From f932d2c8409831cb9cb97a2eb65be93ad4709cd6 Mon Sep 17 00:00:00 2001 From: Michele Di Giorgio Date: Mon, 6 Jul 2020 11:27:21 +0100 Subject: COMPMID-3386: Support memory injection in CLConcatenate functions/kernels Signed-off-by: Georgios Pinitas Change-Id: I611adf4f506d406540e920b0bd6befb4b5108918 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3601 Tested-by: Arm Jenkins Reviewed-by: Michalis Spyrou Comments-Addressed: Arm Jenkins --- .../CL/kernels/CLHeightConcatenateLayerKernel.cpp | 44 +++++++++------------- 1 file changed, 18 insertions(+), 26 deletions(-) (limited to 'src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp') diff --git a/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp b/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp index b3bae4604f..5ab270cc01 100644 --- a/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp +++ b/src/core/CL/kernels/CLHeightConcatenateLayerKernel.cpp @@ -27,19 +27,14 @@ #include "arm_compute/core/CL/CLKernelLibrary.h" #include "arm_compute/core/CL/CLValidate.h" #include "arm_compute/core/CL/ICLTensor.h" -#include "arm_compute/core/CL/OpenCL.h" -#include "arm_compute/core/Error.h" #include "arm_compute/core/Helpers.h" -#include "arm_compute/core/IAccessWindow.h" -#include "arm_compute/core/TensorInfo.h" #include "arm_compute/core/Utils.h" #include "arm_compute/core/Window.h" +#include "arm_compute/core/utils/misc/Cast.h" #include "arm_compute/core/utils/misc/ShapeCalculator.h" #include "support/StringSupport.h" -#include - namespace arm_compute { namespace @@ -77,7 +72,7 @@ Status validate_arguments(const ITensorInfo *input, unsigned int height_offset, } // namespace CLHeightConcatenateLayerKernel::CLHeightConcatenateLayerKernel() - : _input(nullptr), _output(nullptr), _height_offset(0), _num_elems_processed_per_iteration() + : _height_offset(0), _num_elems_processed_per_iteration() { } @@ -89,33 +84,26 @@ Status CLHeightConcatenateLayerKernel::validate(const ITensorInfo *input, unsign return Status{}; } -void CLHeightConcatenateLayerKernel::configure(const ICLTensor *input, unsigned int height_offset, ICLTensor *output) -{ - configure(CLKernelLibrary::get().get_compile_context(), input, height_offset, output); -} - -void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, unsigned int height_offset, ICLTensor *output) +void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_context, ITensorInfo *input, unsigned int height_offset, ITensorInfo *output) { ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); - ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), height_offset, output->info())); + ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, height_offset, output)); - _input = input; - _output = output; _height_offset = height_offset; - auto win_config = validate_and_configure_window(input->info(), output->info(), _num_elems_processed_per_iteration); + auto win_config = validate_and_configure_window(input, output, _num_elems_processed_per_iteration); // Add build options CLBuildOptions build_opts; - build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->info()->element_size())); + build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->element_size())); build_opts.add_option("-DVEC_SIZE=" + support::cpp11::to_string(_num_elems_processed_per_iteration)); build_opts.add_option("-DHEIGHT_OFFSET=" + support::cpp11::to_string(_height_offset)); - build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->info()->dimension(2))); + build_opts.add_option("-DDEPTH=" + support::cpp11::to_string(input->dimension(2))); - if(is_data_type_quantized_asymmetric(input->info()->data_type()) && input->info()->quantization_info() != output->info()->quantization_info()) + if(is_data_type_quantized_asymmetric(input->data_type()) && input->quantization_info() != output->quantization_info()) { - const UniformQuantizationInfo iq_info = input->info()->quantization_info().uniform(); - const UniformQuantizationInfo oq_info = output->info()->quantization_info().uniform(); + const UniformQuantizationInfo iq_info = input->quantization_info().uniform(); + const UniformQuantizationInfo oq_info = output->quantization_info().uniform(); build_opts.add_option("-DOFFSET_IN1=" + float_to_string_with_full_precision(iq_info.offset)); build_opts.add_option("-DOFFSET_OUT=" + float_to_string_with_full_precision(oq_info.offset)); @@ -132,17 +120,21 @@ void CLHeightConcatenateLayerKernel::configure(const CLCompileContext &compile_c ICLKernel::configure_internal(std::get<1>(win_config)); // Set output valid region - output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape())); + output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape())); } -void CLHeightConcatenateLayerKernel::run(const Window &window, cl::CommandQueue &queue) +void CLHeightConcatenateLayerKernel::run_op(const InputTensorMap &inputs, const OutputTensorMap &outputs, + const Window &window, cl::CommandQueue &queue) { ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window); + const auto src = utils::cast::polymorphic_downcast(inputs.at(TensorType::ACL_SRC)); + auto dst = utils::cast::polymorphic_downcast(outputs.at(TensorType::ACL_DST)); + unsigned int idx = 0; - add_4D_tensor_argument(idx, _input, window); - add_4D_tensor_argument(idx, _output, window); + add_4D_tensor_argument(idx, src, window); + add_4D_tensor_argument(idx, dst, window); enqueue(queue, *this, window, lws_hint()); } } // namespace arm_compute -- cgit v1.2.1