| // Copyright 2020 The IREE Authors |
| // |
| // Licensed under the Apache License v2.0 with LLVM Exceptions. |
| // See https://llvm.org/LICENSE.txt for license information. |
| // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| |
| #include "iree/hal/device.h" |
| |
| #include <inttypes.h> |
| #include <string.h> |
| |
| #include "iree/hal/allocator.h" |
| #include "iree/hal/buffer.h" |
| #include "iree/hal/command_buffer.h" |
| #include "iree/hal/detail.h" |
| #include "iree/hal/resource.h" |
| |
| //===----------------------------------------------------------------------===// |
| // iree_hal_device_t |
| //===----------------------------------------------------------------------===// |
| |
| #define _VTABLE_DISPATCH(device, method_name) \ |
| IREE_HAL_VTABLE_DISPATCH(device, iree_hal_device, method_name) |
| |
| IREE_HAL_API_RETAIN_RELEASE(device); |
| |
| IREE_API_EXPORT iree_string_view_t |
| iree_hal_device_id(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| return _VTABLE_DISPATCH(device, id)(device); |
| } |
| |
| IREE_API_EXPORT iree_allocator_t |
| iree_hal_device_host_allocator(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| return _VTABLE_DISPATCH(device, host_allocator)(device); |
| } |
| |
| IREE_API_EXPORT iree_hal_allocator_t* iree_hal_device_allocator( |
| iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| return _VTABLE_DISPATCH(device, device_allocator)(device); |
| } |
| |
| IREE_API_EXPORT void iree_hal_device_replace_allocator( |
| iree_hal_device_t* device, iree_hal_allocator_t* new_allocator) { |
| IREE_ASSERT_ARGUMENT(device); |
| _VTABLE_DISPATCH(device, replace_device_allocator)(device, new_allocator); |
| } |
| |
| IREE_API_EXPORT void iree_hal_device_replace_channel_provider( |
| iree_hal_device_t* device, iree_hal_channel_provider_t* new_provider) { |
| IREE_ASSERT_ARGUMENT(device); |
| _VTABLE_DISPATCH(device, replace_channel_provider)(device, new_provider); |
| } |
| |
| IREE_API_EXPORT |
| iree_status_t iree_hal_device_trim(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, trim)(device); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_query_i64( |
| iree_hal_device_t* device, iree_string_view_t category, |
| iree_string_view_t key, int64_t* out_value) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(out_value); |
| return _VTABLE_DISPATCH(device, query_i64)(device, category, key, out_value); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_query_capabilities( |
| iree_hal_device_t* device, |
| iree_hal_device_capabilities_t* out_capabilities) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(out_capabilities); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| IREE_RETURN_IF_ERROR( |
| _VTABLE_DISPATCH(device, query_capabilities)(device, out_capabilities)); |
| IREE_TRACE_ZONE_END(z0); |
| return iree_ok_status(); |
| } |
| |
| IREE_API_EXPORT const iree_hal_device_topology_info_t* |
| iree_hal_device_topology_info(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| return _VTABLE_DISPATCH(device, topology_info)(device); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_refine_topology_edge( |
| iree_hal_device_t* src_device, iree_hal_device_t* dst_device, |
| iree_hal_topology_edge_t* edge) { |
| IREE_ASSERT_ARGUMENT(src_device); |
| IREE_ASSERT_ARGUMENT(dst_device); |
| IREE_ASSERT_ARGUMENT(edge); |
| return _VTABLE_DISPATCH(src_device, refine_topology_edge)(src_device, |
| dst_device, edge); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_assign_topology_info( |
| iree_hal_device_t* device, |
| const iree_hal_device_topology_info_t* topology_info) { |
| IREE_ASSERT_ARGUMENT(device); |
| return _VTABLE_DISPATCH(device, assign_topology_info)(device, topology_info); |
| } |
| |
| IREE_API_EXPORT iree_hal_semaphore_compatibility_t |
| iree_hal_device_query_semaphore_compatibility(iree_hal_device_t* device, |
| iree_hal_semaphore_t* semaphore) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(semaphore); |
| return _VTABLE_DISPATCH(device, query_semaphore_compatibility)(device, |
| semaphore); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_query_queue_pool_backend( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| iree_hal_queue_pool_backend_t* out_backend) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(out_backend); |
| memset(out_backend, 0, sizeof(*out_backend)); |
| const iree_hal_device_topology_info_t* topology_info = |
| iree_hal_device_topology_info(device); |
| if (!topology_info->topology || !topology_info->frontier.tracker) { |
| return iree_make_status( |
| IREE_STATUS_FAILED_PRECONDITION, |
| "device queue pool backends are unavailable before the device is " |
| "assigned to a device group"); |
| } |
| return _VTABLE_DISPATCH(device, query_queue_pool_backend)( |
| device, queue_affinity, out_backend); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_alloca( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_pool_t* pool, iree_hal_buffer_params_t params, |
| iree_device_size_t allocation_size, iree_hal_alloca_flags_t flags, |
| iree_hal_buffer_t** IREE_RESTRICT out_buffer) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(out_buffer); |
| *out_buffer = NULL; |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_alloca)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, pool, |
