• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /****************************************************************************
2  *
3  * ftcalc.c
4  *
5  *   Arithmetic computations (body).
6  *
7  * Copyright (C) 1996-2020 by
8  * David Turner, Robert Wilhelm, and Werner Lemberg.
9  *
10  * This file is part of the FreeType project, and may only be used,
11  * modified, and distributed under the terms of the FreeType project
12  * license, LICENSE.TXT.  By continuing to use, modify, or distribute
13  * this file you indicate that you have read the license and
14  * understand and accept it fully.
15  *
16  */
17 
18   /**************************************************************************
19    *
20    * Support for 1-complement arithmetic has been totally dropped in this
21    * release.  You can still write your own code if you need it.
22    *
23    */
24 
25   /**************************************************************************
26    *
27    * Implementing basic computation routines.
28    *
29    * FT_MulDiv(), FT_MulFix(), FT_DivFix(), FT_RoundFix(), FT_CeilFix(),
30    * and FT_FloorFix() are declared in freetype.h.
31    *
32    */
33 
34 
35 #include <freetype/ftglyph.h>
36 #include <freetype/fttrigon.h>
37 #include <freetype/internal/ftcalc.h>
38 #include <freetype/internal/ftdebug.h>
39 #include <freetype/internal/ftobjs.h>
40 
41 
42 #ifdef FT_MULFIX_ASSEMBLER
43 #undef FT_MulFix
44 #endif
45 
46 /* we need to emulate a 64-bit data type if a real one isn't available */
47 
48 #ifndef FT_LONG64
49 
50   typedef struct  FT_Int64_
51   {
52     FT_UInt32  lo;
53     FT_UInt32  hi;
54 
55   } FT_Int64;
56 
57 #endif /* !FT_LONG64 */
58 
59 
60   /**************************************************************************
61    *
62    * The macro FT_COMPONENT is used in trace mode.  It is an implicit
63    * parameter of the FT_TRACE() and FT_ERROR() macros, used to print/log
64    * messages during execution.
65    */
66 #undef  FT_COMPONENT
67 #define FT_COMPONENT  calc
68 
69 
70   /* transfer sign, leaving a positive number;                        */
71   /* we need an unsigned value to safely negate INT_MIN (or LONG_MIN) */
72 #define FT_MOVE_SIGN( x, x_unsigned, s ) \
73   FT_BEGIN_STMNT                         \
74     if ( x < 0 )                         \
75     {                                    \
76       x_unsigned = 0U - (x_unsigned);    \
77       s          = -s;                   \
78     }                                    \
79   FT_END_STMNT
80 
81   /* The following three functions are available regardless of whether */
82   /* FT_LONG64 is defined.                                             */
83 
84   /* documentation is in freetype.h */
85 
86   FT_EXPORT_DEF( FT_Fixed )
FT_RoundFix(FT_Fixed a)87   FT_RoundFix( FT_Fixed  a )
88   {
89     return ( ADD_LONG( a, 0x8000L - ( a < 0 ) ) ) & ~0xFFFFL;
90   }
91 
92 
93   /* documentation is in freetype.h */
94 
95   FT_EXPORT_DEF( FT_Fixed )
FT_CeilFix(FT_Fixed a)96   FT_CeilFix( FT_Fixed  a )
97   {
98     return ( ADD_LONG( a, 0xFFFFL ) ) & ~0xFFFFL;
99   }
100 
101 
102   /* documentation is in freetype.h */
103 
104   FT_EXPORT_DEF( FT_Fixed )
FT_FloorFix(FT_Fixed a)105   FT_FloorFix( FT_Fixed  a )
106   {
107     return a & ~0xFFFFL;
108   }
109 
110 #ifndef FT_MSB
111 
112   FT_BASE_DEF ( FT_Int )
FT_MSB(FT_UInt32 z)113   FT_MSB( FT_UInt32 z )
114   {
115     FT_Int  shift = 0;
116 
117 
118     /* determine msb bit index in `shift' */
119     if ( z & 0xFFFF0000UL )
120     {
121       z     >>= 16;
122       shift  += 16;
123     }
124     if ( z & 0x0000FF00UL )
125     {
126       z     >>= 8;
127       shift  += 8;
128     }
129     if ( z & 0x000000F0UL )
130     {
131       z     >>= 4;
132       shift  += 4;
133     }
134     if ( z & 0x0000000CUL )
135     {
136       z     >>= 2;
137       shift  += 2;
138     }
139     if ( z & 0x00000002UL )
140     {
141    /* z     >>= 1; */
142       shift  += 1;
143     }
144 
145     return shift;
146   }
147 
148 #endif /* !FT_MSB */
149 
150 
151   /* documentation is in ftcalc.h */
152 
153   FT_BASE_DEF( FT_Fixed )
