path: root/src/freetype/ftgrays.c

/***************************************************************************/
/*                                                                         */
/*  ftgrays.c                                                              */
/*                                                                         */
/*    A new `perfect' anti-aliasing renderer (body).                       */
/*                                                                         */
/*  Copyright 2000-2018 by                                                 */
/*  David Turner, Robert Wilhelm, and Werner Lemberg.                      */
/*                                                                         */
/*  This file is part of the FreeType project, and may only be used,       */
/*  modified, and distributed under the terms of the FreeType project      */
/*  license, LICENSE.TXT.  By continuing to use, modify, or distribute     */
/*  this file you indicate that you have read the license and              */
/*  understand and accept it fully.                                        */
/*                                                                         */
/***************************************************************************/

  /*************************************************************************/
  /*                                                                       */
  /* This file can be compiled without the rest of the FreeType engine, by */
  /* defining the STANDALONE_ macro when compiling it.  You also need to   */
  /* put the files `ftgrays.h' and `ftimage.h' into the current            */
  /* compilation directory.  Typically, you could do something like        */
  /*                                                                       */
  /* - copy `src/smooth/ftgrays.c' (this file) to your current directory   */
  /*                                                                       */
  /* - copy `include/freetype/ftimage.h' and `src/smooth/ftgrays.h' to the */
  /*   same directory                                                      */
  /*                                                                       */
  /* - compile `ftgrays' with the STANDALONE_ macro defined, as in         */
  /*                                                                       */
  /*     cc -c -DSTANDALONE_ ftgrays.c                                     */
  /*                                                                       */
  /* The renderer can be initialized with a call to                        */
  /* `ft_gray_raster.raster_new'; an anti-aliased bitmap can be generated  */
  /* with a call to `ft_gray_raster.raster_render'.                        */
  /*                                                                       */
  /* See the comments and documentation in the file `ftimage.h' for more   */
  /* details on how the raster works.                                      */
  /*                                                                       */
  /*************************************************************************/

  /*************************************************************************/
  /*                                                                       */
  /* This is a new anti-aliasing scan-converter for FreeType 2.  The       */
  /* algorithm used here is _very_ different from the one in the standard  */
  /* `ftraster' module.  Actually, `ftgrays' computes the _exact_          */
  /* coverage of the outline on each pixel cell.                           */
  /*                                                                       */
  /* It is based on ideas that I initially found in Raph Levien's          */
  /* excellent LibArt graphics library (see http://www.levien.com/libart   */
  /* for more information, though the web pages do not tell anything       */
  /* about the renderer; you'll have to dive into the source code to       */
  /* understand how it works).                                             */
  /*                                                                       */
  /* Note, however, that this is a _very_ different implementation         */
  /* compared to Raph's.  Coverage information is stored in a very         */
  /* different way, and I don't use sorted vector paths.  Also, it doesn't */
  /* use floating point values.                                            */
  /*                                                                       */
  /* This renderer has the following advantages:                           */
  /*                                                                       */
  /* - It doesn't need an intermediate bitmap.  Instead, one can supply a  */
  /*   callback function that will be called by the renderer to draw gray  */
  /*   spans on any target surface.  You can thus do direct composition on */
  /*   any kind of bitmap, provided that you give the renderer the right   */
  /*   callback.                                                           */
  /*                                                                       */
  /* - A perfect anti-aliaser, i.e., it computes the _exact_ coverage on   */
  /*   each pixel cell.                                                    */
  /*                                                                       */
  /* - It performs a single pass on the outline (the `standard' FT2        */
  /*   renderer makes two passes).                                         */
  /*                                                                       */
  /* - It can easily be modified to render to _any_ number of gray levels  */
  /*   cheaply.                                                            */
  /*                                                                       */
  /* - For small (< 20) pixel sizes, it is faster than the standard        */
  /*   renderer.                                                           */
  /*                                                                       */
  /*************************************************************************/


  /*************************************************************************/
  /*                                                                       */
  /* The macro FT_COMPONENT is used in trace mode.  It is an implicit      */
  /* parameter of the FT_TRACE() and FT_ERROR() macros, used to print/log  */
  /* messages during execution.                                            */
  /*                                                                       */
#undef  FT_COMPONENT
#define FT_COMPONENT  trace_smooth


#include "ft2build.h"
#include "ftgrays.h"
#include FT_INTERNAL_OBJECTS_H
#include FT_INTERNAL_DEBUG_H
#include FT_INTERNAL_CALC_H
#include FT_OUTLINE_H

#include "ftsmerrs.h"

#define Smooth_Err_Invalid_Mode     Smooth_Err_Cannot_Render_Glyph
#define Smooth_Err_Memory_Overflow  Smooth_Err_Out_Of_Memory
#define ErrRaster_Memory_Overflow   Smooth_Err_Out_Of_Memory


