diff options
Diffstat (limited to 'libs/ARMComputeEx/src/core/CL/cl_kernels/pad.cl')
-rw-r--r-- | libs/ARMComputeEx/src/core/CL/cl_kernels/pad.cl | 86 |
1 files changed, 86 insertions, 0 deletions
diff --git a/libs/ARMComputeEx/src/core/CL/cl_kernels/pad.cl b/libs/ARMComputeEx/src/core/CL/cl_kernels/pad.cl new file mode 100644 index 000000000..ecf4696e9 --- /dev/null +++ b/libs/ARMComputeEx/src/core/CL/cl_kernels/pad.cl @@ -0,0 +1,86 @@ +/* + * Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved + * Copyright (c) 2016, 2017 ARM Limited. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include "helpers.h" + +#if defined(IW) && defined(IH) && defined(ID) && defined(IB) && defined(DEPTH_OUT) && defined(ZERO_VALUE) +/** Perform space to depth rearrangement of tensor + * + * @attention Data type can be passed using the -DDATA_TYPE compile flag, e.g. -DDATA_TYPE=float + * @attention Output tensor depth should be given as a preprocessor argument using -DDEPTH_OUT=size. e.g. -DDEPTH_OUT=16 + * @attention Input dimensions should be passed as a preprocessor argument using -DIW(width), -DIH(height), -DID(depth) and -DIB(batch). e.g. -DIW = 4 + * @attention The value to be set by pad value using -DZERO_VALUE=value. e.g. -DZERO_VALUE=0 + * + * @param[in] input_ptr Pointer to the source tensor. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32 + * @param[in] input_stride_x Stride of the source tensor in X dimension (in bytes) + * @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes) + * @param[in] input_stride_y Stride of the source tensor in Y dimension (in bytes) + * @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) + * @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes) + * @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes) + * @param[in] input_offset_first_element_in_bytes The offset of the first element in the source tensor + * + * @param[out] output_ptr Pointer to the destination tensor. Supported data types: same as @p inpu +t_ptr + * @param[in] output_stride_x Stride of the destination tensor in X dimension (in bytes) + * @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes) + * @param[in] output_stride_y Stride of the destination tensor in Y dimension (in bytes) + * @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in +bytes) + * @param[in] output_stride_z Stride of the destination tensor in Z dimension (in bytes) + * @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes) + * @param[in] output_stride_w Stride of the destination tensor in W dimension (in bytes) + * @param[in] output_step_w output_stride_w * number of elements along W processed per workitem(in bytes) + * @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination tensor + * + * @param[in] pad_values Padding values for each of the dimensions. Only pad values for Up(for + * batch), Top(for height), Left(for width) and Front(for depth) are + * required. Supported data type: S32 + */ + +__kernel void pad( + TENSOR4D_DECLARATION(input), + TENSOR4D_DECLARATION(output), + const int4 pad_values) + { + Tensor4D in = CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(input, 0); + Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT(output, DEPTH_OUT); + + int index[4]={0}; + + index[0] = get_global_id(0);//W + index[1] = get_global_id(1);//H + index[2] = get_global_id(2) % DEPTH_OUT;//C + index[3] = get_global_id(2) / DEPTH_OUT;//N + + if (index[0] < pad_values.x || index[0] >= (IW + pad_values.x) || + index[1] < pad_values.y || index[1] >= (IH + pad_values.y) || + index[2] < pad_values.z || index[2] >= (ID + pad_values.z) || + index[3] < pad_values.w || index[3] >= (IB + pad_values.w)) + { + *((__global DATA_TYPE *)out.ptr) = (DATA_TYPE)ZERO_VALUE; + } + else + { + *((__global DATA_TYPE *)out.ptr) = *((__global DATA_TYPE *) + tensor4D_offset(&in, index[0] - pad_values.x, + index[1] - pad_values.y, + index[2] - pad_values.z, + index[3] - pad_values.w)); + } + } + +#endif //if defined(IW) && defined(IH) && defined(ID) && defined(IB) && defined(DEPTH_OUT) && defined(ZERO_VALUE) |