... |
... |
@@ -81,57 +81,65 @@ dense_line_to( const FT_Vector* to, dense_worker* worker ) |
81
|
81
|
void
|
82
|
82
|
dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy )
|
83
|
83
|
{
|
84
|
|
- float from_x = worker->prev_x;
|
85
|
|
- float from_y = worker->prev_y;
|
86
|
|
- if ( from_y == toy )
|
87
|
|
- return;
|
88
|
84
|
|
|
85
|
+ FT26D6 fx = worker->prev_x>>2;
|
|
86
|
+ FT26D6 fy = worker->prev_y>>2;
|
|
87
|
+
|
|
88
|
+ FT26D6 from_x = fx;
|
|
89
|
+ FT26D6 from_y = fy;
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+ FT26D6 tx = tox>>2;
|
|
93
|
+ FT26D6 ty = toy>>2;
|
89
|
94
|
|
90
|
|
- from_x /= 256.0;
|
91
|
|
- from_y /= 256.0;
|
92
|
|
- float to_x = tox / 256.0;
|
93
|
|
- float to_y = toy / 256.0;
|
|
95
|
+ if ( fy == ty )
|
|
96
|
+ return;
|
94
|
97
|
|
|
98
|
+ FT26D6 to_x = tx;
|
|
99
|
+ FT26D6 to_y = ty;
|
95
|
100
|
|
96
|
|
- float dir;
|
97
|
|
- if ( from_y < to_y )
|
98
|
|
- dir = 1;
|
99
|
|
- else
|
|
101
|
+ int dir = 1;
|
|
102
|
+ if ( from_y >= to_y )
|
100
|
103
|
{
|
101
|
104
|
dir = -1;
|
102
|
|
- FT_SWAP(from_x, to_x );
|
103
|
|
- FT_SWAP(from_y, to_y );
|
|
105
|
+ FT_SWAP(from_x, to_x);
|
|
106
|
+ FT_SWAP(from_y, to_y);
|
104
|
107
|
}
|
105
|
108
|
|
106
|
109
|
// Clip to the height.
|
107
|
|
- if ( from_y >= worker->m_h || to_y <= 0 )
|
|
110
|
+ if ( from_y >= worker->m_h<<6 || to_y <= 0 )
|
108
|
111
|
return;
|
109
|
112
|
|
110
|
|
- float dxdy = ( to_x - from_x ) / (float)( to_y - from_y );
|
|
113
|
+ FT26D6 deltax,deltay;
|
|
114
|
+ deltax = to_x - from_x;
|
|
115
|
+ deltay = to_y - from_y;
|
|
116
|
+
|
111
|
117
|
if ( from_y < 0 )
|
112
|
118
|
{
|
113
|
|
- from_x -= from_y * dxdy;
|
|
119
|
+ from_x -= from_y * deltax/deltay;
|
114
|
120
|
from_y = 0;
|
115
|
121
|
}
|
116
|
|
- if ( to_y > worker->m_h )
|
|
122
|
+
|
|
123
|
+ if ( to_y > worker->m_h<<6 )
|
117
|
124
|
{
|
118
|
|
- to_x -= ( to_y - worker->m_h ) * dxdy;
|
119
|
|
- to_y = (float)worker->m_h;
|
|
125
|
+ to_x -= (( to_y - worker->m_h<<6 ) * deltax/deltay);
|
|
126
|
+ to_y = worker->m_h<<6;
|
120
|
127
|
}
|
121
|
128
|
|
122
|
|
- float x = from_x;
|
123
|
|
- int y0 = (int)from_y;
|
124
|
|
- int y_limit = (int)ceil( to_y );
|
125
|
|
- float* m_a = worker->m_a;
|
|
129
|
+ int x = from_x;
|
|
130
|
+ int y0 = from_y>>6;
|
|
131
|
+ int y_limit = (to_y + 0x3f)>>6;
|
|
132
|
+
|
|
133
|
+ FT20D12* m_a = worker->m_a;
