aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp')
-rw-r--r--src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp5
1 files changed, 4 insertions, 1 deletions
diff --git a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
index e8efe6f0a9..770740d180 100644
--- a/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
+++ b/src/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.cpp
@@ -171,7 +171,10 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
{
const bool is_dot8_supported = dot8_supported(CLKernelLibrary::get().get_device());
- kernel_name = is_qasymm ? (std::string("depthwise_convolution_3x3_quantized") + (is_dot8_supported ? "_dot8" : "") + "_nchw") : "depthwise_convolution_3x3";
+ kernel_name = is_qasymm ? "dwc_3x3_native_qasymm8" : "depthwise_convolution_3x3";
+ kernel_name += (is_qasymm && is_dot8_supported ? "_dot8" : "");
+ kernel_name += "_nchw";
+
num_elems_written_per_iteration_x = 8 / data_size_from_type(input->data_type());
num_elems_written_per_iteration_y = (is_qasymm && conv_stride_y == 1) ? 2 : 1;
num_elems_read_per_iteration_x = 3 + (num_elems_written_per_iteration_x - 1) * conv_stride_x;