From 5ce897f80a1a6ade8a07d61c7aaaf70d2aa5ee02 Mon Sep 17 00:00:00 2001
From: Georgios Pinitas <georgios.pinitas@arm.com>
Date: Wed, 29 Apr 2020 11:44:10 +0100
Subject: COMPMID-3108: Add Winograd 3x3,4x4 FP16 support for NEON

Change-Id: I20680dc74a3d709297539e2132417308a7aecc9d
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3159
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 .../NEON/kernels/convolution/winograd/padding.cpp  | 26 ++++++++++++++++++++++
 1 file changed, 26 insertions(+)

(limited to 'src/core/NEON/kernels/convolution/winograd/padding.cpp')

diff --git a/src/core/NEON/kernels/convolution/winograd/padding.cpp b/src/core/NEON/kernels/convolution/winograd/padding.cpp
index 46fe57c7c9..04aa472c3c 100644
--- a/src/core/NEON/kernels/convolution/winograd/padding.cpp
+++ b/src/core/NEON/kernels/convolution/winograd/padding.cpp
@@ -85,6 +85,15 @@ template void copy_and_pad_tile(
   unsigned int, unsigned int, unsigned int, unsigned int, float
 );
 
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+template void copy_and_pad_tile(
+    unsigned int, unsigned int, unsigned int,
+    const __fp16 *, unsigned int, unsigned int,
+    __fp16 *, unsigned int, unsigned int,
+    unsigned int, unsigned int, unsigned int, unsigned int, __fp16
+);
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+
 template <unsigned int TileRows, unsigned int TileCols>
 void CopyCropped<TileRows, TileCols>::execute(
   const size_t size,
@@ -163,4 +172,21 @@ template void crop_and_copy_tile(
   unsigned int crop_right
 );
 
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+template void crop_and_copy_tile(
+    unsigned int tile_rows,
+    unsigned int tile_cols,
+    unsigned int n_channels,
+    const __fp16 *inptr,
+    unsigned int in_row_stride,
+    unsigned int in_col_stride,
+    __fp16 *outptr,
+    unsigned int out_row_stride,
+    unsigned int out_col_stride,
+    unsigned int crop_top,
+    unsigned int crop_left,
+    unsigned int crop_bottom,
+    unsigned int crop_right
+);
+#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
 }  // namespace padding
-- 
cgit v1.2.1