|
5003
|
1 |
"{ Encoding: utf8 }"
|
|
|
2 |
|
|
4983
|
3 |
"{ Package: 'stx:libbasic2' }"
|
|
|
4 |
|
|
|
5 |
"{ NameSpace: Smalltalk }"
|
|
|
6 |
|
|
|
7 |
LimitedPrecisionReal variableByteSubclass:#QuadFloat
|
|
|
8 |
instanceVariableNames:''
|
|
|
9 |
classVariableNames:'QuadFloatZero QuadFloatOne Pi E Epsilon NaN PositiveInfinity
|
|
|
10 |
NegativeInfinity Halfpi HalfpiNegative'
|
|
|
11 |
poolDictionaries:''
|
|
|
12 |
category:'Magnitude-Numbers'
|
|
|
13 |
!
|
|
|
14 |
|
|
4991
|
15 |
!QuadFloat primitiveDefinitions!
|
|
|
16 |
%{
|
|
4993
|
17 |
#if __POINTER_SIZE__ == 8
|
|
|
18 |
# define SUPPORT_QUADFLOAT
|
|
|
19 |
#endif
|
|
|
20 |
|
|
|
21 |
#ifdef SUPPORT_QUADFLOAT
|
|
4998
|
22 |
# include <stdint.h>
|
|
|
23 |
|
|
4992
|
24 |
/*----------------------------------------------------------------------------
|
|
|
25 |
| Software floating-point underflow tininess-detection mode.
|
|
|
26 |
*----------------------------------------------------------------------------*/
|
|
|
27 |
enum {
|
|
|
28 |
softfloat_tininess_beforeRounding = 0,
|
|
|
29 |
softfloat_tininess_afterRounding = 1
|
|
|
30 |
};
|
|
|
31 |
|
|
|
32 |
/*----------------------------------------------------------------------------
|
|
|
33 |
| Software floating-point exception flags.
|
|
|
34 |
*----------------------------------------------------------------------------*/
|
|
|
35 |
enum {
|
|
|
36 |
softfloat_flag_inexact = 1,
|
|
|
37 |
softfloat_flag_underflow = 2,
|
|
|
38 |
softfloat_flag_overflow = 4,
|
|
|
39 |
softfloat_flag_infinite = 8,
|
|
|
40 |
softfloat_flag_invalid = 16
|
|
|
41 |
};
|
|
|
42 |
/*----------------------------------------------------------------------------
|
|
|
43 |
| Software floating-point rounding mode. (Mode "odd" is supported only if
|
|
|
44 |
| SoftFloat is compiled with macro 'SOFTFLOAT_ROUND_ODD' defined.)
|
|
|
45 |
*----------------------------------------------------------------------------*/
|
|
|
46 |
enum {
|
|
|
47 |
softfloat_round_near_even = 0,
|
|
|
48 |
softfloat_round_minMag = 1,
|
|
|
49 |
softfloat_round_min = 2,
|
|
|
50 |
softfloat_round_max = 3,
|
|
|
51 |
softfloat_round_near_maxMag = 4,
|
|
|
52 |
softfloat_round_odd = 6
|
|
|
53 |
};
|
|
|
54 |
#define init_detectTininess softfloat_tininess_afterRounding
|
|
|
55 |
|
|
|
56 |
#if defined(LITTLEENDIAN) || defined(__LSB_FIRST__)
|
|
|
57 |
struct uint128 { uint64_t v0, v64; };
|
|
|
58 |
struct uint64_extra { uint64_t extra, v; };
|
|
|
59 |
struct uint128_extra { uint64_t extra; struct uint128 v; };
|
|
|
60 |
#else
|
|
|
61 |
struct uint128 { uint64_t v64, v0; };
|
|
|
62 |
struct uint64_extra { uint64_t v, extra; };
|
|
|
63 |
struct uint128_extra { struct uint128 v; uint64_t extra; };
|
|
|
64 |
#endif
|
|
|
65 |
|
|
|
66 |
typedef unsigned char bool;
|
|
|
67 |
typedef double float64_t;
|
|
|
68 |
union ui64_f64 { uint64_t ui; float64_t f; };
|
|
|
69 |
union ui128_f128 { struct uint128 ui; float128_t f; };
|
|
|
70 |
|
|
|
71 |
/*----------------------------------------------------------------------------
|
|
|
72 |
| The bit pattern for a default generated 128-bit floating-point NaN.
|
|
|
73 |
*----------------------------------------------------------------------------*/
|
|
|
74 |
#define defaultNaNF128UI96 0xFFFF8000
|
|
|
75 |
#define defaultNaNF128UI64 0
|
|
|
76 |
#define defaultNaNF128UI32 0
|
|
|
77 |
#define defaultNaNF128UI0 0
|
|
|
78 |
|
|
|
79 |
struct commonNaN {
|
|
|
80 |
bool sign;
|
|
|
81 |
#ifdef LITTLEENDIAN
|
|
|
82 |
uint64_t v0, v64;
|
|
|
83 |
#else
|
|
|
84 |
uint64_t v64, v0;
|
|
|
85 |
#endif
|
|
|
86 |
};
|
|
|
87 |
struct exp16_sig64 { int_fast16_t exp; uint_fast64_t sig; };
|
|
|
88 |
|
|
4993
|
89 |
#endif // SUPPORT_QUADFLOAT
|
|
|
90 |
|
|
4992
|
91 |
%}
|
|
|
92 |
! !
|
|
|
93 |
|
|
|
94 |
!QuadFloat primitiveVariables!
|
|
|
95 |
%{
|
|
4993
|
96 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
97 |
|
|
4992
|
98 |
uint_fast8_t softfloat_exceptionFlags;
|
|
|
99 |
uint_fast8_t softfloat_roundingMode = softfloat_round_near_even;
|
|
|
100 |
uint_fast8_t softfloat_detectTininess = init_detectTininess;
|
|
|
101 |
uint_fast8_t softfloat_exceptionFlags = 0;
|
|
|
102 |
|
|
|
103 |
const uint_least8_t softfloat_countLeadingZeros8[256] = {
|
|
|
104 |
8, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
|
|
|
105 |
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
|
|
|
106 |
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
107 |
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
108 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
109 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
110 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
111 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
112 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
113 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
114 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
115 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
116 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
117 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
118 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
119 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
|
|
|
120 |
};
|
|
4993
|
121 |
#endif // SUPPORT_QUADFLOAT
|
|
|
122 |
|
|
4991
|
123 |
%}
|
|
|
124 |
! !
|
|
|
125 |
|
|
4992
|
126 |
!QuadFloat primitiveFunctions!
|
|
|
127 |
%{
|
|
4993
|
128 |
#ifdef SUPPORT_QUADFLOAT
|
|
4992
|
129 |
|
|
|
130 |
#if defined(LITTLEENDIAN) || defined(__LSB_FIRST__)
|
|
|
131 |
#define wordIncr 1
|
|
|
132 |
#define indexWord( total, n ) (n)
|
|
|
133 |
#define indexWordHi( total ) ((total) - 1)
|
|
|
134 |
#define indexWordLo( total ) 0
|
|
|
135 |
#define indexMultiword( total, m, n ) (n)
|
|
|
136 |
#define indexMultiwordHi( total, n ) ((total) - (n))
|
|
|
137 |
#define indexMultiwordLo( total, n ) 0
|
|
|
138 |
#define indexMultiwordHiBut( total, n ) (n)
|
|
|
139 |
#define indexMultiwordLoBut( total, n ) 0
|
|
|
140 |
#define INIT_UINTM4( v3, v2, v1, v0 ) { v0, v1, v2, v3 }
|
|
|
141 |
#else
|
|
|
142 |
#define wordIncr -1
|
|
|
143 |
#define indexWord( total, n ) ((total) - 1 - (n))
|
|
|
144 |
#define indexWordHi( total ) 0
|
|
|
145 |
#define indexWordLo( total ) ((total) - 1)
|
|
|
146 |
#define indexMultiword( total, m, n ) ((total) - 1 - (m))
|
|
|
147 |
#define indexMultiwordHi( total, n ) 0
|
|
|
148 |
#define indexMultiwordLo( total, n ) ((total) - (n))
|
|
|
149 |
#define indexMultiwordHiBut( total, n ) 0
|
|
|
150 |
#define indexMultiwordLoBut( total, n ) (n)
|
|
|
151 |
#define INIT_UINTM4( v3, v2, v1, v0 ) { v3, v2, v1, v0 }
|
|
|
152 |
#endif
|
|
|
153 |
|
|
|
154 |
#define signF64UI( a ) ((bool) ((uint64_t) (a)>>63))
|
|
|
155 |
#define expF64UI( a ) ((int_fast16_t) ((a)>>52) & 0x7FF)
|
|
|
156 |
#define fracF64UI( a ) ((a) & UINT64_C( 0x000FFFFFFFFFFFFF ))
|
|
|
157 |
#define softfloat_f64UIToCommonNaN( uiA, zPtr ) if ( ! ((uiA) & UINT64_C( 0x0008000000000000 )) ) softfloat_raiseFlags( softfloat_flag_invalid )
|
|
|
158 |
|
|
|
159 |
#define signF128UI64( a64 ) ((bool) ((uint64_t) (a64)>>63))
|
|
|
160 |
#define expF128UI64( a64 ) ((int_fast32_t) ((a64)>>48) & 0x7FFF)
|
|
|
161 |
#define fracF128UI64( a64 ) ((a64) & UINT64_C( 0x0000FFFFFFFFFFFF ))
|
|
|
162 |
#define packToF128UI64( sign, exp, sig64 ) (((uint_fast64_t) (sign)<<63) + ((uint_fast64_t) (exp)<<48) + (sig64))
|
|
|
163 |
|
|
|
164 |
#define signF128UI96( a96 ) ((bool) ((uint32_t) (a96)>>31))
|
|
|
165 |
#define expF128UI96( a96 ) ((int32_t) ((a96)>>16) & 0x7FFF)
|
|
|
166 |
#define fracF128UI96( a96 ) ((a96) & 0x0000FFFF)
|
|
|
167 |
#define packToF128UI96( sign, exp, sig96 ) (((uint32_t) (sign)<<31) + ((uint32_t) (exp)<<16) + (sig96))
|
|
|
168 |
|
|
|
169 |
#define isNaNF128UI( a64, a0 ) (((~(a64) & UINT64_C( 0x7FFF000000000000 )) == 0) && (a0 || ((a64) & UINT64_C( 0x0000FFFFFFFFFFFF ))))
|
|
|
170 |
#define softfloat_isSigNaNF128UI( uiA64, uiA0 ) ((((uiA64) & UINT64_C( 0x7FFF800000000000 )) == UINT64_C( 0x7FFF000000000000 )) && ((uiA0) || ((uiA64) & UINT64_C( 0x00007FFFFFFFFFFF ))))
|
|
|
171 |
|
|
5000
|
172 |
/*----------------------------------------------------------------------------
|
|
|
173 |
| This function or macro is the same as 'softfloat_shortShiftLeftM' with
|
|
|
174 |
| 'size_words' = 5 (N = 160).
|
|
|
175 |
*----------------------------------------------------------------------------*/
|
|
|
176 |
#define softfloat_shortShiftLeft160M( aPtr, dist, zPtr ) softfloat_shortShiftLeftM( 5, aPtr, dist, zPtr )
|
|
|
177 |
/*----------------------------------------------------------------------------
|
|
|
178 |
| This function or macro is the same as 'softfloat_shiftRightJamM' with
|
|
|
179 |
| 'size_words' = 5 (N = 160).
|
|
|
180 |
*----------------------------------------------------------------------------*/
|
|
|
181 |
#define softfloat_shiftRightJam160M( aPtr, dist, zPtr ) softfloat_shiftRightJamM( 5, aPtr, dist, zPtr )
|
|
|
182 |
/*----------------------------------------------------------------------------
|
|
|
183 |
| This function or macro is the same as 'softfloat_shortShiftLeftM' with
|
|
|
184 |
| 'size_words' = 4 (N = 128).
