summaryrefslogtreecommitdiff
path: root/Blink/Drivers/CMSIS/NN/Include/arm_nnfunctions.h
blob: 96c59c21ba1eb95286269a35664cbe5d2f09aabe (plain)
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
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
/*
 * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
 *
 * SPDX-License-Identifier: Apache-2.0
 *
 * 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
 *
 * 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.
 */

/* ----------------------------------------------------------------------
 * Project:      CMSIS NN Library
 * Title:        arm_nnfunctions.h
 * Description:  Public header file for CMSIS NN Library
 *
 * $Date:        13. July 2018
 * $Revision:    V.1.0.0
 *
 * Target Processor:  Cortex-M cores
 * -------------------------------------------------------------------- */

/**
   \mainpage CMSIS NN Software Library
   *
   * Introduction
   * ------------
   *
   * This user manual describes the CMSIS NN software library,
   * a collection of efficient neural network kernels developed to maximize the 
   * performance and minimize the memory footprint of neural networks on Cortex-M processor cores.
   *
   * The library is divided into a number of functions each covering a specific category:
   * - Neural Network Convolution Functions
   * - Neural Network Activation Functions
   * - Fully-connected Layer Functions
   * - Neural Network Pooling Functions
   * - Softmax Functions
   * - Neural Network Support Functions
   *
   * The library has separate functions for operating on different weight and activation data
   * types including 8-bit integers (q7_t) and 16-bit integers (q15_t). The descrition of the
   * kernels are included in the function description. The implementation details are also 
   * described in this paper [1]. 
   *
   * Block Diagram
   * --------
   * \image html CMSIS-NN-OVERVIEW.PNG
   *
   * Examples
   * --------
   *
   * The library ships with a number of examples which demonstrate how to use the library functions.
   *
   * Pre-processor Macros
   * ------------
   *
   * Each library project have differant pre-processor macros.
   *
   * - ARM_MATH_DSP:
   *
   * Define macro ARM_MATH_DSP, If the silicon supports DSP instructions.
   *
   * - ARM_MATH_BIG_ENDIAN:
   *
   * Define macro ARM_MATH_BIG_ENDIAN to build the library for big endian targets. By default library builds for little endian targets.
   *
   * - ARM_NN_TRUNCATE:
   *
   * Define macro ARM_NN_TRUNCATE to use floor instead of round-to-the-nearest-int for the computation.
   *
   * Copyright Notice
   * ------------
   *
   * Copyright (C) 2010-2018 Arm Limited. All rights reserved.
   *
   * [1] CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs https://arxiv.org/abs/1801.06601
   */

/**
 * @defgroup groupNN Neural Network Functions
 * These functions perform basic operations for neural network layers. 
 */

#ifndef _ARM_NNFUNCTIONS_H
#define _ARM_NNFUNCTIONS_H

#include "arm_nnsupportfunctions.h"
#include "arm_nn_tables.h"

#define USE_INTRINSIC

//#define ARM_NN_TRUNCATE /* This config the rounding model to floor or round to the nearest int */

#ifdef __cplusplus
extern    "C"
{
#endif

/**
 * @defgroup NNConv Neural Network Convolution Functions
 *
 * Perform convolution layer
 *
 * The convolution is implemented in 2 steps: im2col and GEMM
 *
 * im2col is a process of converting each patch of image data into 
 * a column. After im2col, the convolution is computed as matrix-matrix
 * multiplication.
 * 
 * To reduce the memory footprint, the im2col is performed partially.
 * Each iteration, only a few column (i.e., patches) are generated and 
 * computed with GEMM kernels similar to CMSIS-DSP arm_mat_mult functions.
 *
 */

  /**
   * @brief Basic Q7 convolution function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns <code>ARM_MATH_SUCCESS</code> 
   *
   */

    arm_status arm_convolve_HWC_q7_basic(const q7_t * Im_in,
                                         const uint16_t dim_im_in,
                                         const uint16_t ch_im_in,
                                         const q7_t * wt,
                                         const uint16_t ch_im_out,
                                         const uint16_t dim_kernel,
                                         const uint16_t padding,
                                         const uint16_t stride,
                                         const q7_t * bias,
                                         const uint16_t bias_shift,
                                         const uint16_t out_shift,
                                         q7_t * Im_out, 
                                         const uint16_t dim_im_out, 
                                         q15_t * bufferA, 
                                         q7_t * bufferB);

