[v2.7.1][Fix] Fix static check errors detected by TABS-pytorch-AtomGit

M.lintrunner.toml+36-43

	@@ -464,49 +464,42 @@ is_formatter = true
464	# '@{{PATHSFILE}}'	464	# '@{{PATHSFILE}}'
465	# ]	465	# ]
466		466
467	-# [[linter]]	467	+[[linter]]
468	-# code = 'TABS'	468	+code = 'TABS'
469	-# include_patterns = ['**']	469	+include_patterns = ['**']
470	-# exclude_patterns = [	470	+exclude_patterns = [
471	-# '*/.svg',	471	+ '*/.svg',
472	-# '*/Makefile',	472	+ '*/Makefile',
473	-# '/contrib/',	473	+ '/contrib/',
474	-# 'third_party/**',	474	+ 'third_party/**',
475	-# '**/.gitattributes',	475	+ '**/.gitattributes',
476	-# '**/.gitmodules',	476	+ '**/.gitmodules',
477	-# 'fb/**',	477	+ 'fb/**',
478	-# '/fb/',	478	+ '/fb/',
479	-# 'aten/src/ATen/native/vulkan/api/vk_mem_alloc.h',	479	+ 'aten/src/ATen/native/vulkan/api/vk_mem_alloc.h',
480	-# 'test/cpp/jit/upgrader_models/*.ptl',	480	+ 'test/cpp/jit/upgrader_models/*.ptl',
481	-# 'test/cpp/jit/upgrader_models/*.ptl.ff',	481	+ 'test/cpp/jit/upgrader_models/*.ptl.ff',
482	-# '.ci/docker/common/install_rocm_drm.sh',	482	+ '.ci/docker/common/install_rocm_drm.sh',
483	-# '.lintrunner.toml',	483	+ '.lintrunner.toml',
484	-# '*/.patch',	484	+ '*/.patch',
485	-# # NPUGraph logs files	485	+ # NPUGraph logs files
486	-# 'torch_npu/_logging/_internal.py',	486	+ 'torch_npu/_logging/_internal.py',
487	-# 'torch_npu/csrc/core/npu/NPUGraph.cpp',	487	+ '*/.md',
488	-# 'torch_npu/csrc/core/npu/NPUGraph.h',	488	+]
489	-# 'torch_npu/csrc/npu/Graph.cpp',	489	+command = [
490	-# 'torch_npu/csrc/core/npu/NPUCachingAllocator.cpp',	490	+ 'python3',
491	-# 'torch_npu/csrc/core/npu/NPUWorkspaceAllocator.cpp',	491	+ 'tools/linter/adapters/grep_linter.py',
492	-# 'torch_npu/npu/_graph_tree.py',	492	+ # @lint-ignore TXT2
493	-# 'torch_npu/npu/graphs.py',	493	+ '--pattern= ',
494	-# 'torch_npu/utils/_graph_tree.py',	494	+ '--linter-name=TABS',
495	-# ]	495	+ '--error-name=saw some tabs',
496	-# command = [	496	+ '--replace-pattern=s/\t/ /',
497	-# 'python3',	497	+ """--error-description=\
498	-# 'tools/linter/adapters/grep_linter.py',	498	+ This line has tabs; please replace them with spaces.\
499	-# # @lint-ignore TXT2	499	+ """,
500	-# '--pattern= ',	500	+ '--',
501	-# '--linter-name=TABS',	501	+ '@{{PATHSFILE}}'
502	-# '--error-name=saw some tabs',	502	+]
503	-# '--replace-pattern=s/\t/ /',
504	-# """--error-description=\
505	-# This line has tabs; please replace them with spaces.\
506	-# """,
507	-# '--',
508	-# '@{{PATHSFILE}}'
509	-# ]
510		503
511	# [[linter]]	504	# [[linter]]
512	# code = 'C10_UNUSED'	505	# code = 'C10_UNUSED'

