feat(profiler): add CPU-to-NPU correlation for ACLprof-pytorch-AtomGit

MCMakeLists.txt+14-2

	@@ -416,7 +416,7 @@ if (DEFINED BUILD_FXRT)
416	add_subdirectory(${PROJECT_SOURCE_DIR}/fxrt/fxrt)	416	add_subdirectory(${PROJECT_SOURCE_DIR}/fxrt/fxrt)
417	endif()	417	endif()
418		418
419	-if (DEFINED BUILD_GTEST)	419	+if (BUILD_GTEST)
420	enable_testing()	420	enable_testing()
421	SET(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/build/gtest)	421	SET(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/build/gtest)
422	add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/googletest)	422	add_subdirectory(${PROJECT_SOURCE_DIR}/third_party/googletest)
	@@ -426,8 +426,20 @@ if (DEFINED BUILD_GTEST)
426	add_subdirectory(${PROJECT_SOURCE_DIR}/test/cpp/api)	426	add_subdirectory(${PROJECT_SOURCE_DIR}/test/cpp/api)
427	add_executable(test_api ${TORCH_API_TEST_SOURCES})	427	add_executable(test_api ${TORCH_API_TEST_SOURCES})
428		428
		429	+ find_path(KINETO_TEST_INCLUDE_DIR ActivityBuffers.h
		430	+ HINTS "${PYTORCH_INSTALL_DIR}/include/kineto"
		431	+ NO_DEFAULT_PATH
		432	+ DOC "Kineto internal headers matching installed PyTorch (libkineto/src)")
		433	+ if(NOT EXISTS "${KINETO_TEST_INCLUDE_DIR}/ActivityBuffers.h")
		434	+ message(FATAL_ERROR
		435	+ "test_api requires ActivityBuffers.h. Set KINETO_TEST_INCLUDE_DIR to "
		436	+ "libkineto/src from the Kineto version used to build installed PyTorch.")
		437	+ endif()
		438	+ target_include_directories(test_api PRIVATE "${KINETO_TEST_INCLUDE_DIR}")
		439	+ target_compile_definitions(test_api PRIVATE KINETO_NAMESPACE=libkineto)
		440	+
429	target_link_libraries(test_api PUBLIC torch_npu)	441	target_link_libraries(test_api PUBLIC torch_npu)
430	- target_link_libraries(test_api PUBLIC gtest_main gtest)	442	+ target_link_libraries(test_api PUBLIC gtest_main gtest gmock)
431	endif()	443	endif()
432		444
433	if (DEFINED BUILD_LIBTORCH)	445	if (DEFINED BUILD_LIBTORCH)

Mtest/cpp/api/CMakeLists.txt+4-1

	@@ -1,2 +1,5 @@
1	set(TORCH_API_TEST_DIR "${PROJECT_SOURCE_DIR}/test/cpp/api")	1	set(TORCH_API_TEST_DIR "${PROJECT_SOURCE_DIR}/test/cpp/api")
2	-set(TORCH_API_TEST_SOURCES ${TORCH_API_TEST_DIR}/modules.cpp PARENT_SCOPE)	2	+set(TORCH_API_TEST_SOURCES
		3	+ ${TORCH_API_TEST_DIR}/modules.cpp
		4	+ ${TORCH_API_TEST_DIR}/profiler.cpp
		5	+ PARENT_SCOPE)