  /**
   * @brief Basic Q7 convolution function (non-sqaure shape)
   * @param[in]       Im_in        pointer to input tensor
   * @param[in]       dim_im_in_x  input tensor dimention x
   * @param[in]       dim_im_in_y  input tensor dimention y
   * @param[in]       ch_im_in     number of input tensor channels
   * @param[in]       wt           pointer to kernel weights
   * @param[in]       ch_im_out    number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel_x filter kernel size x
   * @param[in]       dim_kernel_y filter kernel size y
   * @param[in]       padding_x    padding size x
   * @param[in]       padding_y    padding size y
   * @param[in]       stride_x     convolution stride x
   * @param[in]       stride_y     convolution stride y
   * @param[in]       bias         pointer to bias
   * @param[in]       bias_shift   amount of left-shift for bias
   * @param[in]       out_shift    amount of right-shift for output
   * @param[in,out]   Im_out       pointer to output tensor
   * @param[in]       dim_im_out_x output tensor dimension x
   * @param[in]       dim_im_out_y output tensor dimension y
   * @param[in,out]   bufferA      pointer to buffer space for input
   * @param[in,out]   bufferB      pointer to buffer space for output
   * @return     The function returns <code>ARM_MATH_SUCCESS</code> 
   */

    arm_status arm_convolve_HWC_q7_basic_nonsquare(const q7_t * Im_in,
                                                  const uint16_t dim_im_in_x,
                                                  const uint16_t dim_im_in_y,
                                                  const uint16_t ch_im_in,
                                                  const q7_t * wt,
                                                  const uint16_t ch_im_out,
                                                  const uint16_t dim_kernel_x,
                                                  const uint16_t dim_kernel_y,
                                                  const uint16_t padding_x,
                                                  const uint16_t padding_y,
                                                  const uint16_t stride_x,
                                                  const uint16_t stride_y,
                                                  const q7_t * bias,
                                                  const uint16_t bias_shift,
                                                  const uint16_t out_shift,
                                                  q7_t * Im_out,
                                                  const uint16_t dim_im_out_x,
                                                  const uint16_t dim_im_out_y,
                                                  q15_t * bufferA,
                                                  q7_t * bufferB);

  /**
   * @brief Basic Q15 convolution function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns <code>ARM_MATH_SUCCESS</code> 
   *
   */

    arm_status arm_convolve_HWC_q15_basic(const q15_t * Im_in,
                                          const uint16_t dim_im_in,
                                          const uint16_t ch_im_in,
                                          const q15_t * wt,
                                          const uint16_t ch_im_out,
                                          const uint16_t dim_kernel,
                                          const uint16_t padding,
                                          const uint16_t stride,
                                          const q15_t * bias,
                                          const uint16_t bias_shift,
                                          const uint16_t out_shift,
                                          q15_t * Im_out, 
                                          const uint16_t dim_im_out, 
                                          q15_t * bufferA, 
                                          q7_t * bufferB);

  /**
   * @brief Fast Q7 convolution function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 4
   *   ch_im_out is multiple of 2
   */

    arm_status arm_convolve_HWC_q7_fast(const q7_t * Im_in,
                                        const uint16_t dim_im_in,
                                        const uint16_t ch_im_in,
                                        const q7_t * wt,
                                        const uint16_t ch_im_out,
                                        const uint16_t dim_kernel,
                                        const uint16_t padding,
                                        const uint16_t stride,
                                        const q7_t * bias,
                                        const uint16_t bias_shift,
                                        const uint16_t out_shift,
                                        q7_t * Im_out, 
                                        const uint16_t dim_im_out, 
                                        q15_t * bufferA, 
                                        q7_t * bufferB);

