Compute Library
 21.08
GEMMInterleaveBlocked.h
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2017-2018 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 #include "GEMM.h"
25 
26 #include "arm_compute/core/Types.h"
27 
28 namespace arm_compute
29 {
30 namespace test
31 {
32 namespace validation
33 {
34 namespace reference
35 {
36 template <typename T>
37 T safe_read(const SimpleTensor<T> &t, int y, int x)
38 {
39  const int stride = t.shape().x();
40  const int M = t.shape().y();
41  const int N = t.shape().x();
42  if((y < M) && (x < N))
43  {
44  return t[y * stride + x];
45  }
46  return 0;
47 }
48 
49 template <typename T>
50 SimpleTensor<T> gemm_interleave_blocked(const SimpleTensor<T> &in, SimpleTensor<T> &out, int int_by, int block, bool transposed)
51 {
52  const int M = out.shape().y();
53  const int N = out.shape().x();
54  for(int y = 0; y < M; y++)
55  {
56  T *out_ptr = &out[y * N];
57  for(int x = 0; x < (N / int_by); x += block)
58  {
59  for(int z = 0; z < int_by; z++)
60  {
61  for(int a = 0; (out_ptr <= &out[y * N + (N - 1)]) && a < block; a++)
62  {
63  if(!transposed)
64  *out_ptr++ = safe_read(in, (y * int_by) + z, x + a);
65  else
66  {
67  const T value = safe_read(in, x + a, (y * int_by) + z);
68  *out_ptr++ = value;
69  }
70  }
71  }
72  }
73  }
74  return out;
75 }
76 
77 template SimpleTensor<uint8_t> gemm_interleave_blocked(const SimpleTensor<uint8_t> &in, SimpleTensor<uint8_t> &out, int int_by, int block, bool transposed);
78 } // namespace reference
79 } // namespace validation
80 } // namespace test
81 } // namespace arm_compute
T safe_read(const SimpleTensor< T > &t, int y, int x)
unsigned int M
TensorShape shape() const override
Shape of the tensor.
Definition: SimpleTensor.h:320
SimpleTensor< T > gemm_interleave_blocked(const SimpleTensor< T > &in, SimpleTensor< T > &out, int int_by, int block, bool transposed)
Copyright (c) 2017-2021 Arm Limited.
unsigned int N
Simple tensor object that stores elements in a consecutive chunk of memory.
Definition: SimpleTensor.h:58