xref: /aosp_15_r20/external/XNNPACK/src/xx-transpose/1x1-memcpy.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Copyright 2021 Google LLC
2 //
3 // This source code is licensed under the BSD-style license found in the
4 // LICENSE file in the root directory of this source tree.
5 
6 #include <assert.h>
7 #include <string.h>
8 
9 #include <xnnpack/common.h>
10 #include <xnnpack/math.h>
11 #include <xnnpack/transpose.h>
12 
xnn_xx_transposev_ukernel__1x1_memcpy(const void * input,void * output,size_t input_row_stride,size_t output_row_stride,size_t input_element_stride,size_t output_element_stride,size_t element_size,size_t block_width,size_t block_height)13 void xnn_xx_transposev_ukernel__1x1_memcpy(
14     const void* input,
15     void* output,
16     size_t input_row_stride,
17     size_t output_row_stride,
18     size_t input_element_stride,
19     size_t output_element_stride,
20     size_t element_size,
21     size_t block_width,
22     size_t block_height)
23 {
24   const size_t input_reset = input_element_stride - block_height * input_row_stride;
25   const size_t output_reset = output_row_stride - block_height * output_element_stride;
26 
27   const void* i = (const void*) input;
28   void* o = (void*) output;
29 
30   do {
31     size_t bh = block_height;
32     for (; bh >= 1; bh -= 1) {
33       memcpy(o, i, element_size);
34       i = (const void*) ((uintptr_t) i + input_row_stride);
35       o = (void*) ((uintptr_t) o + output_element_stride);
36     }
37 
38     i = (const void*) ((uintptr_t) i + input_reset);
39     o = (void*) ((uintptr_t) o + output_reset);
40     block_width -= 1;
41   } while (block_width != 0);
42 }
43