ref: 11d3336af3aeee2a07b38815d2b15a5e69bc99c6
parent: f44ddfda45eaded9956f46ad1c531b61d8447eef
author: Alexei Podtelezhnikov <[email protected]>
date: Thu Sep 1 18:56:24 EDT 2016
[smooth] Simplify span rendering more. It turns out that there is significant cost associated with `FT_Span' creation and calls to `gray_render_span' because it happerns so frequently. This removes these steps from our internal use but leaves it alone for `FT_RASTER_FLAG_DIRECT" to preserve API. The speed gain is about 5%. * src/smooth/ftgrays.c (gray_render_span): Removed. The code is migrated to... (gray_hline): ... here.
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,9 +1,23 @@
+2016-09-01 Alexei Podtelezhnikov <[email protected]>
+
+ [smooth] Simplify span rendering more.
+
+ It turns out that there is significant cost associated with `FT_Span'
+ creation and calls to `gray_render_span' because it happerns so
+ frequently. This removes these steps from our internal use but leaves
+ it alone for `FT_RASTER_FLAG_DIRECT" to preserve API. The speed gain
+ is about 5%.
+
+ * src/smooth/ftgrays.c (gray_render_span): Removed. The code is
+ migrated to...
+ (gray_hline): ... here.
+
2016-08-30 Alexei Podtelezhnikov <[email protected]>
[smooth] Streamline pixmap drawing a bit more.
Zero coverage is unlikely (1 out of 256) to warrant checking. This
- gives 0.5% speed improvement in dendering simple glyphs.
+ gives 0.5% speed improvement in rendering simple glyphs.
* src/smooth/ftgrays.c (gray_hline, gray_render_span): Remove checks.
--- a/src/smooth/ftgrays.c
+++ b/src/smooth/ftgrays.c
@@ -1271,42 +1271,6 @@
static void
- gray_render_span( int y,
- int count,
- const FT_Span* spans,
- gray_PWorker worker )
- {- unsigned char* p = worker->target.origin - y * worker->target.pitch;
-
-
- for ( ; count > 0; count--, spans++ )
- {- unsigned char coverage = spans->coverage;
- unsigned char* q = p + spans->x;
-
-
- /* 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 ( spans->len )
- {- case 7: *q++ = coverage;
- case 6: *q++ = coverage;
- case 5: *q++ = coverage;
- case 4: *q++ = coverage;
- case 3: *q++ = coverage;
- case 2: *q++ = coverage;
- case 1: *q = coverage;
- case 0: break;
- default:
- FT_MEM_SET( q, coverage, spans->len );
- }
- }
- }
-
-
- static void
gray_hline( RAS_ARG_ TCoord x,
TCoord y,
TArea area,
@@ -1342,11 +1306,39 @@
coverage = 255;
}
- span.x = (short)( x + ras.min_ex );
- span.len = (unsigned short)acount;
- span.coverage = (unsigned char)coverage;
+ if ( ras.render_span ) /* for FT_RASTER_FLAG_DIRECT only */
+ {+ span.x = (short)( x + ras.min_ex );
+ span.len = (unsigned short)acount;
+ span.coverage = (unsigned char)coverage;
- ras.render_span( y + ras.min_ey, 1, &span, ras.render_span_data );
+ ras.render_span( y + ras.min_ey, 1, &span, ras.render_span_data );
+ }
+ else
+ {+ unsigned char* q = ras.target.origin -
+ ras.target.pitch * ( y + ras.min_ey ) +
+ x + ras.min_ex;
+
+
+ /* 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++ = coverage;
+ case 6: *q++ = coverage;
+ case 5: *q++ = coverage;
+ case 4: *q++ = coverage;
+ case 3: *q++ = coverage;
+ case 2: *q++ = coverage;
+ case 1: *q = coverage;
+ case 0: break;
+ default:
+ FT_MEM_SET( q, coverage, acount );
+ }
+ }
}
@@ -1960,8 +1952,8 @@
ras.target.pitch = target_map->pitch;
- ras.render_span = (FT_Raster_Span_Func)gray_render_span;
- ras.render_span_data = &ras;
+ ras.render_span = (FT_Raster_Span_Func)NULL;
+ ras.render_span_data = NULL;
}
FT_Outline_Get_CBox( outline, &cbox );