| params, allocation_size, flags, out_buffer); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_dealloca( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_buffer_t* buffer, iree_hal_dealloca_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(buffer); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| |
| // If the buffer has an origin then use it for the deallocation. Some buffers |
| // may not have an origin and we'll use the provided device. |
| const iree_hal_buffer_placement_t placement = |
| iree_hal_buffer_allocation_placement(buffer); |
| if (iree_all_bits_set(flags, IREE_HAL_DEALLOCA_FLAG_PREFER_ORIGIN) && |
| placement.device != NULL) { |
| device = placement.device; |
| queue_affinity = placement.queue_affinity; |
| } |
| |
| // If the buffer was not allocated asynchronously then this is just a barrier. |
| iree_status_t status = iree_ok_status(); |
| if (iree_all_bits_set(placement.flags, |
| IREE_HAL_BUFFER_PLACEMENT_FLAG_ASYNCHRONOUS) && |
| !iree_all_bits_set( |
| placement.flags, |
| IREE_HAL_BUFFER_PLACEMENT_FLAG_INDETERMINATE_LIFETIME)) { |
| status = _VTABLE_DISPATCH(device, queue_dealloca)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| buffer, flags); |
| } else { |
| status = iree_hal_device_queue_barrier( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| IREE_HAL_EXECUTE_FLAG_NONE); |
| } |
| |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_fill( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset, |
| iree_device_size_t length, const void* pattern, |
| iree_host_size_t pattern_length, iree_hal_fill_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(pattern); |
| IREE_ASSERT_ARGUMENT(target_buffer); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| IREE_TRACE_ZONE_APPEND_VALUE_I64(z0, (int64_t)length); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_fill)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| target_buffer, target_offset, length, pattern, pattern_length, flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_update( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| const void* source_buffer, iree_host_size_t source_offset, |
| iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset, |
| iree_device_size_t length, iree_hal_update_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(source_buffer); |
| IREE_ASSERT_ARGUMENT(target_buffer); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| IREE_TRACE_ZONE_APPEND_VALUE_I64(z0, (int64_t)length); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_update)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| source_buffer, source_offset, target_buffer, target_offset, length, |
| flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_copy( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_buffer_t* source_buffer, iree_device_size_t source_offset, |
| iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset, |
| iree_device_size_t length, iree_hal_copy_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(source_buffer); |
| IREE_ASSERT_ARGUMENT(target_buffer); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| IREE_TRACE_ZONE_APPEND_VALUE_I64(z0, (int64_t)length); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_copy)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| source_buffer, source_offset, target_buffer, target_offset, length, |
| flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_read( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_file_t* source_file, uint64_t source_offset, |
| iree_hal_buffer_t* target_buffer, iree_device_size_t target_offset, |
| iree_device_size_t length, iree_hal_read_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(source_file); |
| IREE_ASSERT_ARGUMENT(target_buffer); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_read)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| source_file, source_offset, target_buffer, target_offset, length, flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_write( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_buffer_t* source_buffer, iree_device_size_t source_offset, |
| iree_hal_file_t* target_file, uint64_t target_offset, |
| iree_device_size_t length, iree_hal_write_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(source_buffer); |
| IREE_ASSERT_ARGUMENT(target_file); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_write)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| source_buffer, source_offset, target_file, target_offset, length, flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_host_call( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_host_call_t call, const uint64_t args[4], |
| iree_hal_host_call_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(call.fn); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_host_call)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, call, |
| args, flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_dispatch( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_executable_t* executable, iree_hal_executable_function_t function, |
| const iree_hal_dispatch_config_t config, iree_const_byte_span_t constants, |
| const iree_hal_buffer_ref_list_t bindings, |
| iree_hal_dispatch_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(executable); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_dispatch)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| executable, function, config, constants, bindings, flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_execute( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_command_buffer_t* command_buffer, |
