bool has_exec_fence;
bool has_syncobj;
bool has_syncobj_wait;
+ bool has_syncobj_wait_available;
bool has_context_priority;
bool has_context_isolation;
+ bool has_thread_submit;
bool has_mem_available;
bool has_mmap_offset;
uint64_t gtt_size;
bool has_bindless_images;
/** True if we can use bindless access for samplers */
bool has_bindless_samplers;
+ /** True if we can use timeline semaphores through execbuf */
+ bool has_exec_timeline;
/** True if we can read the GPU timestamp register
*
uint32_t fence_count;
uint32_t fence_array_length;
struct drm_i915_gem_exec_fence * fences;
+ uint64_t * fence_values;
uint32_t temporary_semaphore_count;
uint32_t temporary_semaphore_array_length;
uint32_t sync_fd_semaphore_array_length;
/* Allocated only with non shareable timelines. */
- struct anv_timeline ** wait_timelines;
+ union {
+ struct anv_timeline ** wait_timelines;
+ uint32_t * wait_timeline_syncobjs;
+ };
uint32_t wait_timeline_count;
uint32_t wait_timeline_array_length;
uint64_t * wait_timeline_values;
struct anv_queue {
struct vk_object_base base;
- struct anv_device * device;
+ struct anv_device * device;
- /*
- * A list of struct anv_queue_submit to be submitted to i915.
- */
- struct list_head queued_submits;
+ VkDeviceQueueCreateFlags flags;
+
+ /* Set once from the device api calls. */
+ bool lost_signaled;
+
+ /* Only set once atomically by the queue */
+ int lost;
+ int error_line;
+ const char * error_file;
+ char error_msg[80];
+
+ /*
+ * This mutext protects the variables below.
+ */
+ pthread_mutex_t mutex;
+
+ pthread_t thread;
+ pthread_cond_t cond;
- VkDeviceQueueCreateFlags flags;
+ /*
+ * A list of struct anv_queue_submit to be submitted to i915.
+ */
+ struct list_head queued_submits;
+
+ /* Set to true to stop the submission thread */
+ bool quit;
};
struct anv_pipeline_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,
- uint32_t constant_data_size,
const struct brw_stage_prog_data *prog_data,
uint32_t prog_data_size,
const struct brw_compile_stats *stats,
gl_shader_stage stage,
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 brw_compile_stats *stats,
int fd;
bool can_chain_batches;
bool robust_buffer_access;
+ bool has_thread_submit;
struct anv_device_extension_table enabled_extensions;
struct anv_device_dispatch_table dispatch;
pthread_mutex_t mutex;
pthread_cond_t queue_submit;
int _lost;
+ int lost_reported;
struct gen_batch_decode_ctx decoder_ctx;
/*
void anv_device_init_blorp(struct anv_device *device);
void anv_device_finish_blorp(struct anv_device *device);
-void _anv_device_set_all_queue_lost(struct anv_device *device);
+void _anv_device_report_lost(struct anv_device *device);
VkResult _anv_device_set_lost(struct anv_device *device,
const char *file, int line,
const char *msg, ...)
#define anv_device_set_lost(dev, ...) \
_anv_device_set_lost(dev, __FILE__, __LINE__, __VA_ARGS__)
#define anv_queue_set_lost(queue, ...) \
- _anv_queue_set_lost(queue, __FILE__, __LINE__, __VA_ARGS__)
+ (queue)->device->has_thread_submit ? \
+ _anv_queue_set_lost(queue, __FILE__, __LINE__, __VA_ARGS__) : \
+ _anv_device_set_lost(queue->device, __FILE__, __LINE__, __VA_ARGS__)
static inline bool
anv_device_is_lost(struct anv_device *device)
{
- return unlikely(p_atomic_read(&device->_lost));
+ int lost = p_atomic_read(&device->_lost);
+ if (unlikely(lost && !device->lost_reported))
+ _anv_device_report_lost(device);
+ return lost;
}
VkResult anv_device_query_status(struct anv_device *device);
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);
+uint64_t anv_gem_get_drm_cap(int fd, uint32_t capability);
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_gpu_get_reset_stats(struct anv_device *device,
void anv_gem_syncobj_reset(struct anv_device *device, uint32_t handle);
bool anv_gem_supports_syncobj_wait(int fd);
int anv_gem_syncobj_wait(struct anv_device *device,
- uint32_t *handles, uint32_t num_handles,
+ const uint32_t *handles, uint32_t num_handles,
int64_t abs_timeout_ns, bool wait_all);
+int anv_gem_syncobj_timeline_wait(struct anv_device *device,
+ const uint32_t *handles, const uint64_t *points,
+ uint32_t num_items, int64_t abs_timeout_ns,
+ bool wait_all, bool wait_materialize);
+int anv_gem_syncobj_timeline_signal(struct anv_device *device,
+ const uint32_t *handles, const uint64_t *points,
+ uint32_t num_items);
+int anv_gem_syncobj_timeline_query(struct anv_device *device,
+ const uint32_t *handles, uint64_t *points,
+ uint32_t num_items);
uint64_t anv_vma_alloc(struct anv_device *device,
uint64_t size, uint64_t align,
/** Dynamic offsets for dynamic UBOs and SSBOs */
uint32_t dynamic_offsets[MAX_DYNAMIC_BUFFERS];
- uint64_t push_reg_mask;
+ /* Robust access pushed registers. */
+ uint64_t push_reg_mask[MESA_SHADER_STAGES];
/** Pad out to a multiple of 32 bytes */
uint32_t pad[2];
ANV_SEMAPHORE_TYPE_SYNC_FILE,
ANV_SEMAPHORE_TYPE_DRM_SYNCOBJ,
ANV_SEMAPHORE_TYPE_TIMELINE,
+ ANV_SEMAPHORE_TYPE_DRM_SYNCOBJ_TIMELINE,
};
struct anv_timeline_point {
stage = __builtin_ffs(__tmp) - 1, __tmp; \
__tmp &= ~(1 << (stage)))
+enum anv_shader_reloc {
+ ANV_SHADER_RELOC_CONST_DATA_ADDR_LOW,
+ ANV_SHADER_RELOC_CONST_DATA_ADDR_HIGH,
+};
+
struct anv_pipeline_bind_map {
unsigned char surface_sha1[20];
unsigned char sampler_sha1[20];
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;
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 struct brw_compile_stats *stats, uint32_t num_stats,
return subpass_id;
}
+struct anv_performance_configuration_intel {
+ struct vk_object_base base;
+
+ struct gen_perf_registers *register_config;
+
+ uint64_t config_id;
+};
+
struct gen_perf_config *anv_get_perf(const struct gen_device_info *devinfo, int fd);
void anv_device_perf_init(struct anv_device *device);
void anv_perf_write_pass_results(struct gen_perf_config *perf,
VK_DEFINE_NONDISP_HANDLE_CASTS(anv_ycbcr_conversion, base,
VkSamplerYcbcrConversion,
VK_OBJECT_TYPE_SAMPLER_YCBCR_CONVERSION)
+VK_DEFINE_NONDISP_HANDLE_CASTS(anv_performance_configuration_intel, base,
+ VkPerformanceConfigurationINTEL,
+ VK_OBJECT_TYPE_PERFORMANCE_CONFIGURATION_INTEL)
/* Gen-specific function declarations */
#ifdef genX
projects / mesa.git / blobdiff