1/* 2 * Copyright (c) 2017-2022 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 26#if defined(VEC_SIZE) 27#define VEC_INT VEC_DATA_TYPE(int, VEC_SIZE) 28 29#if defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) 30#define VEC_FLOAT VEC_DATA_TYPE(float, VEC_SIZE) 31#define VEC_QUANT VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE) 32#define CONVERT_RTE(x, type) (convert_##type##_rte((x))) 33#define CONVERT_DOWN(x, type) CONVERT_RTE(x, type) 34inline VEC_QUANT requantize(VEC_QUANT input, float in_offset, float out_offset, float in_scale, float out_scale) 35{ 36 const VEC_FLOAT in_f32 = (CONVERT(input, VEC_FLOAT) - (VEC_FLOAT)((float)in_offset)) * (VEC_FLOAT)((float)in_scale); 37 const VEC_FLOAT out_f32 = in_f32 / ((VEC_FLOAT)(float)out_scale) + ((VEC_FLOAT)((float)out_offset)); 38 const VEC_QUANT res_q8 = CONVERT_SAT(CONVERT_DOWN(out_f32, VEC_INT), VEC_QUANT); 39 return res_q8; 40} 41#endif /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 42 43#if defined(DATA_TYPE) 44#define VEC_TYPE VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE) 45 46#if defined(ELEMENT_SIZE) 47 48#define SELECT_TYPE SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE) 49#define SEQ VEC_OFFS(int, VEC_SIZE) 50 51#if defined(CONCATENATE_WIDTH_X2) 52/** This kernel concatenates two input tensors into the output tensor along the first dimension 53 * 54 * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float 55 * @note Vector size has to be passed at compile time using -DVEC_SIZE. i.e. -DVEC_SIZE=16 56 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE_LEFTOVER=3. It is defined as the remainder between the input's first dimension and VEC_SIZE 57 * 58 * @param[in] src1_ptr Pointer to the source tensor. Supported data types: All. 59 * @param[in] src1_stride_x Stride of the source tensor in X dimension (in bytes) 60 * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 61 * @param[in] src1_stride_y Stride of the source tensor in Y dimension (in bytes) 62 * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 63 * @param[in] src1_stride_z Stride of the source tensor in Z dimension (in bytes) 64 * @param[in] src1_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 65 * @param[in] src1_stride_w Stride of the first source tensor in Z dimension (in bytes) 66 * @param[in] src1_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 67 * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source tensor 68 * @param[in] src2_ptr Pointer to the source tensor. Supported data types: same as @p src1_ptr 69 * @param[in] src2_stride_x Stride of the source tensor in X dimension (in bytes) 70 * @param[in] src2_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 71 * @param[in] src2_stride_y Stride of the source tensor in Y dimension (in bytes) 72 * @param[in] src2_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 73 * @param[in] src2_stride_z Stride of the source tensor in Z dimension (in bytes) 74 * @param[in] src2_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 75 * @param[in] src2_stride_w Stride of the first source tensor in Z dimension (in bytes) 76 * @param[in] src2_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 77 * @param[in] src2_offset_first_element_in_bytes The offset of the first element in the source tensor 78 * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src1_ptr 79 * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) 80 * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) 81 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) 82 * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) 83 * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes) 84 * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) 85 * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) 86 * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) 87 * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor 88 * @param[in] DEPTH Tensor depth 89 * @param[in] INPUT1_WIDTH First input tensor width 90 */ 91__kernel void concatenate_width_x2( 92 TENSOR4D_DECLARATION(src1), 93 TENSOR4D_DECLARATION(src2), 94 TENSOR4D_DECLARATION(dst), 95 const int DEPTH, 96 const int INPUT1_WIDTH) 97{ 98 // Calculate input indices 99 const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0); 100 const int y = get_global_id(1); 101 const int z = get_global_id(2) % (int)DEPTH; 102 const int