|
|
|
185 |
*----------------------------------------------------------------------------*/
|
|
|
186 |
#define softfloat_shortShiftLeft128M( aPtr, dist, zPtr ) softfloat_shortShiftLeftM( 4, aPtr, dist, zPtr )
|
|
|
187 |
|
|
|
188 |
static inline void
|
|
|
189 |
softfloat_raiseFlags( uint_fast8_t flags ) {
|
|
|
190 |
softfloat_exceptionFlags |= flags;
|
|
|
191 |
}
|
|
|
192 |
|
|
|
193 |
#if 1
|
|
|
194 |
static inline void
|
|
|
195 |
softfloat_commonNaNToF128M( const struct commonNaN *aPtr, uint32_t *zWPtr )
|
|
4992
|
196 |
{
|
|
|
197 |
zWPtr[indexWord( 4, 3 )] = defaultNaNF128UI96;
|
|
|
198 |
zWPtr[indexWord( 4, 2 )] = defaultNaNF128UI64;
|
|
|
199 |
zWPtr[indexWord( 4, 1 )] = defaultNaNF128UI32;
|
|
|
200 |
zWPtr[indexWord( 4, 0 )] = defaultNaNF128UI0;
|
|
|
201 |
}
|
|
|
202 |
#else
|
|
|
203 |
# define softfloat_commonNaNToF128M( aPtr, zWPtr ) \
|
|
|
204 |
{ \
|
|
|
205 |
(zWPtr)[indexWord( 4, 3 )] = defaultNaNF128UI96; \
|
|
|
206 |
(zWPtr)[indexWord( 4, 2 )] = defaultNaNF128UI64; \
|
|
|
207 |
(zWPtr)[indexWord( 4, 1 )] = defaultNaNF128UI32; \
|
|
|
208 |
(zWPtr)[indexWord( 4, 0 )] = defaultNaNF128UI0; \
|
|
|
209 |
}
|
|
|
210 |
#endif
|
|
|
211 |
|
|
5000
|
212 |
static inline void
|
|
|
213 |
softfloat_invalidF128M( uint32_t *zWPtr ) {
|
|
|
214 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
215 |
zWPtr[indexWord( 4, 3 )] = defaultNaNF128UI96;
|
|
|
216 |
zWPtr[indexWord( 4, 2 )] = defaultNaNF128UI64;
|
|
|
217 |
zWPtr[indexWord( 4, 1 )] = defaultNaNF128UI32;
|
|
|
218 |
zWPtr[indexWord( 4, 0 )] = defaultNaNF128UI0;
|
|
|
219 |
}
|
|
|
220 |
|
|
|
221 |
static bool
|
|
|
222 |
f128M_isSignalingNaN( const float128_t *aPtr )
|
|
|
223 |
{
|
|
|
224 |
const uint32_t *aWPtr;
|
|
|
225 |
uint32_t uiA96;
|
|
|
226 |
|
|
|
227 |
aWPtr = (const uint32_t *) aPtr;
|
|
|
228 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
229 |
if ( (uiA96 & 0x7FFF8000) != 0x7FFF0000 ) return 0;
|
|
|
230 |
return
|
|
|
231 |
((uiA96 & 0x00007FFF) != 0)
|
|
|
232 |
|| ((aWPtr[indexWord( 4, 2 )] | aWPtr[indexWord( 4, 1 )]
|
|
|
233 |
| aWPtr[indexWord( 4, 0 )])
|
|
|
234 |
!= 0);
|
|
|
235 |
}
|
|
|
236 |
|
|
|
237 |
static void
|
|
|
238 |
softfloat_shortShiftRightJamM(
|
|
|
239 |
uint_fast8_t size_words,
|
|
|
240 |
const uint64_t *aPtr,
|
|
|
241 |
uint_fast8_t dist,
|
|
|
242 |
uint64_t *zPtr
|
|
|
243 |
)
|
|
|
244 |
{
|
|
|
245 |
uint_fast8_t uNegDist;
|
|
|
246 |
unsigned int index, lastIndex;
|
|
|
247 |
uint64_t partWordZ, wordA;
|
|
|
248 |
|
|
|
249 |
uNegDist = -dist;
|
|
|
250 |
index = indexWordLo( size_words );
|
|
|
251 |
lastIndex = indexWordHi( size_words );
|
|
|
252 |
wordA = aPtr[index];
|
|
|
253 |
partWordZ = wordA>>dist;
|
|
|
254 |
if ( partWordZ<<dist != wordA ) partWordZ |= 1;
|
|
|
255 |
while ( index != lastIndex ) {
|
|
|
256 |
wordA = aPtr[index + wordIncr];
|
|
|
257 |
zPtr[index] = wordA<<(uNegDist & 63) | partWordZ;
|
|
|
258 |
index += wordIncr;
|
|
|
259 |
partWordZ = wordA>>dist;
|
|
|
260 |
}
|
|
|
261 |
zPtr[index] = partWordZ;
|
|
4992
|
262 |
}
|
|
|
263 |
|
|
5000
|
264 |
void
|
|
|
265 |
softfloat_shiftRightJamM(
|
|
|
266 |
uint_fast8_t size_words,
|
|
|
267 |
const uint32_t *aPtr,
|
|
|
268 |
uint32_t dist,
|
|
|
269 |
uint32_t *zPtr
|
|
|
270 |
)
|
|
|
271 |
{
|
|
|
272 |
uint32_t wordJam, wordDist, *ptr;
|
|
|
273 |
uint_fast8_t i, innerDist;
|
|
|
274 |
|
|
|
275 |
wordJam = 0;
|
|
|
276 |
wordDist = dist>>5;
|
|
|
277 |
if ( wordDist ) {
|
|
|
278 |
if ( size_words < wordDist ) wordDist = size_words;
|
|
|
279 |
ptr = (uint32_t *) (aPtr + indexMultiwordLo( size_words, wordDist ));
|
|
|
280 |
i = wordDist;
|
|
|
281 |
do {
|
|
|
282 |
wordJam = *ptr++;
|
|
|
283 |
if ( wordJam ) break;
|
|
|
284 |
--i;
|
|
|
285 |
} while ( i );
|
|
|
286 |
ptr = zPtr;
|
|
|
287 |
}
|
|
|
288 |
if ( wordDist < size_words ) {
|
|
|
289 |
aPtr += indexMultiwordHiBut( size_words, wordDist );
|
|
|
290 |
innerDist = dist & 31;
|
|
|
291 |
if ( innerDist ) {
|
|
|
292 |
softfloat_shortShiftRightJamM(
|
|
|
293 |
size_words - wordDist,
|
|
|
294 |
aPtr,
|
|
|
295 |
innerDist,
|
|
|
296 |
zPtr + indexMultiwordLoBut( size_words, wordDist )
|
|
|
297 |
);
|
|
|
298 |
if ( ! wordDist ) goto wordJam;
|
|
|
299 |
} else {
|
|
|
300 |
aPtr += indexWordLo( size_words - wordDist );
|
|
|
301 |
ptr = zPtr + indexWordLo( size_words );
|
|
|
302 |
for ( i = size_words - wordDist; i; --i ) {
|
|
|
303 |
*ptr = *aPtr;
|
|
|
304 |
aPtr += wordIncr;
|
|
|
305 |
ptr += wordIncr;
|
|
|
306 |
}
|
|
|
307 |
}
|
|
|
308 |
ptr = zPtr + indexMultiwordHi( size_words, wordDist );
|
|
|
309 |
}
|
|
|
310 |
do {
|
|
|
311 |
*ptr++ = 0;
|
|
|
312 |
--wordDist;
|
|
|
313 |
} while ( wordDist );
|
|
|
314 |
wordJam:
|
|
|
315 |
if ( wordJam ) zPtr[indexWordLo( size_words )] |= 1;
|
|
|
316 |
|
|
|
317 |
}
|
|
|
318 |
|
|
|
319 |
void
|
|
|
320 |
softfloat_shortShiftLeftM(
|
|
|
321 |
uint_fast8_t size_words,
|
|
|
322 |
const uint32_t *aPtr,
|
|
|
323 |
uint_fast8_t dist,
|
|
|
324 |
uint32_t *zPtr
|
|
|
325 |
)
|
|
|
326 |
{
|
|
|
327 |
uint_fast8_t uNegDist;
|
|
|
328 |
unsigned int index, lastIndex;
|
|
|
329 |
uint32_t partWordZ, wordA;
|
|
|
330 |
|
|
|
331 |
uNegDist = -dist;
|
|
|
332 |
index = indexWordHi( size_words );
|
|
|
333 |
lastIndex = indexWordLo( size_words );
|
|
|
334 |
partWordZ = aPtr[index]<<dist;
|
|
|
335 |
while ( index != lastIndex ) {
|
|
|
336 |
wordA = aPtr[index - wordIncr];
|
|
|
337 |
zPtr[index] = partWordZ | wordA>>(uNegDist & 31);
|
|
|
338 |
index -= wordIncr;
|
|
|
339 |
partWordZ = wordA<<dist;
|
|
|
340 |
}
|
|
|
341 |
zPtr[index] = partWordZ;
|
|
|
342 |
}
|
|
|
343 |
|
|
|
344 |
static void
|
|
|
345 |
softfloat_shiftLeftM(
|
|
|
346 |
uint_fast8_t size_words,
|
|
|
347 |
const uint32_t *aPtr,
|
|
|
348 |
uint32_t dist,
|
|
|
349 |
uint32_t *zPtr
|
|
|
350 |
)
|
|
|
351 |
{
|
|
|
352 |
uint32_t wordDist;
|
|
|
353 |
uint_fast8_t innerDist;
|
|
|
354 |
uint32_t *destPtr;
|
|
|
355 |
uint_fast8_t i;
|
|
|
356 |
|
|
|
357 |
wordDist = dist>>5;
|
|
|
358 |
if ( wordDist < size_words ) {
|
|
|
359 |
aPtr += indexMultiwordLoBut( size_words, wordDist );
|
|
|
360 |
innerDist = dist & 31;
|
|
|
361 |
if ( innerDist ) {
|
|
|
362 |
softfloat_shortShiftLeftM(
|
|
|
363 |
size_words - wordDist,
|
|
|
364 |
aPtr,
|
|
|
365 |
innerDist,
|
|
|
366 |
zPtr + indexMultiwordHiBut( size_words, wordDist )
|
|
|
367 |
);
|
|
|
368 |
if ( ! wordDist ) return;
|
|
|
369 |
} else {
|
|
|
370 |
aPtr += indexWordHi( size_words - wordDist );
|
|
|
371 |
destPtr = zPtr + indexWordHi( size_words );
|
|
|
372 |
for ( i = size_words - wordDist; i; --i ) {
|
|
|
373 |
*destPtr = *aPtr;
|
|
|
374 |
aPtr -= wordIncr;
|
|
|
375 |
destPtr -= wordIncr;
|
|
|
376 |
}
|
|
|
377 |
}
|
|
|
378 |
zPtr += indexMultiwordLo( size_words, wordDist );
|
|
|
379 |
} else {
|
|
|
380 |
wordDist = size_words;
|
|
|
381 |
}
|
|
|
382 |
do {
|
|
|
383 |
*zPtr++ = 0;
|
|
|
384 |
--wordDist;
|
|
|
385 |
} while ( wordDist );
|
|
|
386 |
}
|
|
|
387 |
|
|
|
388 |
/*----------------------------------------------------------------------------
|
|
|
389 |
| This function or macro is the same as 'softfloat_shiftLeftM' with
|
|
|
390 |
| 'size_words' = 4 (N = 128).
