Optimize Neon™ Logistic Activation

- Use a 1d execution window to improve memory access pattern.

Resolves: [COMPMID-5465]
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Change-Id: Ida30669ffa06eb002ca43a6edf15e25a6eaad2f6
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8344
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 12 insertions, 0 deletions

diff --git a/src/core/helpers/WindowHelpers.h b/src/core/helpers/WindowHelpers.h
index c9e5a135c0..eccf7f2d18 100644
--- a/src/core/helpers/WindowHelpers.h
+++ b/src/core/helpers/WindowHelpers.h
@@ -176,6 +176,18 @@ inline Window calculate_max_enlarged_window(const ITensorInfo &info, const Steps
     return calculate_max_enlarged_window(info.valid_region(), steps, border_size);
 }
 
+/** Calculate the squashed or maximum window for the given tensor shape.
+ *
+ * If the tensor data resides continuously in the memory, the tensor can be interpreted
+ * as 1D array and all the dimensions can be squashed together into the x-dimension.
+ * Otherwise, generate the max window for the given tensor shape.
+ *
+ * @param[in] src Tensor info object defining the shape of the input tensor.
+ *
+ * @return The maximum window the kernel can be executed on and the preferred split dimension.
+ */
+std::pair<Window, size_t> calculate_squashed_or_max_window(const ITensorInfo &src);
+
 /** Calculate the squashed or maximum window for the given tensor shapes.
  *
  * If the tensor data resides continuously in the memory, the tensor can be interpreted