64_t *out_signal_values,
uint32_t num_out_semaphores,
- VkFence fence);
+ VkFence fence,
+ int perf_query_pass);
VkResult anv_cmd_buffer_reset(struct anv_cmd_buffer *cmd_buffer);
};
struct anv_fence {
+ struct vk_object_base base;
+
/* Permanent fence state. Every fence has some form of permanent state
* (type != ANV_SEMAPHORE_TYPE_NONE). This may be a BO to fence on (for
* cross-process fences) or it could just be a dummy for use internally.
struct anv_fence *fence);
struct anv_event {
+ struct vk_object_base base;
uint64_t semaphore;
struct anv_state state;
};
};
struct anv_semaphore {
+ struct vk_object_base base;
+
uint32_t refcount;
/* Permanent semaphore state. Every semaphore has some form of permanent
struct anv_semaphore *semaphore);
struct anv_shader_module {
+ struct vk_object_base base;
+
unsigned char sha1[20];
uint32_t size;
char data[0];
struct anv_shader_bin {
uint32_t ref_cnt;
+ gl_shader_stage stage;
+
const struct anv_shader_bin_key *key;
struct anv_state kernel;
struct anv_shader_bin *
anv_shader_bin_create(struct anv_device *device,
+ gl_shader_stage stage,
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,
+ uint32_t prog_data_size,
const struct brw_compile_stats *stats, uint32_t num_stats,
const struct nir_xfb_info *xfb_info,
const struct anv_pipeline_bind_map *bind_map);
anv_shader_bin_destroy(device, shader);
}
-/* 5 possible simultaneous shader stages and FS may have up to 3 binaries */
-#define MAX_PIPELINE_EXECUTABLES 7
-
struct anv_pipeline_executable {
gl_shader_stage stage;
char *disasm;
};
+enum anv_pipeline_type {
+ ANV_PIPELINE_GRAPHICS,
+ ANV_PIPELINE_COMPUTE,
+};
+
struct anv_pipeline {
+ struct vk_object_base base;
+
struct anv_device * device;
+
struct anv_batch batch;
- uint32_t batch_data[512];
struct anv_reloc_list batch_relocs;
- anv_cmd_dirty_mask_t dynamic_state_mask;
- struct anv_dynamic_state dynamic_state;
void * mem_ctx;
+ enum anv_pipeline_type type;
VkPipelineCreateFlags flags;
- struct anv_subpass * subpass;
- bool needs_data_cache;
+ struct util_dynarray executables;
- struct anv_shader_bin * shaders[MESA_SHADER_STAGES];
+ const struct gen_l3_config * l3_config;
+};
- uint32_t num_executables;
- struct anv_pipeline_executable executables[MAX_PIPELINE_EXECUTABLES];
+struct anv_graphics_pipeline {
+ struct anv_pipeline base;
- struct {
- const struct gen_l3_config * l3_config;
- uint32_t total_size;
- } urb;
+ uint32_t batch_data[512];
- VkShaderStageFlags active_stages;
- struct anv_state blend_state;
+ anv_cmd_dirty_mask_t dynamic_state_mask;
+ struct anv_dynamic_state dynamic_state;
- uint32_t vb_used;
- struct anv_pipeline_vertex_binding {
- uint32_t stride;
- bool instanced;
- uint32_t instance_divisor;
- } vb[MAX_VBS];
+ uint32_t topology;
- uint8_t xfb_used;
+ struct anv_subpass * subpass;
- bool primitive_restart;
- uint32_t topology;
+ struct anv_shader_bin * shaders[MESA_SHADER_STAGES];
- uint32_t cs_right_mask;
+ VkShaderStageFlags active_stages;
+ bool primitive_restart;
bool writes_depth;
bool depth_test_enable;
bool writes_stencil;
bool kill_pixel;
bool depth_bounds_test_enable;
+ /* When primitive replication is used, subpass->view_mask will describe what
+ * views to replicate.