|
|
|
391 |
*----------------------------------------------------------------------------*/
|
|
|
392 |
#define softfloat_shiftLeft128M( aPtr, dist, zPtr ) softfloat_shiftLeftM( 4, aPtr, dist, zPtr )
|
|
|
393 |
|
|
4992
|
394 |
bool
|
|
5000
|
395 |
softfloat_isNaNF128M( const uint32_t *aWPtr )
|
|
|
396 |
{
|
|
|
397 |
uint32_t uiA96;
|
|
|
398 |
|
|
|
399 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
400 |
if ( (~uiA96 & 0x7FFF0000) != 0 ) return 0;
|
|
|
401 |
return
|
|
|
402 |
((uiA96 & 0x0000FFFF) != 0)
|
|
|
403 |
|| ((aWPtr[indexWord( 4, 2 )] | aWPtr[indexWord( 4, 1 )]
|
|
|
404 |
| aWPtr[indexWord( 4, 0 )])
|
|
|
405 |
!= 0);
|
|
|
406 |
}
|
|
|
407 |
|
|
|
408 |
static inline uint_fast8_t
|
|
|
409 |
softfloat_countLeadingZeros32( uint32_t a ) {
|
|
|
410 |
return a ? __builtin_clz( a ) : 32;
|
|
|
411 |
}
|
|
|
412 |
|
|
|
413 |
static inline bool
|
|
4992
|
414 |
softfloat_eq128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 ){
|
|
|
415 |
return (a64 == b64) && (a0 == b0);
|
|
|
416 |
}
|
|
|
417 |
|
|
5000
|
418 |
static inline bool
|
|
4992
|
419 |
softfloat_lt128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 ){
|
|
|
420 |
return (a64 < b64) || ((a64 == b64) && (a0 < b0));
|
|
|
421 |
}
|
|
|
422 |
|
|
5000
|
423 |
void
|
|
|
424 |
softfloat_propagateNaNF128M(
|
|
|
425 |
const uint32_t *aWPtr, const uint32_t *bWPtr, uint32_t *zWPtr )
|
|
|
426 |
{
|
|
|
427 |
bool isSigNaNA;
|
|
|
428 |
const uint32_t *ptr;
|
|
|
429 |
bool isSigNaNB;
|
|
|
430 |
uint32_t uiA96, uiB96, wordMagA, wordMagB;
|
|
|
431 |
|
|
|
432 |
isSigNaNA = f128M_isSignalingNaN( (const float128_t *) aWPtr );
|
|
|
433 |
ptr = aWPtr;
|
|
|
434 |
if ( ! bWPtr ) {
|
|
|
435 |
if ( isSigNaNA ) softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
436 |
goto copy;
|
|
|
437 |
}
|
|
|
438 |
isSigNaNB = f128M_isSignalingNaN( (const float128_t *) bWPtr );
|
|
|
439 |
if ( isSigNaNA | isSigNaNB ) {
|
|
|
440 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
441 |
if ( isSigNaNA ) {
|
|
|
442 |
if ( isSigNaNB ) goto returnLargerUIMag;
|
|
|
443 |
if ( softfloat_isNaNF128M( bWPtr ) ) goto copyB;
|
|
|
444 |
goto copy;
|
|
|
445 |
} else {
|
|
|
446 |
if ( softfloat_isNaNF128M( aWPtr ) ) goto copy;
|
|
|
447 |
goto copyB;
|
|
|
448 |
}
|
|
|
449 |
}
|
|
|
450 |
returnLargerUIMag:
|
|
|
451 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
452 |
uiB96 = bWPtr[indexWordHi( 4 )];
|
|
|
453 |
wordMagA = uiA96 & 0x7FFFFFFF;
|
|
|
454 |
wordMagB = uiB96 & 0x7FFFFFFF;
|
|
|
455 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
456 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
457 |
wordMagA = aWPtr[indexWord( 4, 2 )];
|
|
|
458 |
wordMagB = bWPtr[indexWord( 4, 2 )];
|
|
|
459 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
460 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
461 |
wordMagA = aWPtr[indexWord( 4, 1 )];
|
|
|
462 |
wordMagB = bWPtr[indexWord( 4, 1 )];
|
|
|
463 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
464 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
465 |
wordMagA = aWPtr[indexWord( 4, 0 )];
|
|
|
466 |
wordMagB = bWPtr[indexWord( 4, 0 )];
|
|
|
467 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
468 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
469 |
if ( uiA96 < uiB96 ) goto copy;
|
|
|
470 |
copyB:
|
|
|
471 |
ptr = bWPtr;
|
|
|
472 |
copy:
|
|
|
473 |
zWPtr[indexWordHi( 4 )] = ptr[indexWordHi( 4 )] | 0x00008000;
|
|
|
474 |
zWPtr[indexWord( 4, 2 )] = ptr[indexWord( 4, 2 )];
|
|
|
475 |
zWPtr[indexWord( 4, 1 )] = ptr[indexWord( 4, 1 )];
|
|
|
476 |
zWPtr[indexWord( 4, 0 )] = ptr[indexWord( 4, 0 )];
|
|
|
477 |
}
|
|
|
478 |
|
|
|
479 |
static inline uint_fast8_t
|
|
4992
|
480 |
softfloat_countLeadingZeros64( uint64_t a )
|
|
|
481 |
{
|
|
|
482 |
uint_fast8_t count;
|
|
|
483 |
uint32_t a32;
|
|
|
484 |
|
|
|
485 |
count = 0;
|
|
|
486 |
a32 = a>>32;
|
|
|
487 |
if ( ! a32 ) {
|
|
|
488 |
count = 32;
|
|
|
489 |
a32 = a;
|
|
|
490 |
}
|
|
|
491 |
/*------------------------------------------------------------------------
|
|
|
492 |
| From here, result is current count + count leading zeros of `a32'.
|
|
|
493 |
*------------------------------------------------------------------------*/
|
|
|
494 |
if ( a32 < 0x10000 ) {
|
|
|
495 |
count += 16;
|
|
|
496 |
a32 <<= 16;
|
|
|
497 |
}
|
|
|
498 |
if ( a32 < 0x1000000 ) {
|
|
|
499 |
count += 8;
|
|
|
500 |
a32 <<= 8;
|
|
|
501 |
}
|
|
|
502 |
count += softfloat_countLeadingZeros8[a32>>24];
|
|
|
503 |
return count;
|
|
|
504 |
}
|
|
|
505 |
|
|
5000
|
506 |
static inline struct exp16_sig64
|
|
4992
|
507 |
softfloat_normSubnormalF64Sig( uint_fast64_t sig )
|
|
|
508 |
{
|
|
|
509 |
int_fast8_t shiftDist;
|
|
|
510 |
struct exp16_sig64 z;
|
|
|
511 |
|
|
|
512 |
shiftDist = softfloat_countLeadingZeros64( sig ) - 11;
|
|
|
513 |
z.exp = 1 - shiftDist;
|
|
|
514 |
z.sig = sig<<shiftDist;
|
|
|
515 |
return z;
|
|
|
516 |
}
|
|
|
517 |
|
|
5000
|
518 |
static inline struct uint128_extra
|
|
4992
|
519 |
softfloat_shiftRightJam128Extra(
|
|
|
520 |
uint64_t a64, uint64_t a0, uint64_t extra, uint_fast32_t dist )
|
|
|
521 |
{
|
|
|
522 |
uint_fast8_t u8NegDist;
|
|
|
523 |
struct uint128_extra z;
|
|
|
524 |
|
|
|
525 |
u8NegDist = -dist;
|
|
|
526 |
if ( dist < 64 ) {
|
|
|
527 |
z.v.v64 = a64>>dist;
|
|
|
528 |
z.v.v0 = a64<<(u8NegDist & 63) | a0>>dist;
|
|
|
529 |
z.extra = a0<<(u8NegDist & 63);
|
|
|
530 |
} else {
|
|
|
531 |
z.v.v64 = 0;
|
|
|
532 |
if ( dist == 64 ) {
|
|
|
533 |
z.v.v0 = a64;
|
|
|
534 |
z.extra = a0;
|
|
|
535 |
} else {
|
|
|
536 |
extra |= a0;
|
|
|
537 |
if ( dist < 128 ) {
|
|
|
538 |
z.v.v0 = a64>>(dist & 63);
|
|
|
539 |
z.extra = a64<<(u8NegDist & 63);
|
|
|
540 |
} else {
|
|
|
541 |
z.v.v0 = 0;
|
|
|
542 |
z.extra = (dist == 128) ? a64 : (a64 != 0);
|
|
|
543 |
}
|
|
|
544 |
}
|
|
|
545 |
}
|
|
|
546 |
z.extra |= (extra != 0);
|
|
|
547 |
return z;
|
|
|
548 |
}
|
|
|
549 |
|
|
5000
|
550 |
static inline struct uint128_extra
|
|
4992
|
551 |
softfloat_shortShiftRightJam128Extra(
|
|
|
552 |
uint64_t a64, uint64_t a0, uint64_t extra, uint_fast8_t dist )
|
|
|
553 |
{
|
|
|
554 |
uint_fast8_t negDist = -dist;
|
|
|
555 |
struct uint128_extra z;
|
|
|
556 |
z.v.v64 = a64>>dist;
|
|
|
557 |
z.v.v0 = a64<<(negDist & 63) | a0>>dist;
|
|
|
558 |
z.extra = a0<<(negDist & 63) | (extra != 0);
|
|
|
559 |
return z;
|
|
|
560 |
}
|
|
|
561 |
|
|
5000
|
562 |
static inline struct uint128
|
|
|
563 |
softfloat_shiftRightJam128( uint64_t a64, uint64_t a0, uint_fast32_t dist )
|
|
4992
|
564 |
{
|
|
|
565 |
uint_fast8_t u8NegDist;
|
|
|
566 |
struct uint128 z;
|
|
|
567 |
|
|
|
568 |
if ( dist < 64 ) {
|
|
|
569 |
u8NegDist = -dist;
|
|
|
570 |
z.v64 = a64>>dist;
|
|
|
571 |
z.v0 =
|
|
|
572 |
a64<<(u8NegDist & 63) | a0>>dist
|
|
|
573 |
| ((uint64_t) (a0<<(u8NegDist & 63)) != 0);
|
|
|
574 |
} else {
|
|
|
575 |
z.v64 = 0;
|
|
|
576 |
z.v0 =
|
|
|
577 |
(dist < 127)
|
|
|
578 |
? a64>>(dist & 63)
|
|
|
579 |
| (((a64 & (((uint_fast64_t) 1<<(dist & 63)) - 1)) | a0)
|
|
|
580 |
!= 0)
|
|
|
581 |
: ((a64 | a0) != 0);
|
|
|
582 |
}
|
|
|
583 |
return z;
|
|
|
584 |
}
|
|
|
585 |
|
|
5000
|
586 |
static inline struct uint128
|
|
4992
|
587 |
softfloat_shortShiftLeft128( uint64_t a64, uint64_t a0, uint_fast8_t dist )
|
|
|
588 |
{
|
|
|
589 |
struct uint128 z;
|
|
|
590 |
z.v64 = a64<<dist | a0>>(-dist & 63);
|
|
|
591 |
z.v0 = a0<<dist;
|
|
|
592 |
return z;
|
|
|
593 |
}
|
|
|
594 |
|
|
5000
|
595 |
static int_fast8_t
|
|
|
596 |
softfloat_compare128M( const uint32_t *aPtr, const uint32_t *bPtr )
|
|
|
597 |
{
|
|
|
598 |
unsigned int index, lastIndex;
|
|
|
599 |
uint32_t wordA, wordB;
|
|
|
600 |
|
|
|
601 |
index = indexWordHi( 4 );
|
|
|
602 |
lastIndex = indexWordLo( 4 );
|
|
|
603 |
for (;;) {
|
|
|
604 |
wordA = aPtr[index];
|
|
|
605 |
wordB = bPtr[index];
|
|
|
606 |
if ( wordA != wordB ) return (wordA < wordB) ? -1 : 1;
|
|
|
607 |
if ( index == lastIndex ) break;
|
|
|
608 |
index -= wordIncr;
|
|
|
609 |
}
|
|
|
610 |
return 0;
|
|
|
611 |
}
|
|
|
612 |
|
|
|
613 |
static void
|
|
|
614 |
softfloat_roundPackMToF128M(
|
|
|
615 |
bool sign, int32_t exp, uint32_t *extSigPtr, uint32_t *zWPtr )
|
|
|
616 |
{
|
|
|
617 |
uint_fast8_t roundingMode;
|
|
|
618 |
bool roundNearEven;
|
|
|
619 |
uint32_t sigExtra;
|
|
|
620 |
bool doIncrement, isTiny;
|
|
|
621 |
static const uint32_t maxSig[4] =
|
|
|
622 |
INIT_UINTM4( 0x0001FFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF );
|
|
|
623 |
uint32_t ui, uj;
|
|
|
624 |
|
|
|
625 |