  /**
   * @brief Fast Q7 convolution function (non-sqaure shape)
   * @param[in]       Im_in        pointer to input tensor
   * @param[in]       dim_im_in_x  input tensor dimention x
   * @param[in]       dim_im_in_y  input tensor dimention y
   * @param[in]       ch_im_in     number of input tensor channels
   * @param[in]       wt           pointer to kernel weights
   * @param[in]       ch_im_out    number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel_x filter kernel size x
   * @param[in]       dim_kernel_y filter kernel size y
   * @param[in]       padding_x    padding size x
   * @param[in]       padding_y    padding size y
   * @param[in]       stride_x     convolution stride x
   * @param[in]       stride_y     convolution stride y
   * @param[in]       bias         pointer to bias
   * @param[in]       bias_shift   amount of left-shift for bias
   * @param[in]       out_shift    amount of right-shift for output
   * @param[in,out]   Im_out       pointer to output tensor
   * @param[in]       dim_im_out_x output tensor dimension x
   * @param[in]       dim_im_out_y output tensor dimension y
   * @param[in,out]   bufferA      pointer to buffer space for input 
   * @param[in,out]   bufferB      pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 4
   *   ch_im_out is multiple of 2
   */

    arm_status arm_convolve_HWC_q7_fast_nonsquare(const q7_t * Im_in,
                                                  const uint16_t dim_im_in_x,
                                                  const uint16_t dim_im_in_y,
                                                  const uint16_t ch_im_in,
                                                  const q7_t * wt,
                                                  const uint16_t ch_im_out,
                                                  const uint16_t dim_kernel_x,
                                                  const uint16_t dim_kernel_y,
                                                  const uint16_t padding_x,
                                                  const uint16_t padding_y,
                                                  const uint16_t stride_x,
                                                  const uint16_t stride_y,
                                                  const q7_t * bias,
                                                  const uint16_t bias_shift,
                                                  const uint16_t out_shift,
                                                  q7_t * Im_out,
                                                  const uint16_t dim_im_out_x,
                                                  const uint16_t dim_im_out_y,
                                                  q15_t * bufferA,
                                                  q7_t * bufferB);

  /**
   * @brief Fast Q7 version of 1x1 convolution (non-sqaure shape)
   * @param[in]       Im_in        pointer to input tensor
   * @param[in]       dim_im_in_x  input tensor dimention x
   * @param[in]       dim_im_in_y  input tensor dimention y
   * @param[in]       ch_im_in     number of input tensor channels
   * @param[in]       wt           pointer to kernel weights
   * @param[in]       ch_im_out    number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel_x filter kernel size x
   * @param[in]       dim_kernel_y filter kernel size y
   * @param[in]       padding_x    padding size x
   * @param[in]       padding_y    padding size y
   * @param[in]       stride_x     convolution stride x
   * @param[in]       stride_y     convolution stride y
   * @param[in]       bias         pointer to bias
   * @param[in]       bias_shift   amount of left-shift for bias
   * @param[in]       out_shift    amount of right-shift for output
   * @param[in,out]   Im_out       pointer to output tensor
   * @param[in]       dim_im_out_x output tensor dimension x
   * @param[in]       dim_im_out_y output tensor dimension y
   * @param[in,out]   bufferA      pointer to buffer space for input 
   * @param[in,out]   bufferB      pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function implement convolution with 1x1 kernel size (i.e., dim_kernel_x=1
   * and dim_kernel_y=1). It can be used for
   * second half of MobileNets after depthwise separable convolution.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 4
   *   ch_im_out is multiple of 2
   */
    arm_status arm_convolve_1x1_HWC_q7_fast_nonsquare(const q7_t * Im_in,
                                                      const uint16_t dim_im_in_x,
                                                      const uint16_t dim_im_in_y,
                                                      const uint16_t ch_im_in,
                                                      const q7_t * wt,
                                                      const uint16_t ch_im_out,
                                                      const uint16_t dim_kernel_x,
                                                      const uint16_t dim_kernel_y,
                                                      const uint16_t padding_x,
                                                      const uint16_t padding_y,
                                                      const uint16_t stride_x,
                                                      const uint16_t stride_y,
                                                      const q7_t * bias,
                                                      const uint16_t bias_shift,
                                                      const uint16_t out_shift,
                                                      q7_t * Im_out,
                                                      const uint16_t dim_im_out_x,
                                                      const uint16_t dim_im_out_y,
                                                      q15_t * bufferA,
                                                      q7_t * bufferB);

