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
|
#include <assert.h>
#include "caffe2/operators/elementwise_linear_op.h"
#include "caffe2/core/context_gpu.h"
#include "caffe2/operators/operator_fallback_gpu.h"
#include "caffe2/utils/cub_namespace.cuh"
#include <cub/block/block_reduce.cuh>
namespace caffe2 {
namespace {
__global__ void ElementwiseLinearKernel(const int N, const int D,
const float* X_data, const float* a_data, const float* b_data,
float* Y_data) {
CUDA_1D_KERNEL_LOOP(i, N * D) {
int d = i % D;
Y_data[i] = X_data[i] * a_data[d] + b_data[d];
}
}
__global__ void ElementwiseLinearGradientKernel(const int N, const int D,
const float* g_o_data, const float* X_data, const float* a_data,
float* g_X_data, float* g_a_data, float* g_b_data) {
int d = blockIdx.x; // One block per D
float g_a_sum = 0;
float g_b_sum = 0;
for (int n = threadIdx.x; n < N; n += blockDim.x) {
const float gox = g_o_data[n * D + d];
g_X_data[n * D + d] = gox * a_data[d];
g_a_sum += gox * X_data[n * D + d];
g_b_sum += gox;
}
typedef cub::BlockReduce<float, CAFFE_CUDA_NUM_THREADS> BlockReduce;
__shared__ typename BlockReduce::TempStorage temp_storage;
float g_a_sum_tot = BlockReduce(temp_storage).Sum(g_a_sum);
__syncthreads();
float g_b_sum_tot = BlockReduce(temp_storage).Sum(g_b_sum);
if (threadIdx.x == 0) {
g_a_data[d] = g_a_sum_tot;
g_b_data[d] = g_b_sum_tot;
}
}
} // namespace
template<>
bool ElementwiseLinearOp<float, CUDAContext>::RunOnDevice(){
const auto& X = Input(0);
const auto& a = Input(1);
const auto& b = Input(2);
const auto canonical_axis = X.canonical_axis_index(axis_);
const int N = X.size_to_dim(canonical_axis);
const int D = X.size_from_dim(canonical_axis);
CAFFE_ENFORCE_EQ(a.dim(), 1, a.dim());
CAFFE_ENFORCE_EQ(a.dim(0), D, a.dim());
CAFFE_ENFORCE_EQ(b.dim(), 1, b.dim());
CAFFE_ENFORCE_EQ(b.dim(0), D, b.dim());
auto* Y = Output(0, X.sizes(), at::dtype<float>());
ElementwiseLinearKernel<<<
CAFFE_GET_BLOCKS(N * D),
CAFFE_CUDA_NUM_THREADS,
0,
context_.cuda_stream()>>>(
N,
D,
X.data<float>(),
a.data<float>(),
b.data<float>(),
Y->template mutable_data<float>());
C10_CUDA_KERNEL_LAUNCH_CHECK();
return true;
}
template<>
bool ElementwiseLinearGradientOp<float, CUDAContext>::RunOnDevice(){
const auto& g_o = Input(0);
const auto& X = Input(1);
const auto& a = Input(2);
const auto canonical_axis = X.canonical_axis_index(axis_);
const int N = X.size_to_dim(canonical_axis);
const int D = X.size_from_dim(canonical_axis);
CAFFE_ENFORCE_EQ(a.dim(), 1, a.dim());
CAFFE_ENFORCE_EQ(a.dim(0), D, a.dim());
auto* g_X = Output(0, X.sizes(), at::dtype<float>());
auto * g_a = Output(1, a.sizes(), at::dtype<float>());
auto * g_b = Output(2, a.sizes(), at::dtype<float>());
float* g_a_data = g_a->template mutable_data<float>();
float* g_b_data = g_b->template mutable_data<float>();
ElementwiseLinearGradientKernel<<<
D,
CAFFE_CUDA_NUM_THREADS,
0,
context_.cuda_stream()>>>(
N,
D,
g_o.data<float>(),
X.data<float>(),
a.data<float>(),
g_X->template mutable_data<float>(),
g_a_data,
g_b_data);
C10_CUDA_KERNEL_LAUNCH_CHECK();
return true;
}
REGISTER_CUDA_OPERATOR(ElementwiseLinear,
ElementwiseLinearOp<float, CUDAContext>);
REGISTER_CUDA_OPERATOR(ElementwiseLinearGradient,
ElementwiseLinearGradientOp<float, CUDAContext>);
} // namespace caffe2
|
