-
Notifications
You must be signed in to change notification settings - Fork 0
/
generate_cpp.py
446 lines (379 loc) · 19.2 KB
/
generate_cpp.py
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
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
#!/usr/bin/env python
__author__ = "C. Clayton Violand"
__copyright__ = "Copyright 2017"
## Generates .cpp files and Makefiles from template for model inference execution.
## REQUIRED: Caffe .prototxt in: 'models/<model_name>/..'
## Template as 'inference/helper/template.cpp'
##
import os
import re
import sys
import glob
from collections import OrderedDict
from shutil import copyfile
from helper.extract_architecture import extract_architecture
def main():
# Get models.
models = glob.glob('models/*[!.md]')
models = [os.path.basename(m) for m in models]
# For every model.
for m in models:
if os.path.exists("inference/main_%s.cpp" % m):
continue
# Create file from template.
infile = open("inference/helper/template.cpp", "r+").readlines()
# Get model architecture.
prototxt_dir = os.path.join("models", m)
for f in os.listdir(prototxt_dir):
if f.endswith('.prototxt'):
prototxt_file_path = os.path.join(prototxt_dir, f)
a = extract_architecture(prototxt_file_path)
# Line Position
pos = 35
# Triggers
first = True
first_conv = True
# Ticks
tick = 1
pool_tick = 1
fc_tick = 1
scale_tick = 1
counter = 1
# Previous info
last_output = ""
last_type = ""
second_last_output = ""
second_last_type = ""
to_write = []
lines = []
for i in a.keys()[1:]:
i = str(i)
i_type = a[i]['type']
current_index = a.keys().index(i)
if i_type == "convolution":
if "branch1" in last_output:
if not "pad" in a[i].keys():
a[i]['pad'] = 0
if not "stride" in a[i].keys():
a[i]['stride'] = 1
lines.extend([
"const int im_height_%d = im_height_%d;" % (counter, tick-1),
"const int im_width_%d = im_width_%d;" % (counter, tick-1),
"const int im_depth_%d = im_depth_%d;" % (counter, tick-1),
"const int im_size_%d = im_size_%d;" % (counter, tick-1),
""
])
lines.extend([
"const int k_num_%d = %d;" % (counter, a[i]['num_output']),
"const int k_size_%d = %d;" % (counter, a[i]['kernel_size'] * a[i]['kernel_size']),
"const int stride_%d = %d;" % (counter, a[i]['stride']),
"const int k_depth_%d = im_depth_%d;" % (counter, tick),
"",
"const int p1_%d = %d;" % (counter, a[i]['pad']),
"const int p2_%d = %d;" % (counter, a[i]['pad']),
"",
"const int output_height_%d = (im_height_%d + 2 * p1_%d - sqrt(k_size_%d)) / stride_%d + 1;" % (counter, counter, counter, counter, counter),
"const int output_width_%d = (im_width_%d + 2 * p2_%d - sqrt(k_size_%d)) / stride_%d + 1;" % (counter, counter, counter, counter, counter),
"const int output_size_%d = output_height_%d * output_width_%d;" % (counter, counter, counter),
""
])
lines.extend([
'MatrixXd %s_weights = load_csv_arma<MatrixXd>("../weights/%s/%s_weights.csv");' % (i, m, i),
"MatrixXd %s_w = %s_weights;" % (i, i),
"const float %s_min = %s_w.minCoeff();" % (i, i),
"const float %s_max = %s_w.maxCoeff();" % (i, i),
"",
'MatrixXd %s_biases = load_csv_arma<MatrixXd>("../weights/%s/%s_biases.csv");' % (i, m, i),
"VectorXd %s_b(Map<VectorXd>(%s_biases.data(), %s_biases.cols()*%s_biases.rows()));" % (i, i, i, i),
"",
'MatrixXd %s_result_params = load_csv_arma<MatrixXd>("../features/%s/caffe/result_params.csv");' % (i, m),
"const float %s_result_min = %s_result_params(%s, 0);" % (i, i, tick-1),
"const float %s_result_max = %s_result_params(%s, 1);" % (i, i, tick-1),
