1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
|
# Owner(s): ["module: sparse"]
#
# Test to ensure sparsity information propagates properly into traced graph.
#
import sys
import unittest
import torch
from torch._dynamo.config import is_fbcode
from torch._subclasses.fake_tensor import FakeTensor
from torch.testing._internal.common_utils import (
instantiate_parametrized_tests,
parametrize,
run_tests,
subtest,
TestCase,
)
# Various data types (preserved over operations).
DTYPES = [
torch.int64,
torch.float16,
torch.bfloat16,
torch.float32,
torch.float64,
]
# Various index types.
ITYPES = [torch.int32, torch.int64]
# Constructs a subtest for every sparse layout currently supported in torch.sparse.
def all_sparse_layouts(test_name="layout"):
return parametrize(
test_name,
[
subtest(torch.sparse_coo, name="SparseCOO"),
subtest(torch.sparse_csr, name="SparseCSR"),
subtest(torch.sparse_csc, name="SparseCSC"),
subtest(torch.sparse_bsr, name="SparseBSR"),
subtest(torch.sparse_bsc, name="SparseBSC"),
],
)
#
# Various network examples.
#
class IdNet(torch.nn.Module):
def forward(self, x):
return x
class SumNet(torch.nn.Module):
def forward(self, x):
return x.sum()
class EltwiseNet(torch.nn.Module):
def forward(self, x):
return torch.nn.functional.relu(2 * torch.abs(-x))
class ToDenseNet(torch.nn.Module):
def forward(self, x):
return x.to_dense()
class AddNet(torch.nn.Module):
def forward(self, x, y):
return torch.add(x, y)
class SparseActivationCOO(torch.nn.Module):
def forward(self, x):
return [xi.to_sparse() for xi in x]
class SparseActivationCSR(torch.nn.Module):
def forward(self, x):
return [xi.to_sparse_csr() for xi in x]
#
# The test driver.
#
@unittest.skipIf(is_fbcode(), "See torch._dynamo.config")
@unittest.skipIf(
sys.version_info >= (3, 12), "torch.compile is not supported on python 3.12+"
)
class TestSparseProp(TestCase):
def setUp(self):
TestCase.setUp(self)
def assertEqualMeta(self, x, y):
self.assertIsInstance(x, FakeTensor)
self.assertIsInstance(y, torch.Tensor)
# Convert expected value to meta for comparison.
y = y.to("meta")
self.assertEqual(x, y, exact_layout=True, exact_is_coalesced=True)
# When x or y is a meta tensor (say, `x.device == "meta"`), then
# assertEqual(x, y) compares only x and y attributes but skips
# comparing their values. In the case of sparse tensors, this means
# that comparing indices and values attributes are skipped as well,
# which is why we are doing that explicitly below.
if x.layout is torch.strided:
pass
elif x.layout is torch.sparse_coo:
self.assertEqual(x._indices(), y._indices(), exact_layout=True)
self.assertEqual(x._values(), y._values(), exact_layout=True)
else:
if x.layout in {torch.sparse_csr, torch.sparse_bsr}:
x_meta1, y_meta1 = (x.crow_indices(), y.crow_indices())
x_meta2, y_meta2 = (x.col_indices(), y.col_indices())
elif x.layout in {torch.sparse_csc, torch.sparse_bsc}:
x_meta1, y_meta1 = (x.ccol_indices(), y.ccol_indices())
x_meta2, y_meta2 = (x.row_indices(), y.row_indices())
else:
assert 0 # unreachable
self.assertEqual(x_meta1, y_meta1, exact_layout=True)
self.assertEqual(x_meta2, y_meta2, exact_layout=True)
self.assertEqual(x.values(), y.values(), exact_layout=True)
@parametrize("dtype", DTYPES)
@parametrize("itype", ITYPES)
@all_sparse_layouts("layout")
def test_idnet(self, dtype, itype, layout):
net = IdNet()
for sparse_input in self.generate_simple_inputs(
layout,
device="cpu",
dtype=dtype,
index_dtype=itype,
):
