diff options
author | Giorgio Arena <giorgio.arena@arm.com> | 2017-07-21 10:08:48 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:16:42 +0100 |
commit | a2611815f278334c801094d095901d36e111c3f9 (patch) | |
tree | 1a2bf86b37614ce5473669b283dc37c708dd4957 /src/core/NEON | |
parent | a629da13544c361ba9580a233109b2418fd6bb73 (diff) | |
download | ComputeLibrary-a2611815f278334c801094d095901d36e111c3f9.tar.gz |
COMPMID-417 NEON/CL MeanStdDev bugfix using FillBorderKernel
Change-Id: Ic48ba7f69783d0e1e80611264e2bc67d1732436e
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/81293
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Diffstat (limited to 'src/core/NEON')
-rw-r--r-- | src/core/NEON/kernels/NEMeanStdDevKernel.cpp | 10 | ||||
-rw-r--r-- | src/core/NEON/kernels/NESoftmaxLayerKernel.cpp | 2 |
2 files changed, 10 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/NEMeanStdDevKernel.cpp b/src/core/NEON/kernels/NEMeanStdDevKernel.cpp index 4616203d66..1eb7e45e36 100644 --- a/src/core/NEON/kernels/NEMeanStdDevKernel.cpp +++ b/src/core/NEON/kernels/NEMeanStdDevKernel.cpp @@ -85,10 +85,15 @@ std::pair<uint64x1_t, uint64x1_t> accumulate(const Window &window, Iterator &ite } // namespace NEMeanStdDevKernel::NEMeanStdDevKernel() - : _input(nullptr), _mean(nullptr), _stddev(nullptr), _global_sum(nullptr), _global_sum_squared(nullptr), _mtx() + : _input(nullptr), _mean(nullptr), _stddev(nullptr), _global_sum(nullptr), _global_sum_squared(nullptr), _mtx(), _border_size(0) { } +BorderSize NEMeanStdDevKernel::border_size() const +{ + return _border_size; +} + void NEMeanStdDevKernel::configure(const IImage *input, float *mean, uint64_t *global_sum, float *stddev, uint64_t *global_sum_squared) { ARM_COMPUTE_ERROR_ON_TENSOR_NOT_2D(input); @@ -105,6 +110,9 @@ void NEMeanStdDevKernel::configure(const IImage *input, float *mean, uint64_t *g constexpr unsigned int num_elems_processed_per_iteration = 16; + _border_size = BorderSize(std::max(static_cast<int>(num_elems_processed_per_iteration) - static_cast<int>(input->info()->dimension(0)), + static_cast<int>(input->info()->dimension(0) % num_elems_processed_per_iteration))); + // Configure kernel window Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration)); diff --git a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp index 79fcba1dfb..176e3d688e 100644 --- a/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp +++ b/src/core/NEON/kernels/NESoftmaxLayerKernel.cpp @@ -224,7 +224,7 @@ void NELogits1DMaxKernel::configure(const ITensor *input, ITensor *output) _input = input; _output = output; - _border_size = BorderSize(0, input_width % num_elems_processed_per_iteration, 0, 0); + _border_size = BorderSize(0, num_elems_processed_per_iteration - (input_width % num_elems_processed_per_iteration), 0, 0); // Configure kernel window constexpr unsigned int num_elems_written_per_row = 1; |