From 9e73c93bbd49fdd648d8f8cb77df46e7bbc9526d Mon Sep 17 00:00:00 2001 From: Manuel Bottini Date: Tue, 2 Mar 2021 17:40:42 +0000 Subject: Port OpenCL Dequantization to new API Partially resolves: COMPMID-4193 Change-Id: I4e14149d5b0a7f9c0dd3bfce800eaddca1e4d885 Signed-off-by: Manuel Bottini Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5238 Reviewed-by: Georgios Pinitas Tested-by: Arm Jenkins Comments-Addressed: Arm Jenkins --- src/runtime/CL/functions/CLDequantizationLayer.cpp | 39 +++++++++++--- .../CL/functions/CLGenerateProposalsLayer.cpp | 14 ++--- src/runtime/gpu/cl/operators/ClDequantization.cpp | 54 +++++++++++++++++++ src/runtime/gpu/cl/operators/ClDequantization.h | 62 ++++++++++++++++++++++ 4 files changed, 156 insertions(+), 13 deletions(-) create mode 100644 src/runtime/gpu/cl/operators/ClDequantization.cpp create mode 100644 src/runtime/gpu/cl/operators/ClDequantization.h (limited to 'src/runtime') diff --git a/src/runtime/CL/functions/CLDequantizationLayer.cpp b/src/runtime/CL/functions/CLDequantizationLayer.cpp index d358813724..e0381f90ae 100644 --- a/src/runtime/CL/functions/CLDequantizationLayer.cpp +++ b/src/runtime/CL/functions/CLDequantizationLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -23,10 +23,27 @@ */ #include "arm_compute/runtime/CL/functions/CLDequantizationLayer.h" -#include "src/core/CL/kernels/CLDequantizationLayerKernel.h" +#include "arm_compute/core/CL/CLKernelLibrary.h" +#include "arm_compute/core/CL/ICLTensor.h" +#include "arm_compute/core/KernelDescriptors.h" +#include "src/core/CL/ICLKernel.h" +#include "src/runtime/gpu/cl/operators/ClDequantization.h" namespace arm_compute { +struct CLDequantizationLayer::Impl +{ + const ICLTensor *src{ nullptr }; + ICLTensor *dst{ nullptr }; + std::unique_ptr op{ nullptr }; +}; + +CLDequantizationLayer::CLDequantizationLayer() + : _impl(std::make_unique()) +{ +} +CLDequantizationLayer::~CLDequantizationLayer() = default; + void CLDequantizationLayer::configure(const ICLTensor *input, ICLTensor *output) { configure(CLKernelLibrary::get().get_compile_context(), input, output); @@ -34,13 +51,23 @@ void CLDequantizationLayer::configure(const ICLTensor *input, ICLTensor *output) void CLDequantizationLayer::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output) { - auto k = std::make_unique(); - k->configure(compile_context, input, output); - _kernel = std::move(k); + _impl->src = input; + _impl->dst = output; + + _impl->op = std::make_unique(); + _impl->op->configure(compile_context, input->info(), output->info()); } Status CLDequantizationLayer::validate(const ITensorInfo *input, const ITensorInfo *output) { - return CLDequantizationLayerKernel::validate(input, output); + return opencl::ClDequantization::validate(input, output); +} + +void CLDequantizationLayer::run() +{ + ITensorPack pack; + pack.add_tensor(TensorType::ACL_SRC, _impl->src); + pack.add_tensor(TensorType::ACL_DST, _impl->dst); + _impl->op->run(pack); } } // namespace arm_compute diff --git a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp index fb698d5b88..81e24dba08 100644 --- a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp +++ b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp @@ -25,9 +25,9 @@ #include "arm_compute/core/CL/ICLTensor.h" #include "arm_compute/core/Types.h" +#include "arm_compute/runtime/CL/functions/CLDequantizationLayer.h" #include "arm_compute/runtime/CL/functions/CLQuantizationLayer.h" #include "src/core/CL/kernels/CLBoundingBoxTransformKernel.h" -#include "src/core/CL/kernels/CLDequantizationLayerKernel.h" #include "src/core/CL/kernels/CLGenerateProposalsLayerKernel.h" #include "src/core/CL/kernels/CLPadLayerKernel.h" #include "src/core/helpers/AutoConfiguration.h" @@ -43,8 +43,8 @@ CLGenerateProposalsLayer::CLGenerateProposalsLayer(std::shared_ptr()), _bounding_box_kernel(std::make_unique()), _pad_kernel(std::make_unique()), - _dequantize_anchors(std::make_unique()), - _dequantize_deltas(std::make_unique()), + _dequantize_anchors(std::make_unique()), + _dequantize_deltas(std::make_unique()), _quantize_all_proposals(std::make_unique()), _cpp_nms(memory_manager), _is_nhwc(false), @@ -261,10 +261,10 @@ Status CLGenerateProposalsLayer::validate(const ITensorInfo *scores, const ITens