/*------------------------------------------------------------------------
|
|
|
626 |
*------------------------------------------------------------------------*/
|
|
|
627 |
roundingMode = softfloat_roundingMode;
|
|
|
628 |
roundNearEven = (roundingMode == softfloat_round_near_even);
|
|
|
629 |
sigExtra = extSigPtr[indexWordLo( 5 )];
|
|
|
630 |
doIncrement = (0x80000000 <= sigExtra);
|
|
|
631 |
if ( ! roundNearEven && (roundingMode != softfloat_round_near_maxMag) ) {
|
|
|
632 |
doIncrement =
|
|
|
633 |
(roundingMode
|
|
|
634 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
635 |
&& sigExtra;
|
|
|
636 |
}
|
|
|
637 |
/*------------------------------------------------------------------------
|
|
|
638 |
*------------------------------------------------------------------------*/
|
|
|
639 |
if ( 0x7FFD <= (uint32_t) exp ) {
|
|
|
640 |
if ( exp < 0 ) {
|
|
|
641 |
/*----------------------------------------------------------------
|
|
|
642 |
*----------------------------------------------------------------*/
|
|
|
643 |
isTiny =
|
|
|
644 |
(softfloat_detectTininess
|
|
|
645 |
== softfloat_tininess_beforeRounding)
|
|
|
646 |
|| (exp < -1)
|
|
|
647 |
|| ! doIncrement
|
|
|
648 |
|| (softfloat_compare128M(
|
|
|
649 |
extSigPtr + indexMultiwordHi( 5, 4 ), maxSig )
|
|
|
650 |
< 0);
|
|
|
651 |
softfloat_shiftRightJam160M( extSigPtr, -exp, extSigPtr );
|
|
|
652 |
exp = 0;
|
|
|
653 |
sigExtra = extSigPtr[indexWordLo( 5 )];
|
|
|
654 |
if ( isTiny && sigExtra ) {
|
|
|
655 |
softfloat_raiseFlags( softfloat_flag_underflow );
|
|
|
656 |
}
|
|
|
657 |
doIncrement = (0x80000000 <= sigExtra);
|
|
|
658 |
if (
|
|
|
659 |
! roundNearEven
|
|
|
660 |
&& (roundingMode != softfloat_round_near_maxMag)
|
|
|
661 |
) {
|
|
|
662 |
doIncrement =
|
|
|
663 |
(roundingMode
|
|
|
664 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
665 |
&& sigExtra;
|
|
|
666 |
}
|
|
|
667 |
} else if (
|
|
|
668 |
(0x7FFD < exp)
|
|
|
669 |
|| ((exp == 0x7FFD) && doIncrement
|
|
|
670 |
&& (softfloat_compare128M(
|
|
|
671 |
extSigPtr + indexMultiwordHi( 5, 4 ), maxSig )
|
|
|
672 |
== 0))
|
|
|
673 |
) {
|
|
|
674 |
/*----------------------------------------------------------------
|
|
|
675 |
*----------------------------------------------------------------*/
|
|
|
676 |
softfloat_raiseFlags(
|
|
|
677 |
softfloat_flag_overflow | softfloat_flag_inexact );
|
|
|
678 |
if (
|
|
|
679 |
roundNearEven
|
|
|
680 |
|| (roundingMode == softfloat_round_near_maxMag)
|
|
|
681 |
|| (roundingMode
|
|
|
682 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
683 |
) {
|
|
|
684 |
ui = packToF128UI96( sign, 0x7FFF, 0 );
|
|
|
685 |
uj = 0;
|
|
|
686 |
} else {
|
|
|
687 |
ui = packToF128UI96( sign, 0x7FFE, 0x0000FFFF );
|
|
|
688 |
uj = 0xFFFFFFFF;
|
|
|
689 |
}
|
|
|
690 |
zWPtr[indexWordHi( 4 )] = ui;
|
|
|
691 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
692 |
zWPtr[indexWord( 4, 1 )] = uj;
|
|
|
693 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
694 |
return;
|
|
|
695 |
}
|
|
|
696 |
}
|
|
|
697 |
/*------------------------------------------------------------------------
|
|
|
698 |
*------------------------------------------------------------------------*/
|
|
|
699 |
uj = extSigPtr[indexWord( 5, 1 )];
|
|
|
700 |
if ( sigExtra ) {
|
|
|
701 |
softfloat_exceptionFlags |= softfloat_flag_inexact;
|
|
|
702 |
#ifdef SOFTFLOAT_ROUND_ODD
|
|
|
703 |
if ( roundingMode == softfloat_round_odd ) {
|
|
|
704 |
uj |= 1;
|
|
|
705 |
goto noIncrementPackReturn;
|
|
|
706 |
}
|
|
|
707 |
#endif
|
|
|
708 |
}
|
|
|
709 |
if ( doIncrement ) {
|
|
|
710 |
++uj;
|
|
|
711 |
if ( uj ) {
|
|
|
712 |
if ( ! (sigExtra & 0x7FFFFFFF) && roundNearEven ) uj &= ~1;
|
|
|
713 |
zWPtr[indexWord( 4, 2 )] = extSigPtr[indexWord( 5, 3 )];
|
|
|
714 |
zWPtr[indexWord( 4, 1 )] = extSigPtr[indexWord( 5, 2 )];
|
|
|
715 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
716 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
717 |
} else {
|
|
|
718 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
719 |
ui = extSigPtr[indexWord( 5, 2 )] + 1;
|
|
|
720 |
zWPtr[indexWord( 4, 1 )] = ui;
|
|
|
721 |
uj = extSigPtr[indexWord( 5, 3 )];
|
|
|
722 |
if ( ui ) {
|
|
|
723 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
724 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
725 |
} else {
|
|
|
726 |
++uj;
|
|
|
727 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
728 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
729 |
if ( ! uj ) ++ui;
|
|
|
730 |
}
|
|
|
731 |
}
|
|
|
732 |
} else {
|
|
|
733 |
noIncrementPackReturn:
|
|
|
734 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
735 |
ui = extSigPtr[indexWord( 5, 2 )];
|
|
|
736 |
zWPtr[indexWord( 4, 1 )] = ui;
|
|
|
737 |
uj |= ui;
|
|
|
738 |
ui = extSigPtr[indexWord( 5, 3 )];
|
|
|
739 |
zWPtr[indexWord( 4, 2 )] = ui;
|
|
|
740 |
uj |= ui;
|
|
|
741 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
742 |
uj |= ui;
|
|
|
743 |
if ( ! uj ) exp = 0;
|
|
|
744 |
}
|
|
|
745 |
zWPtr[indexWordHi( 4 )] = packToF128UI96( sign, exp, ui );
|
|
|
746 |
}
|
|
|
747 |
|
|
|
748 |
static int
|
|
|
749 |
softfloat_shiftNormSigF128M(
|
|
|
750 |
const uint32_t *wPtr, uint_fast8_t shiftDist, uint32_t *sigPtr )
|
|
|
751 |
{
|
|
|
752 |
uint32_t wordSig;
|
|
|
753 |
int32_t exp;
|
|
|
754 |
uint32_t leadingBit;
|
|
|
755 |
|
|
|
756 |
wordSig = wPtr[indexWordHi( 4 )];
|
|
|
757 |
exp = expF128UI96( wordSig );
|
|
|
758 |
if ( exp ) {
|
|
|
759 |
softfloat_shortShiftLeft128M( wPtr, shiftDist, sigPtr );
|
|
|
760 |
leadingBit = 0x00010000<<shiftDist;
|
|
|
761 |
sigPtr[indexWordHi( 4 )] =
|
|
|
762 |
(sigPtr[indexWordHi( 4 )] & (leadingBit - 1)) | leadingBit;
|
|
|
763 |
} else {
|
|
|
764 |
exp = 16;
|
|
|
765 |
wordSig &= 0x7FFFFFFF;
|
|
|
766 |
if ( ! wordSig ) {
|
|
|
767 |
exp = -16;
|
|
|
768 |
wordSig = wPtr[indexWord( 4, 2 )];
|
|
|
769 |
if ( ! wordSig ) {
|
|
|
770 |
exp = -48;
|
|
|
771 |
wordSig = wPtr[indexWord( 4, 1 )];
|
|
|
772 |
if ( ! wordSig ) {
|
|
|
773 |
wordSig = wPtr[indexWord( 4, 0 )];
|
|
|
774 |
if ( ! wordSig ) return -128;
|
|
|
775 |
exp = -80;
|
|
|
776 |
}
|
|
|
777 |
}
|
|
|
778 |
}
|
|
|
779 |
exp -= softfloat_countLeadingZeros32( wordSig );
|
|
|
780 |
softfloat_shiftLeft128M( wPtr, 1 - exp + shiftDist, sigPtr );
|
|
|
781 |
}
|
|
|
782 |
return exp;
|
|
|
783 |
}
|
|
|
784 |
|
|
|
785 |
bool
|
|
|
786 |
softfloat_tryPropagateNaNF128M(
|
|
|
787 |
const uint32_t *aWPtr, const uint32_t *bWPtr, uint32_t *zWPtr )
|
|
|
788 |
{
|
|
|
789 |
|
|
|
790 |
if ( softfloat_isNaNF128M( aWPtr ) || softfloat_isNaNF128M( bWPtr ) ) {
|
|
|
791 |
softfloat_propagateNaNF128M( aWPtr, bWPtr, zWPtr );
|
|
|
792 |
return 1;
|
|
|
793 |
}
|
|
|
794 |
return 0;
|
|
|
795 |
}
|
|
|
796 |
|
|
|
797 |
static inline struct uint128
|
|
4992
|
798 |
softfloat_add128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 )
|
|
|
799 |
{
|
|
|
800 |
struct uint128 z;
|
|
|
801 |
z.v0 = a0 + b0;
|
|
|
802 |
z.v64 = a64 + b64 + (z.v0 < a0);
|
|
|
803 |
return z;
|
|
|
804 |
}
|
|
|
805 |
|
|
5000
|
806 |
static inline struct uint128
|
|
|
807 |
softfloat_sub128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 )
|
|
4992
|
808 |
{
|
|
|
809 |
struct uint128 z;
|
|
|
810 |
z.v0 = a0 - b0;
|
|
|
811 |
z.v64 = a64 - b64;
|
|
|
812 |
z.v64 -= (a0 < b0);
|
|
|
813 |
return z;
|
|
|
814 |
}
|
|
|
815 |
|
|
5000
|
816 |
static void
|
|
|
817 |
softfloat_mul128MTo256M(
|
|
|
818 |
const uint32_t *aPtr, const uint32_t *bPtr, uint32_t *zPtr )
|
|
|
819 |
{
|
|
|
820 |
uint32_t *lastZPtr, wordB;
|
|
|
821 |
uint64_t dwordProd;
|
|
|
822 |
uint32_t wordZ;
|
|
|
823 |
uint_fast8_t carry;
|
|
|
824 |
|
|
|
825 |
bPtr += indexWordLo( 4 );
|
|
|
826 |
lastZPtr = zPtr + indexMultiwordHi( 8, 5 );
|
|
|
827 |
zPtr += indexMultiwordLo( 8, 5 );
|
|
|
828 |
wordB = *bPtr;
|
|
|
829 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 0 )] * wordB;
|
|
|
830 |
zPtr[indexWord( 5, 0 )] = dwordProd;
|
|
|
831 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 1 )] * wordB + (dwordProd>>32);
|
|
|
832 |
zPtr[indexWord( 5, 1 )] = dwordProd;
|
|
|
833 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 2 )] * wordB + (dwordProd>>32);
|
|
|
834 |
zPtr[indexWord( 5, 2 )] = dwordProd;
|
|
|
835 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 3 )] * wordB + (dwordProd>>32);
|
|
|
836 |
zPtr[indexWord( 5, 3 )] = dwordProd;
|
|
|
837 |
zPtr[indexWord( 5, 4 )] = dwordProd>>32;
|
|
|
838 |
do {
|
|
|
839 |
bPtr += wordIncr;
|
|
|
840 |
zPtr += wordIncr;
|
|
|
841 |
wordB = *bPtr;
|
|
|