  /* as usual, for the speed hungry :-) */

#undef RAS_ARG
#undef RAS_ARG_
#undef RAS_VAR
#undef RAS_VAR_

#ifndef FT_STATIC_RASTER

#define RAS_ARG   gray_PWorker  worker
#define RAS_ARG_  gray_PWorker  worker,

#define RAS_VAR   worker
#define RAS_VAR_  worker,

#else /* FT_STATIC_RASTER */

#define RAS_ARG   void
#define RAS_ARG_  /* empty */
#define RAS_VAR   /* empty */
#define RAS_VAR_  /* empty */

#endif /* FT_STATIC_RASTER */


  /* must be at least 6 bits! */
#define PIXEL_BITS  8

#undef FLOOR
#undef CEILING
#undef TRUNC
#undef SCALED

#define ONE_PIXEL       ( 1 << PIXEL_BITS )
#define TRUNC( x )      ( (TCoord)( (x) >> PIXEL_BITS ) )
#define SUBPIXELS( x )  ( (TPos)(x) * ONE_PIXEL )
#define FLOOR( x )      ( (x) & -ONE_PIXEL )
#define CEILING( x )    ( ( (x) + ONE_PIXEL - 1 ) & -ONE_PIXEL )
#define ROUND( x )      ( ( (x) + ONE_PIXEL / 2 ) & -ONE_PIXEL )

#if PIXEL_BITS >= 6
#define UPSCALE( x )    ( (x) * ( ONE_PIXEL >> 6 ) )
#define DOWNSCALE( x )  ( (x) >> ( PIXEL_BITS - 6 ) )
#else
#define UPSCALE( x )    ( (x) >> ( 6 - PIXEL_BITS ) )
#define DOWNSCALE( x )  ( (x) * ( 64 >> PIXEL_BITS ) )
#endif


  /* Compute `dividend / divisor' and return both its quotient and     */
  /* remainder, cast to a specific type.  This macro also ensures that */
  /* the remainder is always positive.  We use the remainder to keep   */
  /* track of accumulating errors and compensate for them.             */
#define FT_DIV_MOD( type, dividend, divisor, quotient, remainder ) \
  FT_BEGIN_STMNT                                                   \
    (quotient)  = (type)( (dividend) / (divisor) );                \
    (remainder) = (type)( (dividend) % (divisor) );                \
    if ( (remainder) < 0 )                                         \
    {                                                              \
      (quotient)--;                                                \
      (remainder) += (type)(divisor);                              \
    }                                                              \
  FT_END_STMNT

#ifdef  __arm__
  /* Work around a bug specific to GCC which make the compiler fail to */
  /* optimize a division and modulo operation on the same parameters   */
  /* into a single call to `__aeabi_idivmod'.  See                     */
  /*                                                                   */
  /*  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=43721               */
#undef FT_DIV_MOD
#define FT_DIV_MOD( type, dividend, divisor, quotient, remainder ) \
  FT_BEGIN_STMNT                                                   \
    (quotient)  = (type)( (dividend) / (divisor) );                \
    (remainder) = (type)( (dividend) - (quotient) * (divisor) );   \
    if ( (remainder) < 0 )                                         \
    {                                                              \
      (quotient)--;                                                \
      (remainder) += (type)(divisor);                              \
    }                                                              \
  FT_END_STMNT
#endif /* __arm__ */


  /* These macros speed up repetitive divisions by replacing them */
  /* with multiplications and right shifts.                       */
#define FT_UDIVPREP( c, b )                                        \
  long  b ## _r = c ? (long)( FT_ULONG_MAX >> PIXEL_BITS ) / ( b ) \
                    : 0
#define FT_UDIV( a, b )                                        \
  ( ( (unsigned long)( a ) * (unsigned long)( b ## _r ) ) >>   \
    ( sizeof( long ) * FT_CHAR_BIT - PIXEL_BITS ) )


  /*************************************************************************/
  /*                                                                       */
  /*   TYPE DEFINITIONS                                                    */
  /*                                                                       */

  /* don't change the following types to FT_Int or FT_Pos, since we might */
  /* need to define them to "float" or "double" when experimenting with   */
  /* new algorithms                                                       */

  typedef long  TPos;     /* subpixel coordinate               */
  typedef int   TCoord;   /* integer scanline/pixel coordinate */
  typedef int   TArea;    /* cell areas, coordinate products   */


  typedef struct TCell_*  PCell;

  typedef struct  TCell_
  {
    TCoord  x;     /* same with gray_TWorker.ex    */
    TCoord  cover; /* same with gray_TWorker.cover */
    TArea   area;
    PCell   next;

  } TCell;

  typedef struct TPixmap_
  {
    unsigned char*  origin;  /* pixmap origin at the bottom-left */
    int             pitch;   /* pitch to go down one row */

  } TPixmap;

  /* maximum number of gray cells in the buffer */
#if FT_RENDER_POOL_SIZE > 2048
#define FT_MAX_GRAY_POOL  ( FT_RENDER_POOL_SIZE / sizeof ( TCell ) )
#else
#define FT_MAX_GRAY_POOL  ( 2048 / sizeof ( TCell ) )
#endif