|
126
|
134
|
|
127
|
135
|
for ( int y = y0; y < y_limit; y++ )
|
128
|
136
|
{
|
129
|
137
|
int linestart = y * worker->m_w;
|
130
|
|
- float dy = fmin( y + 1.0f, to_y ) - fmax( (float)y, from_y );
|
131
|
|
- float xnext = x + dxdy * dy;
|
132
|
|
- float d = dy * dir;
|
|
138
|
+ FT26D6 dy = FT_MIN( (y + 1)<<6, to_y ) - FT_MAX( y<<6, from_y );
|
|
139
|
+ FT26D6 xnext = x + dy * deltax/deltay;
|
|
140
|
+ FT26D6 d = dy * dir;
|
133
|
141
|
|
134
|
|
- float x0, x1;
|
|
142
|
+ FT26D6 x0, x1;
|
135
|
143
|
if ( x < xnext )
|
136
|
144
|
{
|
137
|
145
|
x0 = x;
|
... |
... |
@@ -143,40 +151,48 @@ dense_render_line( dense_worker* worker, FT_Pos tox, FT_Pos toy ) |
143
|
151
|
x1 = x;
|
144
|
152
|
}
|
145
|
153
|
|
146
|
|
- /*
|
147
|
|
- It's possible for x0 to be negative on the last scanline because of
|
148
|
|
- floating-point inaccuracy That would cause an out-of-bounds array access at
|
149
|
|
- index -1.
|
150
|
|
- */
|
151
|
|
- float x0floor = x0 <= 0.0f ? 0.0f : (float)floor( x0 );
|
152
|
154
|
|
153
|
|
- int x0i = (int)x0floor;
|
154
|
|
- float x1ceil = (float)ceil( x1 );
|
155
|
|
- int x1i = (int)x1ceil;
|
|
155
|
+ int x0i = x0>>6;
|
|
156
|
+ FT26D6 x0floor = x0i<<6;
|
|
157
|
+
|
|
158
|
+
|
|
159
|
+ int x1i = (x1+0x3f)>>6;
|
|
160
|
+ FT26D6 x1ceil = x1i <<6;
|
|
161
|
+
|
156
|
162
|
if ( x1i <= x0i + 1 )
|
157
|
163
|
{
|
158
|
|
- float xmf = 0.5f * ( x + xnext ) - x0floor;
|
159
|
|
- m_a[linestart + x0i] += d - d * xmf;
|
|
164
|
+ FT26D6 xmf = ( ( x + xnext )>>1) - x0floor;
|
|
165
|
+ m_a[linestart + x0i] += d * ((1<<6) - xmf);
|
160
|
166
|
m_a[linestart + ( x0i + 1 )] += d * xmf;
|
161
|
167
|
}
|
162
|
168
|
else
|
163
|
169
|
{
|
164
|
|
- float s = 1.0f / ( x1 - x0 );
|
165
|
|
- float x0f = x0 - x0floor;
|
166
|
|
- float a0 = 0.5f * s * ( 1.0f - x0f ) * ( 1.0f - x0f );
|
167
|
|
- float x1f = x1 - x1ceil + 1.0f;
|
168
|
|
- float am = 0.5f * s * x1f * x1f;
|
|
170
|
+
|
|
171
|
+ FT26D6 oneOverS = x1 - x0;
|
|
172
|
+ FT26D6 x0f = x0 - x0floor;
|
|
173
|
+
|
|
174
|
+
|
|
175
|
+ FT26D6 oneMinusX0f = (1<<6) - x0f;
|
|
176
|
+ FT26D6 a0 = ((oneMinusX0f * oneMinusX0f) >> 1) / oneOverS;
|
|
177
|
+ FT26D6 x1f = x1 - x1ceil + (1<<6);
|
|
178
|
+ FT26D6 am = ((x1f * x1f) >> 1) / oneOverS;
|
|
179
|
+
|
169
|
180
|
m_a[linestart + x0i] += d * a0;
|
|
181
|
+
|
170
|
182
|
if ( x1i == x0i + 2 )
|
171
|
|
