|
5012
|
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
|
|
5014
|
20 |
#if defined(__MSBFIRST__) || defined(__MSB_FIRST__) || defined(MSBFIRST)
|
|
|
21 |
# undef LITTLEENDIAN
|
|
|
22 |
#else
|
|
|
23 |
# define LITTLEENDIAN
|
|
|
24 |
#endif
|
|
4993
|
25 |
|
|
|
26 |
#ifdef SUPPORT_QUADFLOAT
|
|
4998
|
27 |
# include <stdint.h>
|
|
|
28 |
|
|
5013
|
29 |
// The following code is adapted from the Softfloat-3e package,
|
|
|
30 |
// which includes the following license/copyright statement:
|
|
|
31 |
//
|
|
|
32 |
// License for Berkeley SoftFloat Release 3e
|
|
|
33 |
//
|
|
|
34 |
// John R. Hauser
|
|
|
35 |
// 2018 January 20
|
|
|
36 |
//
|
|
|
37 |
// The following applies to the whole of SoftFloat Release 3e as well as to
|
|
|
38 |
// each source file individually.
|
|
|
39 |
//
|
|
|
40 |
// Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018 The Regents of the
|
|
|
41 |
// University of California. All rights reserved.
|
|
|
42 |
//
|
|
|
43 |
// Redistribution and use in source and binary forms, with or without
|
|
|
44 |
// modification, are permitted provided that the following conditions are met:
|
|
|
45 |
//
|
|
|
46 |
// 1. Redistributions of source code must retain the above copyright notice,
|
|
|
47 |
// this list of conditions, and the following disclaimer.
|
|
|
48 |
//
|
|
|
49 |
// 2. Redistributions in binary form must reproduce the above copyright
|
|
|
50 |
// notice, this list of conditions, and the following disclaimer in the
|
|
|
51 |
// documentation and/or other materials provided with the distribution.
|
|
|
52 |
//
|
|
|
53 |
// 3. Neither the name of the University nor the names of its contributors
|
|
|
54 |
// may be used to endorse or promote products derived from this software
|
|
|
55 |
// without specific prior written permission.
|
|
|
56 |
//
|
|
|
57 |
// THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
|
|
|
58 |
// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
|
|
59 |
// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
|
|
|
60 |
// DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
|
|
|
61 |
// DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
|
|
62 |
// (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
|
63 |
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
|
|
|
64 |
// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
65 |
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
|
|
|
66 |
// THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
67 |
//
|
|
|
68 |
|
|
4992
|
69 |
/*----------------------------------------------------------------------------
|
|
|
70 |
| Software floating-point underflow tininess-detection mode.
|
|
|
71 |
*----------------------------------------------------------------------------*/
|
|
|
72 |
enum {
|
|
|
73 |
softfloat_tininess_beforeRounding = 0,
|
|
|
74 |
softfloat_tininess_afterRounding = 1
|
|
|
75 |
};
|
|
|
76 |
|
|
|
77 |
/*----------------------------------------------------------------------------
|
|
|
78 |
| Software floating-point exception flags.
|
|
|
79 |
*----------------------------------------------------------------------------*/
|
|
|
80 |
enum {
|
|
|
81 |
softfloat_flag_inexact = 1,
|
|
|
82 |
softfloat_flag_underflow = 2,
|
|
|
83 |
softfloat_flag_overflow = 4,
|
|
|
84 |
softfloat_flag_infinite = 8,
|
|
|
85 |
softfloat_flag_invalid = 16
|
|
|
86 |
};
|
|
|
87 |
/*----------------------------------------------------------------------------
|
|
|
88 |
| Software floating-point rounding mode. (Mode "odd" is supported only if
|
|
|
89 |
| SoftFloat is compiled with macro 'SOFTFLOAT_ROUND_ODD' defined.)
|
|
|
90 |
*----------------------------------------------------------------------------*/
|
|
|
91 |
enum {
|
|
|
92 |
softfloat_round_near_even = 0,
|
|
|
93 |
softfloat_round_minMag = 1,
|
|
|
94 |
softfloat_round_min = 2,
|
|
|
95 |
softfloat_round_max = 3,
|
|
|
96 |
softfloat_round_near_maxMag = 4,
|
|
|
97 |
softfloat_round_odd = 6
|
|
|
98 |
};
|
|
|
99 |
#define init_detectTininess softfloat_tininess_afterRounding
|
|
|
100 |
|
|
5014
|
101 |
typedef unsigned char bool;
|
|
|
102 |
typedef double float64_t;
|
|
|
103 |
|
|
|
104 |
#if defined(LITTLEENDIAN)
|
|
4992
|
105 |
struct uint128 { uint64_t v0, v64; };
|
|
|
106 |
struct uint64_extra { uint64_t extra, v; };
|
|
|
107 |
struct uint128_extra { uint64_t extra; struct uint128 v; };
|
|
5014
|
108 |
struct extFloat80M { uint64_t signif; uint16_t signExp; };
|
|
4992
|
109 |
#else
|
|
|
110 |
struct uint128 { uint64_t v64, v0; };
|
|
|
111 |
struct uint64_extra { uint64_t v, extra; };
|
|
|
112 |
struct uint128_extra { struct uint128 v; uint64_t extra; };
|
|
5014
|
113 |
struct extFloat80M { uint16_t signExp; uint64_t signif; };
|
|
4992
|
114 |
#endif
|
|
|
115 |
|
|
5014
|
116 |
union ui64_f64 { uint64_t ui; float64_t f; };
|
|
|
117 |
union ui128_f128 { struct uint128 ui; float128_t f; };
|
|
|
118 |
struct exp32_sig128 { int_fast32_t exp; struct uint128 sig; };
|
|
|
119 |
|
|
|
120 |
typedef struct extFloat80M extFloat80_t;
|
|
4992
|
121 |
|
|
|
122 |
/*----------------------------------------------------------------------------
|
|
|
123 |
| The bit pattern for a default generated 128-bit floating-point NaN.
|
|
|
124 |
*----------------------------------------------------------------------------*/
|
|
5014
|
125 |
#define defaultNaNF128UI96_32 0xFFFF8000
|
|
|
126 |
#define defaultNaNF128UI64_32 0
|
|
|
127 |
#define defaultNaNF128UI32_32 0
|
|
|
128 |
#define defaultNaNF128UI0_32 0
|
|
|
129 |
|
|
|
130 |
// #define defaultNaNF128UI64_64 UINT64_C( 0x7FFF800000000000 )
|
|
|
131 |
// #define defaultNaNF128UI0_64 UINT64_C( 0 )
|
|
|
132 |
|
|
5013
|
133 |
/*----------------------------------------------------------------------------
|
|
|
134 |
| The bit pattern for a default generated 128-bit floating-point +INF.
|
|
|
135 |
*----------------------------------------------------------------------------*/
|
|
5014
|
136 |
#define defaultPInfF128UI96_32 0xFFFF8000
|
|
|
137 |
#define defaultPInfF128UI64_32 0
|
|
|
138 |
#define defaultPInfF128UI32_32 0
|
|
|
139 |
#define defaultPInfF128UI0_32 0
|
|
5013
|
140 |
/*----------------------------------------------------------------------------
|
|
|
141 |
| The bit pattern for a default generated 128-bit floating-point -INF.
|
|
|
142 |
*----------------------------------------------------------------------------*/
|
|
5014
|
143 |
#define defaultNInfF128UI96_32 0xFFFF8000
|
|
|
144 |
#define defaultNInfF128UI64_32 0
|
|
|
145 |
#define defaultNInfF128UI32_32 0
|
|
|
146 |
#define defaultNInfF128UI0_32 0
|
|
4992
|
147 |
|
|
|
148 |
struct commonNaN {
|
|
|
149 |
bool sign;
|
|
|
150 |
#ifdef LITTLEENDIAN
|
|
|
151 |
uint64_t v0, v64;
|
|
|
152 |
#else
|
|
|
153 |
uint64_t v64, v0;
|
|
|
154 |
#endif
|
|
|
155 |
};
|
|
|
156 |
struct exp16_sig64 { int_fast16_t exp; uint_fast64_t sig; };
|
|
|
157 |
|
|
4993
|
158 |
#endif // SUPPORT_QUADFLOAT
|
|
|
159 |
|
|
4992
|
160 |
%}
|
|
|
161 |
! !
|
|
|
162 |
|
|
|
163 |
!QuadFloat primitiveVariables!
|
|
|
164 |
%{
|
|
4993
|
165 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
166 |
|
|
4992
|
167 |
uint_fast8_t softfloat_exceptionFlags;
|
|
|
168 |
uint_fast8_t softfloat_roundingMode = softfloat_round_near_even;
|
|
|
169 |
uint_fast8_t softfloat_detectTininess = init_detectTininess;
|
|
|
170 |
uint_fast8_t softfloat_exceptionFlags = 0;
|
|
|
171 |
|
|
|
172 |
const uint_least8_t softfloat_countLeadingZeros8[256] = {
|
|
|
173 |
8, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4,
|
|
|
174 |
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
|
|
|
175 |
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
176 |
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
|
|
|
177 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
178 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
179 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
180 |
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
|
|
|
181 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
182 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
183 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
184 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
185 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
186 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
187 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
|
|
|
188 |
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
|
|
|
189 |
};
|
|
4993
|
190 |
#endif // SUPPORT_QUADFLOAT
|
|
|
191 |
|
|
4991
|
192 |
%}
|
|
|
193 |
! !
|
|
|
194 |
|
|
4992
|
195 |
!QuadFloat primitiveFunctions!
