1 /*
2 * Copyright (c) 2019-2022 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/gemm/ClGemmHelpers.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/OpenCL.h"
29 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
30
31 #include <utility>
32
33 namespace arm_compute
34 {
35 namespace opencl
36 {
37 namespace kernels
38 {
39 namespace gemm
40 {
configure_lhs_rhs_info(unsigned int m,unsigned int n,unsigned int m0,unsigned int n0,unsigned int k0,unsigned int v0,unsigned int h0,bool lhs_interleave,bool rhs_interleave,bool lhs_transpose,bool rhs_transpose,bool export_to_cl_image)41 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_lhs_rhs_info(unsigned int m, unsigned int n, unsigned int m0, unsigned int n0, unsigned int k0, unsigned int v0, unsigned int h0,
42 bool lhs_interleave, bool rhs_interleave, bool lhs_transpose, bool rhs_transpose, bool export_to_cl_image)
43 {
44 ARM_COMPUTE_ERROR_ON(m0 == 0 || n0 == 0);
45 v0 = std::max(std::min(static_cast<int>(m / m0), static_cast<int>(v0)), static_cast<int>(1));
46 h0 = std::max(std::min(static_cast<int>(n / n0), static_cast<int>(h0)), static_cast<int>(1));
47
48 const GEMMLHSMatrixInfo lhs_info(m0, k0, v0, lhs_transpose, lhs_interleave);
49 const GEMMRHSMatrixInfo rhs_info(n0, k0, h0, rhs_transpose, rhs_interleave, export_to_cl_image);
50
51 return std::make_pair(lhs_info, rhs_info);
52 }
53
select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo,GEMMRHSMatrixInfo> info_img,std::pair<GEMMLHSMatrixInfo,GEMMRHSMatrixInfo> info_buf,unsigned int n,unsigned int k,unsigned int b,DataType data_type)54 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_img,
55 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_buf,
56 unsigned int n, unsigned int k, unsigned int b, DataType data_type)
57 {
58 const TensorInfo tensor_rhs_info(TensorShape(n, k, b), 1, data_type);
59 const TensorShape shape = misc::shape_calculator::compute_rhs_reshaped_shape(tensor_rhs_info, info_img.second);
60 const TensorInfo tensor_reshaped_info(shape, 1, data_type);
61
62 if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, info_img.second)))
63 {
64 return info_img;
65 }
66 else
67 {
68 return info_buf;
69 }
70 }
71
update_padding_for_cl_image(ITensorInfo * tensor)72 void update_padding_for_cl_image(ITensorInfo *tensor)
73 {
74 constexpr unsigned int num_floats_per_pixel = 4;
75
76 const unsigned int stride_y_in_elements = tensor->strides_in_bytes()[1] / tensor->element_size();
77 const unsigned int pixel_alignment = get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device());
78
79 ARM_COMPUTE_ERROR_ON_MSG(pixel_alignment == 0, "Cannot retrieve cl_image pitch alignment");
80 if(pixel_alignment == 0)
81 {
82 return;
83 }
84
85 const unsigned int row_pitch_alignment = pixel_alignment * num_floats_per_pixel;
86 const unsigned int round_up_width = ((stride_y_in_elements + row_pitch_alignment - 1) / row_pitch_alignment) * row_pitch_alignment;
87 const unsigned int padding = round_up_width - stride_y_in_elements;
88
89 tensor->extend_padding(PaddingSize(0, tensor->padding().right + padding, 0, 0));
90 }
91
validate_image2d_support_on_rhs(const ITensorInfo & tensor_reshaped_info,const GEMMRHSMatrixInfo & rhs_info)92 Status validate_image2d_support_on_rhs(const ITensorInfo &tensor_reshaped_info, const GEMMRHSMatrixInfo &rhs_info)
93 {
94 if(rhs_info.export_to_cl_image)
95 {
96 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((rhs_info.n0 == 2) || (rhs_info.n0 == 3)) && rhs_info.transpose == false, "Export to cl_image only supported with n0 = 4, 8 or 16");
97 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((rhs_info.k0 == 2) || (rhs_info.k0 == 3)) && rhs_info.transpose == true, "Export to cl_image only supported with k0 = 4, 8 or 16");
98 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(&tensor_reshaped_info, DataType::F32, DataType::F16);
99 ARM_COMPUTE_RETURN_ERROR_ON_MSG(!image2d_from_buffer_supported(CLKernelLibrary::get().get_device()), "The extension cl_khr_image2d_from_buffer is not supported on the target platform");
100 ARM_COMPUTE_RETURN_ERROR_ON_MSG(get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) == 0, "Impossible to retrieve the cl_image pitch alignment");
101
102 // Check the width and height of the output tensor.
103 // Since we cannot create a 3d image from a buffer, the third dimension is collapsed on the second dimension
104 const size_t max_image_w = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>();
105 const size_t max_image_h = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>();
106
107 ARM_COMPUTE_RETURN_ERROR_ON_MSG(tensor_reshaped_info.tensor_shape()[0] > max_image_w * 4, "Not supported width for cl_image");
108 ARM_COMPUTE_RETURN_ERROR_ON_MSG(tensor_reshaped_info.tensor_shape()[1] * tensor_reshaped_info.tensor_shape()[2] > max_image_h, "Not supported height for cl_image");
109 }
110
111 return Status{};
112 }
113
is_mmul_kernel_preferred(const unsigned int m,const unsigned int n,const unsigned int k,const unsigned int b,const DataType data_type,unsigned int & best_m0,unsigned int & best_n0)114 bool is_mmul_kernel_preferred(const unsigned int m, const unsigned int n, const unsigned int k, const unsigned int b,
115 const DataType data_type, unsigned int &best_m0, unsigned int &best_n0)
116 {
117 ARM_COMPUTE_UNUSED(n, k, b, data_type);
118
119 const unsigned int mmul_k0 = 4;
120 best_m0 = 4;
121 best_n0 = 4;
122
123 const unsigned int ceil_to_multiple_m_m0 = ceil_to_multiple(m, best_m0);
124 const unsigned int m_div_m0 = ceil_to_multiple_m_m0 / best_m0;
125 const unsigned int ceil_to_multiple_m_div_m0_mmul_k0 = ceil_to_multiple(m_div_m0, mmul_k0);
126 const unsigned int gws_y = ceil_to_multiple_m_div_m0_mmul_k0 / mmul_k0;
127
128 return ((k % mmul_k0) == 0) && (gws_y > 4);
129 }
130 } // namespace gemm
131 } // namespace kernels
132 } // namespace opencl
133 } // namespace arm_compute
134