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 2016-2018 Intel Corporation
*
* 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 <assert.h>
#include <math.h>
#include "c_types_map.hpp"
#include "type_helpers.hpp"
#include "mkldnn_thread.hpp"
#include "simple_q10n.hpp"
#include "ref_batch_normalization.hpp"
namespace mkldnn {
namespace impl {
namespace cpu {
template <impl::data_type_t data_type>
void ref_batch_normalization_fwd_t<data_type>::execute_forward(
const exec_ctx_t &ctx) const {
/* fast return */
if (this->pd()->has_zero_dim_memory()) return;
auto src = CTX_IN_MEM(const data_t *, MKLDNN_ARG_SRC);
auto scaleshift = CTX_IN_MEM(const float *, MKLDNN_ARG_SCALE_SHIFT);
auto mean = pd()->stats_is_src()
? const_cast<float *>(CTX_IN_MEM(const float *, MKLDNN_ARG_MEAN))
: CTX_OUT_MEM(float *, MKLDNN_ARG_MEAN);
auto variance = pd()->stats_is_src()
? const_cast<float *>(CTX_IN_MEM(const float *, MKLDNN_ARG_VARIANCE))
: CTX_OUT_MEM(float *, MKLDNN_ARG_VARIANCE);
auto dst = CTX_OUT_MEM(data_t *, MKLDNN_ARG_DST);
auto ws = CTX_OUT_MEM(uint8_t *, MKLDNN_ARG_WORKSPACE);
const memory_desc_wrapper data_d(pd()->src_md());
const memory_desc_wrapper scaleshift_d(pd()->weights_md());
const dim_t N = pd()->MB();
const dim_t C = pd()->C();
dim_t H = 1, W = 1, D = 1;
const bool has_spatial = utils::one_of(data_d.ndims(), 4, 5);
if (has_spatial) {
D = pd()->D();
H = pd()->H();
W = pd()->W();
}
const float eps = pd()->desc()->batch_norm_epsilon;
const bool use_scaleshift = pd()->use_scaleshift();;
const bool save_stats = pd()->is_training();
const bool is_training = pd()->is_training();
const bool fuse_bn_relu = pd()->fuse_bn_relu();
const bool calculate_stats = !pd()->stats_is_src();
const bool with_relu = pd()->with_relu_post_op();
auto maybe_post_op = [&](float res) {
return (with_relu && res < 0.0f) ? 0.0f : res;
};
const bool is_3d = data_d.ndims() == 5;
auto data_offset = [&](const memory_desc_wrapper &data_d, dim_t n, dim_t c,
dim_t d, dim_t h, dim_t w) {
if (has_spatial) {
if (is_3d)
return data_d.off(n, c, d, h, w);
else
return data_d.off(n, c, h, w);
} else
return data_d.off(n, c);
};
parallel_nd(C, [&](dim_t c) {
float v_mean = calculate_stats ? 0 : mean[c];
float v_variance = calculate_stats ? 0 : variance[c];
if (calculate_stats) {
for (dim_t n = 0; n < N; ++n)
for (dim_t d = 0; d < D; ++d)
for (dim_t h = 0; h < H; ++h)
for (dim_t w = 0; w < W; ++w)
v_mean += src[data_offset(data_d, n, c, d, h, w)];
v_mean /= W*N*H*D;
for (dim_t n = 0; n < N; ++n)
for (dim_t d = 0; d < D; ++d)
for (dim_t h = 0; h < H; ++h)
for (dim_t w = 0; w < W; ++w) {
float m = src[data_offset(data_d, n, c, d, h, w)] - v_mean;
v_variance += m*m;
}
v_variance /= W*H*N*D;
}
float sqrt_variance = sqrtf(v_variance + eps);
float sm = (use_scaleshift
? scaleshift[scaleshift_d.off(0, c)]
: 1.0f) / sqrt_variance;
float sv = use_scaleshift ? scaleshift[scaleshift_d.off(1, c)] : 0;
for (dim_t n = 0; n < N; ++n)
for (dim_t d = 0; d < D; ++d)
for (dim_t h = 0; h < H; ++h)
for (dim_t w = 0; w < W; ++w) {
auto d_off = data_offset(data_d,n,c,d,h,w);
float bn_res = sm * ((float)src[d_off] - v_mean) + sv;
if (fuse_bn_relu) {
if (bn_res <= 0) {
bn_res = 0;
if (is_training)
ws[d_off] = 0;
} else {
if (is_training)
ws[d_off] = 1;
}
}
if (data_type == data_type::s8) {
dst[d_off] = qz_a1b0<float, data_t>()(maybe_post_op(bn_res));
} else {
dst[d_off] = static_cast<data_t>(maybe_post_op(bn_res));
}
}
if (calculate_stats) {
if (save_stats) {
mean[c] = v_mean;
