• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /****************************************************************************
2  *
3  * ftcalc.h
4  *
5  *   Arithmetic computations (specification).
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 #ifndef FTCALC_H_
20 #define FTCALC_H_
21 
22 
23 #include <freetype/freetype.h>
24 
25 #include "compiler-macros.h"
26 
27 FT_BEGIN_HEADER
28 
29 
30   /**************************************************************************
31    *
32    * FT_MulDiv() and FT_MulFix() are declared in freetype.h.
33    *
34    */
35 
36 #ifndef  FT_CONFIG_OPTION_NO_ASSEMBLER
37   /* Provide assembler fragments for performance-critical functions. */
38   /* These must be defined `static __inline__' with GCC.             */
39 
40 #if defined( __CC_ARM ) || defined( __ARMCC__ )  /* RVCT */
41 
42 #define FT_MULFIX_ASSEMBLER  FT_MulFix_arm
43 
44   /* documentation is in freetype.h */
45 
46   static __inline FT_Int32
FT_MulFix_arm(FT_Int32 a,FT_Int32 b)47   FT_MulFix_arm( FT_Int32  a,
48                  FT_Int32  b )
49   {
50     FT_Int32  t, t2;
51 
52 
53     __asm
54     {
55       smull t2, t,  b,  a           /* (lo=t2,hi=t) = a*b */
56       mov   a,  t,  asr #31         /* a   = (hi >> 31) */
57       add   a,  a,  #0x8000         /* a  += 0x8000 */
58       adds  t2, t2, a               /* t2 += a */
59       adc   t,  t,  #0              /* t  += carry */
60       mov   a,  t2, lsr #16         /* a   = t2 >> 16 */
61       orr   a,  a,  t,  lsl #16     /* a  |= t << 16 */
62     }
63     return a;
64   }
65 
66 #endif /* __CC_ARM || __ARMCC__ */
67 
68 
69 #ifdef __GNUC__
70 
71 #if defined( __arm__ )                                 && \
72     ( !defined( __thumb__ ) || defined( __thumb2__ ) ) && \
73     !( defined( __CC_ARM ) || defined( __ARMCC__ ) )
74 
75 #define FT_MULFIX_ASSEMBLER  FT_MulFix_arm
76 
77   /* documentation is in freetype.h */
78 
79   static __inline__ FT_Int32
FT_MulFix_arm(FT_Int32 a,FT_Int32 b)80   FT_MulFix_arm( FT_Int32  a,
81                  FT_Int32  b )
82   {
83     FT_Int32  t, t2;
84 
85 
86     __asm__ __volatile__ (
87       "smull  %1, %2, %4, %3\n\t"       /* (lo=%1,hi=%2) = a*b */
88       "mov    %0, %2, asr #31\n\t"      /* %0  = (hi >> 31) */
89 #if defined( __clang__ ) && defined( __thumb2__ )
90       "add.w  %0, %0, #0x8000\n\t"      /* %0 += 0x8000 */
91 #else
92       "add    %0, %0, #0x8000\n\t"      /* %0 += 0x8000 */
93 #endif
94       "adds   %1, %1, %0\n\t"           /* %1 += %0 */
95       "adc    %2, %2, #0\n\t"           /* %2 += carry */
96       "mov    %0, %1, lsr #16\n\t"      /* %0  = %1 >> 16 */
97       "orr    %0, %0, %2, lsl #16\n\t"  /* %0 |= %2 << 16 */
98       : "=r"(a), "=&r"(t2), "=&r"(t)
99       : "r"(a), "r"(b)
100       : "cc" );
101     return a;
102   }
103 
104 #endif /* __arm__                      && */
105        /* ( __thumb2__ || !__thumb__ ) && */
106        /* !( __CC_ARM || __ARMCC__ )      */
107 
108 
109 #if defined( __i386__ )
110 
111 #define FT_MULFIX_ASSEMBLER  FT_MulFix_i386
112 
113   /* documentation is in freetype.h */
114 
115   static __inline__ FT_Int32
FT_MulFix_i386(FT_Int32 a,FT_Int32 b)116   FT_MulFix_i386( FT_Int32  a,
117                   FT_Int32  b )
118   {
119     FT_Int32  result;
120 
121 
122     __asm__ __volatile__ (
123       "imul  %%edx\n"
124       "movl  %%edx, %%ecx\n"
125       "sarl  $31, %%ecx\n"
126       "addl  $0x8000, %%ecx\n"
127       "addl  %%ecx, %%eax\n"
128       "adcl  $0, %%edx\n"
129       "shrl  $16, %%eax\n"
130       "shll  $16, %%edx\n"
131       "addl  %%edx, %%eax\n"
132       : "=a"(result), "=d"(b)
133       : "a"(a), "d"(b)
134       : "%ecx", "cc" );
135     return result;
136   }
137 
