diff options
| author | Gian Marco <gianmarco.iodice@arm.com> | 2017-11-29 10:41:38 +0000 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:41:36 +0000 |
| commit | 5420b28d54c5238ed4c4aa2ccf02a7c3855c6760 (patch) | |
| tree | 44d201142f71622eaf9d711e2d38d4e1360586df /src/core/NEON/kernels/NETransposeKernel.cpp | |
| parent | a5e6d67ddad4c88cec87b23063e8493768bb2c61 (diff) | |
| download | ComputeLibrary-5420b28d54c5238ed4c4aa2ccf02a7c3855c6760.tar.gz | |
COMPMID-710 - Fix CLTranspose and NETranspose when the input shape is a
vector
Fixed replacing AccessWindowTranspose with AccessWindowStatic due to the
wrong padding calculation in the X direction within AccessWindowTranspose.
AccessWindowTranspose should be fixed with COMPMID-708
Change-Id: I665e130b897a213ae2bf9da4bf092dd491fa00c6
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111057
Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com <bsgcomp@arm.com>
Reviewed-by: Michel Iwaniec <michel.iwaniec@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NETransposeKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NETransposeKernel.cpp | 10 |
1 files changed, 7 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/NETransposeKernel.cpp b/src/core/NEON/kernels/NETransposeKernel.cpp index 1cfaafe93a..44becf5a4b 100644 --- a/src/core/NEON/kernels/NETransposeKernel.cpp +++ b/src/core/NEON/kernels/NETransposeKernel.cpp @@ -23,10 +23,11 @@ */ #include "arm_compute/core/NEON/kernels/NETransposeKernel.h" -#include "arm_compute/core/AccessWindowTranspose.h" +#include "arm_compute/core/AccessWindowStatic.h" #include "arm_compute/core/Error.h" #include "arm_compute/core/Helpers.h" #include "arm_compute/core/ITensor.h" +#include "arm_compute/core/Utils.h" #include "arm_compute/core/Validate.h" #include <arm_neon.h> @@ -221,8 +222,11 @@ void NETransposeKernel::configure(const ITensor *input, ITensor *output) } // Configure kernel window - Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration, num_elems_processed_per_iteration)); - AccessWindowTranspose output_access(output->info(), 0, 0, num_elems_processed_per_iteration, num_elems_processed_per_iteration); + Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration, num_elems_processed_per_iteration)); + + // TODO (COMPMID-708): Replace AccessWindowStatic with AccessWindowTranspose + AccessWindowStatic output_access(output->info(), 0, 0, ceil_to_multiple(output->info()->dimension(0), num_elems_processed_per_iteration), ceil_to_multiple(output->info()->dimension(1), + num_elems_processed_per_iteration)); update_window_and_padding(win, AccessWindowRectangle(input->info(), 0, 0, num_elems_processed_per_iteration, num_elems_processed_per_iteration), |