From 760bd0397dc89d0ccddbf293c9139a5727b7d877 Mon Sep 17 00:00:00 2001
From: Pauli Oikkonen <pauli.oikkonen@tut.fi>
Date: Wed, 30 Jan 2019 01:33:39 +0200
Subject: [PATCH] Pad the image buffer by 64 bytes from both ends

This will be necessary for an efficient and straightforward
implementation of hor_sad for blocks over 16 pixels wide, because they
cannot use the shuffle trick because inter-lane shuffling is so hard to
do
---
 src/global.h  |  2 ++
 src/image.c   | 12 ++++++++----
 src/kvazaar.h |  3 ++-
 3 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/src/global.h b/src/global.h
index c7e6c8b0..5ced48a1 100644
--- a/src/global.h
+++ b/src/global.h
@@ -239,8 +239,10 @@ typedef int16_t coeff_t;
 #ifdef _MSC_VER
 // Buggy VS2010 throws intellisense warnings if void* is not casted.
   #define MALLOC(type, num) (type *)malloc(sizeof(type) * (num))
+  #define MALLOC_SIMD_PADDED(type, num, padding) (type *)malloc(sizeof(type) * (num) + (padding))
 #else
   #define MALLOC(type, num) malloc(sizeof(type) * (num))
+  #define MALLOC_SIMD_PADDED(type, num, padding) malloc(sizeof(type) * (num) + (padding))
 #endif
 
 // Use memset through FILL and FILL_ARRAY when appropriate, such as when
diff --git a/src/image.c b/src/image.c
index fe626d72..6d973b59 100644
--- a/src/image.c
+++ b/src/image.c
@@ -47,6 +47,8 @@ kvz_picture * kvz_image_alloc(enum kvz_chroma_format chroma_format, const int32_
   assert((width % 2) == 0);
   assert((height % 2) == 0);
 
+  const size_t simd_padding_width = 64;
+
   kvz_picture *im = MALLOC(kvz_picture, 1);
   if (!im) return NULL;
 
@@ -56,12 +58,13 @@ kvz_picture * kvz_image_alloc(enum kvz_chroma_format chroma_format, const int32_
 
   im->chroma_format = chroma_format;
 
-  //Allocate memory
-  im->fulldata = MALLOC(kvz_pixel, (luma_size + 2 * chroma_size));
-  if (!im->fulldata) {
+  //Allocate memory, pad the full data buffer from both ends
+  im->fulldata_buf = MALLOC_SIMD_PADDED(kvz_pixel, (luma_size + 2 * chroma_size), simd_padding_width * 2);
+  if (!im->fulldata_buf) {
     free(im);
     return NULL;
   }
+  im->fulldata = im->fulldata_buf + simd_padding_width / sizeof(kvz_pixel);
 
   im->base_image = im;
   im->refcount = 1; //We give a reference to caller
@@ -110,11 +113,12 @@ void kvz_image_free(kvz_picture *const im)
     // Free our reference to the base image.
     kvz_image_free(im->base_image);
   } else {
-    free(im->fulldata);
+    free(im->fulldata_buf);
   }
 
   // Make sure freed data won't be used.
   im->base_image = NULL;
+  im->fulldata_buf = NULL;
   im->fulldata = NULL;
   im->y = im->u = im->v = NULL;
   im->data[COLOR_Y] = im->data[COLOR_U] = im->data[COLOR_V] = NULL;
diff --git a/src/kvazaar.h b/src/kvazaar.h
index 7209b636..f98a818b 100644
--- a/src/kvazaar.h
+++ b/src/kvazaar.h
@@ -392,7 +392,8 @@ typedef struct kvz_config
  * Function picture_alloc in kvz_api must be used for allocation.
  */
 typedef struct kvz_picture {
-  kvz_pixel *fulldata;         //!< \brief Allocated buffer (only used in the base_image)
+  kvz_pixel *fulldata_buf;     //!< \brief Allocated buffer with padding (only used in the base_image)
+  kvz_pixel *fulldata;         //!< \brief Allocated buffer portion that's actually used
 
   kvz_pixel *y;                //!< \brief Pointer to luma pixel array.
   kvz_pixel *u;                //!< \brief Pointer to chroma U pixel array.