|
|
|
196 |
%{
|
|
4993
|
197 |
#ifdef SUPPORT_QUADFLOAT
|
|
4992
|
198 |
|
|
5011
|
199 |
#if defined(LITTLEENDIAN) || defined(__LSBFIRST__)
|
|
4992
|
200 |
#define wordIncr 1
|
|
|
201 |
#define indexWord( total, n ) (n)
|
|
|
202 |
#define indexWordHi( total ) ((total) - 1)
|
|
|
203 |
#define indexWordLo( total ) 0
|
|
|
204 |
#define indexMultiword( total, m, n ) (n)
|
|
|
205 |
#define indexMultiwordHi( total, n ) ((total) - (n))
|
|
|
206 |
#define indexMultiwordLo( total, n ) 0
|
|
|
207 |
#define indexMultiwordHiBut( total, n ) (n)
|
|
|
208 |
#define indexMultiwordLoBut( total, n ) 0
|
|
|
209 |
#define INIT_UINTM4( v3, v2, v1, v0 ) { v0, v1, v2, v3 }
|
|
|
210 |
#else
|
|
|
211 |
#define wordIncr -1
|
|
|
212 |
#define indexWord( total, n ) ((total) - 1 - (n))
|
|
|
213 |
#define indexWordHi( total ) 0
|
|
|
214 |
#define indexWordLo( total ) ((total) - 1)
|
|
|
215 |
#define indexMultiword( total, m, n ) ((total) - 1 - (m))
|
|
|
216 |
#define indexMultiwordHi( total, n ) 0
|
|
|
217 |
#define indexMultiwordLo( total, n ) ((total) - (n))
|
|
|
218 |
#define indexMultiwordHiBut( total, n ) 0
|
|
|
219 |
#define indexMultiwordLoBut( total, n ) (n)
|
|
|
220 |
#define INIT_UINTM4( v3, v2, v1, v0 ) { v3, v2, v1, v0 }
|
|
|
221 |
#endif
|
|
|
222 |
|
|
|
223 |
#define signF64UI( a ) ((bool) ((uint64_t) (a)>>63))
|
|
|
224 |
#define expF64UI( a ) ((int_fast16_t) ((a)>>52) & 0x7FF)
|
|
|
225 |
#define fracF64UI( a ) ((a) & UINT64_C( 0x000FFFFFFFFFFFFF ))
|
|
|
226 |
#define softfloat_f64UIToCommonNaN( uiA, zPtr ) if ( ! ((uiA) & UINT64_C( 0x0008000000000000 )) ) softfloat_raiseFlags( softfloat_flag_invalid )
|
|
|
227 |
|
|
|
228 |
#define signF128UI64( a64 ) ((bool) ((uint64_t) (a64)>>63))
|
|
|
229 |
#define expF128UI64( a64 ) ((int_fast32_t) ((a64)>>48) & 0x7FFF)
|
|
|
230 |
#define fracF128UI64( a64 ) ((a64) & UINT64_C( 0x0000FFFFFFFFFFFF ))
|
|
|
231 |
#define packToF128UI64( sign, exp, sig64 ) (((uint_fast64_t) (sign)<<63) + ((uint_fast64_t) (exp)<<48) + (sig64))
|
|
|
232 |
|
|
|
233 |
#define signF128UI96( a96 ) ((bool) ((uint32_t) (a96)>>31))
|
|
|
234 |
#define expF128UI96( a96 ) ((int32_t) ((a96)>>16) & 0x7FFF)
|
|
|
235 |
#define fracF128UI96( a96 ) ((a96) & 0x0000FFFF)
|
|
|
236 |
#define packToF128UI96( sign, exp, sig96 ) (((uint32_t) (sign)<<31) + ((uint32_t) (exp)<<16) + (sig96))
|
|
|
237 |
|
|
|
238 |
#define isNaNF128UI( a64, a0 ) (((~(a64) & UINT64_C( 0x7FFF000000000000 )) == 0) && (a0 || ((a64) & UINT64_C( 0x0000FFFFFFFFFFFF ))))
|
|
|
239 |
#define softfloat_isSigNaNF128UI( uiA64, uiA0 ) ((((uiA64) & UINT64_C( 0x7FFF800000000000 )) == UINT64_C( 0x7FFF000000000000 )) && ((uiA0) || ((uiA64) & UINT64_C( 0x00007FFFFFFFFFFF ))))
|
|
5004
|
240 |
#define softfloat_approxRecip32_1( a ) ((uint32_t) (UINT64_C( 0x7FFFFFFFFFFFFFFF ) / (uint32_t) (a)))
|
|
4992
|
241 |
|
|
5014
|
242 |
#define signExtF80UI64( a64 ) ((bool) ((uint16_t) (a64)>>15))
|
|
|
243 |
#define expExtF80UI64( a64 ) ((a64) & 0x7FFF)
|
|
|
244 |
#define packToExtF80UI64( sign, exp ) ((uint_fast16_t) (sign)<<15 | (exp))
|
|
|
245 |
|
|
|
246 |
#define isNaNExtF80UI( a64, a0 ) ((((a64) & 0x7FFF) == 0x7FFF) && ((a0) & UINT64_C( 0x7FFFFFFFFFFFFFFF )))
|
|
|
247 |
|
|
5000
|
248 |
/*----------------------------------------------------------------------------
|
|
|
249 |
| This function or macro is the same as 'softfloat_shortShiftLeftM' with
|
|
|
250 |
| 'size_words' = 5 (N = 160).
|
|
|
251 |
*----------------------------------------------------------------------------*/
|
|
|
252 |
#define softfloat_shortShiftLeft160M( aPtr, dist, zPtr ) softfloat_shortShiftLeftM( 5, aPtr, dist, zPtr )
|
|
|
253 |
/*----------------------------------------------------------------------------
|
|
|
254 |
| This function or macro is the same as 'softfloat_shiftRightJamM' with
|
|
|
255 |
| 'size_words' = 5 (N = 160).
|
|
|
256 |
*----------------------------------------------------------------------------*/
|
|
|
257 |
#define softfloat_shiftRightJam160M( aPtr, dist, zPtr ) softfloat_shiftRightJamM( 5, aPtr, dist, zPtr )
|
|
|
258 |
/*----------------------------------------------------------------------------
|
|
|
259 |
| This function or macro is the same as 'softfloat_shortShiftLeftM' with
|
|
|
260 |
| 'size_words' = 4 (N = 128).
|
|
|
261 |
*----------------------------------------------------------------------------*/
|
|
|
262 |
#define softfloat_shortShiftLeft128M( aPtr, dist, zPtr ) softfloat_shortShiftLeftM( 4, aPtr, dist, zPtr )
|
|
5011
|
263 |
/*----------------------------------------------------------------------------
|
|
|
264 |
| This function or macro is the same as 'softfloat_shortShiftLeftM' with
|
|
|
265 |
| 'size_words' = 3 (N = 96).
|
|
|
266 |
*----------------------------------------------------------------------------*/
|
|
|
267 |
#define softfloat_shortShiftLeft96M( aPtr, dist, zPtr ) softfloat_shortShiftLeftM( 3, aPtr, dist, zPtr )
|
|
5000
|
268 |
|
|
|
269 |
static inline void
|
|
|
270 |
softfloat_raiseFlags( uint_fast8_t flags ) {
|
|
|
271 |
softfloat_exceptionFlags |= flags;
|
|
|
272 |
}
|
|
|
273 |
|
|
|
274 |
#if 1
|
|
|
275 |
static inline void
|
|
5014
|
276 |
softfloat_commonNaNToF128M( uint32_t *zWPtr )
|
|
4992
|
277 |
{
|
|
5014
|
278 |
zWPtr[indexWord( 4, 3 )] = defaultNaNF128UI96_32;
|
|
|
279 |
zWPtr[indexWord( 4, 2 )] = defaultNaNF128UI64_32;
|
|
|
280 |
zWPtr[indexWord( 4, 1 )] = defaultNaNF128UI32_32;
|
|
|
281 |
zWPtr[indexWord( 4, 0 )] = defaultNaNF128UI0_32;
|
|
4992
|
282 |
}
|
|
|
283 |
#else
|
|
5014
|
284 |
# define softfloat_commonNaNToF128M( zWPtr ) \
|
|
4992
|
285 |
{ \
|
|
5014
|
286 |
(zWPtr)[indexWord( 4, 3 )] = defaultNaNF128UI96_32; \
|
|
|
287 |
(zWPtr)[indexWord( 4, 2 )] = defaultNaNF128UI64_32; \
|
|
|
288 |
(zWPtr)[indexWord( 4, 1 )] = defaultNaNF128UI32_32; \
|
|
|
289 |
(zWPtr)[indexWord( 4, 0 )] = defaultNaNF128UI0_32; \
|
|
4992
|
290 |
}
|
|
|
291 |
#endif
|
|
|
292 |
|
|
5014
|
293 |
/*----------------------------------------------------------------------------
|
|
|
294 |
| Assuming the unsigned integer formed from concatenating 'uiA64' and 'uiA0'
|
|
|
295 |
| has the bit pattern of an 80-bit extended floating-point NaN, converts
|
|
|
296 |
| this NaN to the common NaN form, and stores the resulting common NaN at the
|
|
|
297 |
| location pointed to by 'zPtr'. If the NaN is a signaling NaN, the invalid
|
|
|
298 |
| exception is raised.
|
|
|
299 |
*----------------------------------------------------------------------------*/
|
|
|
300 |
#define softfloat_extF80UIToCommonNaN( uiA64, uiA0, zPtr ) if ( ! ((uiA0) & UINT64_C( 0x4000000000000000 )) ) softfloat_raiseFlags( softfloat_flag_invalid )
|
|
|
301 |
|
|
5000
|
302 |
static inline void
|
|
|
303 |
softfloat_invalidF128M( uint32_t *zWPtr ) {
|
|
|
304 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
5014
|
305 |
softfloat_commonNaNToF128M( zWPtr );
|
|
5000
|
306 |
}
|
|
|
307 |
|
|
|
308 |
static bool
|
|
|
309 |
f128M_isSignalingNaN( const float128_t *aPtr )
|
|
|
310 |
{
|
|
|
311 |
const uint32_t *aWPtr;
|
|
|
312 |
uint32_t uiA96;
|
|
|
313 |
|
|
|
314 |
aWPtr = (const uint32_t *) aPtr;
|
|
|
315 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
316 |
if ( (uiA96 & 0x7FFF8000) != 0x7FFF0000 ) return 0;
|
|
|
317 |
return
|
|
|
318 |
((uiA96 & 0x00007FFF) != 0)
|
|
|
319 |
|| ((aWPtr[indexWord( 4, 2 )] | aWPtr[indexWord( 4, 1 )]
|
|
|
320 |
| aWPtr[indexWord( 4, 0 )])
|
|
|
321 |
!= 0);
|
|
|
322 |
}
|
|
|
323 |
|
|
|
324 |
static void
|
|
|
325 |
softfloat_shortShiftRightJamM(
|
|
5014
|
326 |
uint_fast8_t size_words,
|
|
|
327 |
const uint32_t *aPtr,
|
|
|
328 |
uint_fast8_t dist,
|
|
|
329 |
uint32_t *zPtr
|
|
|
330 |
)
|
|
5000
|
331 |
{
|
|
|
332 |
uint_fast8_t uNegDist;
|
|
|
333 |
unsigned int index, lastIndex;
|
|
5014
|
334 |
uint32_t partWordZ, wordA;
|
|
5000
|
335 |
|
|
|
336 |
uNegDist = -dist;
|
|
|
337 |
index = indexWordLo( size_words );
|
|
|
338 |
lastIndex = indexWordHi( size_words );
|
|
|
339 |
wordA = aPtr[index];
|
|
|
340 |
partWordZ = wordA>>dist;
|
|
|
341 |
if ( partWordZ<<dist != wordA ) partWordZ |= 1;
|
|
|
342 |
while ( index != lastIndex ) {
|
|
|
343 |
wordA = aPtr[index + wordIncr];
|
|
5014
|
344 |
zPtr[index] = wordA<<(uNegDist & 31) | partWordZ;
|
|
5000
|
345 |
index += wordIncr;
|
|
|
346 |
partWordZ = wordA>>dist;
|
|
|
347 |
}
|
|
|
348 |
zPtr[index] = partWordZ;
|
|
5014
|
349 |
|
|
4992
|
350 |
}
|
|
|
351 |
|
|
5000
|
352 |
void
|
|
|
