| ... |
... |
@@ -23,11 +23,18 @@ |
|
23
|
23
|
# define FT_SSE4_1 0
|
|
24
|
24
|
#endif
|
|
25
|
25
|
|
|
|
26
|
+#if defined(__ARM_NEON)
|
|
|
27
|
+#define FT_NEON 1
|
|
|
28
|
+else
|
|
|
29
|
+#define FT_NEON 0
|
|
|
30
|
+#endif
|
|
26
|
31
|
|
|
27
|
32
|
#if FT_SSE4_1
|
|
28
|
33
|
|
|
29
|
34
|
#include <immintrin.h>
|
|
30
|
35
|
|
|
|
36
|
+#elif FT_NEON
|
|
|
37
|
+ #include <arm_neon.h>
|
|
31
|
38
|
#endif
|
|
32
|
39
|
|
|
33
|
40
|
#define PIXEL_BITS 8
|
| ... |
... |
@@ -91,7 +98,6 @@ dense_render_line( dense_worker* worker, FT_Pos fromx, FT_Pos fromy, FT_Pos tox, |
|
91
|
98
|
return;
|
|
92
|
99
|
}
|
|
93
|
100
|
|
|
94
|
|
-
|
|
95
|
101
|
void
|
|
96
|
102
|
dense_render_line2( dense_worker* worker, FT_PreLine pl )
|
|
97
|
103
|
{
|
| ... |
... |
@@ -212,8 +218,9 @@ dense_render_line2( dense_worker* worker, FT_PreLine pl ) |
|
212
|
218
|
if ( x1i <= x0i + 1 )
|
|
213
|
219
|
{
|
|
214
|
220
|
FT26D6 xmf = ( ( x + xnext )>>1) - x0floor;
|
|
215
|
|
- m_a[linestart + x0i] += d * ((1<<6) - xmf);
|
|
216
|
|
- m_a[linestart + ( x0i + 1 )] += d * xmf;
|
|
|
221
|
+ FT20D12 dxmf = d*xmf;
|
|
|
222
|
+ m_a[linestart + x0i] += (d * 64) - dxmf;
|
|
|
223
|
+ m_a[linestart + ( x0i + 1 )] += dxmf;
|
|
217
|
224
|
}
|
|
218
|
225
|
else
|
|
219
|
226
|
{
|
| ... |
... |
@@ -252,6 +259,8 @@ dense_render_line2( dense_worker* worker, FT_PreLine pl ) |
|
252
|
259
|
x = xnext;
|
|
253
|
260
|
}
|
|
254
|
261
|
}
|
|
|
262
|
+
|
|
|
263
|
+
|
|
255
|
264
|
}
|
|
256
|
265
|
|
|
257
|
266
|
|
| ... |
... |
@@ -456,74 +465,33 @@ dense_render_glyph( dense_worker* worker, const FT_Bitmap* target, FT_PreLine pl |
|
456
|
465
|
|
|
457
|
466
|
#if FT_SSE4_1
|
|
458
|
467
|
|
|
459
|
|
-__m128i offset = _mm_setzero_si128();
|
|
460
|
|
- __m128i mask = _mm_set1_epi32( 0x0c080400 );
|
|
|
468
|
+ __m128i offset = _mm_setzero_si128();
|
|
461
|
469
|
__m128i nzero = _mm_castps_si128(_mm_set1_ps(-0.0));
|
|
462
|
470
|
|
|
463
|
471
|
for (int i = 0; i < worker->m_h*worker->m_w; i += 4)
|
|
464
|
472
|
{
|
|
465
|
473
|
// load 4 floats from source
|
|
466
|
474
|
|
|
467
|
|
- //printf("%d\n", source[i]);
|
|
468
|
475
|
__m128i x = _mm_load_si128( (__m128i*)&source[i] );
|
|
469
|
476
|
|
|
470
|
477
|
x = _mm_add_epi32( x, _mm_slli_si128( x, 4 ) );
|
|
471
|
478
|
|
|
472
|
|
- // x = _mm_add_epi32(
|
|
473
|
|
- // x, _mm_castps_si128( _mm_shuffle_ps( _mm_setzero_ps(),
|
|
474
|
|
- // _mm_castsi128_ps( x ), 0x40 ) ) );
|
|
475
|
479
|
x = _mm_add_epi32(x, _mm_slli_si128(x,8));
|
|
476
|
480
|
|
|
477
|
481
|
// add the prefsum of previous 4 floats to all current floats
|
|
478
|
482
|
x = _mm_add_epi32( x, offset );
|
|
479
