From 29e27b0544d99e5d98f044a9e606db8abcfb8900 Mon Sep 17 00:00:00 2001 From: Pablo Marquez Tello Date: Thu, 3 Aug 2023 14:47:31 +0100 Subject: Add support for S64 output in NEArgMinMaxLayer * NEArgMinMaxLayer uses NEReductionOperation to compute its result in S32 * We need to call NECast to convert from S32 to S64 * Resolves MLCE-1089 Change-Id: I6fded869b6076d7af1b9b3e70eb384f4ee82fd8a Signed-off-by: Pablo Marquez Tello Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10054 Tested-by: Arm Jenkins Reviewed-by: Viet-Hoa Do Comments-Addressed: Arm Jenkins Benchmark: Arm Jenkins --- src/runtime/NEON/functions/NEArgMinMaxLayer.cpp | 44 +++++++++++++++++++++---- 1 file changed, 38 insertions(+), 6 deletions(-) (limited to 'src/runtime/NEON/functions') diff --git a/src/runtime/NEON/functions/NEArgMinMaxLayer.cpp b/src/runtime/NEON/functions/NEArgMinMaxLayer.cpp index 3876ae6e87..3ac127b02e 100644 --- a/src/runtime/NEON/functions/NEArgMinMaxLayer.cpp +++ b/src/runtime/NEON/functions/NEArgMinMaxLayer.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2021 Arm Limited. + * Copyright (c) 2018-2021, 2023 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -29,22 +29,49 @@ #include "arm_compute/core/TensorInfo.h" #include "arm_compute/core/Types.h" #include "arm_compute/core/Validate.h" +#include "arm_compute/runtime/NEON/functions/NECast.h" +#include "arm_compute/runtime/NEON/functions/NEReductionOperation.h" +#include "arm_compute/runtime/Tensor.h" #include "src/common/utils/Log.h" #include "src/core/NEON/kernels/NEReductionOperationKernel.h" namespace arm_compute { +struct NEArgMinMaxLayer::Impl +{ + MemoryGroup memory_group{}; + std::shared_ptr memory_manager{}; + std::unique_ptr reduction_function{}; + std::unique_ptr cast_function{}; + std::unique_ptr tmp_reduction_result{}; +}; + NEArgMinMaxLayer::~NEArgMinMaxLayer() = default; NEArgMinMaxLayer::NEArgMinMaxLayer(std::shared_ptr memory_manager) - : _reduction_function(std::make_unique()) + : _impl(std::make_unique()) { - ARM_COMPUTE_UNUSED(memory_manager); + _impl->memory_manager = std::move(memory_manager); } + void NEArgMinMaxLayer::configure(ITensor *input, int axis, ITensor *output, const ReductionOperation &op) { ARM_COMPUTE_LOG_PARAMS(input, axis, output, op); - _reduction_function->configure(input, output, axis, op, false); + _impl->reduction_function = std::make_unique(); + if(output->info() && (output->info()->data_type() == DataType::S64 || output->info()->data_type() == DataType::U64)) + { + _impl->memory_group = MemoryGroup(std::move(_impl->memory_manager)); + _impl->cast_function = std::make_unique(); + _impl->tmp_reduction_result = std::make_unique(); + _impl->reduction_function->configure(input, _impl->tmp_reduction_result.get(), axis, op, false); + _impl->cast_function->configure(_impl->tmp_reduction_result.get(), output, ConvertPolicy::SATURATE); + _impl->memory_group.manage(_impl->tmp_reduction_result.get()); + _impl->tmp_reduction_result->allocator()->allocate(); + } + else + { + _impl->reduction_function->configure(input, output, axis, op, false); + } } Status NEArgMinMaxLayer::validate(const ITensorInfo *input, int axis, const ITensorInfo *output, const ReductionOperation &op) @@ -55,7 +82,12 @@ Status NEArgMinMaxLayer::validate(const ITensorInfo *input, int axis, const ITen void NEArgMinMaxLayer::run() { - _reduction_function->run(); + MemoryGroupResourceScope scope_mg(_impl->memory_group); + _impl->reduction_function->run(); + if(_impl->tmp_reduction_result != nullptr) + { + _impl->cast_function->run(); + } } -} // namespace arm_compute \ No newline at end of file +} // namespace arm_compute -- cgit v1.2.1