2 * Copyright (C) 2003-2006 Gabest
3 * http://www.gabest.org
5 * This Program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2, or (at your option)
10 * This Program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with GNU Make; see the file COPYING. If not, write to
17 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
18 * http://www.gnu.org/copyleft/gpl.html
27 #include "Rasterizer.h"
28 #include "SeparableFilter.h"
29 #include "xy_logger.h"
30 #include <boost/flyweight/key_value.hpp>
31 #include "xy_bitmap.h"
33 #ifndef _MAX /* avoid collision with common (nonconforming) macros */
34 #define _MAX (std::max)
35 #define _MIN (std::min)
36 #define _IMPL_MAX std::max
37 #define _IMPL_MIN std::min
39 #define _IMPL_MAX _MAX
40 #define _IMPL_MIN _MIN
44 //NOTE: signed or unsigned affects the result seriously
45 #define COMBINE_AYUV(a, y, u, v) ((((((((int)(a))<<8)|y)<<8)|u)<<8)|v)
47 #define SPLIT_AYUV(color, a, y, u, v) do { \
49 *(u)=((color)>>8) &0xff; \
50 *(y)=((color)>>16)&0xff;\
51 *(a)=((color)>>24)&0xff;\
57 static const int VOLUME_BITS
= 22;//should not exceed 32-8, and better not exceed 31-8
59 ass_synth_priv(const double sigma
);
60 ass_synth_priv(const ass_synth_priv
& priv
);
63 int generate_tables(double sigma
);
74 struct ass_synth_priv_key
76 const double& operator()(const ass_synth_priv
& x
)const
85 ass_tmp_buf(size_t size
);
86 ass_tmp_buf(const ass_tmp_buf
& buf
);
92 struct ass_tmp_buf_get_size
94 const size_t& operator()(const ass_tmp_buf
& buf
)const
100 static const unsigned int maxcolor
= 255;
101 static const unsigned base
= 256;
103 ass_synth_priv::ass_synth_priv(const double sigma
)
112 generate_tables(sigma
);
115 ass_synth_priv::ass_synth_priv(const ass_synth_priv
& priv
):g_r(priv
.g_r
),g_w(priv
.g_w
),sigma(priv
.sigma
)
117 if (this->g_w
> 0 && this != &priv
) {
118 this->g
= (unsigned*)realloc(this->g
, this->g_w
* sizeof(unsigned));
119 this->gt2
= (unsigned*)realloc(this->gt2
, 256 * this->g_w
* sizeof(unsigned));
120 //if (this->g == null || this->gt2 == null) {
123 memcpy(g
, priv
.g
, this->g_w
* sizeof(unsigned));
124 memcpy(gt2
, priv
.gt2
, 256 * this->g_w
* sizeof(unsigned));
128 ass_synth_priv::~ass_synth_priv()
134 int ass_synth_priv::generate_tables(double sigma
)
136 const int TARGET_VOLUME
= 1<<VOLUME_BITS
;
137 const int MAX_VOLUME_ERROR
= VOLUME_BITS
>=22 ? 16 : 1;
139 double a
= -1 / (sigma
* sigma
* 2);
140 double exp_a
= exp(a
);
142 double volume_factor
= 0;
143 double volume_start
= 0, volume_end
= 0;
146 if (this->sigma
== sigma
)
151 this->g_w
= (int)ceil(sigma
*3) | 1;
152 this->g_r
= this->g_w
/ 2;
155 this->g
= (unsigned*)realloc(this->g
, this->g_w
* sizeof(unsigned));
156 this->gt2
= (unsigned*)realloc(this->gt2
, 256 * this->g_w
* sizeof(unsigned));
157 if (this->g
== NULL
|| this->gt2
== NULL
) {
166 double exp_1
= exp_a
;
167 double exp_2
= exp_1
* exp_1
;
168 volume_start
+= exp_0
;
169 for(int i
=0;i
<this->g_r
;++i
)
173 volume_start
+= exp_0
;
174 volume_start
+= exp_0
;
177 // for (i = 0; i < this->g_w; ++i) {
178 // volume_start += exp(a * (i - this->g_r) * (i - this->g_r));
181 volume_end
= (TARGET_VOLUME
+g_w
)/volume_start
;
182 volume_start
= (TARGET_VOLUME
-g_w
)/volume_start
;
185 while( volume_start
+0.000001<volume_end
)
187 volume_factor
= (volume_start
+volume_end
)*0.5;
190 exp_0
= volume_factor
;
192 exp_2
= exp_1
* exp_1
;
194 volume
= static_cast<int>(exp_0
+.5);
195 this->g
[this->g_r
] = volume
;
197 unsigned* p_left
= this->g
+this->g_r
-1;
198 unsigned* p_right
= this->g
+this->g_r
+1;
199 for(int i
=0; i
<this->g_r
;++i
,p_left
--,p_right
++)
203 *p_left
= static_cast<int>(exp_0
+.5);
205 volume
+= (*p_left
<<1);
208 // for (i = 0; i < this->g_w; ++i) {
209 // this->g[i] = (unsigned) ( exp(a * (i - this->g_r) * (i - this->g_r))* volume_factor + .5 );
210 // volume += this->g[i];
213 // volume don't have to be equal to TARGET_VOLUME,
214 // even if volume=TARGET_VOLUME+MAX_VOLUME_ERROR,
215 // max error introducing in later blur operation,
216 // which is (dot_product(g_w, pixel))/TARGET_VOLUME with pixel<256,
217 // would not exceed (MAX_VOLUME_ERROR*256)/TARGET_VOLUME,
218 // as long as MAX_VOLUME_ERROR/TARGET_VOLUME is small enough, error introduced would be kept in safe range
220 // NOTE: when it comes to rounding, no matter how small the error is,
221 // it may result a different rounding output
222 if( volume
>=TARGET_VOLUME
&& volume
< (TARGET_VOLUME
+MAX_VOLUME_ERROR
) )
224 else if(volume
< TARGET_VOLUME
)
226 volume_start
= volume_factor
;
228 else if(volume
>= TARGET_VOLUME
+MAX_VOLUME_ERROR
)
230 volume_end
= volume_factor
;
235 volume_factor
= volume_end
;
237 exp_0
= volume_factor
;
239 exp_2
= exp_1
* exp_1
;
241 volume
= static_cast<int>(exp_0
+.5);
242 this->g
[this->g_r
] = volume
;
244 unsigned* p_left
= this->g
+this->g_r
-1;
245 unsigned* p_right
= this->g
+this->g_r
+1;
246 for(int i
=0; i
<this->g_r
;++i
,p_left
--,p_right
++)
250 *p_left
= static_cast<int>(exp_0
+.5);
252 volume
+= (*p_left
<<1);
255 // for (i = 0; i < this->g_w; ++i) {
256 // this->g[i] = (unsigned) ( exp(a * (i - this->g_r) * (i - this->g_r))* volume_factor + .5 );
257 // volume += this->g[i];
262 for (int mx
= 0; mx
< this->g_w
; mx
++) {
264 unsigned *p_gt2
= this->gt2
+ mx
;
266 for (int i
= 1; i
< 256; i
++) {
267 last_mul
= last_mul
+this->g
[mx
];
271 // this->gt2[this->g_w * i+ mx] = this->g[mx] * i;
278 ass_tmp_buf::ass_tmp_buf(size_t size
)
280 tmp
= (unsigned *)malloc(size
* sizeof(unsigned));
284 ass_tmp_buf::ass_tmp_buf(const ass_tmp_buf
& buf
)
287 tmp
= (unsigned *)malloc(size
* sizeof(unsigned));
290 ass_tmp_buf::~ass_tmp_buf()
296 * \brief gaussian blur. an fast pure c implementation from libass.
298 static void ass_gauss_blur(unsigned char *buffer
, unsigned *tmp2
,
299 int width
, int height
, int stride
, const unsigned *m2
,
305 unsigned char *s
= buffer
;
306 unsigned *t
= tmp2
+ 1;
307 for (y
= 0; y
< height
; y
++) {
308 memset(t
- 1, 0, (width
+ 1) * sizeof(*t
));
310 if(x
< r
)//in case that r < 0
312 const int src
= s
[x
];
314 register unsigned *dstp
= t
+ x
- r
;
316 const unsigned *m3
= m2
+ src
* mwidth
;
318 for (mx
= mwidth
-1; mx
>= r
- x
; mx
--) {
325 for (x
= 1; x
< r
; x
++) {
326 const int src
= s
[x
];
328 register unsigned *dstp
= t
+ x
- r
;
330 const unsigned *m3
= m2
+ src
* mwidth
;
331 for (mx
= r
- x
; mx
< mwidth
; mx
++) {
337 for (; x
< width
- r
; x
++) {
338 const int src
= s
[x
];
340 register unsigned *dstp
= t
+ x
- r
;
342 const unsigned *m3
= m2
+ src
* mwidth
;
343 for (mx
= 0; mx
< mwidth
; mx
++) {
349 for (; x
< width
-1; x
++) {
350 const int src
= s
[x
];
352 register unsigned *dstp
= t
+ x
- r
;
354 const int x2
= r
+ width
- x
;
355 const unsigned *m3
= m2
+ src
* mwidth
;
356 for (mx
= 0; mx
< x2
; mx
++) {
361 if(x
==width
-1) //important: x==width-1 failed, if r==0
363 const int src
= s
[x
];
365 register unsigned *dstp
= t
+ x
- r
;
367 const int x2
= r
+ width
- x
;
368 const unsigned *m3
= m2
+ src
* mwidth
;
370 for (mx
= 0; mx
< x2
; mx
++) {
382 for (x
= 0; x
< width
; x
++) {
384 if(y
< r
)//in case that r<0
386 unsigned *srcp
= t
+ y
* (width
+ 1) + 1;
389 register unsigned *dstp
= srcp
- 1 + (mwidth
-r
+y
)*(width
+ 1);
390 const int src2
= (src
+ (1<<(ass_synth_priv::VOLUME_BITS
-1))) >> ass_synth_priv::VOLUME_BITS
;
391 const unsigned *m3
= m2
+ src2
* mwidth
;
394 *srcp
= (1<<(ass_synth_priv::VOLUME_BITS
-1));
395 for (mx
= mwidth
-1; mx
>=r
- y
; mx
--) {
402 for (y
= 1; y
< r
; y
++) {
403 unsigned *srcp
= t
+ y
* (width
+ 1) + 1;
406 register unsigned *dstp
= srcp
- 1 + width
+ 1;
407 const int src2
= (src
+ (1<<(ass_synth_priv::VOLUME_BITS
-1))) >> ass_synth_priv::VOLUME_BITS
;
408 const unsigned *m3
= m2
+ src2
* mwidth
;
411 *srcp
= (1<<(ass_synth_priv::VOLUME_BITS
-1));
412 for (mx
= r
- y
; mx
< mwidth
; mx
++) {
418 for (; y
< height
- r
; y
++) {
419 unsigned *srcp
= t
+ y
* (width
+ 1) + 1;
422 register unsigned *dstp
= srcp
- 1 - r
* (width
+ 1);
423 const int src2
= (src
+ (1<<(ass_synth_priv::VOLUME_BITS
-1))) >> ass_synth_priv::VOLUME_BITS
;
424 const unsigned *m3
= m2
+ src2
* mwidth
;
427 *srcp
= (1<<(ass_synth_priv::VOLUME_BITS
-1));
428 for (mx
= 0; mx
< mwidth
; mx
++) {
434 for (; y
< height
-1; y
++) {
435 unsigned *srcp
= t
+ y
* (width
+ 1) + 1;
438 const int y2
= r
+ height
- y
;
439 register unsigned *dstp
= srcp
- 1 - r
* (width
+ 1);
440 const int src2
= (src
+ (1<<(ass_synth_priv::VOLUME_BITS
-1))) >> ass_synth_priv::VOLUME_BITS
;
441 const unsigned *m3
= m2
+ src2
* mwidth
;
444 *srcp
= (1<<(ass_synth_priv::VOLUME_BITS
-1));
445 for (mx
= 0; mx
< y2
; mx
++) {
451 if(y
== height
- 1)//important: y == height - 1 failed if r==0
453 unsigned *srcp
= t
+ y
* (width
+ 1) + 1;
456 const int y2
= r
+ height
- y
;
457 register unsigned *dstp
= srcp
- 1 - r
* (width
+ 1);
458 const int src2
= (src
+ (1<<(ass_synth_priv::VOLUME_BITS
-1))) >> ass_synth_priv::VOLUME_BITS
;
459 const unsigned *m3
= m2
+ src2
* mwidth
;
462 *srcp
= (1<<(ass_synth_priv::VOLUME_BITS
-1));
463 for (mx
= 0; mx
< y2
; mx
++) {
475 for (y
= 0; y
< height
; y
++) {
476 for (x
= 0; x
< width
; x
++) {
477 s
[x
] = t
[x
] >> ass_synth_priv::VOLUME_BITS
;
485 * \brief blur with [[1,2,1]. [2,4,2], [1,2,1]] kernel.
