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/validate_examples/cl_gemm.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'tests/validate_examples') diff --git a/tests/validate_examples/cl_gemm.cpp b/tests/validate_examples/cl_gemm.cpp index cdaa33f31a..8b3a103db7 100644 --- a/tests/validate_examples/cl_gemm.cpp +++ b/tests/validate_examples/cl_gemm.cpp @@ -193,7 +193,7 @@ public: init_sgemm_output(tmp_dst, src0, src1, DataType::S32); // Configure GEMMlowp matrix multiply function - mm_gemmlowp.configure(&src0, &src1, &tmp_dst); + mm_gemmlowp.configure(&src0, &src1, nullptr, &tmp_dst); // Configure GEMMlowp output stage mm_gemmlowp_output_stage.configure(&tmp_dst, add_bias ? &biases : nullptr, &dst, dst_multiplier, dst_shift, offset_dst); -- cgit v1.2.1