353 |
softfloat_shiftRightJamM(
|
|
|
354 |
uint_fast8_t size_words,
|
|
|
355 |
const uint32_t *aPtr,
|
|
|
356 |
uint32_t dist,
|
|
|
357 |
uint32_t *zPtr
|
|
|
358 |
)
|
|
|
359 |
{
|
|
|
360 |
uint32_t wordJam, wordDist, *ptr;
|
|
|
361 |
uint_fast8_t i, innerDist;
|
|
|
362 |
|
|
|
363 |
wordJam = 0;
|
|
|
364 |
wordDist = dist>>5;
|
|
|
365 |
if ( wordDist ) {
|
|
|
366 |
if ( size_words < wordDist ) wordDist = size_words;
|
|
|
367 |
ptr = (uint32_t *) (aPtr + indexMultiwordLo( size_words, wordDist ));
|
|
|
368 |
i = wordDist;
|
|
|
369 |
do {
|
|
|
370 |
wordJam = *ptr++;
|
|
|
371 |
if ( wordJam ) break;
|
|
|
372 |
--i;
|
|
|
373 |
} while ( i );
|
|
|
374 |
ptr = zPtr;
|
|
|
375 |
}
|
|
|
376 |
if ( wordDist < size_words ) {
|
|
|
377 |
aPtr += indexMultiwordHiBut( size_words, wordDist );
|
|
|
378 |
innerDist = dist & 31;
|
|
|
379 |
if ( innerDist ) {
|
|
|
380 |
softfloat_shortShiftRightJamM(
|
|
|
381 |
size_words - wordDist,
|
|
|
382 |
aPtr,
|
|
|
383 |
innerDist,
|
|
5014
|
384 |
(zPtr + indexMultiwordLoBut( size_words, wordDist ))
|
|
5000
|
385 |
);
|
|
|
386 |
if ( ! wordDist ) goto wordJam;
|
|
|
387 |
} else {
|
|
|
388 |
aPtr += indexWordLo( size_words - wordDist );
|
|
|
389 |
ptr = zPtr + indexWordLo( size_words );
|
|
|
390 |
for ( i = size_words - wordDist; i; --i ) {
|
|
|
391 |
*ptr = *aPtr;
|
|
|
392 |
aPtr += wordIncr;
|
|
|
393 |
ptr += wordIncr;
|
|
|
394 |
}
|
|
|
395 |
}
|
|
|
396 |
ptr = zPtr + indexMultiwordHi( size_words, wordDist );
|
|
|
397 |
}
|
|
|
398 |
do {
|
|
|
399 |
*ptr++ = 0;
|
|
|
400 |
--wordDist;
|
|
|
401 |
} while ( wordDist );
|
|
|
402 |
wordJam:
|
|
|
403 |
if ( wordJam ) zPtr[indexWordLo( size_words )] |= 1;
|
|
|
404 |
|
|
|
405 |
}
|
|
|
406 |
|
|
|
407 |
void
|
|
|
408 |
softfloat_shortShiftLeftM(
|
|
|
409 |
uint_fast8_t size_words,
|
|
|
410 |
const uint32_t *aPtr,
|
|
|
411 |
uint_fast8_t dist,
|
|
|
412 |
uint32_t *zPtr
|
|
|
413 |
)
|
|
|
414 |
{
|
|
|
415 |
uint_fast8_t uNegDist;
|
|
|
416 |
unsigned int index, lastIndex;
|
|
|
417 |
uint32_t partWordZ, wordA;
|
|
|
418 |
|
|
|
419 |
uNegDist = -dist;
|
|
|
420 |
index = indexWordHi( size_words );
|
|
|
421 |
lastIndex = indexWordLo( size_words );
|
|
|
422 |
partWordZ = aPtr[index]<<dist;
|
|
|
423 |
while ( index != lastIndex ) {
|
|
|
424 |
wordA = aPtr[index - wordIncr];
|
|
|
425 |
zPtr[index] = partWordZ | wordA>>(uNegDist & 31);
|
|
|
426 |
index -= wordIncr;
|
|
|
427 |
partWordZ = wordA<<dist;
|
|
|
428 |
}
|
|
|
429 |
zPtr[index] = partWordZ;
|
|
|
430 |
}
|
|
|
431 |
|
|
|
432 |
static void
|
|
|
433 |
softfloat_shiftLeftM(
|
|
|
434 |
uint_fast8_t size_words,
|
|
|
435 |
const uint32_t *aPtr,
|
|
|
436 |
uint32_t dist,
|
|
|
437 |
uint32_t *zPtr
|
|
|
438 |
)
|
|
|
439 |
{
|
|
|
440 |
uint32_t wordDist;
|
|
|
441 |
uint_fast8_t innerDist;
|
|
|
442 |
uint32_t *destPtr;
|
|
|
443 |
uint_fast8_t i;
|
|
|
444 |
|
|
|
445 |
wordDist = dist>>5;
|
|
|
446 |
if ( wordDist < size_words ) {
|
|
|
447 |
aPtr += indexMultiwordLoBut( size_words, wordDist );
|
|
|
448 |
innerDist = dist & 31;
|
|
|
449 |
if ( innerDist ) {
|
|
|
450 |
softfloat_shortShiftLeftM(
|
|
|
451 |
size_words - wordDist,
|
|
|
452 |
aPtr,
|
|
|
453 |
innerDist,
|
|
|
454 |
zPtr + indexMultiwordHiBut( size_words, wordDist )
|
|
|
455 |
);
|
|
|
456 |
if ( ! wordDist ) return;
|
|
|
457 |
} else {
|
|
|
458 |
aPtr += indexWordHi( size_words - wordDist );
|
|
|
459 |
destPtr = zPtr + indexWordHi( size_words );
|
|
|
460 |
for ( i = size_words - wordDist; i; --i ) {
|
|
|
461 |
*destPtr = *aPtr;
|
|
|
462 |
aPtr -= wordIncr;
|
|
|
463 |
destPtr -= wordIncr;
|
|
|
464 |
}
|
|
|
465 |
}
|
|
|
466 |
zPtr += indexMultiwordLo( size_words, wordDist );
|
|
|
467 |
} else {
|
|
|
468 |
wordDist = size_words;
|
|
|
469 |
}
|
|
|
470 |
do {
|
|
|
471 |
*zPtr++ = 0;
|
|
|
472 |
--wordDist;
|
|
|
473 |
} while ( wordDist );
|
|
|
474 |
}
|
|
|
475 |
|
|
|
476 |
/*----------------------------------------------------------------------------
|
|
|
477 |
| This function or macro is the same as 'softfloat_shiftLeftM' with
|
|
|
478 |
| 'size_words' = 4 (N = 128).
|
|
|
479 |
*----------------------------------------------------------------------------*/
|
|
|
480 |
#define softfloat_shiftLeft128M( aPtr, dist, zPtr ) softfloat_shiftLeftM( 4, aPtr, dist, zPtr )
|
|
|
481 |
|
|
4992
|
482 |
bool
|
|
5000
|
483 |
softfloat_isNaNF128M( const uint32_t *aWPtr )
|
|
|
484 |
{
|
|
|
485 |
uint32_t uiA96;
|
|
|
486 |
|
|
|
487 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
488 |
if ( (~uiA96 & 0x7FFF0000) != 0 ) return 0;
|
|
|
489 |
return
|
|
|
490 |
((uiA96 & 0x0000FFFF) != 0)
|
|
|
491 |
|| ((aWPtr[indexWord( 4, 2 )] | aWPtr[indexWord( 4, 1 )]
|
|
|
492 |
| aWPtr[indexWord( 4, 0 )])
|
|
|
493 |
!= 0);
|
|
|
494 |
}
|
|
|
495 |
|
|
|
496 |
static inline uint_fast8_t
|
|
|
497 |
softfloat_countLeadingZeros32( uint32_t a ) {
|
|
|
498 |
return a ? __builtin_clz( a ) : 32;
|
|
|
499 |
}
|
|
|
500 |
|
|
|
501 |
static inline bool
|
|
4992
|
502 |
softfloat_eq128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 ){
|
|
|
503 |
return (a64 == b64) && (a0 == b0);
|
|
|
504 |
}
|
|
|
505 |
|
|
5000
|
506 |
static inline bool
|
|
4992
|
507 |
softfloat_lt128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 ){
|
|
|
508 |
return (a64 < b64) || ((a64 == b64) && (a0 < b0));
|
|
|
509 |
}
|
|
|
510 |
|
|
5004
|
511 |
static inline bool
|
|
|
512 |
softfloat_le128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 ){
|
|
|
513 |
return (a64 < b64) || ((a64 == b64) && (a0 <= b0));
|
|
|
514 |
}
|
|
|
515 |
|
|
5000
|
516 |
void
|
|
|
517 |
softfloat_propagateNaNF128M(
|
|
|
518 |
const uint32_t *aWPtr, const uint32_t *bWPtr, uint32_t *zWPtr )
|
|
|
519 |
{
|
|
|
520 |
bool isSigNaNA;
|
|
|
521 |
const uint32_t *ptr;
|
|
|
522 |
bool isSigNaNB;
|
|
|
523 |
uint32_t uiA96, uiB96, wordMagA, wordMagB;
|
|
|
524 |
|
|
|
525 |
isSigNaNA = f128M_isSignalingNaN( (const float128_t *) aWPtr );
|
|
|
526 |
ptr = aWPtr;
|
|
|
527 |
if ( ! bWPtr ) {
|
|
|
528 |
if ( isSigNaNA ) softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
529 |
goto copy;
|
|
|
530 |
}
|
|
|
531 |
isSigNaNB = f128M_isSignalingNaN( (const float128_t *) bWPtr );
|
|
|
532 |
if ( isSigNaNA | isSigNaNB ) {
|
|
|
533 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
534 |
if ( isSigNaNA ) {
|
|
|
535 |
if ( isSigNaNB ) goto returnLargerUIMag;
|
|
|
536 |
if ( softfloat_isNaNF128M( bWPtr ) ) goto copyB;
|
|
|
537 |
goto copy;
|
|
|
538 |
} else {
|
|
|
539 |
if ( softfloat_isNaNF128M( aWPtr ) ) goto copy;
|
|
|
540 |
goto copyB;
|
|
|
541 |
}
|
|
|
542 |
}
|
|
|
543 |
returnLargerUIMag:
|
|
|
544 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
545 |
uiB96 = bWPtr[indexWordHi( 4 )];
|
|
|
546 |
wordMagA = uiA96 & 0x7FFFFFFF;
|
|
|
547 |
wordMagB = uiB96 & 0x7FFFFFFF;
|
|
|
548 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
549 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
550 |
wordMagA = aWPtr[indexWord( 4, 2 )];
|
|
|
551 |
wordMagB = bWPtr[indexWord( 4, 2 )];
|
|
|
552 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
553 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
554 |
wordMagA = aWPtr[indexWord( 4, 1 )];
|
|
|
555 |
wordMagB = bWPtr[indexWord( 4, 1 )];
|
|
|
556 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
557 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
558 |
wordMagA = aWPtr[indexWord( 4, 0 )];
|
|
|
559 |
wordMagB = bWPtr[indexWord( 4, 0 )];
|
|
|
560 |
if ( wordMagA < wordMagB ) goto copyB;
|
|
|
561 |
if ( wordMagB < wordMagA ) goto copy;
|
|
|
562 |
if ( uiA96 < uiB96 ) goto copy;
|
|
|
563 |
copyB:
|
|
|
564 |
ptr = bWPtr;
|
|
|
565 |
copy:
|
|
|
566 |
zWPtr[indexWordHi( 4 )] = ptr[indexWordHi( 4 )] | 0x00008000;
|
|
|
567 |
zWPtr[indexWord( 4, 2 )] = ptr[indexWord( 4, 2 )];
|
|
|
568 |
zWPtr[indexWord( 4, 1 )] = ptr[indexWord( 4, 1 )];
|
|
|
569 |
zWPtr[indexWord( 4, 0 )] = ptr[indexWord( 4, 0 )];
|
|
|
570 |
}
|
|
|
571 |
|
|
|
572 |
static inline uint_fast8_t
|
|
4992
|
573 |
softfloat_countLeadingZeros64( uint64_t a )
|
|
|
574 |
{
|
|
|
575 |
uint_fast8_t count;
|
|
|
576 |
uint32_t a32;
|
|
|
577 |
|
|
|
578 |
count = 0;
|
|
|
579 |
a32 = a>>32;
|
|
|
580 |
if ( ! a32 ) {
|
|
|
581 |
count = 32;
|
|
|
582 |
a32 = a;
|
|
|
583 |
}
|
|
|
584 |
/*------------------------------------------------------------------------
|
|
|
585 |
| From here, result is current count + count leading zeros of `a32'.