487 static void be_blur(unsigned char *buf
, unsigned *tmp_base
, int w
, int h
, int stride
)
489 WORD
*col_pix_buf_base
= reinterpret_cast<WORD
*>(xy_malloc(w
*sizeof(WORD
)));
490 WORD
*col_sum_buf_base
= reinterpret_cast<WORD
*>(xy_malloc(w
*sizeof(WORD
)));
491 if(!col_sum_buf_base
|| !col_pix_buf_base
)
493 //ToDo: error handling
496 memset(col_pix_buf_base
, 0, w
*sizeof(WORD
));
497 memset(col_sum_buf_base
, 0, w
*sizeof(WORD
));
498 WORD
*col_pix_buf
= col_pix_buf_base
-2;//for aligment;
499 WORD
*col_sum_buf
= col_sum_buf_base
-2;//for aligment;
502 unsigned char *src
=buf
+y
*stride
;
505 int old_pix
= src
[x
-1];
506 int old_sum
= old_pix
+ src
[x
-2];
507 for ( ; x
< w
; x
++) {
509 int temp2
= old_pix
+ temp1
;
511 temp1
= old_sum
+ temp2
;
513 col_pix_buf
[x
] = temp1
;
518 unsigned char *src
=buf
+y
*stride
;
522 int old_pix
= src
[x
-1];
523 int old_sum
= old_pix
+ src
[x
-2];
524 for ( ; x
< w
; x
++) {
526 int temp2
= old_pix
+ temp1
;
528 temp1
= old_sum
+ temp2
;
531 temp2
= col_pix_buf
[x
] + temp1
;
532 col_pix_buf
[x
] = temp1
;
533 //dst[x-1] = (col_sum_buf[x] + temp2) >> 4;
534 col_sum_buf
[x
] = temp2
;
538 //__m128i round = _mm_set1_epi16(8);
539 for (int y
= 2; y
< h
; y
++) {
540 unsigned char *src
=buf
+y
*stride
;
541 unsigned char *dst
=buf
+(y
-1)*stride
;
545 __m128i old_pix_128
= _mm_cvtsi32_si128(src
[1]);
546 __m128i old_sum_128
= _mm_cvtsi32_si128(src
[0]+src
[1]);
547 for ( ; x
< ((w
-2)&(~7)); x
+=8) {
548 __m128i new_pix
= _mm_loadl_epi64(reinterpret_cast<const __m128i
*>(src
+x
));
549 new_pix
= _mm_unpacklo_epi8(new_pix
, _mm_setzero_si128());
550 __m128i temp
= _mm_slli_si128(new_pix
,2);
551 temp
= _mm_add_epi16(temp
, old_pix_128
);
552 temp
= _mm_add_epi16(temp
, new_pix
);
553 old_pix_128
= _mm_srli_si128(new_pix
,14);
555 new_pix
= _mm_slli_si128(temp
,2);
556 new_pix
= _mm_add_epi16(new_pix
, old_sum_128
);
557 new_pix
= _mm_add_epi16(new_pix
, temp
);
558 old_sum_128
= _mm_srli_si128(temp
, 14);
560 __m128i old_col_pix
= _mm_loadu_si128( reinterpret_cast<const __m128i
*>(col_pix_buf
+x
) );
561 __m128i old_col_sum
= _mm_loadu_si128( reinterpret_cast<const __m128i
*>(col_sum_buf
+x
) );
562 _mm_storeu_si128( reinterpret_cast<__m128i
*>(col_pix_buf
+x
), new_pix
);
563 temp
= _mm_add_epi16(new_pix
, old_col_pix
);
564 _mm_storeu_si128( reinterpret_cast<__m128i
*>(col_sum_buf
+x
), temp
);
566 old_col_sum
= _mm_add_epi16(old_col_sum
, temp
);
567 //old_col_sum = _mm_add_epi16(old_col_sum, round);
568 old_col_sum
= _mm_srli_epi16(old_col_sum
, 4);
569 old_col_sum
= _mm_packus_epi16(old_col_sum
, old_col_sum
);
570 _mm_storel_epi64( reinterpret_cast<__m128i
*>(dst
+x
-1), old_col_sum
);
572 int old_pix
= src
[x
-1];
573 int old_sum
= old_pix
+ src
[x
-2];
574 for ( ; x
< w
; x
++) {
576 int temp2
= old_pix
+ temp1
;
578 temp1
= old_sum
+ temp2
;
581 temp2
= col_pix_buf
[x
] + temp1
;
582 col_pix_buf
[x
] = temp1
;
583 dst
[x
-1] = (col_sum_buf
[x
] + temp2
) >> 4;
584 col_sum_buf
[x
] = temp2
;
588 xy_free(col_sum_buf_base
);
589 xy_free(col_pix_buf_base
);
592 static void Bilinear(unsigned char *buf
, int w
, int h
, int stride
, int x_factor
, int y_factor
)
594 WORD
*col_pix_buf_base
= reinterpret_cast<WORD
*>(xy_malloc(w
*sizeof(WORD
)));
595 if(!col_pix_buf_base
)
597 //ToDo: error handling
600 memset(col_pix_buf_base
, 0, w
*sizeof(WORD
));
602 for (int y
= 0; y
< h
; y
++){
603 unsigned char *src
=buf
+y
*stride
;
605 WORD
*col_pix_buf
= col_pix_buf_base
;
607 for(int x
= 0; x
< w
; x
++)
610 int temp2
= temp1
*x_factor
;
616 temp2
= temp1
*y_factor
;
619 temp1
+= col_pix_buf
[x
];
620 src
[x
] = ((temp1
+32)>>6);
621 col_pix_buf
[x
] = temp2
;
624 xy_free(col_pix_buf_base
);
627 bool Rasterizer::Rasterize(const ScanLineData2
& scan_line_data2
, int xsub
, int ysub
, SharedPtrOverlay overlay
)
629 using namespace ::boost::flyweights
;
636 const ScanLineData
& scan_line_data
= *scan_line_data2
.m_scan_line_data
;
637 if(!scan_line_data
.mWidth
|| !scan_line_data
.mHeight
)
644 int width
= scan_line_data
.mWidth
+ xsub
;
645 int height
= scan_line_data
.mHeight
+ ysub
;
646 overlay
->mfWideOutlineEmpty
= scan_line_data2
.mWideOutline
.empty();
647 if(!overlay
->mfWideOutlineEmpty
)
649 int wide_border
= (scan_line_data2
.mWideBorder
+7)&~7;
651 width
+= 2*wide_border
;
652 height
+= 2*wide_border
;
653 xsub
+= wide_border
;
654 ysub
+= wide_border
;
656 overlay
->mOffsetX
= scan_line_data2
.mPathOffsetX
- xsub
;
657 overlay
->mOffsetY
= scan_line_data2
.mPathOffsetY
- ysub
;
659 overlay
->mWidth
= width
;
660 overlay
->mHeight
= height
;
661 overlay
->mOverlayWidth
= ((width
+7)>>3) + 1;
662 overlay
->mOverlayHeight
= ((height
+7)>>3) + 1;
663 overlay
->mOverlayPitch
= (overlay
->mOverlayWidth
+15)&~15;
665 BYTE
* body
= reinterpret_cast<BYTE
*>(xy_malloc(overlay
->mOverlayPitch
* overlay
->mOverlayHeight
));
670 overlay
->mBody
.reset(body
, xy_free
);
671 memset(body
, 0, overlay
->mOverlayPitch
* overlay
->mOverlayHeight
);
673 if (!overlay
->mfWideOutlineEmpty
)
675 border
= reinterpret_cast<BYTE
*>(xy_malloc(overlay
->mOverlayPitch
* overlay
->mOverlayHeight
));
680 overlay
->mBorder
.reset(border
, xy_free
);
681 memset(border
, 0, overlay
->mOverlayPitch
* overlay
->mOverlayHeight
);
684 // Are we doing a border?
685 const tSpanBuffer
* pOutline
[2] = {&(scan_line_data
.mOutline
), &(scan_line_data2
.mWideOutline
)};
686 for(int i
= countof(pOutline
)-1; i
>= 0; i
--)
688 tSpanBuffer::const_iterator it
= pOutline
[i
]->begin();
689 tSpanBuffer::const_iterator itEnd
= pOutline
[i
]->end();
690 byte
* plan_selected
= i
==0 ? body
: border
;
691 int pitch
= overlay
->mOverlayPitch
;
692 for(; it
!=itEnd
; ++it
)
694 int y
= (int)(((*it
).first
>> 32) - 0x40000000 + ysub
);
695 int x1
= (int)(((*it
).first
& 0xffffffff) - 0x40000000 + xsub
);
696 int x2
= (int)(((*it
).second
& 0xffffffff) - 0x40000000 + xsub
);
700 int last
= (x2
-1)>>3;
701 byte
* dst
= plan_selected
+ (pitch
*(y
>>3) + first
);
706 *dst
+= ((first
+1)<<3) - x1
;
708 while(++first
< last
)
713 *dst
+= x2
- (last
<<3);
722 // @return: true if actually a blur operation has done, or else false and output is leave unset.