Mtest/npu/test_compatibility.py+2-2

	@@ -63,8 +63,8 @@ def is_not_compatibility_for_cpp_api(base_signature: str, file: str):
63	subs += line	63	subs += line
64	if ")" in line and "(" not in line and start_concat:	64	if ")" in line and "(" not in line and start_concat:
65	start_concat = False	65	start_concat = False
66	- subs = re.sub("(?<=\\()[ \n]+", "", subs)	66	+ subs = re.sub("(?<=\\()[ \n]+", "", subs)
67	- subs = re.sub("(?<=,)[ \n]+", " ", subs)	67	+ subs = re.sub("(?<=,)[ \n]+", " ", subs)
68	line = subs	68	line = subs
69	subs = ""	69	subs = ""
70	if not start_concat:	70	if not start_concat:

Mtest/test_npu_sanitizer.py+2-2

	@@ -160,7 +160,7 @@ class TestArgumentHandler(TestCase):
160		160
161	self.assertEqual({out.data_ptr()}, argument_handler.dataptrs_written)	161	self.assertEqual({out.data_ptr()}, argument_handler.dataptrs_written)
162	self.assertEqual({out.data_ptr()}, argument_handler.outputs)	162	self.assertEqual({out.data_ptr()}, argument_handler.outputs)
163	-	163	+
164	def test_equal_reads_inputs_but_no_tensor_output_written(self):	164	def test_equal_reads_inputs_but_no_tensor_output_written(self):
165	"""Data-reading op with non-tensor output should not record tensor writes."""	165	"""Data-reading op with non-tensor output should not record tensor writes."""
166	equal_func = torch.ops.aten.equal.default	166	equal_func = torch.ops.aten.equal.default
	@@ -177,7 +177,7 @@ class TestArgumentHandler(TestCase):
177	self.assertEqual(set(), argument_handler.outputs)	177	self.assertEqual(set(), argument_handler.outputs)
178	self.assertTrue(isinstance(out, bool))	178	self.assertTrue(isinstance(out, bool))
179		179
180	-	180	+
181	class TestRecordStreamHandler(TestCase):	181	class TestRecordStreamHandler(TestCase):
182	def test_erase_stream_removes_recorded_stream(self):	182	def test_erase_stream_removes_recorded_stream(self):
183	"""Communication eraseStream should clear the matching recorded stream."""	183	"""Communication eraseStream should clear the matching recorded stream."""

Mtest/torch_npu_schema.json+2-2

	@@ -576,8 +576,8 @@
576	"signature": "(group=None, rebuild_link=True)"	576	"signature": "(group=None, rebuild_link=True)"
577	},	577	},
578	"torch_npu.distributed.tensor.experimental.context_parallel": {	578	"torch_npu.distributed.tensor.experimental.context_parallel": {
579	- "signature": "(mesh: torch.distributed.device_mesh.DeviceMesh, *, buffers: Optional[list[torch.Tensor]] = None, buffer_seq_dims: Optional[list[int]] = None, no_restore_buffers: Optional[set[torch.Tensor]] = None) -> collections.abc.Generator[None, None, None]"	579	+ "signature": "(mesh: torch.distributed.device_mesh.DeviceMesh, *, buffers: Optional[list[torch.Tensor]] = None, buffer_seq_dims: Optional[list[int]] = None, no_restore_buffers: Optional[set[torch.Tensor]] = None) -> collections.abc.Generator[None, None, None]"
580	- },	580	+ },
581	"torch_npu.distributed.rpc.options.NPUTensorPipeRpcBackendOptions": {	581	"torch_npu.distributed.rpc.options.NPUTensorPipeRpcBackendOptions": {
582	"signature": "(*, num_worker_threads: int = 16, rpc_timeout: float = 60.0, init_method: str = 'env://', device_maps: Optional[Dict[str, Dict[Union[int, str, torch.device], Union[int, str, torch.device]]]] = None, devices: Optional[List[Union[int, str, torch.device]]] = None, _transports: Optional[List] = None, _channels: Optional[List] = None)"	582	"signature": "(*, num_worker_threads: int = 16, rpc_timeout: float = 60.0, init_method: str = 'env://', device_maps: Optional[Dict[str, Dict[Union[int, str, torch.device], Union[int, str, torch.device]]]] = None, devices: Optional[List[Union[int, str, torch.device]]] = None, _transports: Optional[List] = None, _channels: Optional[List] = None)"
583	},	583	},

