Compute Library
 21.05
CLGEMMMatrixMultiplyReshapedKernel.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"
36 #include "src/core/CL/CLUtils.h"
37 #include "src/core/CL/CLValidate.h"
42 #include "support/StringSupport.h"
43 
44 #include <cstddef>
45 #include <cstdint>
46 #include <tuple>
47 
48 using namespace arm_compute;
50 
51 namespace arm_compute
52 {
53 class Coordinates;
54 } // namespace arm_compute
55 
56 namespace
57 {
58 using ElementsProcessed = Steps;
59 
60 Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info,
61  const GEMMRHSMatrixInfo &rhs_info,
62  const GEMMKernelInfo &gemm_info)
63 {
64  ARM_COMPUTE_UNUSED(alpha);
65  ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input0, input1, output);
69  ARM_COMPUTE_RETURN_ERROR_ON_MSG(input0->num_dimensions() > 4, "The number of dimensions for the LHS matrix must be <= 4");
70  ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->num_dimensions() > 3, "The number of dimensions for the RHS matrix must be <= 3");
71  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.k0 != rhs_info.k0);
72  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.transpose == rhs_info.transpose);
73  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");
74  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.k0 > 16);
75  ARM_COMPUTE_RETURN_ERROR_ON(lhs_info.m0 < 2 || lhs_info.m0 > 8);
76  ARM_COMPUTE_RETURN_ERROR_ON_MSG((lhs_info.transpose) && ((lhs_info.m0 & (lhs_info.m0 - 1)) && lhs_info.m0 != 3), "Only 2,3,4,8,16 are supported for m0");
77  ARM_COMPUTE_RETURN_ERROR_ON_MSG((rhs_info.transpose) && ((rhs_info.n0 & (rhs_info.n0 - 1)) && rhs_info.n0 != 3), "Only 2,3,4,8,16 are supported for n0");
78  ARM_COMPUTE_RETURN_ERROR_ON_MSG((gemm_info.reinterpret_input_as_3d || gemm_info.depth_output_gemm3d != 0) && (input2 != nullptr)
79  && (!gemm_info.broadcast_bias),
80  "Bias addition only supported with broadcast mode in case the input or output has to be reinterpreted as 3D");
81  ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.fp_mixed_precision && (input0->data_type() == DataType::F32), "Mixed precision only supported for F16 data type");
83 
84  const unsigned int m = gemm_info.m;
85  const unsigned int n = gemm_info.n;
86  const unsigned int k = gemm_info.k;
87 
88  TensorShape tensor_shape0{ input0->tensor_shape() };
89  tensor_shape0.set(0, k);
90  tensor_shape0.set(1, m);
91 
92  TensorShape tensor_shape1{ input1->tensor_shape() };
93  tensor_shape1.set(0, n);
94  tensor_shape1.set(1, k);
95 
96  if(input2 != nullptr && !(helpers::float_ops::is_zero(beta)))
97  {
98  const unsigned int input2_dim0 = input2->dimension(0);
99  const unsigned int input2_dim1 = input2->dimension(1);
100 
102  if(gemm_info.broadcast_bias)
103  {
104  ARM_COMPUTE_RETURN_ERROR_ON_MSG((input2_dim1 != 1 || input2_dim0 != n), "Incorrect dimension of bias matrix which is to be broadcasted");
105  }
106  else
107  {
108  ARM_COMPUTE_RETURN_ERROR_ON_MSG((input2_dim0 != n || input2_dim1 != m), "Incorrect dimension of bias matrix");
109  }
110  }
111 
112  const TensorInfo tensor_info0 = input0->clone()->set_tensor_shape(tensor_shape0);
113  const TensorInfo tensor_info1 = input1->clone()->set_tensor_shape(tensor_shape1);
114 
115  const TensorInfo tensor_info_reshaped0 = input0->clone()->set_tensor_shape(compute_lhs_reshaped_shape(tensor_info0, lhs_info));
116  const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_rhs_reshaped_shape(tensor_info1, rhs_info));
117 
118  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input0, &tensor_info_reshaped0);
119  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, &tensor_info_reshaped1);
120 
121  if(output->total_size() != 0)
122  {
123  const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(compute_mm_shape(*input0, *input1, gemm_info));
124  ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output);
126  }
127 
128  return Status{};
129 }
130 
131 std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input0, ITensorInfo *input1, ITensorInfo *input2, ITensorInfo *output, const GEMMLHSMatrixInfo &lhs_info,
132  const GEMMRHSMatrixInfo &rhs_info,
133  const GEMMKernelInfo &gemm_info, ElementsProcessed &num_elements_processed)
134 {
135  unsigned int &num_elems_processed_per_iteration_x = num_elements_processed[0];
136  unsigned int &num_elems_processed_per_iteration_y = num_elements_processed[1];
137  bool reinterpret_output_as_3d = gemm_info.depth_output_gemm3d != 0;
138 
139  Window win{};
140  Window win_out{};
141  bool window_changed = false;
142 