723 bool Rasterizer::Blur(const Overlay
& input_overlay
, int fBlur
, double fGaussianBlur
,
724 SharedPtrOverlay output_overlay
)
726 using namespace ::boost::flyweights
;
732 output_overlay
->CleanUp();
734 output_overlay
->mOffsetX
= input_overlay
.mOffsetX
;
735 output_overlay
->mOffsetY
= input_overlay
.mOffsetY
;
736 output_overlay
->mWidth
= input_overlay
.mWidth
;
737 output_overlay
->mHeight
= input_overlay
.mHeight
;
738 output_overlay
->mOverlayWidth
= input_overlay
.mOverlayWidth
;
739 output_overlay
->mOverlayHeight
= input_overlay
.mOverlayHeight
;
740 output_overlay
->mfWideOutlineEmpty
= input_overlay
.mfWideOutlineEmpty
;
743 if(fBlur
|| fGaussianBlur
> 0.1)
745 if (fGaussianBlur
> 0)
746 bluradjust
+= (int)(fGaussianBlur
*3*8 + 0.5) | 1;
749 // Expand the buffer a bit when we're blurring, since that can also widen the borders a bit
750 bluradjust
= (bluradjust
+7)&~7;
752 output_overlay
->mOffsetX
-= bluradjust
;
753 output_overlay
->mOffsetY
-= bluradjust
;
754 output_overlay
->mWidth
+= (bluradjust
<<1);
755 output_overlay
->mHeight
+= (bluradjust
<<1);
756 output_overlay
->mOverlayWidth
+= (bluradjust
>>2);
757 output_overlay
->mOverlayHeight
+= (bluradjust
>>2);
764 output_overlay
->mOverlayPitch
= (output_overlay
->mOverlayWidth
+15)&~15;
766 BYTE
* body
= reinterpret_cast<BYTE
*>(xy_malloc(output_overlay
->mOverlayPitch
* output_overlay
->mOverlayHeight
));
771 output_overlay
->mBody
.reset(body
, xy_free
);
772 memset(body
, 0, output_overlay
->mOverlayPitch
* output_overlay
->mOverlayHeight
);
774 if (!output_overlay
->mfWideOutlineEmpty
)
776 border
= reinterpret_cast<BYTE
*>(xy_malloc(output_overlay
->mOverlayPitch
* output_overlay
->mOverlayHeight
));
781 output_overlay
->mBorder
.reset(border
, xy_free
);
782 memset(border
, 0, output_overlay
->mOverlayPitch
* output_overlay
->mOverlayHeight
);
786 for(int i
= 1; i
>= 0; i
--)
788 byte
* plan_selected
= i
==0 ? body
: border
;
789 const byte
* plan_input
= i
==0 ? input_overlay
.mBody
.get() : input_overlay
.mBorder
.get();
791 plan_selected
+= (bluradjust
>>3) + (bluradjust
>>3)*output_overlay
->mOverlayPitch
;
792 if ( plan_selected
!=NULL
&& plan_input
!=NULL
)
794 for (int j
=0;j
<input_overlay
.mOverlayHeight
;j
++)
796 memcpy(plan_selected
, plan_input
, input_overlay
.mOverlayPitch
);
797 plan_selected
+= output_overlay
->mOverlayPitch
;
798 plan_input
+= input_overlay
.mOverlayPitch
;
803 ass_tmp_buf
tmp_buf( max((output_overlay
->mOverlayPitch
+1)*(output_overlay
->mOverlayHeight
+1),0) );
804 //flyweight<key_value<int, ass_tmp_buf, ass_tmp_buf_get_size>, no_locking> tmp_buf((overlay->mOverlayWidth+1)*(overlay->mOverlayPitch+1));
805 // Do some gaussian blur magic
806 if (fGaussianBlur
> 0.1)//(fGaussianBlur > 0) return true even if fGaussianBlur very small
808 byte
* plan_selected
= output_overlay
->mfWideOutlineEmpty
? body
: border
;
809 flyweight
<key_value
<double, ass_synth_priv
, ass_synth_priv_key
>, no_locking
> fw_priv_blur(fGaussianBlur
);
810 const ass_synth_priv
& priv_blur
= fw_priv_blur
.get();
811 if (output_overlay
->mOverlayWidth
>=priv_blur
.g_w
&& output_overlay
->mOverlayHeight
>=priv_blur
.g_w
)
813 ass_gauss_blur(plan_selected
, tmp_buf
.tmp
, output_overlay
->mOverlayWidth
, output_overlay
->mOverlayHeight
, output_overlay
->mOverlayPitch
,
814 priv_blur
.gt2
, priv_blur
.g_r
, priv_blur
.g_w
);
818 for (int pass
= 0; pass
< fBlur
; pass
++)
820 if(output_overlay
->mOverlayWidth
>= 3 && output_overlay
->mOverlayHeight
>= 3)
822 int pitch
= output_overlay
->mOverlayPitch
;
823 byte
* plan_selected
= output_overlay
->mfWideOutlineEmpty
? body
: border
;
824 be_blur(plan_selected
, tmp_buf
.tmp
, output_overlay
->mOverlayWidth
, output_overlay
->mOverlayHeight
, pitch
);
830 ///////////////////////////////////////////////////////////////////////////
832 static __forceinline
void pixmix(DWORD
*dst
, DWORD color
, DWORD alpha
)
835 // Make sure both a and ia are in range 1..256 for the >>8 operations below to be correct
838 *dst
= ((((*dst
&0x00ff00ff)*ia
+ (color
&0x00ff00ff)*a
)&0xff00ff00)>>8)
839 | ((((*dst
&0x0000ff00)*ia
+ (color
&0x0000ff00)*a
)&0x00ff0000)>>8)
840 | ((((*dst
>>8)&0x00ff0000)*ia
)&0xff000000);
843 static __forceinline
void pixmix2(DWORD
*dst
, DWORD color
, DWORD shapealpha
, DWORD clipalpha
)
845 int a
= (((shapealpha
)*(clipalpha
)*(color
>>24))>>12)&0xff;
848 *dst
= ((((*dst
&0x00ff00ff)*ia
+ (color
&0x00ff00ff)*a
)&0xff00ff00)>>8)
849 | ((((*dst
&0x0000ff00)*ia
+ (color
&0x0000ff00)*a
)&0x00ff0000)>>8)
850 | ((((*dst
>>8)&0x00ff0000)*ia
)&0xff000000);
853 #include <xmmintrin.h>
854 #include <emmintrin.h>
856 static __forceinline
void pixmix_sse2(DWORD
* dst
, DWORD color
, DWORD alpha
)
858 // alpha = (((alpha) * (color>>24)) >> 6) & 0xff;
860 __m128i zero
= _mm_setzero_si128();
861 __m128i a
= _mm_set1_epi32(((alpha
+1) << 16) | (0x100 - alpha
));
862 __m128i d
= _mm_unpacklo_epi8(_mm_cvtsi32_si128(*dst
), zero
);
863 __m128i s
= _mm_unpacklo_epi8(_mm_cvtsi32_si128(color
), zero
);
864 __m128i r
= _mm_unpacklo_epi16(d
, s
);
865 r
= _mm_madd_epi16(r
, a
);
866 r
= _mm_srli_epi32(r
, 8);
867 r
= _mm_packs_epi32(r
, r
);
868 r
= _mm_packus_epi16(r
, r
);
869 *dst
= (DWORD
)_mm_cvtsi128_si32(r
);
872 static __forceinline
void pixmix2_sse2(DWORD
* dst
, DWORD color
, DWORD shapealpha
, DWORD clipalpha
)
874 int alpha
= (((shapealpha
)*(clipalpha
)*(color
>>24))>>12)&0xff;
876 __m128i zero
= _mm_setzero_si128();
877 __m128i a
= _mm_set1_epi32(((alpha
+1) << 16) | (0x100 - alpha
));
878 __m128i d
= _mm_unpacklo_epi8(_mm_cvtsi32_si128(*dst
), zero
);
879 __m128i s
= _mm_unpacklo_epi8(_mm_cvtsi32_si128(color
), zero
);
880 __m128i r
= _mm_unpacklo_epi16(d
, s
);
881 r
= _mm_madd_epi16(r
, a
);
882 r
= _mm_srli_epi32(r
, 8);
883 r
= _mm_packs_epi32(r
, r
);
884 r
= _mm_packus_epi16(r
, r
);
885 *dst
= (DWORD
)_mm_cvtsi128_si32(r
);
888 #include <mmintrin.h>
890 // Calculate a - b clamping to 0 instead of underflowing
891 static __forceinline DWORD
safe_subtract(DWORD a
, DWORD b
)
893 __m64 ap
= _mm_cvtsi32_si64(a
);
894 __m64 bp
= _mm_cvtsi32_si64(b
);
895 __m64 rp
= _mm_subs_pu16(ap
, bp
);
896 DWORD r
= (DWORD
)_mm_cvtsi64_si32(rp
);
899 //return (b > a) ? 0 : a - b;
903 * No aligned requirement
906 void AlphaBlt(byte
* pY
,
907 const byte
* pAlphaMask
,
909 int h
, int w
, int src_stride
, int dst_stride
)
911 __m128i zero
= _mm_setzero_si128();
912 __m128i s
= _mm_set1_epi16(Y
); //s = c 0 c 0 c 0 c 0 c 0 c 0 c 0 c 0
914 if( w
>16 )//IMPORTANT! The result of the following code is undefined with w<15.
916 for( ; h
>0; h
--, pAlphaMask
+= src_stride
, pY
+= dst_stride
)
918 const BYTE
* sa
= pAlphaMask
;
920 const BYTE
* dy_first_mod16
= reinterpret_cast<BYTE
*>((reinterpret_cast<int>(pY
)+15)&~15); //IMPORTANT! w must >= 15
921 const BYTE
* dy_end_mod16
= reinterpret_cast<BYTE
*>(reinterpret_cast<int>(pY
+w
)&~15);
922 const BYTE
* dy_end
= pY
+ w
;
924 for(;dy
< dy_first_mod16
; sa
++, dy
++)
926 *dy
= (*dy
* (256 - *sa
)+ Y
*(*sa
+1))>>8;
928 for(; dy
< dy_end_mod16
; sa
+=8, dy
+=16)
930 __m128i a
= _mm_loadl_epi64((__m128i
*)sa
);
933 __m128i d
= _mm_load_si128((__m128i
*)dy
);
935 //__m128i ones = _mm_cmpeq_epi32(zero,zero); //ones = ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
936 //__m128i ia = _mm_xor_si128(a,ones); //ia = ~a
937 //ia = _mm_unpacklo_epi8(ia,zero); //ia = ~a0 0 ~a1 0 ~a2 0 ~a3 0 ~a4 0 ~a5 0 ~a6 0 ~a7 0
938 a
= _mm_unpacklo_epi8(a
,zero
); //a= a0 0 a1 0 a2 0 a3 0 a4 0 a5 0 a6 0 a7 0
939 __m128i ones
= _mm_set1_epi16(256); //ones = 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
940 __m128i ia
= _mm_sub_epi16(ones
, a
); //ia = 256-a0 ... 256-a7
941 ones
= _mm_srli_epi16(ones
, 8);
942 a
= _mm_add_epi16(a
, ones
); //a= 1+a0 ... 1+a7
944 __m128i dl
= _mm_unpacklo_epi8(d
,zero
); //d = b0 0 b1 0 b2 0 b3 0 b4 0 b5 0 b6 0 b7 0
945 __m128i sl
= _mm_mullo_epi16(s
,a
); //sl = c0*a0 c1*a1 ... c7*a7
947 dl
= _mm_mullo_epi16(dl
,ia
); //d = b0*~a0 b1*~a1 ... b7*~a7
949 dl
= _mm_add_epi16(dl
,sl
); //d = d + sl
950 dl
= _mm_srli_epi16(dl
, 8); //d = d>>8
953 a
= _mm_loadl_epi64((__m128i
*)sa
);
955 a
= _mm_unpacklo_epi8(a
,zero
);
956 ones
= _mm_slli_epi16(ones
, 8);
957 ia
= _mm_sub_epi16(ones
, a
);
958 ones
= _mm_srli_epi16(ones
, 8);
959 a
= _mm_add_epi16(a
,ones
);
961 d
= _mm_unpackhi_epi8(d
,zero
);
962 sl
= _mm_mullo_epi16(s
,a
);
963 d
= _mm_mullo_epi16(d
,ia
);
964 d
= _mm_add_epi16(d
,sl
);
965 d
= _mm_srli_epi16(d
, 8);
967 dl
= _mm_packus_epi16(dl
,d
);
969 _mm_store_si128((__m128i
*)dy
, dl
);
971 for(;dy
< dy_end
; sa
++, dy
++)
973 *dy
= (*dy
* (256 - *sa
)+ Y
*(*sa
+1))>>8;
979 for( ; h
>0; h
--, pAlphaMask
+= src_stride
, pY
+= dst_stride
)
981 const BYTE
* sa
= pAlphaMask
;
983 const BYTE
* dy_end
= pY
+ w
;
985 for(;dy
< dy_end
; sa
++, dy
++)
987 *dy
= (*dy
* (256 - *sa
)+ Y
*(*sa
+1))>>8;
995 * No aligned requirement
998 void AlphaBlt(byte
* pY
,
1001 int h
, int w
, int dst_stride
)
1003 int yPremul
= Y
*(alpha
+1);
1004 int dstAlpha
= 0x100 - alpha
;
1005 if( w
>32 )//IMPORTANT! The result of the following code is undefined with w<15.
