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
|
/******************************************************************************
* Copyright (c) Intel Corporation - All rights reserved. *
* This file is part of the LIBXSMM library. *
* *
* For information on the license, see the LICENSE file. *
* Further information: https://github.com/hfp/libxsmm/ *
* SPDX-License-Identifier: BSD-3-Clause *
******************************************************************************/
/* Alexander Heinecke, Evangelos Georganas, Hans Pabst,
Dhiraj Kalamkar, Ankush Mandal (Intel Corp.)
******************************************************************************/
#include <libxsmm.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#if defined(_OPENMP)
# include <omp.h>
#endif
#define USE_OVERWRITE
/* include c-based dnn library */
#include "../common/dnn_common.h"
#define CHKERR_LIBXSMM_DNN(A) { const int chkerr_libxsmm_dnn_ = A; if (LIBXSMM_DNN_SUCCESS != chkerr_libxsmm_dnn_) { \
fprintf(stderr, "%s\n", libxsmm_dnn_get_error(chkerr_libxsmm_dnn_)); global_status = chkerr_libxsmm_dnn_; } \
}
int main(int argc, char* argv[])
{
unsigned char *naive_input, *naive_input_tmp;
char *naive_filter;
int *naive_output_fp;
int *naive_libxsmm_output;
unsigned char *input_libxsmm;
char *filter_libxsmm;
int *output_libxsmm;
int ifhp, ifwp, ofhp, ofwp, ofh, ofw;
int stride_h, stride_w, pad_h, pad_w, pad_h_in, pad_w_in, pad_h_out, pad_w_out;
naive_conv_t naive_param;
void* scratch;
size_t scratch_size;
/* some parameters we can overwrite via cli,
default is some inner layer of overfeat */
int iters = 10; /* repetitions of benchmark */
int ifw = 14; /* input width, "W" */
int ifh = 18; /* input height, "H" */
int nImg = 32; /* mini-batch size, "N" */
int nIfm = 256; /* number of input feature maps, "C" */
int nOfm = 512; /* number of output feature maps, "K" */
int kh = 3; /* filter height, "R" */
int kw = 3; /* filter width, "S" */
int padh = 1; /* padding in input, height */
int padw = 1; /* padding in input, width */
int stride = 1; /* stride when accessing inputs */
char type = 'A'; /* 'A': ALL, 'F': FP, 'B': BP, 'U', WU */
char format = 'L';
const char *const env_check = getenv("CHECK");
const double check = LIBXSMM_ABS(0 == env_check ? 1 : atof(env_check));
#if defined(_OPENMP)
int nThreads = omp_get_max_threads(); /* number of threads */
#else
int nThreads = 1; /* number of threads */
#endif
int padding_mode = 0; /* padding mode */
unsigned long long l_start, l_end;
double l_total = 0.0;
double lpOps = 0.0; /* number of low precision operations */
int i;
libxsmm_dnn_conv_desc conv_desc;
libxsmm_dnn_layer* libxsmm_handle;
libxsmm_dnn_tensor* libxsmm_input;
libxsmm_dnn_tensor* libxsmm_output;
libxsmm_dnn_tensor* libxsmm_filter;
libxsmm_dnn_tensor_datalayout* libxsmm_layout;
libxsmm_dnn_err_t status;
libxsmm_dnn_err_t global_status = LIBXSMM_DNN_SUCCESS;
libxsmm_matdiff_info norms_fwd, diff;
libxsmm_matdiff_clear(&norms_fwd);
libxsmm_matdiff_clear(&diff);
if (argc > 1 && !strncmp(argv[1], "-h", 3)) {
printf("Usage: %s iters inpWidth inpHeight nImg nIfm nOfm kw kh pad stride type padding_mode\n", argv[0]);
