From 5a1bf629752720a7ba0c88f34249393f7e52ad3c Mon Sep 17 00:00:00 2001 From: Manuel Bottini Date: Mon, 1 Mar 2021 17:39:36 +0000 Subject: Port OpenCL Quantization to new API Partially resolves: COMPMID-4193 Change-Id: Ie8367769c690442a0e30383c67851b50ab7c6742 Signed-off-by: Manuel Bottini Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5231 Reviewed-by: Michalis Spyrou Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- .../CL/functions/CLGenerateProposalsLayer.cpp | 8 +-- src/runtime/CL/functions/CLQuantizationLayer.cpp | 38 ++++++++++-- src/runtime/gpu/cl/operators/ClQuantization.cpp | 53 +++++++++++++++++ src/runtime/gpu/cl/operators/ClQuantization.h | 67 ++++++++++++++++++++++ 4 files changed, 156 insertions(+), 10 deletions(-) create mode 100644 src/runtime/gpu/cl/operators/ClQuantization.cpp create mode 100644 src/runtime/gpu/cl/operators/ClQuantization.h (limited to 'src/runtime') diff --git a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp index 365f95243f..fb698d5b88 100644 --- a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp +++ b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp @@ -25,11 +25,11 @@ #include "arm_compute/core/CL/ICLTensor.h" #include "arm_compute/core/Types.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/CL/kernels/CLQuantizationLayerKernel.h" #include "src/core/helpers/AutoConfiguration.h" namespace arm_compute @@ -45,7 +45,7 @@ CLGenerateProposalsLayer::CLGenerateProposalsLayer(std::shared_ptr()), _dequantize_anchors(std::make_unique()), _dequantize_deltas(std::make_unique()), - _quantize_all_proposals(std::make_unique()), + _quantize_all_proposals(std::make_unique()), _cpp_nms(memory_manager), _is_nhwc(false), _is_qasymm8(false), @@ -270,7 +270,7 @@ Status CLGenerateProposalsLayer::validate(const ITensorInfo *scores, const ITens ARM_COMPUTE_RETURN_ON_ERROR(CLBoundingBoxTransformKernel::validate(&all_anchors_f32_info, &proposals_4_roi_values_f32, &deltas_flattened_f32_info, BoundingBoxTransformInfo(info.im_width(), info.im_height(), 1.f))); - ARM_COMPUTE_RETURN_ON_ERROR(CLQuantizationLayerKernel::validate(&proposals_4_roi_values_f32, &proposals_4_roi_values_quantized)); + ARM_COMPUTE_RETURN_ON_ERROR(CLQuantizationLayer::validate(&proposals_4_roi_values_f32, &proposals_4_roi_values_quantized)); proposals_4_roi_values_to_use = &proposals_4_roi_values_quantized; } else @@ -372,7 +372,7 @@ void CLGenerateProposalsLayer::run() if(_is_qasymm8) { - CLScheduler::get().enqueue(*_quantize_all_proposals, false); + _quantize_all_proposals->run(); } // Non maxima suppression diff --git a/src/runtime/CL/functions/CLQuantizationLayer.cpp b/src/runtime/CL/functions/CLQuantizationLayer.cpp index cb8cabef87..1f6ddb6014 100644 --- a/src/runtime/CL/functions/CLQuantizationLayer.cpp +++ b/src/runtime/CL/functions/CLQuantizationLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2020 Arm Limited. + * Copyright (c) 2017-2021 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -23,10 +23,26 @@ */ #include "arm_compute/runtime/CL/functions/CLQuantizationLayer.h" -#include "src/core/CL/kernels/CLQuantizationLayerKernel.h" +#include "arm_compute/core/CL/CLKernelLibrary.h" +#include "arm_compute/core/CL/ICLTensor.h" +#include "src/core/CL/ICLKernel.h" +#include "src/runtime/gpu/cl/operators/ClQuantization.h" namespace arm_compute { +struct CLQuantizationLayer::Impl +{ + const ICLTensor *src{ nullptr }; + ICLTensor *dst{ nullptr }; + std::unique_ptr op{ nullptr }; +}; + +CLQuantizationLayer::CLQuantizationLayer() + : _impl(std::make_unique()) +{ +} +CLQuantizationLayer::~CLQuantizationLayer() = default; + void CLQuantizationLayer::configure(const ICLTensor *input, ICLTensor *output) { configure(CLKernelLibrary::get().get_compile_context(), input, output); @@ -34,13 +50,23 @@ void CLQuantizationLayer::configure(const ICLTensor *input, ICLTensor *output) void CLQuantizationLayer::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 CLQuantizationLayer::validate(const ITensorInfo *input, const ITensorInfo *output) { - return CLQuantizationLayerKernel::validate(input, output); + return opencl::ClQuantization::validate(input, output); +} + +void CLQuantizationLayer::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/gpu/cl/operators/ClQuantization.cpp b/src/runtime/gpu/cl/operators/ClQuantization.cpp new file mode 100644 index 0000000000..2e753b550e --- /dev/null +++ b/src/runtime/gpu/cl/operators/ClQuantization.cpp @@ -0,0 +1,53 @@ +/* + * 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/ClQuantization.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/runtime/CL/CLScheduler.h" +#include "src/core/gpu/cl/ClCompileContext.h" +#include "src/core/gpu/cl/kernels/ClQuantizationKernel.h" + +namespace arm_compute +{ +namespace opencl +{ +void ClQuantization::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 ClQuantization::validate(const ITensorInfo *src, const ITensorInfo *dst) +{ + return kernels::ClQuantizationKernel::validate(src, dst); +} + +void ClQuantization::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/ClQuantization.h b/src/runtime/gpu/cl/operators/ClQuantization.h new file mode 100644 index 0000000000..d938ff95a0 --- /dev/null +++ b/src/runtime/gpu/cl/operators/ClQuantization.h @@ -0,0 +1,67 @@ +/* + * 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_QUANTIZATION_H +#define ARM_COMPUTE_CL_QUANTIZATION_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 quantize a tensor. This function calls the following OpenCL kernel: + * + * -# @ref kernels::ClQuantizationKernel + */ +class ClQuantization : public IClOperator +{ +public: + /** Constructor */ + ClQuantization() = default; + /** Set the input and output tensors. + * + * @param[in] compile_context The compile context to be used. + * @param[in] src Source tensor. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/32. + * @param[out] dst Destination tensor with the same dimensions of input. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16. + * + * @note Output auto initialization is not supported by this function + */ + 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 CLQuantizationLayer + * + * @param[in] src Input tensor info. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/32. + * @param[in] dst Output tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16. + * + * @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_QUANTIZATION_H */ -- cgit v1.2.1