From d91d7b9826bb877e18498f60dfa2be3621735dec Mon Sep 17 00:00:00 2001
From: homm <homm86@gmail.com>
Date: Sat, 18 Oct 2014 23:56:17 +0400
Subject: [PATCH] significant speed improvement (about 55%)

---
 libImaging/BoxBlur.c | 46 ++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 42 insertions(+), 4 deletions(-)

diff --git a/libImaging/BoxBlur.c b/libImaging/BoxBlur.c
index 2f72e407c..57f97e595 100644
--- a/libImaging/BoxBlur.c
+++ b/libImaging/BoxBlur.c
@@ -202,7 +202,7 @@ HorizontalBoxBlur(Imaging im, Imaging imOut, float floatRadius)
             );
             // Commit.
             for (x = 0; x < im->xsize; x++) {
-                imOut->image8[x][y] = ((UINT8 *)lineOut)[x];
+                imOut->image8[y][x] = ((UINT8 *)lineOut)[x];
             }
         }
     }
@@ -218,7 +218,7 @@ HorizontalBoxBlur(Imaging im, Imaging imOut, float floatRadius)
             );
             // Commit.
             for (x = 0; x < im->xsize; x++) {
-                imOut->image32[x][y] = lineOut[x];
+                imOut->image32[y][x] = lineOut[x];
             }
         }
     }
@@ -230,6 +230,42 @@ HorizontalBoxBlur(Imaging im, Imaging imOut, float floatRadius)
     return imOut;
 }
 
+void
+TransposeImage(Imaging im, Imaging imOut)
+{
+    int x, y, xx, yy, xxsize, yysize;
+    int size = 64;
+
+    if (im->image8)
+    {
+        for (y = 0; y < im->ysize; y += size) {
+            for (x = 0; x < im->xsize; x += size) {
+                yysize = MIN(size, im->ysize - y);
+                xxsize = MIN(size, im->xsize - x);
+                for (yy = 0; yy < yysize; yy++) {
+                    for (xx = 0; xx < xxsize; xx++) {
+                        imOut->image8[x + xx][y + yy] = im->image8[y + yy][x + xx];
+                    }
+                }
+            }
+        }
+    }
+    else
+    {
+        for (y = 0; y < im->ysize; y += size) {
+            for (x = 0; x < im->xsize; x += size) {
+                yysize = MIN(size, im->ysize - y);
+                xxsize = MIN(size, im->xsize - x);
+                for (yy = 0; yy < yysize; yy++) {
+                    for (xx = 0; xx < xxsize; xx++) {
+                        imOut->image32[x + xx][y + yy] = im->image32[y + yy][x + xx];
+                    }
+                }
+            }
+        }
+    }
+}
+
 
 Imaging
 ImagingBoxBlur(Imaging im, Imaging imOut, float radius)
@@ -259,12 +295,14 @@ ImagingBoxBlur(Imaging im, Imaging imOut, float radius)
 
     /* Apply one-dimensional blur.
        HorizontalBoxBlur32 transposes image at same time. */
-    HorizontalBoxBlur(im, temp, radius);
+    HorizontalBoxBlur(im, imOut, radius);
+    TransposeImage(imOut, temp);
 
     /* Blur transposed result from previout step in same direction.
        Reseult will be transposed again. We'll get original image
        blurred in both directions. */
-    HorizontalBoxBlur(temp, imOut, radius);
+    HorizontalBoxBlur(temp, temp, radius);
+    TransposeImage(temp, imOut);
 
     ImagingDelete(temp);