diff options
| author | Gian Marco Iodice <gianmarco.iodice@arm.com> | 2017-06-22 13:00:39 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:14:20 +0100 |
| commit | ec8b45e7333006ae06d834e300c11a4fd1b4c067 (patch) | |
| tree | 6b6b77f74a15c39488bef0ed826258807ffd8bd1 /src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp | |
| parent | 9f89baebb81e6a01ec06fe916564da45eb204f34 (diff) | |
| download | ComputeLibrary-ec8b45e7333006ae06d834e300c11a4fd1b4c067.tar.gz | |
COMPMID-345 - Auto-inizialization for NECol2ImKernel and NEGEMMInterleave4x4Kernel
Added support for 8 bit fixed point in CLTransposeKernel
Change-Id: I8257fa30e90e70825c97c16b0a11af73c426319c
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/78563
Reviewed-by: Moritz Pflanzer <moritz.pflanzer@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp | 14 |
1 files changed, 11 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp index 3ff8b7b201..4505dcb363 100644 --- a/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp +++ b/src/core/NEON/kernels/NEGEMMInterleave4x4Kernel.cpp @@ -133,10 +133,18 @@ NEGEMMInterleave4x4Kernel::NEGEMMInterleave4x4Kernel() void NEGEMMInterleave4x4Kernel::configure(const ITensor *input, ITensor *output) { ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::U8, DataType::S8, DataType::U16, DataType::S16, DataType::U32, DataType::S32, DataType::F16, DataType::F32); - ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QS8, DataType::U8, DataType::S8, DataType::U16, DataType::S16, DataType::U32, DataType::S32, DataType::F16, DataType::F32); + ARM_COMPUTE_ERROR_ON_NULLPTR(output); + + TensorShape output_shape = input->info()->tensor_shape(); + output_shape.set(0, input->info()->dimension(0) * 4); + output_shape.set(1, std::ceil(input->info()->dimension(1) / 4.0f)); + + // Output auto inizialitation if not yet initialized + auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->fixed_point_position()); + + ARM_COMPUTE_ERROR_ON_MISMATCHING_DIMENSIONS(output->info()->tensor_shape(), output_shape); ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); - ARM_COMPUTE_ERROR_ON(output->info()->dimension(0) != input->info()->dimension(0) * 4); - ARM_COMPUTE_ERROR_ON(output->info()->dimension(1) != std::ceil(input->info()->dimension(1) / 4.0f)); + ARM_COMPUTE_ERROR_ON_MISMATCHING_FIXED_POINT(input, output); _input = input; _output = output; |