From 4074c995d2a88684fd4a9d1aa36d51de56bb8dab Mon Sep 17 00:00:00 2001 From: Georgios Pinitas Date: Tue, 30 Jan 2018 18:13:46 +0000 Subject: COMPMID-873: Integrate RSH NEON Depthwise Convolution routine Change-Id: Ida1e9a836bc518bfe5563e16bf7f92bde5fc13f7 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/118472 Tested-by: Jenkins Reviewed-by: Pablo Tello --- src/graph/operations/NESimpleOperations.cpp | 20 +++++++++++++++----- 1 file changed, 15 insertions(+), 5 deletions(-) (limited to 'src/graph') diff --git a/src/graph/operations/NESimpleOperations.cpp b/src/graph/operations/NESimpleOperations.cpp index 88bf3ec0a0..5a00e230ea 100644 --- a/src/graph/operations/NESimpleOperations.cpp +++ b/src/graph/operations/NESimpleOperations.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017 ARM Limited. + * Copyright (c) 2017-2018 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -149,13 +149,23 @@ REGISTER_SIMPLE_OPERATION(NEDepthwiseConvolutionOperation, NEON, OperationType:: auto *biases = ctx.num_inputs() == 3 ? dynamic_cast(ctx.input(2)) : nullptr; auto *out = dynamic_cast(ctx.output(0)); const auto conv_info = ctx.parameter("ConvolutionInfo"); + const auto opt3x3 = ctx.parameter("Optimized3x3"); // Create and configure function std::unique_ptr func; - // TODO (COMPMID-769): Add support for asymmetric padding in NEDepthwiseConvolutionLayer3x3 to enable opt3x3 support - auto depwthwise_conv = arm_compute::support::cpp14::make_unique(); - depwthwise_conv->configure(in, weights, biases, out, conv_info); - func = std::move(depwthwise_conv); + bool run_3x3_opt = opt3x3 && weights->info()->dimension(0) == 3; + if(run_3x3_opt) + { + auto depwthwise_conv = arm_compute::support::cpp14::make_unique(); + depwthwise_conv->configure(in, weights, biases, out, conv_info); + func = std::move(depwthwise_conv); + } + else + { + auto depwthwise_conv = arm_compute::support::cpp14::make_unique(); + depwthwise_conv->configure(in, weights, biases, out, conv_info); + func = std::move(depwthwise_conv); + } // Log info ARM_COMPUTE_LOG_GRAPH_INFO("Instantiating NEDepthwiseConvolutionLayer" -- cgit v1.2.1