Atest/cpp/api/profiler.cpp+941-0

Mtest/profiler/test_experimental_config.py+98-1

	@@ -1,16 +1,107 @@
		1	+import json
		2	+import pickle
1	import unittest	3	import unittest
2	-from unittest.mock import patch	4	+from unittest.mock import PropertyMock, patch
3		5
		6	+from torch._C._profiler import _ExperimentalConfig as KinetoExperimentalConfig
		7	+from torch_npu.profiler import _NPUExperimentalConfig
4	from torch_npu.profiler.experimental_config import supported_ai_core_metrics	8	from torch_npu.profiler.experimental_config import supported_ai_core_metrics
5	from torch_npu.profiler.experimental_config import supported_profiler_level	9	from torch_npu.profiler.experimental_config import supported_profiler_level
6	from torch_npu.profiler.experimental_config import supported_export_type	10	from torch_npu.profiler.experimental_config import supported_export_type
7	from torch_npu.profiler.experimental_config import _ExperimentalConfig	11	from torch_npu.profiler.experimental_config import _ExperimentalConfig
		12	+from torch_npu.profiler.experimental_config import ExportType, ProfilerLevel
8	from torch_npu._C._profiler import _ExperimentalConfig as Cpp_ExperimentalConfig	13	from torch_npu._C._profiler import _ExperimentalConfig as Cpp_ExperimentalConfig
9	from torch_npu.testing.testcase import TestCase, run_tests	14	from torch_npu.testing.testcase import TestCase, run_tests
10	from torch_npu.profiler.analysis.prof_common_func._constant import Constant	15	from torch_npu.profiler.analysis.prof_common_func._constant import Constant
11	from torch_npu.profiler.analysis.prof_common_func._cann_package_manager import CannPackageManager	16	from torch_npu.profiler.analysis.prof_common_func._cann_package_manager import CannPackageManager
12		17
13		18
		19	+class TestNPUExperimentalConfig(unittest.TestCase):
		20	+ def test_backend_only_compatibility(self):
		21	+ self.assertEqual(str(_NPUExperimentalConfig()), '{"backend":"mspti"}')
		22	+ for backend in ["mspti", "aclprof", "invalid", None, 1]:
		23	+ with self.subTest(backend=backend):
		24	+ config = _NPUExperimentalConfig(backend=backend)
		25	+ self.assertEqual(str(config), json.dumps({"backend": backend}, separators=(",", ":")))
		26	+ self.assertEqual(config.__getstate__(), KinetoExperimentalConfig(
		27	+ custom_profiler_config=str(config)).__getstate__())
		28	+
		29	+ def test_configuration_combinations(self):
		30	+ for backend in [None, "mspti", "aclprof"]:
		31	+ for options in [{"profiler_level": ProfilerLevel.Level1}, {"export_type": ExportType.Db},
		32	+ {"profiler_level": ProfilerLevel.Level1, "export_type": [ExportType.Db]}]:
		33	+ with self.subTest(backend=backend, options=options):
		34	+ kwargs = {} if backend is None else {"backend": backend}
		35	+ data = json.loads(str(_NPUExperimentalConfig(kwargs, options)))
		36	+ expected = {"backend": backend or "mspti", **options}
		37	+ if "export_type" in options:
		38	+ expected["export_type"] = [ExportType.Db]
		39	+ self.assertEqual(data, expected)
		40	+
		41	+ def test_supported_levels(self):
		42	+ for level in supported_profiler_level():
		43	+ with self.subTest(level=level):
		44	+ self.assertEqual(json.loads(str(_NPUExperimentalConfig(profiler_level=level)))["profiler_level"], level)
		45	+
		46	+ def test_export_normalization_and_determinism(self):
		47	+ scalar = _NPUExperimentalConfig(export_type=ExportType.Db)
		48	+ self.assertEqual(str(scalar), str(_NPUExperimentalConfig(export_type=[ExportType.Db])))
		49	+ exports = [ExportType.Text, ExportType.Db, ExportType.Text]
		50	+ config = _NPUExperimentalConfig(export_type=exports)
		51	+ self.assertEqual(str(config), '{"backend":"mspti","export_type":["db","text"]}')
		52	+ exports.clear()
		53	+ self.assertEqual(str(config), str(_NPUExperimentalConfig(export_type=[ExportType.Db, ExportType.Text])))
		54	+ self.assertEqual(str(_NPUExperimentalConfig(profiler_level=None, export_type=None)),
		55	+ str(_NPUExperimentalConfig()))
		56	+
		57	+ def test_invalid_profiler_level(self):
		58	+ for level in ["invalid", "Level3", "", 1, True, 1.0, [], {}]:
		59	+ with self.subTest(level=level), self.assertRaisesRegex(ValueError, "profiler_level"):
		60	+ _NPUExperimentalConfig(profiler_level=level)
		61	+
		62	+ def test_invalid_export_type(self):
		63	+ for exports in ["invalid", "db,text", "", 1, True, [], {}, (ExportType.Db,), [ExportType.Db, 1], [[]]]:
		64	+ with self.subTest(exports=exports), self.assertRaisesRegex(ValueError, "export_type"):
		65	+ _NPUExperimentalConfig(export_type=exports)
		66	+
		67	+ def test_standard_channel_and_pickle_roundtrip(self):
		68	+ config = _NPUExperimentalConfig(backend="aclprof", profiler_level=ProfilerLevel.Level1,
		69	+ export_type=ExportType.Db, verbose=True)
		70	+ payload = str(config)
		71	+ self.assertEqual(payload, '{"backend":"aclprof","profiler_level":"Level1","export_type":["db"]}')
		72	+ standard = KinetoExperimentalConfig(custom_profiler_config=payload, verbose=True)
		73	+ self.assertEqual(config.__getstate__(), standard.__getstate__())
		74	+ for value in [config, standard]:
		75	+ restored = pickle.loads(pickle.dumps(value))
		76	+ self.assertEqual(restored.__getstate__(), value.__getstate__())
		77	+ self.assertEqual(str(pickle.loads(pickle.dumps(config))), payload)
		78	+ # Newer PyTorch versions append unrelated fields after the custom config.
		79	+ with patch.object(_NPUExperimentalConfig, "__getstate__", return_value=(*config.__getstate__(), False, False)):
		80	+ self.assertEqual(str(config), payload)
		81	+
		82	+ def test_str_uses_direct_attribute_without_serializing(self):
		83	+ config = _NPUExperimentalConfig()
		84	+ payload = '{"backend":"aclprof"}'
		85	+ with patch.object(_NPUExperimentalConfig, "custom_profiler_config", payload, create=True), \
		86	+ patch.object(_NPUExperimentalConfig, "__getstate__", side_effect=AssertionError("unexpected fallback")) as state:
		87	+ self.assertEqual(str(config), payload)
		88	+ state.assert_not_called()
		89	+
		90	+ def test_str_falls_back_when_attribute_is_unavailable(self):
		91	+ config = _NPUExperimentalConfig(backend="aclprof")
		92	+ payload = '{"backend":"aclprof"}'
		93	+ with patch.object(_NPUExperimentalConfig, "custom_profiler_config", create=True,
		94	+ new_callable=PropertyMock, side_effect=AttributeError):
		95	+ with patch.object(_NPUExperimentalConfig, "__getstate__", return_value=config.__getstate__()) as state:
		96	+ self.assertEqual(str(config), payload)
		97	+ state.assert_called_once_with()
		98	+ self.assertEqual(str(pickle.loads(pickle.dumps(config))), payload)
		99	+
		100	+ def test_json_escaping(self):
		101	+ backend = 'aclprof",\n"profiler_level":"Level2'
		102	+ self.assertEqual(json.loads(str(_NPUExperimentalConfig(backend=backend))), {"backend": backend})
		103	+
		104	+
14	class TestExperimentalConfig(TestCase):	105	class TestExperimentalConfig(TestCase):
15		106
16	@classmethod	107	@classmethod
	@@ -47,6 +138,12 @@ class TestExperimentalConfig(TestCase):
47	experimental_config = _ExperimentalConfig()	138	experimental_config = _ExperimentalConfig()
48	self.assertTrue(isinstance(experimental_config(), Cpp_ExperimentalConfig))	139	self.assertTrue(isinstance(experimental_config(), Cpp_ExperimentalConfig))
49		140
		141	+ def test_kineto_backend_config(self):
		142	+ standard_config = pickle.dumps(KinetoExperimentalConfig())
		143	+ aclprof_config = pickle.dumps(_NPUExperimentalConfig(backend="aclprof"))
		144	+ self.assertNotIn(b'{"backend":"aclprof"}', standard_config)
		145	+ self.assertIn(b'{"backend":"aclprof"}', aclprof_config)
		146	+
50	def test_mstx_domain_switches_will_reset_when_msproftx_and_mstx_not_enabled(self):	147	def test_mstx_domain_switches_will_reset_when_msproftx_and_mstx_not_enabled(self):
51	experimental_config = _ExperimentalConfig(msprof_tx=False,	148	experimental_config = _ExperimentalConfig(msprof_tx=False,
52	mstx=False,	149	mstx=False,