Mtorch_npu/csrc/core/npu/NPUGraph.cpp+2-2

	@@ -25,8 +25,8 @@ namespace {
25	void apply_cache_op_info(aclrtStream stream, bool enabled)	25	void apply_cache_op_info(aclrtStream stream, bool enabled)
26	{	26	{
27	if (!IsGteCANNVersion("8.5.0", "CANN")) {	27	if (!IsGteCANNVersion("8.5.0", "CANN")) {
28	- return;	28	+ return;
29	- }	29	+ }
30	aclrtStreamAttrValue val;	30	aclrtStreamAttrValue val;
31	val.cacheOpInfoSwitch = static_cast<uint32_t>(enabled ? 1u : 0u);	31	val.cacheOpInfoSwitch = static_cast<uint32_t>(enabled ? 1u : 0u);
32	int32_t ret = c10_npu::acl::AclrtSetStreamAttribute(stream, aclrtStreamAttr::ACL_STREAM_ATTR_CACHE_OP_INFO,	32	int32_t ret = c10_npu::acl::AclrtSetStreamAttribute(stream, aclrtStreamAttr::ACL_STREAM_ATTR_CACHE_OP_INFO,

Mtorch_npu/csrc/core/npu/NPUGraph.h+1-1

	@@ -83,7 +83,7 @@ struct TORCH_NPU_API NPUGraph {
83	void capture_begin(	83	void capture_begin(
84	MempoolId_t pool = {0, 0},	84	MempoolId_t pool = {0, 0},
85	aclmdlRICaptureMode capture_mode = aclmdlRICaptureMode::ACL_MODEL_RI_CAPTURE_MODE_GLOBAL,	85	aclmdlRICaptureMode capture_mode = aclmdlRICaptureMode::ACL_MODEL_RI_CAPTURE_MODE_GLOBAL,
86	- bool report_shape = true);	86	+ bool report_shape = true);
87	void capture_end();	87	void capture_end();
88	void replay();	88	void replay();
89	void reset();	89	void reset();

Mtorch_npu/csrc/distributed/ParallelTcpServer.hpp+2-2

	@@ -102,9 +102,9 @@ using ServerProcFn = std::function<StoreMessage(int fd, const StoreMessage &req)
102	class ParallelTcpServer {	102	class ParallelTcpServer {
103	public:	103	public:
104	explicit ParallelTcpServer(uint32_t threadNum, const std::string host, uint16_t port, uint32_t listenThreadNum,	104	explicit ParallelTcpServer(uint32_t threadNum, const std::string host, uint16_t port, uint32_t listenThreadNum,
105	- ServerProcFn process) noexcept;	105	+ ServerProcFn process) noexcept;
106	explicit ParallelTcpServer(uint32_t threadNum, const std::string localSocketPath, uint32_t listenThreadNum,	106	explicit ParallelTcpServer(uint32_t threadNum, const std::string localSocketPath, uint32_t listenThreadNum,
107	- ServerProcFn process) noexcept;	107	+ ServerProcFn process) noexcept;
108		108
109	int Start() noexcept;	109	int Start() noexcept;
110	void Stop() noexcept;	110	void Stop() noexcept;

