Chiebot-Mirror
/
freetype
mirror of https://github.com/freetype/freetype.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

[smooth] Faster bitmap sweeping.

Browse Source 
Selecting the fill rule or checking the direct mode each time we call
`gray_hline' is sub-optimal.  This effectively splits the direct mode
into a separate code path while inlining `gray_hline' and saving 5-7%
of rendering time.

* src/smooth/ftgrays.c (gray_hline): Eliminated in favor of...
(FT_FILL_RULE, FT_GRAY_SET): ... these new macros...
(gray_sweep): ... inlined here.
(gray_sweep_direct): New function that handles the direct span buffer.
(gray_TWorker): Remove the span buffer.
(gray_raster_render, gray_convert_glyph): Updated.
wl/meson-logging
Alexei Podtelezhnikov 4 years ago
parent b070264521
commit 8f43d324d1
2 changed files with 132 additions and 79 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 16
      ChangeLog
  2. 195
      src/smooth/ftgrays.c

16
ChangeLog
Unescape View File

@ -1,3 +1,19 @@
2021-05-07 Alexei Podtelezhnikov <apodtele@gmail.com>
[smooth] Faster bitmap sweeping.
Selecting the fill rule or checking the direct mode each time we call
`gray_hline' is sub-optimal. This effectively splits the direct mode
into a separate code path while inlining `gray_hline' and saving 5-7%
of rendering time.
* src/smooth/ftgrays.c (gray_hline): Eliminated in favor of...
(FT_FILL_RULE, FT_GRAY_SET): ... these new macros...
(gray_sweep): ... inlined here.
(gray_sweep_direct): New function that handles the direct span buffer.
(gray_TWorker): Remove the span buffer.
(gray_raster_render, gray_convert_glyph): Updated.
2021-05-10 Alexei Podtelezhnikov <apodtele@gmail.com>
* src/smooth/ftgrays.c (gray_hline): Simplify even-odd computations.

