Compute Library
 22.05
ClGemmLowpMatrixMultiplyReshapedKernel.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2019-2021 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 
38 
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 using namespace misc::shape_calculator;
49 
50 namespace
51 {
52 using ElementsProcessed = Steps;
53 
54 Status validate_arguments(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst,
55  const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
56 {
57  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src0, src1, dst);
60  ARM_COMPUTE_RETURN_ERROR_ON_MSG(src0->num_dimensions() > 4, "The number of dimensions for the LHS matrix must be <= 4");
61  ARM_COMPUTE_RETURN_ERROR_ON_MSG(src1->num_dimensions() > 3, "The number of dimensions for the RHS matrix must be <= 3");
62  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.transpose);
63  ARM_COMPUTE_RETURN_ERROR_ON(!rhs_info.transpose);
64  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.k0 != rhs_info.k0);
65  ARM_COMPUTE_RETURN_ERROR_ON_MSG(((lhs_info.k0 & (lhs_info.k0 - 1)) && lhs_info.k0 != 3), "Only 2,3,4,8,16 are supported for k0");
66  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.k0 > 16);
67  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.m0 < 2 || lhs_info.m0 > 8);
68  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");
69  ARM_COMPUTE_RETURN_ERROR_ON_MSG(rhs_info.export_to_cl_image, "Export to CLImage not supported for quantized GEMM");
70 
71  const int m = gemm_info.m();
72  const int n = gemm_info.n();
73  const int k = gemm_info.k();
74 
75  TensorShape tensor_shape0{ src0->tensor_shape() };
76  tensor_shape0.set(0, k);
77  tensor_shape0.set(1, m);
78 
79  TensorShape tensor_shape1{ src1->tensor_shape() };
80  tensor_shape1.set(0, n);
81  tensor_shape1.set(1, k);
82 
83  const TensorInfo tensor_info0 = src0->clone()->set_tensor_shape(tensor_shape0);
84  const TensorInfo tensor_info1 = src1->clone()->set_tensor_shape(tensor_shape1);
85 
86  const TensorInfo tensor_info_reshaped0 = src0->clone()->set_tensor_shape(compute_lhs_reshaped_shape(tensor_info0, lhs_info));
87  const TensorInfo tensor_info_reshaped1 = src1->clone()->set_tensor_shape(compute_rhs_reshaped_shape(tensor_info1, rhs_info));
88 
89  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src0, &tensor_info_reshaped0);
90  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(src1, &tensor_info_reshaped1);
91 
92  if(dst->total_size() != 0)
93  {
94  const TensorInfo tensor_info_dst = dst->clone()->set_tensor_shape(compute_mm_shape(*src0, *src1, gemm_info));
97  }
98 
99  return Status{};
100 }
101 
102 std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst,
103  const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info,
104  ElementsProcessed &num_elements_processed)
105 {
106  unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0];
107  unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1];
108  bool reinterpret_output_as_3d = (gemm_info.depth_output_gemm3d() != 0);
109 
110  // dst tensor auto initialization if not yet initialized
111  auto_init_if_empty(*dst, src0->clone()->set_tensor_shape(compute_mm_shape(*src0, *src1, gemm_info)).set_data_type(DataType::S32));
112 
113  TensorInfo tmp_info(*dst);
114  if(reinterpret_output_as_3d)
115  {
116  // Since the dst tensor has to be reinterpreted as 3D and the execute window is based on a 2D GEMM,
117  // the window needs to be constructed on the 2D collapsed version of the tensor
118  TensorShape tmp_shape(dst->tensor_shape());
119  tmp_shape.collapse(2U, 1U);
120  tmp_info.set_tensor_shape(tmp_shape);
121  }
122 
123  // Configure kernel window
124  num_elems_processed_per_iteration_x = rhs_info.n0;
125  num_elems_processed_per_iteration_y = lhs_info.m0;
126  Window win = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
127 
128  // Collapse along the Z direction
129  // This collapse needs to be here in order to tune the Z dimension of LWS
130  Window collapsed = win;
131  const unsigned int dimension_to_collapse = std::min(static_cast<unsigned int>(dst->num_dimensions()), 2u);
132  collapsed = win.collapse(win, dimension_to_collapse);
133 
134  return std::make_pair(Status{}, collapsed);