842 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 0 )] * wordB;
|
|
|
843 |
wordZ = zPtr[indexWord( 5, 0 )] + (uint32_t) dwordProd;
|
|
|
844 |
zPtr[indexWord( 5, 0 )] = wordZ;
|
|
|
845 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
846 |
dwordProd =
|
|
|
847 |
(uint64_t) aPtr[indexWord( 4, 1 )] * wordB + (dwordProd>>32);
|
|
|
848 |
wordZ = zPtr[indexWord( 5, 1 )] + (uint32_t) dwordProd + carry;
|
|
|
849 |
zPtr[indexWord( 5, 1 )] = wordZ;
|
|
|
850 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
851 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
852 |
}
|
|
|
853 |
dwordProd =
|
|
|
854 |
(uint64_t) aPtr[indexWord( 4, 2 )] * wordB + (dwordProd>>32);
|
|
|
855 |
wordZ = zPtr[indexWord( 5, 2 )] + (uint32_t) dwordProd + carry;
|
|
|
856 |
zPtr[indexWord( 5, 2 )] = wordZ;
|
|
|
857 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
858 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
859 |
}
|
|
|
860 |
dwordProd =
|
|
|
861 |
(uint64_t) aPtr[indexWord( 4, 3 )] * wordB + (dwordProd>>32);
|
|
|
862 |
wordZ = zPtr[indexWord( 5, 3 )] + (uint32_t) dwordProd + carry;
|
|
|
863 |
zPtr[indexWord( 5, 3 )] = wordZ;
|
|
|
864 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
865 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
866 |
}
|
|
|
867 |
zPtr[indexWord( 5, 4 )] = (dwordProd>>32) + carry;
|
|
|
868 |
} while ( zPtr != lastZPtr );
|
|
|
869 |
}
|
|
|
870 |
|
|
4992
|
871 |
float128_t
|
|
|
872 |
softfloat_roundPackToF128(
|
|
|
873 |
bool sign,
|
|
|
874 |
int_fast32_t exp,
|
|
|
875 |
uint_fast64_t sig64,
|
|
|
876 |
uint_fast64_t sig0,
|
|
|
877 |
uint_fast64_t sigExtra
|
|
|
878 |
)
|
|
|
879 |
{
|
|
|
880 |
uint_fast8_t roundingMode;
|
|
|
881 |
bool roundNearEven, doIncrement, isTiny;
|
|
|
882 |
struct uint128_extra sig128Extra;
|
|
|
883 |
uint_fast64_t uiZ64, uiZ0;
|
|
|
884 |
struct uint128 sig128;
|
|
|
885 |
union ui128_f128 uZ;
|
|
|
886 |
|
|
|
887 |
/*------------------------------------------------------------------------
|
|
|
888 |
*------------------------------------------------------------------------*/
|
|
|
889 |
roundingMode = softfloat_roundingMode;
|
|
|
890 |
roundNearEven = (roundingMode == softfloat_round_near_even);
|
|
|
891 |
doIncrement = (UINT64_C( 0x8000000000000000 ) <= sigExtra);
|
|
|
892 |
if ( ! roundNearEven && (roundingMode != softfloat_round_near_maxMag) ) {
|
|
|
893 |
doIncrement =
|
|
|
894 |
(roundingMode
|
|
|
895 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
896 |
&& sigExtra;
|
|
|
897 |
}
|
|
|
898 |
/*------------------------------------------------------------------------
|
|
|
899 |
*------------------------------------------------------------------------*/
|
|
|
900 |
if ( 0x7FFD <= (uint32_t) exp ) {
|
|
|
901 |
if ( exp < 0 ) {
|
|
|
902 |
/*----------------------------------------------------------------
|
|
|
903 |
*----------------------------------------------------------------*/
|
|
|
904 |
isTiny =
|
|
|
905 |
(softfloat_detectTininess
|
|
|
906 |
== softfloat_tininess_beforeRounding)
|
|
|
907 |
|| (exp < -1)
|
|
|
908 |
|| ! doIncrement
|
|
|
909 |
|| softfloat_lt128(
|
|
|
910 |
sig64,
|
|
|
911 |
sig0,
|
|
|
912 |
UINT64_C( 0x0001FFFFFFFFFFFF ),
|
|
|
913 |
UINT64_C( 0xFFFFFFFFFFFFFFFF )
|
|
|
914 |
);
|
|
|
915 |
sig128Extra =
|
|
|
916 |
softfloat_shiftRightJam128Extra( sig64, sig0, sigExtra, -exp );
|
|
|
917 |
sig64 = sig128Extra.v.v64;
|
|
|
918 |
sig0 = sig128Extra.v.v0;
|
|
|
919 |
sigExtra = sig128Extra.extra;
|
|
|
920 |
exp = 0;
|
|
|
921 |
if ( isTiny && sigExtra ) {
|
|
|
922 |
softfloat_raiseFlags( softfloat_flag_underflow );
|
|
|
923 |
}
|
|
|
924 |
doIncrement = (UINT64_C( 0x8000000000000000 ) <= sigExtra);
|
|
|
925 |
if (
|
|
|
926 |
! roundNearEven
|
|
|
927 |
&& (roundingMode != softfloat_round_near_maxMag)
|
|
|
928 |
) {
|
|
|
929 |
doIncrement =
|
|
|
930 |
(roundingMode
|
|
|
931 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
932 |
&& sigExtra;
|
|
|
933 |
}
|
|
|
934 |
} else if (
|
|
|
935 |
(0x7FFD < exp)
|
|
|
936 |
|| ((exp == 0x7FFD)
|
|
|
937 |
&& softfloat_eq128(
|
|
|
938 |
sig64,
|
|
|
939 |
sig0,
|
|
|
940 |
UINT64_C( 0x0001FFFFFFFFFFFF ),
|
|
|
941 |
UINT64_C( 0xFFFFFFFFFFFFFFFF )
|
|
|
942 |
)
|
|
|
943 |
&& doIncrement)
|
|
|
944 |
) {
|
|
|
945 |
/*----------------------------------------------------------------
|
|
|
946 |
*----------------------------------------------------------------*/
|
|
|
947 |
softfloat_raiseFlags(
|
|
|
948 |
softfloat_flag_overflow | softfloat_flag_inexact );
|
|
|
949 |
if (
|
|
|
950 |
roundNearEven
|
|
|
951 |
|| (roundingMode == softfloat_round_near_maxMag)
|
|
|
952 |
|| (roundingMode
|
|
|
953 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
954 |
) {
|
|
|
955 |
uiZ64 = packToF128UI64( sign, 0x7FFF, 0 );
|
|
|
956 |
uiZ0 = 0;
|
|
|
957 |
} else {
|
|
|
958 |
uiZ64 =
|
|
|
959 |
packToF128UI64(
|
|
|
960 |
sign, 0x7FFE, UINT64_C( 0x0000FFFFFFFFFFFF ) );
|
|
|
961 |
uiZ0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
|
|
962 |
}
|
|
|
963 |
goto uiZ;
|
|
|
964 |
}
|
|
|
965 |
}
|
|
|
966 |
/*------------------------------------------------------------------------
|
|
|
967 |
*------------------------------------------------------------------------*/
|
|
|
968 |
if ( sigExtra ) {
|
|
|
969 |
softfloat_exceptionFlags |= softfloat_flag_inexact;
|
|
|
970 |
#ifdef SOFTFLOAT_ROUND_ODD
|
|
|
971 |
if ( roundingMode == softfloat_round_odd ) {
|
|
|
972 |
sig0 |= 1;
|
|
|
973 |
goto packReturn;
|
|
|
974 |
}
|
|
|
975 |
#endif
|
|
|
976 |
}
|
|
|
977 |
if ( doIncrement ) {
|
|
|
978 |
sig128 = softfloat_add128( sig64, sig0, 0, 1 );
|
|
|
979 |
sig64 = sig128.v64;
|
|
|
980 |
sig0 =
|
|
|
981 |
sig128.v0
|
|
|
982 |
& ~(uint64_t)
|
|
|
983 |
(! (sigExtra & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
|
|
|
984 |
& roundNearEven);
|
|
|
985 |
} else {
|
|
|
986 |
if ( ! (sig64 | sig0) ) exp = 0;
|
|
|
987 |
}
|
|
|
988 |
/*------------------------------------------------------------------------
|
|
|
989 |
*------------------------------------------------------------------------*/
|
|
|
990 |
packReturn:
|
|
|
991 |
uiZ64 = packToF128UI64( sign, exp, sig64 );
|
|
|
992 |
uiZ0 = sig0;
|
|
|
993 |
uiZ:
|
|
|
994 |
uZ.ui.v64 = uiZ64;
|
|
|
995 |
uZ.ui.v0 = uiZ0;
|
|
|
996 |
return uZ.f;
|
|
|
997 |
}
|
|
|
998 |
|
|
|
999 |
float128_t
|
|
|
1000 |
softfloat_normRoundPackToF128(
|
|
|
1001 |
bool sign, int_fast32_t exp, uint_fast64_t sig64, uint_fast64_t sig0 )
|
|
|
1002 |
{
|
|
|
1003 |
int_fast8_t shiftDist;
|
|
|
1004 |
struct uint128 sig128;
|
|
|
1005 |
union ui128_f128 uZ;
|
|
|
1006 |
uint_fast64_t sigExtra;
|
|
|
1007 |
struct uint128_extra sig128Extra;
|
|
|
1008 |
|
|
|
1009 |
if ( ! sig64 ) {
|
|
|
1010 |
exp -= 64;
|
|
|
1011 |
sig64 = sig0;
|
|
|
1012 |
sig0 = 0;
|
|
|
1013 |
}
|
|
|
1014 |
shiftDist = softfloat_countLeadingZeros64( sig64 ) - 15;
|
|
|
1015 |
exp -= shiftDist;
|
|
|
1016 |
if ( 0 <= shiftDist ) {
|
|
|
1017 |
if ( shiftDist ) {
|
|
|
1018 |
sig128 = softfloat_shortShiftLeft128( sig64, sig0, shiftDist );
|
|
|
1019 |
sig64 = sig128.v64;
|
|
|
1020 |
sig0 = sig128.v0;
|
|
|
1021 |
}
|
|
|
1022 |
if ( (uint32_t) exp < 0x7FFD ) {
|
|
|
1023 |
uZ.ui.v64 = packToF128UI64( sign, sig64 | sig0 ? exp : 0, sig64 );
|
|
|
1024 |
uZ.ui.v0 = sig0;
|
|
|
1025 |
return uZ.f;
|
|
|
1026 |
}
|
|
|
1027 |
sigExtra = 0;
|
|
|
1028 |
} else {
|
|
|
1029 |
sig128Extra =
|
|
|
1030 |
softfloat_shortShiftRightJam128Extra( sig64, sig0, 0, -shiftDist );
|
|
|
1031 |
sig64 = sig128Extra.v.v64;
|
|
|
1032 |
sig0 = sig128Extra.v.v0;
|
|
|
1033 |
sigExtra = sig128Extra.extra;
|
|
|
1034 |
}
|
|
|
1035 |
return softfloat_roundPackToF128( sign, exp, sig64, sig0, sigExtra );
|
|
|
1036 |
}
|
|
|
1037 |
|
|
|
1038 |
struct uint128
|
|
|
1039 |
softfloat_propagateNaNF128UI(
|
|
|
1040 |
uint_fast64_t uiA64,
|
|
|
1041 |
uint_fast64_t uiA0,
|
|
|
1042 |
uint_fast64_t uiB64,
|
|
|
1043 |
uint_fast64_t uiB0
|
|
|
1044 |
)
|
|
|
1045 |
{
|
|
|
1046 |
bool isSigNaNA, isSigNaNB;
|
|
|
1047 |
uint_fast64_t uiNonsigA64, uiNonsigB64, uiMagA64, uiMagB64;
|
|
|
1048 |
struct uint128 uiZ;
|
|
|
1049 |
|
|
|
1050 |
/*------------------------------------------------------------------------
|
|
|
1051 |
*------------------------------------------------------------------------*/
|
|
|
1052 |
isSigNaNA = softfloat_isSigNaNF128UI( uiA64, uiA0 );
|
|
|
1053 |
isSigNaNB = softfloat_isSigNaNF128UI( uiB64, uiB0 );
|
|
|
1054 |
/*------------------------------------------------------------------------
|
|
|
1055 |
| Make NaNs non-signaling.