+ */
+ bool use_primitive_replication;
+
+ struct anv_state blend_state;
+
+ uint32_t vb_used;
+ struct anv_pipeline_vertex_binding {
+ uint32_t stride;
+ bool instanced;
+ uint32_t instance_divisor;
+ } vb[MAX_VBS];
+
struct {
uint32_t sf[7];
uint32_t depth_stencil_state[3];
+ uint32_t clip[4];
} gen7;
struct {
struct {
uint32_t wm_depth_stencil[4];
} gen9;
+};
+
+struct anv_compute_pipeline {
+ struct anv_pipeline base;
+ struct anv_shader_bin * cs;
+ uint32_t cs_right_mask;
+ uint32_t batch_data[9];
uint32_t interface_descriptor_data[8];
};
+#define ANV_DECL_PIPELINE_DOWNCAST(pipe_type, pipe_enum) \
+ static inline struct anv_##pipe_type##_pipeline * \
+ anv_pipeline_to_##pipe_type(struct anv_pipeline *pipeline) \
+ { \
+ assert(pipeline->type == pipe_enum); \
+ return (struct anv_##pipe_type##_pipeline *) pipeline; \
+ }
+
+ANV_DECL_PIPELINE_DOWNCAST(graphics, ANV_PIPELINE_GRAPHICS)
+ANV_DECL_PIPELINE_DOWNCAST(compute, ANV_PIPELINE_COMPUTE)
+
static inline bool
-anv_pipeline_has_stage(const struct anv_pipeline *pipeline,
+anv_pipeline_has_stage(const struct anv_graphics_pipeline *pipeline,
gl_shader_stage stage)
{
return (pipeline->active_stages & mesa_to_vk_shader_stage(stage)) != 0;
}
-#define ANV_DECL_GET_PROG_DATA_FUNC(prefix, stage) \
-static inline const struct brw_##prefix##_prog_data * \
-get_##prefix##_prog_data(const struct anv_pipeline *pipeline) \
-{ \
- if (anv_pipeline_has_stage(pipeline, stage)) { \
- return (const struct brw_##prefix##_prog_data *) \
- pipeline->shaders[stage]->prog_data; \
- } else { \
- return NULL; \
- } \
+#define ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(prefix, stage) \
+static inline const struct brw_##prefix##_prog_data * \
+get_##prefix##_prog_data(const struct anv_graphics_pipeline *pipeline) \
+{ \
+ if (anv_pipeline_has_stage(pipeline, stage)) { \
+ return (const struct brw_##prefix##_prog_data *) \
+ pipeline->shaders[stage]->prog_data; \
+ } else { \
+ return NULL; \
+ } \
}
-ANV_DECL_GET_PROG_DATA_FUNC(vs, MESA_SHADER_VERTEX)
-ANV_DECL_GET_PROG_DATA_FUNC(tcs, MESA_SHADER_TESS_CTRL)
-ANV_DECL_GET_PROG_DATA_FUNC(tes, MESA_SHADER_TESS_EVAL)
-ANV_DECL_GET_PROG_DATA_FUNC(gs, MESA_SHADER_GEOMETRY)
-ANV_DECL_GET_PROG_DATA_FUNC(wm, MESA_SHADER_FRAGMENT)
-ANV_DECL_GET_PROG_DATA_FUNC(cs, MESA_SHADER_COMPUTE)
+ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(vs, MESA_SHADER_VERTEX)
+ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(tcs, MESA_SHADER_TESS_CTRL)
+ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(tes, MESA_SHADER_TESS_EVAL)
+ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(gs, MESA_SHADER_GEOMETRY)
+ANV_DECL_GET_GRAPHICS_PROG_DATA_FUNC(wm, MESA_SHADER_FRAGMENT)
+
+static inline const struct brw_cs_prog_data *
+get_cs_prog_data(const struct anv_compute_pipeline *pipeline)
+{
+ assert(pipeline->cs);
+ return (const struct brw_cs_prog_data *) pipeline->cs->prog_data;
+}
static inline const struct brw_vue_prog_data *
-anv_pipeline_get_last_vue_prog_data(const struct anv_pipeline *pipeline)
+anv_pipeline_get_last_vue_prog_data(const struct anv_graphics_pipeline *pipeline)
{
if (anv_pipeline_has_stage(pipeline, MESA_SHADER_GEOMETRY))
return &get_gs_prog_data(pipeline)->base;
}
VkResult
-anv_pipeline_init(struct anv_pipeline *pipeline, struct anv_device *device,
- struct anv_pipeline_cache *cache,
- const VkGraphicsPipelineCreateInfo *pCreateInfo,