Mtorch_npu/csrc/core/npu/interface/AclInterface.cpp+11-0

	@@ -136,6 +136,7 @@ TORCH_NPU_LOAD_FUNC(aclrtMemset)
136	TORCH_NPU_LOAD_FUNC(aclrtMemsetAsync)	136	TORCH_NPU_LOAD_FUNC(aclrtMemsetAsync)
137	TORCH_NPU_LOAD_FUNC(aclmdlRICaptureThreadExchangeMode)	137	TORCH_NPU_LOAD_FUNC(aclmdlRICaptureThreadExchangeMode)
138	TORCH_NPU_LOAD_FUNC(aclrtGetLogicDevIdByUserDevId)	138	TORCH_NPU_LOAD_FUNC(aclrtGetLogicDevIdByUserDevId)
		139	+TORCH_NPU_LOAD_FUNC(aclrtGetPhyDevIdByUserDevId)
139	TORCH_NPU_LOAD_FUNC(aclrtBinaryLoadFromData)	140	TORCH_NPU_LOAD_FUNC(aclrtBinaryLoadFromData)
140	TORCH_NPU_LOAD_FUNC(aclrtBinaryGetFunction)	141	TORCH_NPU_LOAD_FUNC(aclrtBinaryGetFunction)
141	TORCH_NPU_LOAD_FUNC(aclrtBinaryUnLoad)	142	TORCH_NPU_LOAD_FUNC(aclrtBinaryUnLoad)
	@@ -1943,6 +1944,16 @@ aclError AclrtGetLogicDevIdByUserDevId(const int32_t userDevid, int32_t* const l
1943	TORCH_CHECK(func, "Failed to find function aclrtGetLogicDevIdByUserDevId", PTA_ERROR(ErrCode::NOT_FOUND));	1944	TORCH_CHECK(func, "Failed to find function aclrtGetLogicDevIdByUserDevId", PTA_ERROR(ErrCode::NOT_FOUND));
1944	return func(userDevid, logicDevId);	1945	return func(userDevid, logicDevId);
1945	}	1946	}
		1947	+
		1948	+aclError AclrtGetPhyDevIdByUserDevId(const int32_t userDevId, int32_t* const phyDevId) {
		1949	+ typedef aclError (AclrtGetPhyDevIdByUserDevId)(const int32_t, int32_t const);
		1950	+ static AclrtGetPhyDevIdByUserDevId func = nullptr;
		1951	+ if (func == nullptr) {
		1952	+ func = (AclrtGetPhyDevIdByUserDevId)TORCH_NPU_GET_FUNC(aclrtGetPhyDevIdByUserDevId);
		1953	+ }
		1954	+ TORCH_CHECK(func, "Failed to find function aclrtGetPhyDevIdByUserDevId", PTA_ERROR(ErrCode::NOT_FOUND));
		1955	+ return func(userDevId, phyDevId);
		1956	+}
1946	bool IsExistAclrtLaunchKernelWithHostArgs() {	1957	bool IsExistAclrtLaunchKernelWithHostArgs() {
1947	using Func = decltype(&aclrtLaunchKernelWithHostArgs);	1958	using Func = decltype(&aclrtLaunchKernelWithHostArgs);
1948	static Func func = (Func)TORCH_NPU_GET_FUNC(aclrtLaunchKernelWithHostArgs);	1959	static Func func = (Func)TORCH_NPU_GET_FUNC(aclrtLaunchKernelWithHostArgs);

Mtorch_npu/csrc/core/npu/interface/AclInterface.h+2-0

	@@ -442,6 +442,8 @@ aclError AclrtGetDeviceInfo(uint32_t deviceId, aclrtDevAttr attr, int64_t* value
442		442
443	aclError AclrtGetLogicDevIdByUserDevId(const int32_t userDevid, int32_t* const logicDevId);	443	aclError AclrtGetLogicDevIdByUserDevId(const int32_t userDevid, int32_t* const logicDevId);
444		444
		445	+aclError AclrtGetPhyDevIdByUserDevId(const int32_t userDevId, int32_t* const phyDevId);
		446	+
445	aclError AclrtMemSet(void* devPtr, size_t maxCount, int32_t value, size_t count);	447	aclError AclrtMemSet(void* devPtr, size_t maxCount, int32_t value, size_t count);
446		448
447	aclError AclrtMemSetAsync(void* devPtr, size_t maxCount, int32_t value, size_t count, aclrtStream stream);	449	aclError AclrtMemSetAsync(void* devPtr, size_t maxCount, int32_t value, size_t count, aclrtStream stream);

Mtorch_npu/csrc/framework/utils/NpuUtils.cpp+5-2

	@@ -17,6 +17,7 @@
17	#include "torch_npu/csrc/framework/utils/OpAdapter.h"	17	#include "torch_npu/csrc/framework/utils/OpAdapter.h"
18	#include "torch_npu/csrc/framework/utils/OpPreparation.h"	18	#include "torch_npu/csrc/framework/utils/OpPreparation.h"
19	#ifndef BUILD_LIBTORCH	19	#ifndef BUILD_LIBTORCH
		20	+#include "torch_npu/csrc/profiler/aclprof_activity_profiler.h"
20	#include "torch_npu/csrc/profiler/npu_profiler.h"	21	#include "torch_npu/csrc/profiler/npu_profiler.h"
21	#endif	22	#endif
22		23
	@@ -275,7 +276,8 @@ void NpuUtils::ProfReportMarkDataToNpuProfiler(uint32_t category, const std::str
275	if (data.empty()) {	276	if (data.empty()) {
276	return;	277	return;
277	}	278	}
278	- if (torch_npu::profiler::profDataReportEnable().load(std::memory_order_relaxed)) {	279	+ if (torch_npu::profiler::profDataReportEnable().load(std::memory_order_relaxed) \|\|
		280	+ torch_npu::profiler::aclprofTaskQueueCaptureEnabled()) {
279	torch_npu::profiler::reportMarkDataToNpuProfiler(category, data, correlation_id);	281	torch_npu::profiler::reportMarkDataToNpuProfiler(category, data, correlation_id);
280	}	282	}
281	}	283	}
	@@ -304,7 +306,8 @@ void NpuUtils::ProfReportMarkDataToNpuProfiler(uint32_t category, void* data, si
304	if (C10_UNLIKELY(!data)) {	306	if (C10_UNLIKELY(!data)) {
305	return;	307	return;
306	}	308	}
307	- if (torch_npu::profiler::profDataReportEnable().load(std::memory_order_relaxed)) {	309	+ if (torch_npu::profiler::profDataReportEnable().load(std::memory_order_relaxed) \|\|
		310	+ torch_npu::profiler::aclprofTaskQueueCaptureEnabled()) {
308	static const std::map<int64_t, DqueueCall> DEQUEUE_CALL_FUNC_MAP{	311	static const std::map<int64_t, DqueueCall> DEQUEUE_CALL_FUNC_MAP{
309	{c10_npu::queue::COMPILE_AND_EXECUTE, &DqueueCompileExcute},	312	{c10_npu::queue::COMPILE_AND_EXECUTE, &DqueueCompileExcute},
310	{c10_npu::queue::EXECUTE_OPAPI, &DqueueCompileExcuteOpApi},	313	{c10_npu::queue::EXECUTE_OPAPI, &DqueueCompileExcuteOpApi},