#if defined( _MSC_VER )      /* Visual C++ (and Intel C++) */
  /* We disable the warning `structure was padded due to   */
  /* __declspec(align())' in order to compile cleanly with */
  /* the maximum level of warnings.                        */
#pragma warning( push )
#pragma warning( disable : 4324 )
#endif /* _MSC_VER */

  typedef struct  gray_TWorker_
  {
    ft_jmp_buf  jump_buffer;

    TCoord  ex, ey;
    TCoord  min_ex, max_ex;
    TCoord  min_ey, max_ey;

    TArea   area;
    TCoord  cover;
    int     invalid;

    PCell*      ycells;
    PCell       cells;
    FT_PtrDist  max_cells;
    FT_PtrDist  num_cells;

    TPos    x,  y;

    FT_Outline  outline;
    TPixmap     target;

    FT_Raster_Span_Func  render_span;
    void*                render_span_data;

  } gray_TWorker, *gray_PWorker;

#if defined( _MSC_VER )
#pragma warning( pop )
#endif


#ifndef FT_STATIC_RASTER
#define ras  (*worker)
#else
  static gray_TWorker  ras;
#endif


#ifdef FT_DEBUG_LEVEL_TRACE

  /* to be called while in the debugger --                                */
  /* this function causes a compiler warning since it is unused otherwise */
  static void
  gray_dump_cells( RAS_ARG )
  {
    int  y;


    for ( y = ras.min_ey; y < ras.max_ey; y++ )
    {
      PCell  cell = ras.ycells[y - ras.min_ey];


      printf( "%3d:", y );

      for ( ; cell != NULL; cell = cell->next )
        printf( " (%3d, c:%4d, a:%6d)",
                cell->x, cell->cover, cell->area );
      printf( "\n" );
    }
  }

#endif /* FT_DEBUG_LEVEL_TRACE */


  /*************************************************************************/
  /*                                                                       */
  /* Record the current cell in the table.                                 */
  /*                                                                       */
  static void
  gray_record_cell( RAS_ARG )
  {
    PCell  *pcell, cell;
    TCoord  x = ras.ex;


    pcell = &ras.ycells[ras.ey - ras.min_ey];
    for (;;)
    {
      cell = *pcell;
      if ( !cell || cell->x > x )
        break;

      if ( cell->x == x )
        goto Found;

      pcell = &cell->next;
    }

    if ( ras.num_cells >= ras.max_cells )
      ft_longjmp( ras.jump_buffer, 1 );

    /* insert new cell */
    cell        = ras.cells + ras.num_cells++;
    cell->x     = x;
    cell->area  = ras.area;
    cell->cover = ras.cover;

    cell->next  = *pcell;
    *pcell      = cell;

    return;

  Found:
    /* update old cell */
    cell->area  += ras.area;
    cell->cover += ras.cover;
  }


  /*************************************************************************/
  /*                                                                       */
  /* Set the current cell to a new position.                               */
  /*                                                                       */
  static void
  gray_set_cell( RAS_ARG_ TCoord  ex,
                          TCoord  ey )
  {
    /* Move the cell pointer to a new position.  We set the `invalid'      */
    /* flag to indicate that the cell isn't part of those we're interested */
    /* in during the render phase.  This means that:                       */
    /*                                                                     */
    /* . the new vertical position must be within min_ey..max_ey-1.        */
    /* . the new horizontal position must be strictly less than max_ex     */
    /*                                                                     */
    /* Note that if a cell is to the left of the clipping region, it is    */
    /* actually set to the (min_ex-1) horizontal position.                 */

    if ( ex < ras.min_ex )
      ex = ras.min_ex - 1;

    /* record the current one if it is valid and substantial */
    if ( !ras.invalid && ( ras.area || ras.cover ) )
      gray_record_cell( RAS_VAR );

    ras.area  = 0;
    ras.cover = 0;
    ras.ex    = ex;
    ras.ey    = ey;

    ras.invalid = ( ey >= ras.max_ey || ey < ras.min_ey ||
                    ex >= ras.max_ex );
  }


#ifndef FT_LONG64

  /*************************************************************************/
  /*                                                                       */
  /* Render a scanline as one or more cells.                               */
  /*                                                                       */
  static void
  gray_render_scanline( RAS_ARG_ TCoord  ey,
                                 TPos    x1,
                                 TCoord  y1,
                                 TPos    x2,
                                 TCoord  y2 )
  {
    TCoord  ex1, ex2, fx1, fx2, first, dy, delta, mod;
    TPos    p, dx;
    int     incr;


    ex1 = TRUNC( x1 );
    ex2 = TRUNC( x2 );

    /* trivial case.  Happens often */
    if ( y1 == y2 )
    {
      gray_set_cell( RAS_VAR_ ex2, ey );
      return;
    }

    fx1   = (TCoord)( x1 - SUBPIXELS( ex1 ) );
    fx2   = (TCoord)( x2 - SUBPIXELS( ex2 ) );

