Compute Library
 21.11
ClGemmReshapeRhsMatrixKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 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  */
25 
32 #include "arm_compute/core/Utils.h"
35 #include "src/core/CL/CLValidate.h"
39 #include "support/Cast.h"
40 #include "support/StringSupport.h"
41 
42 namespace arm_compute
43 {
44 namespace opencl
45 {
46 namespace kernels
47 {
48 namespace
49 {
50 Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const GEMMRHSMatrixInfo &rhs_info)
51 {
52  ARM_COMPUTE_RETURN_ERROR_ON(rhs_info.n0 == 0);
53  ARM_COMPUTE_RETURN_ERROR_ON(rhs_info.k0 == 0);
54  ARM_COMPUTE_RETURN_ERROR_ON(rhs_info.h0 == 0);
55  ARM_COMPUTE_RETURN_ERROR_ON_MSG(((rhs_info.n0 & (rhs_info.n0 - 1)) && rhs_info.n0 != 3), "Only 2,3,4,8,16 are supported for n0");
56  ARM_COMPUTE_RETURN_ERROR_ON_MSG(((rhs_info.k0 & (rhs_info.k0 - 1)) && (rhs_info.k0 != 1) && (rhs_info.k0 != 3)), "Only 1,2,3,4,8,16 are supported for k0");
57  ARM_COMPUTE_RETURN_ERROR_ON(rhs_info.n0 > 16);
58  ARM_COMPUTE_RETURN_ERROR_ON(rhs_info.k0 > 16);
59  ARM_COMPUTE_RETURN_ERROR_ON((rhs_info.k0 == 1) && (rhs_info.transpose));
60 
63 
64  if(rhs_info.export_to_cl_image)
65  {
66  const TensorInfo tensor_reshaped_info(misc::shape_calculator::compute_rhs_reshaped_shape(*src, rhs_info), 1, src->data_type());
68  }
69 
70  if(dst->total_size() != 0)
71  {
75  }
76 
77  return Status{};
78 }
79 
80 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst, const GEMMRHSMatrixInfo &rhs_info)
81 {
82  const unsigned int num_elems_processed_per_iteration_x = rhs_info.n0;
83  const unsigned int num_elems_processed_per_iteration_y = rhs_info.k0;
84  bool window_changed = false;
85 
86  // dst auto initialization if not yet initialized
87  auto_init_if_empty(*dst, src->clone()->set_tensor_shape(misc::shape_calculator::compute_rhs_reshaped_shape(*src, rhs_info)));
88 
89  // Configure window
90  Window win = calculate_max_window(*src, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
91 
92  AccessWindowRectangle src_access(src, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
93 
94  window_changed = update_window_and_padding(win, src_access);
95 
96  if(rhs_info.export_to_cl_image)
97  {
99  }
100 
101  // Collapse along the Z direction
102  // This collapse needs to be here in order to tune the Z dimension of LWS
103  Window collapsed = win.collapse(win, Window::DimZ);
104 
105  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
106  return std::make_pair(err, collapsed);
107 }
108 } // namespace
109 
111 {
113 }
114 
116 {
118 
119  // Perform validate step
120  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, rhs_info));
121 
122  // Create build options
123  CLBuildOptions build_opts;
124  build_opts.add_option("-DN0=" + support::cpp11::to_string(rhs_info.n0));
125  build_opts.add_option("-DK0=" + support::cpp11::to_string(rhs_info.k0));
126  build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0));
127  build_opts.add_option_if(rhs_info.transpose, "-DTRANSPOSE");
128  build_opts.add_option_if(rhs_info.interleave, "-DINTERLEAVE");
129  build_opts.add_option("-DSRC_HEIGHT=" + support::cpp11::to_string(src->dimension(1)));
130  build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(src->element_size()));
131 
132  std::string kernel_name("gemm_reshape_rhs_matrix_");
133  kernel_name += rhs_info.transpose ? "t" : "nt";
134 
135  // Create kernel
136  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
137 
138  // Configure kernel window
139  auto win_config = validate_and_configure_window(src, dst, rhs_info);
140  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
141  ICLKernel::configure_internal(win_config.second);
142 }
143 
145 {
146  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, rhs_info));
147  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(src->clone().get(), dst->clone().get(), rhs_info).first);
148 
149  return Status{};
150 }
151 
152 void ClGemmReshapeRhsMatrixKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
153 {
156 
157  const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
158  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
159 
161 
163 
164  do
165  {
166  unsigned int idx = 0;
167  add_3D_tensor_argument(idx, src, slice);
168  add_3D_tensor_argument(idx, dst, slice);
169  enqueue(queue, *this, slice, lws_hint());
170  }
171  while(window.slide_window_slice_3D(slice));
172 }
173 } // namespace kernels
174 } // namespace opencl
175 } // namespace arm_compute
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
#define ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(tensor)
Definition: CLValidate.h:35
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(...)