Atorch_npu/csrc/profiler/aclprof_activity_profiler.cpp+993-0

Atorch_npu/csrc/profiler/aclprof_activity_profiler.h+124-0

@@ -0,0 +1,124 @@
	1	+// Copyright (c) 2026 Huawei Technologies Co., Ltd
	2	+// All rights reserved.
	3	+//
	4	+// Licensed under the BSD 3-Clause License (the "License");
	5	+// you may not use this file except in compliance with the License.
	6	+// You may obtain a copy of the License at
	7	+//
	8	+// https://opensource.org/licenses/BSD-3-Clause
	9	+//
	10	+// Unless required by applicable law or agreed to in writing, software
	11	+// distributed under the License is distributed on an "AS IS" BASIS,
	12	+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	13	+// See the License for the specific language governing permissions and
	14	+// limitations under the License.
	15	+
	16	+#pragma once
	17	+
	18	+#include <cstdint>
	19	+#include <memory>
	20	+#include <mutex>
	21	+#include <optional>
	22	+#include <set>
	23	+#include <string>
	24	+#include <unordered_map>
	25	+#include <vector>
	26	+
	27	+#include "torch_npu/csrc/profiler/npu_activity_profiler.h"
	28	+
	29	+namespace torch_npu {
	30	+namespace profiler {
	31	+
	32	+namespace detail {
	33	+struct AclprofApi;
	34	+struct AclprofEvent;
	35	+} // namespace detail
	36	+
	37	+class AclprofActivityProfilerTestPeer;
	38	+
	39	+TORCH_NPU_API bool aclprofTaskQueueCaptureEnabled();
	40	+TORCH_NPU_API void recordAclprofTaskQueueEvent(
	41	+ uint32_t category,
	42	+ uint64_t taskQueueCorrelation,
	43	+ int64_t monotonicTimestampNs,
	44	+ int64_t pid,
	45	+ int64_t tid,
	46	+ const std::string& name);
	47	+
	48	+class TORCH_NPU_API AclprofActivityProfiler : public NPUActivityProfiler {
	49	+ public:
	50	+ explicit AclprofActivityProfiler(std::set<libkineto::ActivityType> activities);
	51	+ ~AclprofActivityProfiler() override;
	52	+
	53	+ const char* name() const override {
	54	+ return "aclprof";
	55	+ }
	56	+ void start() override;
	57	+ void stop() override;
	58	+ std::vector<std::string> errors() override;
	59	+ void processTrace(libkineto::ActivityLogger& logger) override;
	60	+ void processTrace(
	61	+ libkineto::ActivityLogger& logger,
	62	+ libkineto::getLinkedActivityCallback getLinkedActivity,
	63	+ int64_t startTime,
	64	+ int64_t endTime) override;
	65	+ std::unique_ptr<libkineto::DeviceInfo> getDeviceInfo() override;
	66	+ std::vector<libkineto::ResourceInfo> getResourceInfos() override;
	67	+ std::unique_ptr<libkineto::CpuTraceBuffer> getTraceBuffer() override;
	68	+ void pushCorrelationId(uint64_t id) override;
	69	+ void popCorrelationId() override;
	70	+
	71	+ private:
	72	+ struct TaskQueueMark {
	73	+ uint32_t category;
	74	+ uint64_t taskQueueCorrelation;
	75	+ int64_t timestampNs;
	76	+ int64_t tid;
	77	+ std::string name;
	78	+ std::optional<int32_t> kinetoCorrelation;
	79	+ };
	80	+
	81	+ AclprofActivityProfiler(std::set<libkineto::ActivityType> activities, std::shared_ptr<detail::AclprofApi> api);
	82	+ bool finishCollection(bool reportErrors);
	83	+ void cleanup();
	84	+ void resetCorrelationState();
	85	+ void beginCorrelationCapture();
	86	+ void endCorrelationCapture();
	87	+ void recordTaskQueueEvent(
	88	+ uint32_t category,
	89	+ uint64_t taskQueueCorrelation,
	90	+ int64_t monotonicTimestampNs,
	91	+ int64_t pid,
	92	+ int64_t tid,
	93	+ const std::string& name);
	94	+ void loadTaskQueueClockOffset(const std::string& profile);
	95	+ void resolveCorrelations();
	96	+
	97	+ friend class AclprofActivityProfilerTestPeer;
	98	+ friend void recordAclprofTaskQueueEvent(
	99	+ uint32_t category,
	100	+ uint64_t taskQueueCorrelation,
	101	+ int64_t monotonicTimestampNs,
	102	+ int64_t pid,
	103	+ int64_t tid,
	104	+ const std::string& name);
	105	+ std::set<libkineto::ActivityType> activities_;
	106	+ std::shared_ptr<detail::AclprofApi> api_;
	107	+ std::vector<detail::AclprofEvent> events_;
	108	+ std::vector<std::string> errors_;
	109	+ std::string root_;
	110	+ void* config_ = nullptr;
	111	+ int64_t device_ = -1;
	112	+ std::mutex correlationMutex_;
	113	+ std::unordered_map<int64_t, std::vector<std::optional<int32_t>>> activeCorrelations_;
	114	+ std::vector<TaskQueueMark> taskQueueMarks_;
	115	+ std::optional<int64_t> taskQueueClockOffsetNs_;
	116	+ std::vector<std::optional<int32_t>> linkedCorrelations_;
	117	+ bool initialized_ = false;
	118	+ bool started_ = false;
	119	+ bool captureCorrelations_ = false;
	120	+ bool correlationsResolved_ = false;
	121	+};
	122	+
	123	+} // namespace profiler
	124	+} // namespace torch_npu

Atorch_npu/csrc/profiler/aclprof_activity_profiler_internal.h+42-0

@@ -0,0 +1,42 @@
	1	+#pragma once
	2	+
	3	+#include <cstdint>
	4	+#include <functional>
	5	+#include <string>
	6	+#include <vector>
	7	+
	8	+namespace torch_npu::profiler::detail {
	9	+
	10	+struct AclprofEvent {
	11	+ std::string name;
	12	+ std::string taskType;
	13	+ int64_t startNs = 0;
	14	+ int64_t endNs = 0;
	15	+ int64_t device = 0;
	16	+ int64_t stream = 0;
	17	+ int64_t correlation = 0;
	18	+ uint64_t hostLaunchId = 0;
	19	+ int64_t flowStartNs = 0;
	20	+ int64_t flowStartPid = -1;
	21	+ int64_t flowStartTid = -1;
	22	+};
	23	+
	24	+// physicalDevice is the runtime-resolved physical device for this single-device session.
	25	+std::vector<AclprofEvent> parseAclprofTimeline(
	26	+ const std::string& path,
	27	+ int64_t physicalDevice,
	28	+ std::vector<std::string>& errors);
	29	+
	30	+struct AclprofApi {
	31	+ std::function<int32_t(int32_t*)> getDevice;
	32	+ std::function<int32_t(int32_t, int32_t*)> getPhysicalDevice;
	33	+ std::function<int32_t(const std::string&)> init;
	34	+ std::function<void(uint32_t)> createConfig;
	35	+ std::function<int32_t()> start;
	36	+ std::function<int32_t()> registerDeviceCallback;
	37	+ std::function<int32_t()> stop;
	38	+ std::function<int32_t()> destroyConfig;
	39	+ std::function<int32_t()> finalize;
	40	+};
	41	+
	42	+} // namespace torch_npu::profiler::detail