1007 __m128i zero
= _mm_setzero_si128();
1008 __m128i s
= _mm_set1_epi16(yPremul
); //s = c 0 c 0 c 0 c 0 c 0 c 0 c 0 c 0
1009 __m128i ia
= _mm_set1_epi16(dstAlpha
);
1010 for( ; h
>0; h
--, pY
+= dst_stride
)
1013 const BYTE
* dy_first_mod16
= reinterpret_cast<BYTE
*>((reinterpret_cast<int>(pY
)+15)&~15); //IMPORTANT! w must >= 15
1014 const BYTE
* dy_end_mod16
= reinterpret_cast<BYTE
*>(reinterpret_cast<int>(pY
+w
)&~15);
1015 const BYTE
* dy_end
= pY
+ w
;
1017 for(;dy
< dy_first_mod16
; dy
++)
1019 *dy
= (*dy
* dstAlpha
+ yPremul
)>>8;
1021 for(; dy
< dy_end_mod16
; dy
+=16)
1024 __m128i d
= _mm_load_si128(reinterpret_cast<const __m128i
*>(dy
));
1025 __m128i dl
= _mm_unpacklo_epi8(d
,zero
); //d = b0 0 b1 0 b2 0 b3 0 b4 0 b5 0 b6 0 b7 0
1027 dl
= _mm_mullo_epi16(dl
,ia
); //d = b0*~a0 b1*~a1 ... b7*~a7
1028 dl
= _mm_adds_epu16(dl
,s
); //d = d + s
1029 dl
= _mm_srli_epi16(dl
, 8); //d = d>>8
1031 d
= _mm_unpackhi_epi8(d
,zero
);
1032 d
= _mm_mullo_epi16(d
,ia
);
1033 d
= _mm_adds_epu16(d
,s
);
1034 d
= _mm_srli_epi16(d
, 8);
1036 dl
= _mm_packus_epi16(dl
,d
);
1038 _mm_store_si128(reinterpret_cast<__m128i
*>(dy
), dl
);
1040 for(;dy
< dy_end
; dy
++)
1042 *dy
= (*dy
* dstAlpha
+ yPremul
)>>8;
1048 for( ; h
>0; h
--, pY
+= dst_stride
)
1051 const BYTE
* dy_end
= pY
+ w
;
1053 for(;dy
< dy_end
; dy
++)
1055 *dy
= (*dy
* dstAlpha
+ yPremul
)>>8;
1063 * No aligned requirement
1066 void AlphaBltC(byte
* pY
,
1069 int h
, int w
, int dst_stride
)
1071 int yPremul
= Y
*(alpha
+1);
1072 int dstAlpha
= 0x100 - alpha
;
1074 for( ; h
>0; h
--, pY
+= dst_stride
)
1077 const BYTE
* dy_end
= pY
+ w
;
1079 for(;dy
< dy_end
; dy
++)
1081 *dy
= (*dy
* dstAlpha
+ yPremul
)>>8;
1086 // For CPUID usage in Rasterizer::Draw
1087 #include "../dsutil/vd.h"
1089 void OverlapRegion(tSpanBuffer
& dst
, const tSpanBuffer
& src
, int dx
, int dy
)
1092 temp
.reserve(dst
.size() + src
.size());
1094 tSpanBuffer::iterator itA
= temp
.begin();
1095 tSpanBuffer::iterator itAE
= temp
.end();
1096 tSpanBuffer::const_iterator itB
= src
.begin();
1097 tSpanBuffer::const_iterator itBE
= src
.end();
1098 // Don't worry -- even if dy<0 this will still work! // G: hehe, the evil twin :)
1099 unsigned __int64 offset1
= (((__int64
)dy
)<<32) - dx
;
1100 unsigned __int64 offset2
= (((__int64
)dy
)<<32) + dx
;
1101 while(itA
!= itAE
&& itB
!= itBE
)
1103 if((*itB
).first
+ offset1
< (*itA
).first
)
1105 // B span is earlier. Use it.
1106 unsigned __int64 x1
= (*itB
).first
+ offset1
;
1107 unsigned __int64 x2
= (*itB
).second
+ offset2
;
1109 // B spans don't overlap, so begin merge loop with A first.
1112 // If we run out of A spans or the A span doesn't overlap,
1113 // then the next B span can't either (because B spans don't
1114 // overlap) and we exit.
1115 if(itA
== itAE
|| (*itA
).first
> x2
)
1117 do {x2
= _MAX(x2
, (*itA
++).second
);}
1118 while(itA
!= itAE
&& (*itA
).first
<= x2
);
1119 // If we run out of B spans or the B span doesn't overlap,
1120 // then the next A span can't either (because A spans don't
1121 // overlap) and we exit.
1122 if(itB
== itBE
|| (*itB
).first
+ offset1
> x2
)
1124 do {x2
= _MAX(x2
, (*itB
++).second
+ offset2
);}
1125 while(itB
!= itBE
&& (*itB
).first
+ offset1
<= x2
);
1128 dst
.push_back(tSpan(x1
, x2
));
1132 // A span is earlier. Use it.
1133 unsigned __int64 x1
= (*itA
).first
;
1134 unsigned __int64 x2
= (*itA
).second
;
1136 // A spans don't overlap, so begin merge loop with B first.
1139 // If we run out of B spans or the B span doesn't overlap,
1140 // then the next A span can't either (because A spans don't
1141 // overlap) and we exit.
1142 if(itB
== itBE
|| (*itB
).first
+ offset1
> x2
)
1144 do {x2
= _MAX(x2
, (*itB
++).second
+ offset2
);}
1145 while(itB
!= itBE
&& (*itB
).first
+ offset1
<= x2
);
1146 // If we run out of A spans or the A span doesn't overlap,
1147 // then the next B span can't either (because B spans don't
1148 // overlap) and we exit.
1149 if(itA
== itAE
|| (*itA
).first
> x2
)
1151 do {x2
= _MAX(x2
, (*itA
++).second
);}
1152 while(itA
!= itAE
&& (*itA
).first
<= x2
);
1155 dst
.push_back(tSpan(x1
, x2
));
1158 // Copy over leftover spans.
1160 dst
.push_back(*itA
++);
1163 dst
.push_back(tSpan((*itB
).first
+ offset1
, (*itB
).second
+ offset2
));
1168 // Render a subpicture onto a surface.
1169 // spd is the surface to render on.
1170 // clipRect is a rectangular clip region to render inside.
1171 // pAlphaMask is an alpha clipping mask.
1172 // xsub and ysub ???
1173 // switchpts seems to be an array of fill colours interlaced with coordinates.
1174 // switchpts[i*2] contains a colour and switchpts[i*2+1] contains the coordinate to use that colour from
1175 // fBody tells whether to render the body of the subs.
1176 // fBorder tells whether to render the border of the subs.
1177 SharedPtrByte
Rasterizer::CompositeAlphaMask(const SharedPtrOverlay
& overlay
, const CRect
& clipRect
,
1178 const GrayImage2
* alpha_mask
,
1179 int xsub
, int ysub
, const DWORD
* switchpts
, bool fBody
, bool fBorder
,
1180 CRect
*outputDirtyRect
)
1182 //fix me: check and log error
1183 SharedPtrByte result
;
1184 *outputDirtyRect
= CRect(0, 0, 0, 0);
1185 if (!switchpts
|| !fBody
&& !fBorder
) return result
;
1186 if (fBorder
&& !overlay
->mBorder
) return result
;
1189 if (alpha_mask
!=NULL
)
1191 r
&= CRect(alpha_mask
->left_top
, alpha_mask
->size
);
1194 // Remember that all subtitle coordinates are specified in 1/8 pixels
1195 // (x+4)>>3 rounds to nearest whole pixel.
1196 // ??? What is xsub, ysub, mOffsetX and mOffsetY ?
1197 int x
= (xsub
+ overlay
->mOffsetX
+ 4)>>3;
1198 int y
= (ysub
+ overlay
->mOffsetY
+ 4)>>3;
1199 int w
= overlay
->mOverlayWidth
;
1200 int h
= overlay
->mOverlayHeight
;
1203 if(x
< r
.left
) {xo
= r
.left
-x
; w
-= r
.left
-x
; x
= r
.left
;}
1204 if(y
< r
.top
) {yo
= r
.top
-y
; h
-= r
.top
-y
; y
= r
.top
;}
1205 if(x
+w
> r
.right
) w
= r
.right
-x
;
1206 if(y
+h
> r
.bottom
) h
= r
.bottom
-y
;
1207 // Check if there's actually anything to render
1208 if(w
<= 0 || h
<= 0) return(result
);
1209 outputDirtyRect
->SetRect(x
, y
, x
+w
, y
+h
);
1211 bool fSingleColor
= (switchpts
[1]==0xffffffff);
1214 // Grab the first colour
1215 DWORD color
= switchpts
[0];
1216 byte
* s_base
= (byte
*)xy_malloc(overlay
->mOverlayPitch
* overlay
->mOverlayHeight
);
1217 const byte
* alpha_mask_data
= alpha_mask
!= NULL
? alpha_mask
->data
.get() : NULL
;
1218 const int alpha_mask_pitch
= alpha_mask
!= NULL
? alpha_mask
->pitch
: 0;
1219 if(alpha_mask_data
!=NULL
)
1220 alpha_mask_data
+= alpha_mask
->pitch
* y
+ x
- alpha_mask
->left_top
.y
*alpha_mask
->pitch
- alpha_mask
->left_top
.x
;
1224 overlay
->FillAlphaMash(s_base
, fBody
, fBorder
, xo
, yo
, w
, h
,
1225 alpha_mask_data
, alpha_mask_pitch
,
1231 const DWORD
*sw
= switchpts
;
1232 while( last_x
<w
+xo
)
1234 byte alpha
= sw
[0]>>24;
1235 while( sw
[3]<w
+xo
&& (sw
[2]>>24)==alpha
)
1239 int new_x
= sw
[3] < w
+xo
? sw
[3] : w
+xo
;
1240 overlay
->FillAlphaMash(s_base
, fBody
, fBorder
,
1241 last_x
, yo
, new_x
-last_x
, h
,
1242 alpha_mask_data
, alpha_mask_pitch
,
1248 result
.reset( s_base
, xy_free
);
1254 // draw overlay[clipRect] to bitmap[0,0,w,h]
1256 void Rasterizer::Draw(XyBitmap
* bitmap
, SharedPtrOverlay overlay
, const CRect
& clipRect
, byte
* s_base
,
1257 int xsub
, int ysub
, const DWORD
* switchpts
, bool fBody
, bool fBorder
)
1259 if (!switchpts
|| !fBody
&& !fBorder
) return;
1266 // Limit drawn area to rectangular clip area
1268 // Remember that all subtitle coordinates are specified in 1/8 pixels
1269 // (x+4)>>3 rounds to nearest whole pixel.
1270 int overlayPitch
= overlay
->mOverlayPitch
;
1271 int x
= (xsub
+ overlay
->mOffsetX
+ 4)>>3;
1272 int y
= (ysub
+ overlay
->mOffsetY
+ 4)>>3;
1273 int w
= overlay
->mOverlayWidth
;
1274 int h
= overlay
->mOverlayHeight
;
1277 if(x
< r
.left
) {xo
= r
.left
-x
; w
-= r
.left
-x
; x
= r
.left
;}
1278 if(y
< r
.top
) {yo
= r
.top
-y
; h
-= r
.top
-y
; y
= r
.top
;}
1279 if(x
+w
> r
.right
) w
= r
.right
-x
;
1280 if(y
+h
> r
.bottom
) h
= r
.bottom
-y
;
1281 // Check if there's actually anything to render
1282 if (w
<= 0 || h
<= 0) return;
1283 // must have enough space to draw into
1284 ASSERT(x
>= bitmap
->x
&& y
>= bitmap
->y
&& x
+w
<= bitmap
->x
+ bitmap
->w
&& y
+h
<= bitmap
->y
+ bitmap
->h
);
1287 bool fSSE2
= !!(g_cpuid
.m_flags
& CCpuID::sse2
);
1288 bool fSingleColor
= (switchpts
[1]==0xffffffff);
1289 bool PLANAR
= (bitmap
->type
==XyBitmap::PLANNA
);
1290 int draw_method
= 0;
1292 draw_method
|= DM::SINGLE_COLOR
;
1294 draw_method
|= DM::SSE2
;
1296 draw_method
|= DM::AYUV_PLANAR
;
1299 // Grab the first colour
1300 DWORD color
= switchpts
[0];
1301 const byte
* s
= s_base
+ overlay
->mOverlayPitch
*yo
+ xo
;
1304 if (bitmap
->type
==XyBitmap::PLANNA
)
1305 dst_offset
= bitmap
->pitch
*(y
-bitmap
->y
) + x
- bitmap
->x
;
1307 dst_offset
= bitmap
->pitch
*(y
-bitmap
->y
) + (x
- bitmap
->x
)*4;
1308 unsigned long* dst
= (unsigned long*)((BYTE
*)bitmap
->plans
[0] + dst_offset
);
1310 // Every remaining line in the bitmap to be rendered...
