xref: /aosp_15_r20/external/ComputeLibrary/tests/validation/reference/GEMMTranspose1xW.h (revision c217d954acce2dbc11938adb493fc0abd69584f3)
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>
gemm_transpose_1xW(const SimpleTensor<T> & in)37 SimpleTensor<T> gemm_transpose_1xW(const SimpleTensor<T> &in)
38 {
39     const int         W = 16 / sizeof(T);
40     const TensorShape shape_out(static_cast<size_t>(in.shape().y() * W), static_cast<size_t>(std::ceil(in.shape().x() / static_cast<float>(W))));
41     SimpleTensor<T>   out(shape_out, in.data_type());
42     const int32_t     in_height     = in.shape().y();
43     const int32_t     in_width      = in.shape().x();
44     const int32_t     out_width     = out.shape().x();
45     const T          *in_base_addr  = reinterpret_cast<const T *>(in.data());
46     T                *out_base_addr = reinterpret_cast<T *>(out.data());
47     int               x             = 0;
48     for(; x < in_width; x += W)
49     {
50         for(int y = 0; y < in_height; y++)
51         {
52             const T *in_addr  = (in_base_addr + x + y * in_width);
53             T       *out_addr = (out_base_addr + y * W + (x / W) * out_width);
54 
55             for(int k = 0; k < W; ++k)
56             {
57                 // If the input width is not multiple of W, we fill the reference with 0s
58                 if((x + k) >= in_width)
59                 {
60                     out_addr[k] = T(0);
61                 }
62                 else
63                 {
64                     out_addr[k] = in_addr[k];
65                 }
66             }
67         }
68     }
69     return out;
70 }
71 
72 } // namespace reference
73 } // namespace validation
74 } // namespace test
75 } // namespace arm_compute
76