|
|
|
586 |
*------------------------------------------------------------------------*/
|
|
|
587 |
if ( a32 < 0x10000 ) {
|
|
|
588 |
count += 16;
|
|
|
589 |
a32 <<= 16;
|
|
|
590 |
}
|
|
|
591 |
if ( a32 < 0x1000000 ) {
|
|
|
592 |
count += 8;
|
|
|
593 |
a32 <<= 8;
|
|
|
594 |
}
|
|
|
595 |
count += softfloat_countLeadingZeros8[a32>>24];
|
|
|
596 |
return count;
|
|
|
597 |
}
|
|
|
598 |
|
|
5000
|
599 |
static inline struct exp16_sig64
|
|
4992
|
600 |
softfloat_normSubnormalF64Sig( uint_fast64_t sig )
|
|
|
601 |
{
|
|
|
602 |
int_fast8_t shiftDist;
|
|
|
603 |
struct exp16_sig64 z;
|
|
|
604 |
|
|
|
605 |
shiftDist = softfloat_countLeadingZeros64( sig ) - 11;
|
|
|
606 |
z.exp = 1 - shiftDist;
|
|
|
607 |
z.sig = sig<<shiftDist;
|
|
|
608 |
return z;
|
|
|
609 |
}
|
|
|
610 |
|
|
5000
|
611 |
static inline struct uint128_extra
|
|
4992
|
612 |
softfloat_shiftRightJam128Extra(
|
|
|
613 |
uint64_t a64, uint64_t a0, uint64_t extra, uint_fast32_t dist )
|
|
|
614 |
{
|
|
|
615 |
uint_fast8_t u8NegDist;
|
|
|
616 |
struct uint128_extra z;
|
|
|
617 |
|
|
|
618 |
u8NegDist = -dist;
|
|
|
619 |
if ( dist < 64 ) {
|
|
|
620 |
z.v.v64 = a64>>dist;
|
|
|
621 |
z.v.v0 = a64<<(u8NegDist & 63) | a0>>dist;
|
|
|
622 |
z.extra = a0<<(u8NegDist & 63);
|
|
|
623 |
} else {
|
|
|
624 |
z.v.v64 = 0;
|
|
|
625 |
if ( dist == 64 ) {
|
|
|
626 |
z.v.v0 = a64;
|
|
|
627 |
z.extra = a0;
|
|
|
628 |
} else {
|
|
|
629 |
extra |= a0;
|
|
|
630 |
if ( dist < 128 ) {
|
|
|
631 |
z.v.v0 = a64>>(dist & 63);
|
|
|
632 |
z.extra = a64<<(u8NegDist & 63);
|
|
|
633 |
} else {
|
|
|
634 |
z.v.v0 = 0;
|
|
|
635 |
z.extra = (dist == 128) ? a64 : (a64 != 0);
|
|
|
636 |
}
|
|
|
637 |
}
|
|
|
638 |
}
|
|
|
639 |
z.extra |= (extra != 0);
|
|
|
640 |
return z;
|
|
|
641 |
}
|
|
|
642 |
|
|
5000
|
643 |
static inline struct uint128_extra
|
|
4992
|
644 |
softfloat_shortShiftRightJam128Extra(
|
|
|
645 |
uint64_t a64, uint64_t a0, uint64_t extra, uint_fast8_t dist )
|
|
|
646 |
{
|
|
|
647 |
uint_fast8_t negDist = -dist;
|
|
|
648 |
struct uint128_extra z;
|
|
|
649 |
z.v.v64 = a64>>dist;
|
|
|
650 |
z.v.v0 = a64<<(negDist & 63) | a0>>dist;
|
|
|
651 |
z.extra = a0<<(negDist & 63) | (extra != 0);
|
|
|
652 |
return z;
|
|
|
653 |
}
|
|
|
654 |
|
|
5000
|
655 |
static inline struct uint128
|
|
|
656 |
softfloat_shiftRightJam128( uint64_t a64, uint64_t a0, uint_fast32_t dist )
|
|
4992
|
657 |
{
|
|
|
658 |
uint_fast8_t u8NegDist;
|
|
|
659 |
struct uint128 z;
|
|
|
660 |
|
|
|
661 |
if ( dist < 64 ) {
|
|
|
662 |
u8NegDist = -dist;
|
|
|
663 |
z.v64 = a64>>dist;
|
|
|
664 |
z.v0 =
|
|
|
665 |
a64<<(u8NegDist & 63) | a0>>dist
|
|
|
666 |
| ((uint64_t) (a0<<(u8NegDist & 63)) != 0);
|
|
|
667 |
} else {
|
|
|
668 |
z.v64 = 0;
|
|
|
669 |
z.v0 =
|
|
|
670 |
(dist < 127)
|
|
|
671 |
? a64>>(dist & 63)
|
|
|
672 |
| (((a64 & (((uint_fast64_t) 1<<(dist & 63)) - 1)) | a0)
|
|
|
673 |
!= 0)
|
|
|
674 |
: ((a64 | a0) != 0);
|
|
|
675 |
}
|
|
|
676 |
return z;
|
|
|
677 |
}
|
|
|
678 |
|
|
5000
|
679 |
static inline struct uint128
|
|
4992
|
680 |
softfloat_shortShiftLeft128( uint64_t a64, uint64_t a0, uint_fast8_t dist )
|
|
|
681 |
{
|
|
|
682 |
struct uint128 z;
|
|
|
683 |
z.v64 = a64<<dist | a0>>(-dist & 63);
|
|
|
684 |
z.v0 = a0<<dist;
|
|
|
685 |
return z;
|
|
|
686 |
}
|
|
|
687 |
|
|
5000
|
688 |
static int_fast8_t
|
|
|
689 |
softfloat_compare128M( const uint32_t *aPtr, const uint32_t *bPtr )
|
|
|
690 |
{
|
|
|
691 |
unsigned int index, lastIndex;
|
|
|
692 |
uint32_t wordA, wordB;
|
|
|
693 |
|
|
|
694 |
index = indexWordHi( 4 );
|
|
|
695 |
lastIndex = indexWordLo( 4 );
|
|
|
696 |
for (;;) {
|
|
|
697 |
wordA = aPtr[index];
|
|
|
698 |
wordB = bPtr[index];
|
|
|
699 |
if ( wordA != wordB ) return (wordA < wordB) ? -1 : 1;
|
|
|
700 |
if ( index == lastIndex ) break;
|
|
|
701 |
index -= wordIncr;
|
|
|
702 |
}
|
|
|
703 |
return 0;
|
|
|
704 |
}
|
|
|
705 |
|
|
|
706 |
static void
|
|
|
707 |
softfloat_roundPackMToF128M(
|
|
|
708 |
bool sign, int32_t exp, uint32_t *extSigPtr, uint32_t *zWPtr )
|
|
|
709 |
{
|
|
|
710 |
uint_fast8_t roundingMode;
|
|
|
711 |
bool roundNearEven;
|
|
|
712 |
uint32_t sigExtra;
|
|
|
713 |
bool doIncrement, isTiny;
|
|
|
714 |
static const uint32_t maxSig[4] =
|
|
|
715 |
INIT_UINTM4( 0x0001FFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF );
|
|
|
716 |
uint32_t ui, uj;
|
|
|
717 |
|
|
|
718 |
/*------------------------------------------------------------------------
|
|
|
719 |
*------------------------------------------------------------------------*/
|
|
|
720 |
roundingMode = softfloat_roundingMode;
|
|
|
721 |
roundNearEven = (roundingMode == softfloat_round_near_even);
|
|
|
722 |
sigExtra = extSigPtr[indexWordLo( 5 )];
|
|
|
723 |
doIncrement = (0x80000000 <= sigExtra);
|
|
|
724 |
if ( ! roundNearEven && (roundingMode != softfloat_round_near_maxMag) ) {
|
|
|
725 |
doIncrement =
|
|
|
726 |
(roundingMode
|
|
|
727 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
728 |
&& sigExtra;
|
|
|
729 |
}
|
|
|
730 |
/*------------------------------------------------------------------------
|
|
|
731 |
*------------------------------------------------------------------------*/
|
|
|
732 |
if ( 0x7FFD <= (uint32_t) exp ) {
|
|
|
733 |
if ( exp < 0 ) {
|
|
|
734 |
/*----------------------------------------------------------------
|
|
|
735 |
*----------------------------------------------------------------*/
|
|
|
736 |
isTiny =
|
|
|
737 |
(softfloat_detectTininess
|
|
|
738 |
== softfloat_tininess_beforeRounding)
|
|
|
739 |
|| (exp < -1)
|
|
|
740 |
|| ! doIncrement
|
|
|
741 |
|| (softfloat_compare128M(
|
|
|
742 |
extSigPtr + indexMultiwordHi( 5, 4 ), maxSig )
|
|
|
743 |
< 0);
|
|
|
744 |
softfloat_shiftRightJam160M( extSigPtr, -exp, extSigPtr );
|
|
|
745 |
exp = 0;
|
|
|
746 |
sigExtra = extSigPtr[indexWordLo( 5 )];
|
|
|
747 |
if ( isTiny && sigExtra ) {
|
|
|
748 |
softfloat_raiseFlags( softfloat_flag_underflow );
|
|
|
749 |
}
|
|
|
750 |
doIncrement = (0x80000000 <= sigExtra);
|
|
|
751 |
if (
|
|
|
752 |
! roundNearEven
|
|
|
753 |
&& (roundingMode != softfloat_round_near_maxMag)
|
|
|
754 |
) {
|
|
|
755 |
doIncrement =
|
|
|
756 |
(roundingMode
|
|
|
757 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
758 |
&& sigExtra;
|
|
|
759 |
}
|
|
|
760 |
} else if (
|
|
|
761 |
(0x7FFD < exp)
|
|
|
762 |
|| ((exp == 0x7FFD) && doIncrement
|
|
|
763 |
&& (softfloat_compare128M(
|
|
|
764 |
extSigPtr + indexMultiwordHi( 5, 4 ), maxSig )
|
|
|
765 |
== 0))
|
|
|
766 |
) {
|
|
|
767 |
/*----------------------------------------------------------------
|
|
|
768 |
*----------------------------------------------------------------*/
|
|
|
769 |
softfloat_raiseFlags(
|
|
|
770 |
softfloat_flag_overflow | softfloat_flag_inexact );
|
|
|
771 |
if (
|
|
|
772 |
roundNearEven
|
|
|
773 |
|| (roundingMode == softfloat_round_near_maxMag)
|
|
|
774 |
|| (roundingMode
|
|
|
775 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
776 |
) {
|
|
|
777 |
ui = packToF128UI96( sign, 0x7FFF, 0 );
|
|
|
778 |
uj = 0;
|
|
|
779 |
} else {
|
|
|
780 |
ui = packToF128UI96( sign, 0x7FFE, 0x0000FFFF );
|
|
|
781 |
uj = 0xFFFFFFFF;
|
|
|
782 |
}
|
|
|
783 |
zWPtr[indexWordHi( 4 )] = ui;
|
|
|
784 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
785 |
zWPtr[indexWord( 4, 1 )] = uj;
|
|
|
786 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
787 |
return;
|
|
|
788 |
}
|
|
|
789 |
}
|
|
|
790 |
/*------------------------------------------------------------------------
|
|
|
791 |
*------------------------------------------------------------------------*/
