Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 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 "helpers.h" |
| 25 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 26 | #if defined(DATA_TYPE) && defined(VEC_SIZE) && defined(NUM_GROUPS) && defined(K) && defined(SRC_DIM_Z) |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 27 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 28 | // Check valid VEC_SIZES |
| 29 | #if VEC_SIZE != 4 && VEC_SIZE != 8 && VEC_SIZE != 16 |
| 30 | #error "Only vector sizes 4, 8 and 16 are supported" |
| 31 | #endif // VEC_SIZE != 4 && VEC_SIZE != 8 && VEC_SIZE != 16 |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 32 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 33 | #define TYPE VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE) |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 34 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 35 | #define DIV_MOD_UINT(x, y, div_res, mod_res) \ |
| 36 | ({ \ |
| 37 | div_res = (uint)((x) * (float)(1.0f / (float)(y))); \ |
| 38 | uint r = div_res * (y); \ |
| 39 | mod_res = (x)-r; \ |
| 40 | }) |
| 41 | |
| 42 | /** Performs channel shuffle when the data layout is NCHW. See https://arxiv.org/pdf/1707.01083.pdf for details. |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 43 | * |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 44 | * @note The vector size must be given as a preprocessor argument using -DVEC_SIZE=num. e.g. -DVEC_SIZE=4 |
| 45 | * @note The depth of the tensor must be given as a preprocessor argument using -DSRC_DIM_Z=num. e.g. -DSRC_DIM_Z=64 |
| 46 | * @note The number of groups must be given as a preprocessor argument using -DNUM_GROUPS=num_groups. e.g. -DNUM_GROUPS=2 |
| 47 | * @note The number of channels in each group must be given as a preprocessor argument using -DK=num. e.g. -DK=1 |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 48 | * K is equal to num_channels / num_groups. |
| 49 | * |
Vidhya Sudhan Loganathan | 7485d5a | 2018-07-04 09:34:00 +0100 | [diff] [blame] | 50 | * @param[in] src_ptr Pointer to the source matrix. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 51 | * @param[in] src_stride_x Stride of the first source tensor in X dimension (in bytes) |
| 52 | * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) |
| 53 | * @param[in] src_stride_y Stride of the first source tensor in Y dimension (in bytes) |
| 54 | * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) |
| 55 | * @param[in] src_stride_z Stride of the first source tensor in Z dimension (in bytes) |
| 56 | * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 57 | * @param[in] src_stride_w Stride of the first source tensor in Z dimension (in bytes) |
| 58 | * @param[in] src_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 59 | * @param[in] src_offset_first_element_in_bytes The offset of the first element in the first source tensor |
| 60 | * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr |
| 61 | * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) |
| 62 | * @param[in] dst_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| 63 | * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) |
| 64 | * @param[in] dst_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| 65 | * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) |
| 66 | * @param[in] dst_step_z output_stride_z * number of elements along Z processed per workitem(in bytes) |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 67 | * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) |
| 68 | * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 69 | * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor |
| 70 | */ |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 71 | __kernel void channel_shuffle_nchw(TENSOR4D_DECLARATION(src), |
