-
Notifications
You must be signed in to change notification settings - Fork 0
/
convolution.cc
215 lines (204 loc) · 8.58 KB
/
convolution.cc
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
// Copyright 2023 The OpenXLA Authors.
//
// 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.
#include "xla/service/cpu/runtime/convolution.h"
#include <cstdint>
#include <memory>
#include <vector>
#include "absl/status/status.h"
#include "xla/executable_run_options.h"
#include "xla/service/cpu/runtime_conv2d.h"
#include "xla/service/cpu/runtime_conv3d.h"
namespace xla {
namespace cpu {
using ::xla::runtime::MemrefView;
absl::Status XlaConvolution::operator()(
const ExecutableRunOptions* run_options, MemrefView input,
MemrefView kernel, MemrefView output, int64_t inputBatchDimension,
absl::Span<const int64_t> inputSpatialDimensions,
int64_t inputFeatureDimension,
absl::Span<const int64_t> kernelSpatialDimensions,
int64_t kernelInputFeatureDimension, int64_t kernelOutputFeatureDimension,
absl::Span<const int64_t> outputSpatialDimensions,
absl::Span<const int64_t> window_strides, absl::Span<const int64_t> padding,
absl::Span<const int64_t> lhs_dilation,
absl::Span<const int64_t> rhs_dilation, int64_t feature_group_count) const {
auto size = inputSpatialDimensions.size();
if (size < 1 || size > 3) {
return absl::InvalidArgumentError(
"Only 1D, 2D and 3D convolutions are supported");
}
if (size != kernelSpatialDimensions.size() ||
size != outputSpatialDimensions.size() || size != window_strides.size() ||
size * 2 != padding.size() || size != lhs_dilation.size() ||
size != rhs_dilation.size()) {
return absl::InvalidArgumentError("Number of attributes mismatched");
}
// We lower 1D convolutions into calls to the same Eigen function as 2D
// convolutions, except that we pretend that the 1D convolution is really a 2D
// convolution with the missing dimension set to 1. We also adjust the
// padding, dilation parameters as needed.
std::vector<int64_t> input_dims;
std::vector<int64_t> kernel_dims;
std::vector<int64_t> output_dims;
std::vector<int64_t> strides;
std::vector<int64_t> pad;
std::vector<int64_t> base_dilation;
std::vector<int64_t> window_dilation;
if (size == 1) {
input_dims.push_back(1);
kernel_dims.push_back(1);
output_dims.push_back(1);
strides.push_back(1);
pad.insert(pad.end(), {0, 0});
base_dilation.push_back(1);
window_dilation.push_back(1);
}
for (auto dim : inputSpatialDimensions) {
input_dims.push_back(input.sizes[dim]);
}
for (auto dim : kernelSpatialDimensions) {
kernel_dims.push_back(kernel.sizes[dim]);
}
for (auto dim : outputSpatialDimensions) {
output_dims.push_back(output.sizes[dim]);
}
strides.insert(strides.end(), window_strides.begin(), window_strides.end());
pad.insert(pad.end(), padding.begin(), padding.end());
base_dilation.insert(base_dilation.end(), lhs_dilation.begin(),
lhs_dilation.end());
window_dilation.insert(window_dilation.end(), rhs_dilation.begin(),
rhs_dilation.end());
if (output.dtype == PrimitiveType::F16) {
auto* out = reinterpret_cast<Eigen::half*>(output.data);
auto* lhs = reinterpret_cast<Eigen::half*>(input.data);
auto* rhs = reinterpret_cast<Eigen::half*>(kernel.data);
if (size != 3) {
__xla_cpu_runtime_EigenConv2DF16(
run_options, out, lhs, rhs,
/*input_batch*/ input.sizes[inputBatchDimension],
/*input_rows*/ input_dims[0],
/*input_cols*/ input_dims[1],
/*input_channels*/ input.sizes[inputFeatureDimension],
/*kernel_rows*/ kernel_dims[0],
/*kernel_cols*/ kernel_dims[1],
/*kernel_channels*/ kernel.sizes[kernelInputFeatureDimension],
/*kernel_filters*/ kernel.sizes[kernelOutputFeatureDimension],