"",
])
else:
if not "pad" in a[i].keys():
a[i]['pad'] = 0
if not "stride" in a[i].keys():
a[i]['stride'] = 1
if first == True:
lines.extend([
"const int im_height_1 = %d;" % a['shape']['h'],
"const int im_width_1 = %d;" % a['shape']['w'],
"const int im_depth_1 = %d;" % a['shape']['d'],
"const int im_size_1 = im_height_1*im_width_1;",
""
])
elif second_last_type == "convolution" and last_type != "pooling" or second_last_type == "eltwise" and last_type != "pooling" or last_type == "convolution" or last_type == "eltwise" or last_type == "relu" and second_last_type != "pooling":
lines.extend([
"const int im_height_%d = output_height_%d;" % (counter, tick-1),
"const int im_width_%d = output_width_%d;" % (counter, tick-1),
"const int im_depth_%d = k_num_%d;" % (counter, tick-1),
"const int im_size_%d = im_height_%d * im_width_%d;" % (counter, counter, counter),
""
])
else:
lines.extend([
"const int im_height_%d = static_cast<int>(ceil(static_cast<float>(output_height_%d + 2 * pp1_%d - f_%d ) / s_%d)) + 1;" % (counter, tick-1, pool_tick-1, pool_tick-1, pool_tick-1),
"const int im_width_%d = static_cast<int>(ceil(static_cast<float>(output_width_%d + 2 * pp2_%d - f_%d) / s_%d)) + 1;" % (counter, tick-1, pool_tick-1, pool_tick-1, pool_tick-1),
"const int im_depth_%d = k_num_%d;" % (counter, tick-1),
"const int im_size_%d = im_height_%d * im_width_%d;" % (counter, counter, counter),
""
])
lines.extend([
"const int k_num_%d = %d;" % (counter, a[i]['num_output']),
"const int k_size_%d = %d;" % (counter, a[i]['kernel_size'] * a[i]['kernel_size']),
"const int stride_%d = %d;" % (counter, a[i]['stride']),
"const int k_depth_%d = im_depth_%d;" % (counter, tick),
"",
"const int p1_%d = %d;" % (counter, a[i]['pad']),
"const int p2_%d = %d;" % (counter, a[i]['pad']),
"",
"const int output_height_%d = (im_height_%d + 2 * p1_%d - sqrt(k_size_%d)) / stride_%d + 1;" % (counter, counter, counter, counter, counter),
"const int output_width_%d = (im_width_%d + 2 * p2_%d - sqrt(k_size_%d)) / stride_%d + 1;" % (counter, counter, counter, counter, counter),
"const int output_size_%d = output_height_%d * output_width_%d;" % (counter, counter, counter),
""
])
if first_conv == True:
lines.extend([
'MatrixXd %s_weights = load_csv_arma<MatrixXd>("../weights/%s/%s_weights.csv");' % (i, m, i),
"Map<MatrixXd> %s_w(%s_weights.data(), k_num_%d, k_size_%d * k_depth_%d);" % (i, i, tick, tick, tick),
"const float %s_min = %s_w.minCoeff();" % (i, i),
"const float %s_max = %s_w.maxCoeff();" % (i, i),
"",
'MatrixXd %s_result_params = load_csv_arma<MatrixXd>("../features/%s/caffe/result_params.csv");' % (i, m),
"const float %s_result_min = %s_result_params(%s, 0);" % (i, i, tick-1),
"const float %s_result_max = %s_result_params(%s, 1);" % (i, i, tick-1),
"",
'MatrixXd %s_biases = load_csv_arma<MatrixXd>("../weights/%s/%s_biases.csv");' % (i, m, i),
"VectorXd %s_b(Map<VectorXd>(%s_biases.data(), %s_biases.cols()*%s_biases.rows()));" % (i, i, i, i),
""
])
else:
lines.extend([
'MatrixXd %s_weights = load_csv_arma<MatrixXd>("../weights/%s/%s_weights.csv");' % (i, m, i),
"MatrixXd %s_w = %s_weights;" % (i, i),
"const float %s_min = %s_w.minCoeff();" % (i, i),
"const float %s_max = %s_w.maxCoeff();" % (i, i),
"",
'MatrixXd %s_result_params = load_csv_arma<MatrixXd>("../features/%s/caffe/result_params.csv");' % (i, m),
"const float %s_result_min = %s_result_params(%s, 0);" % (i, i, tick-1),
