hw-native-sys · ChaoZheng109 · Sep 3, 2026 · Sep 2, 2026
diff --git a/src/a2a3/runtime/host_build_graph/aicpu/aicpu_executor.cpp b/src/a2a3/runtime/host_build_graph/aicpu/aicpu_executor.cpp
@@ -8,48 +8,27 @@
  * See LICENSE in the root of the software repository for the full text of the License.
  * -----------------------------------------------------------------------------------------------------------
  */
-#include <unistd.h>
-
 #include <atomic>
-#include <cinttypes>
 #include <cstdint>
 #include <cstdio>
 #include <cstdlib>
 #include <cstring>
-#ifdef __linux__
-#include <sys/mman.h>
-#endif
 
-#include "aicpu/device_time.h"
 #include "aicpu/device_phase_aicpu.h"
-#include "callable_protocol.h"
-#include "dispatch_payload.h"
 #include "runtime.h"
 #include "spin_hint.h"
 
 // Runtime headers (full struct definition for create/destroy + SIMPLER_SCOPE)
 #include "host_build_graph/runtime_core.h"
-#include "host_build_graph/runtime_types.h"
 #include "host_build_graph/shared_memory.h"
 
-// Performance profiling headers
-#include "aicpu/chip_swimlane_collector_aicpu.h"
-#include "aicpu/args_dump_aicpu.h"
-#include "common/chip_swimlane_profiling.h"
 #include "common/unified_log.h"
 
 // Register-based communication
 #include "aicpu/platform_aicpu_affinity.h"
 #include "aicpu/platform_regs.h"
-#include "common/platform_config.h"
 #include "utils/thread_completion_gate.h"
 
-// Core type definitions
-#include "common/core_type.h"
-
-// CoreCallable for resolved dispatch address
-#include "callable.h"
-
 // Scheduler data structures (CoreExecState, CoreTracker, etc.)
 #include "scheduler/scheduler_types.h"
 

diff --git a/src/a2a3/runtime/host_build_graph/host/runtime_maker.cpp b/src/a2a3/runtime/host_build_graph/host/runtime_maker.cpp
@@ -48,13 +48,12 @@
 #include <memory>
 #include <optional>
 #include <string>
-#include <type_traits>
 #include <utility>
 #include <unordered_map>
 #include <vector>
 
+#include "assert_compat.h"
 #include "host_build_graph/runtime_status.h"
-#include "host_build_graph/common.h"
 #include "host_build_graph/dep_gen_host_graph.h"
 #include "host_build_graph/graph_execution.h"
 #include "host_build_graph/host_tensor_access.h"
@@ -72,7 +71,7 @@
 #include "../../../../common/worker/runtime_c_api.h"
 #include "callable.h"
 #include "common/host_log_binding.h"
-#include "common/log_clock.h"
+#include "common/host_phase_kind.h"
 #include "common/platform_config.h"
 #include "common/unified_log.h"
 #include "host_log.h"

diff --git a/src/a2a3/runtime/host_build_graph/orchestration/common.cpp b/src/a2a3/runtime/host_build_graph/orchestration/common.cpp
@@ -8,13 +8,13 @@
  * See LICENSE in the root of the software repository for the full text of the License.
  * -----------------------------------------------------------------------------------------------------------
  */
+#include "assert_compat.h"
 #include "host_build_graph/common.h"
 
 #ifdef __linux__
 #include <cxxabi.h>
 #include <dlfcn.h>
 #include <execinfo.h>
-#include <unistd.h>
 
 #include <array>
 #include <cstring>

diff --git a/src/a2a3/runtime/host_build_graph/orchestration/orchestration_api.h b/src/a2a3/runtime/host_build_graph/orchestration/orchestration_api.h
@@ -30,30 +30,22 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include <array>
-#include <algorithm>
-#include <condition_variable>
 #include <cstring>
 #include <ctime>
 #include <functional>
-#include <mutex>
-#include <thread>
 #include <tuple>
 #include <type_traits>
-#include <utility>
-#include <vector>
 