variance[c] = v_variance;
}
}
});
}
template struct ref_batch_normalization_fwd_t<data_type::f32>;
template struct ref_batch_normalization_fwd_t<data_type::s8>;
template <impl::data_type_t data_type>
void ref_batch_normalization_bwd_t<data_type>::execute_backward(
const exec_ctx_t &ctx) const {
auto src = CTX_IN_MEM(const data_t *, MKLDNN_ARG_SRC);
auto mean = CTX_IN_MEM(const data_t *, MKLDNN_ARG_MEAN);
auto variance = CTX_IN_MEM(const data_t *, MKLDNN_ARG_VARIANCE);
auto diff_dst = CTX_IN_MEM(const data_t *, MKLDNN_ARG_DIFF_DST);
auto scaleshift = CTX_IN_MEM(const data_t *, MKLDNN_ARG_SCALE_SHIFT);
auto ws = CTX_IN_MEM(const uint8_t *, MKLDNN_ARG_WORKSPACE);
auto diff_src = CTX_OUT_MEM(data_t *, MKLDNN_ARG_DIFF_SRC);
auto diff_scaleshift = CTX_OUT_MEM(data_t *, MKLDNN_ARG_DIFF_SCALE_SHIFT);
const memory_desc_wrapper data_d(pd()->src_md());
const memory_desc_wrapper diff_data_d(pd()->diff_src_md());
const memory_desc_wrapper scaleshift_d(pd()->weights_md());
const memory_desc_wrapper diff_scaleshift_d(pd()->diff_weights_md());
const dim_t C = pd()->C();
/* fast return */
if (this->pd()->has_zero_dim_memory()) {
if (diff_scaleshift) {
for (dim_t c = 0; c < C; ++c) {
diff_scaleshift[diff_scaleshift_d.off(0, c)] = 0;
diff_scaleshift[diff_scaleshift_d.off(1, c)] = 0;
}
}
return;
}
const dim_t N = pd()->MB();
dim_t H = 1, W = 1, D = 1;
const bool has_spatial = utils::one_of(data_d.ndims(), 4, 5);
if (has_spatial) {
D = pd()->D();
H = pd()->H();
W = pd()->W();
}
const float eps = pd()->desc()->batch_norm_epsilon;
const bool use_scaleshift = pd()->use_scaleshift();
const bool calculate_diff_stats = !pd()->use_global_stats();
const bool fuse_bn_relu = pd()->fuse_bn_relu();
const bool is_3d = data_d.ndims() == 5;
auto data_offset = [&](const memory_desc_wrapper &data_d, dim_t n, dim_t c,
dim_t d, dim_t h, dim_t w) {
if (has_spatial) {
if (is_3d)
return data_d.off(n, c, d, h, w);
else
return data_d.off(n, c, h, w);
} else
return data_d.off(n, c);
};
parallel_nd(C, [&](dim_t c) {
data_t v_mean = mean[c];
data_t v_variance = variance[c];
data_t sqrt_variance = static_cast<data_t>(1.0f / sqrtf(v_variance + eps));
data_t gamma = use_scaleshift ? scaleshift[scaleshift_d.off(0, c)] : 1;
data_t diff_gamma = data_t(0);
data_t diff_beta = data_t(0);
diff_gamma = 0.0;
diff_beta = 0.0;
for (dim_t n = 0; n < N; ++n)
for (dim_t d = 0; d < D; ++d)
for (dim_t h = 0; h < H; ++h)
for (dim_t w = 0; w < W; ++w) {
const size_t s_off = data_offset(data_d, n, c, d, h, w);
data_t dd = diff_dst[data_offset(diff_data_d, n, c, d, h, w)];
if (fuse_bn_relu && !ws[s_off])
dd = 0;
diff_gamma += (src[s_off] - v_mean) * dd;
diff_beta += dd;
}
diff_gamma *= sqrt_variance;
if (diff_scaleshift) {
diff_scaleshift[diff_scaleshift_d.off(0, c)] = diff_gamma;
diff_scaleshift[diff_scaleshift_d.off(1, c)] = diff_beta;
}
for (dim_t n = 0; n < N; ++n)
for (dim_t d = 0; d < D; ++d)
for (dim_t h = 0; h < H; ++h)
for (dim_t w = 0; w < W; ++w) {
const size_t s_off = data_offset(data_d, n, c, d, h, w);
const size_t dd_off = data_offset(diff_data_d, n, c, d, h, w);
data_t dd = diff_dst[dd_off];
if (fuse_bn_relu && !ws[s_off])
dd = 0;
data_t v_diff_src = dd;
if (calculate_diff_stats) {
v_diff_src -= diff_beta/(D*W*H*N) +
(src[s_off] - v_mean) *
diff_gamma*sqrt_variance/(D*W*H*N);
}
v_diff_src *= gamma*sqrt_variance;
diff_src[dd_off] = v_diff_src;
}
});
}
template struct ref_batch_normalization_bwd_t<data_type::f32>;
}
}
}
// vim: et ts=4 sw=4 cindent cino^=l0,\:0,N-s
|