minhngoc25a
/
freetype2
mirror of git://git.savannah.gnu.org/freetype/freetype2.git
2
1
Fork 2
Code Issues 4 Releases Wiki Activity

* src/smooth/ftgrays.c (gray_render_conic) [SSE2]: Improve flow.

This commit is contained in:
Alexei Podtelezhnikov 2023-09-28 22:56:15 -04:00
parent e0e5b838e5
commit 7b308a29dd
1 changed files with 25 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

50
src/smooth/ftgrays.c
View File

@ -1095,16 +1095,17 @@ typedef ptrdiff_t FT_PtrDist;
return; return;
} }
/* We can calculate the number of necessary bisections because */ /* We can calculate the number of necessary segments because */
/* each bisection predictably reduces deviation exactly 4-fold. */ /* each bisection predictably reduces deviation exactly 4-fold. */
/* Even 32-bit deviation would vanish after 16 bisections. */ /* Even 32-bit deviation would vanish after 16 bisections. */
shift = 0; shift = 16;
do do
{ {
dx >>= 2; dx >>= 2;
shift += 1; shift--;
} while ( dx > ONE_PIXEL / 4 ); } while ( dx > ONE_PIXEL / 4 );
count = 0x10000U >> shift;
/* /*
* The (P0,P1,P2) arc equation, for t in [0,1] range: * The (P0,P1,P2) arc equation, for t in [0,1] range:
@ -1151,9 +1152,8 @@ typedef ptrdiff_t FT_PtrDist;
*/ */
#if FT_SSE2 #if FT_SSE2
/* Experience shows that for small shift values, */ /* Experience shows that for small counts, SSE2 is actually slower. */
/* SSE2 is actually slower. */ if ( count > 4 )
if ( shift > 2 )
{ {
union union
{ {
@ -1169,9 +1169,7 @@ typedef ptrdiff_t FT_PtrDist;
} v; } v;
__m128i a, b; __m128i p, q, r;
__m128i r, q, q2;
__m128i p;
u.i.ax = ax; u.i.ax = ax;
@ -1179,14 +1177,13 @@ typedef ptrdiff_t FT_PtrDist;
u.i.bx = bx; u.i.bx = bx;
u.i.by = by; u.i.by = by;
a = _mm_load_si128( &u.vec.a ); q = _mm_load_si128( &u.vec.b );
b = _mm_load_si128( &u.vec.b ); r = _mm_load_si128( &u.vec.a );
r = _mm_slli_epi64( a, 33 - 2 * shift ); q = _mm_slli_epi64( q, shift + 17);
q = _mm_slli_epi64( b, 33 - shift ); r = _mm_slli_epi64( r, shift + shift );
q2 = _mm_slli_epi64( a, 32 - 2 * shift ); q = _mm_add_epi64( q, r );
r = _mm_add_epi64( r, r );
q = _mm_add_epi64( q2, q );
v.i.px_lo = 0; v.i.px_lo = 0;
v.i.px_hi = p0.x; v.i.px_hi = p0.x;
@ -1195,7 +1192,7 @@ typedef ptrdiff_t FT_PtrDist;
p = _mm_load_si128( &v.vec ); p = _mm_load_si128( &v.vec );
for ( count = 1U << shift; count > 0; count-- ) do
{ {
p = _mm_add_epi64( p, q ); p = _mm_add_epi64( p, q );
q = _mm_add_epi64( q, r ); q = _mm_add_epi64( q, r );
@ -1203,22 +1200,25 @@ typedef ptrdiff_t FT_PtrDist;
_mm_store_si128( &v.vec, p ); _mm_store_si128( &v.vec, p );
gray_render_line( RAS_VAR_ v.i.px_hi, v.i.py_hi ); gray_render_line( RAS_VAR_ v.i.px_hi, v.i.py_hi );
} } while ( --count );
return; return;
} }
#endif /* FT_SSE2 */ #endif /* FT_SSE2 */
rx = LEFT_SHIFT( ax, 33 - 2 * shift ); rx = LEFT_SHIFT( ax, shift + shift );
ry = LEFT_SHIFT( ay, 33 - 2 * shift ); ry = LEFT_SHIFT( ay, shift + shift );
qx = LEFT_SHIFT( bx, 33 - shift ) + LEFT_SHIFT( ax, 32 - 2 * shift ); qx = LEFT_SHIFT( bx, shift + 17 ) + rx;
qy = LEFT_SHIFT( by, 33 - shift ) + LEFT_SHIFT( ay, 32 - 2 * shift ); qy = LEFT_SHIFT( by, shift + 17 ) + ry;
rx *= 2;
ry *= 2;
px = LEFT_SHIFT( p0.x, 32 ); px = LEFT_SHIFT( p0.x, 32 );
py = LEFT_SHIFT( p0.y, 32 ); py = LEFT_SHIFT( p0.y, 32 );
for ( count = 1U << shift; count > 0; count-- ) do
{ {
px += qx; px += qx;
py += qy; py += qy;
@ -1227,7 +1227,7 @@ typedef ptrdiff_t FT_PtrDist;
gray_render_line( RAS_VAR_ (FT_Pos)( px >> 32 ), gray_render_line( RAS_VAR_ (FT_Pos)( px >> 32 ),
(FT_Pos)( py >> 32 ) ); (FT_Pos)( py >> 32 ) );
} } while ( --count );
} }
#else /* !BEZIER_USE_DDA */ #else /* !BEZIER_USE_DDA */