  /**
   * @brief Q7 version of convolution for RGB image
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This kernel is written exclusively for convolution with ch_im_in
   * equals 3. This applies on the first layer of CNNs which has input
   * image with RGB format.
   */

    arm_status arm_convolve_HWC_q7_RGB(const q7_t * Im_in,
                                       const uint16_t dim_im_in,
                                       const uint16_t ch_im_in,
                                       const q7_t * wt,
                                       const uint16_t ch_im_out,
                                       const uint16_t dim_kernel,
                                       const uint16_t padding,
                                       const uint16_t stride,
                                       const q7_t * bias,
                                       const uint16_t bias_shift,
                                       const uint16_t out_shift,
                                       q7_t * Im_out, 
                                       const uint16_t dim_im_out, 
                                       q15_t * bufferA, 
                                       q7_t * bufferB);

  /**
   * @brief Fast Q15 convolution function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 2
   *   ch_im_out is multiple of 2
   */

    arm_status arm_convolve_HWC_q15_fast(const q15_t * Im_in,
                                         const uint16_t dim_im_in,
                                         const uint16_t ch_im_in,
                                         const q15_t * wt,
                                         const uint16_t ch_im_out,
                                         const uint16_t dim_kernel,
                                         const uint16_t padding,
                                         const uint16_t stride,
                                         const q15_t * bias,
                                         const uint16_t bias_shift,
                                         const uint16_t out_shift,
                                         q15_t * Im_out, 
                                         const uint16_t dim_im_out, 
                                         q15_t * bufferA, 
                                         q7_t * bufferB);

  /**
   * @brief Fast Q15 convolution function (non-sqaure shape)
   * @param[in]       Im_in        pointer to input tensor
   * @param[in]       dim_im_in_x  input tensor dimention x
   * @param[in]       dim_im_in_y  input tensor dimention y
   * @param[in]       ch_im_in     number of input tensor channels
   * @param[in]       wt           pointer to kernel weights
   * @param[in]       ch_im_out    number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel_x filter kernel size x
   * @param[in]       dim_kernel_y filter kernel size y
   * @param[in]       padding_x    padding size x
   * @param[in]       padding_y    padding size y
   * @param[in]       stride_x     convolution stride x
   * @param[in]       stride_y     convolution stride y
   * @param[in]       bias         pointer to bias
   * @param[in]       bias_shift   amount of left-shift for bias
   * @param[in]       out_shift    amount of right-shift for output
   * @param[in,out]   Im_out       pointer to output tensor
   * @param[in]       dim_im_out_x output tensor dimension x
   * @param[in]       dim_im_out_y output tensor dimension y
   * @param[in,out]   bufferA      pointer to buffer space for input 
   * @param[in,out]   bufferB      pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * @details
   *
   * <b>Buffer size:</b>
   *
   * bufferA size: 2*ch_im_in*dim_kernel*dim_kernel
   *
   * bufferB size: 0
   *
   * <b>Input dimension constraints:</b>
   *
   * ch_im_in is multiple of 2 
   *
   * ch_im_out is multipe of 2
   *
   */

    arm_status
    arm_convolve_HWC_q15_fast_nonsquare(const q15_t * Im_in,
                              const uint16_t dim_im_in_x,
                              const uint16_t dim_im_in_y,
                              const uint16_t ch_im_in,
                              const q15_t * wt,
                              const uint16_t ch_im_out,
                              const uint16_t dim_kernel_x,
                              const uint16_t dim_kernel_y,
                              const uint16_t padding_x,
                              const uint16_t padding_y,
                              const uint16_t stride_x,
                              const uint16_t stride_y,
                              const q15_t * bias,
                              const uint16_t bias_shift,
                              const uint16_t out_shift,
                              q15_t * Im_out,
                              const uint16_t dim_im_out_x,
                              const uint16_t dim_im_out_y, 
                              q15_t * bufferA, 
                              q7_t * bufferB);
										 
  /**
   * @brief Q7 depthwise separable convolution function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       wt          pointer to kernel weights
   * @param[in]       ch_im_out   number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       bias        pointer to bias
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in,out]   Im_out      pointer to output tensor
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input 
   * @param[in,out]   bufferB     pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 2
   *   ch_im_out is multiple of 2
   */

    arm_status arm_depthwise_separable_conv_HWC_q7(const q7_t * Im_in,
                                                   const uint16_t dim_im_in,
                                                   const uint16_t ch_im_in,
                                                   const q7_t * wt,
                                                   const uint16_t ch_im_out,
                                                   const uint16_t dim_kernel,
                                                   const uint16_t padding,
                                                   const uint16_t stride,
                                                   const q7_t * bias,
                                                   const uint16_t bias_shift,
                                                   const uint16_t out_shift,
                                                   q7_t * Im_out,
                                                   const uint16_t dim_im_out, 
                                                   q15_t * bufferA, 
                                                   q7_t * bufferB);

