diff options
author | Pablo Tello <pablo.tello@arm.com> | 2018-03-23 11:40:05 +0000 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:49:16 +0000 |
commit | fc004494c9cde94fe7b5d67199faf2596e223cf3 (patch) | |
tree | a9b365a9ff7e5aedeb6341cba676b1e8022f42e6 | |
parent | c357c47be8a3f210f9eee9a05cc13f1051b036d3 (diff) | |
download | ComputeLibrary-fc004494c9cde94fe7b5d67199faf2596e223cf3.tar.gz |
COMPMID-959: Removed Interleave blocked kernel.
Change-Id: I775eecbc39da583aae2eb4e033c5930dfc402899
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/125684
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
-rw-r--r-- | arm_compute/core/NEON/NEKernels.h | 1 | ||||
-rw-r--r-- | arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h | 79 | ||||
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h | 3 | ||||
-rw-r--r-- | src/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.cpp | 221 | ||||
-rw-r--r-- | tests/validation/NEON/GEMMLowp.cpp | 27 | ||||
-rw-r--r-- | tests/validation/fixtures/GEMMInterleaveBlockedFixture.h | 114 |
6 files changed, 1 insertions, 444 deletions
diff --git a/arm_compute/core/NEON/NEKernels.h b/arm_compute/core/NEON/NEKernels.h index 7ec74eaccd..31f4881ef5 100644 --- a/arm_compute/core/NEON/NEKernels.h +++ b/arm_compute/core/NEON/NEKernels.h @@ -62,7 +62,6 @@ #include "arm_compute/core/NEON/kernels/NEFloorKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMAssemblyBaseKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h" -#include "arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMLowpOffsetContributionKernel.h" #include "arm_compute/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h" diff --git a/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h b/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h deleted file mode 100644 index f5a1121b04..0000000000 --- a/arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h +++ /dev/null @@ -1,79 +0,0 @@ -/* - * Copyright (c) 2017-2018 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_NEGEMMINTERLEAVEBLOCKEDKERNEL_H__ -#define __ARM_COMPUTE_NEGEMMINTERLEAVEBLOCKEDKERNEL_H__ - -#include "arm_compute/core/NEON/INESimpleKernel.h" - -namespace arm_compute -{ -class ITensor; - -/** NEON kernel to interleave the elements of a matrix - * - * Interleave_Blocked copies a block of values at a time instead of just one. The main use of this is the gemmlowp with the "dot product" - * instruction, where each operation consumes 4 values, so we need to copy blocks of 4 values. - * - */ -class NEGEMMInterleaveBlockedKernel : public INESimpleKernel -{ -public: - const char *name() const override - { - return "NEGEMMInterleaveBlockedKernel"; - } - /* Constructor */ - NEGEMMInterleaveBlockedKernel(); - /** Initialise the kernel's input and output. - * - * @param[in] input Input tensor. Data types supported: U8 - * @param[out] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input. - * @param[in] block_height The height of the blocks to be interleaved. - * @param[in] block_width The width of the blocks to be interleaved. - * @param[in] transpose True if transpose operation must be performed, false otherwise. - */ - void configure(const ITensor *input, ITensor *output, unsigned int block_height, unsigned int block_width, bool transpose); - /** Static function to check if given info will lead to a valid configuration of @ref NEGEMMInterleaveBlockedKernel - * - * @param[in] input Input tensor. Data types supported: U8 - * @param[in] output Output tensor which stores the interleaved matrix. Data type supported: same as @p input. - * @param[in] block_height The height of the blocks to be interleaved. - * @param[in] block_width The width of the blocks to be interleaved. - * @param[in] transpose True if transpose operation must be performed, false otherwise. - * - * @return a status - */ - static Status validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int block_height, unsigned int block_width, bool transpose); - - // Inherited methods overridden: - void run(const Window &window, const ThreadInfo &info) override; - -private: - unsigned int _block_height; - unsigned int _block_width; - bool _transpose; -}; - -} // namespace arm_compute -#endif /*__ARM_COMPUTE_NEGEMMINTERLEAVEBLOCKEDKERNEL_H__*/ diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h index 95776f829a..5e05e93855 100644 --- a/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h @@ -48,7 +48,6 @@ class