Mtorch_npu/csrc/profiler/mspti_activity_profiler.cpp+0-40

	@@ -19,7 +19,6 @@
19	#include "torch_npu/csrc/framework/interface/MsptiInterface.h"	19	#include "torch_npu/csrc/framework/interface/MsptiInterface.h"
20	#include "torch_npu/csrc/core/npu/npu_log.h"	20	#include "torch_npu/csrc/core/npu/npu_log.h"
21	#include "torch_npu/csrc/toolkit/profiler/common/utils.h"	21	#include "torch_npu/csrc/toolkit/profiler/common/utils.h"
22	-#include <torch/csrc/profiler/standalone/privateuse1_profiler.h>
23		22
24	namespace torch_npu {	23	namespace torch_npu {
25	namespace profiler {	24	namespace profiler {
	@@ -47,12 +46,6 @@ bool g_sessionActive = false;
47		46
48	int64_t g_corrCounter = 1LL << 30; // 1,073,741,824 (< 2^32 - 1)	47	int64_t g_corrCounter = 1LL << 30; // 1,073,741,824 (< 2^32 - 1)
49		48
50	-bool privateUse1Requested(const std::set<libkineto::ActivityType>& types) {
51	- using AT = libkineto::ActivityType;
52	- return types.count(AT::CONCURRENT_KERNEL) != 0 \|\| types.count(AT::GPU_MEMCPY) != 0 \|\|
53	- types.count(AT::GPU_MEMSET) != 0 \|\| types.count(AT::GPU_USER_ANNOTATION) != 0;
54	-}
55	-
56	std::vector<msptiActivityKind> computeEnabledKinds(const std::set<libkineto::ActivityType>& acts) {	49	std::vector<msptiActivityKind> computeEnabledKinds(const std::set<libkineto::ActivityType>& acts) {
57	using AT = libkineto::ActivityType;	50	using AT = libkineto::ActivityType;
58	std::vector<msptiActivityKind> ks;	51	std::vector<msptiActivityKind> ks;
	@@ -426,38 +419,5 @@ std::unique_ptr<libkineto::CpuTraceBuffer> MsptiActivityProfiler::getTraceBuffer
426	return nullptr;	419	return nullptr;
427	}	420	}
428		421
429	-std::unique_ptr<libkineto::IActivityProfilerSession> MsptiActivityProfilerPoc::configure(
430	- const std::set<libkineto::ActivityType>& types,
431	- const libkineto::Config&) {
432	- if (!privateUse1Requested(types)) {
433	- return nullptr;
434	- }
435	- return std::make_unique<MsptiActivityProfiler>(types);
436	-}
437	-
438	-std::unique_ptr<libkineto::IActivityProfilerSession> MsptiActivityProfilerPoc::configure(
439	- int64_t,
440	- int64_t,
441	- const std::set<libkineto::ActivityType>& types,
442	- const libkineto::Config& cfg) {
443	- return configure(types, cfg);
444	-}
445	-
446	-// ------------------------------ Registration -------------------------------
447	-// Prefer the upstream macro (stores the factory in PrivateUse1ProfilerRegistry);
448	-#ifdef REGISTER_PRIVATEUSE1_PROFILER
449	-REGISTER_PRIVATEUSE1_PROFILER(MsptiActivityProfilerPoc);
450	-#else
451	-namespace {
452	-struct MsptiAutoRegister {
453	- MsptiAutoRegister() {
454	- libkineto::api().registerProfilerFactory(
455	- []() -> std::unique_ptr<libkineto::IActivityProfiler> { return std::make_unique<MsptiActivityProfilerPoc>(); });
456	- }
457	-};
458	-static MsptiAutoRegister g_mspti_auto_register;
459	-} // namespace
460	-#endif
461	-
462	} // namespace profiler	422	} // namespace profiler
463	} // namespace torch_npu	423	} // namespace torch_npu

Mtorch_npu/csrc/profiler/mspti_activity_profiler.h+6-30

	@@ -1,17 +1,12 @@
1	#pragma once	1	#pragma once
2	-// mspti NPU activity-profiler session + libkineto bridge.	2	+// MSPTI-specific NPU activity-profiler backend.
3	-//
4	-// MsptiActivityProfilerPoc (libkineto::IActivityProfiler) registers for
5	-// ProfilerActivity.PrivateUse1; configure() returns a MsptiActivityProfiler,
6	-// which is the session libkineto drives. The session enables mspti kinds from
7	-// the requested activity-type set and collects NPU kernel and host API records.
8		3
9	#include <memory>	4	#include <memory>
10	#include <set>	5	#include <set>
11	#include <string>	6	#include <string>
12	#include <vector>	7	#include <vector>
13		8
14	-#include "IActivityProfiler.h"	9	+#include "torch_npu/csrc/profiler/npu_activity_profiler.h"
15		10
16	namespace torch_npu {	11	namespace torch_npu {
17	namespace profiler {	12	namespace profiler {
	@@ -31,10 +26,13 @@ struct MsptiEvent {
31	int64_t streamId = 0;	26	int64_t streamId = 0;
32	};	27	};
33		28
34	-class MsptiActivityProfiler : public libkineto::IActivityProfilerSession {	29	+class MsptiActivityProfiler : public NPUActivityProfiler {
35	public:	30	public:
36	explicit MsptiActivityProfiler(std::set<libkineto::ActivityType> activities) : activities_(std::move(activities)) {}	31	explicit MsptiActivityProfiler(std::set<libkineto::ActivityType> activities) : activities_(std::move(activities)) {}
37		32
		33	+ const char* name() const override {
		34	+ return "mspti";
		35	+ }
38	void start() override;	36	void start() override;
39	void stop() override;	37	void stop() override;
40	std::vector<std::string> errors() override;	38	std::vector<std::string> errors() override;
	@@ -56,27 +54,5 @@ class MsptiActivityProfiler : public libkineto::IActivityProfilerSession {
56	std::vector<std::string> errors_;	54	std::vector<std::string> errors_;
57	};	55	};
58		56
59	-class MsptiActivityProfilerPoc : public libkineto::IActivityProfiler {
60	- public:
61	- const std::string& name() const override {
62	- return name_;
63	- }
64	- const std::set<libkineto::ActivityType>& availableActivities() const override {
65	- return avail_;
66	- }
67	- std::unique_ptr<libkineto::IActivityProfilerSession> configure(
68	- const std::set<libkineto::ActivityType>& activity_types,
69	- const libkineto::Config& config) override;
70	- std::unique_ptr<libkineto::IActivityProfilerSession> configure(
71	- int64_t ts_ms,
72	- int64_t duration_ms,
73	- const std::set<libkineto::ActivityType>& activity_types,
74	- const libkineto::Config& config) override;
75	-
76	- private:
77	- std::string name_{"mspti"};
78	- std::set<libkineto::ActivityType> avail_{libkineto::ActivityType::CONCURRENT_KERNEL};
79	-};
80	-
81	} // namespace profiler	57	} // namespace profiler
82	} // namespace torch_npu	58	} // namespace torch_npu

