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
|
#include <torch/csrc/lazy/core/helpers.h>
#include <c10/util/Half.h>
#include <c10/util/irange.h>
#include <torch/csrc/lazy/core/tensor_util.h>
#include <limits>
namespace torch {
namespace lazy {
std::vector<int64_t> DropDimensions(
c10::ArrayRef<int64_t> sizes,
c10::ArrayRef<int64_t> drop_dims) {
std::vector<int64_t> new_dims;
size_t drop_index = 0;
for (const auto i : c10::irange(sizes.size())) {
if (drop_index < drop_dims.size() && i == drop_dims[drop_index]) {
++drop_index;
} else {
new_dims.push_back(sizes[i]);
}
}
TORCH_CHECK(drop_index == drop_dims.size());
return new_dims;
}
int64_t GetCanonicalDimensionIndex(int64_t dim, int64_t rank) {
int64_t min_shape_dim = -rank;
int64_t max_shape_dim = rank - 1;
TORCH_CHECK(
min_shape_dim <= dim && dim <= max_shape_dim,
"Value out of range (expected to be in range of [",
min_shape_dim,
", ",
max_shape_dim,
"], but got ",
dim,
")");
int64_t dim_index = dim < 0 ? rank + dim : dim;
TORCH_CHECK(dim_index >= 0);
TORCH_CHECK(dim_index < rank);
return dim_index;
}
std::vector<int64_t> GetCanonicalDimensionIndices(
c10::ArrayRef<int64_t> dimensions,
int64_t rank) {
std::vector<int64_t> canonical_dim_indices;
for (int64_t dim : dimensions) {
canonical_dim_indices.push_back(GetCanonicalDimensionIndex(dim, rank));
}
return canonical_dim_indices;
}
int64_t GetCanonicalPosition(
c10::ArrayRef<int64_t> dimensions,
int64_t dim,
int64_t pos) {
dim = GetCanonicalDimensionIndex(dim, dimensions.size());
if (pos < 0) {
pos = GetCanonicalDimensionIndex(pos, dimensions[dim]);
} else {
pos = std::min<int64_t>(pos, dimensions[dim]);
}
return pos;
}
std::vector<int64_t> MakeTransposePermutation(
int64_t dim0,
int64_t dim1,
int64_t rank) {
int64_t canonical_dim0 = GetCanonicalDimensionIndex(dim0, rank);
int64_t canonical_dim1 = GetCanonicalDimensionIndex(dim1, rank);
auto permute_dims = Iota<int64_t>(rank);
std::swap(permute_dims[canonical_dim0], permute_dims[canonical_dim1]);
return permute_dims;
}
std::vector<int64_t> GetPromotedShape(
c10::ArrayRef<int64_t> shape1_dims,
c10::ArrayRef<int64_t> shape2_dims) {
std::vector<int64_t> dimensions;
// If the rank of a shape is bigger than then other, fill up the first
// dimensions with the ones of the bigger.
// Example:
// shape1 = [9, 7, 6, 5, 2]
// shape2 = [6, 1, 2]
// Insert [9, 7] into the dimensions vector.
if (shape1_dims.size() > shape2_dims.size()) {
dimensions.insert(
dimensions.end(),
shape1_dims.begin(),
shape1_dims.begin() + (shape1_dims.size() - shape2_dims.size()));
} else if (shape2_dims.size() > shape1_dims.size()) {
dimensions.insert(
dimensions.end(),
shape2_dims.begin(),
shape2_dims.begin() + (shape2_dims.size() - shape1_dims.size()));
}
// For the common dimensions, they must match, or one of them be 1.
size_t min_size = std::min(shape1_dims.size(), shape2_dims.size());
for (const auto i : c10::irange(min_size)) {
int64_t dim1 = shape1_dims[shape1_dims.size() - min_size + i];
int64_t dim2 = shape2_dims[shape2_dims.size() - min_size + i];
TORCH_CHECK(
dim1 == dim2 || dim1 == 1 || dim2 == 1,
"(",
c10::Join(", ", shape1_dims),
") and (",
c10::Join(", ", shape1_dims),
")");
if (dim1 == 0 || dim2 == 0) {
dimensions.push_back(0);
} else {
dimensions.push_back(std::max<int64_t>(dim1, dim2));
}
}
return dimensions;
}
Shape GetPromotedBinaryOpShape(const Shape& shape1, const Shape& shape2) {
return Shape(
promoteTypes(shape1.scalar_type(), shape2.scalar_type()),
GetPromotedShape(shape1.sizes(), shape2.sizes()));
}
std::vector<std::string> StrSplit(c10::string_view text, char delim) {
size_t start = 0;
size_t end = 0;
std::vector<std::string> tokens;
while ((start = text.find_first_not_of(delim, end)) != std::string::npos) {
end = text.find(delim, start);
auto token = text.substr(start, end - start);
tokens.emplace_back(token.begin(), token.end());
}
return tokens;
}
} // namespace lazy
} // namespace torch
|