|
|
|
792 |
uj = extSigPtr[indexWord( 5, 1 )];
|
|
|
793 |
if ( sigExtra ) {
|
|
|
794 |
softfloat_exceptionFlags |= softfloat_flag_inexact;
|
|
|
795 |
#ifdef SOFTFLOAT_ROUND_ODD
|
|
|
796 |
if ( roundingMode == softfloat_round_odd ) {
|
|
|
797 |
uj |= 1;
|
|
|
798 |
goto noIncrementPackReturn;
|
|
|
799 |
}
|
|
|
800 |
#endif
|
|
|
801 |
}
|
|
|
802 |
if ( doIncrement ) {
|
|
|
803 |
++uj;
|
|
|
804 |
if ( uj ) {
|
|
|
805 |
if ( ! (sigExtra & 0x7FFFFFFF) && roundNearEven ) uj &= ~1;
|
|
|
806 |
zWPtr[indexWord( 4, 2 )] = extSigPtr[indexWord( 5, 3 )];
|
|
|
807 |
zWPtr[indexWord( 4, 1 )] = extSigPtr[indexWord( 5, 2 )];
|
|
|
808 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
809 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
810 |
} else {
|
|
|
811 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
812 |
ui = extSigPtr[indexWord( 5, 2 )] + 1;
|
|
|
813 |
zWPtr[indexWord( 4, 1 )] = ui;
|
|
|
814 |
uj = extSigPtr[indexWord( 5, 3 )];
|
|
|
815 |
if ( ui ) {
|
|
|
816 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
817 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
818 |
} else {
|
|
|
819 |
++uj;
|
|
|
820 |
zWPtr[indexWord( 4, 2 )] = uj;
|
|
|
821 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
822 |
if ( ! uj ) ++ui;
|
|
|
823 |
}
|
|
|
824 |
}
|
|
|
825 |
} else {
|
|
|
826 |
noIncrementPackReturn:
|
|
|
827 |
zWPtr[indexWord( 4, 0 )] = uj;
|
|
|
828 |
ui = extSigPtr[indexWord( 5, 2 )];
|
|
|
829 |
zWPtr[indexWord( 4, 1 )] = ui;
|
|
|
830 |
uj |= ui;
|
|
|
831 |
ui = extSigPtr[indexWord( 5, 3 )];
|
|
|
832 |
zWPtr[indexWord( 4, 2 )] = ui;
|
|
|
833 |
uj |= ui;
|
|
|
834 |
ui = extSigPtr[indexWordHi( 5 )];
|
|
|
835 |
uj |= ui;
|
|
|
836 |
if ( ! uj ) exp = 0;
|
|
|
837 |
}
|
|
|
838 |
zWPtr[indexWordHi( 4 )] = packToF128UI96( sign, exp, ui );
|
|
|
839 |
}
|
|
|
840 |
|
|
|
841 |
static int
|
|
|
842 |
softfloat_shiftNormSigF128M(
|
|
|
843 |
const uint32_t *wPtr, uint_fast8_t shiftDist, uint32_t *sigPtr )
|
|
|
844 |
{
|
|
|
845 |
uint32_t wordSig;
|
|
|
846 |
int32_t exp;
|
|
|
847 |
uint32_t leadingBit;
|
|
|
848 |
|
|
|
849 |
wordSig = wPtr[indexWordHi( 4 )];
|
|
|
850 |
exp = expF128UI96( wordSig );
|
|
|
851 |
if ( exp ) {
|
|
|
852 |
softfloat_shortShiftLeft128M( wPtr, shiftDist, sigPtr );
|
|
|
853 |
leadingBit = 0x00010000<<shiftDist;
|
|
|
854 |
sigPtr[indexWordHi( 4 )] =
|
|
|
855 |
(sigPtr[indexWordHi( 4 )] & (leadingBit - 1)) | leadingBit;
|
|
|
856 |
} else {
|
|
|
857 |
exp = 16;
|
|
|
858 |
wordSig &= 0x7FFFFFFF;
|
|
|
859 |
if ( ! wordSig ) {
|
|
|
860 |
exp = -16;
|
|
|
861 |
wordSig = wPtr[indexWord( 4, 2 )];
|
|
|
862 |
if ( ! wordSig ) {
|
|
|
863 |
exp = -48;
|
|
|
864 |
wordSig = wPtr[indexWord( 4, 1 )];
|
|
|
865 |
if ( ! wordSig ) {
|
|
|
866 |
wordSig = wPtr[indexWord( 4, 0 )];
|
|
|
867 |
if ( ! wordSig ) return -128;
|
|
|
868 |
exp = -80;
|
|
|
869 |
}
|
|
|
870 |
}
|
|
|
871 |
}
|
|
|
872 |
exp -= softfloat_countLeadingZeros32( wordSig );
|
|
|
873 |
softfloat_shiftLeft128M( wPtr, 1 - exp + shiftDist, sigPtr );
|
|
|
874 |
}
|
|
|
875 |
return exp;
|
|
|
876 |
}
|
|
|
877 |
|
|
|
878 |
bool
|
|
|
879 |
softfloat_tryPropagateNaNF128M(
|
|
|
880 |
const uint32_t *aWPtr, const uint32_t *bWPtr, uint32_t *zWPtr )
|
|
|
881 |
{
|
|
|
882 |
|
|
|
883 |
if ( softfloat_isNaNF128M( aWPtr ) || softfloat_isNaNF128M( bWPtr ) ) {
|
|
|
884 |
softfloat_propagateNaNF128M( aWPtr, bWPtr, zWPtr );
|
|
|
885 |
return 1;
|
|
|
886 |
}
|
|
|
887 |
return 0;
|
|
|
888 |
}
|
|
|
889 |
|
|
|
890 |
static inline struct uint128
|
|
4992
|
891 |
softfloat_add128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 )
|
|
|
892 |
{
|
|
|
893 |
struct uint128 z;
|
|
|
894 |
z.v0 = a0 + b0;
|
|
|
895 |
z.v64 = a64 + b64 + (z.v0 < a0);
|
|
|
896 |
return z;
|
|
|
897 |
}
|
|
|
898 |
|
|
5000
|
899 |
static inline struct uint128
|
|
|
900 |
softfloat_sub128( uint64_t a64, uint64_t a0, uint64_t b64, uint64_t b0 )
|
|
4992
|
901 |
{
|
|
|
902 |
struct uint128 z;
|
|
|
903 |
z.v0 = a0 - b0;
|
|
|
904 |
z.v64 = a64 - b64;
|
|
|
905 |
z.v64 -= (a0 < b0);
|
|
|
906 |
return z;
|
|
|
907 |
}
|
|
|
908 |
|
|
5004
|
909 |
static inline struct uint128
|
|
|
910 |
softfloat_mul128By32( uint64_t a64, uint64_t a0, uint32_t b )
|
|
|
911 |
{
|
|
|
912 |
union { unsigned __int128 ui; struct uint128 s; } uZ;
|
|
|
913 |
uZ.ui = ((unsigned __int128) a64<<64 | a0) * b;
|
|
|
914 |
return uZ.s;
|
|
|
915 |
}
|
|
|
916 |
|
|
5000
|
917 |
static void
|
|
|
918 |
softfloat_mul128MTo256M(
|
|
|
919 |
const uint32_t *aPtr, const uint32_t *bPtr, uint32_t *zPtr )
|
|
|
920 |
{
|
|
|
921 |
uint32_t *lastZPtr, wordB;
|
|
|
922 |
uint64_t dwordProd;
|
|
|
923 |
uint32_t wordZ;
|
|
|
924 |
uint_fast8_t carry;
|
|
|
925 |
|
|
|
926 |
bPtr += indexWordLo( 4 );
|
|
|
927 |
lastZPtr = zPtr + indexMultiwordHi( 8, 5 );
|
|
|
928 |
zPtr += indexMultiwordLo( 8, 5 );
|
|
|
929 |
wordB = *bPtr;
|
|
|
930 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 0 )] * wordB;
|
|
|
931 |
zPtr[indexWord( 5, 0 )] = dwordProd;
|
|
|
932 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 1 )] * wordB + (dwordProd>>32);
|
|
|
933 |
zPtr[indexWord( 5, 1 )] = dwordProd;
|
|
|
934 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 2 )] * wordB + (dwordProd>>32);
|
|
|
935 |
zPtr[indexWord( 5, 2 )] = dwordProd;
|
|
|
936 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 3 )] * wordB + (dwordProd>>32);
|
|
|
937 |
zPtr[indexWord( 5, 3 )] = dwordProd;
|
|
|
938 |
zPtr[indexWord( 5, 4 )] = dwordProd>>32;
|
|
|
939 |
do {
|
|
|
940 |
bPtr += wordIncr;
|
|
|
941 |
zPtr += wordIncr;
|
|
|
942 |
wordB = *bPtr;
|
|
|
943 |
dwordProd = (uint64_t) aPtr[indexWord( 4, 0 )] * wordB;
|
|
|
944 |
wordZ = zPtr[indexWord( 5, 0 )] + (uint32_t) dwordProd;
|
|
|
945 |
zPtr[indexWord( 5, 0 )] = wordZ;
|
|
|
946 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
947 |
dwordProd =
|
|
|
948 |
(uint64_t) aPtr[indexWord( 4, 1 )] * wordB + (dwordProd>>32);
|
|
|
949 |
wordZ = zPtr[indexWord( 5, 1 )] + (uint32_t) dwordProd + carry;
|
|
|
950 |
zPtr[indexWord( 5, 1 )] = wordZ;
|
|
|
951 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
952 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
953 |
}
|
|
|
954 |
dwordProd =
|
|
|
955 |
(uint64_t) aPtr[indexWord( 4, 2 )] * wordB + (dwordProd>>32);
|
|
|
956 |
wordZ = zPtr[indexWord( 5, 2 )] + (uint32_t) dwordProd + carry;
|
|
|
957 |
zPtr[indexWord( 5, 2 )] = wordZ;
|
|
|
958 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
959 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
960 |
}
|
|
|
961 |
dwordProd =
|
|
|
962 |
(uint64_t) aPtr[indexWord( 4, 3 )] * wordB + (dwordProd>>32);
|
|
|
963 |
wordZ = zPtr[indexWord( 5, 3 )] + (uint32_t) dwordProd + carry;
|
|
|
964 |
zPtr[indexWord( 5, 3 )] = wordZ;
|
|
|
965 |
if ( wordZ != (uint32_t) dwordProd ) {
|
|
|
966 |
carry = (wordZ < (uint32_t) dwordProd);
|
|
|
967 |
}
|
|
|
968 |
zPtr[indexWord( 5, 4 )] = (dwordProd>>32) + carry;
|
|
|
969 |
} while ( zPtr != lastZPtr );
|
|
|
970 |
}
|
|
|
971 |
|
|
4992
|
972 |
float128_t
|
|
|
973 |
softfloat_roundPackToF128(
|
|
|
974 |
bool sign,
|
|
|
975 |
int_fast32_t exp,
|
|
|
976 |
uint_fast64_t sig64,
|
|
|
977 |
uint_fast64_t sig0,
|
|
|
978 |
uint_fast64_t sigExtra
|
|
|
979 |
)
|
|
|
980 |
{
|
|
|
981 |
uint_fast8_t roundingMode;
|
|
|
982 |
bool roundNearEven, doIncrement, isTiny;
|
|
|
983 |
struct uint128_extra sig128Extra;
|
|
|
984 |
uint_fast64_t uiZ64, uiZ0;
|
|
|
985 |
struct uint128 sig128;
|
|
|
986 |
union ui128_f128 uZ;
|
|
|
987 |
|
|
|
988 |