| 72 | TENSOR4D_DECLARATION(dst)) |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 73 | { |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 74 | uint curr_channel = 0; // channel id of input |
| 75 | uint batch_id = 0; // batch id |
| 76 | uint group_id = 0; // group id |
| 77 | uint channel_id = 0; // channel id within the group |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 78 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 79 | // Compute curr_channel and batch_id |
| 80 | DIV_MOD_UINT(get_global_id(2), SRC_DIM_Z, batch_id, curr_channel); |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 81 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 82 | // Compute group_id and channel_id |
| 83 | DIV_MOD_UINT(curr_channel, K, group_id, channel_id); |
| 84 | |
| 85 | const uint x = get_global_id(0) * VEC_SIZE; |
| 86 | const uint y = get_global_id(1) * 2; |
Michele Di Giorgio | efac7c6 | 2018-05-16 00:02:35 +0100 | [diff] [blame] | 87 | const uint z = channel_id * NUM_GROUPS + group_id; |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 88 | |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 89 | // Load the Nx2 block |
| 90 | const __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * src_stride_y + curr_channel * src_stride_z + batch_id * src_stride_w; |
| 91 | TYPE u0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(input_ptr + 0 * src_stride_y)); |
| 92 | TYPE u1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(input_ptr + 1 * src_stride_y)); |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 93 | |
| 94 | // Store blocks |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 95 | __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * dst_stride_y + z * dst_stride_z + batch_id * dst_stride_w; |
| 96 | VSTORE(VEC_SIZE) |
| 97 | (u0, 0, (__global DATA_TYPE *)(output_ptr + 0 * dst_stride_y)); |
| 98 | VSTORE(VEC_SIZE) |
| 99 | (u1, 0, (__global DATA_TYPE *)(output_ptr + 1 * dst_stride_y)); |
Michele Di Giorgio | 7217563 | 2018-05-01 16:52:00 +0100 | [diff] [blame] | 100 | } |
Gian Marco Iodice | 8bab0ee | 2018-09-13 11:51:56 +0100 | [diff] [blame] | 101 | |
| 102 | #if VEC_SIZE == 4 && defined(LAST_ACCESSED) |
| 103 | /** Performs channel shuffle when the data layout is NHWC. See https://arxiv.org/pdf/1707.01083.pdf for details. |
| 104 | * |
| 105 | * @note This implementation is only defined for VEC_SIZE = 4 |
| 106 | * @note This last element accessed along the first dimension must be given as a preprocessor argument using -DLAST_ACCESSED=num. e.g. -DLAST_ACCESSED=64 in order to prevent out-of-bound writes. |
| 107 | * @note The vector size must be given as a preprocessor argument using -DVEC_SIZE=num. e.g. -DVEC_SIZE=4 |
| 108 | * @note The height of the tensor must be given as a preprocessor argument using -DSRC_DIM_Z=num. e.g. -DSRC_DIM_Z=64 |
| 109 | * @note The number of groups must be given as a preprocessor argument using -DNUM_GROUPS=num_groups. e.g. -DNUM_GROUPS=2 |
| 110 | * @note The number of channels in each group must be given as a preprocessor argument using -DK=num. e.g. -DK=1 |
| 111 | * K is equal to num_channels / num_groups. |
| 112 | * |
| 113 | * @param[in] src_ptr Pointer to the source matrix. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 |
| 114 | * @param[in] src_stride_x Stride of the first source tensor in X dimension (in bytes) |
| 115 | * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) |
| 116 | * @param[in] src_stride_y Stride of the first source tensor in Y dimension (in bytes) |
| 117 | * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) |
| 118 | * @param[in] src_stride_z Stride of the first source tensor in Z dimension (in bytes) |
| 119 | * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) |
| 120 | * @param[in] src_stride_w Stride of the first source tensor in Z dimension (in bytes) |
| 121 | * @param[in] src_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) |
