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 --- tests/benchmark/fixtures/GEMMLowpFixture.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'tests/benchmark') diff --git a/tests/benchmark/fixtures/GEMMLowpFixture.h b/tests/benchmark/fixtures/GEMMLowpFixture.h index 46a2f5cc6a..33c6415d20 100644 --- a/tests/benchmark/fixtures/GEMMLowpFixture.h +++ b/tests/benchmark/fixtures/GEMMLowpFixture.h @@ -58,7 +58,7 @@ public: c = create_tensor(shape_dst, DataType::S32, 1, QuantizationInfo(1.0f / 255.0f, 0)); // Create and configure function - gemmlowp.configure(&a, &b, &c); + gemmlowp.configure(&a, &b, nullptr, &c); // Allocate tensors a.allocator()->allocate(); -- cgit v1.2.1