138 #endif /* i386 */
139 
140 #endif /* __GNUC__ */
141 
142 
143 #ifdef _MSC_VER /* Visual C++ */
144 
145 #ifdef _M_IX86
146 
147 #define FT_MULFIX_ASSEMBLER  FT_MulFix_i386
148 
149   /* documentation is in freetype.h */
150 
151   static __inline FT_Int32
FT_MulFix_i386(FT_Int32 a,FT_Int32 b)152   FT_MulFix_i386( FT_Int32  a,
153                   FT_Int32  b )
154   {
155     FT_Int32  result;
156 
157     __asm
158     {
159       mov eax, a
160       mov edx, b
161       imul edx
162       mov ecx, edx
163       sar ecx, 31
164       add ecx, 8000h
165       add eax, ecx
166       adc edx, 0
167       shr eax, 16
168       shl edx, 16
169       add eax, edx
170       mov result, eax
171     }
172     return result;
173   }
174 
175 #endif /* _M_IX86 */
176 
177 #endif /* _MSC_VER */
178 
179 
180 #if defined( __GNUC__ ) && defined( __x86_64__ )
181 
182 #define FT_MULFIX_ASSEMBLER  FT_MulFix_x86_64
183 
184   static __inline__ FT_Int32
FT_MulFix_x86_64(FT_Int32 a,FT_Int32 b)185   FT_MulFix_x86_64( FT_Int32  a,
186                     FT_Int32  b )
187   {
188     /* Temporarily disable the warning that C90 doesn't support */
189     /* `long long'.                                             */
190 #if __GNUC__ > 4 || ( __GNUC__ == 4 && __GNUC_MINOR__ >= 6 )
191 #pragma GCC diagnostic push
192 #pragma GCC diagnostic ignored "-Wlong-long"
193 #endif
194 
195 #if 1
196     /* Technically not an assembly fragment, but GCC does a really good */
197     /* job at inlining it and generating good machine code for it.      */
198     long long  ret, tmp;
199 
200 
201     ret  = (long long)a * b;
202     tmp  = ret >> 63;
203     ret += 0x8000 + tmp;
204 
205     return (FT_Int32)( ret >> 16 );
206 #else
207 
208     /* For some reason, GCC 4.6 on Ubuntu 12.04 generates invalid machine  */
209     /* code from the lines below.  The main issue is that `wide_a' is not  */
210     /* properly initialized by sign-extending `a'.  Instead, the generated */
211     /* machine code assumes that the register that contains `a' on input   */
212     /* can be used directly as a 64-bit value, which is wrong most of the  */
213     /* time.                                                               */
214     long long  wide_a = (long long)a;
215     long long  wide_b = (long long)b;
216     long long  result;
217 
218 
219     __asm__ __volatile__ (
220       "imul %2, %1\n"
221       "mov %1, %0\n"
222       "sar $63, %0\n"
223       "lea 0x8000(%1, %0), %0\n"
224       "sar $16, %0\n"
225       : "=&r"(result), "=&r"(wide_a)
226       : "r"(wide_b)
227       : "cc" );
228 
229     return (FT_Int32)result;
230 #endif
231 
232 #if __GNUC__ > 4 || ( __GNUC__ == 4 && __GNUC_MINOR__ >= 6 )
233 #pragma GCC diagnostic pop
234 #endif
235   }
236 
237 #endif /* __GNUC__ && __x86_64__ */
238 
239 #endif /* !FT_CONFIG_OPTION_NO_ASSEMBLER */
240 
241 
242 #ifdef FT_CONFIG_OPTION_INLINE_MULFIX
243 #ifdef FT_MULFIX_ASSEMBLER
244 #define FT_MulFix( a, b )  FT_MULFIX_ASSEMBLER( (FT_Int32)(a), (FT_Int32)(b) )
245 #endif
246 #endif
247 
248 
249   /**************************************************************************
250    *
251    * @function:
252    *   FT_MulDiv_No_Round
253    *
254    * @description:
255    *   A very simple function used to perform the computation '(a*b)/c'
256    *   (without rounding) with maximum accuracy (it uses a 64-bit
257    *   intermediate integer whenever necessary).
258    *
259    *   This function isn't necessarily as fast as some processor-specific
260    *   operations, but is at least completely portable.
261    *
262    * @input:
263    *   a ::
264    *     The first multiplier.
265    *   b ::
266    *     The second multiplier.
267    *   c ::
268    *     The divisor.
