From be1c017071c81912c78428377d3d95d2da2f966f Mon Sep 17 00:00:00 2001 From: giuros01 Date: Tue, 4 Dec 2018 11:36:08 +0000 Subject: COMPMID-1757: NEON: Implement Tile Change-Id: Ic6a1f55f14d53896725afe426bc2e2acb1546589 Reviewed-on: https://review.mlplatform.org/343 Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas --- arm_compute/core/NEON/NEKernels.h | 1 + arm_compute/core/NEON/kernels/NETileKernel.h | 76 +++++++++++++++ arm_compute/runtime/NEON/NEFunctions.h | 3 +- arm_compute/runtime/NEON/functions/NETile.h | 57 +++++++++++ src/core/NEON/kernels/NETileKernel.cpp | 117 ++++++++++++++++++++++ src/runtime/NEON/functions/NETile.cpp | 42 ++++++++ tests/validation/NEON/Tile.cpp | 141 +++++++++++++++++++++++++++ 7 files changed, 436 insertions(+), 1 deletion(-) create mode 100644 arm_compute/core/NEON/kernels/NETileKernel.h create mode 100644 arm_compute/runtime/NEON/functions/NETile.h create mode 100644 src/core/NEON/kernels/NETileKernel.cpp create mode 100644 src/runtime/NEON/functions/NETile.cpp create mode 100644 tests/validation/NEON/Tile.cpp diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h index 0cfde722d9..49bba3a750 100644 --- a/arm_compute/core/NEON/NEKernels.h +++ b/arm_compute/core/NEON/NEKernels.h @@ -114,6 +114,7 @@ #include "arm_compute/core/NEON/kernels/NESoftmaxLayerKernel.h" #include "arm_compute/core/NEON/kernels/NETableLookupKernel.h" #include "arm_compute/core/NEON/kernels/NEThresholdKernel.h" +#include "arm_compute/core/NEON/kernels/NETileKernel.h" #include "arm_compute/core/NEON/kernels/NETransposeKernel.h" #include "arm_compute/core/NEON/kernels/NEUpsampleLayerKernel.h" #include "arm_compute/core/NEON/kernels/NEWarpKernel.h" diff --git a/arm_compute/core/NEON/kernels/NETileKernel.h b/arm_compute/core/NEON/kernels/NETileKernel.h new file mode 100644 index 0000000000..f842bff7db --- /dev/null +++ b/arm_compute/core/NEON/kernels/NETileKernel.h @@ -0,0 +1,76 @@ +/* + * Copyright (c) 2018 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_NETILEKERNEL_H__ +#define __ARM_COMPUTE_NETILEKERNEL_H__ + +#include "arm_compute/core/NEON/INEKernel.h" + +namespace arm_compute +{ +class ITensor; + +/** NEON kernel to perform a tile operation */ +class NETileKernel : public INEKernel +{ +public: + /** Default constructor */ + NETileKernel(); + /** Prevent instances of this class from being copied (As this class contains pointers). */ + NETileKernel(const NETileKernel &) = delete; + /** Prevent instances of this class from being copied (As this class contains pointers). */ + NETileKernel &operator=(const NETileKernel &) = delete; + /** Allow instances of this class to be moved */ + NETileKernel(NETileKernel &&) = default; + /** Allow instances of this class to be moved */ + NETileKernel &operator=(NETileKernel &&) = default; + const char *name() const override + { + return "NETileKernel"; + } + /** Set the source, destination of the kernel + * + * @param[in] input Source tensor. Data type supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32. + * @param[out] output Destination tensor. Same as @p input + * @param[in] multiples Contains the number of times the input tensor should be replicated on the given dimension. + */ + void configure(const ITensor *input, ITensor *output, const Multiples &multiples); + /** Static function to check if given info will lead to a valid configuration of @ref NETileKernel + * + * @param[in] input Source tensor info. Data type supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32. + * @param[in] output Destination tensor info. Same as @p input + * @param[in] multiples Contains the number of times the input tensor should be replicated on the given dimension. + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples); + + // Inherited