FT_Hypot(FT_Fixed x,FT_Fixed y)154   FT_Hypot( FT_Fixed  x,
155             FT_Fixed  y )
156   {
157     FT_Vector  v;
158 
159 
160     v.x = x;
161     v.y = y;
162 
163     return FT_Vector_Length( &v );
164   }
165 
166 
167 #ifdef FT_LONG64
168 
169 
170   /* documentation is in freetype.h */
171 
172   FT_EXPORT_DEF( FT_Long )
FT_MulDiv(FT_Long a_,FT_Long b_,FT_Long c_)173   FT_MulDiv( FT_Long  a_,
174              FT_Long  b_,
175              FT_Long  c_ )
176   {
177     FT_Int     s = 1;
178     FT_UInt64  a, b, c, d;
179     FT_Long    d_;
180 
181 
182     a = (FT_UInt64)a_;
183     b = (FT_UInt64)b_;
184     c = (FT_UInt64)c_;
185 
186     FT_MOVE_SIGN( a_, a, s );
187     FT_MOVE_SIGN( b_, b, s );
188     FT_MOVE_SIGN( c_, c, s );
189 
190     d = c > 0 ? ( a * b + ( c >> 1 ) ) / c
191               : 0x7FFFFFFFUL;
192 
193     d_ = (FT_Long)d;
194 
195     return s < 0 ? NEG_LONG( d_ ) : d_;
196   }
197 
198 
199   /* documentation is in ftcalc.h */
200 
201   FT_BASE_DEF( FT_Long )
FT_MulDiv_No_Round(FT_Long a_,FT_Long b_,FT_Long c_)202   FT_MulDiv_No_Round( FT_Long  a_,
203                       FT_Long  b_,
204                       FT_Long  c_ )
205   {
206     FT_Int     s = 1;
207     FT_UInt64  a, b, c, d;
208     FT_Long    d_;
209 
210 
211     a = (FT_UInt64)a_;
212     b = (FT_UInt64)b_;
213     c = (FT_UInt64)c_;
214 
215     FT_MOVE_SIGN( a_, a, s );
216     FT_MOVE_SIGN( b_, b, s );
217     FT_MOVE_SIGN( c_, c, s );
218 
219     d = c > 0 ? a * b / c
220               : 0x7FFFFFFFUL;
221 
222     d_ = (FT_Long)d;
223 
224     return s < 0 ? NEG_LONG( d_ ) : d_;
225   }
226 
227 
228   /* documentation is in freetype.h */
229 
230   FT_EXPORT_DEF( FT_Long )
FT_MulFix(FT_Long a_,FT_Long b_)231   FT_MulFix( FT_Long  a_,
232              FT_Long  b_ )
233   {
234 #ifdef FT_MULFIX_ASSEMBLER
235 
236     return FT_MULFIX_ASSEMBLER( (FT_Int32)a_, (FT_Int32)b_ );
237 
238 #else
239 
240     FT_Int64  ab = (FT_Int64)a_ * (FT_Int64)b_;
241 
242     /* this requires arithmetic right shift of signed numbers */
243     return (FT_Long)( ( ab + 0x8000L - ( ab < 0 ) ) >> 16 );
244 
245 #endif /* FT_MULFIX_ASSEMBLER */
246   }
247 
248 
249   /* documentation is in freetype.h */
250 
251   FT_EXPORT_DEF( FT_Long )
FT_DivFix(FT_Long a_,FT_Long b_)252   FT_DivFix( FT_Long  a_,
253              FT_Long  b_ )
254   {
255     FT_Int     s = 1;
256     FT_UInt64  a, b, q;
257     FT_Long    q_;
258 
259 
260     a = (FT_UInt64)a_;
261     b = (FT_UInt64)b_;
262 
263     FT_MOVE_SIGN( a_, a, s );
264     FT_MOVE_SIGN( b_, b, s );
265 
266     q = b > 0 ? ( ( a << 16 ) + ( b >> 1 ) ) / b
267               : 0x7FFFFFFFUL;
268 
269     q_ = (FT_Long)q;
270 
271     return s < 0 ? NEG_LONG( q_ ) : q_;
272   }
273 
274 
275 #else /* !FT_LONG64 */
276 
277 
278   static void
ft_multo64(FT_UInt32 x,FT_UInt32 y,FT_Int64 * z)279   ft_multo64( FT_UInt32  x,
280               FT_UInt32  y,
281               FT_Int64  *z )
282   {
283     FT_UInt32  lo1, hi1, lo2, hi2, lo, hi, i1, i2;
284 
285 
286     lo1 = x & 0x0000FFFFU;  hi1 = x >> 16;
287     lo2 = y & 0x0000FFFFU;  hi2 = y >> 16;
288 
289     lo = lo1 * lo2;
290     i1 = lo1 * hi2;
291     i2 = lo2 * hi1;
292     hi = hi1 * hi2;
293 
294     /* Check carry overflow of i1 + i2 */
295     i1 += i2;
296     hi += (FT_UInt32)( i1 < i2 ) << 16;
297 
298     hi += i1 >> 16;
299     i1  = i1 << 16;
300 
301     /* Check carry overflow of i1 + lo */
302     lo += i1;
303     hi += ( lo < i1 );
304 
305     z->lo = lo;
306     z->hi = hi;
307   }
308 
309 
310   static FT_UInt32
ft_div64by32(FT_UInt32 hi,FT_UInt32 lo,FT_UInt32 y)311   ft_div64by32( FT_UInt32  hi,
312                 FT_UInt32  lo,
313                 FT_UInt32  y )
314   {
315     FT_UInt32  r, q;
316     FT_Int     i;
317 
318 
319     if ( hi >= y )
320       return (FT_UInt32)0x7FFFFFFFL;
321 
322     /* We shift as many bits as we can into the high register, perform     */
323     /* 32-bit division with modulo there, then work through the remaining  */
324     /* bits with long division. This optimization is especially noticeable */
325     /* for smaller dividends that barely use the high register.            */