- const VkAllocationCallbacks *alloc);
+anv_pipeline_init(struct anv_pipeline *pipeline,
+ struct anv_device *device,
+ enum anv_pipeline_type type,
+ VkPipelineCreateFlags flags,
+ const VkAllocationCallbacks *pAllocator);
+
+void
+anv_pipeline_finish(struct anv_pipeline *pipeline,
+ struct anv_device *device,
+ const VkAllocationCallbacks *pAllocator);
VkResult
-anv_pipeline_compile_cs(struct anv_pipeline *pipeline,
+anv_graphics_pipeline_init(struct anv_graphics_pipeline *pipeline, struct anv_device *device,
+ struct anv_pipeline_cache *cache,
+ const VkGraphicsPipelineCreateInfo *pCreateInfo,
+ const VkAllocationCallbacks *alloc);
+
+VkResult
+anv_pipeline_compile_cs(struct anv_compute_pipeline *pipeline,
struct anv_pipeline_cache *cache,
const VkComputePipelineCreateInfo *info,
const struct anv_shader_module *module,
const char *entrypoint,
const VkSpecializationInfo *spec_info);
+struct anv_cs_parameters {
+ uint32_t group_size;
+ uint32_t simd_size;
+ uint32_t threads;
+};
+
+struct anv_cs_parameters
+anv_cs_parameters(const struct anv_compute_pipeline *pipeline);
+
struct anv_format_plane {
enum isl_format isl_format:16;
struct isl_swizzle swizzle;
bool can_ycbcr;
};
+/**
+ * Return the aspect's _format_ plane, not its _memory_ plane (using the
+ * vocabulary of VK_EXT_image_drm_format_modifier). As a consequence, \a
+ * aspect_mask may contain VK_IMAGE_ASPECT_PLANE_*, but must not contain
+ * VK_IMAGE_ASPECT_MEMORY_PLANE_* .
+ */
static inline uint32_t
anv_image_aspect_to_plane(VkImageAspectFlags image_aspects,
VkImageAspectFlags aspect_mask)
};
struct anv_image {
+ struct vk_object_base base;
+
VkImageType type; /**< VkImageCreateInfo::imageType */
/* The original VkFormat provided by the client. This may not match any
* of the actual surface formats.
*/
bool disjoint;
- /* All the formats that can be used when creating views of this image
- * are CCS_E compatible.
- */
- bool ccs_e_compatible;
-
/* Image was created with external format. */
bool external_format;
struct anv_surface shadow_surface;
/**
- * For color images, this is the aux usage for this image when not used
- * as a color attachment.
- *
- * For depth/stencil images, this is set to ISL_AUX_USAGE_HIZ if the
- * image has a HiZ buffer.
+ * The base aux usage for this image. For color images, this can be
+ * either CCS_E or CCS_D depending on whether or not we can reliably
+ * leave CCS on all the time.
*/
enum isl_aux_usage aux_usage;
*/
struct anv_address address;
- /**
- * Address of the main surface used to fill the aux map table. This is
- * used at destruction of the image since the Vulkan spec does not
- * guarantee that the address.bo field we still be valid at destruction.
- */
- uint64_t aux_map_surface_address;
-
/**
* When destroying the image, also free the bo.
* */
VkImageAspectFlagBits aspect)
{
uint32_t plane = anv_image_aspect_to_plane(image->aspects, aspect);
+ if (image->planes[plane].aux_usage == ISL_AUX_USAGE_NONE)
+ return 0;
/* The Gen12 CCS aux surface is represented with only one level. */
- const uint8_t aux_logical_levels =
- image->planes[plane].aux_surface.isl.tiling == ISL_TILING_GEN12_CCS ?
- image->planes[plane].surface.isl.levels :
- image->planes[plane].aux_surface.isl.levels;
-
- return image->planes[plane].aux_surface.isl.size_B > 0 ?
- aux_logical_levels : 0;
+ return image->planes[plane].aux_surface.isl.tiling == ISL_TILING_GEN12_CCS ?