w = get_global_id(2) / (int)DEPTH; 103 const int x1 = min(x, (int)INPUT1_WIDTH - (int)VEC_SIZE); 104 const int x2 = max(x - (int)INPUT1_WIDTH, 0); 105 106 // Calculate inputs and output addresses 107 const __global uchar *dst_addr = dst_ptr + (int)dst_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * (int)dst_stride_y + z * (int)dst_stride_z + w * (int)dst_stride_w; 108 const __global uchar *src1_addr = src1_ptr + (int)src1_offset_first_element_in_bytes + x1 * sizeof(DATA_TYPE) + y * (int)src1_stride_y + z * (int)src1_stride_z + w * (int)src1_stride_w; 109 const __global uchar *src2_addr = src2_ptr + (int)src2_offset_first_element_in_bytes + x2 * sizeof(DATA_TYPE) + y * (int)src2_stride_y + z * (int)src2_stride_z + w * (int)src2_stride_w; 110 111 VEC_TYPE src1_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src1_addr); 112 VEC_TYPE src2_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src2_addr); 113 114#if defined(OFFSET_IN1) && defined(OFFSET_IN2) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_IN2) && defined(SCALE_OUT) 115 src1_values = requantize(src1_values, OFFSET_IN1, OFFSET_OUT, SCALE_IN1, SCALE_OUT); 116 src2_values = requantize(src2_values, OFFSET_IN2, OFFSET_OUT, SCALE_IN2, SCALE_OUT); 117#endif /* defined(OFFSET_IN1) && defined(OFFSET_IN2) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_IN2) && defined(SCALE_OUT) */ 118 const VEC_INT x_coords = SEQ + (VEC_INT)(x); 119 120 // Rotate src1/2_values, if values0 is a combination of src1_values and src2_values. 121 SELECT_TYPE cond = CONVERT(((VEC_INT)x < (VEC_INT)INPUT1_WIDTH) && ((VEC_INT)x > (VEC_INT)(INPUT1_WIDTH - VEC_SIZE)), SELECT_TYPE); 122 src1_values = select(src1_values, ROTATE(src1_values, VEC_SIZE, INPUT1_ROTATE_N), cond); 123 src2_values = select(src2_values, ROTATE(src2_values, VEC_SIZE, INPUT1_ROTATE_N), cond); 124 125 cond = CONVERT(x_coords < (VEC_INT)(INPUT1_WIDTH), SELECT_TYPE); 126 const VEC_TYPE values0 = select(src2_values, src1_values, cond); 127 128 STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 129} 130#endif // defined(CONCATENATE_WIDTH_X2) 131 132#if defined(CONCATENATE_WIDTH_X4) 133/** This kernel concatenates four input tensors into the output tensor along the first dimension 134 * 135 * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float 136 * @note Vector size has to be passed at compile time using -DVEC_SIZE. i.e. -DVEC_SIZE=16 137 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE_LEFTOVER=3. It is defined as the remainder between the input's first dimension and VEC_SIZE 138 * @note Tensor depth should be given as a preprocessor argument using -DDEPTH=size. e.g. -DDEPTH=16 139 * 140 * @param[in] src1_ptr Pointer to the source tensor. Supported data types: All 141 * @param[in] src1_stride_x Stride of the source tensor in X dimension (in bytes) 142 * @param[in] src1_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 143 * @param[in] src1_stride_y Stride of the source tensor in Y dimension (in bytes) 144 * @param[in] src1_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 145 * @param[in] src1_stride_z Stride of the source tensor in Z dimension (in bytes) 146 * @param[in] src1_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 147 * @param[in] src1_stride_w Stride of the first source tensor in Z dimension (in bytes) 148 * @param[in] src1_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 149 * @param[in] src1_offset_first_element_in_bytes The offset of the first element in the source tensor 150 * @param[in] src2_ptr Pointer to the source tensor. Supported data types: same as @p src1_ptr 151 * @param[in] src2_stride_x Stride of the source tensor in X dimension (in bytes) 152 * @param[in] src2_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 153 * @param[in] src2_stride_y Stride of the source tensor in Y dimension (in bytes) 154 * @param[in] src2_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 155 * @param[in] src2_stride_z Stride of the source tensor in Z dimension (in bytes) 156 * @param[in] src2_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 157 * @param[in] src2_stride_w Stride of the first source tensor in Z dimension (in bytes) 158 * @param[in] src2_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 159 * @param[in] src2_offset_first_element_in_bytes The offset of the first element in the source tensor 160 * @param[in] src3_ptr Pointer to the source tensor. Supported data types: same as @p src1_ptr 161 * @param[in] src3_stride_x Stride of the source tensor in X dimension (in bytes) 162 * @param[in] src3_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 163 * @param[in] src3_stride_y Stride of the source tensor in Y dimension (in bytes) 164 * @param[in] src3_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 165 * @param[in] src3_stride_z Stride of the source tensor in Z dimension (in bytes) 166 * @param[in] src3_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 167 * @param[in] src3_stride_w Stride of the first source tensor in Z dimension (in bytes) 168 * @param[in] src3_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 169 * @param[in] src3_offset_first_element_in_bytes The offset of the first element in the source tensor 170 * @param[in] src4_ptr Pointer to the source tensor. Supported data types: same as @p src1_ptr 171 * @param[in] src4_stride_x Stride of the source tensor in X dimension (in bytes) 172 * @param[in] src4_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 173 * @param[in] src4_stride_y Stride of the source tensor in Y dimension (in bytes) 174 * @param[in] src4_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 175 * @param[in] src4_stride_z Stride of the source tensor in Z dimension (in bytes) 176 * @param[in] src4_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 177 * @param[in] src4_stride_w Stride of the first source tensor in Z dimension (in bytes) 178 * @param[in] src4_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 179 * @param[in] src4_offset_first_element_in_bytes The offset of the first element in the source tensor 180 * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src1_ptr 181 * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) 182 * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) 183 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) 184 * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) 185 * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes) 186 * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) 187 * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) 188 * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) 189 * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor 190 * @param[in] DEPTH Tensor depth 191 * @param[in] INPUT1_WIDTH First input tensor width 192 * @param[in] INPUT2_WIDTH Second input tensor width 193 * @param[in] INPUT3_WIDTH Third input tensor width 194 */ 195__kernel void concatenate_width_x4( 196 TENSOR4D_DECLARATION(src1), 197 TENSOR4D_DECLARATION(src2), 198 TENSOR4D_DECLARATION(src3), 199 TENSOR4D_DECLARATION(src4), 200 TENSOR4D_DECLARATION(dst), 201 const int DEPTH, 202 const int INPUT1_WIDTH, 203 const int INPUT2_WIDTH, 204 const int INPUT3_WIDTH) 205{ 206 // Calculate input indices 207 const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0); 208 const int y = get_global_id(1); 209 const int z = get_global_id(2) % (int)DEPTH; 210 const int w = get_global_id(2) / (int)DEPTH; 211 212 const int x1 = min(x, (int)INPUT1_WIDTH - (int)VEC_SIZE); 213 const int x2 = min(max(x - (int)INPUT1_WIDTH, 0), (int)INPUT2_WIDTH - (int)VEC_SIZE); 214 const int x3 = min(max(x - (int)INPUT1_WIDTH - (int)INPUT2_WIDTH, 0), (int)INPUT3_WIDTH - (int)VEC_SIZE); 215 const int x4 = max(x - (int)INPUT1_WIDTH - (int)INPUT2_WIDTH - (int)INPUT3_WIDTH, 0); 216 217 // Calculate inputs and output addresses 218 const __global uchar *dst_addr = dst_ptr + (int)dst_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * (int)dst_stride_y + z * (int)dst_stride_z + w * (int)dst_stride_w; 219 const __global uchar *src1_addr = src1_ptr + (int)src1_offset_first_element_in_bytes + x1 * sizeof(DATA_TYPE) + y * (int)src1_stride_y + z * (int)src1_stride_z + w * (int)src1_stride_w; 220 const __global uchar *src2_addr = src2_ptr + (int)src2_offset_first_element_in_bytes + x2 * sizeof(DATA_TYPE) + y * (int)src2_stride_y + z * (int)src2_stride_z + w * (int)src2_stride_w; 221 const __global uchar *src3_addr = src3_ptr + (int)src3_offset_first_element_in_bytes + x3 * sizeof(DATA_TYPE) + y * (int)src3_stride_y + z * (int)src3_stride_z + w * (int)src3_stride_w; 222 const __global uchar *src4_addr = src4_ptr + (int)src4_offset_first_element_in_bytes + x4 * sizeof(DATA_TYPE) + y * (int)src4_stride_y + z * (int)src4_stride_z + w * (int)src4_stride_w; 223 224 VEC_TYPE src1_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src1_addr); 225 VEC_TYPE src2_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src2_addr); 226 VEC_TYPE src3_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src3_addr); 227 VEC_TYPE src4_values = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src4_addr); 228 229#if defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) && defined(OFFSET_IN2) && defined(SCALE_IN2) && defined(OFFSET_IN3) && defined(SCALE_IN3) && defined(OFFSET_IN4) && defined(SCALE_IN4) 230 src1_values = requantize(src1_values, OFFSET_IN1, OFFSET_OUT, SCALE_IN1, SCALE_OUT); 231 src2_values = requantize(src2_values, OFFSET_IN2, OFFSET_OUT, SCALE_IN2, SCALE_OUT); 232 src3_values = requantize(src3_values, OFFSET_IN3, OFFSET_OUT, SCALE_IN3, SCALE_OUT); 233 src4_values = requantize(src4_values, OFFSET_IN4, OFFSET_OUT, SCALE_IN4, SCALE_OUT); 234#endif /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) && defined(OFFSET_IN2) && defined(SCALE_IN2) && defined(OFFSET_IN3) && defined(SCALE_IN3) && defined(OFFSET_IN4) && defined(SCALE_IN4) */ 235 236 const VEC_INT x_coords = SEQ + (VEC_INT)(x); 237 238 SELECT_TYPE cond_in2 = CONVERT(((VEC_INT)x < (VEC_INT)INPUT1_WIDTH && (VEC_INT)x > (VEC_INT)(INPUT1_WIDTH - VEC_SIZE)), SELECT_TYPE); 239 SELECT_TYPE cond_in3 = CONVERT(((VEC_INT)x < (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH) && (VEC_INT)x > (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH - VEC_SIZE)), SELECT_TYPE); 240 SELECT_TYPE cond_in4 = CONVERT(((VEC_INT)x < (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH + INPUT3_WIDTH) && (VEC_INT)x > (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH + INPUT3_WIDTH - VEC_SIZE)), SELECT_TYPE); 241 242 // Rotate src1/2_values, if values0 is a combination of src1_values and src2_values. 243 src1_values = select(src1_values, ROTATE(src1_values, VEC_SIZE, INPUT1_ROTATE_N), cond_in2); 244 src2_values = select(src2_values, ROTATE(src2_values, VEC_SIZE, INPUT1_ROTATE_N), cond_in2); 245 // Rotate src2/3_values, if values0 is a combination of src2_values and src3_values. 246 src2_values = select(src2_values, ROTATE(src2_values, VEC_SIZE, INPUT2_ROTATE_N), cond_in3); 247 src3_values = select(src3_values, ROTATE(src3_values, VEC_SIZE, INPUT2_ROTATE_N), cond_in3); 248 // Rotate src3/4_values, if values0 is a combination of src3_values and src4_values. 249 src3_values = select(src3_values, ROTATE(src3_values, VEC_SIZE, INPUT3_ROTATE_N), cond_in4); 250 src4_values = select(src4_values, ROTATE(src4_values, VEC_SIZE, INPUT3_ROTATE_N), cond_in4); 251 252 cond_in2 = CONVERT(x_coords < (VEC_INT)(INPUT1_WIDTH), SELECT_TYPE); 253 cond_in3 = CONVERT(x_coords < (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH), SELECT_TYPE); 254 cond_in4 = CONVERT(x_coords < (VEC_INT)(INPUT1_WIDTH + INPUT2_WIDTH + INPUT3_WIDTH), SELECT_TYPE); 255 256 VEC_TYPE values0 = select(src2_values, src1_values, cond_in2); 257 values0 = select(src3_values, values0, cond_in3); 258 values0 = select(src4_values, values0, cond_in4); 259 260 STORE_VECTOR_SELECT(values, DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 261} 262#endif /* defined(CONCATENATE_WIDTH_X4) */ 263#endif /* defined(ELEMENT_SIZE) */ 264 265#if defined(WIDTH_OFFSET) && defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER) 266#if defined(CONCATENATE_WIDTH) 267/** This kernel concatenates the input tensor into the output tensor along the first dimension 268 * 269 * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float 270 * @note Vector size has to be passed at compile time using -DVEC_SIZE. i.e. -DVEC_SIZE=16 271 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE_LEFTOVER=3. It is defined as the remainder between the input's first dimension and VEC_SIZE 272 * @note The offset for the first spatial dimension has to be passed at compile time using -DWIDTH_OFFSET. i.e. -DWIDTH_OFFSET=128 273 * 274 * @param[in] src_ptr Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/F32 275 * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) 276 * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 277 * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) 278 * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 279 * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes) 280 * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 281 * @param[in] src_stride_w Stride of the first source tensor in Z dimension (in bytes) 282 * @param[in] src_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 283 * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor 284 * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr 285 * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) 286 * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) 287 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) 288 * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) 289 * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes) 290 * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) 291 * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) 292 * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) 293 * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor 294 * @param[in] DEPTH Tensor depth 295 */ 296__kernel void concatenate_width( 297 TENSOR4D_DECLARATION(src), 298 TENSOR4D_DECLARATION(dst), 299 const int DEPTH) 300{ 301 // Calculate input indices 302 const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0); 303 const int y = get_global_id(1); 304 const int z = get_global_id(2) % (int)DEPTH; 305 const int w = get_global_id(2) / (int)DEPTH; 306 307 __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * src_stride_y + z * src_stride_z + w * src_stride_w; 308 __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * sizeof(DATA_TYPE) + y * dst_stride_y + z * dst_stride_z + w * dst_stride_w; 309 310 VEC_TYPE source_values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src_addr); 311 312#if defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) 313 const VEC_QUANT out0 = requantize(source_values0, OFFSET_IN1, OFFSET_OUT, SCALE_IN1, SCALE_OUT); 314 STORE_VECTOR_SELECT(out, DATA_TYPE, dst_addr + WIDTH_OFFSET * sizeof(DATA_TYPE), VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 315#else /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 316 STORE_VECTOR_SELECT(source_values, DATA_TYPE, dst_addr + WIDTH_OFFSET * sizeof(DATA_TYPE), VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 317#endif /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 318} 319#endif /* defined(CONCATENATE_WIDTH) */ 320#endif /* defined(WIDTH_OFFSET) && defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER)*/ 321 322#if defined(VEC_SIZE_LEFTOVER) 323#if defined(CONCATENATE_HEIGHT) 324#if defined(HEIGHT_OFFSET) && defined(VEC_SIZE) 325/** This kernel concatenates the input tensor into the output tensor along the second dimension 326 * 327 * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float 328 * @note Vector size has to be passed at compile time using -DVEC_SIZE. i.e. -DVEC_SIZE=16 329 * @note Vector sizes supported are 2,4,8 and 16. 330 * @note The offset for the second spatial dimension has to be passed at compile time using -DHEIGHT_OFFSET. i.e. -DHEIGHT_OFFSET=128 331 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE=3. It is defined as the remainder between the input's first dimension and VEC_SIZE 332 * 333 * @param[in] src_ptr Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/F32 334 * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) 335 * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 336 * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) 337 * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 338 * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes) 339 * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 340 * @param[in] src_stride_w Stride of the first source tensor in Z dimension (in bytes) 341 * @param[in] src_step_w src_stride_z * number of elements along Z processed per workitem(in bytes) 342 * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor 343 * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr 344 * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) 345 * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) 346 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) 347 * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) 348 * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes) 349 * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) 350 * @param[in] dst_stride_w Stride of the destination tensor in Z dimension (in bytes) 351 * @param[in] dst_step_w output_stride_z * number of