【inductor】fix inductor ci bug-pytorch-AtomGit

Mtest/_inductor/test_attncp.py+0-2

	@@ -1,11 +1,9 @@
1	-import unittest
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests	2	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
5	import torch_npu	4	import torch_npu
6		5
7		6
8	-@unittest.skip("This test is not supported yet")
9	class TestAttnCp(TestUtils):	7	class TestAttnCp(TestUtils):
10	shape = (8, 8, 256, 128)	8	shape = (8, 8, 256, 128)
11	dim = -1	9	dim = -1

Mtest/_inductor/test_batch_norm.py+0-2

	@@ -1,11 +1,9 @@
1	-import unittest
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests	2	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
5	import torch_npu	4	import torch_npu
6		5
7		6
8	-@unittest.skip("This test is not supported yet")
9	class TestNativeBatchNorm(TestUtils):	7	class TestNativeBatchNorm(TestUtils):
10	def op_calc(self, input_element):	8	def op_calc(self, input_element):
11	# 创建权重和偏置张量	9	# 创建权重和偏置张量

Mtest/_inductor/test_issue59.py+0-2

	@@ -1,4 +1,3 @@
1	-from unittest import skip
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests	2	from torch.testing._internal.common_utils import run_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
	@@ -20,7 +19,6 @@ class Test_issue59(TestUtils):
20	mean_2 = rsqrt / torch.numel(rsqrt)	19	mean_2 = rsqrt / torch.numel(rsqrt)
21	return mul, add, mean_2	20	return mul, add, mean_2
22		21
23	- @skip("skip ci error")
24	def test_issue59(self):	22	def test_issue59(self):
25	device = 'npu'	23	device = 'npu'
26	x = torch.randn((1, 1024), device=device, dtype=torch.float32)	24	x = torch.randn((1, 1024), device=device, dtype=torch.float32)

Mtest/_inductor/test_npu_device.py+0-2

	@@ -1,4 +1,3 @@
1	-import unittest
2	import torch	1	import torch
3	from torch_npu.npu import device_count	2	from torch_npu.npu import device_count
4	from torch_npu.utils._dynamo_device import NpuInterface, current_device, set_device	3	from torch_npu.utils._dynamo_device import NpuInterface, current_device, set_device
	@@ -8,7 +7,6 @@ from torch_npu._inductor.npu_device import NewNPUDeviceOpOverrides
8	from torch_npu.testing.testcase import TestCase, run_tests	7	from torch_npu.testing.testcase import TestCase, run_tests
9		8
10		9
11	-@unittest.skip("This test is not supported yet")
12	class TestNpuDevice(TestCase):	10	class TestNpuDevice(TestCase):
13	def test_aoti_get_stream(self):	11	def test_aoti_get_stream(self):
14	overrides = NewNPUDeviceOpOverrides()	12	overrides = NewNPUDeviceOpOverrides()

Mtest/_inductor/test_reduction_brocast_add.py+0-2

	@@ -1,11 +1,9 @@
1	-import unittest
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests	2	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
5	import torch_npu	4	import torch_npu
6		5
7		6
8	-@unittest.skip("This test is not supported yet")
9	class TestSumAdd(TestUtils):	7	class TestSumAdd(TestUtils):
10	def foo(self, a, b, dim, shape):	8	def foo(self, a, b, dim, shape):
11	y = a + b	9	y = a + b

Mtest/_inductor/test_renorm.py+0-2

	@@ -1,11 +1,9 @@
1	-import unittest
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests	2	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
5	import torch_npu	4	import torch_npu
6		5
7		6
8	-@unittest.skip("This test is not supported yet")
9	class TestRenorm(TestUtils):	7	class TestRenorm(TestUtils):
10	def op_calc(self, input_element, dim):	8	def op_calc(self, input_element, dim):
11	return torch.renorm(input_element, p=2, dim=dim, maxnorm=5)	9	return torch.renorm(input_element, p=2, dim=dim, maxnorm=5)