"const float %s_result_max = %s_result_params(%s, 1);" % (i, i, tick-1),
"",
'MatrixXd %s_biases = load_csv_arma<MatrixXd>("../weights/%s/%s_biases.csv");' % (i, m, i),
"VectorXd %s_b(Map<VectorXd>(%s_biases.data(), %s_biases.cols()*%s_biases.rows()));" % (i, i, i, i),
""
])
first_conv = False
counter += 1
tick += 1
elif i_type == "pooling":
if not "pad" in a[i].keys():
a[i]['pad'] = 0
if not "stride" in a[i].keys():
a[i]['stride'] = 1
if not "pool" in a[i].keys():
a[i]['pool'] = "max"
lines.extend([
"const int f_%i = %d;" % (pool_tick, a[i]['kernel_size']),
"const int s_%i = %d;" % (pool_tick, a[i]['stride']),
"std::string mode_%i = \"%s\";" % (pool_tick, a[i]['pool']),
""
])
lines.extend([
"const int pp1_%i = %d;" % (pool_tick, a[i]['pad']),
"const int pp2_%i = %d;" % (pool_tick, a[i]['pad']),
""
])
pool_tick += 1
elif i_type == "innerproduct":
lines.extend([
'MatrixXd %s_weights = load_csv_arma<MatrixXd>("../weights/%s/%s_weights.csv");' % (i, m, i),
"",
'MatrixXd %s_biases = load_csv_arma<MatrixXd>("../weights/%s/%s_biases.csv");' % (i, m, i),
"VectorXd %s_b(Map<VectorXd>(%s_biases.data(), %s_biases.cols()*%s_biases.rows()));" % (i, i, i, i),
""
])
fc_tick += 1
elif i_type == "scale":
lines.extend([
'MatrixXd %s_weights = load_csv_arma<MatrixXd>("../weights/%s/%s_weights.csv");' % (i, m, i),
"",
'MatrixXd %s_biases = load_csv_arma<MatrixXd>("../weights/%s/%s_biases.csv");' % (i, m, i),
"VectorXd %s_b(Map<VectorXd>(%s_biases.data(), %s_biases.cols()*%s_biases.rows()));" % (i, i, i, i),
""
])
scale_tick += 1
elif i_type == "batchnorm":
lines.extend([
'MatrixXd %s_mean = load_csv_arma<MatrixXd>("../weights/%s/%s_mean.csv");' % (i, m, i),
'MatrixXd %s_var = load_csv_arma<MatrixXd>("../weights/%s/%s_var.csv");' % (i, m, i),
""
])
scale_tick += 1
first = False
second_last_output = last_output
last_output = i
second_last_type = last_type
last_type = a[i]['type']
# Get input data path.
input_files = os.listdir("inputs/" + m + "/production/")
if (len(input_files) > 1):
for i in range(len(input_files)):
if input_files[i].endswith('.csv'):
input_file_path = "../inputs/" + m + "/production/" + input_files[i]
else:
continue
else:
input_file_path = "../inputs/" + m + "/production/" + input_files[0]
lines.extend([
#"const int im_num = %i;" % a['shape']['n'],
"const int im_num = 1000;",
"",
'ifstream infile;',
"infile.open(\"%s\");" % input_file_path,
""
])
for l in lines:
to_write.append((pos, "\t"+l+"\n"))
pos += 1
pos += 4
lines_input = []
lines_input.extend([
"MatrixXd line = load_csv<MatrixXd>(infile);",
""
])
for l in lines_input:
to_write.append((pos, "\t\t"+l+"\n"))
pos += 1
pos = pos + 7
tick = 1
pool_tick = 1
relu_tick = 1
fc_tick = 1
gemm_tick = 1
eltwise_tick = 1
scale_tick = 1
batchnorm_tick = 1
gemm_string = ""
offline_string = ""
last_output = ""
last_type = ""
second_last_output = ""
second_last_type = ""
lines_two = []
for i in a.keys():
current_index = a.keys().index(i)
if i == "shape":
a['shape']['type'] = "input"
if last_output == "shape":
last_output = "image"
i = str(i)
i_type = a[i]['type']
if i_type == "batchnorm":
lines_two.extend([
"MatrixXd %s = batchnorm(%s, %s_mean, %s_var);" % (i, last_output, i, i),
""
])
batchnorm_tick += 1
if "branch1" in i:
branch_input = second_last_output
if i_type == "scale":
lines_two.extend([
"MatrixXd %s = scale(%s, %s_weights, %s_b);" % (i, last_output, i, i),
""
])
scale_tick += 1
if i_type == "convolution":