methods overridden: + void run(const Window &window, const ThreadInfo &info) override; + +private: + const ITensor *_input; + ITensor *_output; +}; +} // namespace arm_compute +#endif /*__ARM_COMPUTE_NETILEKERNEL_H__ */ diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h index d9c8355178..8873110982 100644 --- a/arm_compute/runtime/NEON/NEFunctions.h +++ b/arm_compute/runtime/NEON/NEFunctions.h @@ -117,6 +117,7 @@ #include "arm_compute/runtime/NEON/functions/NESoftmaxLayer.h" #include "arm_compute/runtime/NEON/functions/NETableLookup.h" #include "arm_compute/runtime/NEON/functions/NEThreshold.h" +#include "arm_compute/runtime/NEON/functions/NETile.h" #include "arm_compute/runtime/NEON/functions/NETranspose.h" #include "arm_compute/runtime/NEON/functions/NEUpsampleLayer.h" #include "arm_compute/runtime/NEON/functions/NEWarpAffine.h" @@ -125,4 +126,4 @@ #include "arm_compute/runtime/NEON/functions/NEWinogradConvolutionLayer.h" #include "arm_compute/runtime/NEON/functions/NEYOLOLayer.h" -#endif /* __ARM_COMPUTE_NEFUNCTIONS_H__ */ \ No newline at end of file +#endif /* __ARM_COMPUTE_NEFUNCTIONS_H__ */ diff --git a/arm_compute/runtime/NEON/functions/NETile.h b/arm_compute/runtime/NEON/functions/NETile.h new file mode 100644 index 0000000000..f637d17aee --- /dev/null +++ b/arm_compute/runtime/NEON/functions/NETile.h @@ -0,0 +1,57 @@ +/* + * Copyright (c) 2018 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_NETILE_H__ +#define __ARM_COMPUTE_NETILE_H__ + +#include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h" + +#include "arm_compute/core/Types.h" + +namespace arm_compute +{ +class ITensor; + +/** Basic function to run @ref NETileKernel */ +class NETile : public INESimpleFunctionNoBorder +{ +public: + /** Set the source, destination of the kernel + * + * @param[in] input Source tensor. Data type supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32. + * @param[out] output Destination tensor. Same as @p input + * @param[in] multiples Contains the number of times the input tensor should be replicated on the given dimension. + */ + void configure(const ITensor *input, ITensor *output, const Multiples &multiples); + /** Static function to check if given info will lead to a valid configuration of @ref NETile + * + * @param[in] input Source tensor info. Data type supported: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32. + * @param[in] output Destination tensor info. Same as @p input + * @param[in] multiples Contains the number of times the input tensor should be replicated on the given dimension. + * + * @return a status + */ + static Status validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples); +}; +} // namespace arm_compute +#endif /* __ARM_COMPUTE_NETILE_H__ */ diff --git a/src/core/NEON/kernels/NETileKernel.cpp b/src/core/NEON/kernels/NETileKernel.cpp new file mode 100644 index 0000000000..dbeacfad94 --- /dev/null +++ b/src/core/NEON/kernels/NETileKernel.cpp @@ -0,0 +1,117 @@ +/* + * Copyright (c) 2018 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 "arm_compute/core/NEON/kernels/NETileKernel.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/core/Helpers.h" +#include "arm_compute/core/ITensor.