diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2017-07-03 12:33:49 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:15:39 +0100 |
| commit | 368da83fdd7406d629e8cca64f3eb0af05437419 (patch) | |
| tree | fadac4142651cb0f86b997c06cbabb1bec622aae /src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp | |
| parent | adffa30de9292c96bf29ff0697ac573270046612 (diff) | |
| download | ComputeLibrary-368da83fdd7406d629e8cca64f3eb0af05437419.tar.gz | |
COMPMID-420, COMPMID-414 - Port CLConvolutionLayer and CLFullyConnectedLayer to use 8 bit fixed point
Change-Id: I1cb1b4d7711ad7b569ee691e13a5df1b3430292b
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/79565
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp')
| -rw-r--r-- | src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp b/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp index 73c8429055..27b215f2c8 100644 --- a/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp +++ b/src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp @@ -56,7 +56,7 @@ void CLGEMMTranspose1xWKernel::configure(const ICLTensor *input, ICLTensor *outp ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape); ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output); - const unsigned int num_elems_processed_per_iteration = max_cl_vector_width / data_size_from_type(input->info()->data_type()); + const unsigned int num_elems_processed_per_iteration = 16 / input->info()->element_size(); const float scale_x = num_elems_processed_per_iteration; ARM_COMPUTE_ERROR_ON((0 == static_cast<int>(input->info()->dimension(0) * (1.f / scale_x)))); |