# Build the traced graph.
prog = torch.export.export(net, (sparse_input,))
# Test arg/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i == 0:
self.assertEqualMeta(meta, sparse_input)
else:
self.assertEqual(meta, None)
@parametrize("dtype", DTYPES)
@parametrize("itype", ITYPES)
@all_sparse_layouts("layout")
def test_sumnet(self, dtype, itype, layout):
net = SumNet()
for sparse_input in self.generate_simple_inputs(
layout,
device="cpu",
dtype=dtype,
index_dtype=itype,
):
result = net(sparse_input)
# Build the traced graph.
prog = torch.export.export(net, (sparse_input,))
# Test arg/sum/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i == 0:
self.assertEqualMeta(meta, sparse_input)
elif i == 1:
self.assertEqualMeta(meta, result)
else:
self.assertEqual(meta, None)
@parametrize("dtype", DTYPES)
@parametrize("itype", ITYPES)
@all_sparse_layouts("layout")
def test_eltwisenet(self, dtype, itype, layout):
net = EltwiseNet()
for sparse_input in self.generate_simple_inputs(
layout,
device="cpu",
dtype=dtype,
index_dtype=itype,
):
result = net(sparse_input)
# Build the traced graph.
prog = torch.export.export(net, (sparse_input,))
# Test arg/neg/abs/mul/relu/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i <= 4:
self.assertEqualMeta(meta, result)
else:
self.assertEqual(meta, None)
@parametrize("dtype", DTYPES)
@parametrize("itype", ITYPES)
@all_sparse_layouts("layout")
def test_todensenet(self, dtype, itype, layout):
net = ToDenseNet()
for sparse_input in self.generate_simple_inputs(
layout,
device="cpu",
dtype=dtype,
index_dtype=itype,
):
result = net(sparse_input)
# Build the traced graph.
prog = torch.export.export(net, (sparse_input,))
# Test arg/todense/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i == 0:
self.assertEqualMeta(meta, sparse_input)
elif i == 1:
self.assertEqualMeta(meta, result)
else:
self.assertEqual(meta, None)
def test_add(self):
net = AddNet()
Y = torch.arange(16, 32, dtype=torch.float32).view(4, 4)
A = torch.tensor(
[
[0.0, 1.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 2.0],
[0.0, 0.0, 1.0, 1.0],
[3.0, 0.0, 3.0, 0.0],
],
dtype=torch.float32,
)
S = A.to_sparse_csr()
result = net(S, Y)
# Build the traced graph.
prog = torch.export.export(net, (S, Y))
# Test args/add/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i == 0:
self.assertEqualMeta(meta, S)
elif i == 1:
self.assertEqualMeta(meta, Y)
elif i == 2:
self.assertEqualMeta(meta, result)
else:
self.assertEqual(meta, None)
def test_activation_coo(self):
net = SparseActivationCOO()
x = [torch.randn(3, 3) for _ in range(3)]
result = net(x)
# Build the traced graph.
prog = torch.export.export(net, args=(x,))
# Test args/to_sparse/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i <= 2:
self.assertEqualMeta(meta, x[i])
elif i <= 5:
self.assertEqualMeta(meta, result[i - 3])
else:
self.assertEqual(meta, None)
def test_activation_csr(self):
net = SparseActivationCSR()
x = [torch.randn(3, 3) for _ in range(3)]
result = net(x)
# Build the traced graph.
prog = torch.export.export(net, args=(x,))
# Test args/to_sparse/output.
for i, node in enumerate(prog.graph.nodes):
meta = node.meta.get("val", None)
if i <= 2:
self.assertEqualMeta(meta, x[i])
elif i <= 5:
self.assertEqualMeta(meta, result[i - 3])
else:
self.assertEqual(meta, None)
instantiate_parametrized_tests(TestSparseProp)
if __name__ == "__main__":
run_tests()
|