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" +#include "arm_compute/core/utils/misc/ShapeCalculator.h" + +namespace arm_compute +{ +namespace +{ +Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples) +{ + ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output); + ARM_COMPUTE_RETURN_ERROR_ON(multiples.size() > 4); + ARM_COMPUTE_RETURN_ERROR_ON(multiples.empty()); + ARM_COMPUTE_RETURN_ERROR_ON(std::any_of(multiples.begin(), multiples.end(), [](uint32_t e) + { + return e == 0; + })); + + // Validate output if initialized + if(output->total_size() != 0) + { + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(misc::shape_calculator::compute_tiled_shape(input->tensor_shape(), multiples), output->tensor_shape()); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); + } + + return Status{}; +} +} // namespace + +NETileKernel::NETileKernel() + : _input(nullptr), _output(nullptr) +{ +} + +void NETileKernel::configure(const ITensor *input, ITensor *output, const Multiples &multiples) +{ + ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); + + // Auto initialize output + TensorShape tiled_shape = misc::shape_calculator::compute_tiled_shape(input->info()->tensor_shape(), multiples); + auto_init_if_empty(*output->info(), tiled_shape, 1, input->info()->data_type()); + + // Validate + ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), multiples)); + + _input = input; + _output = output; + + // Configure window without padding + Window win = calculate_max_window(*output->info()); + INEKernel::configure(win); +} + +Status NETileKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples) +{ + ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, multiples)); + return Status{}; +} + +void NETileKernel::run(const Window &window, const ThreadInfo &info) +{ + ARM_COMPUTE_UNUSED(info); + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); + + Window output_window{ window }; + output_window.set(Window::DimX, Window::Dimension(output_window.x().start(), output_window.x().end(), _input->info()->dimension(0))); + Window out_slice = output_window.first_slice_window_1D(); + + const auto src_shape = _input->info()->tensor_shape(); + do + { + Iterator output_it(_output, out_slice); + + execute_window_loop(out_slice, [&](const Coordinates & id) + { + const size_t x = id.x(); + const size_t y = id.y(); + const size_t z = id.z(); + const size_t w = id[3]; + Coordinates input_coords{ x % src_shape[0], y % src_shape[1], z % src_shape[2], w % src_shape[3] }; + memcpy(output_it.ptr(), _input->ptr_to_element(input_coords), _input->info()->dimension(0) * _input->info()->element_size()); + }, + output_it); + } + while(output_window.slide_window_slice_1D(out_slice)); +} +} // namespace arm_compute diff --git a/src/runtime/NEON/functions/NETile.cpp b/src/runtime/NEON/functions/NETile.cpp new file mode 100644 index 0000000000..0ca4413251 --- /dev/null +++ b/src/runtime/NEON/functions/NETile.cpp @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2018 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 "arm_compute/runtime/NEON/functions/NETile.h" + +#include "arm_compute/core/NEON/kernels/NETileKernel.h" +#include "support/ToolchainSupport.h" + +namespace arm_compute +{ +void NETile::configure(const ITensor *input, ITensor *output, const Multiples &multiples) +{ + auto k = arm_compute::support::cpp14::make_unique(); + k->configure(input, output, multiples); + _kernel = std::move(k); +} + +Status NETile::validate(const ITensorInfo *input, const ITensorInfo *output, const Multiples &multiples) +{ + return NETileKernel::validate(input, output, multiples); +} +} // namespace arm_compute diff --git a/tests/validation/NEON/Tile.cpp b/tests/validation/NEON/Tile.cpp new file mode 100644 index 0000000000..209652009e --- /dev/null +++ b/tests/validation/NEON/Tile.cpp @@ -0,0 +1,141 @@ +/* + * Copyright (c) 2018 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, INNEUDING 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 NEAIM, 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 "arm_compute/core/Types.h" +#include "arm_compute/runtime/NEON/functions/NETile.h" +#include "arm_compute/runtime/Tensor.h" +#include "arm_compute/runtime/TensorAllocator.h" +#include "tests/NEON/Accessor.h" +#include "tests/datasets/ShapeDatasets.h" +#include "tests/framework/Asserts.h" +#include "tests/framework/Macros.h" +#include "tests/framework/datasets/Datasets.h" +#include "tests/validation/Validation.h" +#include "tests/validation/fixtures/TileFixture.h" + +namespace arm_compute +{ +namespace test +{ +namespace validation +{ +namespace +{ +const auto MultiplesDataset = framework::dataset::make("Multiples", { Multiples{ 3 }, + Multiples{ 2, 2 }, + Multiples{ 2, 3, 4, 5 }, + Multiples{ 2, 1, 2, 2 }, + Multiples{ 2, 1, 3 }, + Multiples{ 3, 3, 3 }, + Multiples{ 2, 2, 2 } + }); +} // namespace +TEST_SUITE(NE) +TEST_SUITE(Tile) + +// *INDENT-OFF* +// clang-format off +DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(zip( + framework::dataset::make("InputInfo", { TensorInfo(TensorShape(10, 10), 1, DataType::F32), + TensorInfo(TensorShape(10, 10), 1, DataType::F32), // Mismatching shape + TensorInfo(TensorShape(10, 10), 1, DataType::F16), // Mismatching type + TensorInfo(TensorShape(10, 10), 1, DataType::F32)}), // Wrong multiples + framework::dataset::make("OutputInfo",{ TensorInfo(TensorShape(10, 20), 1, DataType::F32), + TensorInfo(TensorShape(20, 20), 1, DataType::F32), + TensorInfo(TensorShape(20, 20), 1, DataType::F32), + TensorInfo(TensorShape(10, 20), 1, DataType::F32)})), + framework::dataset::make("Multiples",{ Multiples{1, 2}, Multiples{1, 2}, Multiples{0, 1} })), + framework::dataset::make("Expected", {true, false, false, false })), + input_info, output_info, multiples, expected) +{ + const Status status = NETile::validate(&input_info.clone()->set_is_resizable(false), &output_info.clone()->set_is_resizable(false), multiples); + ARM_COMPUTE_EXPECT(bool(status) == expected, framework::LogLevel::ERRORS); +} +// clang-format on +// *INDENT-ON* + +template +using NETileFixture = TileValidationFixture; + +TEST_SUITE(Float) +TEST_SUITE(FP16) +FIXTURE_DATA_TEST_CASE(RunSmall, NETileFixture, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), framework::dataset::make("DataType", DataType::F16)), + MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +FIXTURE_DATA_TEST_CASE(RunLarge, NETileFixture, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), framework::dataset::make("DataType", DataType::F16)), MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +TEST_SUITE_END() // FP16 + +TEST_SUITE(FP32) +FIXTURE_DATA_TEST_CASE(RunSmall, NETileFixture, framework::DatasetMode::PRECOMMIT, combine(combine(datasets::SmallShapes(), framework::dataset::make("DataType", DataType::F32)), + MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +FIXTURE_DATA_TEST_CASE(RunLarge, NETileFixture, framework::DatasetMode::NIGHTLY, combine(combine(datasets::LargeShapes(), framework::dataset::make("DataType", DataType::F32)), + MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +TEST_SUITE_END() // FP32 +TEST_SUITE_END() // Float + +TEST_SUITE(Integer) +TEST_SUITE(S8) +FIXTURE_DATA_TEST_CASE(RunSmall, NETileFixture, framework::DatasetMode::ALL, + combine( + combine(datasets::SmallShapes(), framework::dataset::make("DataType", { DataType::S8 })), + MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +TEST_SUITE_END() // S8 +TEST_SUITE_END() // Integer + +TEST_SUITE(Quantized) +TEST_SUITE(QASYMM8) +FIXTURE_DATA_TEST_CASE(RunSmall, NETileFixture, framework::DatasetMode::ALL, + combine( + combine(datasets::SmallShapes(), framework::dataset::make("DataType", { DataType::QASYMM8 })), + MultiplesDataset)) +{ + // Validate output + validate(Accessor(_target), _reference); +} +TEST_SUITE_END() // QASYMM8 +TEST_SUITE_END() // Quantized + +TEST_SUITE_END() // Tile +TEST_SUITE_END() // NE +} // namespace validation +} // namespace test +} // namespace arm_compute -- cgit v1.2.1