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
|
#include "THCTensorMath.h"
#include "THCGeneral.h"
#include "THCTensorCopy.h"
#include "THCApply.cuh"
#include "THCNumerics.cuh"
#include <cfloat>
void THCudaTensor_cat(THCState *state, THCudaTensor *result, THCudaTensor *ta, THCudaTensor *tb, int dimension)
{
THCudaTensor* inputs[2];
inputs[0] = ta;
inputs[1] = tb;
THCudaTensor_catArray(state, result, inputs, 2, dimension);
}
void THCudaTensor_catArray(THCState *state, THCudaTensor *result, THCudaTensor **inputs, int numInputs, int dimension)
{
THLongStorage *size;
int i, j;
long offset;
int ndim = dimension + 1;
for (i = 0; i < numInputs; i++)
{
ndim = THMax(ndim, THCudaTensor_nDimension(state, inputs[i]));
}
THArgCheck(numInputs > 0, 3, "invalid number of inputs %d", numInputs);
THArgCheck(dimension >= 0, 4, "invalid dimension %d", dimension+1);
size = THLongStorage_newWithSize(ndim);
for(i = 0; i < ndim; i++)
{
long dimSize = i < THCudaTensor_nDimension(state, inputs[0])
? THCudaTensor_size(state, inputs[0], i)
: 1;
if (i == dimension)
{
for (j = 1; j < numInputs; j++)
{
dimSize += i < THCudaTensor_nDimension(state, inputs[j])
? THCudaTensor_size(state, inputs[j], i)
: 1;
}
}
else
{
for (j = 1; j < numInputs; j++)
{
if (dimSize != (i < THCudaTensor_nDimension(state, inputs[j])
? THCudaTensor_size(state, inputs[j], i)
: 1)) {
THLongStorage_free(size);
THError("inconsistent tensor sizes");
}
}
}
size->data[i] = dimSize;
}
THCudaTensor_resize(state, result, size, NULL);
THLongStorage_free(size);
offset = 0;
for (j = 0; j < numInputs; j++)
{
long dimSize = dimension < THCudaTensor_nDimension(state, inputs[j])
? THCudaTensor_size(state, inputs[j], dimension)
: 1;
THCudaTensor *nt = THCudaTensor_newWithTensor(state, result);
THCudaTensor_narrow(state, nt, NULL, dimension, offset, dimSize);
THCudaTensor_copy(state, nt, inputs[j]);
THCudaTensor_free(state, nt);
offset += dimSize;
}
}
struct TensorAddCMulOp {
TensorAddCMulOp(float v) : val(v) {}
__device__ __forceinline__ void
operator()(float* out, float* in1, float* in2) {
*out += val * *in1 * *in2;
}
float val;
};
void THCudaTensor_addcmul(THCState *state, THCudaTensor *self_, THCudaTensor *t, float value, THCudaTensor *src1, THCudaTensor *src2)
{
THAssert(THCudaTensor_checkGPU(state, 4, self_, t, src1, src2));
if(self_ != t)
{
THCudaTensor_resizeAs(state, self_, t);
THCudaTensor_copy(state, self_, t);
}
else
{
THArgCheck(THCudaTensor_nElement(state, self_) == THCudaTensor_nElement(state, src1),
1, "sizes do not match");
}
THArgCheck(THCudaTensor_nElement(state, src1) == THCudaTensor_nElement(state, src2),
3, "sizes do not match");
if (!THC_pointwiseApply3(state, self_, src1, src2, TensorAddCMulOp(value))) {
THArgCheck(false, 2, CUTORCH_DIM_WARNING);
}
THCudaCheck(cudaGetLastError());
}
struct TensorAddCDivOp {
TensorAddCDivOp(float v) : val(v) {}
__device__ __forceinline__ void
operator()(float* out, float* in1, float* in2) {
*out += val * *in1 / *in2;
}
float val;
};
void THCudaTensor_addcdiv(THCState *state, THCudaTensor *self_, THCudaTensor *t, float value, THCudaTensor *src1, THCudaTensor *src2)
{
THAssert(THCudaTensor_checkGPU(state, 4, self_, t, src1, src2));
if(self_ != t)
{
THCudaTensor_resizeAs(state, self_, t);
THCudaTensor_copy(state, self_, t);
}
else
{
THArgCheck(THCudaTensor_nElement(state, self_) == THCudaTensor_nElement(state, src1),
1, "sizes do not match");
}
THArgCheck(THCudaTensor_nElement(state, src1) == THCudaTensor_nElement(state, src2),
3, "sizes do not match");
if (!THC_pointwiseApply3(state, self_, src1, src2, TensorAddCDivOp(value))) {
THArgCheck(false, 2, CUTORCH_DIM_WARNING);
}
THCudaCheck(cudaGetLastError());
}
template <typename T>
struct TensorFillOp {
TensorFillOp(T v) : val(v) {}
__device__ __forceinline__ void operator()(T* v) { *v = val; }
const T val;
};
#include "generic/THCTensorMath.cu"
#include "THCGenerateAllTypes.h"
|
