diff options
author | ramelg01 <ramy.elgammal@arm.com> | 2022-04-07 02:42:52 +0100 |
---|---|---|
committer | Ramy Elgammal <ramy.elgammal@arm.com> | 2022-04-26 15:51:22 +0000 |
commit | 8a164884dddf769643cf3b9f7f94e43cb4f3c20b (patch) | |
tree | 35958dd48b6df1a851c880dad2b2ce285671b611 /src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp | |
parent | c827e99fc46521f43719b0c2d1b6f05d66abf68c (diff) | |
download | ComputeLibrary-8a164884dddf769643cf3b9f7f94e43cb4f3c20b.tar.gz |
Update Neon™ depthwise kernel
- Reduce duplication and simplify overall structure.
- Improve multi-threaded performance by sharing more data
in lower-level caches.
Partially Resolves: COMPMID-5054
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Change-Id: Iac747f39b21c540122fa75218762631c4d787911
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7449
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Andrew Mundy
Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp | 13 |
1 files changed, 3 insertions, 10 deletions
diff --git a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp index ea41529d81..0665fa3a29 100644 --- a/src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp +++ b/src/core/NEON/kernels/arm_conv/depthwise/depthwise_implementation.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2021 Arm Limited. + * Copyright (c) 2021-2022 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -24,7 +24,7 @@ #pragma once -#include "src/core/NEON/kernels/assembly/depthwise.hpp" +#include "depthwise.hpp" #include <cstddef> #include <functional> @@ -136,14 +136,7 @@ UniqueDepthwiseCommon<TInput, TWeight, TOutput> depthwise(const DepthwiseArgs &a { const DepthwiseImplementation<TInput, TWeight, TOutput, OutputStage> *impl = nullptr; const bool success = find_implementation<TInput, TWeight, TOutput, OutputStage>(args, os, impl); - - if(success) - { - auto i = impl->get_instance(args, os); - i->set_name(impl->name); - return UniqueDepthwiseCommon<TInput, TWeight, TOutput>(i); - } - return nullptr; + return UniqueDepthwiseCommon<TInput, TWeight, TOutput>(success ? impl->get_instance(args, os) : nullptr); } } // namespace depthwise |