- m_a[linestart + ( x0i + 1 )] += d * ( 1.0f - a0 - am );
|
|
183
|
+ m_a[linestart + ( x0i + 1 )] += d * ( (1<<6) - a0 - am );
|
172
|
184
|
else
|
173
|
185
|
{
|
174
|
|
- float a1 = s * ( 1.5f - x0f );
|
|
186
|
+ FT26D6 a1 = (((1<<6) + (1<<5) - x0f) << 6) / oneOverS;
|
175
|
187
|
m_a[linestart + ( x0i + 1 )] += d * ( a1 - a0 );
|
176
|
|
- for ( int xi = x0i + 2; xi < x1i - 1; xi++ )
|
177
|
|
- m_a[linestart + xi] += d * s;
|
178
|
|
- float a2 = a1 + ( x1i - x0i - 3 ) * s;
|
179
|
|
- m_a[linestart + ( x1i - 1 )] += d * ( 1.0f - a2 - am );
|
|
188
|
+
|
|
189
|
+ FT26D6 dTimesS = (d << 12) / oneOverS;
|
|
190
|
+
|
|
191
|
+ for ( FT26D6 xi = x0i + 2; xi < x1i - 1; xi++ )
|
|
192
|
+ m_a[linestart + xi] += dTimesS;
|
|
193
|
+
|
|
194
|
+ FT26D6 a2 = a1 + (( x1i - x0i - 3 )<<12)/oneOverS;
|
|
195
|
+ m_a[linestart + ( x1i - 1 )] += d * ( (1<<6) - a2 - am );
|
180
|
196
|
}
|
181
|
197
|
m_a[linestart + x1i] += d * am;
|
182
|
198
|
}
|
... |
... |
@@ -364,48 +380,49 @@ dense_render_glyph( dense_worker* worker, const FT_Bitmap* target ) |
364
|
380
|
FT_Error error = FT_Outline_Decompose( &( worker->outline ),
|
365
|
381
|
&dense_decompose_funcs, worker );
|
366
|
382
|
// Render into bitmap
|
367
|
|
- const float* source = worker->m_a;
|
|
383
|
+ const FT20D12* source = worker->m_a;
|
368
|
384
|
unsigned char* dest = target->buffer;
|
369
|
385
|
unsigned char* dest_end = target->buffer + worker->m_w * worker->m_h;
|
370
|
386
|
|
371
|
|
-#if FT_SSE4_1
|
372
|
|
-
|
373
|
|
- __m128 offset = _mm_setzero_ps();
|
374
|
|
- __m128i mask = _mm_set1_epi32(0x0c080400);
|
375
|
|
- __m128 sign_mask = _mm_set1_ps(-0.f);
|
376
|
|
- for (int i = 0; i < worker->m_h*worker->m_w; i += 4) {
|
377
|
|
- __m128 x = _mm_load_ps(&source[i]);
|
378
|
|
- x = _mm_add_ps(x, _mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(x), 4)));
|
379
|
|
- x = _mm_add_ps(x, _mm_shuffle_ps(_mm_setzero_ps(), x, 0x40));
|
380
|
|
- x = _mm_add_ps(x, offset);
|
381
|
|
- __m128 y = _mm_andnot_ps(sign_mask, x); // fabs(x)
|
382
|
|
- y = _mm_min_ps(y, _mm_set1_ps(1.0f));
|
383
|
|
- y = _mm_mul_ps(y, _mm_set1_ps(255.0f));
|
384
|
|
- __m128i z = _mm_cvtps_epi32(y);
|
385
|
|
- z = _mm_shuffle_epi8(z, mask);
|
386
|
|
- _mm_store_ss((float *)&dest[i], (__m128)z);
|
387
|
|
- offset = _mm_shuffle_ps(x, x, _MM_SHUFFLE(3, 3, 3, 3));
|
388
|
|
- }
|
389
|
|
-
|
390
|
|
-#else /* FT_SSE4_1 */
|
|
387
|
