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
|
/*
* Copyright (c) 2020
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* DNN native backend implementation.
*/
#include "dnn_backend_native.h"
#include "dnn_backend_native_layer_mathbinary.h"
typedef float (*FunType)(float src0, float src1);
static float sub(float src0, float src1)
{
return src0 - src1;
}
static float add(float src0, float src1)
{
return src0 + src1;
}
static float mul(float src0, float src1)
{
return src0 * src1;
}
static float realdiv(float src0, float src1)
{
return src0 / src1;
}
static float minimum(float src0, float src1)
{
return FFMIN(src0, src1);
}
static float floormod(float src0, float src1)
{
return (float)((int)(src0) % (int)(src1));
}
static void math_binary_commutative(FunType pfun, const DnnLayerMathBinaryParams *params, const DnnOperand *input, DnnOperand *output, DnnOperand *operands, const int32_t *input_operand_indexes)
{
int dims_count;
const float *src;
float *dst;
dims_count = ff_calculate_operand_dims_count(output);
src = input->data;
dst = output->data;
if (params->input0_broadcast || params->input1_broadcast) {
for (int i = 0; i < dims_count; ++i) {
dst[i] = pfun(params->v, src[i]);
}
} else {
const DnnOperand *input1 = &operands[input_operand_indexes[1]];
const float *src1 = input1->data;
for (int i = 0; i < dims_count; ++i) {
dst[i] = pfun(src[i], src1[i]);
}
}
}
static void math_binary_not_commutative(FunType pfun, const DnnLayerMathBinaryParams *params, const DnnOperand *input, DnnOperand *output, DnnOperand *operands, const int32_t *input_operand_indexes)
{
int dims_count;
const float *src;
float *dst;
dims_count = ff_calculate_operand_dims_count(output);
src = input->data;
dst = output->data;
if (params->input0_broadcast) {
for (int i = 0; i < dims_count; ++i) {
dst[i] = pfun(params->v, src[i]);
}
} else if (params->input1_broadcast) {
for (int i = 0; i < dims_count; ++i) {
dst[i] = pfun(src[i], params->v);
}
} else {
const DnnOperand *input1 = &operands[input_operand_indexes[1]];
const float *src1 = input1->data;
for (int i = 0; i < dims_count; ++i) {
dst[i] = pfun(src[i], src1[i]);
}
}
}
int ff_dnn_load_layer_math_binary(Layer *layer, AVIOContext *model_file_context, int file_size, int operands_num)
{
DnnLayerMathBinaryParams params = { 0 };
int dnn_size = 0;
int input_index = 0;
params.bin_op = (int32_t)avio_rl32(model_file_context);
dnn_size += 4;
params.input0_broadcast = (int32_t)avio_rl32(model_file_context);
dnn_size += 4;
if (params.input0_broadcast) {
params.v = av_int2float(avio_rl32(model_file_context));
} else {
layer->input_operand_indexes[input_index] = (int32_t)avio_rl32(model_file_context);
if (layer->input_operand_indexes[input_index] >= operands_num) {
return 0;
}
input_index++;
}
dnn_size += 4;
params.input1_broadcast = (int32_t)avio_rl32(model_file_context);
dnn_size += 4;
if (params.input1_broadcast) {
params.v = av_int2float(avio_rl32(model_file_context));
} else {
layer->input_operand_indexes[input_index] = (int32_t)avio_rl32(model_file_context);
if (layer->input_operand_indexes[input_index] >= operands_num) {
return 0;
}
input_index++;
}
dnn_size += 4;
layer->output_operand_index = (int32_t)avio_rl32(model_file_context);
dnn_size += 4;
if (layer->output_operand_index >= operands_num) {
return 0;
}
layer->params = av_memdup(¶ms, sizeof(params));
if (!layer->params)
return 0;
return dnn_size;
}
int ff_dnn_execute_layer_math_binary(DnnOperand *operands, const int32_t *input_operand_indexes,
int32_t output_operand_index, const void *parameters, NativeContext *ctx)
{
const DnnOperand *input = &operands[input_operand_indexes[0]];
DnnOperand *output = &operands[output_operand_index];
const DnnLayerMathBinaryParams *params = parameters;
for (int i = 0; i < 4; ++i)
output->dims[i] = input->dims[i];
output->data_type = input->data_type;
output->length = ff_calculate_operand_data_length(output);
if (output->length <= 0) {
av_log(ctx, AV_LOG_ERROR, "The output data length overflow\n");
return DNN_ERROR;
}
output->data = av_realloc(output->data, output->length);
if (!output->data) {
av_log(ctx, AV_LOG_ERROR, "Failed to reallocate memory for output\n");
return DNN_ERROR;
}
switch (params->bin_op) {
case DMBO_SUB:
math_binary_not_commutative(sub, params, input, output, operands, input_operand_indexes);
return 0;
case DMBO_ADD:
math_binary_commutative(add, params, input, output, operands, input_operand_indexes);
return 0;
case DMBO_MUL:
math_binary_commutative(mul, params, input, output, operands, input_operand_indexes);
return 0;
case DMBO_REALDIV:
math_binary_not_commutative(realdiv, params, input, output, operands, input_operand_indexes);
return 0;
case DMBO_MINIMUM:
math_binary_commutative(minimum, params, input, output, operands, input_operand_indexes);
return 0;
case DMBO_FLOORMOD:
math_binary_not_commutative(floormod, params, input, output, operands, input_operand_indexes);
return 0;
default:
av_log(ctx, AV_LOG_ERROR, "Unmatch math binary operator\n");
return DNN_ERROR;
}
}
|