    /* everything is located in a single cell.  That is easy! */
    /*                                                        */
    if ( ex1 == ex2 )
      goto End;

    /* ok, we'll have to render a run of adjacent cells on the same */
    /* scanline...                                                  */
    /*                                                              */
    dx = x2 - x1;
    dy = y2 - y1;

    if ( dx > 0 )
    {
      p     = ( ONE_PIXEL - fx1 ) * dy;
      first = ONE_PIXEL;
      incr  = 1;
    }
    else
    {
      p     = fx1 * dy;
      first = 0;
      incr  = -1;
      dx    = -dx;
    }

    FT_DIV_MOD( TCoord, p, dx, delta, mod );

    ras.area  += (TArea)( ( fx1 + first ) * delta );
    ras.cover += delta;
    y1        += delta;
    ex1       += incr;
    gray_set_cell( RAS_VAR_ ex1, ey );

    if ( ex1 != ex2 )
    {
      TCoord  lift, rem;


      p = ONE_PIXEL * dy;
      FT_DIV_MOD( TCoord, p, dx, lift, rem );

      do
      {
        delta = lift;
        mod  += rem;
        if ( mod >= (TCoord)dx )
        {
          mod -= (TCoord)dx;
          delta++;
        }

        ras.area  += (TArea)( ONE_PIXEL * delta );
        ras.cover += delta;
        y1        += delta;
        ex1       += incr;
        gray_set_cell( RAS_VAR_ ex1, ey );
      } while ( ex1 != ex2 );
    }

    fx1 = ONE_PIXEL - first;

  End:
    dy = y2 - y1;

    ras.area  += (TArea)( ( fx1 + fx2 ) * dy );
    ras.cover += dy;
  }


  /*************************************************************************/
  /*                                                                       */
  /* Render a given line as a series of scanlines.                         */
  /*                                                                       */
  static void
  gray_render_line( RAS_ARG_ TPos  to_x,
                             TPos  to_y )
  {
    TCoord  ey1, ey2, fy1, fy2, first, delta, mod;
    TPos    p, dx, dy, x, x2;
    int     incr;


    ey1 = TRUNC( ras.y );
    ey2 = TRUNC( to_y );     /* if (ey2 >= ras.max_ey) ey2 = ras.max_ey-1; */

    /* perform vertical clipping */
    if ( ( ey1 >= ras.max_ey && ey2 >= ras.max_ey ) ||
         ( ey1 <  ras.min_ey && ey2 <  ras.min_ey ) )
      goto End;

    fy1 = (TCoord)( ras.y - SUBPIXELS( ey1 ) );
    fy2 = (TCoord)( to_y - SUBPIXELS( ey2 ) );

    /* everything is on a single scanline */
    if ( ey1 == ey2 )
    {
      gray_render_scanline( RAS_VAR_ ey1, ras.x, fy1, to_x, fy2 );
      goto End;
    }

    dx = to_x - ras.x;
    dy = to_y - ras.y;

    /* vertical line - avoid calling gray_render_scanline */
    if ( dx == 0 )
    {
      TCoord  ex     = TRUNC( ras.x );
      TCoord  two_fx = (TCoord)( ( ras.x - SUBPIXELS( ex ) ) << 1 );
      TArea   area;


      if ( dy > 0)
      {
        first = ONE_PIXEL;
        incr  = 1;
      }
      else
      {
        first = 0;
        incr  = -1;
      }

      delta      = first - fy1;
      ras.area  += (TArea)two_fx * delta;
      ras.cover += delta;
      ey1       += incr;

      gray_set_cell( RAS_VAR_ ex, ey1 );

      delta = first + first - ONE_PIXEL;
      area  = (TArea)two_fx * delta;
      while ( ey1 != ey2 )
      {
        ras.area  += area;
        ras.cover += delta;
        ey1       += incr;

        gray_set_cell( RAS_VAR_ ex, ey1 );
      }

      delta      = fy2 - ONE_PIXEL + first;
      ras.area  += (TArea)two_fx * delta;
      ras.cover += delta;

      goto End;
    }

    /* ok, we have to render several scanlines */
    if ( dy > 0)
    {
      p     = ( ONE_PIXEL - fy1 ) * dx;
      first = ONE_PIXEL;
      incr  = 1;
    }
    else
    {
      p     = fy1 * dx;
      first = 0;
      incr  = -1;
      dy    = -dy;
    }

    FT_DIV_MOD( TCoord, p, dy, delta, mod );

    x = ras.x + delta;
    gray_render_scanline( RAS_VAR_ ey1, ras.x, fy1, x, first );

    ey1 += incr;
    gray_set_cell( RAS_VAR_ TRUNC( x ), ey1 );

    if ( ey1 != ey2 )
    {
      TCoord  lift, rem;


      p    = ONE_PIXEL * dx;
      FT_DIV_MOD( TCoord, p, dy, lift, rem );

      do
      {
        delta = lift;
        mod  += rem;
        if ( mod >= (TCoord)dy )
        {
          mod -= (TCoord)dy;
          delta++;
        }

        x2 = x + delta;
        gray_render_scanline( RAS_VAR_ ey1,
                                       x, ONE_PIXEL - first,
                                       x2, first );
        x = x2;

        ey1 += incr;
        gray_set_cell( RAS_VAR_ TRUNC( x ), ey1 );
      } while ( ey1 != ey2 );
    }

    gray_render_scanline( RAS_VAR_ ey1,
                                   x, ONE_PIXEL - first,
                                   to_x, fy2 );