Mtorch_npu/csrc/distributed/ProcessGroupHCCL.cpp+29-29

	@@ -4741,15 +4741,15 @@ c10::intrusive_ptr<c10d::Work> ProcessGroupHCCL::batch_isend_irecv_inner(
4741	tensors_tmp,	4741	tensors_tmp,
4742	[&](at::Tensor& input, at::Tensor& output, HcclComm comm, c10_npu::NPUStream& stream, std::shared_ptr<bool> is_dispatched) {	4742	[&](at::Tensor& input, at::Tensor& output, HcclComm comm, c10_npu::NPUStream& stream, std::shared_ptr<bool> is_dispatched) {
4743	RECORD_FUNCTION("HcclBatchSendRecv", std::vector<c10::IValue>({input}));	4743	RECORD_FUNCTION("HcclBatchSendRecv", std::vector<c10::IValue>({input}));
4744	- auto itemNum = static_cast<uint32_t>(op_type.size());	4744	+ auto itemNum = static_cast<uint32_t>(op_type.size());
4745	- std::vector<void *> tensor_ptr_list;	4745	+ std::vector<void *> tensor_ptr_list;
4746	- std::vector<uint64_t> numel_list;	4746	+ std::vector<uint64_t> numel_list;
4747	- std::vector<HcclDataType> type_list;	4747	+ std::vector<HcclDataType> type_list;
4748	- for (size_t i = 0; i < op_type.size(); ++i) {	4748	+ for (size_t i = 0; i < op_type.size(); ++i) {
4749	- tensor_ptr_list.push_back(tensors[i].data_ptr());	4749	+ tensor_ptr_list.push_back(tensors[i].data_ptr());
4750	- numel_list.push_back(getNumelForHCCL(tensors[i]));	4750	+ numel_list.push_back(getNumelForHCCL(tensors[i]));
4751	- type_list.push_back(getHcclDataType(tensors[i].scalar_type()));	4751	+ type_list.push_back(getHcclDataType(tensors[i].scalar_type()));
4752	- }	4752	+ }
4753		4753
4754	std::vector<uint32_t> remote_rank_list_cast;	4754	std::vector<uint32_t> remote_rank_list_cast;
4755	remote_rank_list_cast.reserve(remote_rank_list.size());	4755	remote_rank_list_cast.reserve(remote_rank_list.size());
	@@ -4761,36 +4761,36 @@ c10::intrusive_ptr<c10d::Work> ProcessGroupHCCL::batch_isend_irecv_inner(
4761	}	4761	}
4762	remote_rank_list_cast.push_back(static_cast<uint32_t>(remote_rank_list[i]));	4762	remote_rank_list_cast.push_back(static_cast<uint32_t>(remote_rank_list[i]));
4763	}	4763	}
4764	- auto hccl_call = [tensor_ptr_list, numel_list, type_list, remote_rank_list_cast, op_type, itemNum, comm, stream, is_dispatched]() -> int {	4764	+ auto hccl_call = [tensor_ptr_list, numel_list, type_list, remote_rank_list_cast, op_type, itemNum, comm, stream, is_dispatched]() -> int {
4765	- HcclSendRecvItem sendRecvInfo[itemNum];	4765	+ HcclSendRecvItem sendRecvInfo[itemNum];
4766	- HcclSendRecvType currType;	4766	+ HcclSendRecvType currType;
4767	- for (size_t i = 0; i < op_type.size(); ++i) {	4767	+ for (size_t i = 0; i < op_type.size(); ++i) {
4768	- if (op_type[i] == "isend") {	4768	+ if (op_type[i] == "isend") {
4769	- currType = HcclSendRecvType::HCCL_SEND;	4769	+ currType = HcclSendRecvType::HCCL_SEND;
4770	- } else if (op_type[i] == "irecv") {	4770	+ } else if (op_type[i] == "irecv") {
4771	- currType = HcclSendRecvType::HCCL_RECV;	4771	+ currType = HcclSendRecvType::HCCL_RECV;
4772	- } else {	4772	+ } else {
4773	- currType = HcclSendRecvType::HCCL_SEND_RECV_RESERVED;	4773	+ currType = HcclSendRecvType::HCCL_SEND_RECV_RESERVED;
4774	- }	4774	+ }
4775	- sendRecvInfo[i] = HcclSendRecvItem{currType,	4775	+ sendRecvInfo[i] = HcclSendRecvItem{currType,
4776	- tensor_ptr_list[i],	4776	+ tensor_ptr_list[i],
4777	- numel_list[i],	4777	+ numel_list[i],
4778	- type_list[i],	4778	+ type_list[i],
4779	- remote_rank_list_cast[i]	4779	+ remote_rank_list_cast[i]
4780	- };	4780	+ };
4781	- }	4781	+ }
4782	#ifndef BUILD_LIBTORCH	4782	#ifndef BUILD_LIBTORCH
4783	torch_npu::profiler::MstxRange range(	4783	torch_npu::profiler::MstxRange range(
4784	getMstxHcclMsg("HcclBatchSendRecv", sendRecvInfo[0].count, sendRecvInfo[0].dataType, comm, stream.id(), -1, -1),	4784	getMstxHcclMsg("HcclBatchSendRecv", sendRecvInfo[0].count, sendRecvInfo[0].dataType, comm, stream.id(), -1, -1),
4785	stream.stream(false), torch_npu::profiler::DOMAIN_COMMUNICATION);	4785	stream.stream(false), torch_npu::profiler::DOMAIN_COMMUNICATION);
4786	#endif	4786	#endif
4787	- if (c10_npu::is_core_control_enabled()) {	4787	+ if (c10_npu::is_core_control_enabled()) {
4788	c10_npu::UseStreamResInCurrentThread(stream.stream(false));	4788	c10_npu::UseStreamResInCurrentThread(stream.stream(false));
4789	}	4789	}
4790	auto hccl_result = hcclBatchIsendIrecv(sendRecvInfo, itemNum, comm, stream.stream(false));	4790	auto hccl_result = hcclBatchIsendIrecv(sendRecvInfo, itemNum, comm, stream.stream(false));
4791	*is_dispatched = true;	4791	*is_dispatched = true;
4792	return hccl_result;	4792	return hccl_result;
4793	- };	4793	+ };
4794	at_npu::native::OpCommand::RunOpApiV3("HcclBatchSendRecv", hccl_call, false, &stream);	4794	at_npu::native::OpCommand::RunOpApiV3("HcclBatchSendRecv", hccl_call, false, &stream);
4795	return HCCL_SUCCESS;	4795	return HCCL_SUCCESS;
4796	},	4796	},

