commit e52a3000d2c13bc1b66ca66b3d12b6b836982394
author Gian Marco Iodice <gianmarco.iodice@arm.com> Wed Apr 11 15:59:10 2018 +0100
committer Anthony Barbier <anthony.barbier@arm.com> Fri Nov 02 16:49:37 2018 +0000
tree 70e8ef5ba216762604f84228805aac9bd65747b6
parent dd03870b63784abe499761da2b26b209b33f2db2

COMPMID-1026 - Add support for 4x4 output tile in CLWinogradConvolutionLayer

The performance achieved can be found at the following confluence page:
https://confluence.arm.com/display/MLENG/GEMM-based+convolution+vs+Winograd-based+convolution+on+OpenCL

Change-Id: I4b690cfdd4eb4ff0cd17b14fdd49ccaa1d1dc85c
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/127729
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

tests/datasets/WinogradOutputTransformDataset.h

diff --git a/tests/datasets/WinogradOutputTransformDataset.h b/tests/datasets/WinogradOutputTransformDataset.h
index aa36f15..b28df39 100644
--- a/tests/datasets/WinogradOutputTransformDataset.h
+++ b/tests/datasets/WinogradOutputTransformDataset.h
@@ -110,6 +110,13 @@
         add_config(TensorShape(1U, 442U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
         add_config(TensorShape(7U, 12U, 16U, 3U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
         add_config(TensorShape(24U, 49U, 16U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+
+        // (4x4, 3x3)
+        add_config(TensorShape(13U, 4U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(10U, 9U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 6U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(10U, 11U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(7U, 117U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(53U, 33U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(7U, 4U, 36U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
+        add_config(TensorShape(24U, 16U, 36U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(14U, 14U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(7U, 12U, 16U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(8U, 10U), PadStrideInfo(1, 1, 0, 0), DataLayout::NCHW));
 
         // (4x4, 5x5)
@@ -127,12 +134,21 @@
 public:
     LargeWinogradOutputTransformDataset()
     {
+        // (2x2, 3x3)
         add_config(TensorShape(64U, 12544U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(32U, 3080U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
         add_config(TensorShape(13U, 756U, 16U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
         add_config(TensorShape(64U, 12544U, 16U, 3U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
         add_config(TensorShape(32U, 3080U, 16U, 2U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
         add_config(TensorShape(13U, 756U, 16U, 5U), WinogradInfo(Size2D(2U, 2U), Size2D(3U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+
+        // (4x4, 3x3)
+        add_config(TensorShape(64U, 3136U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(32U, 784U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 196U, 36U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
+        add_config(TensorShape(64U, 3136U, 36U, 3U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(224U, 224U), PadStrideInfo(1, 1, 1, 1), DataLayout::NCHW));
+        add_config(TensorShape(32U, 784U, 36U, 2U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(112U, 112U), PadStrideInfo(1, 1, 1, 0), DataLayout::NCHW));
+        add_config(TensorShape(13U, 196U, 36U, 5U), WinogradInfo(Size2D(4U, 4U), Size2D(3U, 3U), Size2D(56U, 56U), PadStrideInfo(1, 1, 0, 1), DataLayout::NCHW));
     }
 };
 } // namespace datasets