  End:
    ras.x       = to_x;
    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;


    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;

    ex1 = TRUNC( ras.x );
    ex2 = TRUNC( to_x );

    fx1 = ras.x - SUBPIXELS( ex1 );
    fy1 = ras.y - SUBPIXELS( ey1 );

    dx = to_x - ras.x;
    dy = to_y - ras.y;

    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 */
    {
      TPos  prod = dx * fy1 - dy * fx1;
      FT_UDIVPREP( ex1 != ex2, dx );
      FT_UDIVPREP( ey1 != ey2, 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 )
  {
    TPos  a, b;


    base[4].x = base[2].x;
    b = base[1].x;
    a = base[3].x = ( base[2].x + b ) / 2;
    b = base[1].x = ( base[0].x + b ) / 2;
    base[2].x = ( a + b ) / 2;

    base[4].y = base[2].y;
    b = base[1].y;
    a = base[3].y = ( base[2].y + b ) / 2;
    b = base[1].y = ( base[0].y + b ) / 2;
    base[2].y = ( a + b ) / 2;
  }


  static void
  gray_render_conic( RAS_ARG_ const FT_Vector*  control,
                              const FT_Vector*  to )
  {
    FT_Vector   bez_stack[16 * 2 + 1];  /* enough to accommodate bisections */
    FT_Vector*  arc = bez_stack;
    TPos        dx, dy;
    int         draw, split;


    arc[0].x = UPSCALE( to->x );
    arc[0].y = UPSCALE( to->y );
    arc[1].x = UPSCALE( control->x );
    arc[1].y = UPSCALE( control->y );
    arc[2].x = ras.x;
    arc[2].y = ras.y;

    /* short-cut the arc that crosses the current band */
    if ( ( TRUNC( arc[0].y ) >= ras.max_ey &&
           TRUNC( arc[1].y ) >= ras.max_ey &&
           TRUNC( arc[2].y ) >= ras.max_ey ) ||
         ( TRUNC( arc[0].y ) <  ras.min_ey &&
           TRUNC( arc[1].y ) <  ras.min_ey &&
           TRUNC( arc[2].y ) <  ras.min_ey ) )
    {
      ras.x = arc[0].x;
      ras.y = arc[0].y;
      return;
    }

    dx = FT_ABS( arc[2].x + arc[0].x - 2 * arc[1].x );
    dy = FT_ABS( arc[2].y + arc[0].y - 2 * arc[1].y );
    if ( dx < dy )
      dx = dy;

    /* We can calculate the number of necessary bisections because  */
    /* each bisection predictably reduces deviation exactly 4-fold. */
    /* Even 32-bit deviation would vanish after 16 bisections.      */
    draw = 1;
    while ( dx > ONE_PIXEL / 4 )
    {
      dx   >>= 2;
      draw <<= 1;
    }

    /* We use decrement counter to count the total number of segments */
    /* to draw starting from 2^level. Before each draw we split as    */
    /* many times as there are trailing zeros in the counter.         */
    do
    {
      split = 1;
      while ( ( draw & split ) == 0 )
      {
        gray_split_conic( arc );
        arc += 2;
        split <<= 1;
      }

      gray_render_line( RAS_VAR_ arc[0].x, arc[0].y );
      arc -= 2;

    } while ( --draw );
  }


  static void
  gray_split_cubic( FT_Vector*  base )
  {
    TPos  a, b, c, d;


    base[6].x = base[3].x;
    c = base[1].x;
    d = base[2].x;
    base[1].x = a = ( base[0].x + c ) / 2;
    base[5].x = b = ( base[3].x + d ) / 2;
    c = ( c + d ) / 2;
    base[2].x = a = ( a + c ) / 2;
    base[4].x = b = ( b + c ) / 2;
    base[3].x = ( a + b ) / 2;

    base[6].y = base[3].y;
    c = base[1].y;
    d = base[2].y;
    base[1].y = a = ( base[0].y + c ) / 2;
    base[5].y = b = ( base[3].y + d ) / 2;
    c = ( c + d ) / 2;
    base[2].y = a = ( a + c ) / 2;
    base[4].y = b = ( b + c ) / 2;
    base[3].y = ( a + b ) / 2;
  }


  static void
  gray_render_cubic( RAS_ARG_ const FT_Vector*  control1,
                              const FT_Vector*  control2,
                              const FT_Vector*  to )
  {
    FT_Vector   bez_stack[16 * 3 + 1];  /* enough to accommodate bisections */
    FT_Vector*  arc = bez_stack;
    TPos        dx, dy, dx_, dy_;
    TPos        dx1, dy1, dx2, dy2;
    TPos        L, s, s_limit;


    arc[0].x = UPSCALE( to->x );
    arc[0].y = UPSCALE( to->y );
    arc[1].x = UPSCALE( control2->x );
    arc[1].y = UPSCALE( control2->y );
    arc[2].x = UPSCALE( control1->x );
    arc[2].y = UPSCALE( control1->y );
    arc[3].x = ras.x;
    arc[3].y = ras.y;