135 }
136 } // namespace
137 
139 {
140  _type = CLKernelType::GEMM;
141 }
142 
144  const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
145 {
146  ARM_COMPUTE_ERROR_ON_NULLPTR(src0, src1, dst);
147  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src0, src1, dst, lhs_info, rhs_info, gemm_info));
148 
149  _reinterpret_output_as_3d = (gemm_info.depth_output_gemm3d() != 0);
150  _k = gemm_info.k();
151  _use_dummy_work_items = preferred_dummy_work_items_support(CLKernelLibrary::get().get_device());
152 
153  // Check if we need to slide the matrix B
154  const unsigned int num_dimensionssrc0 = src0->num_dimensions();
155  _slide_matrix_b = (src1->num_dimensions() >= num_dimensionssrc0);
156 
157  auto padding_info = get_padding_info({ src0, src1, dst });
158  ElementsProcessed num_elements_processed{};
159 
160  // Configure kernel window
161  auto win_config = validate_and_configure_window(src0, src1, dst, lhs_info, rhs_info, gemm_info, num_elements_processed);
162  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
163  ICLKernel::configure_internal(win_config.second);
164 
165  // Calculate partial (store instead of load) M0 and partial N0 for the partial blocks at the end of a row/column if any. This is to avoid padding.
166  const unsigned int internal_m = _reinterpret_output_as_3d ? gemm_info.m() : dst->dimension(1);
167 
168  const unsigned int partial_store_m0 = internal_m % lhs_info.m0;
169  const unsigned int partial_store_n0 = gemm_info.n() % rhs_info.n0;
170 
171  // Create build options
172  CLBuildOptions build_opts;
173  build_opts.add_option_if(_reinterpret_output_as_3d, "-DREINTERPRET_OUTPUT_AS_3D");
174  build_opts.add_option_if(_reinterpret_output_as_3d, "-DHEIGHT_GEMM3D=" + support::cpp11::to_string(dst->dimension(1)));
175  build_opts.add_option_if(_reinterpret_output_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(dst->dimension(2)));
176  build_opts.add_option_if(!_slide_matrix_b, "-DMATRIX_B_DEPTH=" + support::cpp11::to_string(src1->dimension(2)));
177  build_opts.add_option_if(lhs_info.interleave, "-DLHS_INTERLEAVE");
178  build_opts.add_option_if(rhs_info.interleave, "-DRHS_INTERLEAVE");
179  build_opts.add_option_if(_use_dummy_work_items, "-DDUMMY_WORK_ITEMS");
180  build_opts.add_option("-DM=" + support::cpp11::to_string(gemm_info.m()));
181  build_opts.add_option("-DN=" + support::cpp11::to_string(gemm_info.n()));
182  build_opts.add_option("-DM0=" + support::cpp11::to_string(lhs_info.m0));
183  build_opts.add_option("-DN0=" + support::cpp11::to_string(rhs_info.n0));
184  build_opts.add_option("-DK0=" + support::cpp11::to_string(lhs_info.k0));
185  build_opts.add_option("-DV0=" + support::cpp11::to_string(lhs_info.v0));
186  build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0));
187  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src0->data_type()));
188  build_opts.add_option("-DACC_DATA_TYPE=" + get_cl_dot8_acc_type_from_data_type(src0->data_type()));
189  build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string(partial_store_m0));
190  build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(partial_store_n0));
191 
192  std::string kernel_name("gemmlowp_mm_reshaped_");
193  kernel_name += lhs_info.transpose ? "lhs_t_" : "lhs_nt_";
194  kernel_name += rhs_info.transpose ? "rhs_t" : "rhs_nt";
195 
196  // Create kernel
197  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
198 
199  // Set config_id for enabling LWS tuning
200  _config_id = kernel_name;
201  _config_id += "_";
202  _config_id += dot8_supported(CLKernelLibrary::get().get_device()) ? "_dot8" : "";
203  _config_id += "_";
204  _config_id += (_reinterpret_output_as_3d ? "3do_" : "");
205  _config_id += support::cpp11::to_string(dst->dimension(1));
206  _config_id += "_";
207  _config_id += support::cpp11::to_string(dst->dimension(0));
208  _config_id += "_";
209  _config_id += support::cpp11::to_string(gemm_info.k());
210  _config_id += "_";
211  _config_id += support::cpp11::to_string(dst->dimension(2));
212  _config_id += "_";
213  _config_id += support::cpp11::to_string(lhs_info.m0);
214  _config_id += "_";
215  _config_id += support::cpp11::to_string(rhs_info.n0);
216  _config_id += "_";