+//#if FT_SSE4_1
|
|
388
|
+
|
|
389
|
+ // __m128 offset = _mm_setzero_ps();
|
|
390
|
+ // __m128i mask = _mm_set1_epi32(0x0c080400);
|
|
391
|
+ // __m128 sign_mask = _mm_set1_ps(-0.f);
|
|
392
|
+ // for (int i = 0; i < worker->m_h*worker->m_w; i += 4) {
|
|
393
|
+ // __m128 x = _mm_load_ps(&source[i]);
|
|
394
|
+ // x = _mm_add_ps(x, _mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(x), 4)));
|
|
395
|
+ // x = _mm_add_ps(x, _mm_shuffle_ps(_mm_setzero_ps(), x, 0x40));
|
|
396
|
+ // x = _mm_add_ps(x, offset);
|
|
397
|
+ // __m128 y = _mm_andnot_ps(sign_mask, x); // fabs(x)
|
|
398
|
+ // y = _mm_min_ps(y, _mm_set1_ps(1.0f));
|
|
399
|
+ // y = _mm_mul_ps(y, _mm_set1_ps(255.0f));
|
|
400
|
+ // __m128i z = _mm_cvtps_epi32(y);
|
|
401
|
+ // z = _mm_shuffle_epi8(z, mask);
|
|
402
|
+ // _mm_store_ss((float *)&dest[i], (__m128)z);
|
|
403
|
+ // offset = _mm_shuffle_ps(x, x, _MM_SHUFFLE(3, 3, 3, 3));
|
|
404
|
+ // }
|
|
405
|
+
|
|
406
|
+//#else /* FT_SSE4_1 */
|
|
407
|
+
|
|
408
|
+ FT20D12 value = 0;
|
391
|
409
|
|
392
|
|
- float value = 0.0f;
|
393
|
410
|
while ( dest < dest_end )
|
394
|
411
|
{
|
395
|
412
|
value += *source++;
|
396
|
|
- if ( value > 0.0f )
|
397
|
|
- {
|
398
|
|
- int n = (int)( fabs( value ) * 255.0f + 0.5f );
|
399
|
|
- if ( n > 255 )
|
400
|
|
- n = 255;
|
|
413
|
+
|
|
414
|
+ if(value > 0){
|
|
415
|
+ int n = value >>4;
|
|
416
|
+
|
|
417
|
+ if(n>255)n=255;
|
401
|
418
|
*dest = (unsigned char)n;
|
402
|
|
- }
|
403
|
|
- else
|
|
419
|
+ }else{
|
404
|
420
|
*dest = 0;
|
|
421
|
+ }
|
405
|
422
|
dest++;
|
406
|
423
|
}
|
407
|
424
|
|
408
|
|
-#endif /* FT_SSE4_1 */
|
|
425
|
+//#endif /* FT_SSE4_1 */
|
409
|
426
|
|
410
|
427
|
free(worker->m_a);
|
411
|
428
|
return error;
|
... |
... |
@@ -444,10 +461,10 @@ dense_raster_render( FT_Raster raster, const FT_Raster_Params* params ) |
444
|
461
|
|
445
|
462
|
int size = worker->m_w * worker->m_h + 4;
|
446
|
463
|
|
447
|
|
- worker->m_a = malloc( sizeof( float ) * size );
|
|
464
|
+ worker->m_a = malloc( sizeof( FT20D12 ) * size );
|
448
|
465
|
worker->m_a_size = size;
|
449
|
466
|
|
450
|
|
- memset( worker->m_a, 0, ( sizeof( float ) * size ) );
|
|
467
|
+ memset( worker->m_a, 0, ( sizeof( FT20D12 ) * size ) );
|
451
|
468
|
/* exit if nothing to do */
|
452
|
469
|
if ( worker->m_w <= worker->m_origin_x || worker->m_h <= worker->m_origin_y )
|
453
|
470
|
{
|