ITensor; * * otherwise if the DOT product instruction is available: * - * -# @ref NEGEMMInterleaveBlockedKernel * -# @ref NEGEMMLowpOffsetContributionKernel * */ @@ -85,7 +84,7 @@ public: */ static Status validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *output, const GEMMInfo &gemm_info = GEMMInfo()); - // Inherited methods overridden: + // Inherited methods overridden void run() override; private: diff --git a/src/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.cpp b/src/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.cpp deleted file mode 100644 index 768dd8b33e..0000000000 --- a/src/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.cpp +++ /dev/null @@ -1,221 +0,0 @@ -/* - * Copyright (c) 2017 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/NEGEMMInterleaveBlockedKernel.h" - -#include "arm_compute/core/Error.h" -#include "arm_compute/core/Helpers.h" -#include "arm_compute/core/ITensor.h" -#include "arm_compute/core/NEON/INEKernel.h" -#include "arm_compute/core/Types.h" -#include "arm_compute/core/Validate.h" -#include "arm_compute/core/Window.h" - -#include <arm_neon.h> -#include <cstddef> -#include <cstdint> -#include <tuple> - -using namespace arm_compute; - -namespace -{ -TensorShape get_output_shape(const ITensorInfo *input, unsigned int block_height) -{ - TensorShape output_shape = input->tensor_shape(); - const float interleave_by_f32 = block_height; - output_shape.set(0, input->dimension(0) * interleave_by_f32); - output_shape.set(1, std::ceil(static_cast<float>(input->dimension(1)) / interleave_by_f32)); - return output_shape; -} - -Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, unsigned int block_width, unsigned int block_height) -{ - ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8); - ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(block_height < 1, "Block height must be greater than 0"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(block_width < 1, "Block window must be greater than 0"); - - if(output->total_size() != 0) - { - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), get_output_shape(input, block_height)); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); - ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output); - } - - return Status{}; -} - -std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, unsigned int block_width, unsigned int block_height) -{ - const unsigned int num_elems_processed_per_iteration_x = block_width; - const unsigned int num_elems_processed_per_iteration_y = block_height; - bool window_changed = false; - - // Configure kernel window - Window win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y)); - const float scaley_factor = 1.f / block_height; - - AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y); - window_changed = window_changed || update_window_and_padding(win, input_access); - - // Configure window in case of configured output - if(output->total_size() != 0) - { - AccessWindowRectangle output_access(output, - 0, 0, - num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y, - 1, num_elems_processed_per_iteration_y, scaley_factor); - window_changed = window_changed || update_window_and_padding(win, output_access); - output_access.set_valid_region(win, input->valid_region()); - } - - Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{}; - return std::make_pair(err, win); -} - -inline void gemm_interleave_blocked_transposed_8bit(const ITensor *input, ITensor *output, const Window &window, unsigned int block_width, unsigned int block_height) -{ - const size_t in_stride = input->info()->strides_in_bytes()[1]; - - const unsigned int in_height = input->info()->dimension(1); - const unsigned int in_width = input->info()->dimension(0); - - const float scale_y_factor = 1.f / float(block_height); - - // Set window for output tensor - Window win_out(window); - win_out.scale(Window::DimY, scale_y_factor); - Iterator in(input, window); - - win_out.set_dimension_step(Window::DimX, block_width * block_height); - Iterator out(output, win_out); - - execute_window_loop(window, [&](const Coordinates &) - { - std::fill_n(out.ptr(), block_width * block_height, 0); - }, - out); - - execute_window_loop(window, [&](const Coordinates & id) - { - for(unsigned int z = id.y(); (z < in_width) && z < (id.y() + block_height); ++z) - { - int j = (z - id.y()) * block_width; - for(unsigned int b = id.x(); (b < in_height) && (b < (id.x() + block_width)); ++b) - { - *(out.ptr() + j++) = *(input->buffer() + b * in_stride + z); - } - } - }, - in, out); -} - -inline void gemm_interleave_blocked_8bit(const ITensor *input, ITensor *output, const Window &window, unsigned int block_width, unsigned int block_height) -{ - const size_t in_stride = input->info()->strides_in_bytes()[1]; - - const unsigned int in_height = input->info()->dimension(1); - const unsigned int in_width = input->info()->dimension(0); - - const float scale_y_factor = 1.f / float(block_height); - - // Set window for output tensor - Window win_out(window); - win_out.scale(Window::DimY, scale_y_factor); - Iterator in(input, window); - - win_out.set_dimension_step(Window::DimX, block_width * block_height); - Iterator out(output, win_out); - - execute_window_loop(window, [&](const Coordinates &) - { - std::fill_n(out.ptr(), block_width * block_height, 0); - }, - out); - - execute_window_loop(window, [&](const Coordinates & id) - { - for(unsigned int z = id.y(); (z < in_height) && z < (id.y() + block_height); ++z) - { - int j = (z - id.y()) * block_width; - for(unsigned int b = id.x(); (b < in_width) && (b < (id.x() + block_width)); ++b) - { - *(out.ptr() + j++) = *(input->buffer() + z * in_stride + b); - } - } - }, - in, out); -} -} // namespace - -NEGEMMInterleaveBlockedKernel::NEGEMMInterleaveBlockedKernel() - : _block_height(0), _block_width(0), _transpose(false) -{ -} - -void NEGEMMInterleaveBlockedKernel::configure(const ITensor *input, ITensor *output, unsigned int block_height, unsigned int block_width, bool transpose) -{ - ARM_COMPUTE_ERROR_ON_NULLPTR(input, output); - - // Output auto inizialitation if not yet initialized - auto_init_if_empty(*output->info(), get_output_shape(input->info(), block_height), 1, input->info()->data_type(), input->info()->fixed_point_position()); - - // Perform validation step - ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), block_width, block_height)); - - _input = input; - _output = output; - _block_height = block_height; - _block_width = block_width; - _transpose = transpose; - - // Configure kernel window - auto win_config = validate_and_configure_window(input->info(), output->info(), block_width, block_height); - ARM_COMPUTE_ERROR_THROW_ON(win_config.first); - INEKernel::configure(win_config.second); -} - -Status NEGEMMInterleaveBlockedKernel::validate(const ITensorInfo *input, const ITensorInfo *output, unsigned int block_height, unsigned int block_width, bool transpose) -{ - ARM_COMPUTE_UNUSED(transpose); - ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, block_width, block_height)); - ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get(), block_width, block_height).first); - - return Status{}; -} - -void NEGEMMInterleaveBlockedKernel::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); - if(_transpose) - { - gemm_interleave_blocked_transposed_8bit(_input, _output, window, _block_width, _block_height); - } - else - { - gemm_interleave_blocked_8bit(_input, _output, window, _block_width, _block_height); - } -} diff --git a/tests/validation/NEON/GEMMLowp.cpp b/tests/validation/NEON/GEMMLowp.cpp index 471ec3faaf..ed24d618f5 100644 --- a/tests/validation/NEON/GEMMLowp.cpp +++ b/tests/validation/NEON/GEMMLowp.cpp @@ -21,7 +21,6 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#include "arm_compute/core/NEON/kernels/NEGEMMInterleaveBlockedKernel.h" #include "arm_compute/core/Types.h" #include "arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h" #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h" @@ -38,7 +37,6 @@ #include "tests/framework/Macros.h" #include "tests/framework/datasets/Datasets.h" #include "tests/validation/Validation.h" -#include "tests/validation/fixtures/GEMMInterleaveBlockedFixture.h" #include "tests/validation/fixtures/GEMMLowpAssemblyFixture.h" #include "tests/validation/fixtures/GEMMLowpFixture.h" @@ -50,8 +48,6 @@ namespace validation { namespace { -const auto data_int_blk = framework::dataset::make("M", 8, 12) * framework::dataset::make("N", 8, 12) * framework::dataset::make("by", 8, 13) * framework::dataset::make("block", 4, 9); -const auto data_int_blk_tr = framework::dataset::make("M", 8, 17) * framework::dataset::make("N", 8, 14) * framework::dataset::make("by", 12) * framework::dataset::make("block", 4); const auto data_matrix_multiply = framework::dataset::make("M", 12, 20) * framework::dataset::make("N", 12, 20) * framework::dataset::make("K", 