| 122 | * @param[in] src_offset_first_element_in_bytes The offset of the first element in the first source tensor |
| 123 | * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr |
| 124 | * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) |
| 125 | * @param[in] dst_step_x output_stride_x * number of elements along X processed per workitem(in bytes) |
| 126 | * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) |
| 127 | * @param[in] dst_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) |
| 128 | * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) |
| 129 | * @param[in] dst_step_z output_stride_z * number of elements along Z processed per workitem(in bytes) |
| 130 | * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) |
| 131 | * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) |
| 132 | * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor |
| 133 | */ |
| 134 | __kernel void channel_shuffle_nhwc(TENSOR4D_DECLARATION(src), |
| 135 | TENSOR4D_DECLARATION(dst)) |
| 136 | { |
| 137 | const uint curr_channel = min((uint)(get_global_id(0) * VEC_SIZE), (uint)LAST_ACCESSED); // input feature map |
| 138 | uint channel_id0 = 0; |
| 139 | uint channel_id1 = 0; |
| 140 | uint channel_id2 = 0; |
| 141 | uint channel_id3 = 0; |
| 142 | uint group_id0 = 0; |
| 143 | uint group_id1 = 0; |
| 144 | uint group_id2 = 0; |
| 145 | uint group_id3 = 0; |
| 146 | uint y = 0; |
| 147 | uint batch_id = 0; |
| 148 | |
| 149 | // Compute curr_channel and batch_id |
| 150 | DIV_MOD_UINT(get_global_id(2), (uint)SRC_DIM_Z, batch_id, y); |
| 151 | |
| 152 | // Compute group_id and channel_id |
| 153 | DIV_MOD_UINT(curr_channel + (uint)0, K, group_id0, channel_id0); |
| 154 | DIV_MOD_UINT(curr_channel + (uint)1, K, group_id1, channel_id1); |
| 155 | DIV_MOD_UINT(curr_channel + (uint)2, K, group_id2, channel_id2); |
| 156 | DIV_MOD_UINT(curr_channel + (uint)3, K, group_id3, channel_id3); |
| 157 | |
| 158 | const uint x = get_global_id(1) * 2; |
| 159 | const uint z0 = channel_id0 * (uint)NUM_GROUPS + group_id0; |
| 160 | const uint z1 = channel_id1 * (uint)NUM_GROUPS + group_id1; |
| 161 | const uint z2 = channel_id2 * (uint)NUM_GROUPS + group_id2; |
| 162 | const uint z3 = channel_id3 * (uint)NUM_GROUPS + group_id3; |
| 163 | |
| 164 | // Load the Nx2 block |
| 165 | const __global uchar *input_ptr = src_ptr + src_offset_first_element_in_bytes + curr_channel * sizeof(DATA_TYPE) + x * src_stride_y + y * src_stride_z + batch_id * src_stride_w; |
| 166 | TYPE u0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(input_ptr + 0 * src_stride_y)); |
| 167 | TYPE u1 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)(input_ptr + 1 * src_stride_y)); |
| 168 | |
| 169 | // Store blocks |
| 170 | __global uchar *output_ptr = dst_ptr + dst_offset_first_element_in_bytes + x * dst_stride_y + y * dst_stride_z + batch_id * dst_stride_w; |
| 171 | *((__global DATA_TYPE *)(output_ptr + (uint)0 * dst_stride_y + z0 * sizeof(DATA_TYPE))) = u0.s0; |
| 172 | *((__global DATA_TYPE *)(output_ptr + (uint)0 * dst_stride_y + z1 * sizeof(DATA_TYPE))) = u0.s1; |
| 173 | *((__global DATA_TYPE *)(output_ptr + (uint)0 * dst_stride_y + z2 * sizeof(DATA_TYPE))) = u0.s2; |
| 174 | *((__global DATA_TYPE *)(output_ptr + (uint)0 * dst_stride_y + z3 * sizeof(DATA_TYPE))) = u0.s3; |
| 175 | *((__global DATA_TYPE *)(output_ptr + (uint)1 * dst_stride_y + z0 * sizeof(DATA_TYPE))) = u1.s0; |
| 176 | *((__global DATA_TYPE *)(output_ptr + (uint)1 * dst_stride_y + z1 * sizeof(DATA_TYPE))) = u1.s1; |
| 177 | *((__global DATA_TYPE *)(output_ptr + (uint)1 * dst_stride_y + z2 * sizeof(DATA_TYPE))) = u1.s2; |
| 178 | *((__global DATA_TYPE *)(output_ptr + (uint)1 * dst_stride_y + z3 * sizeof(DATA_TYPE))) = u1.s3; |
| 179 | } |
| 180 | #endif // VEC_SIZE == 4 && defined(LAST_ACCESSED) |
| 181 | #endif // defined(DATA_TYPE) && defined(VEC_SIZE) && defined(NUM_GROUPS) && defined(K) && defined(SRC_DIM_Z) |