Pablo Tello | 2fdc409 | 2017-11-23 15:50:08 +0000 | [diff] [blame] | 1 | /* |
Michele Di Giorgio | d9eaf61 | 2020-07-08 11:12:57 +0100 | [diff] [blame] | 2 | * Copyright (c) 2017-2018 Arm Limited. |
Pablo Tello | 2fdc409 | 2017-11-23 15:50:08 +0000 | [diff] [blame] | 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" |
Pablo Tello | 2fdc409 | 2017-11-23 15:50:08 +0000 | [diff] [blame] | 27 | |
| 28 | namespace arm_compute |
| 29 | { |
| 30 | namespace test |
| 31 | { |
| 32 | namespace validation |
| 33 | { |
| 34 | namespace reference |
| 35 | { |
| 36 | template <typename T> |
| 37 | SimpleTensor<T> gemm_interleave_4x4(const SimpleTensor<T> &in, SimpleTensor<T> &out) |
| 38 | { |
| 39 | const T *mtx_in = reinterpret_cast<const T *>(in.data()); |
| 40 | T *mtx_ref = reinterpret_cast<T *>(out.data()); |
| 41 | const int32_t in_rows = in.shape().y(); |
| 42 | const int32_t in_cols = in.shape().x(); |
| 43 | const int32_t out_stride = out.shape().x(); |
| 44 | int32_t y = 0; |
| 45 | for(; y <= (in_rows - 4); y += 4) |
| 46 | { |
| 47 | const T *in_ptr = &mtx_in[y * in_cols]; |
| 48 | |
| 49 | for(int32_t x = 0; x < in_cols; x++) |
| 50 | { |
| 51 | const T tmp[4] = { in_ptr[x + 0 * in_cols], |
| 52 | in_ptr[x + 1 * in_cols], |
| 53 | in_ptr[x + 2 * in_cols], |
| 54 | in_ptr[x + 3 * in_cols] |
| 55 | }; |
| 56 | |
| 57 | T *dst = &mtx_ref[static_cast<size_t>(x * 4.f) + static_cast<size_t>(std::ceil(y / 4.f)) * out_stride]; |
| 58 | memcpy(dst, tmp, sizeof(T) * 4); |
| 59 | } |
| 60 | } |
| 61 | |
| 62 | // Leftover along the Y direction |
| 63 | const int32_t leftover_y = in_rows - y; |
| 64 | |
| 65 | if(leftover_y != 0) |
| 66 | { |
| 67 | const T *in_ptr = &mtx_in[y * in_cols]; |
| 68 | |
| 69 | for(int32_t x = 0; x < in_cols; x++) |
| 70 | { |
| 71 | T tmp[4] = { 0, 0, 0, 0 }; |
| 72 | |
| 73 | for(int32_t k = 0; k < leftover_y; k++) |
| 74 | { |
| 75 | tmp[k] = in_ptr[k * in_cols + x]; |
| 76 | } |
| 77 | T *dst = &mtx_ref[static_cast<size_t>(x * 4.f) + static_cast<size_t>(std::ceil(y / 4.f)) * out_stride]; |
| 78 | memcpy(dst, tmp, sizeof(T) * 4); |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | return out; |
| 83 | } |
| 84 | |
| 85 | } // namespace reference |
| 86 | } // namespace validation |
| 87 | } // namespace test |
| 88 | } // namespace arm_compute |