326 
327     i = 31 - FT_MSB( hi );
328     r = ( hi << i ) | ( lo >> ( 32 - i ) ); lo <<= i; /* left 64-bit shift */
329     q = r / y;
330     r -= q * y;   /* remainder */
331 
332     i = 32 - i;   /* bits remaining in low register */
333     do
334     {
335       q <<= 1;
336       r   = ( r << 1 ) | ( lo >> 31 ); lo <<= 1;
337 
338       if ( r >= y )
339       {
340         r -= y;
341         q |= 1;
342       }
343     } while ( --i );
344 
345     return q;
346   }
347 
348 
349   static void
FT_Add64(FT_Int64 * x,FT_Int64 * y,FT_Int64 * z)350   FT_Add64( FT_Int64*  x,
351             FT_Int64*  y,
352             FT_Int64  *z )
353   {
354     FT_UInt32  lo, hi;
355 
356 
357     lo = x->lo + y->lo;
358     hi = x->hi + y->hi + ( lo < x->lo );
359 
360     z->lo = lo;
361     z->hi = hi;
362   }
363 
364 
365   /*  The FT_MulDiv function has been optimized thanks to ideas from     */
366   /*  Graham Asher and Alexei Podtelezhnikov.  The trick is to optimize  */
367   /*  a rather common case when everything fits within 32-bits.          */
368   /*                                                                     */
369   /*  We compute 'a*b+c/2', then divide it by 'c' (all positive values). */
370   /*                                                                     */
371   /*  The product of two positive numbers never exceeds the square of    */
372   /*  its mean values.  Therefore, we always avoid the overflow by       */
373   /*  imposing                                                           */
374   /*                                                                     */
375   /*    (a + b) / 2 <= sqrt(X - c/2)    ,                                */
376   /*                                                                     */
377   /*  where X = 2^32 - 1, the maximum unsigned 32-bit value, and using   */
378   /*  unsigned arithmetic.  Now we replace `sqrt' with a linear function */
379   /*  that is smaller or equal for all values of c in the interval       */
380   /*  [0;X/2]; it should be equal to sqrt(X) and sqrt(3X/4) at the       */
381   /*  endpoints.  Substituting the linear solution and explicit numbers  */
382   /*  we get                                                             */
383   /*                                                                     */
384   /*    a + b <= 131071.99 - c / 122291.84    .                          */
385   /*                                                                     */
386   /*  In practice, we should use a faster and even stronger inequality   */
387   /*                                                                     */
388   /*    a + b <= 131071 - (c >> 16)                                      */
389   /*                                                                     */
390   /*  or, alternatively,                                                 */
391   /*                                                                     */
392   /*    a + b <= 129894 - (c >> 17)    .                                 */
393   /*                                                                     */
394   /*  FT_MulFix, on the other hand, is optimized for a small value of    */
395   /*  the first argument, when the second argument can be much larger.   */
396   /*  This can be achieved by scaling the second argument and the limit  */
397   /*  in the above inequalities.  For example,                           */
398   /*                                                                     */
399   /*    a + (b >> 8) <= (131071 >> 4)                                    */
400   /*                                                                     */
401   /*  covers the practical range of use. The actual test below is a bit  */
402   /*  tighter to avoid the border case overflows.                        */
403   /*                                                                     */
404   /*  In the case of FT_DivFix, the exact overflow check                 */
405   /*                                                                     */
406   /*    a << 16 <= X - c/2                                               */
