commit 4b90865ab985d571f70c60583cdfb8c7a65f1670
author Gian Marco Iodice <gianmarco.iodice@arm.com> Thu Oct 18 10:21:02 2018 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:55:45 2018 +0000
tree f116a4ffef5f5e823689dd00c1e5c9d987f3d295
parent c55beee7ef70fa08a5d217619083b288a74fcb27

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 <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp

diff --git a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
index fb6d4a1..fc65469 100644
--- a/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
+++ b/src/runtime/NEON/functions/NEGEMMConvolutionLayer.cpp
@@ -111,7 +111,7 @@
         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 @@
         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
     {