COMPMID-345: Added support for 5x5 kernels in NEDirectConvolution
Change-Id: I25cd8f057566b59ce40e2acf14714e83a286ae4e
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/83791
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
diff --git a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
index 872fae3..9d2775a 100644
--- a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
@@ -48,8 +48,14 @@
NEDirectConvolutionLayer();
/** Set the input, weights, biases and output tensors.
*
+ * @note: DirectConvolution only works in the following configurations:
+ * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = QS8/QS16/F16/F32
+ * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = QS8/F16/F32
+ * 5x5 convolution with stride_x = 1/2/3, stride_y = 1/2/3 data type = F32
+ *
* @param[in, out] input Input tensor. Data types supported: QS8/QS16/F16/F32.
* @param[in] weights Set of kernels to convolve the input volume.
+ * Supported sizes: 1x1, 3x3 and 5x5.
* The 3rd dimension must be the same as the input's volume 3rd dimension.
* Data type supported: Same as @p input.
* @param[in] bias Set of biases. Data type supported: Same as @p input.