269    *
270    * @return:
271    *   The result of '(a*b)/c'.  This function never traps when trying to
272    *   divide by zero; it simply returns 'MaxInt' or 'MinInt' depending on
273    *   the signs of 'a' and 'b'.
274    */
275   FT_BASE( FT_Long )
276   FT_MulDiv_No_Round( FT_Long  a,
277                       FT_Long  b,
278                       FT_Long  c );
279 
280 
281   /*
282    * A variant of FT_Matrix_Multiply which scales its result afterwards.  The
283    * idea is that both `a' and `b' are scaled by factors of 10 so that the
284    * values are as precise as possible to get a correct result during the
285    * 64bit multiplication.  Let `sa' and `sb' be the scaling factors of `a'
286    * and `b', respectively, then the scaling factor of the result is `sa*sb'.
287    */
288   FT_BASE( void )
289   FT_Matrix_Multiply_Scaled( const FT_Matrix*  a,
290                              FT_Matrix        *b,
291                              FT_Long           scaling );
292 
293 
294   /*
295    * Check a matrix.  If the transformation would lead to extreme shear or
296    * extreme scaling, for example, return 0.  If everything is OK, return 1.
297    *
298    * Based on geometric considerations we use the following inequality to
299    * identify a degenerate matrix.
300    *
301    *   50 * abs(xx*yy - xy*yx) < xx^2 + xy^2 + yx^2 + yy^2
302    *
303    * Value 50 is heuristic.
304    */
305   FT_BASE( FT_Bool )
306   FT_Matrix_Check( const FT_Matrix*  matrix );
307 
308 
309   /*
310    * A variant of FT_Vector_Transform.  See comments for
311    * FT_Matrix_Multiply_Scaled.
312    */
313   FT_BASE( void )
314   FT_Vector_Transform_Scaled( FT_Vector*        vector,
315                               const FT_Matrix*  matrix,
316                               FT_Long           scaling );
317 
318 
319   /*
320    * This function normalizes a vector and returns its original length.  The
321    * normalized vector is a 16.16 fixed-point unit vector with length close
322    * to 0x10000.  The accuracy of the returned length is limited to 16 bits
323    * also.  The function utilizes quick inverse square root approximation
324    * without divisions and square roots relying on Newton's iterations
325    * instead.
326    */
327   FT_BASE( FT_UInt32 )
328   FT_Vector_NormLen( FT_Vector*  vector );
329 
330 
331   /*
332    * Return -1, 0, or +1, depending on the orientation of a given corner.  We
333    * use the Cartesian coordinate system, with positive vertical values going
334    * upwards.  The function returns +1 if the corner turns to the left, -1 to
335    * the right, and 0 for undecidable cases.
336    */
337   FT_BASE( FT_Int )
338   ft_corner_orientation( FT_Pos  in_x,
339                          FT_Pos  in_y,
340                          FT_Pos  out_x,
341                          FT_Pos  out_y );
342 
343 
344   /*
345    * Return TRUE if a corner is flat or nearly flat.  This is equivalent to
346    * saying that the corner point is close to its neighbors, or inside an
347    * ellipse defined by the neighbor focal points to be more precise.
348    */
349   FT_BASE( FT_Int )
350   ft_corner_is_flat( FT_Pos  in_x,
351                      FT_Pos  in_y,
352                      FT_Pos  out_x,
353                      FT_Pos  out_y );
354 
355 
356   /*
357    * Return the most significant bit index.
358    */
359 
360 #ifndef  FT_CONFIG_OPTION_NO_ASSEMBLER
361 
362 #if defined( __GNUC__ )                                          && \
363     ( __GNUC__ > 3 || ( __GNUC__ == 3 && __GNUC_MINOR__ >= 4 ) )
364 
365 #if FT_SIZEOF_INT == 4
366 
367 #define FT_MSB( x )  ( 31 - __builtin_clz( x ) )
368 
369 #elif FT_SIZEOF_LONG == 4
370 
371 #define FT_MSB( x )  ( 31 - __builtin_clzl( x ) )
372 
373 #endif /* __GNUC__ */
374 
375 
376 #elif defined( _MSC_VER ) && ( _MSC_VER >= 1400 )
377 
378 #if FT_SIZEOF_INT == 4
379 
380 #include <intrin.h>
381 #pragma intrinsic( _BitScanReverse )
382 
383   static __inline FT_Int32
FT_MSB_i386(FT_UInt32 x)384   FT_MSB_i386( FT_UInt32  x )
385   {
386     unsigned long  where;
387 
388 
389     _BitScanReverse( &where, x );
390 
391     return (FT_Int32)where;
392   }
393 
394 #define FT_MSB( x )  ( FT_MSB_i386( x ) )
395 
396 #endif
397 
398 #endif /* _MSC_VER */
399 
400 
401 #endif /* !FT_CONFIG_OPTION_NO_ASSEMBLER */
402 
403 #ifndef FT_MSB
404 
405   FT_BASE( FT_Int )
406   FT_MSB( FT_UInt32  z );
407 
408 #endif
409 
410 
411   /*
412    * Return sqrt(x*x+y*y), which is the same as `FT_Vector_Length' but uses
413    * two fixed-point arguments instead.