  /**
   * @brief Q7 depthwise separable convolution function (non-square shape)
   * @param[in]       Im_in         pointer to input tensor
   * @param[in]       dim_im_in_x   input tensor dimention x
   * @param[in]       dim_im_in_y   input tensor dimention y
   * @param[in]       ch_im_in      number of input tensor channels
   * @param[in]       wt            pointer to kernel weights
   * @param[in]       ch_im_out     number of filters, i.e., output tensor channels
   * @param[in]       dim_kernel_x  filter kernel size x
   * @param[in]       dim_kernel_y  filter kernel size y
   * @param[in]       padding_x     padding sizes x
   * @param[in]       padding_y     padding sizes y
   * @param[in]       stride_x      convolution stride x
   * @param[in]       stride_y      convolution stride y
   * @param[in]       bias          pointer to bias
   * @param[in]       bias_shift    amount of left-shift for bias
   * @param[in]       out_shift     amount of right-shift for output
   * @param[in,out]   Im_out        pointer to output tensor
   * @param[in]       dim_im_out_x  output tensor dimension x
   * @param[in]       dim_im_out_y  output tensor dimension y
   * @param[in,out]   bufferA       pointer to buffer space for input 
   * @param[in,out]   bufferB       pointer to buffer space for output
   * @return     The function returns either
   * <code>ARM_MATH_SIZE_MISMATCH</code> or <code>ARM_MATH_SUCCESS</code> based on the outcome of size checking.
   *
   * This function is the version with full list of optimization tricks, but with
   * some contraints:
   *   ch_im_in is multiple of 2
   *   ch_im_out is multiple of 2
   */
    arm_status arm_depthwise_separable_conv_HWC_q7_nonsquare(const q7_t * Im_in,
                                                             const uint16_t dim_im_in_x,
                                                             const uint16_t dim_im_in_y,
                                                             const uint16_t ch_im_in,
                                                             const q7_t * wt,
                                                             const uint16_t ch_im_out,
                                                             const uint16_t dim_kernel_x,
                                                             const uint16_t dim_kernel_y,
                                                             const uint16_t padding_x,
                                                             const uint16_t padding_y,
                                                             const uint16_t stride_x,
                                                             const uint16_t stride_y,
                                                             const q7_t * bias,
                                                             const uint16_t bias_shift,
                                                             const uint16_t out_shift,
                                                             q7_t * Im_out,
                                                             const uint16_t dim_im_out_x,
                                                             const uint16_t dim_im_out_y,
                                                             q15_t * bufferA,
                                                             q7_t * bufferB);


/**
 * @defgroup FC Fully-connected Layer Functions
 *
 * Perform fully-connected layer
 *
 * Fully-connected layer is basically a matrix-vector multiplication
 * with bias. The matrix is the weights and the input/output vectors
 * are the activation values. Supported {weight, activation} precisions
 * include {8-bit, 8-bit}, {16-bit, 16-bit}, and {8-bit, 16-bit}.
 *
 * Here we have two types of kernel functions. The basic function
 * implements the function using regular GEMV approach. The opt functions
 * operates with weights in interleaved formats. 
 *
 */

  /**
   * @brief Q7 basic fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_q7(const q7_t * pV,
                                      const q7_t * pM,
                                      const uint16_t dim_vec,
                                      const uint16_t num_of_rows,
                                      const uint16_t bias_shift,
                                      const uint16_t out_shift, 
                                      const q7_t * bias, 
                                      q7_t * pOut, 
                                      q15_t * vec_buffer);

  /**
   * @brief Q7 opt fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_q7_opt(const q7_t * pV,
                                          const q7_t * pM,
                                          const uint16_t dim_vec,
                                          const uint16_t num_of_rows,
                                          const uint16_t bias_shift,
                                          const uint16_t out_shift, 
                                          const q7_t * bias, 
                                          q7_t * pOut, 
                                          q15_t * vec_buffer);

