| commit | 0d11b70fbfa95431dacd7dce02403cf90bc688d5 | [log] [tgz] |
|---|---|---|
| author | Pablo Tello <pablo.tello@arm.com> | Fri Oct 08 11:22:24 2021 +0100 |
| committer | Pablo Marquez Tello <pablo.tello@arm.com> | Mon Oct 11 16:41:45 2021 +0000 |
| tree | df504645a1a3935ea27fc4e710374f74820e35be | |
| parent | 16def8d847b9cdf9c13319771bf5a3b97d4970cf [diff] |
Fix precision issue in ChannelShuffleKernel * Partially resolves COMPMID-4831 Change-Id: I2100432c51f2a4d2f2524ed7a3574035632c092c Signed-off-by: Pablo Tello <pablo.tello@arm.com> Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6403 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Freddie Liardet <frederick.liardet@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/NEON/kernels/NEChannelShuffleLayerKernel.cpp b/src/core/NEON/kernels/NEChannelShuffleLayerKernel.cpp index 008ad7c..85ae974 100644 --- a/src/core/NEON/kernels/NEChannelShuffleLayerKernel.cpp +++ b/src/core/NEON/kernels/NEChannelShuffleLayerKernel.cpp
@@ -103,7 +103,7 @@ const size_t row_size = input->info()->dimension(width_idx) * input->info()->element_size(); const unsigned int K = input->info()->dimension(channel_idx) / num_groups; - const float rK = 1.f / K; + const double rK = 1.0 / K; Iterator in(input, win);