Atorch_npu/csrc/profiler/npu_activity_profiler.cpp+169-0

@@ -0,0 +1,169 @@
	1	+// Copyright (c) 2026 Huawei Technologies Co., Ltd
	2	+// All rights reserved.
	3	+//
	4	+// Licensed under the BSD 3-Clause License (the "License");
	5	+// you may not use this file except in compliance with the License.
	6	+// You may obtain a copy of the License at
	7	+//
	8	+// https://opensource.org/licenses/BSD-3-Clause
	9	+//
	10	+// Unless required by applicable law or agreed to in writing, software
	11	+// distributed under the License is distributed on an "AS IS" BASIS,
	12	+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	13	+// See the License for the specific language governing permissions and
	14	+// limitations under the License.
	15	+
	16	+#include "torch_npu/csrc/profiler/npu_activity_profiler.h"
	17	+
	18	+#include <set>
	19	+#include <utility>
	20	+
	21	+#include <nlohmann/json.hpp>
	22	+#include <c10/util/Exception.h>
	23	+
	24	+#include "libkineto.h"
	25	+#include "torch_npu/csrc/profiler/aclprof_activity_profiler.h"
	26	+#include "torch_npu/csrc/profiler/mspti_activity_profiler.h"
	27	+#include <torch/csrc/profiler/standalone/privateuse1_profiler.h>
	28	+
	29	+namespace torch_npu {
	30	+namespace profiler {
	31	+namespace {
	32	+
	33	+bool privateUse1Requested(const std::set<libkineto::ActivityType>& types) {
	34	+ using AT = libkineto::ActivityType;
	35	+ return types.count(AT::CONCURRENT_KERNEL) != 0 \|\| types.count(AT::GPU_MEMCPY) != 0 \|\|
	36	+ types.count(AT::GPU_MEMSET) != 0 \|\| types.count(AT::GPU_USER_ANNOTATION) != 0;
	37	+}
	38	+
	39	+NPUProfilerConfig parseExperimentalConfig(const std::string& customConfig) {
	40	+ NPUProfilerConfig result;
	41	+ const auto config = nlohmann::json::parse(customConfig, nullptr, false);
	42	+ if (!config.is_object()) {
	43	+ return result;
	44	+ }
	45	+ const auto backend = config.find("backend");
	46	+ if (backend != config.end() && backend->is_string() && backend->get<std::string>() == "aclprof") {
	47	+ result.backend = "aclprof";
	48	+ }
	49	+ const auto level = config.find("profiler_level");
	50	+ if (level != config.end()) {
	51	+ TORCH_CHECK(
	52	+ level->is_string() &&
	53	+ (level == "Level0" \|\| level == "Level1" \|\| level == "Level2" \|\| level == "Level_none"),
	54	+ "Invalid NPU profiler_level");
	55	+ result.profilerLevel = level->get<std::string>();
	56	+ }
	57	+ const auto exports = config.find("export_type");
	58	+ if (exports != config.end()) {
	59	+ TORCH_CHECK(exports->is_array() && !exports->empty(), "Invalid NPU export_type: expected a nonempty array");
	60	+ std::set<std::string> values;
	61	+ for (const auto& value : *exports) {
	62	+ TORCH_CHECK(value.is_string() && (value == "db" \|\| value == "text"), "Invalid NPU export_type");
	63	+ values.insert(value.get<std::string>());
	64	+ }
	65	+ result.exportTypes.assign(values.begin(), values.end());
	66	+ }
	67	+ return result;
	68	+}
	69	+
	70	+} // namespace
	71	+
	72	+std::unique_ptr<NPUActivityProfiler> NPUActivityProfilerFactory::create(
	73	+ const std::set<libkineto::ActivityType>& activities,
	74	+ const libkineto::Config& config) {
	75	+ if (!privateUse1Requested(activities)) {
	76	+ return nullptr;
	77	+ }
	78	+ auto options = parseExperimentalConfig(config.getCustomConfig());
	79	+ std::unique_ptr<NPUActivityProfiler> profiler;
	80	+ if (options.backend == "aclprof") {
	81	+ profiler = std::make_unique<AclprofActivityProfiler>(activities);
	82	+ } else {
	83	+ profiler = std::make_unique<MsptiActivityProfiler>(activities);
	84	+ }
	85	+ profiler->config_ = std::move(options);
	86	+ return profiler;
	87	+}
	88	+
	89	+void PrivateUse1ProfilerManager::start() {
	90	+ profiler_->start();
	91	+ status_ = profiler_->status();
	92	+}
	93	+
	94	+void PrivateUse1ProfilerManager::stop() {
	95	+ profiler_->stop();
	96	+ status_ = profiler_->status();
	97	+}
	98	+
	99	+std::vector<std::string> PrivateUse1ProfilerManager::errors() {
	100	+ return profiler_->errors();
	101	+}
	102	+
	103	+void PrivateUse1ProfilerManager::processTrace(libkineto::ActivityLogger& logger) {
	104	+ profiler_->processTrace(logger);
	105	+}
	106	+
	107	+void PrivateUse1ProfilerManager::processTrace(
	108	+ libkineto::ActivityLogger& logger,
	109	+ libkineto::getLinkedActivityCallback getLinkedActivity,
	110	+ int64_t startTime,
	111	+ int64_t endTime) {
	112	+ profiler_->processTrace(logger, std::move(getLinkedActivity), startTime, endTime);
	113	+}
	114	+
	115	+std::unique_ptr<libkineto::DeviceInfo> PrivateUse1ProfilerManager::getDeviceInfo() {
	116	+ return profiler_->getDeviceInfo();
	117	+}
	118	+
	119	+std::vector<libkineto::ResourceInfo> PrivateUse1ProfilerManager::getResourceInfos() {
	120	+ return profiler_->getResourceInfos();
	121	+}
	122	+
	123	+std::unique_ptr<libkineto::CpuTraceBuffer> PrivateUse1ProfilerManager::getTraceBuffer() {
	124	+ return profiler_->getTraceBuffer();
	125	+}
	126	+
	127	+void PrivateUse1ProfilerManager::pushCorrelationId(uint64_t id) {
	128	+ profiler_->pushCorrelationId(id);
	129	+}
	130	+
	131	+void PrivateUse1ProfilerManager::popCorrelationId() {
	132	+ profiler_->popCorrelationId();
	133	+}
	134	+
	135	+std::unique_ptr<libkineto::IActivityProfilerSession> NPUActivityProfilerPlugin::configure(
	136	+ const std::set<libkineto::ActivityType>& activityTypes,
	137	+ const libkineto::Config& config) {
	138	+ auto profiler = NPUActivityProfilerFactory::create(activityTypes, config);
	139	+ if (profiler == nullptr) {
	140	+ return nullptr;
	141	+ }
	142	+ return std::make_unique<PrivateUse1ProfilerManager>(std::move(profiler));
	143	+}
	144	+
	145	+std::unique_ptr<libkineto::IActivityProfilerSession> NPUActivityProfilerPlugin::configure(
	146	+ int64_t,
	147	+ int64_t,
	148	+ const std::set<libkineto::ActivityType>& activityTypes,
	149	+ const libkineto::Config& config) {
	150	+ return configure(activityTypes, config);
	151	+}
	152	+
	153	+#ifdef REGISTER_PRIVATEUSE1_PROFILER
	154	+REGISTER_PRIVATEUSE1_PROFILER(NPUActivityProfilerPlugin);
	155	+#else
	156	+namespace {
	157	+struct NPUProfilerAutoRegister {
	158	+ NPUProfilerAutoRegister() {
	159	+ libkineto::api().registerProfilerFactory([]() -> std::unique_ptr<libkineto::IActivityProfiler> {
	160	+ return std::make_unique<NPUActivityProfilerPlugin>();
	161	+ });
	162	+ }
	163	+};
	164	+static NPUProfilerAutoRegister g_npu_profiler_auto_register;
	165	+} // namespace
	166	+#endif
	167	+
	168	+} // namespace profiler
	169	+} // namespace torch_npu