143  // Output tensor auto initialization if not yet initialized
144  auto_init_if_empty(*output, input0->clone()->set_tensor_shape(compute_mm_shape(*input0, *input1, gemm_info)));
145 
146  TensorInfo tmp_info(*output);
147 
148  if(reinterpret_output_as_3d)
149  {
150  // Since the output tensor has to be reinterpreted as 3D and the execute window is based on a 2D GEMM,
151  // the window needs to be constructed on the 2D collapsed version of the tensor
152  TensorShape tmp_shape(output->tensor_shape());
153  tmp_shape.collapse(2U, 1U);
154  tmp_info.set_tensor_shape(tmp_shape);
155  }
156 
157  // Configure kernel window
158  num_elems_processed_per_iteration_x = rhs_info.n0;
159  num_elems_processed_per_iteration_y = lhs_info.m0;
160 
161  win = calculate_max_window(tmp_info, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
162  win_out = calculate_max_window(*output, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
163 
164  if(input2 != nullptr)
165  {
166  const int bias_processed_per_iteration_x = num_elems_processed_per_iteration_x;
167 
168  const int bias_processed_per_iteration_y = gemm_info.broadcast_bias ? 1 : num_elems_processed_per_iteration_y;
169 
170  AccessWindowStatic input2_access(input2, 0, 0,
171  ceil_to_multiple(input2->dimension(0), bias_processed_per_iteration_x),
172  ceil_to_multiple(input2->dimension(1), bias_processed_per_iteration_y));
173 
174  window_changed = update_window_and_padding(win, input2_access); // window used by the execute_window_loop
175  }
176 
177  // Collapse along the Z direction
178  // This collapse needs to be here in order to tune the Z dimension of LWS
179  Window collapsed = win;
180  const unsigned int dimension_to_collapse = std::min(static_cast<unsigned int>(output->num_dimensions()), 2u);
181  collapsed = win.collapse(win, dimension_to_collapse);
182 
183  Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
184  return std::make_pair(err, collapsed);
185 }
186 } // namespace
187 
189  : _input0(nullptr), _input1(nullptr), _input2(nullptr), _output(nullptr), _slide_matrix_b(true), _reinterpret_output_as_3d(false), _use_dummy_work_items(false), _add_bias(false),
190  _broadcast_bias(false), _export_to_cl_image(false), _k(1)
191 {
192 }
193 
194 void CLGEMMMatrixMultiplyReshapedKernel::configure(const ICLTensor *input0, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, float alpha, float beta,
195  const GEMMLHSMatrixInfo &lhs_info,
196  const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
197 {
198  configure(CLKernelLibrary::get().get_compile_context(), input0, input1, input2, output, alpha, beta, lhs_info, rhs_info, gemm_info);
199 }
200 
201 void CLGEMMMatrixMultiplyReshapedKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input0, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, float alpha,
202  float beta,
203  const GEMMLHSMatrixInfo &lhs_info,
204  const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
205 {
206  ARM_COMPUTE_ERROR_ON_NULLPTR(input0, input1, output);
207 
208  ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input0->info(), input1->info(), (input2 != nullptr ? input2->info() : nullptr), output->info(), alpha, beta, lhs_info, rhs_info, gemm_info));
209 
210  auto padding_info = get_padding_info({ input0, output });
211  _input0 = input0;
212  _input1 = input1;
213  _input2 = helpers::float_ops::is_zero(beta) ? nullptr : input2;
214  _output = output;
215  _reinterpret_output_as_3d = gemm_info.depth_output_gemm3d != 0;
216  _use_dummy_work_items = preferred_dummy_work_items_support(CLKernelLibrary::get().get_device());
217  _add_bias = _input2 != nullptr;
218  _broadcast_bias = gemm_info.broadcast_bias;
219  _export_to_cl_image = rhs_info.export_to_cl_image;
220  _k = gemm_info.k;
221 
222  // Check if we need to slide the matrix B
223  const unsigned int num_dimensions_input0 = _input0->info()->num_dimensions();
224  _slide_matrix_b = (_input1->info()->num_dimensions() >= num_dimensions_input0);
225 
226  ElementsProcessed num_elements_processed{};
227 
228  // Configure kernel window
229  auto win_config = validate_and_configure_window(input0->info(), input1->info(), input2 != nullptr ? input2->info() : nullptr, output->info(), lhs_info, rhs_info, gemm_info, num_elements_processed);
230  ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
231  ICLKernel::configure_internal(win_config.second);
232 
233  const bool enable_mixed_precision = gemm_info.fp_mixed_precision;
234  const DataType data_type = input0->info()->data_type();
235 
236  // 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.
