diff options
| author | Gunes Bayir <gunes.bayir@arm.com> | 2022-09-04 21:00:10 +0100 |
|---|---|---|
| committer | Gunes Bayir <gunes.bayir@arm.com> | 2022-09-09 09:29:43 +0000 |
| commit | 0eed305680ade0c48d07f592c4c4a8aaaad077b7 (patch) | |
| tree | ec4aa5c2e66135d377b5a34f5cf03f97462424c0 /src/cpu/kernels/CpuScaleKernel.cpp | |
| parent | d11de9861e6c32fa389f503e037098f50ffed156 (diff) | |
| download | ComputeLibrary-0eed305680ade0c48d07f592c4c4a8aaaad077b7.tar.gz | |
Optimize FP32/16 Bilinear Scale Kernel for Neon™
This patch removes index and weight pre-computations where it's not used and reduces some calculations inside the inner-most loop of Scale.
Resolves: COMPMID-5452
Change-Id: Ie149b1b76a90a8cb659ada0f97aef78caf69932f
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8220
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels/CpuScaleKernel.cpp')
| -rw-r--r-- | src/cpu/kernels/CpuScaleKernel.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/cpu/kernels/CpuScaleKernel.cpp b/src/cpu/kernels/CpuScaleKernel.cpp index c9e858fc02..e7386a385a 100644 --- a/src/cpu/kernels/CpuScaleKernel.cpp +++ b/src/cpu/kernels/CpuScaleKernel.cpp @@ -140,12 +140,12 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dx, const I ARM_COMPUTE_RETURN_ERROR_ON(output_width == 0); ARM_COMPUTE_RETURN_ERROR_ON(output_height == 0); - if(info.interpolation_policy == InterpolationPolicy::NEAREST_NEIGHBOR) + if(info.interpolation_policy == InterpolationPolicy::NEAREST_NEIGHBOR && offsets != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(offsets, 1, DataType::S32); } - if(info.interpolation_policy == InterpolationPolicy::BILINEAR) + if(info.interpolation_policy == InterpolationPolicy::BILINEAR && offsets != nullptr) { ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(offsets, 1, DataType::S32); if(dx != nullptr && dy != nullptr) |