diff options
author | Pablo Tello <pablo.tello@arm.com> | 2017-11-15 13:28:27 +0000 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:35:24 +0000 |
commit | 181e65145d153210ec5587a42d2938e27e1d5b01 (patch) | |
tree | 70115705382ec4997d2f1ff44a33224f50ace38a /arm_compute/runtime/NEON | |
parent | bc8fb0634339dfd662f4b2d825f74615b8a69bac (diff) | |
download | ComputeLibrary-181e65145d153210ec5587a42d2938e27e1d5b01.tar.gz |
COMPMID-675: NEGEMMLowp Assembly Integration
Added support for S8 input in NEGEMMLowp Matrix Multiply Kernel.
Added a new function to run assembly kernels such that A*B=C (no offsets involved)
Added new tests for the assembly gemmlowp kernels (no offsets)
Integrated the assembly kernel for the A57
Change-Id: Ib3e39c1f3f7f1baa0d39be69485f61cd18e3c9b3
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/95864
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON')
-rw-r--r-- | arm_compute/runtime/NEON/NEFunctions.h | 1 | ||||
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h | 69 |
2 files changed, 70 insertions, 0 deletions
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h index 118603b20b..203df33692 100644 --- a/arm_compute/runtime/NEON/NEFunctions.h +++ b/arm_compute/runtime/NEON/NEFunctions.h @@ -60,6 +60,7 @@ #include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h" #include "arm_compute/runtime/NEON/functions/NEGEMM.h" #include "arm_compute/runtime/NEON/functions/NEGEMMInterleave4x4.h" +#include "arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h" #include "arm_compute/runtime/NEON/functions/NEGEMMLowpMatrixMultiplyCore.h" #include "arm_compute/runtime/NEON/functions/NEGEMMLowpOutputStage.h" #include "arm_compute/runtime/NEON/functions/NEGEMMTranspose1xW.h" diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h b/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h new file mode 100644 index 0000000000..3b6aa1c7db --- /dev/null +++ b/arm_compute/runtime/NEON/functions/NEGEMMLowpAssemblyMatrixMultiplyCore.h @@ -0,0 +1,69 @@ + +/* + * 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_NEGEMMLOWPASSEMBLYMATRIXMULTIPLYCORE_H__ +#define __ARM_COMPUTE_NEGEMMLOWPASSEMBLYMATRIXMULTIPLYCORE_H__ + +#include "arm_compute/core/NEON/INEKernel.h" +#include "arm_compute/runtime/IFunction.h" +#include "arm_compute/runtime/IMemoryManager.h" +#include "arm_compute/runtime/MemoryGroup.h" +#include "arm_compute/runtime/Tensor.h" + +#include <memory> + +namespace arm_compute +{ +class ITensor; + +/** Basic function to execute matrix multiply assembly kernels. + * +*/ +class NEGEMMLowpAssemblyMatrixMultiplyCore : public IFunction +{ +public: + /** Constructor */ + NEGEMMLowpAssemblyMatrixMultiplyCore(std::shared_ptr<IMemoryManager> memory_manager = nullptr); + /** Initialise the kernel's inputs, output + * + * @param[in] a First input tensor (Matrix A). Data type supported: U8, S8. + * @param[in] b Second input tensor (Matrix B). Data type supported: same as @p a + * @param[out] output Output tensor. Data type supported: Data type supported: S32 + */ + void configure(const ITensor *a, const ITensor *b, ITensor *output); + + // Inherited methods overridden: + void run() override; + +private: + MemoryGroup _memory_group; + std::unique_ptr<INEKernel> _mm_kernel; + std::unique_ptr<INEKernel> _mtx_a_reshape_kernel; + std::unique_ptr<INEKernel> _mtx_b_reshape_kernel; + Tensor _tmp_a; + Tensor _tmp_b; + Tensor _workspace; +}; +} +#endif /*__ARM_COMPUTE_NEGEMMLOWPASSEMBLYMATRIXMULTIPLYCORE_H__ */ |