237  const unsigned int internal_m = _reinterpret_output_as_3d ? gemm_info.m : output->info()->dimension(1);
238 
239  const unsigned int partial_store_m0 = internal_m % lhs_info.m0;
240  const unsigned int partial_store_n0 = gemm_info.n % rhs_info.n0;
241 
242  // Create build options
243  CLBuildOptions build_opts;
244  build_opts.add_option_if(!(helpers::float_ops::is_one(alpha)), "-DALPHA=" + float_to_string_with_full_precision(alpha));
245  build_opts.add_option_if(_input2 != nullptr, "-DBETA=" + float_to_string_with_full_precision(beta));
246  build_opts.add_option_if(helpers::float_ops::is_one(beta), "-DUNIT_BETA");
247  build_opts.add_option_if(_reinterpret_output_as_3d, "-DREINTERPRET_OUTPUT_AS_3D");
248  build_opts.add_option_if(_reinterpret_output_as_3d, "-DHEIGHT_GEMM3D=" + support::cpp11::to_string(output->info()->dimension(1)));
249  build_opts.add_option_if(_reinterpret_output_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(output->info()->dimension(2)));
250  build_opts.add_option_if(gemm_info.broadcast_bias, "-DBROADCAST_BIAS");
251  build_opts.add_option_if(!_slide_matrix_b, "-DMATRIX_B_DEPTH=" + support::cpp11::to_string(input1->info()->dimension(2)));
252  build_opts.add_option_if(lhs_info.interleave, "-DLHS_INTERLEAVE");
253  build_opts.add_option_if(rhs_info.interleave, "-DRHS_INTERLEAVE");
254  build_opts.add_option_if(lhs_info.transpose, "-DLHS_TRANSPOSE");
255  build_opts.add_option_if(_use_dummy_work_items, "-DDUMMY_WORK_ITEMS");
256  build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DACTIVATION_TYPE=" + lower_string(string_from_activation_func(gemm_info.activation_info.activation())));
257  build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DA_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.a()));
258  build_opts.add_option_if(gemm_info.activation_info.enabled(), "-DB_VAL=" + float_to_string_with_full_precision(gemm_info.activation_info.b()));
259  build_opts.add_option_if(enable_mixed_precision, "-DMIXED_PRECISION");
260  build_opts.add_option_if(rhs_info.export_to_cl_image, "-DOPENCL_IMAGE_SUPPORT");
261  build_opts.add_option("-DRHS_HEIGHT=" + support::cpp11::to_string(input1->info()->dimension(1)));
262  build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(data_type));
263  build_opts.add_option("-DDATA_TYPE_ACCUMULATOR=" + (enable_mixed_precision ? get_cl_type_from_data_type(DataType::F32) : get_cl_type_from_data_type(data_type)));
264  build_opts.add_option("-DM=" + support::cpp11::to_string(gemm_info.m));
265  build_opts.add_option("-DN=" + support::cpp11::to_string(gemm_info.n));
266  build_opts.add_option("-DK=" + support::cpp11::to_string(gemm_info.k));
267  build_opts.add_option("-DM0=" + support::cpp11::to_string(lhs_info.m0));
268  build_opts.add_option("-DN0=" + support::cpp11::to_string(rhs_info.n0));
269  build_opts.add_option("-DK0=" + support::cpp11::to_string(lhs_info.k0));
270  build_opts.add_option("-DV0=" + support::cpp11::to_string(lhs_info.v0));
271  build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0));
272  build_opts.add_option("-DPARTIAL_STORE_M0=" + support::cpp11::to_string(partial_store_m0));
273  build_opts.add_option("-DPARTIAL_STORE_N0=" + support::cpp11::to_string(partial_store_n0));
274 
275  std::string kernel_name("gemm_mm_reshaped_");
276  kernel_name += lhs_info.transpose ? "lhs_t_" : "lhs_nt_";
277  kernel_name += rhs_info.transpose ? "rhs_t" : "rhs_nt";
278  kernel_name += rhs_info.export_to_cl_image ? "_texture" : "";
279 
280  // Create kernel
281  _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
282 
283  // Set config_id for enabling LWS tuning
284  _config_id = kernel_name;
285  _config_id += "_";
286  _config_id += (_add_bias ? "add_bias_" : "");
287  _config_id += (_broadcast_bias ? "broadcast_bias_" : "");
288  _config_id += (_reinterpret_output_as_3d ? "3do_" : "");
289  _config_id += (gemm_info.activation_info.enabled() ? "fused_activation_" : "");
290  _config_id += lower_string(string_from_data_type(input0->info()->data_type()));
291  _config_id += "_";
292  _config_id += (enable_mixed_precision ? "mixed_precision_" : "");