|
|
|
1056 |
*------------------------------------------------------------------------*/
|
|
|
1057 |
uiNonsigA64 = uiA64 | UINT64_C( 0x0000800000000000 );
|
|
|
1058 |
uiNonsigB64 = uiB64 | UINT64_C( 0x0000800000000000 );
|
|
|
1059 |
/*------------------------------------------------------------------------
|
|
|
1060 |
*------------------------------------------------------------------------*/
|
|
|
1061 |
if ( isSigNaNA | isSigNaNB ) {
|
|
|
1062 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
1063 |
if ( isSigNaNA ) {
|
|
|
1064 |
if ( isSigNaNB ) goto returnLargerMag;
|
|
|
1065 |
if ( isNaNF128UI( uiB64, uiB0 ) ) goto returnB;
|
|
|
1066 |
goto returnA;
|
|
|
1067 |
} else {
|
|
|
1068 |
if ( isNaNF128UI( uiA64, uiA0 ) ) goto returnA;
|
|
|
1069 |
goto returnB;
|
|
|
1070 |
}
|
|
|
1071 |
}
|
|
|
1072 |
returnLargerMag:
|
|
|
1073 |
uiMagA64 = uiA64 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
|
|
|
1074 |
uiMagB64 = uiB64 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
|
|
|
1075 |
if ( uiMagA64 < uiMagB64 ) goto returnB;
|
|
|
1076 |
if ( uiMagB64 < uiMagA64 ) goto returnA;
|
|
|
1077 |
if ( uiA0 < uiB0 ) goto returnB;
|
|
|
1078 |
if ( uiB0 < uiA0 ) goto returnA;
|
|
|
1079 |
if ( uiNonsigA64 < uiNonsigB64 ) goto returnA;
|
|
|
1080 |
returnB:
|
|
|
1081 |
uiZ.v64 = uiNonsigB64;
|
|
|
1082 |
uiZ.v0 = uiB0;
|
|
|
1083 |
return uiZ;
|
|
|
1084 |
returnA:
|
|
|
1085 |
uiZ.v64 = uiNonsigA64;
|
|
|
1086 |
uiZ.v0 = uiA0;
|
|
|
1087 |
return uiZ;
|
|
|
1088 |
|
|
|
1089 |
}
|
|
|
1090 |
|
|
|
1091 |
void
|
|
|
1092 |
f64_to_f128M( float64_t a, float128_t *zPtr )
|
|
|
1093 |
{
|
|
|
1094 |
uint32_t *zWPtr;
|
|
|
1095 |
union ui64_f64 uA;
|
|
|
1096 |
uint64_t uiA;
|
|
|
1097 |
bool sign;
|
|
|
1098 |
int_fast16_t exp;
|
|
|
1099 |
uint64_t frac;
|
|
|
1100 |
struct commonNaN commonNaN;
|
|
|
1101 |
uint32_t uiZ96;
|
|
|
1102 |
struct exp16_sig64 normExpSig;
|
|
|
1103 |
|
|
|
1104 |
/*------------------------------------------------------------------------
|
|
|
1105 |
*------------------------------------------------------------------------*/
|
|
|
1106 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1107 |
/*------------------------------------------------------------------------
|
|
|
1108 |
*------------------------------------------------------------------------*/
|
|
|
1109 |
uA.f = a;
|
|
|
1110 |
uiA = uA.ui;
|
|
|
1111 |
sign = signF64UI( uiA );
|
|
|
1112 |
exp = expF64UI( uiA );
|
|
|
1113 |
frac = fracF64UI( uiA );
|
|
|
1114 |
/*------------------------------------------------------------------------
|
|
|
1115 |
*------------------------------------------------------------------------*/
|
|
|
1116 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1117 |
if ( exp == 0x7FF ) {
|
|
|
1118 |
if ( frac ) {
|
|
|
1119 |
softfloat_f64UIToCommonNaN( uiA, &commonNaN );
|
|
|
1120 |
softfloat_commonNaNToF128M( &commonNaN, zWPtr );
|
|
|
1121 |
return;
|
|
|
1122 |
}
|
|
|
1123 |
uiZ96 = packToF128UI96( sign, 0x7FFF, 0 );
|
|
|
1124 |
goto uiZ;
|
|
|
1125 |
}
|
|
|
1126 |
/*------------------------------------------------------------------------
|
|
|
1127 |
*------------------------------------------------------------------------*/
|
|
|
1128 |
if ( ! exp ) {
|
|
|
1129 |
if ( ! frac ) {
|
|
|
1130 |
uiZ96 = packToF128UI96( sign, 0, 0 );
|
|
|
1131 |
goto uiZ;
|
|
|
1132 |
}
|
|
|
1133 |
normExpSig = softfloat_normSubnormalF64Sig( frac );
|
|
|
1134 |
exp = normExpSig.exp - 1;
|
|
|
1135 |
frac = normExpSig.sig;
|
|
|
1136 |
}
|
|
|
1137 |
/*------------------------------------------------------------------------
|
|
|
1138 |
*------------------------------------------------------------------------*/
|
|
|
1139 |
zWPtr[indexWord( 4, 1 )] = (uint32_t) frac<<28;
|
|
|
1140 |
frac >>= 4;
|
|
|
1141 |
zWPtr[indexWordHi( 4 )] = packToF128UI96( sign, exp + 0x3C00, frac>>32 );
|
|
|
1142 |
zWPtr[indexWord( 4, 2 )] = frac;
|
|
|
1143 |
return;
|
|
|
1144 |
/*------------------------------------------------------------------------
|
|
|
1145 |
*------------------------------------------------------------------------*/
|
|
|
1146 |
uiZ:
|
|
|
1147 |
zWPtr[indexWord( 4, 3 )] = uiZ96;
|
|
|
1148 |
zWPtr[indexWord( 4, 2 )] = 0;
|
|
|
1149 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1150 |
|
|
|
1151 |
}
|
|
|
1152 |
|
|
|
1153 |
static float128_t
|
|
|
1154 |
softfloat_addMagsF128(
|
|
|
1155 |
uint64_t uiA64,
|
|
|
1156 |
uint64_t uiA0,
|
|
|
1157 |
uint64_t uiB64,
|
|
|
1158 |
uint64_t uiB0,
|
|
|
1159 |
bool signZ
|
|
|
1160 |
)
|
|
|
1161 |
{
|
|
|
1162 |
int32_t expA;
|
|
|
1163 |
struct uint128 sigA;
|
|
|
1164 |
int32_t expB;
|
|
|
1165 |
struct uint128 sigB;
|
|
|
1166 |
int32_t expDiff;
|
|
|
1167 |
struct uint128 uiZ, sigZ;
|
|
|
1168 |
int32_t expZ;
|
|
|
1169 |
uint64_t sigZExtra;
|
|
|
1170 |
struct uint128_extra sig128Extra;
|
|
|
1171 |
union ui128_f128 uZ;
|
|
|
1172 |
|
|
|
1173 |
expA = expF128UI64( uiA64 );
|
|
|
1174 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1175 |
sigA.v0 = uiA0;
|
|
|
1176 |
expB = expF128UI64( uiB64 );
|
|
|
1177 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1178 |
sigB.v0 = uiB0;
|
|
|
1179 |
expDiff = expA - expB;
|
|
|
1180 |
if ( ! expDiff ) {
|
|
|
1181 |
if ( expA == 0x7FFF ) {
|
|
|
1182 |
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1183 |
uiZ.v64 = uiA64;
|
|
|
1184 |
uiZ.v0 = uiA0;
|
|
|
1185 |
goto uiZ;
|
|
|
1186 |
}
|
|
|
1187 |
sigZ = softfloat_add128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
|
|
|
1188 |
if ( ! expA ) {
|
|
|
1189 |
uiZ.v64 = packToF128UI64( signZ, 0, sigZ.v64 );
|
|
|
1190 |
uiZ.v0 = sigZ.v0;
|
|
|
1191 |
goto uiZ;
|
|
|
1192 |
}
|
|
|
1193 |
expZ = expA;
|
|
|
1194 |
sigZ.v64 |= UINT64_C( 0x0002000000000000 );
|
|
|
1195 |
sigZExtra = 0;
|
|
|
1196 |
goto shiftRight1;
|
|
|
1197 |
}
|
|
|
1198 |
if ( expDiff < 0 ) {
|
|
|
1199 |
if ( expB == 0x7FFF ) {
|
|
|
1200 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1201 |
uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
|
|
|
1202 |
uiZ.v0 = 0;
|
|
|
1203 |
goto uiZ;
|
|
|
1204 |
}
|
|
|
1205 |
expZ = expB;
|
|
|
1206 |
if ( expA ) {
|
|
|
1207 |
sigA.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1208 |
} else {
|
|
|
1209 |
++expDiff;
|
|
|
1210 |
sigZExtra = 0;
|
|
|
1211 |
if ( ! expDiff ) goto newlyAligned;
|
|
|
1212 |
}
|
|
|
1213 |
sig128Extra =
|
|
|
1214 |
softfloat_shiftRightJam128Extra( sigA.v64, sigA.v0, 0, -expDiff );
|
|
|
1215 |
sigA = sig128Extra.v;
|
|
|
1216 |
sigZExtra = sig128Extra.extra;
|
|
|
1217 |
} else {
|
|
|
1218 |
if ( expA == 0x7FFF ) {
|
|
|
1219 |
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
|
|
|
1220 |
uiZ.v64 = uiA64;
|
|
|
1221 |
uiZ.v0 = uiA0;
|
|
|
1222 |
goto uiZ;
|
|
|
1223 |
}
|
|
|
1224 |
expZ = expA;
|
|
|
1225 |
if ( expB ) {
|
|
|
1226 |
sigB.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1227 |
} else {
|
|
|
1228 |
--expDiff;
|
|
|
1229 |
sigZExtra = 0;
|
|
|
1230 |
if ( ! expDiff ) goto newlyAligned;
|
|
|
1231 |
}
|
|
|
1232 |
sig128Extra =
|
|
|
1233 |
softfloat_shiftRightJam128Extra( sigB.v64, sigB.v0, 0, expDiff );
|
|
|
1234 |
sigB = sig128Extra.v;
|
|
|
1235 |
sigZExtra = sig128Extra.extra;
|
|
|
1236 |
}
|
|
|
1237 |
newlyAligned:
|
|
|
1238 |
sigZ =
|
|
|
1239 |
softfloat_add128(
|
|
|
1240 |
sigA.v64 | UINT64_C( 0x0001000000000000 ),
|
|
|
1241 |
sigA.v0,
|
|
|
1242 |
sigB.v64,
|
|
|
1243 |
sigB.v0
|
|
|
1244 |
);
|
|
|
1245 |
--expZ;
|
|
|
1246 |
if ( sigZ.v64 < UINT64_C( 0x0002000000000000 ) ) goto roundAndPack;
|
|
|
1247 |
++expZ;
|
|
|
1248 |
shiftRight1:
|
|
|
1249 |
sig128Extra =
|
|
|
1250 |
softfloat_shortShiftRightJam128Extra(
|
|
|
1251 |
sigZ.v64, sigZ.v0, sigZExtra, 1 );
|
|
|
1252 |
sigZ = sig128Extra.v;
|
|
|
1253 |
sigZExtra = sig128Extra.extra;
|
|
|
1254 |
roundAndPack:
|
|
|
1255 |
return
|
|
|
1256 |
softfloat_roundPackToF128( signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
|
|
|
1257 |
propagateNaN:
|
|
|
1258 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
1259 |
uiZ:
|
|
|
1260 |
uZ.ui = uiZ;
|
|
|
1261 |
return uZ.f;
|
|
|
1262 |
|
|
|
1263 |
}
|
|
|
1264 |
|
|
|
1265 |
float128_t
|
|
|
1266 |
softfloat_subMagsF128(
|
|
|
1267 |
uint_fast64_t uiA64,
|
|
|
1268 |
uint_fast64_t uiA0,
|
|
|
1269 |
uint_fast64_t uiB64,
|
|
|
1270 |
uint_fast64_t uiB0,
|
|
|
1271 |
bool signZ
|
|
|
1272 |
)
|
|
|
1273 |
{
|
|
|
1274 |
int_fast32_t expA;
|
|
|
1275 |
struct uint128 sigA;
|
|
|
1276 |
int_fast32_t expB;
|
|
|
1277 |
struct uint128 sigB, sigZ;
|
|
|
1278 |
int_fast32_t expDiff, expZ;
|
|
|
1279 |
struct uint128 uiZ;
|
|
|
1280 |
union ui128_f128 uZ;
|
|
|
1281 |
|
|
|
1282 |
expA = expF128UI64( uiA64 );
|
|
|
1283 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1284 |
sigA.v0 = uiA0;
|
|
|
1285 |
expB = expF128UI64( uiB64 );
|
|
|
1286 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1287 |
sigB.v0 = uiB0;
|
|
|
1288 |
sigA = softfloat_shortShiftLeft128( sigA.v64, sigA.v0, 4 );
|
|
|
1289 |
sigB = softfloat_shortShiftLeft128( sigB.v64, sigB.v0, 4 );
|
|
|
1290 |
expDiff = expA - expB;
|
|
|
1291 |
if ( 0 < expDiff ) goto expABigger;
|
|
|
1292 |
if ( expDiff < 0 ) goto expBBigger;
|
|
|
1293 |
if ( expA == 0x7FFF ) {
|
|
|
1294 |
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1295 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
1296 |
uiZ.v64 = defaultNaNF128UI64;
|
|
|
1297 |
uiZ.v0 = defaultNaNF128UI0;
|
|
|
1298 |
goto uiZ;
|
|
|
1299 |
}
|
|
|
1300 |
expZ = expA;
|
|
|
1301 |
if ( ! expZ ) expZ = 1;
|
|
|
1302 |
if ( sigB.v64 < sigA.v64 ) goto aBigger;
|
|
|
1303 |
if ( sigA.v64 < sigB.v64 ) goto bBigger;