407   /*                                                                     */
408   /*  is scaled down by 2^16 and we use                                  */
409   /*                                                                     */
410   /*    a <= 65535 - (c >> 17)    .                                      */
411 
412   /* documentation is in freetype.h */
413 
414   FT_EXPORT_DEF( FT_Long )
FT_MulDiv(FT_Long a_,FT_Long b_,FT_Long c_)415   FT_MulDiv( FT_Long  a_,
416              FT_Long  b_,
417              FT_Long  c_ )
418   {
419     FT_Int     s = 1;
420     FT_UInt32  a, b, c;
421 
422 
423     /* XXX: this function does not allow 64-bit arguments */
424 
425     a = (FT_UInt32)a_;
426     b = (FT_UInt32)b_;
427     c = (FT_UInt32)c_;
428 
429     FT_MOVE_SIGN( a_, a, s );
430     FT_MOVE_SIGN( b_, b, s );
431     FT_MOVE_SIGN( c_, c, s );
432 
433     if ( c == 0 )
434       a = 0x7FFFFFFFUL;
435 
436     else if ( a + b <= 129894UL - ( c >> 17 ) )
437       a = ( a * b + ( c >> 1 ) ) / c;
438 
439     else
440     {
441       FT_Int64  temp, temp2;
442 
443 
444       ft_multo64( a, b, &temp );
445 
446       temp2.hi = 0;
447       temp2.lo = c >> 1;
448 
449       FT_Add64( &temp, &temp2, &temp );
450 
451       /* last attempt to ditch long division */
452       a = ( temp.hi == 0 ) ? temp.lo / c
453                            : ft_div64by32( temp.hi, temp.lo, c );
454     }
455 
456     a_ = (FT_Long)a;
457 
458     return s < 0 ? NEG_LONG( a_ ) : a_;
459   }
460 
461 
462   FT_BASE_DEF( FT_Long )
FT_MulDiv_No_Round(FT_Long a_,FT_Long b_,FT_Long c_)463   FT_MulDiv_No_Round( FT_Long  a_,
464                       FT_Long  b_,
465                       FT_Long  c_ )
466   {
467     FT_Int     s = 1;
468     FT_UInt32  a, b, c;
469 
470 
471     /* XXX: this function does not allow 64-bit arguments */
472 
473     a = (FT_UInt32)a_;
474     b = (FT_UInt32)b_;
475     c = (FT_UInt32)c_;
476 
477     FT_MOVE_SIGN( a_, a, s );
478     FT_MOVE_SIGN( b_, b, s );
479     FT_MOVE_SIGN( c_, c, s );
480 
481     if ( c == 0 )
482       a = 0x7FFFFFFFUL;
483 
484     else if ( a + b <= 131071UL )
485       a = a * b / c;
486 
487     else
488     {
489       FT_Int64  temp;
490 
491 
492       ft_multo64( a, b, &temp );
493 
494       /* last attempt to ditch long division */
495       a = ( temp.hi == 0 ) ? temp.lo / c
496                            : ft_div64by32( temp.hi, temp.lo, c );
497     }
498 
499     a_ = (FT_Long)a;
500 
501     return s < 0 ? NEG_LONG( a_ ) : a_;
502   }
503 
504 
505   /* documentation is in freetype.h */
506 
507   FT_EXPORT_DEF( FT_Long )
FT_MulFix(FT_Long a_,FT_Long b_)508   FT_MulFix( FT_Long  a_,
509              FT_Long  b_ )
510   {
511 #ifdef FT_MULFIX_ASSEMBLER
512 
513     return FT_MULFIX_ASSEMBLER( a_, b_ );
514 
515 #elif 0
516 
517     /*
518      * This code is nonportable.  See comment below.
519      *
520      * However, on a platform where right-shift of a signed quantity fills
521      * the leftmost bits by copying the sign bit, it might be faster.
522      */
523 
524     FT_Long    sa, sb;
525     FT_UInt32  a, b;
526 
527 
528     /*
529      * This is a clever way of converting a signed number `a' into its
530      * absolute value (stored back into `a') and its sign.  The sign is
531      * stored in `sa'; 0 means `a' was positive or zero, and -1 means `a'
532      * was negative.  (Similarly for `b' and `sb').
533      *
534      * Unfortunately, it doesn't work (at least not portably).
535      *
536      * It makes the assumption that right-shift on a negative signed value
537      * fills the leftmost bits by copying the sign bit.  This is wrong.
538      * According to K&R 2nd ed, section `A7.8 Shift Operators' on page 206,
539      * the result of right-shift of a negative signed value is
540      * implementation-defined.  At least one implementation fills the
541      * leftmost bits with 0s (i.e., it is exactly the same as an unsigned
542      * right shift).  This means that when `a' is negative, `sa' ends up
543      * with the value 1 rather than -1.  After that, everything else goes
544      * wrong.
