SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016, 2017 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 | |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 26 | #if defined(FIXED_POINT_POSITION) |
| 27 | #include "fixed_point.h" |
| 28 | |
| 29 | #define ADD_OP(a, b) ADD_SAT_OP_EXPAND((a), (b), DATA_TYPE_PROMOTED, 8) |
| 30 | #define MUL_OP(a, b) MUL_SAT_OP_EXPAND(CONVERT((a), VEC_DATA_TYPE(DATA_TYPE_PROMOTED, 8)), CONVERT((b), VEC_DATA_TYPE(DATA_TYPE_PROMOTED, 8)), DATA_TYPE_PROMOTED, 8, FIXED_POINT_POSITION) |
| 31 | |
| 32 | // There is no need to have a larger intermediate type for qs32 because all the arguments are already promoted |
| 33 | MULQ_SAT_IMPL(qs32x8, qs32x8) |
| 34 | |
| 35 | #else /* FIXED_POINT_POSITION */ |
steniu01 | db00668 | 2017-08-09 16:26:22 +0100 | [diff] [blame] | 36 | #undef CONVERT_SAT |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 37 | |
| 38 | #define ADD_OP(a, b) ((a) + (b)) |
| 39 | #define MUL_OP(a, b) ((a) * (b)) |
| 40 | #define CONVERT_SAT(a, b) ((a)) |
| 41 | |
| 42 | #endif /* FIXED_POINT_POSITION */ |
| 43 | |
Gian Marco Iodice | 1c8409d | 2017-09-06 17:24:25 +0100 | [diff] [blame] | 44 | #if defined(DATA_TYPE) && defined(DATA_SIZE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH) |
| 45 | |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 46 | #if STRIDE_X == 3 |
| 47 | #define INPUT_PIXEL_STR(data_size) extract_input_stride3_##data_size |
| 48 | #define INPUT_PIXEL(data_size) INPUT_PIXEL_STR(data_size) |
| 49 | #elif STRIDE_X == 2 |
| 50 | #define INPUT_PIXEL(data_size) extract_input_stride2 |
| 51 | #elif STRIDE_X == 1 |
| 52 | #define INPUT_PIXEL(data_size) extract_input_stride1 |
| 53 | #else /* STRIDE_X not equals 1, 2 or 3 */ |
| 54 | #error "Only support strides 1, 2 and 3" |
| 55 | #endif /* STRIDE_X == 3 */ |
| 56 | |
| 57 | /** Extracts a 1D horizontal vector from the input tensor with stride as 1. |
| 58 | * |
| 59 | * @param[in] input_pixel Pointer to the first pixel. |
| 60 | * |
| 61 | * @return extracted input pixels. |
| 62 | */ |
| 63 | inline VEC_DATA_TYPE(DATA_TYPE, 8) extract_input_stride1(__global const DATA_TYPE *input_pixel) |
| 64 | { |
| 65 | return vload8(0, input_pixel); |
| 66 | } |
| 67 | |
| 68 | /** Extracts a 1D horizontal vector from the input tensor with stride as 2. |
| 69 | * |
| 70 | * @param[in] input_pixel Pointer to the first pixel. |
| 71 | * |
| 72 | * @return extracted input pixels. |
| 73 | */ |
| 74 | inline VEC_DATA_TYPE(DATA_TYPE, 8) extract_input_stride2(__global const DATA_TYPE *input_pixel) |
| 75 | { |
| 76 | VEC_DATA_TYPE(DATA_TYPE, 16) |
| 77 | temp = vload16(0, input_pixel); |
| 78 | return temp.s02468ace; |
| 79 | } |
| 80 | |
| 81 | /** Extracts a 1D horizontal vector from the input tensor with stride as 3 and 32-bit data size. |
| 82 | * |
| 83 | * @param[in] input_pixel Pointer to the first pixel. |
| 84 | * |
| 85 | * @return extracted input pixels. |
| 86 | */ |
| 87 | inline VEC_DATA_TYPE(DATA_TYPE, 8) extract_input_stride3_32(__global const DATA_TYPE *input_pixel) |
| 88 | { |
| 89 | VEC_DATA_TYPE(DATA_TYPE, 4) |
| 90 | temp1 = vload4(0, input_pixel); |
| 91 | VEC_DATA_TYPE(DATA_TYPE, 4) |
| 92 | temp2 = vload4(0, input_pixel + 6); |
| 93 | VEC_DATA_TYPE(DATA_TYPE, 4) |
| 94 | temp3 = vload4(0, input_pixel + 12); |
| 95 | VEC_DATA_TYPE(DATA_TYPE, 4) |
| 96 | temp4 = vload4(0, input_pixel + 18); |
| 97 | return (VEC_DATA_TYPE(DATA_TYPE, 8))(temp1.s03, temp2.s03, temp3.s03, temp4.s03); |
