From 3580c7564c8ae68987dbe4717f99a952bc5d7aad Mon Sep 17 00:00:00 2001 From: SiCong Li Date: Wed, 14 Oct 2020 17:00:56 +0100 Subject: COMPMID-3708 Remove OpenCL padding: CLCopyKernel [Patch2] * Remove the user-supplied padding from CLCopyKernel Note that this padding is different from the internal "padding" in the original task, as it is user-supplied instead of internal. This user-supplied padding interface is removed simply because it has been replaced by a more capable CLPadLayerKernel, and is not used anywhere else. Signed-off-by: SiCong Li Change-Id: Ib53e76efd7d043ee79dcd47ca734c6dc685da43e Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4194 Tested-by: Arm Jenkins Reviewed-by: Michele Di Giorgio Comments-Addressed: Arm Jenkins --- src/core/CL/cl_kernels/copy_tensor.cl | 54 ----------------------------------- 1 file changed, 54 deletions(-) (limited to 'src/core/CL/cl_kernels/copy_tensor.cl') diff --git a/src/core/CL/cl_kernels/copy_tensor.cl b/src/core/CL/cl_kernels/copy_tensor.cl index 95da9a3cd3..9c90969827 100644 --- a/src/core/CL/cl_kernels/copy_tensor.cl +++ b/src/core/CL/cl_kernels/copy_tensor.cl @@ -23,60 +23,6 @@ */ #include "helpers.h" -#if defined(PAD00) && defined(PAD10) && defined(PAD20) && defined(PAD21) && defined(PAD30) && defined(DATA_TYPE) && defined(VEC_SIZE) // Compile time constants - -/** Perform a padded copy of input tensor to the output tensor. Padding values are defined at compile time - * - * @attention The following variables must be passed at compile time: - * -# -DPAD{d}{0,1} = padding before{0} and after{1} dimension d (d < 4) - * -# -DDEPTH = The third dimension (depth) of the tensor (it is needed only if d == 3) - * -# -DDATA_TYPE = Input and output datatypes. - * - * @param[in] in_ptr Pointer to the source tensor. Supported data types: All - * @param[in] in_stride_x Stride of the source tensor in X dimension (in bytes) - * @param[in] in_step_x input_stride_x * number of elements along X processed per workitem(in bytes) - * @param[in] in_stride_y Stride of the source tensor in Y dimension (in bytes) - * @param[in] in_step_y input_stride_y * number of elements along Y processed per workitem(in bytes) - * @param[in] in_stride_z Stride of the source tensor in Z dimension (in bytes) - * @param[in] in_step_z input_stride_z * number of elements along Z processed per workitem(in bytes) - * @param[in] in_offset_first_element_in_bytes The offset of the first element in the source tensor - * @param[out] out_ptr Pointer to the destination tensor. Supported data types: same as @p in_ptr - * @param[in] out_stride_x Stride of the destination tensor in X dimension (in bytes) - * @param[in] out_step_x output_stride_x * number of elements along X processed per workitem(in bytes) - * @param[in] out_stride_y Stride of the destination tensor in Y dimension (in bytes) - * @param[in] out_step_y output_stride_y * number of elements along Y processed per workitem(in bytes) - * @param[in] out_stride_z Stride of the source tensor in Z dimension (in bytes) - * @param[in] out_step_z output_stride_z * number of elements along Z processed per workitem(in bytes) - * @param[in] out_offset_first_element_in_bytes The offset of the first element in the destination tensor - */ -__kernel void copy_pad_tensor( - TENSOR3D_DECLARATION(in), - TENSOR3D_DECLARATION(out)) - -{ - Tensor3D in = CONVERT_TO_TENSOR3D_STRUCT(in); - Tensor3D out = CONVERT_TO_TENSOR3D_STRUCT(out); - - const int offset_x = PAD00; - const int offset_y = PAD10; - const int offset_z = PAD20; - -#if PAD30 > 0 - const size_t in_batch = get_global_id(2) / DEPTH; - const int total_depth = DEPTH + PAD20 + PAD21; - const int offset_w = PAD30 * total_depth + in_batch * (PAD20 + PAD21); -#else // PAD30 == 0 - const int offset_w = 0; -#endif // PAD30 - - VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE) - data = VLOAD(VEC_SIZE)(0, (__global DATA_TYPE *)in.ptr); - - VSTORE(VEC_SIZE) - (data, 0, (__global DATA_TYPE *)tensor3D_offset(&out, offset_x, offset_y, offset_z + offset_w)); -} -#endif // Compile time constants - #if defined(DATA_TYPE) && defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER) /** Performs a copy of input tensor to the output tensor. * -- cgit v1.2.1