Mtest/_inductor/test_var_mean.py+0-2

	@@ -1,4 +1,3 @@
1	-from unittest import skip
2	import torch	1	import torch
3	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests	2	from torch.testing._internal.common_utils import run_tests, parametrize, instantiate_parametrized_tests
4	from testutils import TestUtils	3	from testutils import TestUtils
	@@ -10,7 +9,6 @@ class TestVarMean(TestUtils):
10	return torch.var_mean(input_element, dim)	9	return torch.var_mean(input_element, dim)
11		10
12	# case：The shape must not be too large	11	# case：The shape must not be too large
13	- @skip("skip ci codegen error")
14	@parametrize('shape', [(8, 64, 128)])	12	@parametrize('shape', [(8, 64, 128)])
15	@parametrize('dim', [0, 1, 2, (0, 2), (0, 1)])	13	@parametrize('dim', [0, 1, 2, (0, 2), (0, 1)])
16	@parametrize('dtype', ['float32'])	14	@parametrize('dtype', ['float32'])

Mtorch_npu/_inductor/codegen/ir.py+5-4

	@@ -166,12 +166,13 @@ def substituted_dims_in_indexing(self, indexing, kernel, range_tree_nodes_substi
166	return substituted	166	return substituted
167		167
168		168
169	-def generate_body_indexing(body, indices):	169	+def generate_body_indexing(body, indices, allow_same_symbol_in_index=False):
170	index = list(itertools.chain.from_iterable(indices))	170	index = list(itertools.chain.from_iterable(indices))
171	if not (len(index) == len(body.var_ranges)):	171	if not (len(index) == len(body.var_ranges)):
172	raise RuntimeError("assert len(index) == len(body.var_ranges), (index, body.var_ranges)")	172	raise RuntimeError("assert len(index) == len(body.var_ranges), (index, body.var_ranges)")
173	- if not (all(v not in body.var_ranges for v in index)):	173	+ if not allow_same_symbol_in_index:
174	- raise RuntimeError("assert all(v not in body.var_ranges for v in index)")	174	+ if not (all(v not in body.var_ranges for v in index)):
		175	+ raise RuntimeError("assert all(v not in body.var_ranges for v in index)")
175		176
176	replacements = dict(zip(body.var_ranges.keys(), index))	177	replacements = dict(zip(body.var_ranges.keys(), index))
177	indexing_map = dict(zip(index, body.var_ranges.keys()))	178	indexing_map = dict(zip(index, body.var_ranges.keys()))
	@@ -193,7 +194,7 @@ def transform_dims_in_indexing(self, indices):
193	# select tiling axis, recover missing dimensions,	194	# select tiling axis, recover missing dimensions,
194	def loopbody__call__(self, *indices, allow_same_symbol_in_index=False):	195	def loopbody__call__(self, *indices, allow_same_symbol_in_index=False):
195	if self.indexing is None:	196	if self.indexing is None:
196	- generate_body_indexing(self, indices)	197	+ generate_body_indexing(self, indices, allow_same_symbol_in_index)
197	result = self.root_block()	198	result = self.root_block()
198	self.indexing = None	199	self.indexing = None
199	return result	200	return result

Mtorch_npu/_inductor/codegen/triton.py+4-1

	@@ -1185,7 +1185,10 @@ class NPUIndexTritonKernel(TritonKernel):
1185	) -> Union[CSEVariable, Tuple[CSEVariable, ...]]:	1185	) -> Union[CSEVariable, Tuple[CSEVariable, ...]]:
1186	if not self.inside_reduction:	1186	if not self.inside_reduction:
1187	raise RuntimeError("assert self.inside_reduction")	1187	raise RuntimeError("assert self.inside_reduction")
1188	- masks = {f"{node.symbol()}_mask" for node in self.sorted_axis}	1188	+ if self.persistent_reduction and self.numof_reduction_axis() == 1:
		1189	+ masks = {f"{node.symbol()}_mask" for node in self.sorted_axis if node.name[0] != "r"}
		1190	+ else:
		1191	+ masks = {f"{node.symbol()}_mask" for node in self.sorted_axis}
1189	self.filter_masks(masks)	1192	self.filter_masks(masks)
1190	masks = sorted(masks)	1193	masks = sorted(masks)
1191	if self._load_mask:	1194	if self._load_mask: