From 4b90865ab985d571f70c60583cdfb8c7a65f1670 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Thu, 18 Oct 2018 10:21:02 +0100 Subject: COMPMID-1413 - Improve the performance of GEMMLowp with 8 bit dot product on OpenCL COMPMID-1424 - Add dot product support for CLDepthwise QASYMM8 3x3 NHWC non-unit stride With this patch we are able to improve the performance of MobileNet v1-qasymm8 by 37 % Tried to use the dot product instruction in CLDepthwise QASYMM8 3x3 NHWC non-unit stride but I have not seen any benefit (maybe because we have few arithemtic operation and we do not have more load instructions). However Depthwise convolution has been improved by 30% Change-Id: Id768a99c2e53a04276707e427af5d0ec93419ada Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/155082 Tested-by: bsgcomp Reviewed-by: Georgios Pinitas --- src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp') diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp index fb6d4a1847..fc65469488 100644 --- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp +++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp @@ -111,7 +111,7 @@ void NEGEMMConvolutionLayer::configure_mm(const ITensor *input, const ITensor *w input->info()->set_quantization_info(QuantizationInfo(input_quantization_info.scale, -input_quantization_info.offset)); weights->info()->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset)); - _mm_gemmlowp.configure(input, weights, output, GEMMInfo(false, false, true /* Reshape weights only for the first run*/)); + _mm_gemmlowp.configure(input, weights, nullptr, output, GEMMInfo(false, false, true /* Reshape weights only for the first run*/)); // Revert back QuantizatioInfo as input and weights could be used in other convolution layers input->info()->set_quantization_info(input_quantization_info); @@ -143,7 +143,7 @@ Status NEGEMMConvolutionLayer::validate_mm(const ITensorInfo *input, const ITens weights_qa->set_quantization_info(QuantizationInfo(weights_quantization_info.scale, -weights_quantization_info.offset)); // Perform validation step on GEMMLowp - return NEGEMMLowpMatrixMultiplyCore::validate(input_qa.get(), weights_qa.get(), output, gemm_info); + return NEGEMMLowpMatrixMultiplyCore::validate(input_qa.get(), weights_qa.get(), nullptr, output, gemm_info); } else { -- cgit v1.2.1