| 98 | } |
| 99 | |
| 100 | /** Extracts a 1D horizontal vector from the input tensor with stride as 3 and 16-bit data size. |
| 101 | * |
| 102 | * @param[in] input_pixel Pointer to the first pixel. |
| 103 | * |
| 104 | * @return extracted input pixels. |
| 105 | */ |
| 106 | inline VEC_DATA_TYPE(DATA_TYPE, 8) extract_input_stride3_16(__global const DATA_TYPE *input_pixel) |
| 107 | { |
| 108 | VEC_DATA_TYPE(DATA_TYPE, 8) |
| 109 | temp1 = vload8(0, input_pixel); |
| 110 | VEC_DATA_TYPE(DATA_TYPE, 8) |
| 111 | temp2 = vload8(0, input_pixel + 8); |
| 112 | VEC_DATA_TYPE(DATA_TYPE, 8) |
| 113 | temp3 = vload8(0, input_pixel + 16); |
| 114 | return (VEC_DATA_TYPE(DATA_TYPE, 8))(temp1.s036, temp2.s147, temp3.s25); |
| 115 | } |
| 116 | |
| 117 | /** Extracts a 1D horizontal vector from the input tensor with stride as 3 and 8-bit data size. |
| 118 | * |
| 119 | * @param[in] input_pixel Pointer to the first pixel. |
| 120 | * |
| 121 | * @return extracted input pixels. |
| 122 | */ |
| 123 | inline VEC_DATA_TYPE(DATA_TYPE, 8) extract_input_stride3_8(__global const DATA_TYPE *input_pixel) |
| 124 | { |
| 125 | VEC_DATA_TYPE(DATA_TYPE, 16) |
| 126 | temp1 = vload16(0, input_pixel); |
| 127 | VEC_DATA_TYPE(DATA_TYPE, 16) |
| 128 | temp2 = vload16(0, input_pixel + 12); |
| 129 | return (VEC_DATA_TYPE(DATA_TYPE, 8))(temp1.s0369, temp2.s0369); |
| 130 | } |
| 131 | |
| 132 | /** This kernel performs a direct convolution to convolve the low three dimensions. |
| 133 | * |
| 134 | * @note The data type must be passed at compile time using -DDATA_TYPE: e.g. -DDATA_TYPE=float |
| 135 | * @note The data size must be passed at compile time using -DDATA_SIZE e.g. -DDATA_SIZE=32 |
Gian Marco Iodice | 5cb4d6a | 2017-08-08 10:53:00 +0100 | [diff] [blame] | 136 | * @note The convolution stride x must be passed at compile time using -DSTRIDE_X e.g. -DSTRIDE_X=1 |
| 137 | * @note The third dimensions of the weights tensors must be passed at compile time using -DWEIGHTS_DEPTH |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 138 | * @note In case biases will be added to the convolution -DHAS_BIAS has to be passed to append the final matrix with 1 in each row. |
| 139 | * |
Gian Marco Iodice | 5cb4d6a | 2017-08-08 10:53:00 +0100 | [diff] [blame] | 140 | * @param[in] src_ptr Pointer to the source tensor. Supported data types: F16/F32 |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 141 | * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) |
| 142 | * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) |
| 143 | * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) |
| 144 | * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) |
| 145 | * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes) |
| 146 | * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) |
| 147 | * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor |
| 148 | * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr |
| 149 | * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) |
| 150 | * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) |
| 151 | * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) |
| 152 | * @param[in] dst_step_y dst_stride_y * number of elements along Z processed per workitem(in bytes) |
| 153 | * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) |
| 154 | * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) |
| 155 | * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor |
| 156 | * @param[out] weights_ptr Pointer to the weights tensor. Supported data types: same as @p weights_ptr |
| 157 | * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) |
| 158 | * @param[in] weights_step_x weights_stride_x * number of elements along X processed per workitem(in bytes) |
| 159 | * @param[in] weights_stride_y Stride of the weights tensor in Y dimension (in bytes) |
| 160 | * @param[in] weights_step_y weights_stride_y * number of elements along y processed per workitem(in bytes) |
| 161 | * @param[in] weights_stride_z Stride of the weights tensor in Z dimension (in bytes) |
| 162 | * @param[in] weights_step_z weights_stride_z * number of elements along Z processed per workitem(in bytes) |
| 163 | * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor |
| 164 | * @param[in] biases_ptr Pointer to the biases tensor. Same as @p src_ptr |
| 165 | * @param[in] biases_stride_x Stride of the biases tensor in X dimension (in bytes) |
| 166 | * @param[in] biases_step_x biases_stride_x * number of elements along X processed per workitem(in bytes) |
| 167 | * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the biases tensor |
Gian Marco Iodice | 5cb4d6a | 2017-08-08 10:53:00 +0100 | [diff] [blame] | 168 | * @param[in] weights_stride_w Stride of the weights tensor in the 4th dimension |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 169 | */ |
| 170 | __kernel void direct_convolution1x1( |
| 171 | TENSOR3D_DECLARATION(src), |
| 172 | TENSOR3D_DECLARATION(dst), |
| 173 | TENSOR3D_DECLARATION(weights), |
| 174 | #ifdef HAS_BIAS |
| 175 | VECTOR_DECLARATION(biases), |
| 176 | #endif /* defined(HAS_BIAS) */ |
Gian Marco Iodice | 5cb4d6a | 2017-08-08 10:53:00 +0100 | [diff] [blame] | 177 | unsigned int weights_stride_w) |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 178 | { |
| 179 | Image src = CONVERT_TO_IMAGE_STRUCT(src); |
| 180 | Tensor3D weights = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(weights); |
| 181 | Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst); |
| 182 | |
| 183 | #ifdef HAS_BIAS |
| 184 | Vector biases = CONVERT_TO_VECTOR_STRUCT_NO_STEP(biases); |
| 185 | #endif /* defined(HAS_BIAS) */ |
| 186 | |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 187 | VEC_DATA_TYPE(DATA_TYPE_PROMOTED, 8) |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 188 | pixels = 0; |
| 189 | |
| 190 | const uint z_index = get_global_id(2); |
| 191 | |
| 192 | weights.ptr += z_index * weights_stride_w; |
| 193 | |
Gian Marco Iodice | 744b5ed | 2017-10-06 15:44:27 +0100 | [diff] [blame] | 194 | for(volatile int d = 0; d < WEIGHTS_DEPTH; ++d) |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 195 | { |
| 196 | DATA_TYPE weight = *(__global DATA_TYPE *)weights.ptr; |
| 197 | VEC_DATA_TYPE(DATA_TYPE, 8) |
| 198 | input_pixel = INPUT_PIXEL(DATA_SIZE)((__global DATA_TYPE *)src.ptr); |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 199 | pixels = ADD_OP(pixels, MUL_OP((VEC_DATA_TYPE(DATA_TYPE, 8))weight, input_pixel)); |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 200 | src.ptr += src_stride_z; |
| 201 | weights.ptr += weights_stride_z; |
| 202 | } |
| 203 | |
| 204 | #ifdef HAS_BIAS |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 205 | pixels = ADD_OP(pixels, (VEC_DATA_TYPE(DATA_TYPE_PROMOTED, 8)) * ((__global DATA_TYPE *)(vector_offset(&biases, z_index)))); |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 206 | #endif /* defined(HAS_BIAS) */ |
