COMPMID-1995: Fix 32-bit NEDepthwiseConvolution errors.

-Updates padding handling in assembly depthwise kernels.
-Fixes 32-bit runs issues for depthwise convolution.

Change-Id: I3fe6369397c1d13f5629dd34c068ce4af53c95cd
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/939
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 8 insertions, 0 deletions

diff --git a/arm_compute/core/NEON/kernels/convolution/depthwise/depthwise_quantized.hpp b/arm_compute/core/NEON/kernels/convolution/depthwise/depthwise_quantized.hpp
index 4c1d883a70..47fa60139f 100644
--- a/arm_compute/core/NEON/kernels/convolution/depthwise/depthwise_quantized.hpp
+++ b/arm_compute/core/NEON/kernels/convolution/depthwise/depthwise_quantized.hpp
@@ -109,6 +109,14 @@ class QAsymm8DepthwiseConvolution : public DepthwiseConvolutionBase<
       unsigned int out_col_stride
     );
 
+    template <nck::ActivationFunction Activation>
+    void execute_tile(
+      int n_channels,
+      const void* packed_params,
+      const uint8_t* inptrs[Base::inner_tile_rows][Base::inner_tile_cols],
+      uint8_t* outptrs[Base::output_tile_rows][Base::output_tile_cols]
+    );
+
   private:
     // Quantization parameters
     const qasymm8::QAsymm8Params _weights_quant, _inputs_quant, _output_quant;