return 0;
}
srand(1);
/* reading new values from cli */
i = 1;
if (argc > i) iters = atoi(argv[i++]);
if (argc > i) ifw = atoi(argv[i++]);
if (argc > i) ifh = atoi(argv[i++]);
if (argc > i) nImg = atoi(argv[i++]);
if (argc > i) nIfm = atoi(argv[i++]);
if (argc > i) nOfm = atoi(argv[i++]);
if (argc > i) kw = atoi(argv[i++]);
if (argc > i) kh = atoi(argv[i++]);
if (argc > i) padw = atoi(argv[i++]);
if (argc > i) padh = atoi(argv[i++]);
if (argc > i) stride = atoi(argv[i++]);
if (argc > i) type = *(argv[i++]);
if (argc > i) format = *(argv[i++]);
if (argc > i) padding_mode = atoi(argv[i++]);
if (type != 'A' && type != 'F' && type != 'B' && type != 'U') {
printf("type needs to be 'A' (All), 'F' (FP only), 'B' (BP only), 'U' (WU only)\n");
return 0;
}
if (format != 'L') {
printf("format needs to be 'L'\n");
return 0;
}
stride_w = stride;
stride_h = stride;
pad_w = padw;
pad_h = padh;
if (0 == padding_mode) {
pad_h_in = 0;
pad_w_in = 0;
pad_h_out = 0;
pad_w_out = 0;
}
else {
/* TODO: change "1" to "0" if "padding_mode = -1" is acknowledged */
if (1 < padding_mode) pad_w = padding_mode;
pad_h_in = pad_h;
pad_w_in = pad_w;
pad_h_out = pad_h;
pad_w_out = pad_w;
}
/* deriving some values for naive code */
ofh = (ifh + 2 * pad_h - kh) / stride_h + 1;
ofw = (ifw + 2 * pad_w - kw) / stride_w + 1;
ifhp = ifh + 2 * pad_h_in;
ifwp = ifw + 2 * pad_w_in;
ofhp = ofh + 2 * pad_h_out;
ofwp = ofw + 2 * pad_w_out;
/* set struct for naive convolution */
naive_param.nImg = nImg;
naive_param.nIfm = nIfm;
naive_param.nOfm = nOfm;
naive_param.ifhp = ifhp;
naive_param.ifwp = ifwp;
naive_param.ifh = ifh;
naive_param.ifw = ifw;
naive_param.ofhp = ofhp;
naive_param.ofwp = ofwp;
naive_param.ofh = ofh;
naive_param.ofw = ofw;
naive_param.pad_h = pad_h;
naive_param.pad_w = pad_w;
naive_param.pad_h_in = pad_h_in;
naive_param.pad_w_in = pad_w_in;
naive_param.pad_h_out = pad_h_out;
naive_param.pad_w_out = pad_w_out;
naive_param.kh = kh;
naive_param.kw = kw;
naive_param.stride_h = stride_h;
naive_param.stride_w = stride_w;
/* print some summary */
printf("##########################################\n");
printf("# Setting Up Common #\n");
printf("##########################################\n");
printf("PARAMS: W:%d H:%d N:%d C:%d K:%d R:%d S:%d P:%d Q:%d STRIDE:%d\n", ifw, ifh, nImg, nIfm, nOfm, kw, kh, ofh, ofw, stride);
printf("PARAMS: ITERS:%d", iters); if (LIBXSMM_FEQ(0, check)) printf(" Threads:%d\n", nThreads); else printf("\n");
printf(" InImg %dx%d Padded (%dx%d)\n", ifh, ifw, ifhp, ifwp);
printf("OutImg %dx%d Padded (%dx%d)\n", ofh, ofw, ofhp, ofwp);
printf("SIZE Input (MB): %10.2f MiB\n", (double)(nImg*nIfm*ifhp*ifwp*sizeof(unsigned char))/(1024.0*1024.0) );
printf("SIZE Output (MB): %10.2f MiB\n", (double)(nImg*nOfm*ofhp*ofwp*sizeof(int))/(1024.0*1024.0) );
printf("SIZE Input (1): %10.2f MiB\n", (double)(1*nIfm*ifhp*ifwp* sizeof(unsigned char))/(1024.0*1024.0) );
printf("SIZE Output (1): %10.2f MiB\n", (double)(1*nOfm*ofhp*ofwp* sizeof(int))/(1024.0*1024.0) );
printf("SIZE Weight : %10.2f MiB\n", (double)(nIfm*nOfm*kw*kh* sizeof(char))/(1024.0*1024.0) );