545      */
546     sa = ( a_ >> ( sizeof ( a_ ) * 8 - 1 ) );
547     a  = ( a_ ^ sa ) - sa;
548     sb = ( b_ >> ( sizeof ( b_ ) * 8 - 1 ) );
549     b  = ( b_ ^ sb ) - sb;
550 
551     a = (FT_UInt32)a_;
552     b = (FT_UInt32)b_;
553 
554     if ( a + ( b >> 8 ) <= 8190UL )
555       a = ( a * b + 0x8000U ) >> 16;
556     else
557     {
558       FT_UInt32  al = a & 0xFFFFUL;
559 
560 
561       a = ( a >> 16 ) * b + al * ( b >> 16 ) +
562           ( ( al * ( b & 0xFFFFUL ) + 0x8000UL ) >> 16 );
563     }
564 
565     sa ^= sb;
566     a   = ( a ^ sa ) - sa;
567 
568     return (FT_Long)a;
569 
570 #else /* 0 */
571 
572     FT_Int     s = 1;
573     FT_UInt32  a, b;
574 
575 
576     /* XXX: this function does not allow 64-bit arguments */
577 
578     a = (FT_UInt32)a_;
579     b = (FT_UInt32)b_;
580 
581     FT_MOVE_SIGN( a_, a, s );
582     FT_MOVE_SIGN( b_, b, s );
583 
584     if ( a + ( b >> 8 ) <= 8190UL )
585       a = ( a * b + 0x8000UL ) >> 16;
586     else
587     {
588       FT_UInt32  al = a & 0xFFFFUL;
589 
590 
591       a = ( a >> 16 ) * b + al * ( b >> 16 ) +
592           ( ( al * ( b & 0xFFFFUL ) + 0x8000UL ) >> 16 );
593     }
594 
595     a_ = (FT_Long)a;
596 
597     return s < 0 ? NEG_LONG( a_ ) : a_;
598 
599 #endif /* 0 */
600 
601   }
602 
603 
604   /* documentation is in freetype.h */
605 
606   FT_EXPORT_DEF( FT_Long )
FT_DivFix(FT_Long a_,FT_Long b_)607   FT_DivFix( FT_Long  a_,
608              FT_Long  b_ )
609   {
610     FT_Int     s = 1;
611     FT_UInt32  a, b, q;
612     FT_Long    q_;
613 
614 
615     /* XXX: this function does not allow 64-bit arguments */
616 
617     a = (FT_UInt32)a_;
618     b = (FT_UInt32)b_;
619 
620     FT_MOVE_SIGN( a_, a, s );
621     FT_MOVE_SIGN( b_, b, s );
622 
623     if ( b == 0 )
624     {
625       /* check for division by 0 */
626       q = 0x7FFFFFFFUL;
627     }
628     else if ( a <= 65535UL - ( b >> 17 ) )
629     {
630       /* compute result directly */
631       q = ( ( a << 16 ) + ( b >> 1 ) ) / b;
632     }
633     else
634     {
635       /* we need more bits; we have to do it by hand */
636       FT_Int64  temp, temp2;
637 
638 
639       temp.hi  = a >> 16;
640       temp.lo  = a << 16;
641       temp2.hi = 0;
642       temp2.lo = b >> 1;
643 
644       FT_Add64( &temp, &temp2, &temp );
645       q = ft_div64by32( temp.hi, temp.lo, b );
646     }
647 
648     q_ = (FT_Long)q;
649 
650     return s < 0 ? NEG_LONG( q_ ) : q_;
651   }
652 
653 
654 #endif /* !FT_LONG64 */
655 
656 
657   /* documentation is in ftglyph.h */
658 
659   FT_EXPORT_DEF( void )
FT_Matrix_Multiply(const FT_Matrix * a,FT_Matrix * b)660   FT_Matrix_Multiply( const FT_Matrix*  a,
661                       FT_Matrix        *b )
662   {
663     FT_Fixed  xx, xy, yx, yy;
664 
665 
666     if ( !a || !b )
667       return;
668 
669     xx = ADD_LONG( FT_MulFix( a->xx, b->xx ),
670                    FT_MulFix( a->xy, b->yx ) );
671     xy = ADD_LONG( FT_MulFix( a->xx, b->xy ),
672                    FT_MulFix( a->xy, b->yy ) );
673     yx = ADD_LONG( FT_MulFix( a->yx, b->xx ),
674                    FT_MulFix( a->yy, b->yx ) );
675     yy = ADD_LONG( FT_MulFix( a->yx, b->xy ),
676                    FT_MulFix( a->yy, b->yy ) );
677 
678     b->xx = xx;
679     b->xy = xy;
680     b->yx = yx;
681     b->yy = yy;
682   }
683 
684 
685   /* documentation is in ftglyph.h */
686 
687   FT_EXPORT_DEF( FT_Error )
FT_Matrix_Invert(FT_Matrix * matrix)688   FT_Matrix_Invert( FT_Matrix*  matrix )
689   {
690     FT_Pos  delta, xx, yy;
691 
692 
693     if ( !matrix )
694       return FT_THROW( Invalid_Argument );
695 
696     /* compute discriminant */
697     delta = FT_MulFix( matrix->xx, matrix->yy ) -
698             FT_MulFix( matrix->xy, matrix->yx );
699 
700     if ( !delta )
701       return FT_THROW( Invalid_Argument );  /* matrix can't be inverted */
702 
