[freetype2] master 6eb6158: [smooth] Faster alternative line renderer.

[Alexei Podtelezhnikov]

branch: master
commit 6eb6158dd787dfbd7576904f99346f5b18a72ee7
Author: Alexei Podtelezhnikov <address@hidden>
Commit: Alexei Podtelezhnikov <address@hidden>

    [smooth] Faster alternative line renderer.
    
    This implementation renders the entire line segment at once without
    subdividing it into scanlines.  The main speed improvement comes from
    reducing the number of divisions to just two per line segment, which
    is a bare minimum to calculate cell coverage in a smooth rasterizer.
    Notably, the progression from cell to cell does not itself require any
    divisions at all.  The speed improvement is more noticeable at larger
    sizes.
    
    * src/smooth/ftgrays.c (gray_render_line): New implementation.
---
 ChangeLog            |   14 +++++
 src/smooth/ftgrays.c |  153 +++++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 165 insertions(+), 2 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index f167eae..f076420 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,17 @@
+2015-10-01  Alexei Podtelezhnikov  <address@hidden>
+
+       [smooth] Faster alternative line renderer.
+
+       This implementation renders the entire line segment at once without
+       subdividing it into scanlines.  The main speed improvement comes from
+       reducing the number of divisions to just two per line segment, which
+       is a bare minimum to calculate cell coverage in a smooth rasterizer.
+       Notably, the progression from cell to cell does not itself require any
+       divisions at all.  The speed improvement is more noticeable at larger
+       sizes.
+
+       * src/smooth/ftgrays.c (gray_render_line): New implementation.
+
 2015-10-06  Werner Lemberg  <address@hidden>
 
        [cff] Return correct PS names from pure CFF (#46130).
diff --git a/src/smooth/ftgrays.c b/src/smooth/ftgrays.c
index ba29445..9a43c07 100644
--- a/src/smooth/ftgrays.c
+++ b/src/smooth/ftgrays.c
@@ -135,8 +135,10 @@
 #include <string.h>
 #include <setjmp.h>
 #include <limits.h>
-#define FT_UINT_MAX  UINT_MAX
-#define FT_INT_MAX   INT_MAX
+#define FT_CHAR_BIT   CHAR_BIT
+#define FT_UINT_MAX   UINT_MAX
+#define FT_INT_MAX    INT_MAX
+#define FT_ULONG_MAX  ULONG_MAX
 
 #define ft_memset   memset
 
@@ -366,6 +368,15 @@ typedef ptrdiff_t  FT_PtrDist;
 #endif /* __arm__ */
 
 
+  /* These macros speed up repetitive divisions by replacing them */
+  /* with multiplications and right shifts.                       */ 
+#define FT_UDIVPREP( b )                                       \
+  long  b ## _r = (long)( FT_ULONG_MAX >> PIXEL_BITS ) / ( b )
+#define FT_UDIV( a, b )                                        \
+  ( ( (unsigned long)( a ) * (unsigned long)( b ## _r ) ) >>   \
+    ( sizeof( long ) * FT_CHAR_BIT - PIXEL_BITS ) )
+
+
   /*************************************************************************/
   /*                                                                       */
   /*   TYPE DEFINITIONS                                                    */
@@ -678,6 +689,7 @@ typedef ptrdiff_t  FT_PtrDist;
     gray_set_cell( RAS_VAR_ ex, ey );
   }
 
+#if 0
 
   /*************************************************************************/
   /*                                                                       */
@@ -910,6 +922,143 @@ typedef ptrdiff_t  FT_PtrDist;
     ras.y       = to_y;
   }
 
