aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp
diff options
context:
space:
mode:
authorGian Marco Iodice <gianmarco.iodice@arm.com>2019-05-14 10:14:08 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2019-06-03 12:09:49 +0000
commit43a129e94df41f9ac8bc78b702da5a387ada0494 (patch)
tree900463d2235bc59e5492a934f33a949aa629a40e /src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp
parentd7dd15c445397ab879439de6659859db09f4b752 (diff)
downloadComputeLibrary-43a129e94df41f9ac8bc78b702da5a387ada0494.tar.gz
COMPMID-2379: Use the macros available in gemm_helpers.h in GEMMLowp OpenCL kernels
Change-Id: I09923a068bff36d42a3f2c1084ffa8bf218187b9 Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com> Reviewed-on: https://review.mlplatform.org/c/1260 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp')
-rw-r--r--src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp3
1 files changed, 2 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp
index a8c1704d91..050b792c4e 100644
--- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.cpp
@@ -214,7 +214,6 @@ void CLGEMMLowpMatrixMultiplyReshapedKernel::configure(const ICLTensor *input0,
std::string kernel_name("gemmlowp_mm_reshaped_");
kernel_name += lhs_info.transpose ? "lhs_t_" : "lhs_nt_";
kernel_name += rhs_info.transpose ? "rhs_t" : "rhs_nt";
- kernel_name += dot8_supported(CLKernelLibrary::get().get_device()) ? "_dot8" : "";
// Create kernel
_kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name, build_opts.options()));
@@ -222,6 +221,8 @@ void CLGEMMLowpMatrixMultiplyReshapedKernel::configure(const ICLTensor *input0,
// Set config_id for enabling LWS tuning
_config_id = kernel_name;
_config_id += "_";
+ _config_id += dot8_supported(CLKernelLibrary::get().get_device()) ? "_dot8" : "";
+ _config_id += "_";
_config_id += (_reinterpret_output_as_3d ? "3do_" : "");
_config_id += support::cpp11::to_string(output->info()->dimension(1));
_config_id += "_";