|
|
|
1304 |
if ( sigB.v0 < sigA.v0 ) goto aBigger;
|
|
|
1305 |
if ( sigA.v0 < sigB.v0 ) goto bBigger;
|
|
|
1306 |
uiZ.v64 =
|
|
|
1307 |
packToF128UI64(
|
|
|
1308 |
(softfloat_roundingMode == softfloat_round_min), 0, 0 );
|
|
|
1309 |
uiZ.v0 = 0;
|
|
|
1310 |
goto uiZ;
|
|
|
1311 |
expBBigger:
|
|
|
1312 |
if ( expB == 0x7FFF ) {
|
|
|
1313 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1314 |
uiZ.v64 = packToF128UI64( signZ ^ 1, 0x7FFF, 0 );
|
|
|
1315 |
uiZ.v0 = 0;
|
|
|
1316 |
goto uiZ;
|
|
|
1317 |
}
|
|
|
1318 |
if ( expA ) {
|
|
|
1319 |
sigA.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1320 |
} else {
|
|
|
1321 |
++expDiff;
|
|
|
1322 |
if ( ! expDiff ) goto newlyAlignedBBigger;
|
|
|
1323 |
}
|
|
|
1324 |
sigA = softfloat_shiftRightJam128( sigA.v64, sigA.v0, -expDiff );
|
|
|
1325 |
newlyAlignedBBigger:
|
|
|
1326 |
expZ = expB;
|
|
|
1327 |
sigB.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1328 |
bBigger:
|
|
|
1329 |
signZ = ! signZ;
|
|
|
1330 |
sigZ = softfloat_sub128( sigB.v64, sigB.v0, sigA.v64, sigA.v0 );
|
|
|
1331 |
goto normRoundPack;
|
|
|
1332 |
expABigger:
|
|
|
1333 |
if ( expA == 0x7FFF ) {
|
|
|
1334 |
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
|
|
|
1335 |
uiZ.v64 = uiA64;
|
|
|
1336 |
uiZ.v0 = uiA0;
|
|
|
1337 |
goto uiZ;
|
|
|
1338 |
}
|
|
|
1339 |
if ( expB ) {
|
|
|
1340 |
sigB.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1341 |
} else {
|
|
|
1342 |
--expDiff;
|
|
|
1343 |
if ( ! expDiff ) goto newlyAlignedABigger;
|
|
|
1344 |
}
|
|
|
1345 |
sigB = softfloat_shiftRightJam128( sigB.v64, sigB.v0, expDiff );
|
|
|
1346 |
newlyAlignedABigger:
|
|
|
1347 |
expZ = expA;
|
|
|
1348 |
sigA.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1349 |
aBigger:
|
|
|
1350 |
sigZ = softfloat_sub128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
|
|
|
1351 |
normRoundPack:
|
|
|
1352 |
return softfloat_normRoundPackToF128( signZ, expZ - 5, sigZ.v64, sigZ.v0 );
|
|
|
1353 |
propagateNaN:
|
|
|
1354 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
1355 |
uiZ:
|
|
|
1356 |
uZ.ui = uiZ;
|
|
|
1357 |
return uZ.f;
|
|
|
1358 |
|
|
|
1359 |
}
|
|
|
1360 |
|
|
|
1361 |
void
|
|
|
1362 |
f128M_add( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1363 |
{
|
|
|
1364 |
const uint64_t *aWPtr, *bWPtr;
|
|
|
1365 |
uint64_t uiA64, uiA0;
|
|
|
1366 |
int signA;
|
|
|
1367 |
uint64_t uiB64, uiB0;
|
|
|
1368 |
int signB;
|
|
|
1369 |
|
|
|
1370 |
aWPtr = (const uint64_t *) aPtr;
|
|
|
1371 |
bWPtr = (const uint64_t *) bPtr;
|
|
|
1372 |
uiA64 = aWPtr[indexWord( 2, 1 )];
|
|
|
1373 |
uiA0 = aWPtr[indexWord( 2, 0 )];
|
|
|
1374 |
signA = signF128UI64( uiA64 );
|
|
|
1375 |
uiB64 = bWPtr[indexWord( 2, 1 )];
|
|
|
1376 |
uiB0 = bWPtr[indexWord( 2, 0 )];
|
|
|
1377 |
signB = signF128UI64( uiB64 );
|
|
|
1378 |
|
|
|
1379 |
if ( signA == signB ) {
|
|
|
1380 |
*zPtr = softfloat_addMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1381 |
} else {
|
|
|
1382 |
*zPtr = softfloat_subMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1383 |
}
|
|
4999
|
1384 |
}
|
|
4992
|
1385 |
|
|
4999
|
1386 |
void
|
|
|
1387 |
f128M_sub( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1388 |
{
|
|
|
1389 |
const uint64_t *aWPtr, *bWPtr;
|
|
|
1390 |
uint_fast64_t uiA64, uiA0;
|
|
|
1391 |
bool signA;
|
|
|
1392 |
uint_fast64_t uiB64, uiB0;
|
|
|
1393 |
bool signB;
|
|
|
1394 |
|
|
|
1395 |
aWPtr = (const uint64_t *) aPtr;
|
|
|
1396 |
bWPtr = (const uint64_t *) bPtr;
|
|
|
1397 |
uiA64 = aWPtr[indexWord( 2, 1 )];
|
|
|
1398 |
uiA0 = aWPtr[indexWord( 2, 0 )];
|
|
|
1399 |
signA = signF128UI64( uiA64 );
|
|
|
1400 |
uiB64 = bWPtr[indexWord( 2, 1 )];
|
|
|
1401 |
uiB0 = bWPtr[indexWord( 2, 0 )];
|
|
|
1402 |
signB = signF128UI64( uiB64 );
|
|
|
1403 |
if ( signA == signB ) {
|
|
|
1404 |
*zPtr = softfloat_subMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1405 |
} else {
|
|
|
1406 |
*zPtr = softfloat_addMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1407 |
}
|
|
4992
|
1408 |
}
|
|
4999
|
1409 |
|
|
5000
|
1410 |
void
|
|
|
1411 |
f128M_mul( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1412 |
{
|
|
|
1413 |
const uint32_t *aWPtr, *bWPtr;
|
|
|
1414 |
uint32_t *zWPtr;
|
|
|
1415 |
uint32_t uiA96;
|
|
|
1416 |
int32_t expA;
|
|
|
1417 |
uint32_t uiB96;
|
|
|
1418 |
int32_t expB;
|
|
|
1419 |
bool signZ;
|
|
|
1420 |
const uint32_t *ptr;
|
|
|
1421 |
uint32_t uiZ96, sigA[4];
|
|
|
1422 |
uint_fast8_t shiftDist;
|
|
|
1423 |
uint32_t sigB[4];
|
|
|
1424 |
int32_t expZ;
|
|
|
1425 |
uint32_t sigProd[8], *extSigZPtr;
|
|
|
1426 |
|
|
|
1427 |
/*------------------------------------------------------------------------
|
|
|
1428 |
*------------------------------------------------------------------------*/
|
|
|
1429 |
aWPtr = (const uint32_t *) aPtr;
|
|
|
1430 |
bWPtr = (const uint32_t *) bPtr;
|
|
|
1431 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1432 |
/*------------------------------------------------------------------------
|
|
|
1433 |
*------------------------------------------------------------------------*/
|
|
|
1434 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
1435 |
expA = expF128UI96( uiA96 );
|
|
|
1436 |
uiB96 = bWPtr[indexWordHi( 4 )];
|
|
|
1437 |
expB = expF128UI96( uiB96 );
|
|
|
1438 |
signZ = signF128UI96( uiA96 ) ^ signF128UI96( uiB96 );
|
|
|
1439 |
/*------------------------------------------------------------------------
|
|
|
1440 |
*------------------------------------------------------------------------*/
|
|
|
1441 |
if ( (expA == 0x7FFF) || (expB == 0x7FFF) ) {
|
|
|
1442 |
if ( softfloat_tryPropagateNaNF128M( aWPtr, bWPtr, zWPtr ) ) return;
|
|
|
1443 |
ptr = aWPtr;
|
|
|
1444 |
if ( ! expA ) goto possiblyInvalid;
|
|
|
1445 |
if ( ! expB ) {
|
|
|
1446 |
ptr = bWPtr;
|
|
|
1447 |
possiblyInvalid:
|
|
|
1448 |
if (
|
|
|
1449 |
! fracF128UI96( ptr[indexWordHi( 4 )] )
|
|
|
1450 |
&& ! (ptr[indexWord( 4, 2 )] | ptr[indexWord( 4, 1 )]
|
|
|
1451 |
| ptr[indexWord( 4, 0 )])
|
|
|
1452 |
) {
|
|
|
1453 |
softfloat_invalidF128M( zWPtr );
|
|
|
1454 |
return;
|
|
|
1455 |
}
|
|
|
1456 |
}
|
|
|
1457 |
uiZ96 = packToF128UI96( signZ, 0x7FFF, 0 );
|
|
|
1458 |
goto uiZ96;
|
|
|
1459 |
}
|
|
|
1460 |
/*------------------------------------------------------------------------
|
|
|
1461 |
*------------------------------------------------------------------------*/
|
|
|
1462 |
if ( expA ) {
|
|
|
1463 |
sigA[indexWordHi( 4 )] = fracF128UI96( uiA96 ) | 0x00010000;
|
|
|
1464 |
sigA[indexWord( 4, 2 )] = aWPtr[indexWord( 4, 2 )];
|
|
|
1465 |
sigA[indexWord( 4, 1 )] = aWPtr[indexWord( 4, 1 )];
|
|
|
1466 |
sigA[indexWord( 4, 0 )] = aWPtr[indexWord( 4, 0 )];
|
|
|
1467 |
} else {
|
|
|
1468 |
expA = softfloat_shiftNormSigF128M( aWPtr, 0, sigA );
|
|
|
1469 |
if ( expA == -128 ) goto zero;
|
|
|
1470 |
}
|
|
|
1471 |
if ( expB ) {
|
|
|
1472 |
sigB[indexWordHi( 4 )] = fracF128UI96( uiB96 ) | 0x00010000;
|
|
|
1473 |
sigB[indexWord( 4, 2 )] = bWPtr[indexWord( 4, 2 )];
|
|
|
1474 |
sigB[indexWord( 4, 1 )] = bWPtr[indexWord( 4, 1 )];
|
|
|
1475 |
sigB[indexWord( 4, 0 )] = bWPtr[indexWord( 4, 0 )];
|
|
|
1476 |
} else {
|
|
|
1477 |
expB = softfloat_shiftNormSigF128M( bWPtr, 0, sigB );
|
|
|
1478 |
if ( expB == -128 ) goto zero;
|
|
|
1479 |
}
|
|
|
1480 |
/*------------------------------------------------------------------------
|
|
|
1481 |
*------------------------------------------------------------------------*/
|
|
|
1482 |
expZ = expA + expB - 0x4000;
|
|
|
1483 |
softfloat_mul128MTo256M( sigA, sigB, sigProd );
|
|
|
1484 |
if (
|
|
|
1485 |
sigProd[indexWord( 8, 2 )]
|
|
|
1486 |
|| (sigProd[indexWord( 8, 1 )] | sigProd[indexWord( 8, 0 )])
|
|
|
1487 |
) {
|
|
|
1488 |
sigProd[indexWord( 8, 3 )] |= 1;
|
|
|
1489 |
}
|
|
|
1490 |
extSigZPtr = &sigProd[indexMultiwordHi( 8, 5 )];
|
|
|
1491 |
shiftDist = 16;
|
|
|
1492 |
if ( extSigZPtr[indexWordHi( 5 )] & 2 ) {
|
|
|
1493 |
++expZ;
|
|
|
1494 |
shiftDist = 15;
|
|
|
1495 |
}
|
|
|
1496 |
softfloat_shortShiftLeft160M( extSigZPtr, shiftDist, extSigZPtr );
|
|
|
1497 |
softfloat_roundPackMToF128M( signZ, expZ, extSigZPtr, zWPtr );
|
|
|
1498 |
return;
|
|
|
1499 |
/*------------------------------------------------------------------------
|
|
|
1500 |
*------------------------------------------------------------------------*/
|
|
|
1501 |
zero:
|
|
|
1502 |
uiZ96 = packToF128UI96( signZ, 0, 0 );
|
|
|
1503 |
uiZ96:
|
|
|
1504 |
zWPtr[indexWordHi( 4 )] = uiZ96;
|
|
|
1505 |
zWPtr[indexWord( 4, 2 )] = 0;
|
|
|
1506 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1507 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1508 |
}
|
|
|
1509 |
|
|
4993
|
1510 |
#endif // SUPPORT_QUADFLOAT
|
|
4992
|
1511 |
|
|
|
1512 |
%}
|
|
|
1513 |
! !
|
|
4991
|
1514 |
|
|
4983
|
1515 |
!QuadFloat class methodsFor:'documentation'!
|
|
|
1516 |
|
|
|
1517 |
documentation
|
|
|
1518 |
"
|
|
|
1519 |
QuadFloats represent rational numbers with limited precision
|
|
|
1520 |
and are mapped to IEEE quadruple precision format (128bit).
|
|
|
1521 |
If the underlying cpu supports them natively, the machine format (long double) is
|
|
|
1522 |
used. Otherwise, a software emulation is done, which is much slower.
|
|
|
1523 |
Thus only use them, if you really need the additional precision;
|
|
|
1524 |
if not, use Float (which are doubles) or LongFloats which usually have IEEE extended precision (80bit).
|
|
|
1525 |
|
|
|
1526 |
QuadFloats give you definite 128 bit quadruple floats,
|
|
|
1527 |
thus, code using quadFloats is guaranteed to be portable from one architecture to another.