293  _config_id += support::cpp11::to_string(output->info()->dimension(1));
294  _config_id += "_";
295  _config_id += support::cpp11::to_string(output->info()->dimension(0));
296  _config_id += "_";
297  _config_id += support::cpp11::to_string(gemm_info.k);
298  _config_id += "_";
299  _config_id += support::cpp11::to_string(output->info()->dimension(2));
300  _config_id += "_";
301  _config_id += support::cpp11::to_string(lhs_info.m0);
302  _config_id += "_";
303  _config_id += support::cpp11::to_string(rhs_info.n0);
304  _config_id += "_";
305  _config_id += support::cpp11::to_string(lhs_info.k0);
306  _config_id += "_";
307  _config_id += support::cpp11::to_string(lhs_info.v0);
308  _config_id += "_";
309  _config_id += support::cpp11::to_string(rhs_info.h0);
310  _config_id += "_";
311  _config_id += support::cpp11::to_string(lhs_info.interleave);
312  _config_id += "_";
313  _config_id += support::cpp11::to_string(rhs_info.interleave);
314 
316 }
317 
318 Status CLGEMMMatrixMultiplyReshapedKernel::validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float alpha, float beta,
319  const GEMMLHSMatrixInfo &lhs_info,
320  const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
321 {
322  ElementsProcessed num_elements_processed{};
323  ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input0, input1, input2, output, alpha, beta, lhs_info, rhs_info, gemm_info));
324  ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input0->clone().get(),
325  input1->clone().get(),
326  input2 != nullptr ? input2->clone().get() : nullptr,
327  output->clone().get(),
328  lhs_info,
329  rhs_info,
330  gemm_info,
331  num_elements_processed)
332  .first);
333 
334  return Status{};
335 }
336 
337 void CLGEMMMatrixMultiplyReshapedKernel::run(const Window &window, cl::CommandQueue &queue)
338 {
341 
342  if(_input1->info()->num_dimensions() < 3)
343  {
344  // The stride_z for matrix B must be zero if we do not slice
345  ARM_COMPUTE_ERROR_ON(_input1->info()->strides_in_bytes()[3] != 0);
346  }
347 
349  Window slice_matrix_b = slice;
350 
351  slice_matrix_b.set(Window::DimX, Window::Dimension(0, 1, 1));
352  slice_matrix_b.set(Window::DimY, Window::Dimension(0, 1, 1));
353 
354  const unsigned int total_cross_plane_pad = _output->info()->padding().top + _output->info()->padding().bottom;
355 
356  cl::Image2D input1_image2d;
357 
358  if(_export_to_cl_image)
359  {
360  const TensorShape shape2d(_input1->info()->dimension(0) / 4, _input1->info()->dimension(1) * _input1->info()->dimension(2));
361  const size_t image_row_pitch = _input1->info()->strides_in_bytes()[1];
362 
363  input1_image2d = create_image2d_from_buffer(CLKernelLibrary::get().context(), _input1->cl_buffer(), shape2d, _input1->info()->data_type(), image_row_pitch);
364  }
365 
366  do
367  {
368  Window slice_b = slice;
369  // Don't slice matrix B along the z dimension if matrix B has just 2 dimensions and matrix A more than 2
370  // This scenario can happen when the matrix multiplication is used to perform a convolution operation
371  if(!_slide_matrix_b)
372  {
373  slice_b = slice_matrix_b;
374  }
375 
376  unsigned int idx = 0;
377 
378  // LHS buffer
379  add_2D_tensor_argument(idx, _input0, slice);
380 
381  // RHS buffer or RHS OpenCL image (_export_to_cl_image == true)
382  if(_export_to_cl_image)
383  {
384  _kernel.setArg(idx++, input1_image2d);
385  }
386  else
387  {
388  add_2D_tensor_argument(idx, _input1, slice_b);
389  }
390 
391  // Bias buffer (_add_bias == true)
392  add_2D_tensor_argument_if(_add_bias, idx, _input2, slice);
393 
394  // Output buffer
395  add_2D_tensor_argument(idx, _output, slice);
396 
397  // K dimension (not used if _export_to_cl_image == true)
398  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_k));
399 
400  // LHS stride_z
401  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input0->info()->strides_in_bytes()[2]));
402 
403  // RHS stride_z (not used if _export_to_cl_image == true)
404  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input1->info()->strides_in_bytes()[2]));