+ image->planes[plane].surface.isl.levels :
+ image->planes[plane].aux_surface.isl.levels;
}
/* Returns the number of auxiliary buffer layers attached to an image. */
}
static inline struct anv_address
-anv_image_get_clear_color_addr(const struct anv_device *device,
+anv_image_get_clear_color_addr(UNUSED const struct anv_device *device,
const struct anv_image *image,
VkImageAspectFlagBits aspect)
{
if (!(image->aspects & VK_IMAGE_ASPECT_DEPTH_BIT))
return false;
+ /* For Gen8-11, there are some restrictions around sampling from HiZ.
+ * The Skylake PRM docs for RENDER_SURFACE_STATE::AuxiliarySurfaceMode
+ * say:
+ *
+ * "If this field is set to AUX_HIZ, Number of Multisamples must
+ * be MULTISAMPLECOUNT_1, and Surface Type cannot be SURFTYPE_3D."
+ */
+ if (image->type == VK_IMAGE_TYPE_3D)
+ return false;
+
/* Allow this feature on BDW even though it is disabled in the BDW devinfo
* struct. There's documentation which suggests that this feature actually
* reduces performance on BDW, but it has only been observed to help so
void
anv_image_mcs_op(struct anv_cmd_buffer *cmd_buffer,
const struct anv_image *image,
- enum isl_format format,
+ enum isl_format format, struct isl_swizzle swizzle,
VkImageAspectFlagBits aspect,
uint32_t base_layer, uint32_t layer_count,
enum isl_aux_op mcs_op, union isl_color_value *clear_value,
void
anv_image_ccs_op(struct anv_cmd_buffer *cmd_buffer,
const struct anv_image *image,
- enum isl_format format,
+ enum isl_format format, struct isl_swizzle swizzle,
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,
uint32_t base_level, uint32_t level_count,
uint32_t base_layer, uint32_t layer_count);
+enum isl_aux_state
+anv_layout_to_aux_state(const struct gen_device_info * const devinfo,
+ const struct anv_image *image,
+ const VkImageAspectFlagBits aspect,
+ const VkImageLayout layout);
+
enum isl_aux_usage
anv_layout_to_aux_usage(const struct gen_device_info * const devinfo,
const struct anv_image *image,
const VkImageAspectFlagBits aspect,
+ const VkImageUsageFlagBits usage,
const VkImageLayout layout);
enum anv_fast_clear_type
}
struct anv_image_view {
+ struct vk_object_base base;
+
const struct anv_image *image; /**< VkImageViewCreateInfo::image */
VkImageAspectFlags aspect_mask;
const VkAllocationCallbacks* alloc,
VkImage *pImage);
-const struct anv_surface *
-anv_image_get_surface_for_aspect_mask(const struct anv_image *image,
- VkImageAspectFlags aspect_mask);
-
enum isl_format
anv_isl_format_for_descriptor_type(VkDescriptorType type);
}
+/* Haswell border color is a bit of a disaster. Float and unorm formats use a
+ * straightforward 32-bit float color in the first 64 bytes. Instead of using
+ * a nice float/integer union like Gen8+, Haswell specifies the integer border
+ * color as a separate entry /after/ the float color. The layout of this entry
+ * also depends on the format's bpp (with extra hacks for RG32), and overlaps.
+ *
+ * Since we don't know the format/bpp, we can't make any of the border colors
+ * containing '1' work for all formats, as it would be in the wrong place for
+ * some of them. We opt to make 32-bit integers work as this seems like the
+ * most common option. Fortunately, transparent black works regardless, as
+ * all zeroes is the same in every bit-size.
+ */
+struct hsw_border_color {
+ float float32[4];
+ uint32_t _pad0[12];
+ uint32_t uint32[4];
+ uint32_t _pad1[108];
+};
+
+struct gen8_border_color {
+ union {
+ float float32[4];
+ uint32_t uint32[4];
+ };
+ /* Pad out to 64 bytes */
+ uint32_t _pad[12];
+};
+
struct anv_ycbcr_conversion {
+ struct vk_object_base base;
+
const struct anv_format * format;
VkSamplerYcbcrModelConversion ycbcr_model;
VkSamplerYcbcrRange ycbcr_range;
};
struct anv_sampler {
+ struct vk_object_base base;
+
uint32_t state[3][4];
uint32_t n_planes;
struct anv_ycbcr_conversion *conversion;
* and with a 32-byte stride for use as bindless samplers.
