add test for TORCH_ACL_INIT_CONFIG

已合并

add test for TORCH_ACL_INIT_CONFIG_PATH #39953

huangyunlong创建于 7月3日从huangyunlong2022/pytorch： 2.9aclj 合入到 Ascend/pytorch： v2.9.0

add test for TORCH_ACL_INIT_CONFIG_PATH #39953

已合并

huangyunlong创建于 7月3日

从huangyunlong2022/pytorch 2.9aclj合入到Ascend/pytorch v2.9.0

共 2 个文件变更+145-9

Mtest/npu/test_option.py+133-3

	@@ -1,13 +1,14 @@
		1	+import os
1	import sys	2	import sys
2	import subprocess	3	import subprocess
		4	+import unittest
		5	+from functools import wraps
3		6
4	import torch	7	import torch
5	import torch_npu	8	import torch_npu
6		9
7	-import torch_npu.npu.utils as utils
8	-
9	from torch_npu.testing.testcase import TestCase, run_tests	10	from torch_npu.testing.testcase import TestCase, run_tests
10	-from torch_npu.testing.common_utils import SupportedDevices	11	+from torch_npu.testing.common_utils import SupportedDevices, SkipIfNotGteCANNVersion
11		12
12		13
13	class TestOption(TestCase):	14	class TestOption(TestCase):
	@@ -111,5 +112,134 @@ class TestAclOpInitMode(TestCase):
111	self.assertIn(self.INVALID_VALUE_WARN, stderr)	112	self.assertIn(self.INVALID_VALUE_WARN, stderr)
112		113
113		114
		115	+def _skipIfLazy(fn):
		116	+ @wraps(fn)
		117	+ def wrapper(slf, args, *kwargs):
		118	+ if torch_npu.npu.utils._is_gte_cann_version('8.3.RC1'):
		119	+ raise unittest.SkipTest(
		120	+ "Test only for non-lazy set_device mode (CANN < 8.3.RC1)")
		121	+ return fn(slf, args, *kwargs)
		122	+ return wrapper
		123	+
		124	+
		125	+class TestAclInitConfigPath(TestCase):
		126	+
		127	+ LACKS_DEFAULT_DEVICE_MSG = "lacks 'defaultDevice'"
		128	+ HAS_DEFAULT_DEVICE_MSG = "contains 'defaultDevice'"
		129	+ PARSE_FAILED_MSG = "Failed to parse user acl json"
		130	+ INVALID_PATH_MSG = "is invalid"
		131	+ OPEN_FAILED_MSG = "Failed to open user acl json"
		132	+
		133	+ def _make_temp_json(self, content):
		134	+ import tempfile
		135	+ tmp = tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False)
		136	+ tmp.write(content)
		137	+ tmp.close()
		138	+ self.addCleanup(os.unlink, tmp.name)
		139	+ return tmp.name
		140	+
		141	+ def _run_with_env(self, json_path):
		142	+ test_script = (
		143	+ f"import os; "
		144	+ f"os.environ['TORCH_ACL_INIT_CONFIG_PATH']='{json_path}'; "
		145	+ f"import torch; import torch_npu; torch_npu.npu.set_device(0)"
		146	+ )
		147	+ result = subprocess.run(
		148	+ [sys.executable, '-c', test_script],
		149	+ capture_output=True, text=True
		150	+ )
		151	+ return result
		152	+
		153	+ # ---- 设备无关：路径无效/解析失败 ----
		154	+
		155	+ def test_invalid_json(self):
		156	+ """非法 JSON 格式应抛 RuntimeError"""
		157	+ json_path = self._make_temp_json('not a json{{{')
		158	+ ret = self._run_with_env(json_path)
		159	+ self.assertNotEqual(ret.returncode, 0)
		160	+ self.assertIn(self.PARSE_FAILED_MSG, ret.stderr,
		161	+ "Invalid JSON should raise RuntimeError")
		162	+
		163	+ def test_nonexistent_path(self):
		164	+ """不存在的路径应抛 RuntimeError"""
		165	+ json_path = '/tmp/nonexistent_acl_config_for_test.json'
		166	+ ret = self._run_with_env(json_path)
		167	+ self.assertNotEqual(ret.returncode, 0)
		168	+ self.assertIn(self.INVALID_PATH_MSG, ret.stderr,
		169	+ "Non-existent path should raise RuntimeError")
		170	+
		171	+ def test_empty_json_file(self):
		172	+ """空 JSON 文件应抛 RuntimeError"""