405 
406  // Bias stride_z (if _add_bias == true)
407  if(_add_bias)
408  {
409  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_input2->info()->strides_in_bytes()[2]));
410  }
411 
412  // Output stride_z
413  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(_output->info()->strides_in_bytes()[2]));
414 
415  // Cross-plan padding (if _reinterpret_output_as_3d = true)
416  if(_reinterpret_output_as_3d)
417  {
418  _kernel.setArg<cl_uint>(idx++, static_cast<unsigned int>(total_cross_plane_pad));
419  }
420 
421  // Dispatch kernel
422  enqueue(queue, *this, slice, lws_hint(), _use_dummy_work_items);
423  }
425 }
unsigned int top
top of the border
Definition: Types.h:375
virtual size_t num_dimensions() const =0
The number of dimensions of the tensor (rank)
bool is_one(float a, float epsilon=0.00001f)
Checks if the input floating point number is 1.0f checking if the difference is within a range define...
Definition: float_ops.h:97
Window calculate_max_window(const ValidRegion &valid_region, const Steps &steps, bool skip_border, BorderSize border_size)
bool broadcast_bias
Flag used to broadcast the bias addition.
#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
static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
Static function to check if given info will lead to a valid configuration of CLGEMMMatrixMultiplyResh...
Shape of a tensor.
Definition: TensorShape.h:39
bool fp_mixed_precision
Flag used to indicate wider accumulators (32 bit instead of 16 for FP16).
void run(const Window &window, cl::CommandQueue &queue) override
Enqueue the OpenCL kernel to process the given window on the passed OpenCL command queue.
Descriptor used by the GEMM kernels.
void add_2D_tensor_argument_if(bool cond, unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx ...
Definition: ICLKernel.h:159
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:489
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.
void configure(const ICLTensor *input0, const ICLTensor *input1, const ICLTensor *input2, ICLTensor *output, float alpha, float beta, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMKernelInfo &gemm_info)
Initialise the kernel's input and output.
unsigned int depth_output_gemm3d
Depth of the output tensor in case is reinterpreted as 3D.
cl::NDRange lws_hint() const
Return the Local-Workgroup-Size hint.
Definition: ICLKernel.h:276
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:361
#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.
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.
1 channel, 1 F32 per channel
#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
const std::string & string_from_activation_func(ActivationLayerInfo::ActivationFunction act)
Translates a given activation function to a string.
Definition: Utils.cpp:163
static CLKernelLibrary & get()
Access the KernelLibrary singleton.
GEMM LHS (Left Hand Side) matrix information.
Definition: Types.h:1903
Store the tensor's metadata.
Definition: ITensorInfo.h:40
#define ARM_COMPUTE_ERROR_THROW_ON(status)
Definition: Error.h:455
Describe one of the image's dimensions with a start, end and step.
Definition: Window.h:77
unsigned int bottom
bottom of the border
Definition: Types.h:377
Status class.
Definition: Error.h:52
std::string lower_string(const std::string &val)
Lower a given string.
Definition: Utils.cpp:326
#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:1928
Copyright (c) 2017-2021 Arm Limited.
1 channel, 1 F16 per channel
Implementation of a static rectangular access pattern.
#define ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(...)
Definition: Validate.h:159
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:1913
const DataType data_type
Definition: Im2Col.cpp:150
Window collapse(const Window &full_window, size_t first, size_t last=Coordinates::num_max_dimensions) const
Collapse the dimensions between first and last.
Definition: Window.inl:111
unsigned int k0
Number of partial accumulations performed by the matrix multiplication.
Definition: Types.h:1926
unsigned int m
Number of LHS rows.
