• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/ClWeightsReshapeKernel.h"
25 #include "arm_compute/core/CL/ICLTensor.h"
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
28 #include "src/core/helpers/AutoConfiguration.h"
29 #include "src/core/helpers/WindowHelpers.h"
30 #include "support/Cast.h"
31 #include "support/StringSupport.h"
32 
33 namespace arm_compute
34 {
35 using namespace misc::shape_calculator;
36 namespace opencl
37 {
38 namespace kernels
39 {
40 namespace
41 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * biases,const ITensorInfo * output,unsigned int num_groups)42 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *biases, const ITensorInfo *output, unsigned int num_groups)
43 {
44     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
45     ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
46     ARM_COMPUTE_RETURN_ERROR_ON(num_groups == 0);
47     ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::NHWC && num_groups > 1);
48     ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4 && num_groups > 1);
49     ARM_COMPUTE_RETURN_ERROR_ON((input->dimension(3) % num_groups) != 0);
50 
51     if(biases != nullptr)
52     {
53         ARM_COMPUTE_RETURN_ERROR_ON(!is_data_type_float(input->data_type()));
54         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, biases);
55         ARM_COMPUTE_RETURN_ERROR_ON((input->num_dimensions() == 4) && (biases->num_dimensions() != 1));
56         ARM_COMPUTE_RETURN_ERROR_ON((input->num_dimensions() == 5) && (biases->num_dimensions() != 2));
57         ARM_COMPUTE_RETURN_ERROR_ON((input->num_dimensions() == 4) && (biases->dimension(0) != input->tensor_shape()[3]));
58         ARM_COMPUTE_RETURN_ERROR_ON((input->num_dimensions() == 5) && (biases->dimension(0) != input->tensor_shape()[3] || biases->dimension(1) != input->tensor_shape()[4]));
59     }
60 
61     // Checks performed when output is configured
62     if(output->total_size() != 0)
63     {
64         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_weights_reshaped_shape(*input, biases != nullptr, num_groups));
65         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
66         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
67     }
68 
69     return Status{};
70 }
71 } // namespace
72 
ClWeightsReshapeKernel()73 ClWeightsReshapeKernel::ClWeightsReshapeKernel()
74 {
75     _type = CLKernelType::ELEMENTWISE;
76 }
77 
configure(const ClCompileContext & compile_context,const ITensorInfo * src,const ITensorInfo * biases,ITensorInfo * dst,unsigned int num_groups)78 void ClWeightsReshapeKernel::configure(const ClCompileContext &compile_context, const ITensorInfo *src, const ITensorInfo *biases, ITensorInfo *dst, unsigned int num_groups)
79 {
80     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
81 
82     // Output tensor auto inizialitation if not yet initialized
83     auto_init_if_empty(*dst, src->clone()->set_tensor_shape(compute_weights_reshaped_shape(*src, (biases != nullptr), num_groups)));
84 
85     // Perform validation step
86     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, biases, dst, num_groups));
87     auto padding_info = get_padding_info({ src, biases, dst });
88 
89     const DataType data_type = src->data_type();
90 
91     // Create build options
92     CLBuildOptions build_opts;
93     build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(data_type)));
94     build_opts.add_option("-DNUM_GROUPS=" + support::cpp11::to_string(num_groups));
95     build_opts.add_option_if(biases != nullptr, "-DHAS_BIAS");
96 
97     // Create kernel
98     _kernel = create_kernel(compile_context, "reshape_to_columns", build_opts.options());
99 
100     // Configure window
101     Window win = calculate_max_window(*src, Steps());
102     ICLKernel::configure_internal(win);
103 
104     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
105 }
106 
validate(const ITensorInfo * src,const ITensorInfo * biases,const ITensorInfo * dst,unsigned int num_groups)107 Status ClWeightsReshapeKernel::validate(const ITensorInfo *src, const ITensorInfo *biases, const ITensorInfo *dst, unsigned int num_groups)
108 {
109     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, biases, dst, num_groups));
110     return Status{};
111 }
112 
run_op(ITensorPack & tensors,const Window & window,cl::CommandQueue & queue)113 void ClWeightsReshapeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
114 {
115     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
116     ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
117 
118     auto src    = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
119     auto biases = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_BIAS));
120     auto dst    = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
121 
122     Window out_window;
123     out_window.use_tensor_dimensions(dst->info()->tensor_shape());
124 
125     Window in_slice  = window.first_slice_window_3D();
126     Window out_slice = out_window.first_slice_window_2D();
127 
128     Window biases_window;
129     Window biases_slice;
130 
131     unsigned int idx = num_arguments_per_3D_tensor() + num_arguments_per_2D_tensor();
132     idx += (biases != nullptr) ? num_arguments_per_1D_tensor() : 0;
133     _kernel.setArg<cl_uint>(idx++, src->info()->dimension(0));
134     _kernel.setArg<cl_uint>(idx++, src->info()->dimension(1));
135     _kernel.setArg<cl_uint>(idx++, src->info()->dimension(2));
136     _kernel.setArg<cl_uint>(idx++, src->info()->dimension(3));
137     _kernel.setArg<cl_uint>(idx++, dst->info()->strides_in_bytes().z());
138 
139     if(biases != nullptr)
140     {
141         biases_window.use_tensor_dimensions(biases->info()->tensor_shape());
142         biases_slice = biases_window.first_slice_window_1D();
143     }
144 
145     do
146     {
147         // Set arguments
148         unsigned idx = 0;
149         add_3D_tensor_argument(idx, src, in_slice);
150         add_2D_tensor_argument(idx, dst, out_slice);
151         if(biases != nullptr)
152         {
153             add_1D_tensor_argument(idx, biases, biases_slice);
154             ARM_COMPUTE_UNUSED(biases_window.slide_window_slice_1D(biases_slice));
155         }
156 
157         // Run kernel
158         enqueue(queue, *this, in_slice, lws_hint());
159     }
160     while(window.slide_window_slice_4D(in_slice) && out_window.slide_window_slice_2D(out_slice));
161 }
162 } // namespace kernels
163 } // namespace opencl
164 } // namespace arm_compute
165