		173	+ json_path = self._make_temp_json('')
		174	+ ret = self._run_with_env(json_path)
		175	+ self.assertNotEqual(ret.returncode, 0)
		176	+ self.assertIn(self.PARSE_FAILED_MSG, ret.stderr,
		177	+ "Empty JSON file should raise RuntimeError")
		178	+
		179	+ # ---- non-lazy 模式（CANN < 8.3.RC1） ----
		180	+
		181	+ @_skipIfLazy
		182	+ def test_valid_json_non_lazy(self):
		183	+ """non-lazy 模式：不含 defaultDevice 的用户 JSON 正常使用"""
		184	+ json_path = self._make_temp_json('{"dump":{"dump_scene":"lite_exception"}}')
		185	+ ret = self._run_with_env(json_path)
		186	+ self.assertEqual(ret.returncode, 0,
		187	+ f"Valid JSON in non-lazy should succeed, got: {ret.stderr}")
		188	+
		189	+ @_skipIfLazy
		190	+ def test_valid_json_with_default_device_in_non_lazy(self):
		191	+ """non-lazy 模式：含 defaultDevice 应抛 RuntimeError"""
		192	+ json_path = self._make_temp_json(
		193	+ '{"dump":{"dump_scene":"lite_exception"},"defaultDevice":{"default_device":"0"}}'
		194	+ )
		195	+ ret = self._run_with_env(json_path)
		196	+ self.assertNotEqual(ret.returncode, 0)
		197	+ self.assertIn(self.HAS_DEFAULT_DEVICE_MSG, ret.stderr,
		198	+ "Non-lazy mode with defaultDevice should raise RuntimeError")
		199	+
		200	+ # ---- lazy 模式（CANN >= 8.3.RC1） ----
		201	+
		202	+ @SkipIfNotGteCANNVersion('8.3.RC1')
		203	+ def test_valid_json_with_default_device_in_lazy(self):
		204	+ """lazy 模式：含合法 defaultDevice 正常使用"""
		205	+ json_path = self._make_temp_json(
		206	+ '{"dump":{"dump_scene":"lite_exception"},"defaultDevice":{"default_device":"0"}}'
		207	+ )
		208	+ ret = self._run_with_env(json_path)
		209	+ self.assertEqual(ret.returncode, 0,
		210	+ f"Valid JSON in lazy should succeed, got: {ret.stderr}")
		211	+
		212	+ @SkipIfNotGteCANNVersion('8.3.RC1')
		213	+ def test_json_without_default_device_in_lazy(self):
		214	+ """lazy 模式：不含 defaultDevice 应抛 RuntimeError"""
		215	+ json_path = self._make_temp_json('{"dump":{"dump_scene":"lite_exception"}}')
		216	+ ret = self._run_with_env(json_path)
		217	+ self.assertNotEqual(ret.returncode, 0)
		218	+ self.assertIn(self.LACKS_DEFAULT_DEVICE_MSG, ret.stderr,
		219	+ "Lazy mode without defaultDevice should raise RuntimeError")
		220	+
		221	+ @SkipIfNotGteCANNVersion('8.3.RC1')
		222	+ def test_json_wrong_default_device_value_in_lazy(self):
		223	+ """lazy 模式：default_device 值不为 '0' 应抛 RuntimeError"""
		224	+ json_path = self._make_temp_json(
		225	+ '{"defaultDevice":{"default_device":"1"}}'
		226	+ )
		227	+ ret = self._run_with_env(json_path)
		228	+ self.assertNotEqual(ret.returncode, 0)
		229	+ self.assertIn('requires', ret.stderr,
		230	+ "default_device='1' should raise RuntimeError")
		231	+
		232	+ @SkipIfNotGteCANNVersion('8.3.RC1')
		233	+ def test_json_default_device_wrong_type(self):
		234	+ """lazy 模式：defaultDevice 非 object 应抛 RuntimeError"""
		235	+ json_path = self._make_temp_json(
		236	+ '{"defaultDevice":"not_an_object","dump":{"dump_scene":"lite_exception"}}'
		237	+ )
		238	+ ret = self._run_with_env(json_path)
		239	+ self.assertNotEqual(ret.returncode, 0)
		240	+ self.assertIn(self.LACKS_DEFAULT_DEVICE_MSG, ret.stderr,
		241	+ "defaultDevice as string should raise RuntimeError")
		242	+
		243	+
114	if __name__ == "__main__":	244	if __name__ == "__main__":
115	run_tests()	245	run_tests()