1313 case DM::SINGLE_COLOR
| DM::SSE2
| 0*DM::AYUV_PLANAR
:
1317 for(int wt
=0; wt
<w
; ++wt
)
1318 // The <<6 is due to pixmix expecting the alpha parameter to be
1319 // the multiplication of two 6-bit unsigned numbers but we
1320 // only have one here. (No alpha mask.)
1321 pixmix_sse2(&dst
[wt
], color
, s
[wt
]);
1323 dst
= (unsigned long *)((char *)dst
+ bitmap
->pitch
);
1327 case DM::SINGLE_COLOR
| 0*DM::SSE2
| 0*DM::AYUV_PLANAR
:
1331 for(int wt
=0; wt
<w
; ++wt
)
1332 pixmix(&dst
[wt
], color
, s
[wt
]);
1334 dst
= (unsigned long *)((char *)dst
+ bitmap
->pitch
);
1338 case 0*DM::SINGLE_COLOR
| DM::SSE2
| 0*DM::AYUV_PLANAR
:
1342 const DWORD
*sw
= switchpts
;
1343 for(int wt
=0; wt
<w
; ++wt
)
1345 // xo is the offset (usually negative) we have moved into the image
1346 // So if we have passed the switchpoint (?) switch to another colour
1347 // (So switchpts stores both colours *and* coordinates?)
1348 if(wt
+xo
>= sw
[1]) {while(wt
+xo
>= sw
[1]) sw
+= 2; color
= sw
[-2];}
1349 pixmix_sse2(&dst
[wt
], color
, s
[wt
]);
1352 dst
= (unsigned long *)((char *)dst
+ bitmap
->pitch
);
1356 case 0*DM::SINGLE_COLOR
| 0*DM::SSE2
| 0*DM::AYUV_PLANAR
:
1360 const DWORD
*sw
= switchpts
;
1361 for(int wt
=0; wt
<w
; ++wt
)
1363 if(wt
+xo
>= sw
[1]) {while(wt
+xo
>= sw
[1]) sw
+= 2; color
= sw
[-2];}
1364 pixmix(&dst
[wt
], color
, s
[wt
]);
1367 dst
= (unsigned long *)((char *)dst
+ bitmap
->pitch
);
1371 case DM::SINGLE_COLOR
| DM::SSE2
| DM::AYUV_PLANAR
:
1373 unsigned char* dst_A
= bitmap
->plans
[0] + dst_offset
;
1374 unsigned char* dst_Y
= bitmap
->plans
[1] + dst_offset
;
1375 unsigned char* dst_U
= bitmap
->plans
[2] + dst_offset
;
1376 unsigned char* dst_V
= bitmap
->plans
[3] + dst_offset
;
1378 AlphaBlt(dst_Y
, s
, ((color
)>>16)&0xff, h
, w
, overlayPitch
, bitmap
->pitch
);
1379 AlphaBlt(dst_U
, s
, ((color
)>>8)&0xff, h
, w
, overlayPitch
, bitmap
->pitch
);
1380 AlphaBlt(dst_V
, s
, ((color
))&0xff, h
, w
, overlayPitch
, bitmap
->pitch
);
1381 AlphaBlt(dst_A
, s
, 0, h
, w
, overlayPitch
, bitmap
->pitch
);
1384 case 0*DM::SINGLE_COLOR
| DM::SSE2
| DM::AYUV_PLANAR
:
1386 unsigned char* dst_A
= bitmap
->plans
[0] + dst_offset
;
1387 unsigned char* dst_Y
= bitmap
->plans
[1] + dst_offset
;
1388 unsigned char* dst_U
= bitmap
->plans
[2] + dst_offset
;
1389 unsigned char* dst_V
= bitmap
->plans
[3] + dst_offset
;
1391 const DWORD
*sw
= switchpts
;
1396 int new_x
= sw
[3] < w
+xo
? sw
[3] : w
+xo
;
1399 if( new_x
< last_x
)
1401 AlphaBlt(dst_Y
, s
+ last_x
- xo
, (color
>>16)&0xff, h
, new_x
-last_x
, overlayPitch
, bitmap
->pitch
);
1402 AlphaBlt(dst_U
, s
+ last_x
- xo
, (color
>>8)&0xff, h
, new_x
-last_x
, overlayPitch
, bitmap
->pitch
);
1403 AlphaBlt(dst_V
, s
+ last_x
- xo
, (color
)&0xff, h
, new_x
-last_x
, overlayPitch
, bitmap
->pitch
);
1404 AlphaBlt(dst_A
, s
+ last_x
- xo
, 0, h
, new_x
-last_x
, overlayPitch
, bitmap
->pitch
);
1406 dst_A
+= new_x
- last_x
;
1407 dst_Y
+= new_x
- last_x
;
1408 dst_U
+= new_x
- last_x
;
1409 dst_V
+= new_x
- last_x
;
1414 case DM::SINGLE_COLOR
| 0*DM::SSE2
| DM::AYUV_PLANAR
:
1416 // char * debug_dst=(char*)dst;int h2 = h;
1417 // XY_DO_ONCE( xy_logger::write_file("G:\\b2_rt", (char*)&color, sizeof(color)) );
1418 // XY_DO_ONCE( xy_logger::write_file("G:\\b2_rt", debug_dst, (h2-1)*spd.pitch) );
1419 // debug_dst += spd.pitch*spd.h;
1420 // XY_DO_ONCE( xy_logger::write_file("G:\\b2_rt", debug_dst, (h2-1)*spd.pitch) );
1421 // debug_dst += spd.pitch*spd.h;
1422 // XY_DO_ONCE( xy_logger::write_file("G:\\b2_rt", debug_dst, (h2-1)*spd.pitch) );
1423 // debug_dst += spd.pitch*spd.h;
1424 // XY_DO_ONCE( xy_logger::write_file("G:\\b2_rt", debug_dst, (h2-1)*spd.pitch) );
1425 // debug_dst=(char*)dst;
1427 unsigned char* dst_A
= bitmap
->plans
[0] + dst_offset
;
1428 unsigned char* dst_Y
= bitmap
->plans
[1] + dst_offset
;
1429 unsigned char* dst_U
= bitmap
->plans
[2] + dst_offset
;
1430 unsigned char* dst_V
= bitmap
->plans
[3] + dst_offset
;
1433 for(int wt
=0; wt
<w
; ++wt
)
1435 DWORD temp
= COMBINE_AYUV(dst_A
[wt
], dst_Y
[wt
], dst_U
[wt
], dst_V
[wt
]);
1436 pixmix(&temp
, color
, s
[wt
]);
1437 SPLIT_AYUV(temp
, dst_A
+wt
, dst_Y
+wt
, dst_U
+wt
, dst_V
+wt
);
1440 dst_A
+= bitmap
->pitch
;
1441 dst_Y
+= bitmap
->pitch
;
1442 dst_U
+= bitmap
->pitch
;
1443 dst_V
+= bitmap
->pitch
;
1445 // XY_DO_ONCE( xy_logger::write_file("G:\\a2_rt", debug_dst, (h2-1)*spd.pitch) );
1446 // debug_dst += spd.pitch*spd.h;
1447 // XY_DO_ONCE( xy_logger::write_file("G:\\a2_rt", debug_dst, (h2-1)*spd.pitch) );
1448 // debug_dst += spd.pitch*spd.h;
1449 // XY_DO_ONCE( xy_logger::write_file("G:\\a2_rt", debug_dst, (h2-1)*spd.pitch) );
1450 // debug_dst += spd.pitch*spd.h;
1451 // XY_DO_ONCE( xy_logger::write_file("G:\\a2_rt", debug_dst, (h2-1)*spd.pitch) );
1454 case 0*DM::SINGLE_COLOR
| 0*DM::SSE2
| DM::AYUV_PLANAR
:
1456 unsigned char* dst_A
= bitmap
->plans
[0] + dst_offset
;
1457 unsigned char* dst_Y
= bitmap
->plans
[1] + dst_offset
;
1458 unsigned char* dst_U
= bitmap
->plans
[2] + dst_offset
;
1459 unsigned char* dst_V
= bitmap
->plans
[3] + dst_offset
;
1462 const DWORD
*sw
= switchpts
;
1463 for(int wt
=0; wt
<w
; ++wt
)
1465 if(wt
+xo
>= sw
[1]) {while(wt
+xo
>= sw
[1]) sw
+= 2; color
= sw
[-2];}
1466 DWORD temp
= COMBINE_AYUV(dst_A
[wt
], dst_Y
[wt
], dst_U
[wt
], dst_V
[wt
]);
1467 pixmix(&temp
, color
, (s
[wt
]*(color
>>24))>>8);
1468 SPLIT_AYUV(temp
, dst_A
+wt
, dst_Y
+wt
, dst_U
+wt
, dst_V
+wt
);
1471 dst_A
+= bitmap
->pitch
;
1472 dst_Y
+= bitmap
->pitch
;
1473 dst_U
+= bitmap
->pitch
;
1474 dst_V
+= bitmap
->pitch
;
1479 // Remember to EMMS!
1480 // Rendering fails in funny ways if we don't do this.
1485 void Rasterizer::FillSolidRect(SubPicDesc
& spd
, int x
, int y
, int nWidth
, int nHeight
, DWORD argb
)
1487 bool fSSE2
= !!(g_cpuid
.m_flags
& CCpuID::sse2
);
1488 bool AYUV_PLANAR
= (spd
.type
==MSP_AYUV_PLANAR
);
1489 int draw_method
= 0;
1491 draw_method
|= DM::SSE2
;
1493 draw_method
|= DM::AYUV_PLANAR
;
1495 switch (draw_method
)
1497 case DM::SSE2
| 0*DM::AYUV_PLANAR
:
1499 for (int wy
=y
; wy
<y
+nHeight
; wy
++) {
1500 DWORD
* dst
= (DWORD
*)((BYTE
*)spd
.bits
+ spd
.pitch
* wy
) + x
;
1501 for(int wt
=0; wt
<nWidth
; ++wt
) {
1502 pixmix_sse2(&dst
[wt
], argb
, argb
>>24);
1507 case 0*DM::SSE2
| 0*DM::AYUV_PLANAR
:
1509 for (int wy
=y
; wy
<y
+nHeight
; wy
++) {
1510 DWORD
* dst
= (DWORD
*)((BYTE
*)spd
.bits
+ spd
.pitch
* wy
) + x
;
1511 for(int wt
=0; wt
<nWidth
; ++wt
) {
1512 pixmix(&dst
[wt
], argb
, argb
>>24);
1517 case DM::SSE2
| DM::AYUV_PLANAR
:
1519 BYTE
* dst
= reinterpret_cast<BYTE
*>(spd
.bits
) + spd
.pitch
* y
+ x
;
1521 BYTE
* dst_Y
= dst_A
+ spd
.pitch
*spd
.h
;
1522 BYTE
* dst_U
= dst_Y
+ spd
.pitch
*spd
.h
;
1523 BYTE
* dst_V
= dst_U
+ spd
.pitch
*spd
.h
;
1524 AlphaBlt(dst_Y
, argb
>>24, ((argb
)>>16)&0xff, nHeight
, nWidth
, spd
.pitch
);
1525 AlphaBlt(dst_U
, argb
>>24, ((argb
)>>8)&0xff, nHeight
, nWidth
, spd
.pitch
);
1526 AlphaBlt(dst_V
, argb
>>24, ((argb
))&0xff, nHeight
, nWidth
, spd
.pitch
);
1527 AlphaBlt(dst_A
, argb
>>24, 0, nHeight
, nWidth
, spd
.pitch
);
1530 case 0*DM::SSE2
| DM::AYUV_PLANAR
:
1532 BYTE
* dst
= reinterpret_cast<BYTE
*>(spd
.bits
) + spd
.pitch
* y
+ x
;
1534 BYTE
* dst_Y
= dst_A
+ spd
.pitch
*spd
.h
;
1535 BYTE
* dst_U
= dst_Y
+ spd
.pitch
*spd
.h
;
1536 BYTE
* dst_V
= dst_U
+ spd
.pitch
*spd
.h
;
1537 AlphaBltC(dst_Y
, argb
>>24, ((argb
)>>16)&0xff, nHeight
, nWidth
, spd
.pitch
);
1538 AlphaBltC(dst_U
, argb
>>24, ((argb
)>>8)&0xff, nHeight
, nWidth
, spd
.pitch
);
1539 AlphaBltC(dst_V
, argb
>>24, ((argb
))&0xff, nHeight
, nWidth
, spd
.pitch
);
1540 AlphaBltC(dst_A
, argb
>>24, 0, nHeight
, nWidth
, spd
.pitch
);
1547 ///////////////////////////////////////////////////////////////
1551 void Overlay::_DoFillAlphaMash(byte
* outputAlphaMask
, const byte
* pBody
, const byte
* pBorder
, int x
, int y
, int w
, int h
,
1552 const byte
* pAlphaMask
, int pitch
, DWORD color_alpha
)
1554 pBody
= pBody
!=NULL
? pBody
+ y
*mOverlayPitch
+ x
: NULL
;
1555 pBorder
= pBorder
!=NULL
? pBorder
+ y
*mOverlayPitch
+ x
: NULL
;
1556 byte
* dst
= outputAlphaMask
+ y
*mOverlayPitch
+ x
;
1558 const int x0
= ((reinterpret_cast<int>(dst
)+3)&~3) - reinterpret_cast<int>(dst
) < w
?