/*------------------------------------------------------------------------
|
|
|
989 |
*------------------------------------------------------------------------*/
|
|
|
990 |
roundingMode = softfloat_roundingMode;
|
|
|
991 |
roundNearEven = (roundingMode == softfloat_round_near_even);
|
|
|
992 |
doIncrement = (UINT64_C( 0x8000000000000000 ) <= sigExtra);
|
|
|
993 |
if ( ! roundNearEven && (roundingMode != softfloat_round_near_maxMag) ) {
|
|
|
994 |
doIncrement =
|
|
|
995 |
(roundingMode
|
|
|
996 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
997 |
&& sigExtra;
|
|
|
998 |
}
|
|
|
999 |
/*------------------------------------------------------------------------
|
|
|
1000 |
*------------------------------------------------------------------------*/
|
|
|
1001 |
if ( 0x7FFD <= (uint32_t) exp ) {
|
|
|
1002 |
if ( exp < 0 ) {
|
|
|
1003 |
/*----------------------------------------------------------------
|
|
|
1004 |
*----------------------------------------------------------------*/
|
|
|
1005 |
isTiny =
|
|
|
1006 |
(softfloat_detectTininess
|
|
|
1007 |
== softfloat_tininess_beforeRounding)
|
|
|
1008 |
|| (exp < -1)
|
|
|
1009 |
|| ! doIncrement
|
|
|
1010 |
|| softfloat_lt128(
|
|
|
1011 |
sig64,
|
|
|
1012 |
sig0,
|
|
|
1013 |
UINT64_C( 0x0001FFFFFFFFFFFF ),
|
|
|
1014 |
UINT64_C( 0xFFFFFFFFFFFFFFFF )
|
|
|
1015 |
);
|
|
|
1016 |
sig128Extra =
|
|
|
1017 |
softfloat_shiftRightJam128Extra( sig64, sig0, sigExtra, -exp );
|
|
|
1018 |
sig64 = sig128Extra.v.v64;
|
|
|
1019 |
sig0 = sig128Extra.v.v0;
|
|
|
1020 |
sigExtra = sig128Extra.extra;
|
|
|
1021 |
exp = 0;
|
|
|
1022 |
if ( isTiny && sigExtra ) {
|
|
|
1023 |
softfloat_raiseFlags( softfloat_flag_underflow );
|
|
|
1024 |
}
|
|
|
1025 |
doIncrement = (UINT64_C( 0x8000000000000000 ) <= sigExtra);
|
|
|
1026 |
if (
|
|
|
1027 |
! roundNearEven
|
|
|
1028 |
&& (roundingMode != softfloat_round_near_maxMag)
|
|
|
1029 |
) {
|
|
|
1030 |
doIncrement =
|
|
|
1031 |
(roundingMode
|
|
|
1032 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
1033 |
&& sigExtra;
|
|
|
1034 |
}
|
|
|
1035 |
} else if (
|
|
|
1036 |
(0x7FFD < exp)
|
|
|
1037 |
|| ((exp == 0x7FFD)
|
|
|
1038 |
&& softfloat_eq128(
|
|
|
1039 |
sig64,
|
|
|
1040 |
sig0,
|
|
|
1041 |
UINT64_C( 0x0001FFFFFFFFFFFF ),
|
|
|
1042 |
UINT64_C( 0xFFFFFFFFFFFFFFFF )
|
|
|
1043 |
)
|
|
|
1044 |
&& doIncrement)
|
|
|
1045 |
) {
|
|
|
1046 |
/*----------------------------------------------------------------
|
|
|
1047 |
*----------------------------------------------------------------*/
|
|
|
1048 |
softfloat_raiseFlags(
|
|
|
1049 |
softfloat_flag_overflow | softfloat_flag_inexact );
|
|
|
1050 |
if (
|
|
|
1051 |
roundNearEven
|
|
|
1052 |
|| (roundingMode == softfloat_round_near_maxMag)
|
|
|
1053 |
|| (roundingMode
|
|
|
1054 |
== (sign ? softfloat_round_min : softfloat_round_max))
|
|
|
1055 |
) {
|
|
|
1056 |
uiZ64 = packToF128UI64( sign, 0x7FFF, 0 );
|
|
|
1057 |
uiZ0 = 0;
|
|
|
1058 |
} else {
|
|
5011
|
1059 |
uiZ64 = packToF128UI64( sign, 0x7FFE, UINT64_C( 0x0000FFFFFFFFFFFF ) );
|
|
4992
|
1060 |
uiZ0 = UINT64_C( 0xFFFFFFFFFFFFFFFF );
|
|
|
1061 |
}
|
|
|
1062 |
goto uiZ;
|
|
|
1063 |
}
|
|
|
1064 |
}
|
|
|
1065 |
/*------------------------------------------------------------------------
|
|
|
1066 |
*------------------------------------------------------------------------*/
|
|
|
1067 |
if ( sigExtra ) {
|
|
|
1068 |
softfloat_exceptionFlags |= softfloat_flag_inexact;
|
|
|
1069 |
#ifdef SOFTFLOAT_ROUND_ODD
|
|
|
1070 |
if ( roundingMode == softfloat_round_odd ) {
|
|
|
1071 |
sig0 |= 1;
|
|
|
1072 |
goto packReturn;
|
|
|
1073 |
}
|
|
|
1074 |
#endif
|
|
|
1075 |
}
|
|
|
1076 |
if ( doIncrement ) {
|
|
|
1077 |
sig128 = softfloat_add128( sig64, sig0, 0, 1 );
|
|
|
1078 |
sig64 = sig128.v64;
|
|
|
1079 |
sig0 =
|
|
|
1080 |
sig128.v0
|
|
|
1081 |
& ~(uint64_t)
|
|
|
1082 |
(! (sigExtra & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
|
|
|
1083 |
& roundNearEven);
|
|
|
1084 |
} else {
|
|
|
1085 |
if ( ! (sig64 | sig0) ) exp = 0;
|
|
|
1086 |
}
|
|
|
1087 |
/*------------------------------------------------------------------------
|
|
|
1088 |
*------------------------------------------------------------------------*/
|
|
|
1089 |
packReturn:
|
|
|
1090 |
uiZ64 = packToF128UI64( sign, exp, sig64 );
|
|
|
1091 |
uiZ0 = sig0;
|
|
|
1092 |
uiZ:
|
|
|
1093 |
uZ.ui.v64 = uiZ64;
|
|
|
1094 |
uZ.ui.v0 = uiZ0;
|
|
|
1095 |
return uZ.f;
|
|
|
1096 |
}
|
|
|
1097 |
|
|
|
1098 |
float128_t
|
|
5004
|
1099 |
softfloat_normRoundPackToF128(
|
|
4992
|
1100 |
bool sign, int_fast32_t exp, uint_fast64_t sig64, uint_fast64_t sig0 )
|
|
|
1101 |
{
|
|
|
1102 |
int_fast8_t shiftDist;
|
|
|
1103 |
struct uint128 sig128;
|
|
|
1104 |
union ui128_f128 uZ;
|
|
|
1105 |
uint_fast64_t sigExtra;
|
|
|
1106 |
struct uint128_extra sig128Extra;
|
|
|
1107 |
|
|
|
1108 |
if ( ! sig64 ) {
|
|
|
1109 |
exp -= 64;
|
|
|
1110 |
sig64 = sig0;
|
|
|
1111 |
sig0 = 0;
|
|
|
1112 |
}
|
|
|
1113 |
shiftDist = softfloat_countLeadingZeros64( sig64 ) - 15;
|
|
|
1114 |
exp -= shiftDist;
|
|
|
1115 |
if ( 0 <= shiftDist ) {
|
|
|
1116 |
if ( shiftDist ) {
|
|
|
1117 |
sig128 = softfloat_shortShiftLeft128( sig64, sig0, shiftDist );
|
|
|
1118 |
sig64 = sig128.v64;
|
|
|
1119 |
sig0 = sig128.v0;
|
|
|
1120 |
}
|
|
|
1121 |
if ( (uint32_t) exp < 0x7FFD ) {
|
|
|
1122 |
uZ.ui.v64 = packToF128UI64( sign, sig64 | sig0 ? exp : 0, sig64 );
|
|
|
1123 |
uZ.ui.v0 = sig0;
|
|
|
1124 |
return uZ.f;
|
|
|
1125 |
}
|
|
|
1126 |
sigExtra = 0;
|
|
|
1127 |
} else {
|
|
|
1128 |
sig128Extra =
|
|
|
1129 |
softfloat_shortShiftRightJam128Extra( sig64, sig0, 0, -shiftDist );
|
|
|
1130 |
sig64 = sig128Extra.v.v64;
|
|
|
1131 |
sig0 = sig128Extra.v.v0;
|
|
|
1132 |
sigExtra = sig128Extra.extra;
|
|
|
1133 |
}
|
|
|
1134 |
return softfloat_roundPackToF128( sign, exp, sig64, sig0, sigExtra );
|
|
|
1135 |
}
|
|
|
1136 |
|
|
5004
|
1137 |
struct exp32_sig128
|
|
|
1138 |
softfloat_normSubnormalF128Sig( uint_fast64_t sig64, uint_fast64_t sig0 )
|
|
|
1139 |
{
|
|
|
1140 |
int_fast8_t shiftDist;
|
|
|
1141 |
struct exp32_sig128 z;
|
|
|
1142 |
|
|
|
1143 |
if ( ! sig64 ) {
|
|
|
1144 |
shiftDist = softfloat_countLeadingZeros64( sig0 ) - 15;
|
|
|
1145 |
z.exp = -63 - shiftDist;
|
|
|
1146 |
if ( shiftDist < 0 ) {
|
|
|
1147 |
z.sig.v64 = sig0>>-shiftDist;
|
|
|
1148 |
z.sig.v0 = sig0<<(shiftDist & 63);
|
|
|
1149 |
} else {
|
|
|
1150 |
z.sig.v64 = sig0<<shiftDist;
|
|
|
1151 |
z.sig.v0 = 0;
|
|
|
1152 |
}
|
|
|
1153 |
} else {
|
|
|
1154 |
shiftDist = softfloat_countLeadingZeros64( sig64 ) - 15;
|
|
|
1155 |
z.exp = 1 - shiftDist;
|
|
|
1156 |
z.sig = softfloat_shortShiftLeft128( sig64, sig0, shiftDist );
|
|
|
1157 |
}
|
|
|
1158 |
return z;
|
|
|
1159 |
}
|
|
|
1160 |
|
|
4992
|
1161 |
struct uint128
|
|
|
1162 |
softfloat_propagateNaNF128UI(
|
|
|
1163 |
uint_fast64_t uiA64,
|
|
|
1164 |
uint_fast64_t uiA0,
|
|
|
1165 |
uint_fast64_t uiB64,
|
|
|
1166 |
uint_fast64_t uiB0
|
|
|
1167 |
)
|
|
|
1168 |
{
|
|
|
1169 |
bool isSigNaNA, isSigNaNB;
|
|
|
1170 |
uint_fast64_t uiNonsigA64, uiNonsigB64, uiMagA64, uiMagB64;
|
|
|
1171 |
struct uint128 uiZ;
|
|
|
1172 |
|
|
|
1173 |
/*------------------------------------------------------------------------
|
|
|
1174 |
*------------------------------------------------------------------------*/
|
|
|
1175 |
isSigNaNA = softfloat_isSigNaNF128UI( uiA64, uiA0 );
|
|
|
1176 |
isSigNaNB = softfloat_isSigNaNF128UI( uiB64, uiB0 );
|
|
|
1177 |
/*------------------------------------------------------------------------
|
|
|
1178 |
| Make NaNs non-signaling.
