1 /*
2 * Copyright (c) 2017-2021 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 "src/gpu/cl/kernels/ClTransposeKernel.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/TensorInfo.h"
31 #include "arm_compute/core/Utils.h"
32 #include "arm_compute/core/Validate.h"
33 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
34 #include "src/core/CL/CLValidate.h"
35 #include "src/core/helpers/AutoConfiguration.h"
36 #include "src/core/helpers/WindowHelpers.h"
37 #include "support/Cast.h"
38 #include "support/StringSupport.h"
39
40 namespace arm_compute
41 {
42 namespace opencl
43 {
44 namespace kernels
45 {
ClTransposeKernel()46 ClTransposeKernel::ClTransposeKernel()
47 {
48 _type = CLKernelType::ELEMENTWISE;
49 }
50
configure(const CLCompileContext & compile_context,const ITensorInfo * src,ITensorInfo * dst)51 void ClTransposeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
52 {
53 ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
54
55 // Output auto initialization if not yet initialized
56 const TensorShape dst_shape = misc::shape_calculator::compute_transposed_shape(*src);
57 auto_init_if_empty(*dst, src->clone()->set_tensor_shape(dst_shape));
58
59 ARM_COMPUTE_ERROR_THROW_ON(ClTransposeKernel::validate(src, dst));
60 auto padding_info = get_padding_info({ src, dst });
61
62 // Create kernel
63 const unsigned int vec_size_x = adjust_vec_size(max_cl_vector_width / src->element_size(), src->dimension(0));
64 const int vec_size_x_leftovers = src->dimension(0) % vec_size_x;
65 const unsigned int vec_size_y = adjust_vec_size(max_cl_vector_width / src->element_size(), src->dimension(1));
66 const int vec_size_y_leftovers = src->dimension(1) % vec_size_y;
67
68 CLBuildOptions build_opts;
69 build_opts.add_option("-DDATA_TYPE_IN_BYTES=" + support::cpp11::to_string(src->element_size()));
70 build_opts.add_option("-DVEC_SIZE_X=" + support::cpp11::to_string(vec_size_x));
71 build_opts.add_option("-DVEC_SIZE_LEFTOVER_X=" + support::cpp11::to_string(vec_size_x_leftovers));
72 build_opts.add_option("-DVEC_SIZE_Y=" + support::cpp11::to_string(vec_size_y));
73 build_opts.add_option("-DVEC_SIZE_LEFTOVER_Y=" + support::cpp11::to_string(vec_size_y_leftovers));
74
75 _kernel = create_kernel(compile_context, "transpose", build_opts.options());
76
77 // Configure kernel window
78 Window win = calculate_max_window(*src, Steps(vec_size_x, vec_size_y));
79 ICLKernel::configure_internal(win, cl::NDRange(2, 8));
80 ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
81 }
82
validate(const ITensorInfo * src,const ITensorInfo * dst)83 Status ClTransposeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
84 {
85 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
86 ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
87 ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
88 ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->num_dimensions() > 2, "Transpose up to 2-D src tensor is supported");
89
90 // Validate configured dst
91 if(dst->total_size() != 0)
92 {
93 const TensorInfo dst_info = src->clone()->set_tensor_shape(misc::shape_calculator::compute_transposed_shape(*src));
94 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(dst, &dst_info);
95
96 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
97 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
98 }
99
100 return Status{};
101 }
102
run_op(ITensorPack & tensors,const Window & window,cl::CommandQueue & queue)103 void ClTransposeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
104 {
105 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
106 ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
107
108 const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
109 auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
110
111 Window slice = window.first_slice_window_2D();
112
113 do
114 {
115 unsigned int idx = 0;
116 add_2D_tensor_argument(idx, src, slice);
117 add_2D_tensor_argument(idx, dst, slice);
118 enqueue(queue, *this, slice, lws_hint());
119 }
120 while(window.slide_window_slice_2D(slice));
121 }
122 } // namespace kernels
123 } // namespace opencl
124 } // namespace arm_compute