if "branch1" in last_output and "scale" in last_output:
lines_two.extend([
"MatrixXd %s;" % i,
"float gemm_time_%i;" % tick,
"float offline_time_%i;" % tick,
"std::tie(%s, gemm_time_%i, offline_time_%i) = convolve(%s, im_size_%i, im_height_%i, im_width_%i, im_depth_%i, k_size_%i, stride_%i, %s_b, p1_%i, p2_%i, %s_w, output_size_%i, mode, %s_min, %s_max, %s_result_min, %s_result_max);" % (i, tick, tick, branch_input, tick, tick, tick, tick, tick, tick, i, tick, tick, i, tick, i, i, i, i),
""
])
else:
lines_two.extend([
"MatrixXd %s;" % i,
"float gemm_time_%i;" % tick,
"float offline_time_%i;" % tick,
"std::tie(%s, gemm_time_%i, offline_time_%i) = convolve(%s, im_size_%i, im_height_%i, im_width_%i, im_depth_%i, k_size_%i, stride_%i, %s_b, p1_%i, p2_%i, %s_w, output_size_%i, mode, %s_min, %s_max, %s_result_min, %s_result_max);" % (i, tick, tick, last_output, tick, tick, tick, tick, tick, tick, i, tick, tick, i, tick, i, i, i, i),
""
])
offline_string += " - offline_time_%s" % gemm_tick
gemm_string += " + gemm_time_%s" % gemm_tick
gemm_tick += 1
tick += 1
elif "branch1" in i and "scale" in i:
eltwise_input = i
elif i_type == "pooling":
lines_two.extend([
"MatrixXd %s = pool(%s, f_%i, s_%i, output_width_%i, output_height_%i, pp1_%i, pp2_%i, mode_%i);" % (i, last_output, pool_tick, pool_tick, tick-1, tick-1, pool_tick, pool_tick, pool_tick),
""
])
pool_tick += 1
elif i_type == "relu":
lines_two.extend([
"MatrixXd %s = relu(%s);" % (i, last_output),
""
])
relu_tick += 1
if last_type == "eltwise":
eltwise_input = i
elif i_type == "innerproduct":
lines_two.extend([
"MatrixXd %s = fully_connect(%s, %s.rows(), %s_weights, %s_b);" % (i, last_output, last_output, i, i),
""
])
fc_tick += 1
elif i_type == "eltwise":
lines_two.extend([
"MatrixXd %s = eltwise(%s, %s);" % (i, last_output, eltwise_input),
""
])
eltwise_tick += 1
second_last_output = last_output
last_output = i
second_last_type = last_type
last_type = a[i]['type']
for l in lines_two:
to_write.append((pos, "\t\t"+l+"\n"))
pos += 1
lines_three = []
lines_three.extend([
"run_time_total += (run_time%s);" % offline_string,
"gemm_time_total += 0.0%s;" % gemm_string,
""
])
name_tick = 1
last_item = a.keys()[-1]
for j in a.keys()[1:]:
j_type = a[j]['type']
if j == last_item:
lines_three.extend([
'std::string name_%i = "../features/%s/" + mode + "/%s_" + std::to_string(i) + ".csv";' % (name_tick, m, j),
'write_to_csv(name_%i, %s);' % (name_tick, j),
])
name_tick += 1
pos = pos + 3
for l in lines_three:
to_write.append((pos, "\t\t"+l+"\n"))
pos += 1
# Add the content.
for (index, string) in to_write:
infile.insert(index, string)
# Write to file.
outfile_path = os.path.join("inference", "main_"+ m +".cpp")
outfile = open(outfile_path, "w")
[outfile.write(l) for l in infile]
outfile.close()
# Makefile.
makefile_template = open("inference/helper/makefile_template", "r+").readlines()
makefile_template.insert(3, "SOURCES= main_%s.cpp helper/writer.cpp ../layers/convolution_layer/convolution.cpp ../layers/convolution_layer/lowp.cpp ../layers/pooling_layer/pooling.cpp ../layers/fully_connected_layer/fully_connected.cpp ../layers/relu_layer/relu.cpp ../layers/eltwise_layer/eltwise.cpp ../layers/scale_layer/scale.cpp ../layers/batchnorm_layer/batchnorm.cpp" % m)
makefile_path = os.path.join("inference", "Makefile." + m)
makefile = open(makefile_path, "w")
[makefile.write(l) for l in makefile_template]
makefile.close()
if __name__ == "__main__":
main()