[Git][freetype/freetype][gsoc-anurag-2022-final] [dense] Add optimization for division

[Anurag Thakur (@AdbhutDev)]

Anurag Thakur pushed to branch gsoc-anurag-2022-final at FreeType / FreeType

Commits:

• 6b161ceb
  by Anurag Thakur at 2022-11-19T13:22:15+05:30

  [dense] Add optimization for division
  
  * src/dense/ftdense.c: FT_UDIV, FT_UDIVPREP macros taken from smooth
  rasterizer, help optimize fixed-point division
  

1 changed file:

• src/dense/ftdense.c

Changes:

src/dense/ftdense.c

---

...	...	@@ -41,6 +41,11 @@
41	41	#define FT_MAX( a, b ) ( (a) > (b) ? (a) : (b) )
42	42	#define FT_ABS( a ) ( (a) < 0 ? -(a) : (a) )
43	43
	44	+// TODO: Fix types
	45	+#define FT_UDIVPREP( c, b ) \
	46	+ FT26D6 b ## _r = c ? (FT26D6)0xFFFFFFFF / ( b ) : 0
	47	+#define FT_UDIV( a, b ) \
	48	+ (FT26D6)( ( (FT26D6)( a ) * (FT26D6)( b ## _r ) ) >> 32 )
44	49
45	50	typedef struct dense_TRaster_
46	51	{
...	...	@@ -114,6 +119,10 @@ dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy )
114	119	deltax = to_x - from_x;
115	120	deltay = to_y - from_y;
116	121
	122	+ FT_UDIVPREP(from_x != to_x, deltax);
	123	+
	124	+ FT_UDIVPREP(from_y != to_y, deltay);
	125	+
117	126	if ( from_y < 0 )
118	127	{
119	128	from_x -= from_y * deltax/deltay;
...	...	@@ -164,7 +173,7 @@ dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy )
164	173	{
165	174	int linestart = y * worker->m_w;
166	175	FT26D6 dy = FT_MIN( (y + 1)<<6, to_y ) - FT_MAX( y<<6, from_y );
167		- FT26D6 xnext = x + dy * deltax/deltay;
	176	+ FT26D6 xnext = x + FT_UDIV((dy*deltax), deltay);
168	177	FT26D6 d = dy * dir;
169	178
170	179	FT26D6 x0, x1;
...	...	@@ -197,13 +206,16 @@ dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy )
197	206	{
198	207
199	208	FT26D6 oneOverS = x1 - x0;
	209	+
	210	+ FT_UDIVPREP(x1 != x0, oneOverS);
	211	+
200	212	FT26D6 x0f = x0 - x0floor;
201	213
202	214
203	215	FT26D6 oneMinusX0f = (1<<6) - x0f;
204		- FT26D6 a0 = ((oneMinusX0f * oneMinusX0f) >> 1) / oneOverS;
	216	+ FT26D6 a0 = FT_UDIV(((oneMinusX0f * oneMinusX0f) >> 1), oneOverS);
205	217	FT26D6 x1f = x1 - x1ceil + (1<<6);
206		- FT26D6 am = ((x1f * x1f) >> 1) / oneOverS;
	218	+ FT26D6 am = FT_UDIV(((x1f * x1f) >> 1) , oneOverS);
207	219
208	220	m_a[linestart + x0i] += d * a0;
209	221
...	...	@@ -211,15 +223,15 @@ dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy )
211	223	m_a[linestart + ( x0i + 1 )] += d * ( (1<<6) - a0 - am );
212	224	else
213	225	{
214		- FT26D6 a1 = (((1<<6) + (1<<5) - x0f) << 6) / oneOverS;
	226	+ FT26D6 a1 = FT_UDIV((((1<<6) + (1<<5) - x0f) << 6) , oneOverS);
215	227	m_a[linestart + ( x0i + 1 )] += d * ( a1 - a0 );
216	228
217		- FT26D6 dTimesS = (d << 12) / oneOverS;
	229	+ FT26D6 dTimesS = FT_UDIV((d << 12) , oneOverS);
218	230
219	231	for ( FT26D6 xi = x0i + 2; xi < x1i - 1; xi++ )
220	232	m_a[linestart + xi] += dTimesS;
221	233
222		- FT26D6 a2 = a1 + (( x1i - x0i - 3 )<<12)/oneOverS;
	234	+ FT26D6 a2 = a1 + FT_UDIV((( x1i - x0i - 3 )<<12),oneOverS);
223	235	m_a[linestart + ( x1i - 1 )] += d * ( (1<<6) - a2 - am );
224	236	}
225	237	m_a[linestart + x1i] += d * am;

—
View it on GitLab.
You're receiving this email because of your account on gitlab.freedesktop.org. Manage all notifications · Help