 // Type headers needed by orchestration
-#include "host_build_graph/common.h"            // framework_bind_runtime / framework_current_runtime
-#include "common/host_phase_kind.h"             // HostPhaseKind, for the phase records below
-#include "host_build_graph/graph_cache.h"       // Graph Execution key and result helpers
-#include "host_build_graph/graph_host_state.h"  // GRAPH_MAX_DEFINITIONS
-#include "host_build_graph/runtime_ops.h"       // RuntimeOps, RuntimeContext forward declaration
-#include "host_build_graph/runtime_types.h"     // SIMPLER_ERROR_*
-#include "host_build_graph/submit_types.h"      // MixedKernels, INVALID_KERNEL_ID, subtask slots
-#include "types.h"                              // Arg, TaskOutputTensors, TensorArgType
-#include "task_args.h"                          // ChipStorageTaskArgs, simpler::hbg::Tensor
-#include "tensor.h"                             // simpler::hbg::Tensor, TensorCreateInfo
+#include "assert_compat.h"                    // debug_assert
+#include "host_build_graph/common.h"          // framework_bind_runtime / framework_current_runtime
+#include "common/host_phase_kind.h"           // HostPhaseKind, for the phase records below
+#include "host_build_graph/graph_cache.h"     // Graph Execution key and result helpers
+#include "host_build_graph/runtime_ops.h"     // RuntimeOps, RuntimeContext forward declaration
+#include "host_build_graph/runtime_status.h"  // SIMPLER_ERROR_*
+#include "host_build_graph/submit_types.h"    // MixedKernels, INVALID_KERNEL_ID, subtask slots
+#include "types.h"                            // Arg, TaskOutputTensors, TensorArgType
+#include "tensor.h"                           // simpler::hbg::Tensor, TensorCreateInfo
 
 // =============================================================================
 // simpler::hbg::Tensor Factory Helpers

diff --git a/src/a2a3/runtime/host_build_graph/runtime/aicore_completion_mailbox.h b/src/a2a3/runtime/host_build_graph/runtime/aicore_completion_mailbox.h
@@ -14,7 +14,6 @@
 #include <atomic>
 #include <cstdint>
 
-#include "aicore_completion_mailbox_types.h"
 #include "host_build_graph/constants.h"
 #include "host_build_graph/task_id.h"
 

diff --git a/src/a2a3/runtime/host_build_graph/runtime/async_wait.h b/src/a2a3/runtime/host_build_graph/runtime/async_wait.h
@@ -12,14 +12,13 @@
 #pragma once
 
 #include <atomic>
-#include <cstddef>
 #include <cstdint>
 
-#include "aicpu/platform_regs.h"
 #include "backend/sdma/sdma_completion_scheduler.h"
-#include "intrinsic.h"
 #include "aicore_completion_mailbox.h"
+#include "aicore_completion_mailbox_types.h"
 #include "host_build_graph/completion_token.h"
+#include "host_build_graph/runtime_status.h"
 #include "host_build_graph/runtime_types.h"
 
 struct SchedulerState;

diff --git a/src/a2a3/runtime/host_build_graph/runtime/backend/sdma/sdma_completion_scheduler.h b/src/a2a3/runtime/host_build_graph/runtime/backend/sdma/sdma_completion_scheduler.h
@@ -11,12 +11,11 @@
 
 #pragma once
 
-#include <cstddef>
 #include <cstdint>
 
-#include "aicpu/platform_regs.h"
-#include "aicore_completion_mailbox.h"
+#include "aicpu/cache_maintenance.h"
 #include "host_build_graph/completion_token.h"
+#include "host_build_graph/constants.h"
 #include "host_build_graph/runtime_status.h"
 
 inline uintptr_t sdma_completion_cache_line(const volatile void *addr) {

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler.h b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler.h
@@ -34,18 +34,23 @@
 
 #include <atomic>
 
-#include "common/core_type.h"
 #include "common/memory_barrier.h"
 #include "utils/device_arena.h"
 #include "aicpu/platform_regs.h"  // get_reg_ptr / RegId for the early-dispatch doorbell
 #include "async_wait.h"
 #include "host_build_graph/graph_execution.h"
 #include "host_build_graph/task_id.h"
-#include "host_build_graph/task_allocator.h"
+#include "host_build_graph/runtime_status.h"
 #include "host_build_graph/runtime_types.h"
 #include "host_build_graph/shared_memory.h"
+// Defines the SIMPLER_*_PROFILING levels the conditionals below test. An #if on an
+// undefined macro evaluates to 0, so a profiling block reached through a transitive
+// include would switch itself off silently if that path ever went away.
+#include "profiling_config.h"
 