414    */
415   FT_BASE( FT_Fixed )
416   FT_Hypot( FT_Fixed  x,
417             FT_Fixed  y );
418 
419 
420 #if 0
421 
422   /**************************************************************************
423    *
424    * @function:
425    *   FT_SqrtFixed
426    *
427    * @description:
428    *   Computes the square root of a 16.16 fixed-point value.
429    *
430    * @input:
431    *   x ::
432    *     The value to compute the root for.
433    *
434    * @return:
435    *   The result of 'sqrt(x)'.
436    *
437    * @note:
438    *   This function is not very fast.
439    */
440   FT_BASE( FT_Int32 )
441   FT_SqrtFixed( FT_Int32  x );
442 
443 #endif /* 0 */
444 
445 
446 #define INT_TO_F26DOT6( x )    ( (FT_Long)(x) * 64  )    /* << 6  */
447 #define INT_TO_F2DOT14( x )    ( (FT_Long)(x) * 16384 )  /* << 14 */
448 #define INT_TO_FIXED( x )      ( (FT_Long)(x) * 65536 )  /* << 16 */
449 #define F2DOT14_TO_FIXED( x )  ( (FT_Long)(x) * 4 )      /* << 2  */
450 #define FIXED_TO_INT( x )      ( FT_RoundFix( x ) >> 16 )
451 
452 #define ROUND_F26DOT6( x )     ( ( (x) + 32 - ( x < 0 ) ) & -64 )
453 
454   /*
455    * The following macros have two purposes.
456    *
457    * - Tag places where overflow is expected and harmless.
458    *
459    * - Avoid run-time sanitizer errors.
460    *
461    * Use with care!
462    */
463 #define ADD_INT( a, b )                           \
464           (FT_Int)( (FT_UInt)(a) + (FT_UInt)(b) )
465 #define SUB_INT( a, b )                           \
466           (FT_Int)( (FT_UInt)(a) - (FT_UInt)(b) )
467 #define MUL_INT( a, b )                           \
468           (FT_Int)( (FT_UInt)(a) * (FT_UInt)(b) )
469 #define NEG_INT( a )                              \
470           (FT_Int)( (FT_UInt)0 - (FT_UInt)(a) )
471 
472 #define ADD_LONG( a, b )                             \
473           (FT_Long)( (FT_ULong)(a) + (FT_ULong)(b) )
474 #define SUB_LONG( a, b )                             \
475           (FT_Long)( (FT_ULong)(a) - (FT_ULong)(b) )
476 #define MUL_LONG( a, b )                             \
477           (FT_Long)( (FT_ULong)(a) * (FT_ULong)(b) )
478 #define NEG_LONG( a )                                \
479           (FT_Long)( (FT_ULong)0 - (FT_ULong)(a) )
480 
481 #define ADD_INT32( a, b )                               \
482           (FT_Int32)( (FT_UInt32)(a) + (FT_UInt32)(b) )
483 #define SUB_INT32( a, b )                               \
484           (FT_Int32)( (FT_UInt32)(a) - (FT_UInt32)(b) )
485 #define MUL_INT32( a, b )                               \
486           (FT_Int32)( (FT_UInt32)(a) * (FT_UInt32)(b) )
487 #define NEG_INT32( a )                                  \
488           (FT_Int32)( (FT_UInt32)0 - (FT_UInt32)(a) )
489 
490 #ifdef FT_LONG64
491 
492 #define ADD_INT64( a, b )                               \
493           (FT_Int64)( (FT_UInt64)(a) + (FT_UInt64)(b) )
494 #define SUB_INT64( a, b )                               \
495           (FT_Int64)( (FT_UInt64)(a) - (FT_UInt64)(b) )
496 #define MUL_INT64( a, b )                               \
497           (FT_Int64)( (FT_UInt64)(a) * (FT_UInt64)(b) )
498 #define NEG_INT64( a )                                  \
499           (FT_Int64)( (FT_UInt64)0 - (FT_UInt64)(a) )
500 
501 #endif /* FT_LONG64 */
502 
503 
504 FT_END_HEADER
505 
506 #endif /* FTCALC_H_ */
507 
508 
509 /* END */
510