1559 ((reinterpret_cast<int>(dst
)+3)&~3) - reinterpret_cast<int>(dst
) : w
; //IMPORTANT! Should not exceed w.
1560 const int x00
= ((reinterpret_cast<int>(dst
)+15)&~15) - reinterpret_cast<int>(dst
) < w
?
1561 ((reinterpret_cast<int>(dst
)+15)&~15) - reinterpret_cast<int>(dst
) : w
;//IMPORTANT! Should not exceed w.
1562 const int x_end00
= ((reinterpret_cast<int>(dst
)+w
)&~15) - reinterpret_cast<int>(dst
);
1563 const int x_end0
= ((reinterpret_cast<int>(dst
)+w
)&~3) - reinterpret_cast<int>(dst
);
1564 const int x_end
= w
;
1566 __m64 color_alpha_64
= _mm_set1_pi16(color_alpha
);
1567 __m128i color_alpha_128
= _mm_set1_epi16(color_alpha
);
1569 if(pAlphaMask
==NULL
&& pBody
!=NULL
&& pBorder
!=NULL
)
1574 mov eax, color_alpha
1576 punpcklwd XMM3, XMM3
1577 pshufd XMM3, XMM3, 0
1585 int temp
= pBorder
[j
]-pBody
[j
];
1586 temp
= temp
<0 ? 0 : temp
;
1587 dst
[j
] = (temp
* color_alpha
)>>6;
1591 __m64 border
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBorder
+j
));
1592 __m64 body
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBody
+j
));
1593 border
= _mm_subs_pu8(border
, body
);
1594 __m64 zero
= _mm_setzero_si64();
1595 border
= _mm_unpacklo_pi8(border
, zero
);
1596 border
= _mm_mullo_pi16(border
, color_alpha_64
);
1597 border
= _mm_srli_pi16(border
, 6);
1598 border
= _mm_packs_pu16(border
,border
);
1599 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(border
);
1601 __m128i zero
= _mm_setzero_si128();
1602 for( ;j
<x_end00
;j
+=16)
1604 __m128i border
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pBorder
+j
));
1605 __m128i body
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pBody
+j
));
1606 border
= _mm_subs_epu8(border
,body
);
1607 __m128i srchi
= border
;
1608 border
= _mm_unpacklo_epi8(border
, zero
);
1609 srchi
= _mm_unpackhi_epi8(srchi
, zero
);
1610 border
= _mm_mullo_epi16(border
, color_alpha_128
);
1611 srchi
= _mm_mullo_epi16(srchi
, color_alpha_128
);
1612 border
= _mm_srli_epi16(border
, 6);
1613 srchi
= _mm_srli_epi16(srchi
, 6);
1614 border
= _mm_packus_epi16(border
, srchi
);
1615 _mm_storeu_si128(reinterpret_cast<__m128i
*>(dst
+j
), border
);
1617 for( ;j
<x_end0
;j
+=4)
1619 __m64 border
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBorder
+j
));
1620 __m64 body
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBody
+j
));
1621 border
= _mm_subs_pu8(border
, body
);
1622 __m64 zero
= _mm_setzero_si64();
1623 border
= _mm_unpacklo_pi8(border
, zero
);
1624 border
= _mm_mullo_pi16(border
, color_alpha_64
);
1625 border
= _mm_srli_pi16(border
, 6);
1626 border
= _mm_packs_pu16(border
,border
);
1627 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(border
);
1631 int temp
= pBorder
[j
]-pBody
[j
];
1632 temp
= temp
<0 ? 0 : temp
;
1633 dst
[j
] = (temp
* color_alpha
)>>6;
1635 pBody
+= mOverlayPitch
;
1636 pBorder
+= mOverlayPitch
;
1637 //pAlphaMask += pitch;
1638 dst
+= mOverlayPitch
;
1641 else if( ((pBody
==NULL
) + (pBorder
==NULL
))==1 && pAlphaMask
==NULL
)
1643 const BYTE
* src1
= pBody
!=NULL
? pBody
: pBorder
;
1649 dst
[j
] = (src1
[j
] * color_alpha
)>>6;
1653 __m64 src
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(src1
+j
));
1654 __m64 zero
= _mm_setzero_si64();
1655 src
= _mm_unpacklo_pi8(src
, zero
);
1656 src
= _mm_mullo_pi16(src
, color_alpha_64
);
1657 src
= _mm_srli_pi16(src
, 6);
1658 src
= _mm_packs_pu16(src
,src
);
1659 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(src
);
1661 __m128i zero
= _mm_setzero_si128();
1662 for( ;j
<x_end00
;j
+=16)
1664 __m128i src
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(src1
+j
));
1665 __m128i srchi
= src
;
1666 src
= _mm_unpacklo_epi8(src
, zero
);
1667 srchi
= _mm_unpackhi_epi8(srchi
, zero
);
1668 src
= _mm_mullo_epi16(src
, color_alpha_128
);
1669 srchi
= _mm_mullo_epi16(srchi
, color_alpha_128
);
1670 src
= _mm_srli_epi16(src
, 6);
1671 srchi
= _mm_srli_epi16(srchi
, 6);
1672 src
= _mm_packus_epi16(src
, srchi
);
1673 _mm_storeu_si128(reinterpret_cast<__m128i
*>(dst
+j
), src
);
1675 for( ;j
<x_end0
;j
+=4)
1677 __m64 src
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(src1
+j
));
1678 __m64 zero
= _mm_setzero_si64();
1679 src
= _mm_unpacklo_pi8(src
, zero
);
1680 src
= _mm_mullo_pi16(src
, color_alpha_64
);
1681 src
= _mm_srli_pi16(src
, 6);
1682 src
= _mm_packs_pu16(src
,src
);
1683 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(src
);
1687 dst
[j
] = (src1
[j
] * color_alpha
)>>6;
1689 src1
+= mOverlayPitch
;
1690 //pAlphaMask += pitch;
1691 dst
+= mOverlayPitch
;
1694 else if( ((pBody
==NULL
) + (pBorder
==NULL
))==1 && pAlphaMask
!=NULL
)
1696 const BYTE
* src1
= pBody
!=NULL
? pBody
: pBorder
;
1702 dst
[j
] = (src1
[j
] * pAlphaMask
[j
] * color_alpha
)>>12;
1706 __m64 src
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(src1
+j
));
1707 __m64 mask
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pAlphaMask
+j
));
1708 __m64 zero
= _mm_setzero_si64();
1709 src
= _mm_unpacklo_pi8(src
, zero
);
1710 src
= _mm_mullo_pi16(src
, color_alpha_64
);
1711 mask
= _mm_unpacklo_pi8(zero
, mask
); //important!
1712 src
= _mm_mulhi_pi16(src
, mask
); //important!
1713 src
= _mm_srli_pi16(src
, 12+8-16); //important!
1714 src
= _mm_packs_pu16(src
,src
);
1715 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(src
);
1717 __m128i zero
= _mm_setzero_si128();
1718 for( ;j
<x_end00
;j
+=16)
1720 __m128i src
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(src1
+j
));
1721 __m128i mask
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pAlphaMask
+j
));
1722 __m128i srchi
= src
;
1723 __m128i maskhi
= mask
;
1724 src
= _mm_unpacklo_epi8(src
, zero
);
1725 srchi
= _mm_unpackhi_epi8(srchi
, zero
);
1726 mask
= _mm_unpacklo_epi8(zero
, mask
); //important!
1727 maskhi
= _mm_unpackhi_epi8(zero
, maskhi
);
1728 src
= _mm_mullo_epi16(src
, color_alpha_128
);
1729 srchi
= _mm_mullo_epi16(srchi
, color_alpha_128
);
1730 src
= _mm_mulhi_epu16(src
, mask
); //important!
1731 srchi
= _mm_mulhi_epu16(srchi
, maskhi
);
1732 src
= _mm_srli_epi16(src
, 12+8-16); //important!
1733 srchi
= _mm_srli_epi16(srchi
, 12+8-16);
1734 src
= _mm_packus_epi16(src
, srchi
);
1735 _mm_storeu_si128(reinterpret_cast<__m128i
*>(dst
+j
), src
);
1737 for( ;j
<x_end0
;j
+=4)
1739 __m64 src
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(src1
+j
));
1740 __m64 mask
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pAlphaMask
+j
));
1741 __m64 zero
= _mm_setzero_si64();
1742 src
= _mm_unpacklo_pi8(src
, zero
);
1743 src
= _mm_mullo_pi16(src
, color_alpha_64
);
1744 mask
= _mm_unpacklo_pi8(zero
, mask
); //important!
1745 src
= _mm_mulhi_pi16(src
, mask
); //important!
1746 src
= _mm_srli_pi16(src
, 12+8-16); //important!
1747 src
= _mm_packs_pu16(src
,src
);
1748 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(src
);
1752 dst
[j
] = (src1
[j
] * pAlphaMask
[j
] * color_alpha
)>>12;
1754 src1
+= mOverlayPitch
;
1755 pAlphaMask
+= pitch
;
1756 dst
+= mOverlayPitch
;
1759 else if( pAlphaMask
!=NULL
&& pBody
!=NULL
&& pBorder
!=NULL
)
1766 int temp
= pBorder
[j
]-pBody
[j
];
1767 temp
= temp
<0 ? 0 : temp
;
1768 dst
[j
] = (temp
* pAlphaMask
[j
] * color_alpha
)>>12;
1772 __m64 border
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBorder
+j
));
1773 __m64 body
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBody
+j
));
1774 border
= _mm_subs_pu8(border
, body
);
1775 __m64 mask
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pAlphaMask
+j
));
1776 __m64 zero
= _mm_setzero_si64();
1777 border
= _mm_unpacklo_pi8(border
, zero
);
1778 border
= _mm_mullo_pi16(border
, color_alpha_64
);
1779 mask
= _mm_unpacklo_pi8(zero
, mask
); //important!
1780 border
= _mm_mulhi_pi16(border
, mask
); //important!
1781 border
= _mm_srli_pi16(border
, 12+8-16); //important!
1782 border
= _mm_packs_pu16(border
,border
);
1783 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(border
);
1785 __m128i zero
= _mm_setzero_si128();
1786 for( ;j
<x_end00
;j
+=16)
1788 __m128i border
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pBorder
+j
));
1789 __m128i body
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pBody
+j
));
1790 border
= _mm_subs_epu8(border
,body
);
1792 __m128i mask
= _mm_loadu_si128(reinterpret_cast<const __m128i
*>(pAlphaMask
+j
));
1793 __m128i srchi
= border
;
1794 __m128i maskhi
= mask
;
1795 border
= _mm_unpacklo_epi8(border
, zero
);
1796 srchi
= _mm_unpackhi_epi8(srchi
, zero
);
1797 mask
= _mm_unpacklo_epi8(zero
, mask
); //important!