elements along Z processed per workitem(in bytes) 352 * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor 353 * @param[in] DEPTH Tensor depth 354 */ 355__kernel void concatenate_height( 356 TENSOR4D_DECLARATION(src), 357 TENSOR4D_DECLARATION(dst), 358 const int DEPTH) 359{ 360 const int x_offs = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0) * sizeof(DATA_TYPE); 361 362 __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offs + get_global_id(1) * src_stride_y + (get_global_id(2) % DEPTH) * src_stride_z + (get_global_id( 363 2) / DEPTH) * src_stride_w; 364 __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offs + get_global_id(1) * dst_stride_y + (get_global_id(2) % DEPTH) * dst_stride_z + (get_global_id( 365 2) / DEPTH) * dst_stride_w; 366 367 VEC_TYPE source_values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src_addr); 368 369#if defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) 370 const VEC_QUANT out0 = requantize(source_values0, OFFSET_IN1, OFFSET_OUT, SCALE_IN1, SCALE_OUT); 371 STORE_VECTOR_SELECT(out, DATA_TYPE, dst_addr + HEIGHT_OFFSET * dst_stride_y, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 372#else /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 373 STORE_VECTOR_SELECT(source_values, DATA_TYPE, dst_addr + HEIGHT_OFFSET * dst_stride_y, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 374#endif /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 375} 376#endif /* defined(CONCATENATE_HEIGHT) */ 377#endif /* defined(HEIGHT_OFFSET) */ 378 379#if defined(CONCATENATE) 380/** This kernel concatenates the input tensor into the output tensor along the third dimension 381 * 382 * @note The data type has to be passed at compile time using -DDATA_TYPE. i.e. -DDATA_TYPE=float 383 * @note Vector size has to be passed at compile time using -DVEC_SIZE. i.e. -DVEC_SIZE=16 384 * @note Leftover vector size has to be passed at compile time using -DVEC_SIZE_LEFTOVER. e.g. -DVEC_SIZE=3. It is defined as the remainder between the input's first dimension and VEC_SIZE 385 * 386 * @param[in] src_ptr Pointer to the source tensor. Supported data types: All 387 * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) 388 * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) 389 * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) 390 * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) 391 * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes) 392 * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) 393 * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor 394 * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr 395 * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) 396 * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) 397 * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) 398 * @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes) 399 * @param[in] dst_stride_z Stride of the source tensor in Z dimension (in bytes) 400 * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) 401 * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor 402 * @param[in] offsets The offsets to the first valid element of the output tensor in bytes 403 */ 404__kernel void concatenate( 405 TENSOR3D_DECLARATION(src), 406 TENSOR3D_DECLARATION(dst), 407 int offset) 408{ 409 uint x_offs = max((int)(get_global_id(0) * VEC_SIZE * sizeof(DATA_TYPE) - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE * sizeof(DATA_TYPE)), 0); 410 411 __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x_offs + get_global_id(1) * src_stride_y + get_global_id(2) * src_stride_z; 412 __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x_offs + get_global_id(1) * dst_stride_y + get_global_id(2) * dst_stride_z; 413 414 VEC_TYPE source_values0 = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)src_addr); 415 416#if defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) 417 source_values0 = requantize(source_values0, OFFSET_IN1, OFFSET_OUT, SCALE_IN1, SCALE_OUT); 418#endif /* defined(OFFSET_IN1) && defined(OFFSET_OUT) && defined(SCALE_IN1) && defined(SCALE_OUT) */ 419 420 STORE_VECTOR_SELECT(source_values, DATA_TYPE, dst_addr + offset, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0) 421} 422#endif // defined(CONCATENATE) 423#endif /* defined(VEC_SIZE_LEFTOVER) */ 424#endif /* defined(DATA_TYPE) */ 425#endif /* defined(VEC_SIZE) */ 426