195
src/smooth/ftgrays.c
Unescape View File

@ -392,6 +392,42 @@ typedef ptrdiff_t FT_PtrDist;
( sizeof( long ) * FT_CHAR_BIT - PIXEL_BITS ) )
/* Scale area and apply fill rule to calculate the coverage byte. */
/* The top fill bit is used for the non-zero rule. The eighth */
/* fill bit is used for the even-odd rule. The higher coverage */
/* bytes are either clamped for the non-zero-rule or discarded */
/* later for the even-odd rule. */
#define FT_FILL_RULE( coverage, area, fill ) \
FT_BEGIN_STMNT \
coverage = (int)( area >> ( PIXEL_BITS * 2 + 1 - 8 ) ); \
if ( coverage & fill ) \
coverage = ~coverage; \
if ( coverage > 255 && fill & INT_MIN ) \
coverage = 255; \
FT_END_STMNT
/* It is faster to write small spans byte-by-byte than calling */
/* `memset'. This is mainly due to the cost of the function call. */
#define FT_GRAY_SET( d, s, count ) \
FT_BEGIN_STMNT \
unsigned char* q = d; \
unsigned char c = (unsigned char)s; \
switch ( count ) \
{ \
case 7: *q++ = c; /* fall through */ \
case 6: *q++ = c; /* fall through */ \
case 5: *q++ = c; /* fall through */ \
case 4: *q++ = c; /* fall through */ \
case 3: *q++ = c; /* fall through */ \
case 2: *q++ = c; /* fall through */ \
case 1: *q = c; /* fall through */ \
case 0: break; \
default: FT_MEM_SET( d, s, count ); \
} \
FT_END_STMNT
/**************************************************************************
*
* TYPE DEFINITIONS
@ -463,8 +499,6 @@ typedef ptrdiff_t FT_PtrDist;
FT_Raster_Span_Func render_span;
void* render_span_data;
FT_Span spans[FT_MAX_GRAY_SPANS];
int num_spans;
} gray_TWorker, *gray_PWorker;
@ -1171,94 +1205,62 @@ typedef ptrdiff_t FT_PtrDist;
static void
gray_hline( RAS_ARG_ TCoord x,
TCoord y,
TArea coverage,
TCoord acount )
gray_sweep( RAS_ARG )
{
/* scale the coverage from 0..(ONE_PIXEL*ONE_PIXEL*2) to 0..256 */
coverage >>= PIXEL_BITS * 2 + 1 - 8;
int fill = ras.outline.flags & FT_OUTLINE_EVEN_ODD_FILL ? 0x100 : INT_MIN;
int coverage;
int y;
/* compute the line's coverage depending on the outline fill rule */
if ( ras.outline.flags & FT_OUTLINE_EVEN_ODD_FILL )
{
if ( coverage & 256 ) /* odd bit */
coverage = ~coverage;
/* higher bits discarded below */
}
else /* default non-zero winding rule */
for ( y = ras.min_ey; y < ras.max_ey; y++ )
{
if ( coverage < 0 )
coverage = ~coverage; /* the same as -coverage - 1 */
PCell cell = ras.ycells[y - ras.min_ey];
TCoord x = ras.min_ex;
TArea cover = 0;
TArea area;
if ( coverage >= 256 )
coverage = 255;
}
unsigned char* line = ras.target.origin - ras.target.pitch * y;
if ( ras.num_spans >= 0 ) /* for FT_RASTER_FLAG_DIRECT only */
{
FT_Span* span = ras.spans + ras.num_spans++;
for ( ; cell != NULL; cell = cell->next )
{
if ( cover != 0 && cell->x > x )
{
FT_FILL_RULE( coverage, cover, fill );
FT_GRAY_SET( line + x, coverage, cell->x - x );
}
span->x = (short)x;
span->len = (unsigned short)acount;
span->coverage = (unsigned char)coverage;
cover += (TArea)cell->cover * ( ONE_PIXEL * 2 );
area = cover - cell->area;
if ( ras.num_spans == FT_MAX_GRAY_SPANS )
{
/* flush the span buffer and reset the count */
ras.render_span( y, ras.num_spans, ras.spans, ras.render_span_data );
ras.num_spans = 0;
}
}
else
{
unsigned char* q = ras.target.origin - ras.target.pitch * y + x;
unsigned char c = (unsigned char)coverage;
if ( area != 0 && cell->x >= ras.min_ex )
{
FT_FILL_RULE( coverage, area, fill );
line[cell->x] = (unsigned char)coverage;
}
x = cell->x + 1;
}
/* 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 )
if ( cover != 0 ) /* only if cropped */
{
case 7:
*q++ = c;
/* fall through */
case 6:
*q++ = c;
/* fall through */
case 5:
*q++ = c;
/* fall through */
case 4:
*q++ = c;
/* fall through */
case 3:
*q++ = c;
/* fall through */
case 2:
*q++ = c;
/* fall through */
case 1:
*q = c;
/* fall through */
case 0:
break;
default:
FT_MEM_SET( q, c, acount );
FT_FILL_RULE( coverage, cover, fill );
FT_GRAY_SET( line + x, coverage, ras.max_ex - x );
}
}
}
static void
gray_sweep( RAS_ARG )
gray_sweep_direct( RAS_ARG )
{
int fill = ras.outline.flags & FT_OUTLINE_EVEN_ODD_FILL ? 0x100 : INT_MIN;
int coverage;
int y;
FT_Span span[FT_MAX_GRAY_SPANS];
int n = 0;
for ( y = ras.min_ey; y < ras.max_ey; y++ )
{
@ -1271,25 +1273,59 @@ typedef ptrdiff_t FT_PtrDist;
for ( ; cell != NULL; cell = cell->next )
{
if ( cover != 0 && cell->x > x )
gray_hline( RAS_VAR_ x, y, cover, cell->x - x );
{
FT_FILL_RULE( coverage, cover, fill );
span[n].coverage = (unsigned char)coverage;
span[n].x = (short)x;
span[n].len = (unsigned short)( cell->x - x );
if ( ++n == FT_MAX_GRAY_SPANS )
{
/* flush the span buffer and reset the count */
ras.render_span( y, n, span, ras.render_span_data );
n = 0;
}
}
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 );
{
FT_FILL_RULE( coverage, area, fill );
span[n].coverage = (unsigned char)coverage;
span[n].x = (short)cell->x;
span[n].len = 1;
if ( ++n == FT_MAX_GRAY_SPANS )
{
/* flush the span buffer and reset the count */
ras.render_span( y, n, span, ras.render_span_data );
n = 0;
}
}
x = cell->x + 1;
}
if ( cover != 0 )
gray_hline( RAS_VAR_ x, y, cover, ras.max_ex - x );
if ( cover != 0 ) /* only if cropped */
{
FT_FILL_RULE( coverage, cover, fill );
span[n].coverage = (unsigned char)coverage;
span[n].x = (short)x;
span[n].len = (unsigned short)( ras.max_ex - x );
++n;
}
if ( ras.num_spans > 0 ) /* for FT_RASTER_FLAG_DIRECT only */
if ( n )
{
/* flush the span buffer and reset the count */
ras.render_span( y, ras.num_spans, ras.spans, ras.render_span_data );
ras.num_spans = 0;
ras.render_span( y, n, span, ras.render_span_data );
n = 0;
}
}
}
@ -1688,7 +1724,10 @@ typedef ptrdiff_t FT_PtrDist;
if ( !error )
{
gray_sweep( RAS_VAR );
if ( ras.render_span ) /* for FT_RASTER_FLAG_DIRECT only */
gray_sweep_direct( RAS_VAR );
else
gray_sweep( RAS_VAR );
band--;
continue;
}
@ -1757,7 +1796,6 @@ typedef ptrdiff_t FT_PtrDist;
ras.render_span = (FT_Raster_Span_Func)params->gray_spans;
ras.render_span_data = params->user;
ras.num_spans = 0;
ras.min_ex = params->clip_box.xMin;
ras.min_ey = params->clip_box.yMin;
@ -1787,7 +1825,6 @@ typedef ptrdiff_t FT_PtrDist;
ras.render_span = (FT_Raster_Span_Func)NULL;
ras.render_span_data = NULL;
ras.num_spans = -1; /* invalid */
ras.min_ex = 0;
ras.min_ey = 0;

Write Preview
Loading…
Cancel
Save