  /**
   * @brief Q15 basic fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_q15(const q15_t * pV,
                                       const q15_t * pM,
                                       const uint16_t dim_vec,
                                       const uint16_t num_of_rows,
                                       const uint16_t bias_shift,
                                       const uint16_t out_shift, 
                                       const q15_t * bias, 
                                       q15_t * pOut, 
                                       q15_t * vec_buffer);

  /**
   * @brief Q15 opt fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_q15_opt(const q15_t * pV,
                                           const q15_t * pM,
                                           const uint16_t dim_vec,
                                           const uint16_t num_of_rows,
                                           const uint16_t bias_shift,
                                           const uint16_t out_shift,
                                           const q15_t * bias, 
                                           q15_t * pOut, 
                                           q15_t * vec_buffer);

  /**
   * @brief Mixed Q15-Q7 fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_mat_q7_vec_q15(const q15_t * pV,
                                                  const q7_t * pM,
                                                  const uint16_t dim_vec,
                                                  const uint16_t num_of_rows,
                                                  const uint16_t bias_shift,
                                                  const uint16_t out_shift,
                                                  const q7_t * bias, 
                                                  q15_t * pOut, 
                                                  q15_t * vec_buffer);

  /**
   * @brief Mixed Q15-Q7 opt fully-connected layer function
   * @param[in]       pV          pointer to input vector
   * @param[in]       pM          pointer to matrix weights
   * @param[in]       dim_vec     length of the vector
   * @param[in]       num_of_rows number of rows in weight matrix
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        pointer to bias
   * @param[in,out]   pOut        pointer to output vector
   * @param[in,out]   vec_buffer  pointer to buffer space for input
   * @return     The function returns <code>ARM_MATH_SUCCESS</code>
   *
   */

    arm_status arm_fully_connected_mat_q7_vec_q15_opt(const q15_t * pV,
                                                      const q7_t * pM,
                                                      const uint16_t dim_vec,
                                                      const uint16_t num_of_rows,
                                                      const uint16_t bias_shift,
                                                      const uint16_t out_shift,
                                                      const q7_t * bias, 
                                                      q15_t * pOut, 
                                                      q15_t * vec_buffer);

/**
 * @brief Matrix-Multiplication Kernels for Convolution
 *
 * These functions are used within convolution layer functions for 
 * matrix multiplication.
 * 
 * The implementation is similar to CMSIS-DSP arm_mat_mult functions
 * with one Q7 and one Q15 operands. The Q15 operand is the im2col
 * output which is always with 2 columns.
 *
 */

  /**
   * @brief Matrix-multiplication function for convolution
   * @param[in]       pA          pointer to operand A
   * @param[in]       pInBuffer   pointer to operand B, always conssists of 2 vectors
   * @param[in]       ch_im_out   numRow of A
   * @param[in]       numCol_A    numCol of A
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        the bias
   * @param[in,out]   pOut        pointer to output
   * @return     The function returns the incremented output pointer
   */

    q7_t     *arm_nn_mat_mult_kernel_q7_q15(const q7_t * pA,
                                            const q15_t * pInBuffer,
                                            const uint16_t ch_im_out,
                                            const uint16_t numCol_A,
                                            const uint16_t bias_shift,
                                            const uint16_t out_shift, 
                                            const q7_t * bias, 
                                            q7_t * pOut);

  /**
   * @brief Matrix-multiplication function for convolution with reordered columns
   * @param[in]       pA          pointer to operand A
   * @param[in]       pInBuffer   pointer to operand B, always conssists of 2 vectors
   * @param[in]       ch_im_out   numRow of A
   * @param[in]       numCol_A    numCol of A
   * @param[in]       bias_shift  amount of left-shift for bias
   * @param[in]       out_shift   amount of right-shift for output
   * @param[in]       bias        the bias
   * @param[in,out]   pOut        pointer to output
   * @return     The function returns the incremented output pointer
   */

    q7_t     *arm_nn_mat_mult_kernel_q7_q15_reordered(const q7_t * pA,
                                                      const q15_t * pInBuffer,
                                                      const uint16_t ch_im_out,
                                                      const uint16_t numCol_A,
                                                      const uint16_t bias_shift,
                                                      const uint16_t out_shift, 
                                                      const q7_t * bias, 
                                                      q7_t * pOut);