217  _config_id += support::cpp11::to_string(lhs_info.k0);
218  _config_id += "_";
219  _config_id += support::cpp11::to_string(lhs_info.v0);
220  _config_id += "_";
221  _config_id += support::cpp11::to_string(rhs_info.h0);
222  _config_id += "_";
223  _config_id += support::cpp11::to_string(lhs_info.interleave);
224  _config_id += "_";
225  _config_id += support::cpp11::to_string(rhs_info.interleave);
226 
228 }
229 
231  const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
232 {
233  ElementsProcessed num_elements_processed{};
234  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src0, src1, dst, lhs_info, rhs_info, gemm_info));
236  src1->clone().get(),
237  dst->clone().get(),
238  lhs_info,
239  rhs_info,
240  gemm_info,
241  num_elements_processed)
242  .first);
243 
244  return Status{};
245 }
246 
247 void ClGemmLowpMatrixMultiplyReshapedKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
248 {
251 
252  const auto src0 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0));
253  const auto src1 = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1));
254  auto dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
255 
256  if(src1->info()->num_dimensions() < 3)
257  {
258  // The stride_z for matrix B must be zero if we do not slice
259  ARM_COMPUTE_ERROR_ON(src1->info()->strides_in_bytes()[3] != 0);
260  }
261 
263  Window slice_matrix_b = slice;
264 
265  slice_matrix_b.set(Window::DimX, Window::Dimension(0, 1, 1));
266  slice_matrix_b.set(Window::DimY, Window::Dimension(0, 1, 1));
267 
268  if(_reinterpret_output_as_3d)
269  {
270  // Pass bottom paddings to the kernel if the dst has to be reinterpreted as 3D tensor
271  const unsigned int idx0 = 3 * num_arguments_per_2D_tensor() + 4;
272  const unsigned int total_cross_plane_pad = dst->info()->padding().top + dst->info()->padding().bottom;
273  _kernel.setArg<cl_uint>(idx0, static_cast<unsigned int>(total_cross_plane_pad));
274  }
275 
276  do
277  {
278  Window slice_b = slice;
279  // Don't slice matrix B along the z dimension if matrix B has just 2 dimensions and matrix A more than 2
280  // This scenario can happen when the matrix multiplication is used to perform a convolution operation
281  if(!_slide_matrix_b)
282  {
283  slice_b = slice_matrix_b;
284  }
285 
286  unsigned int idx = 0;
287  add_2D_tensor_argument(idx, src0, slice);
288  add_2D_tensor_argument(idx, src1, slice_b);
289  add_2D_tensor_argument(idx, dst, slice);
290  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_k));
291  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src0->info()->strides_in_bytes()[2]));
292  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(src1->info()->strides_in_bytes()[2]));
293  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(dst->info()->strides_in_bytes()[2]));
294  enqueue(queue, *this, slice, lws_hint(), _use_dummy_work_items);
295  }
296  while(window.slide_window_slice_3D(slice));
297 }
298 } // namespace kernels
299 } // namespace opencl
300 } // namespace arm_compute
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
const Window & window() const
The maximum window the kernel can be executed on.
Definition: IKernel.cpp:28
bool dot8_supported(const cl::Device &device)
Helper function to check whether the cl_arm_integer_dot_product_int8 extension is supported...
Definition: CLHelpers.cpp:241
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...
virtual size_t dimension(size_t index) const =0
Return the size of the requested dimension.
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.
unsigned int v0
Number of vertical blocks of size (m0xk0) stored on the same output row.
Definition: Types.h:2063
bool preferred_dummy_work_items_support(const cl::Device &device)
Helper function to check if "dummy work-items" are preferred to have a power of two NDRange In case d...
Definition: CLHelpers.cpp:363
GEMM reshape information class.
Definition: Types.h:1910
std::string get_cl_dot8_acc_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL dot8 accumulator type.
Definition: CLHelpers.cpp:175
#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.
virtual DataType data_type() const =0
Data type used for each element of the tensor.
