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/core/CL/CLHelpers.cpp | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'src/core/CL/CLHelpers.cpp') diff --git a/src/core/CL/CLHelpers.cpp b/src/core/CL/CLHelpers.cpp index 5c435ddc22..0947d58973 100644 --- a/src/core/CL/CLHelpers.cpp +++ b/src/core/CL/CLHelpers.cpp @@ -144,7 +144,12 @@ bool fp16_supported(const cl::Device &device) bool dot8_supported(const cl::Device &device) { - return device_supports_extension(device, "cl_arm_integer_dot_product_int8"); + std::string device_name = device.getInfo(); + const GPUTarget gpu_target = get_target_from_name(device_name); + + // SW_WORKAROUND: Workaround for DDK revision r14p0.to enable cl_arm_integer_dot_product_int8 + std::set sw_workaround_issue = {GPUTarget::G76}; + return (device_supports_extension(device, "cl_arm_integer_dot_product_int8") || sw_workaround_issue.count(gpu_target) != 0); } bool dot8_acc_supported(const cl::Device &device) -- cgit v1.2.1