/* allocate data */
naive_input = (unsigned char*)libxsmm_aligned_malloc( nImg*nIfm*ifhp*ifwp*sizeof(unsigned char), 2097152);
naive_output_fp = (int* )libxsmm_aligned_malloc( nImg*nOfm*ofhp*ofwp*sizeof(int), 2097152);
naive_libxsmm_output = (int* )libxsmm_aligned_malloc( nImg*nOfm*ofhp*ofwp*sizeof(int), 2097152);
naive_filter = (char*)libxsmm_aligned_malloc( nOfm*nIfm*kh*kw* sizeof(char), 2097152);
input_libxsmm = (unsigned char*)libxsmm_aligned_malloc( nImg*nIfm*ifhp*ifwp*sizeof(unsigned char), 2097152);
filter_libxsmm = (char*)libxsmm_aligned_malloc( nOfm*nIfm*kh*kw* sizeof(char), 2097152);
output_libxsmm = (int*) libxsmm_aligned_malloc( nImg*nOfm*ofhp*ofwp*sizeof(int), 2097152);
/* initialize data */
naive_input_tmp = (unsigned char*)libxsmm_aligned_malloc( nImg*nIfm*ifhp*ifwp*sizeof(unsigned char), 2097152);
zero_buf_uint8(naive_input, nImg*nIfm*ifhp*ifwp);
if (padding_mode == 0 ) {
init_buf_uint8(naive_input, nImg*nIfm*ifhp*ifwp, 0, 0);
} else {
init_buf_uint8(naive_input_tmp, nImg*nIfm*ifh*ifw, 0, 0);
copy_internal_nchw_uint8( naive_input , naive_input_tmp, nImg, nIfm, ifh, ifw, pad_h, pad_w);
}
init_buf_int8(naive_filter, nOfm*nIfm*kh*kw, 0, 0);
zero_buf_int32(naive_output_fp, nImg*nOfm*ofhp*ofwp);
zero_buf_int32(output_libxsmm, nImg*nOfm*ofhp*ofwp);
zero_buf_int32(naive_libxsmm_output, nImg*nOfm*ofhp*ofwp);
if (LIBXSMM_NEQ(0, check)) {
printf("##########################################\n");
printf("# Computing Reference ... #\n");
printf("##########################################\n");
/* run naive convolutions */
if (type == 'A' || type == 'F') {
naive_conv_fp_int8int32(&naive_param, naive_input, naive_output_fp, naive_filter);
}
printf("##########################################\n");
printf("# Computing Reference ... done #\n");
printf("##########################################\n");
}
printf("\n");
printf("##########################################\n");
printf("# Setting Up (custom-Storage) #\n");
printf("##########################################\n");
/* setup LIBXSMM handle */
conv_desc.N = nImg;
conv_desc.C = nIfm;
conv_desc.H = ifh;
conv_desc.W = ifw;
conv_desc.K = nOfm;
conv_desc.R = kh;
conv_desc.S = kw;
conv_desc.u = stride_h;
conv_desc.v = stride_w;
conv_desc.pad_h = pad_h;
conv_desc.pad_w = pad_w;
conv_desc.pad_h_in = pad_h_in;
conv_desc.pad_w_in = pad_w_in;
conv_desc.pad_h_out = pad_h_out;
conv_desc.pad_w_out = pad_w_out;
conv_desc.threads = nThreads;
conv_desc.algo = LIBXSMM_DNN_CONV_ALGO_DIRECT;
conv_desc.buffer_format = LIBXSMM_DNN_TENSOR_FORMAT_LIBXSMM;
conv_desc.filter_format = LIBXSMM_DNN_TENSOR_FORMAT_LIBXSMM;
conv_desc.fuse_ops = LIBXSMM_DNN_CONV_FUSE_NONE;
conv_desc.options = LIBXSMM_DNN_CONV_OPTION_OVERWRITE;
conv_desc.datatype_in = LIBXSMM_DNN_DATATYPE_I8;
conv_desc.datatype_out = LIBXSMM_DNN_DATATYPE_I32;
libxsmm_handle = libxsmm_dnn_create_conv_layer( conv_desc, &status );
CHKERR_LIBXSMM_DNN( status );
/* setup LIBXSMM buffers and filter */
libxsmm_layout = libxsmm_dnn_create_tensor_datalayout( libxsmm_handle, LIBXSMM_DNN_INPUT, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_input = libxsmm_dnn_link_tensor( libxsmm_layout, input_libxsmm, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_dnn_destroy_tensor_datalayout( libxsmm_layout );