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:403
unsigned int n
Number of RHS columns.
const std::string & string_from_data_type(DataType dt)
Convert a data type identity into a string.
Definition: Utils.cpp:135
static constexpr size_t DimX
Alias for dimension 0 also known as X dimension.
Definition: Window.h:43
bool update_window_and_padding(Window &win, Ts &&... patterns)
Update window and padding size for each of the access patterns.
Definition: WindowHelpers.h:46
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
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:1918
std::string float_to_string_with_full_precision(float val)
Create a string with the float in full precision.
Definition: Utils.h:1061
virtual const TensorShape & tensor_shape() const =0
Size for each dimension of the tensor.
auto ceil_to_multiple(S value, T divisor) -> decltype(((value+divisor - 1)/divisor) *divisor)
Computes the smallest number larger or equal to value that is a multiple of divisor.
Definition: Utils.h:71
unsigned int n0
Number of columns processed by the matrix multiplication.
Definition: Types.h:1925
Class to describe a number of elements in each dimension.
Definition: Steps.h:40
Coordinates of an item.
Definition: Coordinates.h:37
std::string kernel_name
std::string get_cl_type_from_data_type(const DataType &dt)
Translates a tensor data type to the appropriate OpenCL type.
Definition: CLHelpers.cpp:37
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...
bool reinterpret_input_as_3d
Flag used to reinterpret the input as 3D.
virtual std::unique_ptr< T > clone() const =0
Provide a clone of the current object of class T.
virtual ITensorInfo * info() const =0
Interface to be implemented by the child class to return the tensor's metadata.
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
virtual PaddingSize padding() const =0
Padding of tensor.
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
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:504
CLCompileContext class.
static constexpr size_t DimY
Alias for dimension 1 also known as Y dimension.
Definition: Window.h:45
void add_2D_tensor_argument(unsigned int &idx, const ICLTensor *tensor, const Window &window)
Add the passed 2D tensor's parameters to the object's kernel's arguments starting from the index idx.
Definition: ICLKernel.h:148
Interface for OpenCL tensor.
Definition: ICLTensor.h:42
virtual size_t total_size() const =0
Returns the total size of the tensor in bytes.
#define ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(...)
Definition: Validate.h:439
#define ARM_COMPUTE_CREATE_ERROR(error_code, msg)
Creates an error with a given message.
Definition: Error.h:159
Manages all the OpenCL kernels compilation and caching, provides accessors for the OpenCL Context.
#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
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *output_stage)
Wrapper to configure the Khronos OpenCL C++ header.
unsigned int k
Number of LHS columns or RHS rows.
bool is_zero(float a, float epsilon=0.00001f)
Checks if the input floating point number is 0.0f checking if the difference is within a range define...
Definition: float_ops.h:109
#define ARM_COMPUTE_RETURN_ERROR_ON_MSG(cond, msg)
If the condition is true, an error is returned.
Definition: Error.h:244
virtual const cl::Buffer & cl_buffer() const =0
Interface to be implemented by the child class to return a reference to the OpenCL buffer containing ...
#define ARM_COMPUTE_ERROR_ON_NULLPTR(...)
Definition: Validate.h:157
Store the tensor's metadata.
Definition: TensorInfo.h:43
unsigned int k0
Number of partial accumulations performed by the matrix multiplication.
Definition: Types.h:1911
cl::Image2D create_image2d_from_buffer(const cl::Context &ctx, const cl::Buffer &buffer, const TensorShape &shape2d, DataType data_type, size_t image_row_pitch)
Create a cl::Image2D object from an OpenCL buffer.
Definition: CLUtils.cpp:29
unsigned int m0
Number of rows processed by the matrix multiplication.
Definition: Types.h:1910
virtual const Strides & strides_in_bytes() const =0
The strides in bytes for accessing each dimension of the tensor.
Window first_slice_window_3D() const
First 3D slice of the window.
Definition: Window.h:291
DataType
Available data types.
Definition: Types.h:77
Describe a multidimensional execution window.
Definition: Window.h:39
void collapse(size_t n, size_t first=0)
Collapse the first n dimensions.
Definition: TensorShape.h:133
TensorShape & set(size_t dimension, size_t value, bool apply_dim_correction=true, bool increase_dim_unit=true)
Accessor to set the value of one of the dimensions.
Definition: TensorShape.h:79
#define ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(f, s)
Definition: Validate.h:201
SimpleTensor< T > slice(const SimpleTensor< T > &src, Coordinates starts, Coordinates ends)