|
483
|
|
|
480
|
|
-
|
|
481
|
|
-
|
|
482
|
|
- // __m128 y = _mm_mul_ps(_mm_castsi128_ps(x), _mm_set1_ps(255.9));
|
|
483
|
|
-
|
|
484
|
|
- // y = _mm_andnot_ps(_mm_castsi128_ps(nzero), y);
|
|
485
|
|
-
|
|
486
|
|
- // __m128i z = _mm_cvttps_epi32(y);
|
|
487
|
|
- // z = _mm_packus_epi16(_mm_packs_epi32(z, nzero), nzero);
|
|
488
|
|
-
|
|
489
|
|
-
|
|
490
|
|
-
|
|
491
|
|
- // int yu = ;
|
|
492
|
|
- // *((int*)dest+i) = yu;
|
|
493
|
|
-
|
|
494
|
|
-
|
|
495
|
|
- // take absolute value
|
|
496
|
|
- //__m128i y = _mm_abs_epi32( x ); // fabs(x)
|
|
497
|
|
-
|
|
498
|
|
-
|
|
499
|
|
- // cap max value to 1
|
|
500
|
|
- //y = _mm_min_epi32( _mm_srli_epi32( y, 4 ), _mm_set1_epi32( 255 ) );
|
|
501
|
|
- //__m128i y = _mm_abs_epi32(_mm_srai_epi32( x , 4 ));
|
|
502
|
484
|
__m128i y = _mm_srli_epi32( _mm_abs_epi32( x) , 4 );
|
|
503
|
|
-
|
|
504
|
|
- // reduce to 255
|
|
505
|
|
- // y =
|
|
506
|
|
-
|
|
507
|
|
- // // shuffle
|
|
508
|
|
- //y = _mm_shuffle_epi8( y, mask );
|
|
509
|
485
|
|
|
510
|
486
|
y = _mm_packus_epi16(_mm_packs_epi32(y, nzero), nzero);
|
|
511
|
|
- //__m128i z = _mm_packus_epi16(_mm_packs_epi32(z, nzero), nzero);
|
|
512
|
487
|
|
|
513
|
488
|
// int* ptr = (int*)&dest[i];
|
|
514
|
489
|
_mm_storeu_si32(&dest[i], y);
|
|
515
|
|
- //*(int*)&dest[i] = *(int*)&y;
|
|
516
|
|
- //*(int*)&dest[i] = _mm_extract_epi32(y, 0);
|
|
517
|
490
|
|
|
518
|
|
- //_mm_store_ss( (float*)&dest[i], _mm_castsi128_ps(y) );
|
|
519
|
|
-
|
|
520
|
|
- // store the current prefix sum in offset
|
|
521
|
|
- // offset = _mm_castps_si128( _mm_shuffle_ps( _mm_castsi128_ps( x ),
|
|
522
|
|
- // _mm_castsi128_ps( x ),
|
|
523
|
|
- // _MM_SHUFFLE( 3, 3, 3, 3 ) ) );
|
|
524
|
491
|
offset = _mm_shuffle_epi32(x,_MM_SHUFFLE( 3, 3, 3, 3 ) );
|
|
525
|
|
- //offset = _mm_set1_epi32(_mm_extract_epi32(x, 3));
|
|
526
|
492
|
}
|
|
|
493
|
+#elif FT_NEON
|
|
|
494
|
+
|
|
527
|
495
|
|
|
528
|
496
|
#else /* FT_SSE4_1 */
|
|
529
|
497
|
|
| ... |
... |
@@ -588,10 +556,10 @@ dense_raster_render( FT_Raster raster, const FT_Raster_Params* params ) |
|
588
|
556
|
|
|
589
|
557
|
int size = (worker->m_w * worker->m_h + 3) & ~3;
|
|
590
|
558
|
|
|
591
|
|
- worker->m_a = malloc( sizeof( FT20D12 ) * size );
|
|
|
559
|
+ worker->m_a = calloc( size, sizeof( FT20D12 ));
|
|
592
|
560
|
worker->m_a_size = size;
|
|
593
|
561
|
|
|
594
|
|
- memset( worker->m_a, 0, ( sizeof( FT20D12 ) * size ) );
|
|
|
562
|
+ //memset( worker->m_a, 0, ( sizeof( FT20D12 ) * size ) );
|
|
595
|
563
|
/* exit if nothing to do */
|
|
596
|
564
|
if ( worker->m_w <= worker->m_origin_x || worker->m_h <= worker->m_origin_y )
|
|
597
|
565
|
{
|