| 207 | |
Michalis Spyrou | def665a | 2017-08-14 11:26:37 +0100 | [diff] [blame] | 208 | vstore8(CONVERT_SAT(pixels, VEC_DATA_TYPE(DATA_TYPE, 8)), 0, (__global DATA_TYPE *)dst.ptr); |
SiCong Li | c51b72f | 2017-07-28 14:46:20 +0100 | [diff] [blame] | 209 | } |
steniu01 | db00668 | 2017-08-09 16:26:22 +0100 | [diff] [blame] | 210 | #endif // defined(DATA_TYPE) && defined(DATA_SIZE) && defined(STRIDE_X) && defined(WEIGHTS_DEPTH) |
Gian Marco Iodice | 1c8409d | 2017-09-06 17:24:25 +0100 | [diff] [blame] | 211 | |
| 212 | #if defined(WEIGHTS_DEPTH) |
| 213 | |
| 214 | #define CONVOLUTION1x1_BIFROST(acc, src, weight_value) \ |
| 215 | ({ \ |
| 216 | acc.s0 = mad(src.s0, weight_value, acc.s0); \ |
| 217 | acc.s1 = mad(src.s1, weight_value, acc.s1); \ |
| 218 | acc.s2 = mad(src.s2, weight_value, acc.s2); \ |
| 219 | acc.s3 = mad(src.s3, weight_value, acc.s3); \ |
| 220 | }) |
| 221 | |
| 222 | /** An optimized direct convolution 1x1 OpenCL kernel for Bifrost architectures when the data type is F32 |
| 223 | * |
| 224 | * @note This OpenCL kernel works only with stride_x and stride_y equal to 1 |
| 225 | * @note The third dimensions of the weights tensors must be passed at compile time using -DWEIGHTS_DEPTH |
| 226 | * @note In case biases, -DHAS_BIAS must to be passed at compile |
| 227 | * |
| 228 | * @param[in] src_ptr Pointer to the source tensor. Supported data types: F32 |
| 229 | * @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes) |
| 230 | * @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes) |
| 231 | * @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes) |
| 232 | * @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes) |
| 233 | * @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes) |
| 234 | * @param[in] src_step_z src_stride_z * number of elements along Z processed per workitem(in bytes) |
| 235 | * @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor |
| 236 | * @param[out] dst_ptr Pointer to the destination tensor. Supported data types: same as @p src_ptr |
| 237 | * @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes) |
| 238 | * @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes) |
| 239 | * @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes) |
| 240 | * @param[in] dst_step_y dst_stride_y * number of elements along Z processed per workitem(in bytes) |
| 241 | * @param[in] dst_stride_z Stride of the destination tensor in Z dimension (in bytes) |
| 242 | * @param[in] dst_step_z dst_stride_z * number of elements along Z processed per workitem(in bytes) |
| 243 | * @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor |
| 244 | * @param[out] weights_ptr Pointer to the weights tensor. Supported data types: same as @p weights_ptr |
| 245 | * @param[in] weights_stride_x Stride of the weights tensor in X dimension (in bytes) |
| 246 | * @param[in] weights_step_x weights_stride_x * number of elements along X processed per workitem(in bytes) |
| 247 | * @param[in] weights_stride_y Stride of the weights tensor in Y dimension (in bytes) |
| 248 | * @param[in] weights_step_y weights_stride_y * number of elements along y processed per workitem(in bytes) |
| 249 | * @param[in] weights_stride_z Stride of the weights tensor in Z dimension (in bytes) |
| 250 | * @param[in] weights_step_z weights_stride_z * number of elements along Z processed per workitem(in bytes) |