| iree_hal_buffer_binding_table_t binding_table, |
| iree_hal_execute_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT( |
| !wait_semaphore_list.count || |
| (wait_semaphore_list.semaphores && wait_semaphore_list.payload_values)); |
| IREE_ASSERT_ARGUMENT(!signal_semaphore_list.count || |
| (signal_semaphore_list.semaphores && |
| signal_semaphore_list.payload_values)); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| |
| // TODO(benvanik): move into devices instead? then a synchronous/inline device |
| // could assert the waits are resolved instead of blanket failing on an |
| // already-resolved semaphore. This would make using stream-ordered |
| // allocations easier. |
| if (wait_semaphore_list.count > 0 && command_buffer && |
| iree_all_bits_set(iree_hal_command_buffer_mode(command_buffer), |
| IREE_HAL_COMMAND_BUFFER_MODE_ALLOW_INLINE_EXECUTION)) { |
| // Inline command buffers are not allowed to wait (as they could have |
| // already been executed!). This is a requirement of the API so we |
| // validate it across all backends even if they don't support inline |
| // execution and ignore it. |
| IREE_TRACE_ZONE_END(z0); |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "inline command buffer submitted with a wait; inline command " |
| "buffers must be ready to execute immediately"); |
| } |
| |
| // Validate command buffer bindings against the provided binding tables. |
| // This will error out if a binding table is required but not provided or if |
| // any binding in the table does not match the requirements of the command |
| // buffer as recorded. |
| if (command_buffer) { |
| IREE_RETURN_AND_END_ZONE_IF_ERROR( |
| z0, iree_hal_command_buffer_validate_submission(command_buffer, |
| binding_table)); |
| } |
| |
| iree_status_t status = _VTABLE_DISPATCH(device, queue_execute)( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, |
| command_buffer, binding_table, flags); |
| |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_barrier( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity, |
| const iree_hal_semaphore_list_t wait_semaphore_list, |
| const iree_hal_semaphore_list_t signal_semaphore_list, |
| iree_hal_execute_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = iree_hal_device_queue_execute( |
| device, queue_affinity, wait_semaphore_list, signal_semaphore_list, NULL, |
| iree_hal_buffer_binding_table_empty(), flags); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_queue_flush( |
| iree_hal_device_t* device, iree_hal_queue_affinity_t queue_affinity) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = |
| _VTABLE_DISPATCH(device, queue_flush)(device, queue_affinity); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_wait_semaphores( |
| iree_hal_device_t* device, iree_async_wait_mode_t wait_mode, |
| const iree_hal_semaphore_list_t semaphore_list, iree_timeout_t timeout, |
| iree_async_wait_flags_t flags) { |
| IREE_ASSERT_ARGUMENT(device); |
| if (semaphore_list.count == 0) return iree_ok_status(); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| // HAL semaphores embed async semaphores at offset 0 (toll-free bridge). |
| iree_status_t status = iree_async_semaphore_multi_wait( |
| wait_mode, (iree_async_semaphore_t**)semaphore_list.semaphores, |
| semaphore_list.payload_values, semaphore_list.count, timeout, flags, |
| iree_allocator_system()); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_profiling_begin( |
| iree_hal_device_t* device, |
| const iree_hal_device_profiling_options_t* options) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(options); |
| |
| const iree_hal_device_profiling_flags_t supported_flags = |
| IREE_HAL_DEVICE_PROFILING_FLAG_LIGHTWEIGHT_STATISTICS; |
| if (iree_any_bit_set(options->flags, ~supported_flags)) { |
| return iree_make_status(IREE_STATUS_INVALID_ARGUMENT, |
| "unsupported profile option flags 0x%x", |
| options->flags & ~supported_flags); |
| } |
| const iree_hal_profile_capture_filter_flags_t supported_filter_flags = |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_EXECUTABLE_FUNCTION_PATTERN | |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_COMMAND_BUFFER_ID | |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_COMMAND_INDEX | |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_PHYSICAL_DEVICE_ORDINAL | |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_QUEUE_ORDINAL; |
| if (iree_any_bit_set(options->capture_filter.flags, |
| ~supported_filter_flags)) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "unsupported profile capture filter bits 0x%x", |
| options->capture_filter.flags & ~supported_filter_flags); |
| } |
| if (options->capture_filter.reserved0 != 0) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "profile capture filter reserved fields must be zero"); |
| } |
| if (iree_any_bit_set( |
| options->capture_filter.flags, |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_EXECUTABLE_FUNCTION_PATTERN) && |
| iree_string_view_is_empty( |
| options->capture_filter.executable_function_pattern)) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "profile capture executable function filter must not be empty"); |
| } |
| if (iree_any_bit_set( |
| options->capture_filter.flags, |
| IREE_HAL_PROFILE_CAPTURE_FILTER_FLAG_COMMAND_BUFFER_ID) && |
