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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
|
/* Copyright 2016 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#if GOOGLE_CUDA
#define EIGEN_USE_GPU
#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
#include "tensorflow/core/kernels/ops_util.h"
#include "tensorflow/core/kernels/transpose_functor.h"
#include "tensorflow/core/util/cuda_kernel_helper.h"
// TODO(yangzihao): Remove the dependency of conv_2d.h once we move all
// GPU util functions and transpose kernels into separate files.
#include "tensorflow/core/kernels/conv_2d.h"
typedef Eigen::GpuDevice GPUDevice;
namespace tensorflow {
namespace internal {
template <typename T, bool conjugate>
__global__ void TransposeKernel(int nthreads, const T* src, const int32* buf,
const int32 ndims, T* dst) {
const int32* in_strides = buf;
const int32* out_strides = buf + ndims;
const int32* perm = buf + ndims * 2;
CUDA_1D_KERNEL_LOOP(o_idx, nthreads) {
int32 i_idx = 0;
int32 t = o_idx;
for (int32 i = 0; i < ndims; ++i) {
const int32 ratio = t / out_strides[i];
t -= ratio * out_strides[i];
i_idx += ratio * in_strides[perm[i]];
}
if (conjugate) {
dst[o_idx] = Eigen::numext::conj(ldg(src + i_idx));
} else {
dst[o_idx] = ldg(src + i_idx);
}
}
}
template <typename T, bool conjugate>
void TransposeSimple(const GPUDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
// Ensures we can use 32-bit index.
const int64 nelem = in.NumElements();
CHECK_LT(nelem, kint32max) << "Tensor too large to transpose on GPU";
// Pack strides and permutation into one buffer.
const int32 ndims = in.dims();
gtl::InlinedVector<int32, 24> host_buf(ndims * 3);
gtl::InlinedVector<int32, 8> in_strides = ComputeStride<int32>(in.shape());
gtl::InlinedVector<int32, 8> out_strides = ComputeStride<int32>(out->shape());
// Dimension permutation.
for (int i = 0; i < ndims; ++i) {
host_buf[i] = in_strides[i];
host_buf[ndims + i] = out_strides[i];
host_buf[ndims * 2 + i] = perm[i];
}
// Copies the input strides, output strides and permutation to the device.
auto num_bytes = sizeof(int64) * host_buf.size();
auto dev_buf = d.allocate(num_bytes);
// NOTE: host_buf is not allocated by CudaHostAllocator, and
// therefore we are doing a sync copy effectively.
d.memcpyHostToDevice(dev_buf, host_buf.data(), num_bytes);
// Launch kernel to q[...] = p[...].
const T* p = reinterpret_cast<const T*>(in.tensor_data().data());
T* q = reinterpret_cast<T*>(const_cast<char*>((out->tensor_data().data())));
CudaLaunchConfig cfg = GetCudaLaunchConfig(nelem, d);
TransposeKernel<T, conjugate>
<<<cfg.block_count, cfg.thread_per_block, 0, d.stream()>>>(
cfg.virtual_thread_count, p, reinterpret_cast<const int32*>(dev_buf),
ndims, q);
// Safe to deallocate immediately after the kernel launch.
d.deallocate(dev_buf);
}
// TransposeUsingTile tries to reduce the dimension of the input tensor to 3 and
// then call special kernels to swap either dimension 1 and dimension 2 or
// dimension 0 and dimension 2. It returns true if the operation is success,
// false otherwise.
template <typename T, bool conjugate = false>
struct TransposeUsingTile {
static bool run(const Eigen::GpuDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
// First try to reduce the dimensions of the input tensor.