+#else
+
+  /*************************************************************************/
+  /*                                                                       */
+  /* Render a straight line across multiple cells in any direction.        */
+  /*                                                                       */
+  static void
+  gray_render_line( RAS_ARG_ TPos  to_x,
+                             TPos  to_y )
+  {
+    TPos    dx, dy, fx1, fy1, fx2, fy2;
+    TCoord  ex1, ex2, ey1, ey2;
+
+
+    ex1 = TRUNC( ras.x );
+    ex2 = TRUNC( to_x );
+    ey1 = TRUNC( ras.y );
+    ey2 = TRUNC( to_y );
+
+    /* perform vertical clipping */
+    if ( ( ey1 >= ras.max_ey && ey2 >= ras.max_ey ) ||
+         ( ey1 <  ras.min_ey && ey2 <  ras.min_ey ) )
+      goto End;
+
+    dx = to_x - ras.x;
+    dy = to_y - ras.y;
+
+    fx1 = ras.x - SUBPIXELS( ex1 );
+    fy1 = ras.y - SUBPIXELS( ey1 );
+
+    if ( ex1 == ex2 && ey1 == ey2 )       /* inside one cell */
+      ;
+    else if ( dy == 0 ) /* ex1 != ex2 */  /* any horizontal line */
+    {
+      ex1 = ex2;
+      gray_set_cell( RAS_VAR_ ex1, ey1 );
+    }
+    else if ( dx == 0 )
+    {
+      if ( dy > 0 )                       /* vertical line up */
+        do
+        {
+          fy2 = ONE_PIXEL;
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * fx1 * 2;
+          fy1 = 0;
+          ey1++;
+          gray_set_cell( RAS_VAR_ ex1, ey1 );
+        } while ( ey1 != ey2 );
+      else                                /* vertical line down */
+        do
+        {
+          fy2 = 0;
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * fx1 * 2;
+          fy1 = ONE_PIXEL;
+          ey1--;
+          gray_set_cell( RAS_VAR_ ex1, ey1 );
+        } while ( ey1 != ey2 );
+    }
+    else                                  /* any other line */
+    {
+      TArea  prod = dx * fy1 - dy * fx1;
+      FT_UDIVPREP( dx );
+      FT_UDIVPREP( dy );
+
+
+      /* The fundamental value `prod' determines which side and the  */
+      /* exact coordinate where the line exits current cell.  It is  */
+      /* also easily updated when moving from one cell to the next.  */
+      do
+      {
+        if      ( prod                                   <= 0 &&
+                  prod - dx * ONE_PIXEL                  >  0 ) /* left */
+        {
+          fx2 = 0;
+          fy2 = (TPos)FT_UDIV( -prod, -dx );
+          prod -= dy * ONE_PIXEL;
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * ( fx1 + fx2 );
+          fx1 = ONE_PIXEL;
+          fy1 = fy2;
+          ex1--;
+        }
+        else if ( prod - dx * ONE_PIXEL                  <= 0 &&
+                  prod - dx * ONE_PIXEL + dy * ONE_PIXEL >  0 ) /* up */
+        {
+          prod -= dx * ONE_PIXEL;
+          fx2 = (TPos)FT_UDIV( -prod, dy );
+          fy2 = ONE_PIXEL;
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * ( fx1 + fx2 );
+          fx1 = fx2;
+          fy1 = 0;
+          ey1++;
+        }
+        else if ( prod - dx * ONE_PIXEL + dy * ONE_PIXEL <= 0 &&
+                  prod                  + dy * ONE_PIXEL >= 0 ) /* right */
+        {
+          prod += dy * ONE_PIXEL;
+          fx2 = ONE_PIXEL;
+          fy2 = (TPos)FT_UDIV( prod, dx );
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * ( fx1 + fx2 );
+          fx1 = 0;
+          fy1 = fy2;
+          ex1++;
+        }
+        else /* ( prod                  + dy * ONE_PIXEL <  0 &&
+                  prod                                   >  0 )    down */
+        {
+          fx2 = (TPos)FT_UDIV( prod, -dy );
+          fy2 = 0;
+          prod += dx * ONE_PIXEL;
+          ras.cover += ( fy2 - fy1 );
+          ras.area  += ( fy2 - fy1 ) * ( fx1 + fx2 );
+          fx1 = fx2;
+          fy1 = ONE_PIXEL;
+          ey1--;
+        }
+
+        gray_set_cell( RAS_VAR_ ex1, ey1 );
+      } while ( ex1 != ex2 || ey1 != ey2 );
+    }
+
+    fx2 = to_x - SUBPIXELS( ex2 );
+    fy2 = to_y - SUBPIXELS( ey2 );
+
+    ras.cover += ( fy2 - fy1 );
+    ras.area  += ( fy2 - fy1 ) * ( fx1 + fx2 );
+
+  End:
+    ras.x       = to_x;
+    ras.y       = to_y;
+  }
+
+#endif
 
   static void
   gray_split_conic( FT_Vector*  base )