16); } // namespace @@ -79,29 +75,6 @@ TEST_SUITE_END() TEST_SUITE_END() TEST_SUITE(GEMMLowp) - -TEST_SUITE(INTERLEAVE_BLOCKED) - -using NEInterleaveBlocked = NESynthetizeFunction<NEGEMMInterleaveBlockedKernel>; -using NEGEMMInterleaveBlockedFixture = GEMMInterleaveBlockedValidationFixture<Tensor, Accessor, NEInterleaveBlocked>; -FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMInterleaveBlockedFixture, framework::DatasetMode::PRECOMMIT, data_int_blk) -{ - // Validate output - validate(Accessor(_target), _reference); -} -TEST_SUITE_END() - -TEST_SUITE(INTERLEAVE_BLOCKED_TRANSPOSED) -using NEInterleaveBlockedTransposed = NESynthetizeFunction<NEGEMMInterleaveBlockedKernel>; -using NEGEMMInterleaveBlockedTransposedFixture = GEMMInterleaveBlockedValidationFixture<Tensor, Accessor, NEInterleaveBlockedTransposed, true>; -FIXTURE_DATA_TEST_CASE(RunSmall, NEGEMMInterleaveBlockedTransposedFixture, framework::DatasetMode::PRECOMMIT, data_int_blk_tr) -{ - // Validate output - validate(Accessor(_target), _reference); -} - -TEST_SUITE_END() - TEST_SUITE(MatrixMultiplyCore) using NEGEMMLowpMatrixMultiplyCoreFixture = GEMMLowpMatrixMultiplyCoreValidationFixture<Tensor, Accessor, NEGEMMLowpMatrixMultiplyCore>; diff --git a/tests/validation/fixtures/GEMMInterleaveBlockedFixture.h b/tests/validation/fixtures/GEMMInterleaveBlockedFixture.h deleted file mode 100644 index 488324dcc8..0000000000 --- a/tests/validation/fixtures/GEMMInterleaveBlockedFixture.h +++ /dev/null @@ -1,114 +0,0 @@ -/* - * Copyright (c) 2017 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_TEST_GEMM_INTERLEAVE_BLOCKED_FIXTURE -#define ARM_COMPUTE_TEST_GEMM_INTERLEAVE_BLOCKED_FIXTURE - -#include "arm_compute/core/TensorShape.h" -#include "arm_compute/core/Types.h" -#include "tests/AssetsLibrary.h" -#include "tests/Globals.h" -#include "tests/IAccessor.h" -#include "tests/framework/Asserts.h" -#include "tests/framework/Fixture.h" -#include "tests/validation/Helpers.h" -#include "tests/validation/reference/GEMMInterleaveBlocked.h" - -#include <random> - -namespace arm_compute -{ -namespace test -{ -namespace validation -{ -template <typename TensorType, typename AccessorType, typename FunctionType, bool Transposed = false> -class GEMMInterleaveBlockedValidationFixture : public framework::Fixture -{ -public: - template <typename...> - void setup(size_t x, size_t y, int int_by, int block) - { - const float interleave_by_f32 = int_by; - const TensorShape shape_a(x, y); - const TensorShape shape_b(static_cast<size_t>(x * interleave_by_f32), static_cast<size_t>(std::ceil(y / interleave_by_f32))); - _target = compute_target(shape_a, shape_b, int_by, block); - _reference = compute_reference(shape_a, shape_b, int_by, block); - } - -protected: - template <typename U> - void fill(U &&tensor, int i) - { - ARM_COMPUTE_ERROR_ON(tensor.data_type() != DataType::U8); - std::uniform_int_distribution<> distribution(0, 255); - library->fill(tensor, distribution, i); - } - - TensorType compute_target(const TensorShape &shape_a, const TensorShape &shape_b, int int_by, int block) - { - // Create tensors - TensorType a = create_tensor<TensorType>(shape_a, DataType::U8, 1); - TensorType b = create_tensor<TensorType>(shape_b, DataType::U8, 1); - - // Create and configure function - FunctionType f; - f.configure(&a, &b, int_by, block, Transposed); - - ARM_COMPUTE_EXPECT(a.info()->is_resizable(), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(b.info()->is_resizable(), framework::LogLevel::ERRORS); - - // Allocate tensors - a.allocator()->allocate(); - b.allocator()->allocate(); - - ARM_COMPUTE_EXPECT(!a.info()->is_resizable(), framework::LogLevel::ERRORS); - ARM_COMPUTE_EXPECT(!b.info()->is_resizable(), framework::LogLevel::ERRORS); - - // Fill tensors - fill(AccessorType(a), 0); - - // Compute GEMM function - f.run(); - return b; - } - - SimpleTensor<uint8_t> compute_reference(const TensorShape &shape_a, const TensorShape &shape_b, int int_by, int block) - { - // Create reference - SimpleTensor<uint8_t> a{ shape_a, DataType::U8, 1 }; - SimpleTensor<uint8_t> b{ shape_b, DataType::U8, 1 }; - - // Fill reference - fill(a, 0); - return reference::gemm_interleave_blocked<uint8_t>(a, b, int_by, block, Transposed); - } - - TensorType _target{}; - SimpleTensor<uint8_t> _reference{}; -}; - -} // namespace validation -} // namespace test -} // namespace arm_compute -#endif /* ARM_COMPUTE_TEST_GEMM_INTERLEAVE_BLOCKED_FIXTURE */ |