Mtorch_npu/csrc/core/npu/sys_ctrl/npu_sys_ctrl.cpp+12-6

	@@ -116,13 +116,15 @@ std::string GetAclConfigJsonPath()
116	std::string json_path = std::string(acl_init_path);	116	std::string json_path = std::string(acl_init_path);
117	std::string json_path_str = torch_npu::toolkit::profiler::Utils::RealPath(json_path);	117	std::string json_path_str = torch_npu::toolkit::profiler::Utils::RealPath(json_path);
118	if (json_path_str.empty()) {	118	if (json_path_str.empty()) {
119	- TORCH_CHECK(false, "TORCH_ACL_INIT_CONFIG_PATH ", acl_init_path,	119	+ TORCH_CHECK(false, "TORCH_ACL_INIT_CONFIG_PATH ", acl_init_path, " is invalid. ",
120	- " is invalid.", PTA_ERROR(ErrCode::UNAVAIL));	120	+ "Please set the environment variable 'TORCH_ACL_INIT_CONFIG_PATH' to a valid JSON config file path.",
		121	+ PTA_ERROR(ErrCode::UNAVAIL));
121	}	122	}
122		123
123	std::ifstream config_file(json_path_str);	124	std::ifstream config_file(json_path_str);
124	if (!config_file.is_open()) {	125	if (!config_file.is_open()) {
125	TORCH_CHECK(false, "Failed to open user acl json ", json_path_str,	126	TORCH_CHECK(false, "Failed to open user acl json ", json_path_str,
		127	+ ". Please ensure the file exists and has read permission.",
126	PTA_ERROR(ErrCode::UNAVAIL));	128	PTA_ERROR(ErrCode::UNAVAIL));
127	}	129	}
128		130
	@@ -132,13 +134,15 @@ std::string GetAclConfigJsonPath()
132	config_file >> config;	134	config_file >> config;
133	} catch (const std::exception& e) {	135	} catch (const std::exception& e) {
134	TORCH_CHECK(false, "Failed to parse user acl json ", json_path_str,	136	TORCH_CHECK(false, "Failed to parse user acl json ", json_path_str,
135	- " : ", e.what(), PTA_ERROR(ErrCode::UNAVAIL));	137	+ ". Please check that the file contains valid JSON syntax. Error: ", e.what(),
		138	+ PTA_ERROR(ErrCode::UNAVAIL));
136	}	139	}
137		140
138	if (c10_npu::is_lazy_set_device()) {	141	if (c10_npu::is_lazy_set_device()) {
139	if (!config.contains("defaultDevice") \|\| !config["defaultDevice"].is_object()) {	142	if (!config.contains("defaultDevice") \|\| !config["defaultDevice"].is_object()) {
140	TORCH_CHECK(false, "User acl json ", json_path_str,	143	TORCH_CHECK(false, "User acl json ", json_path_str,
141	- " lacks 'defaultDevice' object, required for lazy set_device mode.",	144	+ " lacks 'defaultDevice' object, required for lazy set_device mode. ",
		145	+ "Please add: \"defaultDevice\": {\"default_device\": \"0\"}",
142	PTA_ERROR(ErrCode::VALUE));	146	PTA_ERROR(ErrCode::VALUE));
143	}	147	}
144	const auto& default_dev = config["defaultDevice"];	148	const auto& default_dev = config["defaultDevice"];
	@@ -146,13 +150,15 @@ std::string GetAclConfigJsonPath()
146	!default_dev["default_device"].is_string() \|\|	150	!default_dev["default_device"].is_string() \|\|
147	default_dev["default_device"].get<std::string>() != "0") {	151	default_dev["default_device"].get<std::string>() != "0") {
148	TORCH_CHECK(false, "User acl json ", json_path_str,	152	TORCH_CHECK(false, "User acl json ", json_path_str,
149	- " requires 'defaultDevice.default_device' to be \"0\" in lazy set_device mode.",	153	+ " requires 'defaultDevice.default_device' to be \"0\" in lazy set_device mode. ",
		154	+ "Please set: \"defaultDevice\": {\"default_device\": \"0\"}",
150	PTA_ERROR(ErrCode::VALUE));	155	PTA_ERROR(ErrCode::VALUE));
151	}	156	}
152	} else {	157	} else {
153	if (config.contains("defaultDevice")) {	158	if (config.contains("defaultDevice")) {
154	TORCH_CHECK(false, "User acl json ", json_path_str,	159	TORCH_CHECK(false, "User acl json ", json_path_str,
155	- " contains 'defaultDevice' which is not expected in non-lazy set_device mode.",	160	+ " contains 'defaultDevice' which is not expected in non-lazy set_device mode. ",
		161	+ "Please remove the 'defaultDevice' field from your JSON config.",
156	PTA_ERROR(ErrCode::VALUE));	162	PTA_ERROR(ErrCode::VALUE));
157	}	163	}
158	}	164	}