Atorch_npu/csrc/profiler/npu_activity_profiler.h+126-0

@@ -0,0 +1,126 @@
	1	+// Copyright (c) 2026 Huawei Technologies Co., Ltd
	2	+// All rights reserved.
	3	+//
	4	+// Licensed under the BSD 3-Clause License (the "License");
	5	+// you may not use this file except in compliance with the License.
	6	+// You may obtain a copy of the License at
	7	+//
	8	+// https://opensource.org/licenses/BSD-3-Clause
	9	+//
	10	+// Unless required by applicable law or agreed to in writing, software
	11	+// distributed under the License is distributed on an "AS IS" BASIS,
	12	+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	13	+// See the License for the specific language governing permissions and
	14	+// limitations under the License.
	15	+
	16	+#pragma once
	17	+
	18	+#include <memory>
	19	+#include <set>
	20	+#include <string>
	21	+#include <utility>
	22	+#include <vector>
	23	+
	24	+#include "IActivityProfiler.h"
	25	+#include "torch_npu/csrc/core/npu/NPUMacros.h"
	26	+
	27	+namespace torch_npu {
	28	+namespace profiler {
	29	+
	30	+struct NPUProfilerConfig {
	31	+ std::string backend = "mspti";
	32	+ // Empty fields mean unspecified; neither field affects backend selection.
	33	+ std::string profilerLevel;
	34	+ std::vector<std::string> exportTypes;
	35	+};
	36	+
	37	+class NPUActivityProfiler {
	38	+ public:
	39	+ virtual ~NPUActivityProfiler() = default;
	40	+
	41	+ virtual const char* name() const = 0;
	42	+ virtual void start() = 0;
	43	+ virtual void stop() = 0;
	44	+ virtual std::vector<std::string> errors() = 0;
	45	+ virtual void processTrace(libkineto::ActivityLogger& logger) = 0;
	46	+ virtual void processTrace(
	47	+ libkineto::ActivityLogger& logger,
	48	+ libkineto::getLinkedActivityCallback getLinkedActivity,
	49	+ int64_t startTime,
	50	+ int64_t endTime) = 0;
	51	+ virtual std::unique_ptr<libkineto::DeviceInfo> getDeviceInfo() = 0;
	52	+ virtual std::vector<libkineto::ResourceInfo> getResourceInfos() = 0;
	53	+ virtual std::unique_ptr<libkineto::CpuTraceBuffer> getTraceBuffer() = 0;
	54	+ virtual void pushCorrelationId(uint64_t) {}
	55	+ virtual void popCorrelationId() {}
	56	+
	57	+ libkineto::TraceStatus status() const {
	58	+ return status_;
	59	+ }
	60	+
	61	+ const NPUProfilerConfig& experimentalConfig() const {
	62	+ return config_;
	63	+ }
	64	+
	65	+ protected:
	66	+ libkineto::TraceStatus status_ = libkineto::TraceStatus::READY;
	67	+
	68	+ private:
	69	+ friend class NPUActivityProfilerFactory;
	70	+ NPUProfilerConfig config_;
	71	+};
	72	+
	73	+class TORCH_NPU_API NPUActivityProfilerFactory {
	74	+ public:
	75	+ static std::unique_ptr<NPUActivityProfiler> create(
	76	+ const std::set<libkineto::ActivityType>& activities,
	77	+ const libkineto::Config& config);
	78	+};
	79	+
	80	+class TORCH_NPU_API PrivateUse1ProfilerManager : public libkineto::IActivityProfilerSession {
	81	+ public:
	82	+ explicit PrivateUse1ProfilerManager(std::unique_ptr<NPUActivityProfiler> profiler) : profiler_(std::move(profiler)) {}
	83	+
	84	+ void start() override;
	85	+ void stop() override;
	86	+ std::vector<std::string> errors() override;
	87	+ void processTrace(libkineto::ActivityLogger& logger) override;
	88	+ void processTrace(
	89	+ libkineto::ActivityLogger& logger,
	90	+ libkineto::getLinkedActivityCallback getLinkedActivity,
	91	+ int64_t startTime,
	92	+ int64_t endTime) override;
	93	+ std::unique_ptr<libkineto::DeviceInfo> getDeviceInfo() override;
	94	+ std::vector<libkineto::ResourceInfo> getResourceInfos() override;
	95	+ std::unique_ptr<libkineto::CpuTraceBuffer> getTraceBuffer() override;
	96	+ void pushCorrelationId(uint64_t id) override;
	97	+ void popCorrelationId() override;
	98	+
	99	+ private:
	100	+ std::unique_ptr<NPUActivityProfiler> profiler_;
	101	+};
	102	+
	103	+class TORCH_NPU_API NPUActivityProfilerPlugin : public libkineto::IActivityProfiler {
	104	+ public:
	105	+ const std::string& name() const override {
	106	+ return name_;
	107	+ }
	108	+ const std::set<libkineto::ActivityType>& availableActivities() const override {
	109	+ return avail_;
	110	+ }
	111	+ std::unique_ptr<libkineto::IActivityProfilerSession> configure(
	112	+ const std::set<libkineto::ActivityType>& activityTypes,
	113	+ const libkineto::Config& config) override;
	114	+ std::unique_ptr<libkineto::IActivityProfilerSession> configure(
	115	+ int64_t tsMs,
	116	+ int64_t durationMs,
	117	+ const std::set<libkineto::ActivityType>& activityTypes,
	118	+ const libkineto::Config& config) override;
	119	+
	120	+ private:
	121	+ std::string name_{"mspti"};
	122	+ std::set<libkineto::ActivityType> avail_{libkineto::ActivityType::CONCURRENT_KERNEL};
	123	+};
	124	+
	125	+} // namespace profiler
	126	+} // namespace torch_npu