-#include "aicpu/device_time.h"  // get_sys_cnt_aicpu (used by early-dispatch doorbell timing too)
+#if SIMPLER_ORCH_PROFILING || SIMPLER_SCHED_PROFILING
+#include "aicpu/device_time.h"  // get_sys_cnt_aicpu, used only by the timed blocks below
+#endif
 #if SIMPLER_SCHED_PROFILING
 #define SCHED_CYCLE_START() uint64_t _st0 = get_sys_cnt_aicpu(), _st1
 #define SCHED_CYCLE_LAP(acc)        \

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_cold_path.cpp b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_cold_path.cpp
@@ -13,6 +13,7 @@
 #include <cinttypes>
 #include <cstdio>
 
+#include "assert_compat.h"
 #include "common/unified_log.h"
 #include "aicpu/device_time.h"
 #include "aicpu/chip_swimlane_collector_aicpu.h"
@@ -22,6 +23,7 @@
 #include "common/memory_barrier.h"
 #include "common/chip_swimlane_profiling.h"
 #include "common/platform_config.h"
+#include "host_build_graph/runtime_status.h"
 #include "host_build_graph/runtime_core.h"
 #include "host_build_graph/shared_memory.h"
 #include "runtime.h"

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_completion.cpp b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_completion.cpp
@@ -10,14 +10,12 @@
  */
 #include "scheduler_context.h"
 
-#include <algorithm>
-
 #include "common/unified_log.h"
 #include "aicpu/device_time.h"
-#include "aicpu/platform_regs.h"
 #include "common/chip_swimlane_profiling.h"
 #include "common/memory_barrier.h"
 #include "common/platform_config.h"
+#include "host_build_graph/runtime_status.h"
 #include "host_build_graph/runtime_core.h"
 #include "runtime.h"
 #include "spin_hint.h"

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_context.h b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_context.h
@@ -10,6 +10,7 @@
  */
 #pragma once
 
+#include "assert_compat.h"
 #include "aicpu/device_phase_aicpu.h"
 #include "aicpu/platform_regs.h"
 #include "common/chip_swimlane_profiling.h"
@@ -19,7 +20,6 @@
 
 #include "scheduler/scheduler.h"
 
-#include "aicore_completion_mailbox.h"
 #include "dispatch_payload.h"
 
 // runtime.h cannot be included here — it pulls in Handshake, which this header

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_dispatch.cpp b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_dispatch.cpp
@@ -14,17 +14,18 @@
 #include <cinttypes>
 #include <limits>
 
-#include "host_build_graph/common.h"  // debug_assert
+#include "assert_compat.h"  // debug_assert
+#include "host_build_graph/common.h"
 
 #include "common/unified_log.h"
 #include "aicpu/aicpu_device_config.h"
 #include "aicpu/device_time.h"
-#include "aicpu/platform_regs.h"
 #include "callable.h"
 #include "common/chip_swimlane_profiling.h"
 #include "common/memory_barrier.h"
 #include "common/platform_config.h"
 #include "host_build_graph/async_poll_phase_accumulator.h"
+#include "host_build_graph/runtime_status.h"
 #include "host_build_graph/runtime_core.h"
 #include "runtime.h"
 #include "spin_hint.h"

diff --git a/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_types.h b/src/a2a3/runtime/host_build_graph/runtime/scheduler/scheduler_types.h
@@ -14,6 +14,7 @@
 #include <cstddef>
 #include <cstdint>
 
+#include "assert_compat.h"
 #include "common/core_type.h"
 #include "common/platform_config.h"
 #include "host_build_graph/runtime_types.h"

diff --git a/src/a5/runtime/host_build_graph/aicpu/aicpu_executor.cpp b/src/a5/runtime/host_build_graph/aicpu/aicpu_executor.cpp
@@ -8,29 +8,21 @@
  * See LICENSE in the root of the software repository for the full text of the License.
  * -----------------------------------------------------------------------------------------------------------
  */
-#include <unistd.h>
-
 #include <atomic>
-#include <algorithm>
 #include <cinttypes>
 #include <cstdint>
 #include <cstdio>
 #include <cstdlib>
 #include <cstring>
-#ifdef __linux__
-#include <sys/mman.h>
-#endif
 
 #include "aicore_lifecycle.h"
 #include "aicore_scheduler_error.h"
 #include "aicore_scheduler_state.h"
 #include "aicpu/aicpu_device_config.h"
 #include "aicpu/cache_maintenance.h"
-#include "aicpu/args_dump_aicpu.h"
 #include "aicpu/chip_swimlane_collector_aicpu.h"
 #include "aicpu/device_phase_aicpu.h"
 #include "aicpu/device_time.h"
-#include "aicpu/pmu_collector_aicpu.h"
 #include "host_build_graph/runtime_status.h"
 #include "host_build_graph/shared_memory.h"
 #include "runtime.h"

diff --git a/src/a5/runtime/host_build_graph/host/runtime_maker.cpp b/src/a5/runtime/host_build_graph/host/runtime_maker.cpp
@@ -50,13 +50,12 @@
 #include <mutex>
 #include <optional>
 #include <string>
-#include <type_traits>
 #include <utility>
 #include <unordered_map>
 #include <vector>
 