TransposePermsVec new_perm;
TransposeDimsVec new_dims;
ReduceTransposeDimensions(in.shape(), perm, &new_perm, &new_dims);
// Only use special GPU kernel when dimension is 2 or 3.
int dims = new_dims.size();
if (dims < 2 || dims > 3) return false;
auto in_data = reinterpret_cast<const T*>(in.tensor_data().data());
auto out_data =
reinterpret_cast<T*>(const_cast<char*>(out->tensor_data().data()));
switch (dims) {
case 2:
if (new_perm[0] == 1 && new_perm[1] == 0) {
// Add the first dimension size as 1.
new_dims.insert(new_dims.begin(), 1);
tensorflow::functor::SwapDimension1And2InTensor3<GPUDevice, T,
conjugate>()(
d, in_data, new_dims, out_data);
return true;
}
break;
case 3:
if (new_perm == TransposePermsVec({0, 2, 1})) {
tensorflow::functor::SwapDimension1And2InTensor3<GPUDevice, T,
conjugate>()(
d, in_data, new_dims, out_data);
return true;
} else if (new_perm == TransposePermsVec({2, 1, 0})) {
tensorflow::functor::SwapDimension0And2InTensor3<GPUDevice, T,
conjugate>()(
d, in_data, new_dims, out_data);
return true;
} else {
// do not handle other 3D permutations
return false;
}
break;
default:
return false;
}
return false;
}
};
template <bool conjugate>
struct TransposeUsingTile<complex64, conjugate> {
static bool run(const Eigen::GpuDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
if (!conjugate) {
return TransposeUsingTile<uint64>::run(d, in, perm, out);
} else {
return TransposeUsingTile<float2, true>::run(d, in, perm, out);
}
}
};
template <bool conjugate>
struct TransposeUsingTile<complex128, conjugate> {
static bool run(const Eigen::GpuDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
if (!conjugate) {
return TransposeUsingTile<float4>::run(d, in, perm, out);
} else {
return TransposeUsingTile<double2, true>::run(d, in, perm, out);
}
}
};
} // namespace internal
// Transpose kernel specialized for GPU Device.
template <typename T, bool conjugate>
struct Transpose<GPUDevice, T, conjugate> {
static void run(const GPUDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
switch (in.dims()) {
case 2:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 2>(d, in, perm, conjugate,
out);
}
break;
case 3:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 3>(d, in, perm, conjugate,
out);
}
break;
case 4:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 4>(d, in, perm, conjugate,
out);
}
break;
case 5:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 5>(d, in, perm, conjugate,
out);
}
break;
case 6:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 6>(d, in, perm, conjugate,
out);
}
break;
case 7:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 7>(d, in, perm, conjugate,
out);
}
break;
case 8:
if (!internal::TransposeUsingTile<T, conjugate>::run(d, in, perm,
out)) {
internal::TransposeUsingEigen<GPUDevice, T, 8>(d, in, perm, conjugate,
out);
}
break;
default:
internal::TransposeSimple<T, conjugate>(d, in, perm, out);
break;
}
}
};
template <bool conjugate>
struct Transpose<GPUDevice, string, conjugate> {
static void run(const GPUDevice& d, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
LOG(FATAL) << "Transpose of DT_STRING tensor not supported on GPU.";
}
};
// Explicit instantiation.
template struct Transpose<GPUDevice, string, false>;
template <>
Status DoTranspose(const GPUDevice& device, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
return internal::DoTransposeImpl(device, in, perm, /*conjugate=*/false, out);
}
template <>
Status DoConjugateTranspose(const GPUDevice& device, const Tensor& in,
const gtl::ArraySlice<int32> perm, Tensor* out) {
return internal::DoTransposeImpl(device, in, perm, /*conjugate=*/true, out);
}
template <>
Status DoMatrixTranspose(const GPUDevice& device, const Tensor& in,
Tensor* out) {
return internal::DoMatrixTransposeImpl(device, in, /*conjugate=*/false, out);
}
template <>
Status DoConjugateMatrixTranspose(const GPUDevice& device, const Tensor& in,
Tensor* out) {
return internal::DoMatrixTransposeImpl(device, in, /*conjugate=*/true, out);
}
} // namespace tensorflow
#endif // GOOGLE_CUDA
|