/*output_rows*/ output_dims[0],
/*output_cols*/ output_dims[1],
/*row_stride*/ strides[0],
/*col_stride*/ strides[1],
/*padding_top*/ pad[0],
/*padding_bottom*/ pad[1],
/*padding_left*/ pad[2],
/*padding_right*/ pad[3],
/*lhs_row_dilation*/ base_dilation[0],
/*lhs_col_dilation*/ base_dilation[1],
/*rhs_row_dilation*/ window_dilation[0],
/*rhs_col_dilation*/ window_dilation[1], feature_group_count);
} else {
__xla_cpu_runtime_EigenConv3DF16(
run_options, out, lhs, rhs,
/*input_batch*/ input.sizes[inputBatchDimension],
/*input_x*/ input_dims[0],
/*input_y*/ input_dims[1],
/*input_z*/ input_dims[2],
/*input_channels*/ input.sizes[inputFeatureDimension],
/*kernel_x*/ kernel_dims[0],
/*kernel_y*/ kernel_dims[1],
/*kernel_z*/ kernel_dims[2],
/*kernel_channels*/ kernel.sizes[kernelInputFeatureDimension],
/*kernel_filters*/ kernel.sizes[kernelOutputFeatureDimension],
/*output_x*/ output_dims[0],
/*output_y*/ output_dims[1],
/*output_z*/ output_dims[2],
/*x_stride*/ strides[0],
/*y_stride*/ strides[1],
/*z_stride*/ strides[2],
/*padding_x_before*/ pad[0],
/*padding_x_after*/ pad[1],
/*padding_y_before*/ pad[2],
/*padding_y_after*/ pad[3],
/*padding_z_before*/ pad[4],
/*padding_z_after*/ pad[5],
/*lhs_x_dilation*/ base_dilation[0],
/*lhs_y_dilation*/ base_dilation[1],
/*lhs_z_dilation*/ base_dilation[2],
/*rhs_x_dilation*/ window_dilation[0],
/*rhs_y_dilation*/ window_dilation[1],
/*rhs_z_dilation*/ window_dilation[2], feature_group_count);
}
} else {
auto* out = reinterpret_cast<float*>(output.data);
auto* lhs = reinterpret_cast<float*>(input.data);
auto* rhs = reinterpret_cast<float*>(kernel.data);
if (size != 3) {
__xla_cpu_runtime_EigenConv2DF32(
run_options, out, lhs, rhs,
/*input_batch*/ input.sizes[inputBatchDimension],
/*input_rows*/ input_dims[0],
/*input_cols*/ input_dims[1],
/*input_channels*/ input.sizes[inputFeatureDimension],
/*kernel_rows*/ kernel_dims[0],
/*kernel_cols*/ kernel_dims[1],
/*kernel_channels*/ kernel.sizes[kernelInputFeatureDimension],
/*kernel_filters*/ kernel.sizes[kernelOutputFeatureDimension],
/*output_rows*/ output_dims[0],
/*output_cols*/ output_dims[1],
/*row_stride*/ strides[0],
/*col_stride*/ strides[1],
/*padding_top*/ pad[0],
/*padding_bottom*/ pad[1],
/*padding_left*/ pad[2],
/*padding_right*/ pad[3],
/*lhs_row_dilation*/ base_dilation[0],
/*lhs_col_dilation*/ base_dilation[1],
/*rhs_row_dilation*/ window_dilation[0],
/*rhs_col_dilation*/ window_dilation[1], feature_group_count);
} else {
__xla_cpu_runtime_EigenConv3DF32(
run_options, out, lhs, rhs,
/*input_batch*/ input.sizes[inputBatchDimension],
/*input_x*/ input_dims[0],
/*input_y*/ input_dims[1],
/*input_z*/ input_dims[2],
/*input_channels*/ input.sizes[inputFeatureDimension],
/*kernel_x*/ kernel_dims[0],
/*kernel_y*/ kernel_dims[1],
/*kernel_z*/ kernel_dims[2],
/*kernel_channels*/ kernel.sizes[kernelInputFeatureDimension],
/*kernel_filters*/ kernel.sizes[kernelOutputFeatureDimension],
/*output_x*/ output_dims[0],
/*output_y*/ output_dims[1],
/*output_z*/ output_dims[2],
/*x_stride*/ strides[0],
/*y_stride*/ strides[1],
/*z_stride*/ strides[2],
/*padding_x_before*/ pad[0],
/*padding_x_after*/ pad[1],
/*padding_y_before*/ pad[2],
/*padding_y_after*/ pad[3],
/*padding_z_before*/ pad[4],
/*padding_z_after*/ pad[5],
/*lhs_x_dilation*/ base_dilation[0],
/*lhs_y_dilation*/ base_dilation[1],
/*lhs_z_dilation*/ base_dilation[2],
/*rhs_x_dilation*/ window_dilation[0],
/*rhs_y_dilation*/ window_dilation[1],
/*rhs_z_dilation*/ window_dilation[2], feature_group_count);
}
}
return absl::OkStatus();
}
} // namespace cpu
} // namespace xla