    /* short-cut the arc that crosses the current band */
    if ( ( TRUNC( arc[0].y ) >= ras.max_ey &&
           TRUNC( arc[1].y ) >= ras.max_ey &&
           TRUNC( arc[2].y ) >= ras.max_ey &&
           TRUNC( arc[3].y ) >= ras.max_ey ) ||
         ( TRUNC( arc[0].y ) <  ras.min_ey &&
           TRUNC( arc[1].y ) <  ras.min_ey &&
           TRUNC( arc[2].y ) <  ras.min_ey &&
           TRUNC( arc[3].y ) <  ras.min_ey ) )
    {
      ras.x = arc[0].x;
      ras.y = arc[0].y;
      return;
    }

    for (;;)
    {
      /* Decide whether to split or draw. See `Rapid Termination          */
      /* Evaluation for Recursive Subdivision of Bezier Curves' by Thomas */
      /* F. Hain, at                                                      */
      /* http://www.cis.southalabama.edu/~hain/general/Publications/Bezier/Camera-ready%20CISST02%202.pdf */

      /* dx and dy are x and y components of the P0-P3 chord vector. */
      dx = dx_ = arc[3].x - arc[0].x;
      dy = dy_ = arc[3].y - arc[0].y;

      L = FT_HYPOT( dx_, dy_ );

      /* Avoid possible arithmetic overflow below by splitting. */
      if ( L > 32767 )
        goto Split;

      /* Max deviation may be as much as (s/L) * 3/4 (if Hain's v = 1). */
      s_limit = L * (TPos)( ONE_PIXEL / 6 );

      /* s is L * the perpendicular distance from P1 to the line P0-P3. */
      dx1 = arc[1].x - arc[0].x;
      dy1 = arc[1].y - arc[0].y;
      s = FT_ABS( SUB_LONG( MUL_LONG( dy, dx1 ), MUL_LONG( dx, dy1 ) ) );

      if ( s > s_limit )
        goto Split;

      /* s is L * the perpendicular distance from P2 to the line P0-P3. */
      dx2 = arc[2].x - arc[0].x;
      dy2 = arc[2].y - arc[0].y;
      s = FT_ABS( SUB_LONG( MUL_LONG( dy, dx2 ), MUL_LONG( dx, dy2 ) ) );

      if ( s > s_limit )
        goto Split;

      /* Split super curvy segments where the off points are so far
         from the chord that the angles P0-P1-P3 or P0-P2-P3 become
         acute as detected by appropriate dot products. */
      if ( dx1 * ( dx1 - dx ) + dy1 * ( dy1 - dy ) > 0 ||
           dx2 * ( dx2 - dx ) + dy2 * ( dy2 - dy ) > 0 )
        goto Split;

      gray_render_line( RAS_VAR_ arc[0].x, arc[0].y );

      if ( arc == bez_stack )
        return;

      arc -= 3;
      continue;

    Split:
      gray_split_cubic( arc );
      arc += 3;
    }
  }


  static int
  gray_move_to( const FT_Vector*  to,
                gray_PWorker      worker )
  {
    TPos  x, y;


    /* start to a new position */
    x = UPSCALE( to->x );
    y = UPSCALE( to->y );

    gray_set_cell( RAS_VAR_ TRUNC( x ), TRUNC( y ) );

    ras.x = x;
    ras.y = y;
    return 0;
  }


  static int
  gray_line_to( const FT_Vector*  to,
                gray_PWorker      worker )
  {
    gray_render_line( RAS_VAR_ UPSCALE( to->x ), UPSCALE( to->y ) );
    return 0;
  }


  static int
  gray_conic_to( const FT_Vector*  control,
                 const FT_Vector*  to,
                 gray_PWorker      worker )
  {
    gray_render_conic( RAS_VAR_ control, to );
    return 0;
  }


  static int
  gray_cubic_to( const FT_Vector*  control1,
                 const FT_Vector*  control2,
                 const FT_Vector*  to,
                 gray_PWorker      worker )
  {
    gray_render_cubic( RAS_VAR_ control1, control2, to );
    return 0;
  }


  static void
  gray_hline( RAS_ARG_ TCoord  x,
                       TCoord  y,
                       TArea   coverage,
                       TCoord  acount )
  {
    /* scale the coverage from 0..(ONE_PIXEL*ONE_PIXEL*2) to 0..256  */
    coverage >>= PIXEL_BITS * 2 + 1 - 8;
    if ( coverage < 0 )
      coverage = -coverage - 1;