*/
struct anv_state bindless_state;
+
+ struct anv_state custom_border_color;
};
struct anv_framebuffer {
+ struct vk_object_base base;
+
uint32_t width;
uint32_t height;
uint32_t layers;
};
struct anv_render_pass {
+ struct vk_object_base base;
+
uint32_t attachment_count;
uint32_t subpass_count;
/* An array of subpass_count+1 flushes, one per subpass boundary */
#define ANV_PIPELINE_STATISTICS_MASK 0x000007ff
+#define OA_SNAPSHOT_SIZE (256)
+#define ANV_KHR_PERF_QUERY_SIZE (ALIGN(sizeof(uint64_t), 64) + 2 * OA_SNAPSHOT_SIZE)
+
struct anv_query_pool {
+ struct vk_object_base base;
+
VkQueryType type;
VkQueryPipelineStatisticFlags pipeline_statistics;
/** Stride between slots, in bytes */
/** Number of slots in this query pool */
uint32_t slots;
struct anv_bo * bo;
+
+ /* Perf queries : */
+ struct anv_bo reset_bo;
+ uint32_t n_counters;
+ struct gen_perf_counter_pass *counter_pass;
+ uint32_t n_passes;
+ struct gen_perf_query_info **pass_query;
};
+static inline uint32_t khr_perf_query_preamble_offset(struct anv_query_pool *pool,
+ uint32_t pass)
+{
+ return pass * ANV_KHR_PERF_QUERY_SIZE + 8;
+}
+
int anv_get_instance_entrypoint_index(const char *name);
int anv_get_device_entrypoint_index(const char *name);
int anv_get_physical_device_entrypoint_index(const char *name);
const struct anv_instance_extension_table *instance,
const struct anv_device_extension_table *device);
+void *anv_resolve_device_entrypoint(const struct gen_device_info *devinfo,
+ uint32_t index);
void *anv_lookup_entrypoint(const struct gen_device_info *devinfo,
const char *name);
struct gen_perf_config *anv_get_perf(const struct gen_device_info *devinfo, int fd);
void anv_device_perf_init(struct anv_device *device);
-
-#define ANV_DEFINE_HANDLE_CASTS(__anv_type, __VkType) \
- \
- static inline struct __anv_type * \
- __anv_type ## _from_handle(__VkType _handle) \
- { \
- return (struct __anv_type *) _handle; \
- } \
- \
- static inline __VkType \
- __anv_type ## _to_handle(struct __anv_type *_obj) \
- { \
- return (__VkType) _obj; \
- }
-
-#define ANV_DEFINE_NONDISP_HANDLE_CASTS(__anv_type, __VkType) \
- \
- static inline struct __anv_type * \
- __anv_type ## _from_handle(__VkType _handle) \
- { \
- return (struct __anv_type *)(uintptr_t) _handle; \
- } \
- \
- static inline __VkType \
- __anv_type ## _to_handle(struct __anv_type *_obj) \
- { \
- return (__VkType)(uintptr_t) _obj; \
- }
+void anv_perf_write_pass_results(struct gen_perf_config *perf,
+ struct anv_query_pool *pool, uint32_t pass,
+ const struct gen_perf_query_result *accumulated_results,
+ union VkPerformanceCounterResultKHR *results);
#define ANV_FROM_HANDLE(__anv_type, __name, __handle) \
- struct __anv_type *__name = __anv_type ## _from_handle(__handle)
-
-ANV_DEFINE_HANDLE_CASTS(anv_cmd_buffer, VkCommandBuffer)
-ANV_DEFINE_HANDLE_CASTS(anv_device, VkDevice)
-ANV_DEFINE_HANDLE_CASTS(anv_instance, VkInstance)
-ANV_DEFINE_HANDLE_CASTS(anv_physical_device, VkPhysicalDevice)
-ANV_DEFINE_HANDLE_CASTS(anv_queue, VkQueue)
-
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_cmd_pool, VkCommandPool)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer, VkBuffer)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer_view, VkBufferView)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_pool, VkDescriptorPool)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set, VkDescriptorSet)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set_layout, VkDescriptorSetLayout)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_update_template, VkDescriptorUpdateTemplate)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_device_memory, VkDeviceMemory)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_fence, VkFence)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_event, VkEvent)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_framebuffer, VkFramebuffer)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image, VkImage)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_image_view, VkImageView);
