| 245 | } |
| 246 | |
| 247 | void ClMatMulNativeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) |
| 248 | { |
| 249 | ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); |
| 250 | ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); |
| 251 | |
| 252 | const ICLTensor *lhs = |
| 253 | utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_0)); |
| 254 | const ICLTensor *rhs = |
| 255 | utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC_1)); |
| 256 | const ICLTensor *bias = utils::cast::polymorphic_downcast<const ICLTensor *>( |
| 257 | tensors.get_const_tensor(TensorType::ACL_SRC_2)); // nullptr if bias is not present |
| 258 | ICLTensor *dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST)); |
| 259 | ARM_COMPUTE_ERROR_ON_NULLPTR(lhs, rhs, dst); |
| 260 | ARM_COMPUTE_LOG_PARAMS(lhs, rhs, bias, dst); |
| 261 | |
| 262 | unsigned int idx = 0; |
| 263 | Window window_collapsed = window.collapse(ICLKernel::window(), Window::DimZ); |
| 264 | |
| 265 | add_3d_tensor_nhw_argument(idx, lhs); |
| 266 | |
| 267 | cl::Image2D rhs_cl_image; |
| 268 | if (_export_rhs_to_cl_image) |
| 269 | { |
| 270 | const size_t image_w = rhs->info()->dimension(0) / 4; |
| 271 | const size_t image_h = rhs->info()->tensor_shape().total_size() / rhs->info()->dimension(0); |
| 272 | const TensorShape shape2d(image_w, image_h); |
| 273 | const size_t image_row_pitch = rhs->info()->strides_in_bytes()[1]; |
| 274 | |
| 275 | // Export cl_buffer to cl_image |
| 276 | rhs_cl_image = create_image2d_from_buffer(CLKernelLibrary::get().context(), rhs->cl_buffer(), shape2d, |
| 277 | rhs->info()->data_type(), image_row_pitch, CLImage2DType::ReadOnly); |
| 278 | _kernel.setArg(idx++, rhs_cl_image); |
| 279 | } |
| 280 | |
| 281 | add_3d_tensor_nhw_argument(idx, rhs); |
| 282 | if (bias != nullptr) |
| 283 | { |
| 284 | add_3d_tensor_nhw_argument(idx, bias); |
| 285 | } |
| 286 | add_3d_tensor_nhw_argument(idx, dst); |
| 287 | |
| 288 | enqueue(queue, *this, window_collapsed, lws_hint()); |
| 289 | } |
| 290 | |
| 291 | } // namespace kernels |
| 292 | } // namespace opencl |
nothing calls this directly
no test coverage detected