    /* compute the line's coverage depending on the outline fill rule */
    if ( ras.outline.flags & FT_OUTLINE_EVEN_ODD_FILL )
    {
      coverage &= 511;

      if ( coverage >= 256 )
        coverage = 511 - coverage;
    }
    else
    {
      /* normal non-zero winding rule */
      if ( coverage >= 256 )
        coverage = 255;
    }

    if ( ras.render_span )  /* for FT_RASTER_FLAG_DIRECT only */
    {
      FT_Span  span;


      span.x        = (short)x;
      span.len      = (unsigned short)acount;
      span.coverage = (unsigned char)coverage;

      ras.render_span( y, 1, &span, ras.render_span_data );
    }
    else
    {
      unsigned char*  q = ras.target.origin - ras.target.pitch * y + x;
      unsigned char   c = (unsigned char)coverage;


      /* For small-spans it is faster to do it by ourselves than
       * calling `memset'.  This is mainly due to the cost of the
       * function call.
       */
      switch ( acount )
      {
      case 7: *q++ = c;
      case 6: *q++ = c;
      case 5: *q++ = c;
      case 4: *q++ = c;
      case 3: *q++ = c;
      case 2: *q++ = c;
      case 1: *q   = c;
      case 0: break;
      default:
        FT_MEM_SET( q, c, acount );
      }
    }
  }


  static void
  gray_sweep( RAS_ARG )
  {
    int  y;


    for ( y = ras.min_ey; y < ras.max_ey; y++ )
    {
      PCell   cell  = ras.ycells[y - ras.min_ey];
      TCoord  x     = ras.min_ex;
      TArea   cover = 0;
      TArea   area;


      for ( ; cell != NULL; cell = cell->next )
      {
        if ( cover != 0 && cell->x > x )
          gray_hline( RAS_VAR_ x, y, cover, cell->x - x );

        cover += (TArea)cell->cover * ( ONE_PIXEL * 2 );
        area   = cover - cell->area;

        if ( area != 0 && cell->x >= ras.min_ex )
          gray_hline( RAS_VAR_ cell->x, y, area, 1 );

        x = cell->x + 1;
      }

      if ( cover != 0 )
        gray_hline( RAS_VAR_ x, y, cover, ras.max_ex - x );
    }
  }


  FT_DEFINE_OUTLINE_FUNCS(
    func_interface,

    (FT_Outline_MoveTo_Func) gray_move_to,   /* move_to  */
    (FT_Outline_LineTo_Func) gray_line_to,   /* line_to  */
    (FT_Outline_ConicTo_Func)gray_conic_to,  /* conic_to */
    (FT_Outline_CubicTo_Func)gray_cubic_to,  /* cubic_to */

    0,                                       /* shift    */
    0                                        /* delta    */
  )


  static int
  gray_convert_glyph_inner( RAS_ARG )
  {

    volatile int  error = 0;

    if ( ft_setjmp( ras.jump_buffer ) == 0 )
    {
      error = FT_Outline_Decompose( &ras.outline, &func_interface, &ras );
      if ( !ras.invalid )
        gray_record_cell( RAS_VAR );

      FT_TRACE7(( "band [%d..%d]: %d cell%s\n",
                  ras.min_ey,
                  ras.max_ey,
                  ras.num_cells,
                  ras.num_cells == 1 ? "" : "s" ));
    }
    else
    {
      error = FT_THROW( Memory_Overflow );

      FT_TRACE7(( "band [%d..%d]: to be bisected\n",
                  ras.min_ey, ras.max_ey ));
    }

    return error;
  }


  static int
  gray_convert_glyph( RAS_ARG )
  {
    const TCoord  yMin = ras.min_ey;
    const TCoord  yMax = ras.max_ey;
    const TCoord  xMin = ras.min_ex;
    const TCoord  xMax = ras.max_ex;

    TCell    buffer[FT_MAX_GRAY_POOL];
    size_t   height = (size_t)( yMax - yMin );
    size_t   n = FT_MAX_GRAY_POOL / 8;
    TCoord   y;
    TCoord   bands[32];  /* enough to accommodate bisections */
    TCoord*  band;


    /* set up vertical bands */
    if ( height > n )
    {
      /* two divisions rounded up */
      n       = ( height + n - 1 ) / n;
      height  = ( height + n - 1 ) / n;
    }

    /* memory management */
    n = ( height * sizeof ( PCell ) + sizeof ( TCell ) - 1 ) / sizeof ( TCell );

    ras.cells     = buffer + n;
    ras.max_cells = (FT_PtrDist)( FT_MAX_GRAY_POOL - n );
    ras.ycells    = (PCell*)buffer;

    for ( y = yMin; y < yMax; )
    {
      ras.min_ey = y;
      y         += height;
      ras.max_ey = FT_MIN( y, yMax );

      band    = bands;
      band[1] = xMin;
      band[0] = xMax;

      do
      {
        TCoord  width = band[0] - band[1];
        int     error;