1798 maskhi
= _mm_unpackhi_epi8(zero
, maskhi
);
1799 border
= _mm_mullo_epi16(border
, color_alpha_128
);
1800 srchi
= _mm_mullo_epi16(srchi
, color_alpha_128
);
1801 border
= _mm_mulhi_epu16(border
, mask
); //important!
1802 srchi
= _mm_mulhi_epu16(srchi
, maskhi
);
1803 border
= _mm_srli_epi16(border
, 12+8-16); //important!
1804 srchi
= _mm_srli_epi16(srchi
, 12+8-16);
1805 border
= _mm_packus_epi16(border
, srchi
);
1806 _mm_storeu_si128(reinterpret_cast<__m128i
*>(dst
+j
), border
);
1808 for( ;j
<x_end0
;j
+=4)
1810 __m64 border
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBorder
+j
));
1811 __m64 body
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pBody
+j
));
1812 border
= _mm_subs_pu8(border
, body
);
1813 __m64 mask
= _mm_cvtsi32_si64(*reinterpret_cast<const int*>(pAlphaMask
+j
));
1814 __m64 zero
= _mm_setzero_si64();
1815 border
= _mm_unpacklo_pi8(border
, zero
);
1816 border
= _mm_mullo_pi16(border
, color_alpha_64
);
1817 mask
= _mm_unpacklo_pi8(zero
, mask
); //important!
1818 border
= _mm_mulhi_pi16(border
, mask
); //important!
1819 border
= _mm_srli_pi16(border
, 12+8-16); //important!
1820 border
= _mm_packs_pu16(border
,border
);
1821 *reinterpret_cast<int*>(dst
+j
) = _mm_cvtsi64_si32(border
);
1825 int temp
= pBorder
[j
]-pBody
[j
];
1826 temp
= temp
<0 ? 0 : temp
;
1827 dst
[j
] = (temp
* pAlphaMask
[j
] * color_alpha
)>>12;
1829 pBody
+= mOverlayPitch
;
1830 pBorder
+= mOverlayPitch
;
1831 pAlphaMask
+= pitch
;
1832 dst
+= mOverlayPitch
;
1837 //should NOT happen!
1841 for(int j
=0;j
<x_end
;j
++)
1845 dst
+= mOverlayPitch
;
1850 void Overlay::FillAlphaMash( byte
* outputAlphaMask
, bool fBody
, bool fBorder
, int x
, int y
, int w
, int h
, const byte
* pAlphaMask
, int pitch
, DWORD color_alpha
)
1852 if(!fBorder
&& fBody
&& pAlphaMask
==NULL
)
1854 _DoFillAlphaMash(outputAlphaMask
, mBody
.get(), NULL
, x
, y
, w
, h
, pAlphaMask
, pitch
, color_alpha
);
1856 else if(/*fBorder &&*/ fBody
&& pAlphaMask
==NULL
)
1858 _DoFillAlphaMash(outputAlphaMask
, NULL
, mBorder
.get(), x
, y
, w
, h
, pAlphaMask
, pitch
, color_alpha
);
1860 else if(!fBody
&& fBorder
/* pAlphaMask==NULL or not*/)
1862 _DoFillAlphaMash(outputAlphaMask
, mBody
.get(), mBorder
.get(), x
, y
, w
, h
, pAlphaMask
, pitch
, color_alpha
);
1864 else if(!fBorder
&& fBody
&& pAlphaMask
!=NULL
)
1866 _DoFillAlphaMash(outputAlphaMask
, mBody
.get(), NULL
, x
, y
, w
, h
, pAlphaMask
, pitch
, color_alpha
);
1868 else if(fBorder
&& fBody
&& pAlphaMask
!=NULL
)
1870 _DoFillAlphaMash(outputAlphaMask
, NULL
, mBorder
.get(), x
, y
, w
, h
, pAlphaMask
, pitch
, color_alpha
);
1879 Overlay
* Overlay::GetSubpixelVariance(unsigned int xshift
, unsigned int yshift
)
1881 Overlay
* overlay
= new Overlay();
1889 overlay
->mOffsetX
= mOffsetX
- xshift
;
1890 overlay
->mOffsetY
= mOffsetY
- yshift
;
1891 overlay
->mWidth
= mWidth
+ xshift
;
1892 overlay
->mHeight
= mHeight
+ yshift
;
1894 overlay
->mOverlayWidth
= ((overlay
->mWidth
+7)>>3) + 1;
1895 overlay
->mOverlayHeight
= ((overlay
->mHeight
+ 7)>>3) + 1;
1896 overlay
->mOverlayPitch
= (overlay
->mOverlayWidth
+15)&~15;
1899 overlay
->mfWideOutlineEmpty
= mfWideOutlineEmpty
;
1901 if (overlay
->mOverlayPitch
* overlay
->mOverlayHeight
<=0)
1906 BYTE
* body
= reinterpret_cast<BYTE
*>(xy_malloc(overlay
->mOverlayPitch
* overlay
->mOverlayHeight
));
1911 overlay
->mBody
.reset(body
, xy_free
);
1912 BYTE
* border
= NULL
;
1913 if (!overlay
->mfWideOutlineEmpty
)
1915 border
= reinterpret_cast<BYTE
*>(xy_malloc(overlay
->mOverlayPitch
* overlay
->mOverlayHeight
));
1920 overlay
->mBorder
.reset(border
, xy_free
);
1923 if(overlay
->mOverlayPitch
==mOverlayPitch
&& overlay
->mOverlayHeight
>=mOverlayHeight
)
1927 memcpy(body
, mBody
.get(), mOverlayPitch
* mOverlayHeight
);
1928 memset(body
+mOverlayPitch
*mOverlayHeight
, 0, mOverlayPitch
* (overlay
->mOverlayHeight
-mOverlayHeight
));
1930 else if ( (!!body
)!=(!!mBody
)/*==NULL*/)
1935 if (border
&& mBorder
)
1937 memcpy(border
, mBorder
.get(), mOverlayPitch
* mOverlayHeight
);
1938 memset(border
+mOverlayPitch
*mOverlayHeight
, 0, mOverlayPitch
* (overlay
->mOverlayHeight
-mOverlayHeight
));
1940 else if ( (!!border
)!=(!!mBorder
)/*==NULL*/ )
1947 memset(body
, 0, overlay
->mOverlayPitch
* overlay
->mOverlayHeight
);
1949 const byte
* src
= mBody
.get();
1950 for (int i
=0;i
<mOverlayHeight
;i
++)
1952 memcpy(dst
, src
, mOverlayPitch
);
1953 dst
+= overlay
->mOverlayPitch
;
1954 src
+= mOverlayPitch
;
1956 if (!overlay
->mfWideOutlineEmpty
)
1958 ASSERT(border
&& mBorder
);
1959 memset(border
, 0, overlay
->mOverlayPitch
* overlay
->mOverlayHeight
);
1961 src
= mBorder
.get();
1962 for (int i
=0;i
<mOverlayHeight
;i
++)
1964 memcpy(dst
, src
, mOverlayPitch
);
1965 dst
+= overlay
->mOverlayPitch
;
1966 src
+= mOverlayPitch
;
1971 // Bilinear(overlay->mpOverlayBuffer.base, overlay->mOverlayWidth, 2*overlay->mOverlayHeight, overlay->mOverlayPitch, xshift, yshift);
1972 Bilinear(body
, overlay
->mOverlayWidth
, overlay
->mOverlayHeight
, overlay
->mOverlayPitch
, xshift
, yshift
);
1973 if (!overlay
->mfWideOutlineEmpty
)
1975 Bilinear(border
, overlay
->mOverlayWidth
, overlay
->mOverlayHeight
, overlay
->mOverlayPitch
, xshift
, yshift
);
1980 ///////////////////////////////////////////////////////////////
1984 PathData::PathData():mpPathTypes(NULL
), mpPathPoints(NULL
), mPathPoints(0)
1988 PathData::PathData( const PathData
& src
):mpPathTypes(NULL
), mpPathPoints(NULL
), mPathPoints(src
.mPathPoints
)
1990 //TODO: deal with the case that src.mPathPoints<0
1993 mpPathTypes
= static_cast<BYTE
*>(malloc(mPathPoints
* sizeof(BYTE
)));
1994 mpPathPoints
= static_cast<POINT
*>(malloc(mPathPoints
* sizeof(POINT
)));
1998 memcpy(mpPathTypes
, src
.mpPathTypes
, mPathPoints
*sizeof(BYTE
));
1999 memcpy(mpPathPoints
, src
.mpPathPoints
, mPathPoints
*sizeof(POINT
));
2003 const PathData
& PathData::operator=( const PathData
& src
)
2007 if(mPathPoints
!=src
.mPathPoints
&& src
.mPathPoints
>0)
2010 mPathPoints
= src
.mPathPoints
;
2011 mpPathTypes
= static_cast<BYTE
*>(malloc(mPathPoints
* sizeof(BYTE
)));
2012 mpPathPoints
= static_cast<POINT
*>(malloc(mPathPoints
* sizeof(POINT
)));//better than realloc
2014 if(src
.mPathPoints
>0)
2016 memcpy(mpPathTypes
, src
.mpPathTypes
, mPathPoints
*sizeof(BYTE
));
2017 memcpy(mpPathPoints
, src
.mpPathPoints
, mPathPoints
*sizeof(POINT
));
2023 PathData::~PathData()
2028 bool PathData::operator==( const PathData
& rhs
) const
2030 return (this==&rhs
) || (
2031 mPathPoints
==rhs
.mPathPoints
2032 && !memcmp(mpPathTypes
, rhs
.mpPathTypes
, mPathPoints
* sizeof(BYTE
) )
2033 && !memcmp(mpPathPoints
, rhs
.mpPathPoints
, mPathPoints
* sizeof(POINT
) )
2037 void PathData::_TrashPath()
2047 mpPathPoints
= NULL
;
2052 bool PathData::BeginPath(HDC hdc
)
2055 return !!::BeginPath(hdc
);
2058 bool PathData::EndPath(HDC hdc
)
2063 mPathPoints
= GetPath(hdc
, NULL
, NULL
, 0);
2066 mpPathTypes
= (BYTE
*)malloc(sizeof(BYTE
) * mPathPoints
);
2067 mpPathPoints
= (POINT
*)malloc(sizeof(POINT
) * mPathPoints
);
2068 if(mPathPoints
== GetPath(hdc
, mpPathPoints
, mpPathTypes
, mPathPoints
))
2075 bool PathData::PartialBeginPath(HDC hdc
, bool bClearPath
)
2079 return !!::BeginPath(hdc
);
2082 bool PathData::PartialEndPath(HDC hdc
, long dx
, long dy
)
2090 nPoints
= GetPath(hdc
, NULL
, NULL
, 0);
2093 pNewTypes
= (BYTE
*)realloc(mpPathTypes
, (mPathPoints
+ nPoints
) * sizeof(BYTE
));
2094 pNewPoints
= (POINT
*)realloc(mpPathPoints
, (mPathPoints
+ nPoints
) * sizeof(POINT
));
2096 mpPathTypes
= pNewTypes
;
2098 mpPathPoints
= pNewPoints
;
2099 BYTE
* pTypes
= new BYTE
[nPoints
];
2100 POINT
* pPoints
= new POINT
[nPoints
];
2101 if(pNewTypes
&& pNewPoints
&& nPoints
== GetPath(hdc
, pPoints
, pTypes
, nPoints
))
2103 for(int i
= 0; i
< nPoints
; ++i
)
2105 mpPathPoints
[mPathPoints
+ i
].x
= pPoints
[i
].x
+ dx
;
2106 mpPathPoints
[mPathPoints
+ i
].y
= pPoints
[i
].y
+ dy
;
2107 mpPathTypes
[mPathPoints
+ i
] = pTypes
[i
];
2109 mPathPoints
+= nPoints
;
2123 void PathData::AlignLeftTop(CPoint
*left_top
, CSize
*size
)
2129 for(int i
=0; i
<mPathPoints
; ++i
)
2131 int ix
= mpPathPoints
[i
].x
;
2132 int iy
= mpPathPoints
[i
].y
;
2133 if(ix
< minx
) minx
= ix
;
2134 if(ix
> maxx
) maxx
= ix
;
2135 if(iy
< miny
) miny
= iy
;
2136 if(iy
> maxy
) maxy
= iy
;
2138 if(minx
> maxx
|| miny
> maxy
)
2141 *left_top
= CPoint(0, 0);
2142 *size
= CSize(0, 0);
2145 minx
= (minx
>> 3) & ~7;
2146 miny
= (miny
>> 3) & ~7;
2147 maxx
= (maxx
+ 7) >> 3;
2148 maxy
= (maxy
+ 7) >> 3;
2149 for(int i
=0; i
<mPathPoints
; ++i
)
2151 mpPathPoints
[i
].x
-= minx
*8;
2152 mpPathPoints
[i
].y
-= miny
*8;
2154 *left_top
= CPoint(minx
, miny
);
2155 *size
= CSize(maxx
+1-minx
, maxy
+1-miny
);
2159 //////////////////////////////////////////////////////////////////////////
2163 ScanLineData::ScanLineData()
2167 ScanLineData::~ScanLineData()
2171 void ScanLineData::_ReallocEdgeBuffer(int edges
)
2173 mEdgeHeapSize
= edges
;
2174 mpEdgeBuffer
= (Edge
*)realloc(mpEdgeBuffer
, sizeof(Edge
)*edges
);
2177 void ScanLineData::_EvaluateBezier(const PathData
& path_data
, int ptbase
, bool fBSpline
)
2179 const POINT
* pt0
= path_data
.mpPathPoints
+ ptbase
;
2180 const POINT
* pt1
= path_data
.mpPathPoints
+ ptbase
+ 1;
2181 const POINT
* pt2
= path_data
.mpPathPoints
+ ptbase
+ 2;
2182 const POINT
* pt3
= path_data
.mpPathPoints
+ ptbase
+ 3;
2191 double cx3
, cx2
, cx1
, cx0
, cy3
, cy2
, cy1
, cy0
;
2198 double _1div6
= 1.0/6.0;
2199 cx3
= _1div6
*(- x0
+3*x1
-3*x2
+x3
);
2200 cx2
= _1div6
*( 3*x0
-6*x1
+3*x2
);
2201 cx1
= _1div6
*(-3*x0
+3*x2
);
2202 cx0
= _1div6
*( x0
+4*x1
+1*x2
);
2203 cy3
= _1div6
*(- y0
+3*y1
-3*y2
+y3
);
2204 cy2
= _1div6
*( 3*y0
-6*y1
+3*y2
);
2205 cy1
= _1div6
*(-3*y0
+3*y2
);
2206 cy0
= _1div6
*( y0
+4*y1
+1*y2
);
2214 cx3
= - x0
+3*x1
-3*x2
+x3
;
2215 cx2
= 3*x0
-6*x1
+3*x2
;
2218 cy3
= - y0
+3*y1
-3*y2
+y3
;
2219 cy2
= 3*y0
-6*y1
+3*y2
;
2224 // This equation is from Graphics Gems I.
