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
|
# Must happen before importing caffe2.python.*
import caffe2.python.fakelowp.init_shared_libs # noqa
import datetime
import numpy as np
from hypothesis import given, settings, example
from hypothesis import strategies as st
from caffe2.python import core, workspace
from caffe2.python.onnx.onnxifi import onnxifi_caffe2_net
from caffe2.python.fakelowp.test_utils import print_test_debug_info
import caffe2.python.serialized_test.serialized_test_util as serial
# Test that parallel chunks behave the same way as the serial one
workspace.GlobalInit(
[
"caffe2",
"--glow_global_fp16=1",
"--glow_global_fused_scale_offset_fp16=1",
"--glow_global_force_sls_fp16_accum=1",
"--glow_nnpi_num_parallel_chunks=2",
"--glow_use_dag_optimizer=false",
"--glow_dump_graph=true",
]
)
class Fusions(serial.SerializedTestCase):
def _get_scale_zp(self, tensor):
tensor_max = np.max(tensor)
tensor_min = min(0, np.min(tensor))
scale = np.float32(np.float16((tensor_max - tensor_min) / 255.0))
if scale < 1e-6:
scale = np.float32(1e-6)
zero_point = 0 - tensor_min / scale
zero_point = int(round(np.clip(zero_point, 0, 255.0)))
return (scale, zero_point)
@given(
scale=st.floats(1e-4, 1e2),
zp=st.integers(-128, 128),
rand_seed=st.integers(0, 65534),
m=st.integers(32, 64),
k=st.integers(1000, 6000),
n=st.integers(200, 600),
)
# @example(m=64, k=5423, n=553, scale=1e-3, zp=120, rand_seed=1)
@settings(deadline=datetime.timedelta(seconds=1000), max_examples=1)
def test_ParallelFC(self, m, k, n, scale, zp, rand_seed):
np.random.seed(rand_seed)
workspace.ResetWorkspace()
# Y = W_T * X + b
X_fp32 = np.random.uniform(-1, 1, size=(m, k)).astype(np.float16) \
.astype(np.float32)
W_fp32 = np.random.uniform(-1, 1, size=(n, k)).astype(np.float32)
b_fp32 = np.zeros((n,), dtype=np.float32)
X_scale, X_zero_point = self._get_scale_zp(X_fp32)
workspace.FeedBlob("X", X_fp32)
workspace.FeedBlob("W", W_fp32)
workspace.FeedBlob("b", b_fp32)
workspace.RunOperatorOnce(
core.CreateOperator(
"Int8FCPackWeight",
["W"],
["W_int8"],
engine="DNNLOWP",
save_unpacked_weights=True,
in_scale=X_scale,
)
)
ref_net = core.Net("net")
ref_net.Int8QuantizeNNPI(
["X"],
["X_int8"],
Y_scale=X_scale,
Y_zero_point=X_zero_point
)
ref_net.Int8FCFakeAcc32NNPI(
["X_int8", "W_int8", "b"],
["Y_int8"],
Y_scale=X_scale,
Y_zero_point=X_zero_point,
)
ref_net.Int8Relu(
["Y_int8"],
["Y_relu"],
Y_zero_point=X_zero_point,
Y_scale=X_scale,
)
ref_net.Int8DequantizeNNPI(
["Y_relu"],
["Y"]
)
ref_net.Proto().external_output.append("Y")
# run ref_net
workspace.RunNetOnce(ref_net)
Y_fbgemm = workspace.FetchBlob("Y")
# run onnxifi net
ref_net.Proto().op[0].type = "Int8Quantize"
ref_net.Proto().op[1].type = "Int8FC"
ref_net.Proto().op[2].type = "Int8Relu"
ref_net.Proto().op[3].type = "Int8Dequantize"
net_onnxified = onnxifi_caffe2_net(
ref_net.Proto(),
{},
debug=True,
adjust_batch=False,
use_onnx=False,
weight_names=["W_int8", "b"],
)
num_onnxified_ops = sum(
1 if o.type == "Onnxifi" else 0 for o in net_onnxified.op
)
print(net_onnxified)
np.testing.assert_equal(num_onnxified_ops, 1)
workspace.CreateNet(net_onnxified)
workspace.RunNet(net_onnxified.name)
Y_glow = workspace.FetchBlob("Y")
if not np.allclose(Y_glow, Y_fbgemm):
diff_Y = np.abs(Y_glow - Y_fbgemm)
print_test_debug_info(
"int8_fc",
{
"seed": rand_seed,
"n": n,
"X": X_fp32,
"W": W_fp32,
"b": b_fp32,
"Y_fbgemm": Y_fbgemm,
"Y_glow": Y_glow,
"diff": diff_Y,
"maxdiff": diff_Y.max(axis=1),
},
)
assert 0
|