if(is_qasymm8) { TensorInfo all_anchors_f32_info(anchors->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32)); - ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayerKernel::validate(&all_anchors_info, &all_anchors_f32_info)); + ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayer::validate(&all_anchors_info, &all_anchors_f32_info)); TensorInfo deltas_flattened_f32_info(deltas->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32)); - ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayerKernel::validate(&deltas_flattened_info, &deltas_flattened_f32_info)); + ARM_COMPUTE_RETURN_ON_ERROR(CLDequantizationLayer::validate(&deltas_flattened_info, &deltas_flattened_f32_info)); TensorInfo proposals_4_roi_values_f32(deltas->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true).set_data_type(DataType::F32)); ARM_COMPUTE_RETURN_ON_ERROR(CLBoundingBoxTransformKernel::validate(&all_anchors_f32_info, &proposals_4_roi_values_f32, &deltas_flattened_f32_info, @@ -363,8 +363,8 @@ void CLGenerateProposalsLayer::run() if(_is_qasymm8) { - CLScheduler::get().enqueue(*_dequantize_anchors, false); - CLScheduler::get().enqueue(*_dequantize_deltas, false); + _dequantize_anchors->run(); + _dequantize_deltas->run(); } // Build the boxes diff --git a/src/runtime/gpu/cl/operators/ClDequantization.cpp b/src/runtime/gpu/cl/operators/ClDequantization.cpp new file mode 100644 index 0000000000..df3203d2e1 --- /dev/null +++ b/src/runtime/gpu/cl/operators/ClDequantization.cpp @@ -0,0 +1,54 @@ +/* + * Copyright (c) 2021 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "src/runtime/gpu/cl/operators/ClDequantization.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/runtime/CL/CLScheduler.h" +#include "src/core/CL/kernels/CLFillBorderKernel.h" +#include "src/core/gpu/cl/ClCompileContext.h" +#include "src/core/gpu/cl/kernels/ClDequantizationKernel.h" + +namespace arm_compute +{ +namespace opencl +{ +void ClDequantization::configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst) +{ + auto k = std::make_unique(); + k->configure(compile_context, src, dst); + _kernel = std::move(k); +} + +Status ClDequantization::validate(const ITensorInfo *src, const ITensorInfo *dst) +{ + return kernels::ClDequantizationKernel::validate(src, dst); +} + +void ClDequantization::run(ITensorPack &tensors) +{ + ARM_COMPUTE_ERROR_ON_MSG(tensors.empty(), "No inputs provided"); + CLScheduler::get().enqueue_op(*_kernel.get(), tensors); +} +} // namespace opencl +} // namespace arm_compute diff --git a/src/runtime/gpu/cl/operators/ClDequantization.h b/src/runtime/gpu/cl/operators/ClDequantization.h new file mode 100644 index 0000000000..a696b73d2e --- /dev/null +++ b/src/runtime/gpu/cl/operators/ClDequantization.h @@ -0,0 +1,62 @@ +/* + * Copyright (c) 2021 Arm Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#ifndef ARM_COMPUTE_CL_DEQUANTIZATION_H +#define ARM_COMPUTE_CL_DEQUANTIZATION_H + +#include "arm_compute/core/KernelDescriptors.h" +#include "src/core/gpu/cl/ClCompileContext.h" +#include "src/runtime/gpu/cl/IClOperator.h" + +namespace arm_compute +{ +namespace opencl +{ +/** Basic function to run @ref kernels::ClDequantizationKernel that dequantizes an input tensor */ +class ClDequantization : public IClOperator +{ +public: + /** Constructor */ + ClDequantization() = default; + /** Set the input and output tensors. + * + * @param[in] compile_context The compile context to be used. + * @param[in] src Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QSYMM8/QSYMM16. + * @param[out] dst Destination tensor info with the same dimensions of @p src. Data type supported: F16/F32. + */ + void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst); + /** Static function to check if given info will lead to a valid configuration of @ref CLDequantizationLayer + * + * @param[in] src Input tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QSYMM8/QSYMM16. + * @param[in] dst Output tensor info. Data type supported: F16/F32. + * + * @return a status + */ + static Status validate(const ITensorInfo *src, const ITensorInfo *dst); + + // Inherited method overridden + void run(ITensorPack &tensors) override; +}; +} // namespace opencl +} // namespace arm_compute +#endif /* ARM_COMPUTE_CL_DEQUANTIZATION_H */ -- cgit v1.2.1