Mtorch_npu/csrc/distributed/ProcessGroupHCCL.hpp+7-7

	@@ -444,7 +444,7 @@ public:
444	std::vector<std::pair<c10::weak_intrusive_ptr<c10::StorageImpl>, c10_npu::NPUStream>> recorded_outputs_;	444	std::vector<std::pair<c10::weak_intrusive_ptr<c10::StorageImpl>, c10_npu::NPUStream>> recorded_outputs_;
445		445
446	std::vector<at::Tensor> lazy_destroy_tensors_;	446	std::vector<at::Tensor> lazy_destroy_tensors_;
447	-	447	+
448	std::vector<at::Tensor> stashed_for_allocator_safety_;	448	std::vector<at::Tensor> stashed_for_allocator_safety_;
449	// unique id used to tell the trace buffer that this	449	// unique id used to tell the trace buffer that this
450	// work has completed	450	// work has completed
	@@ -627,14 +627,14 @@ public:
627	const c10d::ReduceOptions& opts = c10d::ReduceOptions());	627	const c10d::ReduceOptions& opts = c10d::ReduceOptions());
628		628
629	c10::intrusive_ptr<c10d::Work> batch_isend_irecv(	629	c10::intrusive_ptr<c10d::Work> batch_isend_irecv(
630	- std::vector<std::string>& op_type,	630	+ std::vector<std::string>& op_type,
631	- std::vector<at::Tensor>& tensors,	631	+ std::vector<at::Tensor>& tensors,
632	- std::vector<int64_t> remote_rank_list);	632	+ std::vector<int64_t> remote_rank_list);
633		633
634	c10::intrusive_ptr<c10d::Work> batch_isend_irecv_inner(	634	c10::intrusive_ptr<c10d::Work> batch_isend_irecv_inner(
635	- std::vector<std::string>& op_type,	635	+ std::vector<std::string>& op_type,
636	- std::vector<at::Tensor>& tensors,	636	+ std::vector<at::Tensor>& tensors,
637	- std::vector<int64_t> remote_rank_list);	637	+ std::vector<int64_t> remote_rank_list);
638		638
639	at::Tensor byte_alignment(at::Tensor& tensors) const;	639	at::Tensor byte_alignment(at::Tensor& tensors) const;
640		640

Mtorch_npu/csrc/npu/Graph.cpp+1-1

	@@ -743,7 +743,7 @@ void TORCH_NPU_API THNPGraph_init(PyObject* module) {
743	[](c10_npu::NPUGraph& self,	743	[](c10_npu::NPUGraph& self,
744	std::optional<c10_npu::MempoolId_t> pool_opt,	744	std::optional<c10_npu::MempoolId_t> pool_opt,
745	std::string capture_error_mode,	745	std::string capture_error_mode,
746	- bool report_shape) {	746	+ bool report_shape) {
747	aclmdlRICaptureMode capture_mode;	747	aclmdlRICaptureMode capture_mode;
748	c10_npu::MempoolId_t pool = pool_opt.has_value()	748	c10_npu::MempoolId_t pool = pool_opt.has_value()
749	? pool_opt.value() : c10_npu::MempoolId_t{0, 0};	749	? pool_opt.value() : c10_npu::MempoolId_t{0, 0};