| 251 | * @param[in] weights_offset_first_element_in_bytes The offset of the first element in the weights tensor |
| 252 | * @param[in] biases_ptr Pointer to the biases tensor. Same as @p src_ptr |
| 253 | * @param[in] biases_stride_x Stride of the biases tensor in X dimension (in bytes) |
| 254 | * @param[in] biases_step_x biases_stride_x * number of elements along X processed per workitem(in bytes) |
| 255 | * @param[in] biases_offset_first_element_in_bytes The offset of the first element in the biases tensor |
| 256 | * @param[in] weights_stride_w Stride of the weights tensor in the 4th dimension |
| 257 | */ |
| 258 | __kernel void direct_convolution1x1_f32_bifrost( |
| 259 | TENSOR3D_DECLARATION(src), |
| 260 | TENSOR3D_DECLARATION(dst), |
| 261 | TENSOR3D_DECLARATION(weights), |
| 262 | #ifdef HAS_BIAS |
| 263 | VECTOR_DECLARATION(biases), |
| 264 | #endif /* defined(HAS_BIAS) */ |
| 265 | unsigned int weights_stride_w) |
| 266 | { |
| 267 | // Get the kernel index |
| 268 | const int kernel_index = get_global_id(2); |
| 269 | |
| 270 | Image src = CONVERT_TO_IMAGE_STRUCT(src); |
| 271 | Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst); |
| 272 | |
| 273 | float4 acc0 = 0.0f; |
| 274 | float4 acc1 = 0.0f; |
| 275 | float4 acc2 = 0.0f; |
| 276 | float4 acc3 = 0.0f; |
| 277 | |
| 278 | __global uchar *weights_addr = (__global uchar *)(weights_ptr + weights_offset_first_element_in_bytes + kernel_index * weights_stride_w); |
| 279 | __global uchar *src_addr = (__global uchar *)offset(&src, 0, 0); |
| 280 | |
| 281 | for(ushort d = 0; d < (ushort)WEIGHTS_DEPTH; ++d) |
| 282 | { |
| 283 | // Load the weights |
| 284 | float weight = *((__global float *)weights_addr); |
| 285 | |
| 286 | // Load values from row0 of input tensor |
| 287 | float4 src0 = vload4(0, (__global float *)(src_addr + 0 * src_stride_y)); |
| 288 | float4 src1 = vload4(0, (__global float *)(src_addr + 1 * src_stride_y)); |
| 289 | float4 src2 = vload4(0, (__global float *)(src_addr + 2 * src_stride_y)); |
| 290 | float4 src3 = vload4(0, (__global float *)(src_addr + 3 * src_stride_y)); |
| 291 | |
| 292 | CONVOLUTION1x1_BIFROST(acc0, src0, weight); |
| 293 | CONVOLUTION1x1_BIFROST(acc1, src1, weight); |
| 294 | CONVOLUTION1x1_BIFROST(acc2, src2, weight); |
| 295 | CONVOLUTION1x1_BIFROST(acc3, src3, weight); |
| 296 | |
| 297 | src_addr += src_stride_z; |
| 298 | weights_addr += weights_stride_z; |
| 299 | } |
| 300 | |
| 301 | #ifdef HAS_BIAS |
| 302 | Vector biases = CONVERT_TO_VECTOR_STRUCT_NO_STEP(biases); |
| 303 | |
| 304 | float bias = (float) * ((__global float *)(vector_offset(&biases, kernel_index))); |
| 305 | |
| 306 | acc0.s0 += bias; |
| 307 | acc0.s1 += bias; |
| 308 | acc0.s2 += bias; |
| 309 | acc0.s3 += bias; |
| 310 | acc1.s0 += bias; |
| 311 | acc1.s1 += bias; |
| 312 | acc1.s2 += bias; |
| 313 | acc1.s3 += bias; |
| 314 | acc2.s0 += bias; |
| 315 | acc2.s1 += bias; |
| 316 | acc2.s2 += bias; |
| 317 | acc2.s3 += bias; |
| 318 | acc3.s0 += bias; |
| 319 | acc3.s1 += bias; |
| 320 | acc3.s2 += bias; |
| 321 | acc3.s3 += bias; |
| 322 | #endif /* defined(HAS_BIAS) */ |
| 323 | |
| 324 | vstore4(acc0, 0, (__global float *)(dst.ptr + 0 * dst_stride_y)); |
| 325 | vstore4(acc1, 0, (__global float *)(dst.ptr + 1 * dst_stride_y)); |
| 326 | vstore4(acc2, 0, (__global float *)(dst.ptr + 2 * dst_stride_y)); |
| 327 | vstore4(acc3, 0, (__global float *)(dst.ptr + 3 * dst_stride_y)); |
| 328 | } |
| 329 | #endif // defined(WEIGHTS_DEPTH) |