From 368da83fdd7406d629e8cca64f3eb0af05437419 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Mon, 3 Jul 2017 12:33:49 +0100 Subject: 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 Reviewed-by: Georgios Pinitas --- src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/core/CL/kernels/CLGEMMTranspose1xWKernel.cpp') 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(input->info()->dimension(0) * (1.f / scale_x)))); -- cgit v1.2.1