        FT_MEM_ZERO( ras.ycells, height * sizeof ( PCell ) );

        ras.num_cells = 0;
        ras.invalid   = 1;
        ras.min_ex    = band[1];
        ras.max_ex    = band[0];

        error = gray_convert_glyph_inner( RAS_VAR );

        if ( !error )
        {
          gray_sweep( RAS_VAR );
          band--;
          continue;
        }
        else if ( error != ErrRaster_Memory_Overflow )
          return 1;

        /* render pool overflow; we will reduce the render band by half */
        width >>= 1;

        /* this should never happen even with tiny rendering pool */
        if ( width == 0 )
        {
          FT_TRACE7(( "gray_convert_glyph: rotten glyph\n" ));
          return 1;
        }

        band++;
        band[1]  = band[0];
        band[0] += width;
      } while ( band >= bands );
    }

    return 0;
  }


  static int
  gray_raster_render( const FT_Raster_Params*  params )
  {
    const FT_Outline*  outline    = (const FT_Outline*)params->source;
    const FT_Bitmap*   target_map = params->target;
    FT_BBox            cbox, clip;

#ifndef FT_STATIC_RASTER
    gray_TWorker  worker[1];
#endif

    /* this version does not support monochrome rendering */
    if ( !( params->flags & FT_RASTER_FLAG_AA ) )
      return FT_THROW( Invalid_Mode );

    if ( !outline )
      return FT_THROW( Invalid_Outline );

    /* return immediately if the outline is empty */
    if ( outline->n_points == 0 || outline->n_contours <= 0 )
      return 0;

    if ( !outline->contours || !outline->points )
      return FT_THROW( Invalid_Outline );

    if ( outline->n_points !=
           outline->contours[outline->n_contours - 1] + 1 )
      return FT_THROW( Invalid_Outline );

    ras.outline = *outline;

    if ( params->flags & FT_RASTER_FLAG_DIRECT )
    {
      if ( !params->gray_spans )
        return 0;

      ras.render_span      = (FT_Raster_Span_Func)params->gray_spans;
      ras.render_span_data = params->user;
    }
    else
    {
      /* if direct mode is not set, we must have a target bitmap */
      if ( !target_map )
        return FT_THROW( Invalid_Argument );

      /* nothing to do */
      if ( !target_map->width || !target_map->rows )
        return 0;

      if ( !target_map->buffer )
        return FT_THROW( Invalid_Argument );

      if ( target_map->pitch < 0 )
        ras.target.origin = target_map->buffer;
      else
        ras.target.origin = target_map->buffer
              + ( target_map->rows - 1 ) * (unsigned int)target_map->pitch;

      ras.target.pitch = target_map->pitch;

      ras.render_span      = (FT_Raster_Span_Func)NULL;
      ras.render_span_data = NULL;
    }

    FT_Outline_Get_CBox( outline, &cbox );

    /* reject too large outline coordinates */
    if ( cbox.xMin < -0x1000000L || cbox.xMax > 0x1000000L ||
         cbox.yMin < -0x1000000L || cbox.yMax > 0x1000000L )
      return FT_THROW( Invalid_Outline );

    /* truncate the bounding box to integer pixels */
    cbox.xMin = cbox.xMin >> 6;
    cbox.yMin = cbox.yMin >> 6;
    cbox.xMax = ( cbox.xMax + 63 ) >> 6;
    cbox.yMax = ( cbox.yMax + 63 ) >> 6;

    /* compute clipping box */
    if ( !( params->flags & FT_RASTER_FLAG_DIRECT ) )
    {
      /* compute clip box from target pixmap */
      clip.xMin = 0;
      clip.yMin = 0;
      clip.xMax = (FT_Pos)target_map->width;
      clip.yMax = (FT_Pos)target_map->rows;
    }
    else if ( params->flags & FT_RASTER_FLAG_CLIP )
      clip = params->clip_box;
    else
    {
      clip.xMin = -32768L;
      clip.yMin = -32768L;
      clip.xMax =  32767L;
      clip.yMax =  32767L;
    }

    /* clip to target bitmap, exit if nothing to do */
    ras.min_ex = FT_MAX( cbox.xMin, clip.xMin );
    ras.min_ey = FT_MAX( cbox.yMin, clip.yMin );
    ras.max_ex = FT_MIN( cbox.xMax, clip.xMax );
    ras.max_ey = FT_MIN( cbox.yMax, clip.yMax );

    if ( ras.max_ex <= ras.min_ex || ras.max_ey <= ras.min_ey )
      return 0;

    return gray_convert_glyph( RAS_VAR );
  }


  FT_DEFINE_RASTER_FUNCS(
    ft_grays_raster,

    (FT_Raster_Render_Func)  gray_raster_render    /* raster_render   */
  )


/* END */


/* Local Variables: */
/* coding: utf-8    */
/* End:             */