libxsmm_layout = libxsmm_dnn_create_tensor_datalayout( libxsmm_handle, LIBXSMM_DNN_OUTPUT, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_output = libxsmm_dnn_link_tensor( libxsmm_layout, output_libxsmm, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_dnn_destroy_tensor_datalayout( libxsmm_layout );
libxsmm_layout = libxsmm_dnn_create_tensor_datalayout( libxsmm_handle, LIBXSMM_DNN_FILTER, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_filter = libxsmm_dnn_link_tensor( libxsmm_layout, filter_libxsmm, &status ); CHKERR_LIBXSMM_DNN( status );
libxsmm_dnn_destroy_tensor_datalayout( libxsmm_layout );
/* copy in data to LIBXSMM format */
/* we can also use the layout functions and set the data on our
own external to the library, @TODO, we plan to add an example here */
CHKERR_LIBXSMM_DNN( libxsmm_dnn_copyin_tensor( libxsmm_input, (void*)naive_input, LIBXSMM_DNN_TENSOR_FORMAT_NCHW ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_zero_tensor( libxsmm_output ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_copyin_tensor( libxsmm_filter, (void*)naive_filter, LIBXSMM_DNN_TENSOR_FORMAT_KCRS ) );
/* bind buffers and filter to handle */
CHKERR_LIBXSMM_DNN( libxsmm_dnn_bind_tensor( libxsmm_handle, libxsmm_input, LIBXSMM_DNN_REGULAR_INPUT ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_bind_tensor( libxsmm_handle, libxsmm_output, LIBXSMM_DNN_REGULAR_OUTPUT ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_bind_tensor( libxsmm_handle, libxsmm_filter, LIBXSMM_DNN_REGULAR_FILTER ) );
/* let's allocate and bind scratch */
scratch_size = libxsmm_dnn_get_scratch_size( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_ALL, &status );
CHKERR_LIBXSMM_DNN( status );
scratch = libxsmm_aligned_scratch( scratch_size, 2097152 );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_bind_scratch( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_ALL, scratch ) );
/* set scratch to bogus to make sure that libxsmm takes care of zeroing internally */
if ((type == 'A' || type == 'F') && LIBXSMM_NEQ(0, check)) {
printf("##############################################\n");
printf("# Check Correctness - FWD (custom-Storage) #\n");
printf("##############################################\n");
/* run LIBXSMM convolutions */
#if defined(_OPENMP)
# pragma omp parallel
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
CHKERR_LIBXSMM_DNN( libxsmm_dnn_execute_st( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_FWD, 0, tid ) );
}
/* copy out data */
CHKERR_LIBXSMM_DNN( libxsmm_dnn_copyout_tensor( libxsmm_output, (void*)naive_libxsmm_output, LIBXSMM_DNN_TENSOR_FORMAT_NCHW ) );
/* compare */
libxsmm_matdiff(&norms_fwd, LIBXSMM_DATATYPE_I32, nImg*nOfm*ofhp*ofwp, 1, naive_output_fp, naive_libxsmm_output, 0, 0);
printf("L1 reference : %.25g\n", norms_fwd.l1_ref);
printf("L1 test : %.25g\n", norms_fwd.l1_tst);
printf("L2 abs.error : %.24f\n", norms_fwd.l2_abs);
printf("L2 rel.error : %.24f\n", norms_fwd.l2_rel);
printf("Linf abs.error: %.24f\n", norms_fwd.linf_abs);