703     matrix->xy = -FT_DivFix( matrix->xy, delta );
704     matrix->yx = -FT_DivFix( matrix->yx, delta );
705 
706     xx = matrix->xx;
707     yy = matrix->yy;
708 
709     matrix->xx = FT_DivFix( yy, delta );
710     matrix->yy = FT_DivFix( xx, delta );
711 
712     return FT_Err_Ok;
713   }
714 
715 
716   /* documentation is in ftcalc.h */
717 
718   FT_BASE_DEF( void )
FT_Matrix_Multiply_Scaled(const FT_Matrix * a,FT_Matrix * b,FT_Long scaling)719   FT_Matrix_Multiply_Scaled( const FT_Matrix*  a,
720                              FT_Matrix        *b,
721                              FT_Long           scaling )
722   {
723     FT_Fixed  xx, xy, yx, yy;
724 
725     FT_Long   val = 0x10000L * scaling;
726 
727 
728     if ( !a || !b )
729       return;
730 
731     xx = ADD_LONG( FT_MulDiv( a->xx, b->xx, val ),
732                    FT_MulDiv( a->xy, b->yx, val ) );
733     xy = ADD_LONG( FT_MulDiv( a->xx, b->xy, val ),
734                    FT_MulDiv( a->xy, b->yy, val ) );
735     yx = ADD_LONG( FT_MulDiv( a->yx, b->xx, val ),
736                    FT_MulDiv( a->yy, b->yx, val ) );
737     yy = ADD_LONG( FT_MulDiv( a->yx, b->xy, val ),
738                    FT_MulDiv( a->yy, b->yy, val ) );
739 
740     b->xx = xx;
741     b->xy = xy;
742     b->yx = yx;
743     b->yy = yy;
744   }
745 
746 
747   /* documentation is in ftcalc.h */
748 
749   FT_BASE_DEF( FT_Bool )
FT_Matrix_Check(const FT_Matrix * matrix)750   FT_Matrix_Check( const FT_Matrix*  matrix )
751   {
752     FT_Matrix  m;
753     FT_Fixed   val[4];
754     FT_Fixed   nonzero_minval, maxval;
755     FT_Fixed   temp1, temp2;
756     FT_UInt    i;
757 
758 
759     if ( !matrix )
760       return 0;
761 
762     val[0] = FT_ABS( matrix->xx );
763     val[1] = FT_ABS( matrix->xy );
764     val[2] = FT_ABS( matrix->yx );
765     val[3] = FT_ABS( matrix->yy );
766 
767     /*
768      * To avoid overflow, we ensure that each value is not larger than
769      *
770      *   int(sqrt(2^31 / 4)) = 23170  ;
771      *
772      * we also check that no value becomes zero if we have to scale.
773      */
774 
775     maxval         = 0;
776     nonzero_minval = FT_LONG_MAX;
777 
778     for ( i = 0; i < 4; i++ )
779     {
780       if ( val[i] > maxval )
781         maxval = val[i];
782       if ( val[i] && val[i] < nonzero_minval )
783         nonzero_minval = val[i];
784     }
785 
786     /* we only handle 32bit values */
787     if ( maxval > 0x7FFFFFFFL )
788       return 0;
789 
790     if ( maxval > 23170 )
791     {
792       FT_Fixed  scale = FT_DivFix( maxval, 23170 );
793 
794 
795       if ( !FT_DivFix( nonzero_minval, scale ) )
796         return 0;    /* value range too large */
797 
798       m.xx = FT_DivFix( matrix->xx, scale );
799       m.xy = FT_DivFix( matrix->xy, scale );
800       m.yx = FT_DivFix( matrix->yx, scale );
801       m.yy = FT_DivFix( matrix->yy, scale );
802     }
803     else
804       m = *matrix;
805 
806     temp1 = FT_ABS( m.xx * m.yy - m.xy * m.yx );
807     temp2 = m.xx * m.xx + m.xy * m.xy + m.yx * m.yx + m.yy * m.yy;
808 
809     if ( temp1 == 0         ||
810          temp2 / temp1 > 50 )
811       return 0;
812 
813     return 1;
814   }
815 
816 
817   /* documentation is in ftcalc.h */
818 
819   FT_BASE_DEF( void )
FT_Vector_Transform_Scaled(FT_Vector * vector,const FT_Matrix * matrix,FT_Long scaling)820   FT_Vector_Transform_Scaled( FT_Vector*        vector,
821                               const FT_Matrix*  matrix,
822                               FT_Long           scaling )
823   {
824     FT_Pos   xz, yz;
825 
826     FT_Long  val = 0x10000L * scaling;
827 
828 
829     if ( !vector || !matrix )
830       return;
831 
832     xz = ADD_LONG( FT_MulDiv( vector->x, matrix->xx, val ),
833                    FT_MulDiv( vector->y, matrix->xy, val ) );
834     yz = ADD_LONG( FT_MulDiv( vector->x, matrix->yx, val ),
835                    FT_MulDiv( vector->y, matrix->yy, val ) );