Mtorch_npu/csrc/profiler/npu_profiler.cpp+11-0

	@@ -8,6 +8,7 @@
8	#include "torch_npu/csrc/core/npu/npu_log.h"	8	#include "torch_npu/csrc/core/npu/npu_log.h"
9	#include "torch_npu/csrc/core/npu/NPUException.h"	9	#include "torch_npu/csrc/core/npu/NPUException.h"
10	#include "torch_npu/csrc/core/npu/interface/AclInterface.h"	10	#include "torch_npu/csrc/core/npu/interface/AclInterface.h"
		11	+#include "torch_npu/csrc/profiler/aclprof_activity_profiler.h"
11	#include "torch_npu/csrc/profiler/npu_profiler.h"	12	#include "torch_npu/csrc/profiler/npu_profiler.h"
12		13
13	#include "torch_npu/csrc/toolkit/profiler/common/utils.h"	14	#include "torch_npu/csrc/toolkit/profiler/common/utils.h"
	@@ -386,6 +387,16 @@ void finalizeNpuProfiler() {
386	}	387	}
387		388
388	void reportMarkDataToNpuProfiler(uint32_t category, const std::string& msg, uint64_t correlation_id) {	389	void reportMarkDataToNpuProfiler(uint32_t category, const std::string& msg, uint64_t correlation_id) {
		390	+ recordAclprofTaskQueueEvent(
		391	+ category,
		392	+ correlation_id,
		393	+ static_cast<int64_t>(Utils::GetClockMonotonicRawNs()),
		394	+ static_cast<int64_t>(Utils::GetPid()),
		395	+ static_cast<int64_t>(Utils::GetTid()),
		396	+ msg);
		397	+ if (!profDataReportEnable().load(std::memory_order_relaxed)) {
		398	+ return;
		399	+ }
389	ProfilerMgr::GetInstance()->UploadWithLock(std::make_unique<torch_npu::toolkit::profiler::OpMarkData>(	400	ProfilerMgr::GetInstance()->UploadWithLock(std::make_unique<torch_npu::toolkit::profiler::OpMarkData>(
390	static_cast<int64_t>(Utils::GetClockTime()), category, correlation_id, Utils::GetTid(), Utils::GetPid(), msg));	401	static_cast<int64_t>(Utils::GetClockTime()), category, correlation_id, Utils::GetTid(), Utils::GetPid(), msg));
391	}	402	}

Mtorch_npu/profiler/__init__.py+3-1

	@@ -9,8 +9,10 @@ from .scheduler import Schedule as schedule
9	from .scheduler import ProfilerAction	9	from .scheduler import ProfilerAction
10	from .experimental_config import _ExperimentalConfig, supported_profiler_level, supported_ai_core_metrics, \	10	from .experimental_config import _ExperimentalConfig, supported_profiler_level, supported_ai_core_metrics, \
11	supported_export_type, ProfilerLevel, AiCMetrics, ExportType, HostSystem	11	supported_export_type, ProfilerLevel, AiCMetrics, ExportType, HostSystem
		12	+from ._kineto_config import _NPUExperimentalConfig
12	from ._non_intrusive_profile import _NonIntrusiveProfile	13	from ._non_intrusive_profile import _NonIntrusiveProfile
13		14
14	__all__ = ["profile", "ProfilerActivity", "supported_activities", "tensorboard_trace_handler", "schedule",	15	__all__ = ["profile", "ProfilerActivity", "supported_activities", "tensorboard_trace_handler", "schedule",
15	- "ProfilerAction", "_ExperimentalConfig", "supported_profiler_level", "supported_ai_core_metrics",	16	+ "ProfilerAction", "_ExperimentalConfig", "supported_profiler_level",
		17	+ "supported_ai_core_metrics",
16	"supported_export_type", "ProfilerLevel", "AiCMetrics", "ExportType", "HostSystem"]	18	"supported_export_type", "ProfilerLevel", "AiCMetrics", "ExportType", "HostSystem"]

Atorch_npu/profiler/_kineto_config.py+56-0

@@ -0,0 +1,56 @@
	1	+# Copyright (c) 2026 Huawei Technologies Co., Ltd
	2	+# All rights reserved.
	3	+#
	4	+# Licensed under the BSD 3-Clause License (the "License");
	5	+# you may not use this file except in compliance with the License.
	6	+# You may obtain a copy of the License at
	7	+#
	8	+# https://opensource.org/licenses/BSD-3-Clause
	9	+#
	10	+# Unless required by applicable law or agreed to in writing, software
	11	+# distributed under the License is distributed on an "AS IS" BASIS,
	12	+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
	13	+# See the License for the specific language governing permissions and
	14	+# limitations under the License.
	15	+
	16	+import json
	17	+
	18	+from torch._C._profiler import _ExperimentalConfig as ExperimentalConfig
	19	+
	20	+from .experimental_config import supported_export_type, supported_profiler_level
	21	+
	22	+
	23	+class _NPUExperimentalConfig(ExperimentalConfig):
	24	+ """Ascend-specific options for ``torch.profiler.profile``.
	25	+
	26	+ Only ``backend="aclprof"`` selects the ACLprof stub. ``profiler_level``
	27	+ and ``export_type`` are retained for future implementations; they do not
	28	+ enable collection or export. ``None`` leaves these options unspecified.
	29	+ Exports accept an ExportType string or a nonempty list of those strings.
	30	+ Invalid levels/exports raise ValueError; invalid backends fall back to MSPTI.
	31	+ """
	32	+
	33	+ def __init__(self, , backend: str = "mspti", profiler_level=None, export_type=None, *kwargs):
	34	+ config = {"backend": backend}
	35	+ if profiler_level is not None:
	36	+ if not isinstance(profiler_level, str) or profiler_level not in supported_profiler_level():
	37	+ raise ValueError("profiler_level must be a supported ProfilerLevel value")
	38	+ config["profiler_level"] = profiler_level
	39	+ if export_type is not None:
	40	+ exports = [export_type] if isinstance(export_type, str) else export_type
	41	+ if not isinstance(exports, list) or not exports or any(
	42	+ not isinstance(value, str) or value not in supported_export_type() for value in exports
	43	+ ):
	44	+ raise ValueError("export_type must be an ExportType value or a nonempty list of them")
	45	+ config["export_type"] = sorted(set(exports))
	46	+ super().__init__(
	47	+ custom_profiler_config=json.dumps(config, separators=(",", ":")),
	48	+ **kwargs,
	49	+ )
	50	+
	51	+ def __str__(self):
	52	+ # Older PyTorch builds expose this field only through their pickle state.
	53	+ custom_config = getattr(self, "custom_profiler_config", None)
	54	+ if custom_config is not None:
	55	+ return custom_config
	56	+ return self.__getstate__()[11]