|
|
|
1528 |
|
|
|
1529 |
Representation:
|
|
4991
|
1530 |
128bit quadruple IEEE floats (16bytes);
|
|
|
1531 |
112 bit mantissa,
|
|
|
1532 |
16 bit exponent,
|
|
|
1533 |
34 decimal digits (approx.)
|
|
4983
|
1534 |
|
|
4992
|
1535 |
On Sparc CPUs, this is a native supported type (long double) and fast;
|
|
|
1536 |
on x86 CPUs, this is emulated and slow.
|
|
|
1537 |
|
|
4983
|
1538 |
Mixed mode arithmetic:
|
|
4991
|
1539 |
quadFloat op anyFloat -> longFloat
|
|
|
1540 |
longFloat op complex -> complex
|
|
4983
|
1541 |
|
|
|
1542 |
Range and precision of storage formats: see LimitedPrecisionReal >> documentation
|
|
|
1543 |
|
|
|
1544 |
[author:]
|
|
4991
|
1545 |
Claus Gittinger
|
|
4983
|
1546 |
|
|
|
1547 |
[see also:]
|
|
4991
|
1548 |
Number
|
|
|
1549 |
Float ShortFloat LongFloat Fraction FixedPoint Integer Complex
|
|
|
1550 |
FloatArray DoubleArray
|
|
|
1551 |
https://en.wikipedia.org/wiki/Extended_precision
|
|
4983
|
1552 |
"
|
|
|
1553 |
! !
|
|
|
1554 |
|
|
|
1555 |
!QuadFloat class methodsFor:'instance creation'!
|
|
|
1556 |
|
|
|
1557 |
basicNew
|
|
|
1558 |
"return a new quadFloat - here we return 0.0
|
|
|
1559 |
- QuadFloats are usually NOT created this way ...
|
|
|
1560 |
Its implemented here to allow things like binary store & load
|
|
|
1561 |
of quadFloats.
|
|
|
1562 |
(but it is not a good idea to store the bits of a float - the reader might have a
|
|
|
1563 |
totally different representation - so floats should be
|
|
|
1564 |
binary stored in a device independent format)."
|
|
|
1565 |
|
|
|
1566 |
%{ /* NOCONTEXT */
|
|
4993
|
1567 |
#ifdef SUPPORT_QUADFLOAT
|
|
4983
|
1568 |
OBJ newFloat;
|
|
4991
|
1569 |
|
|
|
1570 |
if (sizeof(long double) == sizeof(float128_t)) {
|
|
|
1571 |
__qMKLFLOAT(newFloat, 0.0); /* OBJECT ALLOCATION */
|
|
4983
|
1572 |
} else {
|
|
4991
|
1573 |
float128_t qf;
|
|
4992
|
1574 |
f64_to_f128M(0.0, &qf);
|
|
4991
|
1575 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
4983
|
1576 |
}
|
|
|
1577 |
RETURN (newFloat);
|
|
4993
|
1578 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
1579 |
%}.
|
|
|
1580 |
self error:'QuadFloats not supported on this patform'
|
|
4983
|
1581 |
|
|
|
1582 |
"Created: / 06-06-2019 / 17:18:58 / Claus Gittinger"
|
|
|
1583 |
!
|
|
|
1584 |
|
|
|
1585 |
fromFloat:aFloat
|
|
|
1586 |
"return a new quadFloat, given a float value"
|
|
|
1587 |
|
|
|
1588 |
%{ /* NOCONTEXT */
|
|
4993
|
1589 |
#ifdef SUPPORT_QUADFLOAT
|
|
4983
|
1590 |
OBJ newFloat;
|
|
4991
|
1591 |
float128_t f;
|
|
4983
|
1592 |
|
|
|
1593 |
if (__isFloatLike(aFloat)) {
|
|
4992
|
1594 |
float64_t f = __floatVal(aFloat);
|
|
4991
|
1595 |
float128_t qf;
|
|
|
1596 |
|
|
4992
|
1597 |
f64_to_f128M(f, &qf);
|
|
4991
|
1598 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
1599 |
RETURN (newFloat);
|
|
4983
|
1600 |
}
|
|
4993
|
1601 |
#endif /* SUPPORT_QUADFLOAT */
|
|
4983
|
1602 |
%}.
|
|
4993
|
1603 |
aFloat isFloat ifTrue:[
|
|
4997
|
1604 |
self errorUnsupported
|
|
4993
|
1605 |
].
|
|
4997
|
1606 |
ArgumentError raise
|
|
4983
|
1607 |
|
|
|
1608 |
"
|
|
|
1609 |
QuadFloat fromFloat:123.0
|
|
|
1610 |
123.0 asQuadFloat
|
|
|
1611 |
123 asQuadFloat
|
|
|
1612 |
"
|
|
|
1613 |
|
|
|
1614 |
"Created: / 06-06-2019 / 18:01:03 / Claus Gittinger"
|
|
|
1615 |
! !
|
|
|
1616 |
|
|
|
1617 |
!QuadFloat class methodsFor:'coercing & converting'!
|
|
|
1618 |
|
|
|
1619 |
coerce:aNumber
|
|
|
1620 |
"convert the argument aNumber into an instance of the receiver's class and return it."
|
|
|
1621 |
|
|
|
1622 |
^ aNumber asQuadFloat.
|
|
|
1623 |
|
|
|
1624 |
"Created: / 06-06-2019 / 16:51:01 / Claus Gittinger"
|
|
|
1625 |
! !
|
|
|
1626 |
|
|
|
1627 |
!QuadFloat class methodsFor:'constants'!
|
|
|
1628 |
|
|
|
1629 |
NaN
|
|
|
1630 |
"return a shortFloat which represents not-a-Number (i.e. an invalid number)"
|
|
|
1631 |
|
|
|
1632 |
NaN isNil ifTrue:[
|
|
4991
|
1633 |
NaN := super NaN
|
|
4983
|
1634 |
].
|
|
|
1635 |
^ NaN
|
|
|
1636 |
|
|
|
1637 |
"
|
|
|
1638 |
self NaN
|
|
|
1639 |
"
|
|
|
1640 |
|
|
|
1641 |
"Created: / 06-06-2019 / 16:56:09 / Claus Gittinger"
|
|
|
1642 |
!
|
|
|
1643 |
|
|
|
1644 |
e
|
|
|
1645 |
"return the constant e as quadFloat"
|
|
|
1646 |
|
|
|
1647 |
E isNil ifTrue:[
|
|
4991
|
1648 |
"/ eDigits has enough digits for 128bit IEEE quads
|
|
|
1649 |
"/ do not use as a literal constant here - we cannot depend on the underlying C-compiler here...
|
|
|
1650 |
E := self readFrom:(Number eDigits)
|
|
4983
|
1651 |
].
|
|
|
1652 |
^ E
|
|
|
1653 |
|
|
|
1654 |
"Created: / 06-06-2019 / 17:01:54 / Claus Gittinger"
|
|
|
1655 |
!
|
|
|
1656 |
|
|
|
1657 |
pi
|
|
|
1658 |
"return the constant pi as quadFloat"
|
|
|
1659 |
|
|
|
1660 |
Pi isNil ifTrue:[
|
|
4991
|
1661 |
"/ piDigits has enough digits for 128bit IEEE quads
|
|
|
1662 |
"/ do not use as a literal constant here - we cannot depend on the underlying C-compiler here...
|
|
|
1663 |
Pi := self readFrom:(Number piDigits)
|
|
4983
|
1664 |
].
|
|
|
1665 |
^ Pi
|
|
|
1666 |
|
|
|
1667 |
"Created: / 06-06-2019 / 17:09:51 / Claus Gittinger"
|
|
|
1668 |
! !
|
|
|
1669 |
|
|
5003
|
1670 |
!QuadFloat class methodsFor:'error reportng'!
|
|
4994
|
1671 |
|
|
4997
|
1672 |
errorUnsupported
|
|
|
1673 |
self error:'QuadFloats not supported on this patform'
|
|
|
1674 |
|
|
5003
|
1675 |
"Created: / 07-06-2019 / 02:44:39 / Claus Gittinger"
|
|
4994
|
1676 |
! !
|
|
|
1677 |
|
|
|
1678 |
!QuadFloat methodsFor:'arithmetic'!
|
|
|
1679 |
|
|
|
1680 |
* aNumber
|
|
|
1681 |
"return the product of the receiver and the argument."
|
|
|
1682 |
|
|
|
1683 |
^ aNumber productFromQuadFloat:self
|
|
|
1684 |
!
|
|
|
1685 |
|
|
|
1686 |
+ aNumber
|
|
|
1687 |
"return the sum of the receiver and the argument, aNumber"
|
|
|
1688 |
|
|
|
1689 |
^ aNumber sumFromQuadFloat:self
|
|
|
1690 |
!
|
|
|
1691 |
|
|
|
1692 |
- aNumber
|
|
|
1693 |
"return the difference of the receiver and the argument, aNumber"
|
|
|
1694 |
|
|
|
1695 |
^ aNumber differenceFromQuadFloat:self
|
|
|
1696 |
!
|
|
|
1697 |
|
|
|
1698 |
/ aNumber
|
|
|
1699 |
"return the quotient of the receiver and the argument, aNumber"
|
|
|
1700 |
|
|
|
1701 |
aNumber isZero ifTrue:[
|
|
|
1702 |
"
|
|
|
1703 |
No, you shalt not divide by zero
|
|
|
1704 |
"
|
|
|
1705 |
^ ZeroDivide raiseRequestWith:thisContext.
|
|
|
1706 |
].
|
|
|
1707 |
^ aNumber quotientFromQuadFloat:self
|
|
|
1708 |
!
|
|
|
1709 |
|
|
|
1710 |
rem: aNumber
|
|
|
1711 |
"return the floating point remainder of the receiver and the argument, aNumber"
|
|
|
1712 |
|
|
|
1713 |
aNumber isZero ifTrue:[
|
|
|
1714 |
"
|
|
|
1715 |
No, you shalt not divide by zero
|
|
|
1716 |
"
|
|
|
1717 |
^ ZeroDivide raiseRequestWith:thisContext.
|
|
|
1718 |
].
|
|
|
1719 |
^ aNumber remainderFromLongFloat:self
|
|
|
1720 |
! !
|
|
|
1721 |
|
|
5003
|
1722 |
!QuadFloat methodsFor:'double dispatching'!
|
|
|
1723 |
|
|
|
1724 |
differenceFromQuadFloat:aQuadFloat
|
|
|
1725 |
%{
|
|
|
1726 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
1727 |
OBJ newFloat;
|
|
|
1728 |
float128_t result, myVal, argVal;
|
|
|
1729 |
|
|
|
1730 |
myVal = __quadFloatVal(self);
|
|
|
1731 |
argVal = __quadFloatVal(aQuadFloat);
|
|
|
1732 |
f128M_sub( &myVal, &argVal, &result );
|
|
|
1733 |
__qMKQFLOAT(newFloat, result);
|
|
|
1734 |
RETURN ( newFloat );
|
|
|
1735 |
#endif // SUPPORT_QUADFLOAT
|
|
|
1736 |
%}.
|
|
|
1737 |
self errorUnsupported
|
|
|
1738 |
!
|
|
|
1739 |
|
|
|
1740 |
sumFromQuadFloat:aQuadFloat
|
|
|
1741 |
"sent when aQuadFloat does not know how to add the receiver, self"
|
|
|
1742 |
|
|
|
1743 |
%{
|
|
|
1744 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
1745 |
OBJ newFloat;
|
|
|
1746 |
float128_t result, myVal, argVal;
|
|
|
1747 |
|
|
|
1748 |
myVal = __quadFloatVal(self);
|
|
|
1749 |
argVal = __quadFloatVal(aQuadFloat);
|
|
|
1750 |
f128M_add( &myVal, &argVal, &result );
|
|
|
1751 |
__qMKQFLOAT(newFloat, result);
|
|
|
1752 |
RETURN ( newFloat );
|
|
|
1753 |
#endif // SUPPORT_QUADFLOAT
|
|
|
1754 |
%}.
|
|
|
1755 |
self errorUnsupported
|
|
|
1756 |
! !
|
|
|
1757 |
|
|
|
1758 |
!QuadFloat methodsFor:'error reportng'!
|
|
|
1759 |
|
|
|
1760 |
errorUnsupported
|
|
|
1761 |
self class errorUnsupported
|
|
|
1762 |
|
|
|
1763 |
"Modified: / 07-06-2019 / 02:44:51 / Claus Gittinger"
|
|
|
1764 |
! !
|
|
|
1765 |
|
|
4983
|
1766 |
!QuadFloat class methodsFor:'documentation'!
|
|
|
1767 |
|
|
|
1768 |
version_CVS
|
|
|
1769 |
^ '$Header$'
|
|
|
1770 |
! !
|
|
5003
|
1771 |
|