COMPMID-477 - Optimizing CLDirectConvolution 3x3 on OpenCL and added the auto configuration
Change-Id: I3c8384dcbc9d7786943134bb658dafb35356d90d
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/83253
Reviewed-by: Steven Niu <steven.niu@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
diff --git a/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerKernel.h b/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerKernel.h
index e0dac98..5672782 100644
--- a/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEDirectConvolutionLayerKernel.h
@@ -48,6 +48,10 @@
~NEDirectConvolutionLayerKernel() = default;
/** Set the input, weights, and output tensors.
*
+ * @note: DirectConvolution only works in the following configurations:
+ * 1x1 convolution with stride_x = 1/2/3, stride_y = 1/2/3
+ * 3x3 convolution with stride_x = 1/2/3, stride_y = 1/2/3
+ *
* @param[in] input The input tensor to convolve. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs. Data types supported: QS8/QS16/F16/F32.
* @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].