|
|
|
1179 |
*------------------------------------------------------------------------*/
|
|
|
1180 |
uiNonsigA64 = uiA64 | UINT64_C( 0x0000800000000000 );
|
|
|
1181 |
uiNonsigB64 = uiB64 | UINT64_C( 0x0000800000000000 );
|
|
|
1182 |
/*------------------------------------------------------------------------
|
|
|
1183 |
*------------------------------------------------------------------------*/
|
|
|
1184 |
if ( isSigNaNA | isSigNaNB ) {
|
|
|
1185 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
1186 |
if ( isSigNaNA ) {
|
|
|
1187 |
if ( isSigNaNB ) goto returnLargerMag;
|
|
|
1188 |
if ( isNaNF128UI( uiB64, uiB0 ) ) goto returnB;
|
|
|
1189 |
goto returnA;
|
|
|
1190 |
} else {
|
|
|
1191 |
if ( isNaNF128UI( uiA64, uiA0 ) ) goto returnA;
|
|
|
1192 |
goto returnB;
|
|
|
1193 |
}
|
|
|
1194 |
}
|
|
|
1195 |
returnLargerMag:
|
|
|
1196 |
uiMagA64 = uiA64 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
|
|
|
1197 |
uiMagB64 = uiB64 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
|
|
|
1198 |
if ( uiMagA64 < uiMagB64 ) goto returnB;
|
|
|
1199 |
if ( uiMagB64 < uiMagA64 ) goto returnA;
|
|
|
1200 |
if ( uiA0 < uiB0 ) goto returnB;
|
|
|
1201 |
if ( uiB0 < uiA0 ) goto returnA;
|
|
|
1202 |
if ( uiNonsigA64 < uiNonsigB64 ) goto returnA;
|
|
|
1203 |
returnB:
|
|
|
1204 |
uiZ.v64 = uiNonsigB64;
|
|
|
1205 |
uiZ.v0 = uiB0;
|
|
|
1206 |
return uiZ;
|
|
|
1207 |
returnA:
|
|
|
1208 |
uiZ.v64 = uiNonsigA64;
|
|
|
1209 |
uiZ.v0 = uiA0;
|
|
|
1210 |
return uiZ;
|
|
|
1211 |
|
|
|
1212 |
}
|
|
|
1213 |
|
|
5014
|
1214 |
#if __POINTER_SIZE__ == 4
|
|
|
1215 |
|
|
4992
|
1216 |
void
|
|
5014
|
1217 |
i32_to_f128M( int32_t a, float128_t *zPtr )
|
|
5011
|
1218 |
{
|
|
5014
|
1219 |
uint32_t *zWPtr;
|
|
|
1220 |
uint32_t uiZ96, uiZ64;
|
|
|
1221 |
bool sign;
|
|
|
1222 |
uint32_t absA;
|
|
|
1223 |
int_fast8_t shiftDist;
|
|
|
1224 |
uint64_t normAbsA;
|
|
|
1225 |
|
|
|
1226 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1227 |
uiZ96 = 0;
|
|
|
1228 |
uiZ64 = 0;
|
|
|
1229 |
if ( a ) {
|
|
|
1230 |
sign = (a < 0);
|
|
|
1231 |
absA = sign ? -(uint32_t) a : (uint32_t) a;
|
|
|
1232 |
shiftDist = softfloat_countLeadingZeros32( absA ) + 17;
|
|
|
1233 |
normAbsA = (uint64_t) absA<<shiftDist;
|
|
|
1234 |
uiZ96 = packToF128UI96( sign, 0x402E - shiftDist, normAbsA>>32 );
|
|
|
1235 |
uiZ64 = normAbsA;
|
|
|
1236 |
}
|
|
|
1237 |
zWPtr[indexWord( 4, 3 )] = uiZ96;
|
|
|
1238 |
zWPtr[indexWord( 4, 2 )] = uiZ64;
|
|
|
1239 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1240 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1241 |
}
|
|
|
1242 |
|
|
|
1243 |
#else
|
|
|
1244 |
|
|
|
1245 |
void
|
|
|
1246 |
i64_to_f128M( int64_t a, float128_t *zPtr )
|
|
|
1247 |
{
|
|
|
1248 |
uint32_t *zWPtr;
|
|
|
1249 |
uint32_t uiZ96, uiZ64;
|
|
|
1250 |
bool sign;
|
|
|
1251 |
uint64_t absA;
|
|
5011
|
1252 |
uint_fast8_t shiftDist;
|
|
|
1253 |
uint32_t *ptr;
|
|
|
1254 |
|
|
|
1255 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1256 |
uiZ96 = 0;
|
|
|
1257 |
uiZ64 = 0;
|
|
|
1258 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1259 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1260 |
if ( a ) {
|
|
5014
|
1261 |
sign = (a < 0);
|
|
|
1262 |
absA = sign ? -(uint64_t) a : (uint64_t) a;
|
|
|
1263 |
shiftDist = softfloat_countLeadingZeros64( absA ) + 17;
|
|
5011
|
1264 |
if ( shiftDist < 32 ) {
|
|
|
1265 |
ptr = zWPtr + indexMultiwordHi( 4, 3 );
|
|
|
1266 |
ptr[indexWord( 3, 2 )] = 0;
|
|
5014
|
1267 |
ptr[indexWord( 3, 1 )] = absA>>32;
|
|
|
1268 |
ptr[indexWord( 3, 0 )] = absA;
|
|
5011
|
1269 |
softfloat_shortShiftLeft96M( ptr, shiftDist, ptr );
|
|
|
1270 |
ptr[indexWordHi( 3 )] =
|
|
5014
|
1271 |
packToF128UI96(
|
|
|
1272 |
sign, 0x404E - shiftDist, ptr[indexWordHi( 3 )] );
|
|
5011
|
1273 |
return;
|
|
|
1274 |
}
|
|
5014
|
1275 |
absA <<= shiftDist - 32;
|
|
|
1276 |
uiZ96 = packToF128UI96( sign, 0x404E - shiftDist, absA>>32 );
|
|
|
1277 |
uiZ64 = absA;
|
|
5011
|
1278 |
}
|
|
|
1279 |
zWPtr[indexWord( 4, 3 )] = uiZ96;
|
|
|
1280 |
zWPtr[indexWord( 4, 2 )] = uiZ64;
|
|
|
1281 |
}
|
|
|
1282 |
|
|
5014
|
1283 |
#endif
|
|
|
1284 |
|
|
5011
|
1285 |
void
|
|
4992
|
1286 |
f64_to_f128M( float64_t a, float128_t *zPtr )
|
|
|
1287 |
{
|
|
|
1288 |
uint32_t *zWPtr;
|
|
|
1289 |
union ui64_f64 uA;
|
|
|
1290 |
uint64_t uiA;
|
|
|
1291 |
bool sign;
|
|
|
1292 |
int_fast16_t exp;
|
|
|
1293 |
uint64_t frac;
|
|
|
1294 |
struct commonNaN commonNaN;
|
|
|
1295 |
uint32_t uiZ96;
|
|
|
1296 |
struct exp16_sig64 normExpSig;
|
|
|
1297 |
|
|
|
1298 |
/*------------------------------------------------------------------------
|
|
|
1299 |
*------------------------------------------------------------------------*/
|
|
|
1300 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1301 |
/*------------------------------------------------------------------------
|
|
|
1302 |
*------------------------------------------------------------------------*/
|
|
|
1303 |
uA.f = a;
|
|
|
1304 |
uiA = uA.ui;
|
|
|
1305 |
sign = signF64UI( uiA );
|
|
|
1306 |
exp = expF64UI( uiA );
|
|
|
1307 |
frac = fracF64UI( uiA );
|
|
|
1308 |
/*------------------------------------------------------------------------
|
|
|
1309 |
*------------------------------------------------------------------------*/
|
|
|
1310 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1311 |
if ( exp == 0x7FF ) {
|
|
|
1312 |
if ( frac ) {
|
|
5014
|
1313 |
// NaN
|
|
|
1314 |
softfloat_commonNaNToF128M( zWPtr );
|
|
4992
|
1315 |
return;
|
|
|
1316 |
}
|
|
|
1317 |
uiZ96 = packToF128UI96( sign, 0x7FFF, 0 );
|
|
|
1318 |
goto uiZ;
|
|
|
1319 |
}
|
|
|
1320 |
/*------------------------------------------------------------------------
|
|
|
1321 |
*------------------------------------------------------------------------*/
|
|
|
1322 |
if ( ! exp ) {
|
|
|
1323 |
if ( ! frac ) {
|
|
|
1324 |
uiZ96 = packToF128UI96( sign, 0, 0 );
|
|
|
1325 |
goto uiZ;
|
|
|
1326 |
}
|
|
|
1327 |
normExpSig = softfloat_normSubnormalF64Sig( frac );
|
|
|
1328 |
exp = normExpSig.exp - 1;
|
|
|
1329 |
frac = normExpSig.sig;
|
|
|
1330 |
}
|
|
|
1331 |
/*------------------------------------------------------------------------
|
|
|
1332 |
*------------------------------------------------------------------------*/
|
|
|
1333 |
zWPtr[indexWord( 4, 1 )] = (uint32_t) frac<<28;
|
|
|
1334 |
frac >>= 4;
|
|
|
1335 |
zWPtr[indexWordHi( 4 )] = packToF128UI96( sign, exp + 0x3C00, frac>>32 );
|
|
|
1336 |
zWPtr[indexWord( 4, 2 )] = frac;
|
|
|
1337 |
return;
|
|
|
1338 |
/*------------------------------------------------------------------------
|
|
|
1339 |
*------------------------------------------------------------------------*/
|
|
|
1340 |
uiZ:
|
|
|
1341 |
zWPtr[indexWord( 4, 3 )] = uiZ96;
|
|
|
1342 |
zWPtr[indexWord( 4, 2 )] = 0;
|
|
|
1343 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1344 |
|
|
|
1345 |
}
|
|
|
1346 |
|
|
5014
|
1347 |
float128_t
|
|
|
1348 |
extF80_to_f128( extFloat80_t a )
|
|
|
1349 |
{
|
|
|
1350 |
union { struct extFloat80M s; extFloat80_t f; } uA;
|
|
|
1351 |
uint_fast16_t uiA64;
|
|
|
1352 |
uint_fast64_t uiA0;
|
|
|
1353 |
uint_fast16_t exp;
|
|
|
1354 |
uint_fast64_t frac;
|
|
|
1355 |
struct commonNaN commonNaN;
|
|
|
1356 |
struct uint128 uiZ;
|
|
|
1357 |
bool sign;
|
|
|
1358 |
struct uint128 frac128;
|
|
|
1359 |
union ui128_f128 uZ;
|
|
|
1360 |
|
|
|
1361 |
uA.f = a;
|
|
|
1362 |
uiA64 = uA.s.signExp;
|
|
|
1363 |
uiA0 = uA.s.signif;
|
|
|
1364 |
exp = expExtF80UI64( uiA64 );
|
|
|
1365 |
frac = uiA0 & UINT64_C( 0x7FFFFFFFFFFFFFFF );
|
|
|
1366 |
if ( (exp == 0x7FFF) && frac ) {
|
|
|
1367 |
softfloat_extF80UIToCommonNaN( uiA64, uiA0, &commonNaN );
|
|
|
1368 |
softfloat_commonNaNToF128M( (uint32_t*) &uiZ );
|
|
|
1369 |
} else {
|
|
|
1370 |
sign = signExtF80UI64( uiA64 );
|
|
|
1371 |
frac128 = softfloat_shortShiftLeft128( 0, frac, 49 );
|
|
|
1372 |
uiZ.v64 = packToF128UI64( sign, exp, frac128.v64 );