| options->capture_filter.command_buffer_id == 0) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "profile capture command buffer filter must be nonzero"); |
| } |
| |
| if (options->counter_set_count != 0) { |
| if (!options->counter_sets) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "hardware counter set selections require a counter_sets array"); |
| } |
| if (!iree_hal_device_profiling_options_requests_counters(options)) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "hardware counter set selections require a counter profiling data " |
| "family"); |
| } |
| for (iree_host_size_t i = 0; i < options->counter_set_count; ++i) { |
| const iree_hal_profile_counter_set_selection_t* counter_set = |
| &options->counter_sets[i]; |
| if (counter_set->counter_name_count == 0 || !counter_set->counter_names) { |
| return iree_make_status(IREE_STATUS_INVALID_ARGUMENT, |
| "hardware counter set %" PRIhsz |
| " must request at least one counter name", |
| i); |
| } |
| for (iree_host_size_t j = 0; j < counter_set->counter_name_count; ++j) { |
| iree_string_view_t counter_name = counter_set->counter_names[j]; |
| if (iree_string_view_is_empty(counter_name) || !counter_name.data) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "hardware counter set %" PRIhsz |
| " has an empty counter name at index %" PRIhsz, |
| i, j); |
| } |
| } |
| } |
| } |
| if (iree_hal_device_profiling_options_requests_counters(options) && |
| options->counter_set_count == 0) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "counter profiling requires at least one counter set selection"); |
| } |
| |
| const bool data_requested = |
| options->data_families != IREE_HAL_DEVICE_PROFILING_DATA_NONE || |
| iree_hal_device_profiling_options_requests_lightweight_statistics( |
| options); |
| if (data_requested && !options->sink) { |
| return iree_make_status( |
| IREE_STATUS_INVALID_ARGUMENT, |
| "HAL-native profiling with requested data requires a profile sink"); |
| } |
| if (!data_requested) { |
| return iree_ok_status(); |
| } |
| |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = |
| _VTABLE_DISPATCH(device, profiling_begin)(device, options); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t |
| iree_hal_device_profiling_flush(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, profiling_flush)(device); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t |
| iree_hal_device_profiling_end(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, profiling_end)(device); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_external_capture_begin( |
| iree_hal_device_t* device, |
| const iree_hal_device_external_capture_options_t* options) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_ASSERT_ARGUMENT(options); |
| if (iree_string_view_is_empty(options->provider)) { |
| return iree_make_status(IREE_STATUS_INVALID_ARGUMENT, |
| "external capture provider must be specified"); |
| } |
| |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = |
| _VTABLE_DISPATCH(device, external_capture_begin)(device, options); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_status_t |
| iree_hal_device_external_capture_end(iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = _VTABLE_DISPATCH(device, external_capture_end)(device); |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| //===----------------------------------------------------------------------===// |
| // iree_hal_device_list_t |
| //===----------------------------------------------------------------------===// |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_list_allocate( |
| iree_host_size_t capacity, iree_allocator_t host_allocator, |
| iree_hal_device_list_t** out_list) { |
| IREE_ASSERT_ARGUMENT(out_list); |
| *out_list = NULL; |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_hal_device_list_t* list = NULL; |
| iree_host_size_t total_size = |
| sizeof(*list) + capacity * sizeof(list->devices[0]); |
| IREE_RETURN_AND_END_ZONE_IF_ERROR( |
| z0, iree_allocator_malloc(host_allocator, total_size, (void**)&list)); |
| list->host_allocator = host_allocator; |
| list->capacity = capacity; |
| list->count = 0; |
| *out_list = list; |
| IREE_TRACE_ZONE_END(z0); |
| return iree_ok_status(); |
| } |
| |
| IREE_API_EXPORT void iree_hal_device_list_free(iree_hal_device_list_t* list) { |
| if (!list) return; |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_allocator_t host_allocator = list->host_allocator; |
| for (iree_host_size_t i = 0; i < list->count; ++i) { |
| iree_hal_device_release(list->devices[i]); |
| } |
| iree_allocator_free(host_allocator, list); |
| IREE_TRACE_ZONE_END(z0); |
| } |
| |
| IREE_API_EXPORT iree_status_t iree_hal_device_list_push_back( |
| iree_hal_device_list_t* list, iree_hal_device_t* device) { |
| IREE_ASSERT_ARGUMENT(list); |
| IREE_ASSERT_ARGUMENT(device); |
| IREE_TRACE_ZONE_BEGIN(z0); |
| iree_status_t status = iree_ok_status(); |
| if (list->count + 1 <= list->capacity) { |
| iree_hal_device_retain(device); |
| list->devices[list->count++] = device; |
| } else { |
| status = iree_make_status(IREE_STATUS_RESOURCE_EXHAUSTED, |
| "list capacity %" PRIhsz |
| " reached; no more devices can be added", |
| list->capacity); |
| } |
| IREE_TRACE_ZONE_END(z0); |
| return status; |
| } |
| |
| IREE_API_EXPORT iree_hal_device_t* iree_hal_device_list_at( |
| const iree_hal_device_list_t* list, iree_host_size_t i) { |
| IREE_ASSERT_ARGUMENT(list); |
| return i < list->count ? list->devices[i] : NULL; |
| } |