commit a9676118fd2a0e5bc916969af83ecee049bae76b
author Michalis Spyrou <michalis.spyrou@arm.com> Thu Feb 22 18:07:43 2018 +0000
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:48:35 2018 +0000
tree f67ff64d962a2f802700f6b26a9ab160c04c721d
parent 2bc74410251dcbaf17a7c5447317aa6d0171972a

COMPMID-886 Don't use LWS hints by default for GPU post Mali-G72

Change-Id: I64cb2d7f9513d69aebd9307a803b1b2c9c0e04c3
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/121929
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>

src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.cpp

diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.cpp
index 29564b3..7a47bcc 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3Kernel.cpp
@@ -118,7 +118,7 @@
 
     // Configure the local work size for Bifrost with a value obtained
     // via exhaustive autotuning for the MobileNets tensor shapes.
-    const GPUTarget gpu_target = get_arch_from_target(get_target());
+    const GPUTarget gpu_target = get_target();
 
     // Configure kernel window
     unsigned int num_elems_read_per_iteration_x    = 0;
@@ -151,7 +151,7 @@
                 break;
         }
     }
-    else if(input->info()->data_type() == DataType::F32 && gpu_target == GPUTarget::BIFROST)
+    else if(input->info()->data_type() == DataType::F32 && gpu_target_is_in(gpu_target, GPUTarget::G71, GPUTarget::G72))
     {
         if(_conv_stride_x == 1 && _conv_stride_y == 1)
         {