printf("Linf rel.error: %.24f\n", norms_fwd.linf_rel);
printf("Check-norm : %.24f\n", norms_fwd.normf_rel);
libxsmm_matdiff_reduce(&diff, &norms_fwd);
}
if (type == 'A' || type == 'F') {
printf("##########################################\n");
printf("# Performance - FWD (custom-Storage) #\n");
printf("##########################################\n");
/* run LIBXSMM convolution for performance */
for (i = 0; i < 10; ++i) {
#if defined(_OPENMP)
# pragma omp parallel
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
libxsmm_dnn_execute_st( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_FWD, 0, tid );
}
}
l_start = libxsmm_timer_tick();
for (i = 0; i < iters; ++i) {
#if defined(_OPENMP)
# pragma omp parallel
#endif
{
#if defined(_OPENMP)
const int tid = omp_get_thread_num();
#else
const int tid = 0;
#endif
libxsmm_dnn_execute_st( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_FWD, 0, tid );
}
}
l_end = libxsmm_timer_tick();
l_total = libxsmm_timer_duration(l_start, l_end);
lpOps = (double)nImg * (double)nIfm * (double)nOfm * (double)ofh * (double)ofw * (double)(2 * kh * kw) * (double)iters;
printf("GOP = %.5g\n", lpOps*1e-9/(double)iters);
printf("fp time = %.5g\n", ((double)(l_total/iters)));
printf("GOPS = %.5g\n", (lpOps*1e-9)/l_total);
printf("PERFDUMP,FP,%s,%i,%i,%i,%i,%i,%i,%i,%i,%i,%i,%i,%.5g,%.5g,%f,%f,%f,%f,%f,%f,%f\n", LIBXSMM_VERSION, nThreads, nImg, nIfm, nOfm,
ifw, ifh, kw, kh, stride, padw, padh, ((double)(l_total/iters)), (lpOps*1e-9)/l_total, norms_fwd.l1_ref, norms_fwd.l1_tst,
norms_fwd.l2_abs, norms_fwd.l2_rel, norms_fwd.linf_abs, norms_fwd.linf_rel, norms_fwd.normf_rel);
}
/* clean-up */
CHKERR_LIBXSMM_DNN( libxsmm_dnn_release_scratch( libxsmm_handle, LIBXSMM_DNN_COMPUTE_KIND_ALL ) );
libxsmm_free(scratch);
CHKERR_LIBXSMM_DNN( libxsmm_dnn_release_tensor( libxsmm_handle, LIBXSMM_DNN_REGULAR_INPUT ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_release_tensor( libxsmm_handle, LIBXSMM_DNN_REGULAR_OUTPUT ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_release_tensor( libxsmm_handle, LIBXSMM_DNN_REGULAR_FILTER ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_destroy_tensor( libxsmm_input ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_destroy_tensor( libxsmm_output ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_destroy_tensor( libxsmm_filter ) );
CHKERR_LIBXSMM_DNN( libxsmm_dnn_destroy_conv_layer( libxsmm_handle ) );
/* deallocate data */
libxsmm_free(naive_input);
libxsmm_free(naive_output_fp);
libxsmm_free(naive_libxsmm_output);
libxsmm_free(naive_filter);
libxsmm_free(input_libxsmm);
libxsmm_free(output_libxsmm);
libxsmm_free(filter_libxsmm);
{ const char *const env_check_scale = getenv("CHECK_SCALE");
const double check_scale = LIBXSMM_ABS(0 == env_check_scale ? 1.0 : atof(env_check_scale));
if (LIBXSMM_NEQ(0, check) && (check < 100.0 * check_scale * diff.normf_rel) && (global_status == LIBXSMM_DNN_SUCCESS)) {
fprintf(stderr, "FAILED with an error of %f%%!\n", 100.0 * diff.normf_rel);
exit(EXIT_FAILURE);
}
}
/* some empty lines at the end */
printf("\n\n\n");
return global_status;
}
|