Definition: Validate.h:606
void enqueue(cl::CommandQueue &queue, ICLKernel &kernel, const Window &window, const cl::NDRange &lws_hint=CLKernelLibrary::get().default_ndrange(), bool use_dummy_work_items=false)
Add the kernel to the command queue with the given window.
Definition: ICLKernel.cpp:32
const StringSet & options() const
Gets the current options list set.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:318
#define ARM_COMPUTE_RETURN_ON_ERROR(status)
Checks if a status contains an error and returns it.
Definition: Error.h:204
std::string to_string(T &&value)
Convert integer and float values to string.
unsigned int h0
Number of horizontal blocks of size (k0xn0) stored on the same output row.
Definition: Types.h:1962
void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue...
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
Status class.
Definition: Error.h:52
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
void configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, const GEMMRHSMatrixInfo &rhs_info)
Initialise the kernel&#39;s input and output.
void update_padding_for_cl_image(ITensorInfo *tensor)
Update padding required to export the OpenCL buffer to OpenCL image2d.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(...)
Definition: Validate.h:284
void add_3D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 3D tensor&#39;s parameters to the object&#39;s kernel&#39;s arguments starting from the index idx...
Definition: ICLKernel.h:214
bool transpose
True if the (k0xn0) block has to be transposed before been stored.
Definition: Types.h:1963
SimpleTensor< float > src
Definition: DFT.cpp:155
Copyright (c) 2017-2021 Arm Limited.
void add_option(std::string option)
Adds option to the existing build option list.
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
unsigned int k0
Number of partial accumulations performed by the matrix multiplication.
Definition: Types.h:1961
cl::Kernel create_kernel(const CLCompileContext &ctx, const std::string &kernel_name, const std::set< std::string > &build_opts=std::set< std::string >())
Creates an opencl kernel using a compile context.
Definition: CLHelpers.cpp:391
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
Status validate_image2d_support_on_rhs(const ITensorInfo &tensor_reshaped_info, const GEMMRHSMatrixInfo &rhs_info)
Utility function to validate the image2d OpenCL object support on the RHS reshaped matrix...
GEMM RHS (Right Hand Side) matrix information.
Definition: Types.h:1953
static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const GEMMRHSMatrixInfo &rhs_info)
Static function to check if given info will lead to a valid configuration.
unsigned int n0
Number of columns processed by the matrix multiplication.
Definition: Types.h:1960
TensorShape compute_rhs_reshaped_shape(const ITensorInfo &a, const GEMMRHSMatrixInfo &rhs_info)
Calculate the Right Hand Side matrix reshaped shape.
bool auto_init_if_empty(ITensorInfo &info, const TensorShape &shape, int num_channels, DataType data_type, QuantizationInfo quantization_info=QuantizationInfo())
Auto initialize the tensor info (shape, number of channels and data type) if the current assignment i...
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
virtual size_t element_size() const =0
Element size in bytes calculated as data_size() * num_channels()
Elementeise CL kernel type.
Definition: CLTypes.h:84
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:335
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
CLCompileContext class.
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
static constexpr size_t DimZ
Alias for dimension 2 also known as Z dimension.
Definition: Window.h:47
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
std::string get_cl_unsigned_type_from_element_size(size_t element_size)
Translates the element size to an unsigned integer data type.
Definition: CLHelpers.cpp:105
Wrapper to configure the Khronos OpenCL C++ header.
bool interleave
True if the h0 (k0xn0) blocks have to be interleaved in the output row.
Definition: Types.h:1964
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
Tensor packing service.
Definition: ITensorPack.h:39
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
std::string kernel_name
Describe a multidimensional execution window.
Definition: Window.h:39
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)