#ifdef __cplusplus
}
#endif

/*
 *  Other functions
 *  These layers are typically not timing critical
 *  Basic implementation is supported here
 */

#ifdef __cplusplus
extern    "C"
{
#endif

/**
 * @defgroup Acti Neural Network Activation Functions
 *
 * Perform activation layers, including ReLU (Rectified Linear Unit),
 * sigmoid and tanh
 *
 */

  /**
   * @brief Q7 RELU function
   * @param[in,out]   data        pointer to input
   * @param[in]       size        number of elements
   * @return none.
   */

    void      arm_relu_q7(q7_t * data, uint16_t size);

  /**
   * @brief Q15 RELU function
   * @param[in,out]   data        pointer to input
   * @param[in]       size        number of elements
   * @return none.
   */

    void      arm_relu_q15(q15_t * data, uint16_t size);

  /**
   * @brief Q7 neural network activation function using direct table look-up
   * @param[in,out]   data        pointer to input
   * @param[in]       size        number of elements
   * @param[in]       int_width   bit-width of the integer part, assume to be smaller than 3
   * @param[in]       type        type of activation functions
   * @return none.
   */

    void      arm_nn_activations_direct_q7(q7_t * data, uint16_t size, uint16_t int_width, 
                                           arm_nn_activation_type type);

  /**
   * @brief Q15 neural network activation function using direct table look-up
   * @param[in,out]   data        pointer to input
   * @param[in]       size        number of elements
   * @param[in]       int_width   bit-width of the integer part, assume to be smaller than 3
   * @param[in]       type        type of activation functions
   * @return none.
   */

    void      arm_nn_activations_direct_q15(q15_t * data, uint16_t size, uint16_t int_width,
                                            arm_nn_activation_type type);

/**
 * @defgroup Pooling Neural Network Pooling Functions
 *
 * Perform pooling functions, including max pooling and average pooling
 *
 */

  /**
   * @brief Q7 max pooling function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input
   * @param[in,out]   Im_out      pointer to output tensor
   * @return none.
   *
   */

    void      arm_maxpool_q7_HWC(q7_t * Im_in,
                                 const uint16_t dim_im_in,
                                 const uint16_t ch_im_in,
                                 const uint16_t dim_kernel,
                                 const uint16_t padding,
                                 const uint16_t stride, 
                                 const uint16_t dim_im_out, 
                                 q7_t * bufferA, 
                                 q7_t * Im_out);

  /**
   * @brief Q7 average pooling function
   * @param[in]       Im_in       pointer to input tensor
   * @param[in]       dim_im_in   input tensor dimention
   * @param[in]       ch_im_in    number of input tensor channels
   * @param[in]       dim_kernel  filter kernel size
   * @param[in]       padding     padding sizes
   * @param[in]       stride      convolution stride
   * @param[in]       dim_im_out  output tensor dimension
   * @param[in,out]   bufferA     pointer to buffer space for input
   * @param[in,out]   Im_out      pointer to output tensor
   * @return none.
   *
   */

    void      arm_avepool_q7_HWC(q7_t * Im_in,
                                 const uint16_t dim_im_in,
                                 const uint16_t ch_im_in,
                                 const uint16_t dim_kernel,
                                 const uint16_t padding,
                                 const uint16_t stride, 
                                 const uint16_t dim_im_out, 
                                 q7_t * bufferA, 
                                 q7_t * Im_out);

/**
 * @defgroup Softmax Softmax Functions
 *
 * EXP(2) based softmax function
 *
 */

  /**
   * @brief Q7 softmax function
   * @param[in]       vec_in      pointer to input vector
   * @param[in]       dim_vec     input vector dimention
   * @param[out]      p_out       pointer to output vector
   * @return none.
   *
   */

    void      arm_softmax_q7(const q7_t * vec_in, const uint16_t dim_vec, q7_t * p_out);

  /**
   * @brief Q15 softmax function
   * @param[in]       vec_in      pointer to input vector
   * @param[in]       dim_vec     input vector dimention
   * @param[out]      p_out       pointer to output vector
   * @return none.
   *
   */

    void      arm_softmax_q15(const q15_t * vec_in, const uint16_t dim_vec, q15_t * p_out);

#ifdef __cplusplus
}
#endif

#endif