836 
837     vector->x = xz;
838     vector->y = yz;
839   }
840 
841 
842   /* documentation is in ftcalc.h */
843 
844   FT_BASE_DEF( FT_UInt32 )
FT_Vector_NormLen(FT_Vector * vector)845   FT_Vector_NormLen( FT_Vector*  vector )
846   {
847     FT_Int32   x_ = vector->x;
848     FT_Int32   y_ = vector->y;
849     FT_Int32   b, z;
850     FT_UInt32  x, y, u, v, l;
851     FT_Int     sx = 1, sy = 1, shift;
852 
853 
854     x = (FT_UInt32)x_;
855     y = (FT_UInt32)y_;
856 
857     FT_MOVE_SIGN( x_, x, sx );
858     FT_MOVE_SIGN( y_, y, sy );
859 
860     /* trivial cases */
861     if ( x == 0 )
862     {
863       if ( y > 0 )
864         vector->y = sy * 0x10000;
865       return y;
866     }
867     else if ( y == 0 )
868     {
869       if ( x > 0 )
870         vector->x = sx * 0x10000;
871       return x;
872     }
873 
874     /* Estimate length and prenormalize by shifting so that */
875     /* the new approximate length is between 2/3 and 4/3.   */
876     /* The magic constant 0xAAAAAAAAUL (2/3 of 2^32) helps  */
877     /* achieve this in 16.16 fixed-point representation.    */
878     l = x > y ? x + ( y >> 1 )
879               : y + ( x >> 1 );
880 
881     shift  = 31 - FT_MSB( l );
882     shift -= 15 + ( l >= ( 0xAAAAAAAAUL >> shift ) );
883 
884     if ( shift > 0 )
885     {
886       x <<= shift;
887       y <<= shift;
888 
889       /* re-estimate length for tiny vectors */
890       l = x > y ? x + ( y >> 1 )
891                 : y + ( x >> 1 );
892     }
893     else
894     {
895       x >>= -shift;
896       y >>= -shift;
897       l >>= -shift;
898     }
899 
900     /* lower linear approximation for reciprocal length minus one */
901     b = 0x10000 - (FT_Int32)l;
902 
903     x_ = (FT_Int32)x;
904     y_ = (FT_Int32)y;
905 
906     /* Newton's iterations */
907     do
908     {
909       u = (FT_UInt32)( x_ + ( x_ * b >> 16 ) );
910       v = (FT_UInt32)( y_ + ( y_ * b >> 16 ) );
911 
912       /* Normalized squared length in the parentheses approaches 2^32. */
913       /* On two's complement systems, converting to signed gives the   */
914       /* difference with 2^32 even if the expression wraps around.     */
915       z = -(FT_Int32)( u * u + v * v ) / 0x200;
916       z = z * ( ( 0x10000 + b ) >> 8 ) / 0x10000;
917 
918       b += z;
919 
920     } while ( z > 0 );
921 
922     vector->x = sx < 0 ? -(FT_Pos)u : (FT_Pos)u;
923     vector->y = sy < 0 ? -(FT_Pos)v : (FT_Pos)v;
924 
925     /* Conversion to signed helps to recover from likely wrap around */
926     /* in calculating the prenormalized length, because it gives the */
927     /* correct difference with 2^32 on two's complement systems.     */
928     l = (FT_UInt32)( 0x10000 + (FT_Int32)( u * x + v * y ) / 0x10000 );
929     if ( shift > 0 )
930       l = ( l + ( 1 << ( shift - 1 ) ) ) >> shift;
931     else
932       l <<= -shift;
933 
934     return l;
935   }
936 
937 
938 #if 0
939 
940   /* documentation is in ftcalc.h */
941 
942   FT_BASE_DEF( FT_Int32 )
943   FT_SqrtFixed( FT_Int32  x )
944   {
945     FT_UInt32  root, rem_hi, rem_lo, test_div;
946     FT_Int     count;
947 
948 
949     root = 0;
950 
951     if ( x > 0 )
952     {
953       rem_hi = 0;
954       rem_lo = (FT_UInt32)x;
955       count  = 24;
956       do
957       {
958         rem_hi   = ( rem_hi << 2 ) | ( rem_lo >> 30 );
959         rem_lo <<= 2;
960         root   <<= 1;
961         test_div = ( root << 1 ) + 1;
962 
963         if ( rem_hi >= test_div )
964         {
965           rem_hi -= test_div;
966           root   += 1;
967         }
968       } while ( --count );
969     }
970 
971     return (FT_Int32)root;
972   }
973 
974 #endif /* 0 */
975 
976 
977   /* documentation is in ftcalc.h */
978 
979   FT_BASE_DEF( FT_Int )
ft_corner_orientation(FT_Pos in_x,FT_Pos in_y,FT_Pos out_x,FT_Pos out_y)980   ft_corner_orientation( FT_Pos  in_x,