2226 // The idea is that since we're approximating a cubic curve with lines,
2227 // any error we incur is due to the curvature of the line, which we can
2228 // estimate by calculating the maximum acceleration of the curve. For
2229 // a cubic, the acceleration (second derivative) is a line, meaning that
2230 // the absolute maximum acceleration must occur at either the beginning
2231 // (|c2|) or the end (|c2+c3|). Our bounds here are a little more
2232 // conservative than that, but that's okay.
2234 // If the acceleration of the parametric formula is zero (c2 = c3 = 0),
2235 // that component of the curve is linear and does not incur any error.
2236 // If a=0 for both X and Y, the curve is a line segment and we can
2237 // use a step size of 1.
2238 double maxaccel1
= fabs(2*cy2
) + fabs(6*cy3
);
2239 double maxaccel2
= fabs(2*cx2
) + fabs(6*cx3
);
2240 double maxaccel
= maxaccel1
> maxaccel2
? maxaccel1
: maxaccel2
;
2242 if(maxaccel
> 8.0) h
= sqrt(8.0 / maxaccel
);
2243 if(!fFirstSet
) {firstp
.x
= (LONG
)cx0
; firstp
.y
= (LONG
)cy0
; lastp
= firstp
; fFirstSet
= true;}
2244 for(double t
= 0; t
< 1.0; t
+= h
)
2246 double x
= cx0
+ t
*(cx1
+ t
*(cx2
+ t
*cx3
));
2247 double y
= cy0
+ t
*(cy1
+ t
*(cy2
+ t
*cy3
));
2248 _EvaluateLine(lastp
.x
, lastp
.y
, (int)x
, (int)y
);
2250 double x
= cx0
+ cx1
+ cx2
+ cx3
;
2251 double y
= cy0
+ cy1
+ cy2
+ cy3
;
2252 _EvaluateLine(lastp
.x
, lastp
.y
, (int)x
, (int)y
);
2255 void ScanLineData::_EvaluateLine(const PathData
& path_data
, int pt1idx
, int pt2idx
)
2257 const POINT
* pt1
= path_data
.mpPathPoints
+ pt1idx
;
2258 const POINT
* pt2
= path_data
.mpPathPoints
+ pt2idx
;
2259 _EvaluateLine(pt1
->x
, pt1
->y
, pt2
->x
, pt2
->y
);
2262 void ScanLineData::_EvaluateLine(int x0
, int y0
, int x1
, int y1
)
2264 if(lastp
.x
!= x0
|| lastp
.y
!= y0
)
2266 _EvaluateLine(lastp
.x
, lastp
.y
, x0
, y0
);
2268 if(!fFirstSet
) {firstp
.x
= x0
; firstp
.y
= y0
; fFirstSet
= true;}
2273 __int64 xacc
= (__int64
)x0
<< 13;
2276 int y
= ((y0
+ 3)&~7) + 4;
2281 __int64 invslope
= (__int64(x1
- x0
) << 16) / dy
;
2282 while(mEdgeNext
+ y1
+ 1 - iy
> mEdgeHeapSize
)
2283 _ReallocEdgeBuffer(mEdgeHeapSize
*2);
2284 xacc
+= (invslope
* (y
- y0
)) >> 3;
2287 int ix
= (int)((xacc
+ 32768) >> 16);
2288 mpEdgeBuffer
[mEdgeNext
].next
= mpScanBuffer
[iy
];
2289 mpEdgeBuffer
[mEdgeNext
].posandflag
= ix
*2 + 1;
2290 mpScanBuffer
[iy
] = mEdgeNext
++;
2296 else if(y1
< y0
) // up
2298 __int64 xacc
= (__int64
)x1
<< 13;
2301 int y
= ((y1
+ 3)&~7) + 4;
2306 __int64 invslope
= (__int64(x0
- x1
) << 16) / dy
;
2307 while(mEdgeNext
+ y0
+ 1 - iy
> mEdgeHeapSize
)
2308 _ReallocEdgeBuffer(mEdgeHeapSize
*2);
2309 xacc
+= (invslope
* (y
- y1
)) >> 3;
2312 int ix
= (int)((xacc
+ 32768) >> 16);
2313 mpEdgeBuffer
[mEdgeNext
].next
= mpScanBuffer
[iy
];
2314 mpEdgeBuffer
[mEdgeNext
].posandflag
= ix
*2;
2315 mpScanBuffer
[iy
] = mEdgeNext
++;
2323 bool ScanLineData::ScanConvert(const PathData
& path_data
, const CSize
& size
)
2325 int lastmoveto
= -1;
2327 // Drop any outlines we may have.
2329 // Determine bounding box
2330 if(!path_data
.mPathPoints
)
2332 mWidth
= mHeight
= 0;
2337 // Initialize edge buffer. We use edge 0 as a sentinel.
2339 mEdgeHeapSize
= 2048;
2340 mpEdgeBuffer
= (Edge
*)malloc(sizeof(Edge
)*mEdgeHeapSize
);
2341 // Initialize scanline list.
2342 mpScanBuffer
= new unsigned int[mHeight
];
2343 memset(mpScanBuffer
, 0, mHeight
*sizeof(unsigned int));
2344 // Scan convert the outline. Yuck, Bezier curves....
2345 // Unfortunately, Windows 95/98 GDI has a bad habit of giving us text
2346 // paths with all but the first figure left open, so we can't rely
2347 // on the PT_CLOSEFIGURE flag being used appropriately.
2349 firstp
.x
= firstp
.y
= 0;
2350 lastp
.x
= lastp
.y
= 0;
2351 for(i
=0; i
<path_data
.mPathPoints
; ++i
)
2353 BYTE t
= path_data
.mpPathTypes
[i
] & ~PT_CLOSEFIGURE
;
2357 if(lastmoveto
>= 0 && firstp
!= lastp
)
2358 _EvaluateLine(lastp
.x
, lastp
.y
, firstp
.x
, firstp
.y
);
2361 lastp
= path_data
.mpPathPoints
[i
];
2366 if(path_data
.mPathPoints
- (i
-1) >= 2) _EvaluateLine(path_data
, i
-1, i
);
2369 if(path_data
.mPathPoints
- (i
-1) >= 4) _EvaluateBezier(path_data
, i
-1, false);
2373 if(path_data
.mPathPoints
- (i
-1) >= 4) _EvaluateBezier(path_data
, i
-1, true);
2376 case PT_BSPLINEPATCHTO
:
2377 if(path_data
.mPathPoints
- (i
-3) >= 4) _EvaluateBezier(path_data
, i
-3, true);
2381 if(lastmoveto
>= 0 && firstp
!= lastp
)
2382 _EvaluateLine(lastp
.x
, lastp
.y
, firstp
.x
, firstp
.y
);
2383 // Convert the edges to spans. We couldn't do this before because some of
2384 // the regions may have winding numbers >+1 and it would have been a pain
2385 // to try to adjust the spans on the fly. We use one heap to detangle
2386 // a scanline's worth of edges from the singly-linked lists, and another
2387 // to collect the actual scans.
2388 std::vector
<int> heap
;
2389 mOutline
.reserve(mEdgeNext
/ 2);
2391 for(y
=0; y
<mHeight
; ++y
)
2394 // Detangle scanline into edge heap.
2395 for(unsigned ptr
= (unsigned)(mpScanBuffer
[y
]&0xffffffff); ptr
; ptr
= mpEdgeBuffer
[ptr
].next
)
2397 heap
.push_back(mpEdgeBuffer
[ptr
].posandflag
);
2399 // Sort edge heap. Note that we conveniently made the opening edges
2400 // one more than closing edges at the same spot, so we won't have any
2401 // problems with abutting spans.
2402 std::sort(heap
.begin(), heap
.end()/*begin() + heap.size()*/);
2403 // Process edges and add spans. Since we only check for a non-zero
2404 // winding number, it doesn't matter which way the outlines go!
2405 std::vector
<int>::iterator itX1
= heap
.begin();
2406 std::vector
<int>::iterator itX2
= heap
.end(); // begin() + heap.size();
2408 for(; itX1
!= itX2
; ++itX1
)
2421 mOutline
.push_back(std::pair
<__int64
,__int64
>((y
<<32)+x1
+0x4000000040000000i
64, (y
<<32)+x2
+0x4000000040000000i
64)); // G: damn Avery, this is evil! :)
2426 // Dump the edge and scan buffers, since we no longer need them.
2428 delete [] mpScanBuffer
;
2433 using namespace std
;
2435 void ScanLineData::DeleteOutlines()
2440 bool ScanLineData2::CreateWidenedRegion(int rx
, int ry
)
2444 mWideBorder
= max(rx
,ry
);
2445 mWideOutline
.clear();
2447 const tSpanBuffer
& out_line
= m_scan_line_data
->mOutline
;
2450 // Do a half circle.
2451 // _OverlapRegion mirrors this so both halves are done.
2452 for(int y
= -ry
; y
<= ry
; ++y
)
2454 int x
= (int)(0.5 + sqrt(float(ry
*ry
- y
*y
)) * float(rx
)/float(ry
));
2455 OverlapRegion(mWideOutline
, out_line
, x
, y
);
2458 else if (ry
== 0 && rx
> 0)
2460 // There are artifacts if we don't make at least two overlaps of the line, even at same Y coord
2461 OverlapRegion(mWideOutline
, out_line
, rx
, 0);
2462 OverlapRegion(mWideOutline
, out_line
, rx
, 0);