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_cache, VkPipelineCache)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline, VkPipeline)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_layout, VkPipelineLayout)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_query_pool, VkQueryPool)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_render_pass, VkRenderPass)
-ANV_DEFINE_NONDISP_HANDLE_CASTS(anv_sampler, VkSampler)
-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, VkSamplerYcbcrConversion)
+ VK_FROM_HANDLE(__anv_type, __name, __handle)
+
+VK_DEFINE_HANDLE_CASTS(anv_cmd_buffer, base, VkCommandBuffer,
+ VK_OBJECT_TYPE_COMMAND_BUFFER)
+VK_DEFINE_HANDLE_CASTS(anv_device, vk.base, VkDevice, VK_OBJECT_TYPE_DEVICE)
+VK_DEFINE_HANDLE_CASTS(anv_instance, base, VkInstance, VK_OBJECT_TYPE_INSTANCE)
+VK_DEFINE_HANDLE_CASTS(anv_physical_device, base, VkPhysicalDevice,
+ VK_OBJECT_TYPE_PHYSICAL_DEVICE)
+VK_DEFINE_HANDLE_CASTS(anv_queue, base, VkQueue, VK_OBJECT_TYPE_QUEUE)
+
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_cmd_pool, base, VkCommandPool,
+ VK_OBJECT_TYPE_COMMAND_POOL)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer, base, VkBuffer,
+ VK_OBJECT_TYPE_BUFFER)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_buffer_view, base, VkBufferView,
+ VK_OBJECT_TYPE_BUFFER_VIEW)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_pool, base, VkDescriptorPool,
+ VK_OBJECT_TYPE_DESCRIPTOR_POOL)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set, base, VkDescriptorSet,
+ VK_OBJECT_TYPE_DESCRIPTOR_SET)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_set_layout, base,
+ VkDescriptorSetLayout,
+ VK_OBJECT_TYPE_DESCRIPTOR_SET_LAYOUT)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_descriptor_update_template, base,
+ VkDescriptorUpdateTemplate,
+ VK_OBJECT_TYPE_DESCRIPTOR_UPDATE_TEMPLATE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_device_memory, base, VkDeviceMemory,
+ VK_OBJECT_TYPE_DEVICE_MEMORY)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_fence, base, VkFence, VK_OBJECT_TYPE_FENCE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_event, base, VkEvent, VK_OBJECT_TYPE_EVENT)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_framebuffer, base, VkFramebuffer,
+ VK_OBJECT_TYPE_FRAMEBUFFER)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_image, base, VkImage, VK_OBJECT_TYPE_IMAGE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_image_view, base, VkImageView,
+ VK_OBJECT_TYPE_IMAGE_VIEW);
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_cache, base, VkPipelineCache,
+ VK_OBJECT_TYPE_PIPELINE_CACHE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline, base, VkPipeline,
+ VK_OBJECT_TYPE_PIPELINE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_pipeline_layout, base, VkPipelineLayout,
+ VK_OBJECT_TYPE_PIPELINE_LAYOUT)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_query_pool, base, VkQueryPool,
+ VK_OBJECT_TYPE_QUERY_POOL)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_render_pass, base, VkRenderPass,
+ VK_OBJECT_TYPE_RENDER_PASS)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_sampler, base, VkSampler,
+ VK_OBJECT_TYPE_SAMPLER)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_semaphore, base, VkSemaphore,
+ VK_OBJECT_TYPE_SEMAPHORE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_shader_module, base, VkShaderModule,
+ VK_OBJECT_TYPE_SHADER_MODULE)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_ycbcr_conversion, base,
+ VkSamplerYcbcrConversion,
+ VK_OBJECT_TYPE_SAMPLER_YCBCR_CONVERSION)
/* Gen-specific function declarations */
#ifdef genX
projects / mesa.git / blobdiff