+#include "assert_compat.h"
 #include "host_build_graph/runtime_status.h"
-#include "host_build_graph/common.h"
 #include "host_build_graph/dep_gen_host_graph.h"
 #include "host_build_graph/graph_execution.h"
 #include "scheduler/scheduler_graph.h"
@@ -76,7 +75,7 @@
 #include "../../../../common/worker/runtime_c_api.h"
 #include "callable.h"
 #include "common/host_log_binding.h"
-#include "common/log_clock.h"
+#include "common/host_phase_kind.h"
 #include "common/platform_config.h"
 #include "common/unified_log.h"
 #include "host_log.h"

diff --git a/src/a5/runtime/host_build_graph/orchestration/common.cpp b/src/a5/runtime/host_build_graph/orchestration/common.cpp
@@ -8,13 +8,13 @@
  * See LICENSE in the root of the software repository for the full text of the License.
  * -----------------------------------------------------------------------------------------------------------
  */
+#include "assert_compat.h"
 #include "host_build_graph/common.h"
 
 #ifdef __linux__
 #include <cxxabi.h>
 #include <dlfcn.h>
 #include <execinfo.h>
-#include <unistd.h>
 
 #include <array>
 #include <cstring>

diff --git a/src/a5/runtime/host_build_graph/orchestration/orchestration_api.h b/src/a5/runtime/host_build_graph/orchestration/orchestration_api.h
@@ -30,30 +30,22 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include <array>
-#include <algorithm>
-#include <condition_variable>
 #include <cstring>
 #include <ctime>
 #include <functional>
-#include <mutex>
-#include <thread>
 #include <tuple>
 #include <type_traits>
-#include <utility>
-#include <vector>
 
 // Type headers needed by orchestration
-#include "host_build_graph/common.h"            // framework_bind_runtime / framework_current_runtime
-#include "common/host_phase_kind.h"             // HostPhaseKind, for the phase records below
-#include "host_build_graph/graph_cache.h"       // Graph Execution key and result helpers
-#include "host_build_graph/graph_host_state.h"  // GRAPH_MAX_DEFINITIONS
-#include "host_build_graph/runtime_ops.h"       // RuntimeOps, RuntimeContext forward declaration
-#include "host_build_graph/runtime_types.h"     // SIMPLER_ERROR_*
-#include "host_build_graph/submit_types.h"      // MixedKernels, INVALID_KERNEL_ID, subtask slots
-#include "types.h"                              // Arg, TaskOutputTensors, TensorArgType
-#include "task_args.h"                          // ChipStorageTaskArgs, simpler::hbg::Tensor
-#include "tensor.h"                             // simpler::hbg::Tensor, TensorCreateInfo
+#include "assert_compat.h"                    // debug_assert
+#include "host_build_graph/common.h"          // framework_bind_runtime / framework_current_runtime
+#include "common/host_phase_kind.h"           // HostPhaseKind, for the phase records below
+#include "host_build_graph/graph_cache.h"     // Graph Execution key and result helpers
+#include "host_build_graph/runtime_ops.h"     // RuntimeOps, RuntimeContext forward declaration
+#include "host_build_graph/runtime_status.h"  // SIMPLER_ERROR_*
+#include "host_build_graph/submit_types.h"    // MixedKernels, INVALID_KERNEL_ID, subtask slots
+#include "types.h"                            // Arg, TaskOutputTensors, TensorArgType
+#include "tensor.h"                           // simpler::hbg::Tensor, TensorCreateInfo
 
 // =============================================================================
 // simpler::hbg::Tensor Factory Helpers

diff --git a/src/a5/runtime/host_build_graph/runtime/aicore_completion_mailbox.h b/src/a5/runtime/host_build_graph/runtime/aicore_completion_mailbox.h
@@ -14,7 +14,6 @@
 #include <atomic>
 #include <cstdint>
 
-#include "aicore_completion_mailbox_types.h"
 #include "host_build_graph/constants.h"
 #include "host_build_graph/task_id.h"