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
|
#include "caffe2/operators/utility_ops.h"
#include "caffe2/core/operator.h"
#include "caffe2/ideep/ideep_utils.h"
using namespace caffe2;
namespace {
class CopyCPUToIDEEPOp final : public IDEEPOperator {
public:
USE_SIMPLE_IDEEP_CTOR_DTOR(CopyCPUToIDEEPOp);
USE_IDEEP_DEF_ALIASES();
bool RunOnDevice() override {
const auto& X = OperatorBase::Input<Tensor>(0, CPU);
auto* Y = OperatorBase::OutputBlob(0);
itensor::dims src_dims(X.sizes().begin(), X.sizes().end());
if (!(Y->template IsType<itensor>() &&
Y->Get<itensor>().get_data_type() == itensor::data_type::f32) ||
Y->Get<itensor>().get_dims() != src_dims) {
Y->Reset(new itensor());
Y->GetMutable<itensor>()->resize(src_dims, itensor::data_type::f32);
}
Y->GetMutable<itensor>()->feed_from(
src_dims, itensor::data_type::f32, X.raw_data());
return true;
}
};
class IDEEPCopyOp final : public IDEEPOperator {
public:
USE_SIMPLE_IDEEP_CTOR_DTOR(IDEEPCopyOp);
USE_IDEEP_DEF_ALIASES();
bool RunOnDevice() override {
const auto& X = OperatorBase::Input<itensor>(0);
auto* Y = Output(0);
if (X != *Y) {
Y->reinit_like(X);
ideep::direct_copy::compute(X, *Y);
}
return true;
}
};
class CopyIDEEPToCPUOp final : public IDEEPOperator {
public:
USE_SIMPLE_IDEEP_CTOR_DTOR(CopyIDEEPToCPUOp);
USE_IDEEP_DEF_ALIASES();
bool RunOnDevice() override {
const auto& input_blob = OperatorBase::InputBlob(0);
if (BlobIsTensorType(input_blob, CPU)) {
VLOG(2) << "Directing sharing of TensorCPU";
const auto& X = OperatorBase::Input<Tensor>(0, CPU);
OutputTensorCopyFrom(0, at::device(CPU), X);
} else {
const auto& X = OperatorBase::Input<itensor>(0);
if (X.get_data_type() == itensor::data_type::f32) {
std::vector<int64_t> dims;
// NOLINTNEXTLINE(clang-diagnostic-sign-compare)
for (int i = 0; i < X.get_dims().size(); ++i) {
dims.push_back(X.get_dims()[i]);
}
auto* Y =
OperatorBase::OutputTensor(0, dims, at::dtype<float>().device(CPU));
itensor temp_ten(
X.get_desc().to_default_format(),
Y->template mutable_data<float>());
X.reorder_to(temp_ten);
} else {
CAFFE_THROW("Unsupported ideep type: ",
static_cast<int>(X.get_data_type()));
}
}
return true;
}
};
class IDEEPWeightedSumOp : public IDEEPOperator {
public:
USE_IDEEP_DEF_ALIASES();
USE_IDEEP_OPERATOR_FUNCTIONS();
IDEEPWeightedSumOp(const OperatorDef& operator_def, Workspace* ws)
: IDEEPOperator(operator_def, ws) {}
bool RunOnDevice() override {
CAFFE_ENFORCE_EQ(InputSize() % 2, 0);
auto ndims = Input(0).ndims();
auto nelems = Input(0).get_nelems();
auto w_nelems = Input(1).get_nelems();
CAFFE_ENFORCE_GT(nelems, 0);
CAFFE_ENFORCE_EQ(w_nelems, 1);
auto* output = Output(0);
std::vector<float> scales;
scales.reserve(InputSize() / 2);
std::vector<itensor> inputs;
inputs.reserve(InputSize() / 2);
for (int i = 0; i < InputSize(); i += 2) {
auto& X = Input(i);
CAFFE_ENFORCE(X.ndims() == ndims);
CAFFE_ENFORCE(X.get_nelems() == nelems);
CAFFE_ENFORCE(Input(i + 1).get_nelems() == w_nelems);
inputs.push_back(X);
auto scale = static_cast<float *>(Input(i + 1).get_data_handle());
scales.push_back(scale[0]);
}
ideep::sum::compute(scales, inputs, *output);
return true;
}
};
REGISTER_IDEEP_OPERATOR(CopyCPUToIDEEP, CopyCPUToIDEEPOp);
REGISTER_IDEEP_OPERATOR(CopyIDEEPToCPU, CopyIDEEPToCPUOp);
REGISTER_IDEEP_OPERATOR(Copy, IDEEPCopyOp);
REGISTER_IDEEP_OPERATOR(WeightedSum, IDEEPWeightedSumOp);
// NOLINTNEXTLINE(clang-diagnostic-unused-function,cppcoreguidelines-avoid-non-const-global-variables)
OPERATOR_SCHEMA(CopyCPUToIDEEP)
.NumInputs(1)
.NumOutputs(1)
.Input(0, "cpu_blob", "The input TensorCPU to copy")
.Output(0, "ideep_blob", "The output IDEEP tensort to copy to");
// NOLINTNEXTLINE(clang-diagnostic-unused-function,cppcoreguidelines-avoid-non-const-global-variables)
OPERATOR_SCHEMA(CopyIDEEPToCPU)
.NumInputs(1)
.NumOutputs(1)
.Input(0, "ideep_blob", "The input IDEEP tensort to copy")
.Output(0, "cpu_blob", "The output TensorCPU to copy to");
} // namespace
|
