/
utils.h
170 lines (163 loc) · 11.5 KB
/
utils.h
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
#include "paddle/phi/backends/cpu/cpu_context.h"
#include "paddle/phi/backends/gpu/gpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/copy_kernel.h"
#include "paddle/phi/kernels/empty_kernel.h"
#define DEFINE_SPARSE_UNARY_KERNEL(dense_kernel_func) \
namespace phi { \
namespace sparse { \
\
template <typename T, typename Context> \
void SparseCoo##dense_kernel_func(const Context& dev_ctx, \
const SparseCooTensor& x, \
SparseCooTensor* out) { \
DenseTensor non_zero_indices = \
phi::EmptyLike<T, Context>(dev_ctx, x.non_zero_indices()); \
DenseTensor non_zero_elements = \
phi::EmptyLike<T, Context>(dev_ctx, x.non_zero_elements()); \
phi::Copy(dev_ctx, \
x.non_zero_indices(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_indices); \
phi::dense_kernel_func<T, Context>( \
dev_ctx, x.non_zero_elements(), &non_zero_elements); \
out->SetMember(non_zero_indices, non_zero_elements, x.dims(), true); \
} \
\
template <typename T, typename Context> \
void SparseCsr##dense_kernel_func(const Context& dev_ctx, \
const SparseCsrTensor& x, \
SparseCsrTensor* out) { \
DenseTensor non_zero_crows = \
phi::EmptyLike<T, Context>(dev_ctx, x.non_zero_crows()); \
DenseTensor non_zero_cols = \
phi::EmptyLike<T, Context>(dev_ctx, x.non_zero_cols()); \
DenseTensor non_zero_elements = \
phi::EmptyLike<T, Context>(dev_ctx, x.non_zero_elements()); \
phi::Copy(dev_ctx, \
x.non_zero_crows(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_crows); \
phi::Copy(dev_ctx, \
x.non_zero_cols(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_cols); \
phi::dense_kernel_func<T, Context>( \
dev_ctx, x.non_zero_elements(), &non_zero_elements); \
out->SetMember( \
non_zero_crows, non_zero_cols, non_zero_elements, x.dims()); \
} \
} \
}
#define DEFINE_SPARSE_UNARY_GRAD_KERNEL(dense_kernel_func) \
namespace phi { \
namespace sparse { \
\
template <typename T, typename Context> \
void SparseCoo##dense_kernel_func(const Context& dev_ctx, \
const SparseCooTensor& x_or_out, \
const SparseCooTensor& out_grad, \
SparseCooTensor* x_grad) { \
DenseTensor non_zero_indices = \
phi::EmptyLike<T, Context>(dev_ctx, x_or_out.non_zero_indices()); \
DenseTensor non_zero_elements = \
phi::EmptyLike<T, Context>(dev_ctx, x_or_out.non_zero_elements()); \
phi::Copy(dev_ctx, \
x_or_out.non_zero_indices(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_indices); \
phi::dense_kernel_func<T, Context>(dev_ctx, \
x_or_out.non_zero_elements(), \
out_grad.non_zero_elements(), \
&non_zero_elements); \
x_grad->SetMember( \
non_zero_indices, non_zero_elements, x_or_out.dims(), true); \
} \
\
template <typename T, typename Context> \
void SparseCsr##dense_kernel_func(const Context& dev_ctx, \
const SparseCsrTensor& x_or_out, \
const SparseCsrTensor& out_grad, \
SparseCsrTensor* out) { \
DenseTensor non_zero_crows = \
phi::EmptyLike<T, Context>(dev_ctx, x_or_out.non_zero_crows()); \
DenseTensor non_zero_cols = \
phi::EmptyLike<T, Context>(dev_ctx, x_or_out.non_zero_cols()); \
DenseTensor non_zero_elements = \
phi::EmptyLike<T, Context>(dev_ctx, x_or_out.non_zero_elements()); \
phi::Copy(dev_ctx, \
x_or_out.non_zero_crows(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_crows); \
phi::Copy(dev_ctx, \
x_or_out.non_zero_cols(), \
dev_ctx.GetPlace(), \
false, \
&non_zero_cols); \
phi::dense_kernel_func<T, Context>(dev_ctx, \
x_or_out.non_zero_elements(), \
out_grad.non_zero_elements(), \
&non_zero_elements); \
out->SetMember( \
non_zero_crows, non_zero_cols, non_zero_elements, x_or_out.dims()); \
} \
} \
}
#define REGISTER_CPU_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func) \
PD_REGISTER_KERNEL(sparse_coo_##kernel_name, \
CPU, \
ALL_LAYOUT, \
phi::sparse::SparseCoo##dense_kernel_func, \
float, \
double) { \
kernel->InputAt(0).SetDataLayout(phi::DataLayout::SPARSE_COO); \
} \
PD_REGISTER_KERNEL(sparse_csr_##kernel_name, \
CPU, \
ALL_LAYOUT, \
phi::sparse::SparseCsr##dense_kernel_func, \
float, \
double) { \
kernel->InputAt(0).SetDataLayout(phi::DataLayout::SPARSE_CSR); \
}
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
#define REGISTER_GPU_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func) \
PD_REGISTER_KERNEL(sparse_coo_##kernel_name, \
GPU, \
ALL_LAYOUT, \
phi::sparse::SparseCoo##dense_kernel_func, \
float, \
double, \
phi::dtype::float16) { \
kernel->InputAt(0).SetDataLayout(phi::DataLayout::SPARSE_COO); \
} \
\
PD_REGISTER_KERNEL(sparse_csr_##kernel_name, \
GPU, \
ALL_LAYOUT, \
phi::sparse::SparseCsr##dense_kernel_func, \
float, \
double, \
phi::dtype::float16) { \
kernel->InputAt(0).SetDataLayout(phi::DataLayout::SPARSE_CSR); \
}
#else
// This macro definition is empty when GPU is disabled
#define REGISTER_GPU_SPARSE_UNARY_KERNEL(sparse_kernel_name, dense_kernel_func)
#endif
#define REGISTER_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func) \
REGISTER_CPU_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func) \
REGISTER_GPU_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func)
#define DEFINE_AND_REGISTER_SPARSE_UNARY_KERNEL(kernel_name, \
dense_kernel_func) \
DEFINE_SPARSE_UNARY_KERNEL(dense_kernel_func) \
REGISTER_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func)
#define DEFINE_AND_REGISTER_SPARSE_UNARY_GRAD_KERNEL(kernel_name, \
dense_kernel_func) \
DEFINE_SPARSE_UNARY_GRAD_KERNEL(dense_kernel_func) \
REGISTER_SPARSE_UNARY_KERNEL(kernel_name, dense_kernel_func)