xref: /aosp_15_r20/external/ComputeLibrary/src/gpu/cl/kernels/ClReshapeKernel.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
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 "src/gpu/cl/kernels/ClReshapeKernel.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/CL/OpenCL.h"
30 #include "arm_compute/core/Helpers.h"
31 #include "arm_compute/core/TensorInfo.h"
32 #include "arm_compute/core/Utils.h"
33 #include "src/core/CL/CLValidate.h"
34 #include "src/core/helpers/WindowHelpers.h"
35 #include "support/Cast.h"
36 
37 #include <string>
38 
39 /** [ClReshapeKernel Kernel] **/
40 namespace arm_compute
41 {
42 namespace opencl
43 {
44 namespace kernels
45 {
46 namespace
47 {
validate_arguments(const ITensorInfo * src,const ITensorInfo * dst)48 Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
49 {
50     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
51     ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
52     ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
53 
54     if(dst->tensor_shape().total_size() != 0)
55     {
56         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
57         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
58         ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() != dst->tensor_shape().total_size());
59     }
60 
61     return Status{};
62 }
63 } // namespace
64 
ClReshapeKernel()65 ClReshapeKernel::ClReshapeKernel()
66 {
67     _type = CLKernelType::ELEMENTWISE;
68 }
69 
configure(const CLCompileContext & compile_context,const ITensorInfo * src,ITensorInfo * dst)70 void ClReshapeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
71 {
72     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
73     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst));
74 
75     auto padding_info = get_padding_info({ src, dst });
76 
77     // Create kernel
78     std::set<std::string> build_opts = { "-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(src->element_size()) };
79     _kernel                          = create_kernel(compile_context, "reshape_layer", build_opts);
80 
81     // Add static arguments
82     const cl_int2 src_shape =
83     {
84         {
85             static_cast<cl_int>(src->tensor_shape()[0]),
86             static_cast<cl_int>(src->tensor_shape()[1])
87         }
88     };
89     const cl_int2 dst_shape =
90     {
91         {
92             static_cast<cl_int>(dst->tensor_shape()[0]),
93             static_cast<cl_int>(dst->tensor_shape()[1])
94         }
95     };
96     unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the src and dst parameters
97     _kernel.setArg<cl_int2>(idx++, src_shape);
98     _kernel.setArg<cl_int2>(idx++, dst_shape);
99 
100     // Configure kernel window
101     Window win = calculate_max_window(*dst);
102     ICLKernel::configure_internal(win);
103 
104     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
105 }
106 
validate(const ITensorInfo * src,const ITensorInfo * dst)107 Status ClReshapeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
108 {
109     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst));
110 
111     return Status{};
112 }
113 
run_op(ITensorPack & tensors,const Window & window,cl::CommandQueue & queue)114 void ClReshapeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
115 {
116     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
117     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
118 
119     Window window_collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
120     Window slice            = window_collapsed.first_slice_window_3D();
121 
122     const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
123     auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
124 
125     // Set srcs
126     unsigned int idx = 0;
127     add_3D_tensor_argument(idx, src, window_collapsed);
128     add_3D_tensor_argument(idx, dst, window_collapsed);
129     enqueue(queue, *this, slice, lws_hint());
130 }
131 } // namespace kernels
132 } // namespace opencl
133 } // namespace arm_compute
134 /** [ClReshapeKernel Kernel] **/
135