TensorShape compute_mm_shape(const ITensorInfo &input0, const ITensorInfo &input1, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info)
Calculate the matrix multiplication output shape of two tensors.
unsigned int h0
Number of horizontal blocks of size (k0xn0) stored on the same output row.
Definition: Types.h:2078
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
GEMM LHS (Left Hand Side) matrix information.
Definition: Types.h:2054
Store the tensor&#39;s metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Describe one of the image&#39;s dimensions with a start, end and step.
Definition: Window.h:79
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context...
void configure(const CLCompileContext &compile_context, const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
Initialise the kernel&#39;s input and dst.
Status class.
Definition: Error.h:52
Status validate_arguments(const ITensorInfo *src, const ITensorInfo *weights, const ITensorInfo *dst, const PadStrideInfo &conv_info)
#define ARM_COMPUTE_RETURN_ERROR_ON(cond)
If the condition is true, an error is returned.
Definition: Error.h:296
bool transpose
True if the (k0xn0) block has to be transposed before been stored.
Definition: Types.h:2079
bool interleave
True if the v0 (m0xk0) blocks have to be interleaved in the output row.
Definition: Types.h:2065
Copyright (c) 2017-2022 Arm Limited.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
1 channel, 1 S32 per channel
void add_option(std::string option)
Adds option to the existing build option list.
bool transpose
True if the (m0xk0) block has to be transposed before been stored.
Definition: Types.h:2064
int n() const
Number of matrix B columns.
Definition: Types.h:1948
const ITensor * get_const_tensor(int id) const
Get constant tensor of a given id.
Definition: ITensorPack.cpp:54
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
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
TensorShape compute_lhs_reshaped_shape(const ITensorInfo &a, const GEMMLHSMatrixInfo &lhs_info, bool reinterpret_input_as_3d=false)
Calculate the Left Hand Side matrix reshaped shape.
GEMM RHS (Right Hand Side) matrix information.
Definition: Types.h:2069
unsigned int n0
Number of columns processed by the matrix multiplication.
Definition: Types.h:2076
quantized, asymmetric fixed-point 8-bit number unsigned
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:39
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.
int k() const
Number of matrix A columns or matrix B rows.
Definition: Types.h:1956
void add_option_if(bool cond, std::string option)
Adds option if a given condition is true;.
void set(size_t dimension, const Dimension &dim)
Set the values of a given dimension.
Definition: Window.inl:49
bool slide_window_slice_3D(Window &slice) const
Slide the passed 3D window slice.
Definition: Window.h:348
#define ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(k)
Definition: Validate.h:915
bool has_padding_changed(const std::unordered_map< const ITensorInfo *, PaddingSize > &padding_map)
Check if the previously stored padding info has changed after configuring a kernel.
Definition: Utils.cpp:601
CLCompileContext class.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
std::pair< Status, Window > validate_and_configure_window(ITensorInfo *src, ITensorInfo *dst)
ITensor * get_tensor(int id)
Get tensor of a given id from the pac.
Definition: ITensorPack.cpp:64
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:439
int m() const
Number of matrix A rows.
Definition: Types.h:1940
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(...)
Definition: Validate.h:541
#define ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(t, c,...)
Definition: Validate.h:788
std::unordered_map< const ITensorInfo *, PaddingSize > get_padding_info(std::initializer_list< const ITensorInfo *> infos)
Stores padding information before configuring a kernel.
Definition: Utils.cpp:586
Wrapper to configure the Khronos OpenCL C++ header.
int depth_output_gemm3d() const
Depth (third dimension) of the output tensor to be used with the GEMM3D kernel.
Definition: Types.h:1983
bool interleave
True if the h0 (k0xn0) blocks have to be interleaved in the output row.
Definition: Types.h:2080
#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
unsigned int k0
Number of partial accumulations performed by the matrix multiplication.
Definition: Types.h:2062
static Status validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info)
Static function to check if given info will lead to a valid configuration.
unsigned int m0
Number of rows processed by the matrix multiplication.
Definition: Types.h:2061
quantized, asymmetric fixed-point 8-bit number signed
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:304
std::string kernel_name
Describe a multidimensional execution window.
Definition: Window.h:39
Convolution using GEMM.
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)