981                          FT_Pos  in_y,
982                          FT_Pos  out_x,
983                          FT_Pos  out_y )
984   {
985     /* we silently ignore overflow errors since such large values */
986     /* lead to even more (harmless) rendering errors later on     */
987 
988 #ifdef FT_LONG64
989 
990     FT_Int64  delta = SUB_INT64( MUL_INT64( in_x, out_y ),
991                                  MUL_INT64( in_y, out_x ) );
992 
993 
994     return ( delta > 0 ) - ( delta < 0 );
995 
996 #else
997 
998     FT_Int  result;
999 
1000 
1001     if ( ADD_LONG( FT_ABS( in_x ), FT_ABS( out_y ) ) <= 131071L &&
1002          ADD_LONG( FT_ABS( in_y ), FT_ABS( out_x ) ) <= 131071L )
1003     {
1004       FT_Long  z1 = MUL_LONG( in_x, out_y );
1005       FT_Long  z2 = MUL_LONG( in_y, out_x );
1006 
1007 
1008       if ( z1 > z2 )
1009         result = +1;
1010       else if ( z1 < z2 )
1011         result = -1;
1012       else
1013         result = 0;
1014     }
1015     else /* products might overflow 32 bits */
1016     {
1017       FT_Int64  z1, z2;
1018 
1019 
1020       /* XXX: this function does not allow 64-bit arguments */
1021       ft_multo64( (FT_UInt32)in_x, (FT_UInt32)out_y, &z1 );
1022       ft_multo64( (FT_UInt32)in_y, (FT_UInt32)out_x, &z2 );
1023 
1024       if ( z1.hi > z2.hi )
1025         result = +1;
1026       else if ( z1.hi < z2.hi )
1027         result = -1;
1028       else if ( z1.lo > z2.lo )
1029         result = +1;
1030       else if ( z1.lo < z2.lo )
1031         result = -1;
1032       else
1033         result = 0;
1034     }
1035 
1036     /* XXX: only the sign of return value, +1/0/-1 must be used */
1037     return result;
1038 
1039 #endif
1040   }
1041 
1042 
1043   /* documentation is in ftcalc.h */
1044 
1045   FT_BASE_DEF( FT_Int )
ft_corner_is_flat(FT_Pos in_x,FT_Pos in_y,FT_Pos out_x,FT_Pos out_y)1046   ft_corner_is_flat( FT_Pos  in_x,
1047                      FT_Pos  in_y,
1048                      FT_Pos  out_x,
1049                      FT_Pos  out_y )
1050   {
1051     FT_Pos  ax = in_x + out_x;
1052     FT_Pos  ay = in_y + out_y;
1053 
1054     FT_Pos  d_in, d_out, d_hypot;
1055 
1056 
1057     /* The idea of this function is to compare the length of the */
1058     /* hypotenuse with the `in' and `out' length.  The `corner'  */
1059     /* represented by `in' and `out' is flat if the hypotenuse's */
1060     /* length isn't too large.                                   */
1061     /*                                                           */
1062     /* This approach has the advantage that the angle between    */
1063     /* `in' and `out' is not checked.  In case one of the two    */
1064     /* vectors is `dominant', this is, much larger than the      */
1065     /* other vector, we thus always have a flat corner.          */
1066     /*                                                           */
1067     /*                hypotenuse                                 */
1068     /*       x---------------------------x                       */
1069     /*        \                      /                           */
1070     /*         \                /                                */
1071     /*      in  \          /  out                                */
1072     /*           \    /                                          */
1073     /*            o                                              */
1074     /*              Point                                        */
1075 
1076     d_in    = FT_HYPOT(  in_x,  in_y );
1077     d_out   = FT_HYPOT( out_x, out_y );
1078     d_hypot = FT_HYPOT(    ax,    ay );
1079 
1080     /* now do a simple length comparison: */
1081     /*                                    */
1082     /*   d_in + d_out < 17/16 d_hypot     */
1083 
1084     return ( d_in + d_out - d_hypot ) < ( d_hypot >> 4 );
1085   }
1086 
1087 
1088 /* END */
1089