|
|
|
1373 |
uiZ.v0 = frac128.v0;
|
|
|
1374 |
}
|
|
|
1375 |
uZ.ui = uiZ;
|
|
|
1376 |
return uZ.f;
|
|
|
1377 |
|
|
|
1378 |
}
|
|
|
1379 |
|
|
4992
|
1380 |
static float128_t
|
|
|
1381 |
softfloat_addMagsF128(
|
|
|
1382 |
uint64_t uiA64,
|
|
|
1383 |
uint64_t uiA0,
|
|
|
1384 |
uint64_t uiB64,
|
|
|
1385 |
uint64_t uiB0,
|
|
|
1386 |
bool signZ
|
|
|
1387 |
)
|
|
|
1388 |
{
|
|
|
1389 |
int32_t expA;
|
|
|
1390 |
struct uint128 sigA;
|
|
|
1391 |
int32_t expB;
|
|
|
1392 |
struct uint128 sigB;
|
|
|
1393 |
int32_t expDiff;
|
|
|
1394 |
struct uint128 uiZ, sigZ;
|
|
|
1395 |
int32_t expZ;
|
|
|
1396 |
uint64_t sigZExtra;
|
|
|
1397 |
struct uint128_extra sig128Extra;
|
|
|
1398 |
union ui128_f128 uZ;
|
|
|
1399 |
|
|
|
1400 |
expA = expF128UI64( uiA64 );
|
|
|
1401 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1402 |
sigA.v0 = uiA0;
|
|
|
1403 |
expB = expF128UI64( uiB64 );
|
|
|
1404 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1405 |
sigB.v0 = uiB0;
|
|
|
1406 |
expDiff = expA - expB;
|
|
|
1407 |
if ( ! expDiff ) {
|
|
|
1408 |
if ( expA == 0x7FFF ) {
|
|
|
1409 |
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1410 |
uiZ.v64 = uiA64;
|
|
|
1411 |
uiZ.v0 = uiA0;
|
|
|
1412 |
goto uiZ;
|
|
|
1413 |
}
|
|
|
1414 |
sigZ = softfloat_add128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
|
|
|
1415 |
if ( ! expA ) {
|
|
|
1416 |
uiZ.v64 = packToF128UI64( signZ, 0, sigZ.v64 );
|
|
|
1417 |
uiZ.v0 = sigZ.v0;
|
|
|
1418 |
goto uiZ;
|
|
|
1419 |
}
|
|
|
1420 |
expZ = expA;
|
|
|
1421 |
sigZ.v64 |= UINT64_C( 0x0002000000000000 );
|
|
|
1422 |
sigZExtra = 0;
|
|
|
1423 |
goto shiftRight1;
|
|
|
1424 |
}
|
|
|
1425 |
if ( expDiff < 0 ) {
|
|
|
1426 |
if ( expB == 0x7FFF ) {
|
|
|
1427 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1428 |
uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
|
|
|
1429 |
uiZ.v0 = 0;
|
|
|
1430 |
goto uiZ;
|
|
|
1431 |
}
|
|
|
1432 |
expZ = expB;
|
|
|
1433 |
if ( expA ) {
|
|
|
1434 |
sigA.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1435 |
} else {
|
|
|
1436 |
++expDiff;
|
|
|
1437 |
sigZExtra = 0;
|
|
|
1438 |
if ( ! expDiff ) goto newlyAligned;
|
|
|
1439 |
}
|
|
|
1440 |
sig128Extra =
|
|
|
1441 |
softfloat_shiftRightJam128Extra( sigA.v64, sigA.v0, 0, -expDiff );
|
|
|
1442 |
sigA = sig128Extra.v;
|
|
|
1443 |
sigZExtra = sig128Extra.extra;
|
|
|
1444 |
} else {
|
|
|
1445 |
if ( expA == 0x7FFF ) {
|
|
|
1446 |
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
|
|
|
1447 |
uiZ.v64 = uiA64;
|
|
|
1448 |
uiZ.v0 = uiA0;
|
|
|
1449 |
goto uiZ;
|
|
|
1450 |
}
|
|
|
1451 |
expZ = expA;
|
|
|
1452 |
if ( expB ) {
|
|
|
1453 |
sigB.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1454 |
} else {
|
|
|
1455 |
--expDiff;
|
|
|
1456 |
sigZExtra = 0;
|
|
|
1457 |
if ( ! expDiff ) goto newlyAligned;
|
|
|
1458 |
}
|
|
|
1459 |
sig128Extra =
|
|
|
1460 |
softfloat_shiftRightJam128Extra( sigB.v64, sigB.v0, 0, expDiff );
|
|
|
1461 |
sigB = sig128Extra.v;
|
|
|
1462 |
sigZExtra = sig128Extra.extra;
|
|
|
1463 |
}
|
|
|
1464 |
newlyAligned:
|
|
|
1465 |
sigZ =
|
|
|
1466 |
softfloat_add128(
|
|
|
1467 |
sigA.v64 | UINT64_C( 0x0001000000000000 ),
|
|
|
1468 |
sigA.v0,
|
|
|
1469 |
sigB.v64,
|
|
|
1470 |
sigB.v0
|
|
|
1471 |
);
|
|
|
1472 |
--expZ;
|
|
|
1473 |
if ( sigZ.v64 < UINT64_C( 0x0002000000000000 ) ) goto roundAndPack;
|
|
|
1474 |
++expZ;
|
|
|
1475 |
shiftRight1:
|
|
|
1476 |
sig128Extra =
|
|
|
1477 |
softfloat_shortShiftRightJam128Extra(
|
|
|
1478 |
sigZ.v64, sigZ.v0, sigZExtra, 1 );
|
|
|
1479 |
sigZ = sig128Extra.v;
|
|
|
1480 |
sigZExtra = sig128Extra.extra;
|
|
|
1481 |
roundAndPack:
|
|
|
1482 |
return
|
|
|
1483 |
softfloat_roundPackToF128( signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
|
|
|
1484 |
propagateNaN:
|
|
|
1485 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
1486 |
uiZ:
|
|
|
1487 |
uZ.ui = uiZ;
|
|
|
1488 |
return uZ.f;
|
|
|
1489 |
|
|
|
1490 |
}
|
|
|
1491 |
|
|
|
1492 |
float128_t
|
|
|
1493 |
softfloat_subMagsF128(
|
|
|
1494 |
uint_fast64_t uiA64,
|
|
|
1495 |
uint_fast64_t uiA0,
|
|
|
1496 |
uint_fast64_t uiB64,
|
|
|
1497 |
uint_fast64_t uiB0,
|
|
|
1498 |
bool signZ
|
|
|
1499 |
)
|
|
|
1500 |
{
|
|
|
1501 |
int_fast32_t expA;
|
|
|
1502 |
struct uint128 sigA;
|
|
|
1503 |
int_fast32_t expB;
|
|
|
1504 |
struct uint128 sigB, sigZ;
|
|
|
1505 |
int_fast32_t expDiff, expZ;
|
|
|
1506 |
struct uint128 uiZ;
|
|
|
1507 |
union ui128_f128 uZ;
|
|
|
1508 |
|
|
|
1509 |
expA = expF128UI64( uiA64 );
|
|
|
1510 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1511 |
sigA.v0 = uiA0;
|
|
|
1512 |
expB = expF128UI64( uiB64 );
|
|
|
1513 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1514 |
sigB.v0 = uiB0;
|
|
|
1515 |
sigA = softfloat_shortShiftLeft128( sigA.v64, sigA.v0, 4 );
|
|
|
1516 |
sigB = softfloat_shortShiftLeft128( sigB.v64, sigB.v0, 4 );
|
|
|
1517 |
expDiff = expA - expB;
|
|
|
1518 |
if ( 0 < expDiff ) goto expABigger;
|
|
|
1519 |
if ( expDiff < 0 ) goto expBBigger;
|
|
|
1520 |
if ( expA == 0x7FFF ) {
|
|
|
1521 |
if ( sigA.v64 | sigA.v0 | sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1522 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
5014
|
1523 |
softfloat_commonNaNToF128M( (uint32_t*)&uiZ );
|
|
|
1524 |
// uiZ.v64 = defaultNaNF128UI64;
|
|
|
1525 |
// uiZ.v0 = defaultNaNF128UI0;
|
|
4992
|
1526 |
goto uiZ;
|
|
|
1527 |
}
|
|
|
1528 |
expZ = expA;
|
|
|
1529 |
if ( ! expZ ) expZ = 1;
|
|
|
1530 |
if ( sigB.v64 < sigA.v64 ) goto aBigger;
|
|
|
1531 |
if ( sigA.v64 < sigB.v64 ) goto bBigger;
|
|
|
1532 |
if ( sigB.v0 < sigA.v0 ) goto aBigger;
|
|
|
1533 |
if ( sigA.v0 < sigB.v0 ) goto bBigger;
|
|
|
1534 |
uiZ.v64 =
|
|
|
1535 |
packToF128UI64(
|
|
|
1536 |
(softfloat_roundingMode == softfloat_round_min), 0, 0 );
|
|
|
1537 |
uiZ.v0 = 0;
|
|
|
1538 |
goto uiZ;
|
|
|
1539 |
expBBigger:
|
|
|
1540 |
if ( expB == 0x7FFF ) {
|
|
|
1541 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1542 |
uiZ.v64 = packToF128UI64( signZ ^ 1, 0x7FFF, 0 );
|
|
|
1543 |
uiZ.v0 = 0;
|
|
|
1544 |
goto uiZ;
|
|
|
1545 |
}
|
|
|
1546 |
if ( expA ) {
|
|
|
1547 |
sigA.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1548 |
} else {
|
|
|
1549 |
++expDiff;
|
|
|
1550 |
if ( ! expDiff ) goto newlyAlignedBBigger;
|
|
|
1551 |
}
|
|
|
1552 |
sigA = softfloat_shiftRightJam128( sigA.v64, sigA.v0, -expDiff );
|
|
|
1553 |
newlyAlignedBBigger:
|
|
|
1554 |
expZ = expB;
|
|
|
1555 |
sigB.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1556 |
bBigger:
|
|
|
1557 |
signZ = ! signZ;
|
|
|
1558 |
sigZ = softfloat_sub128( sigB.v64, sigB.v0, sigA.v64, sigA.v0 );
|
|
|
1559 |
goto normRoundPack;
|
|
|
1560 |
expABigger:
|
|
|
1561 |
if ( expA == 0x7FFF ) {
|
|
|
1562 |
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
|
|
|
1563 |
uiZ.v64 = uiA64;
|
|
|
1564 |
uiZ.v0 = uiA0;
|
|
|
1565 |
goto uiZ;
|
|
|
1566 |
}
|
|
|
1567 |
if ( expB ) {
|
|
|
1568 |
sigB.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1569 |
} else {
|
|
|
1570 |
--expDiff;
|
|
|
1571 |
if ( ! expDiff ) goto newlyAlignedABigger;
|
|
|
1572 |
}
|
|
|
1573 |
sigB = softfloat_shiftRightJam128( sigB.v64, sigB.v0, expDiff );
|
|
|
1574 |
newlyAlignedABigger:
|
|
|
1575 |
expZ = expA;
|
|
|
1576 |
sigA.v64 |= UINT64_C( 0x0010000000000000 );
|
|
|
1577 |
aBigger:
|
|
|
1578 |
sigZ = softfloat_sub128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 );
|
|
|
1579 |
normRoundPack:
|
|
|
1580 |
return softfloat_normRoundPackToF128( signZ, expZ - 5, sigZ.v64, sigZ.v0 );
|
|
|
1581 |
propagateNaN:
|
|
|
1582 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
1583 |
uiZ:
|
|
|
1584 |
uZ.ui = uiZ;
|
|
|
1585 |
return uZ.f;
|
|
|
1586 |
}
|
|
|
1587 |
|
|
|
1588 |
void
|
|
|
1589 |
f128M_add( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1590 |
{
|
|
|
1591 |
const uint64_t *aWPtr, *bWPtr;
|
|
|
1592 |
uint64_t uiA64, uiA0;
|
|
|
1593 |
int signA;
|
|
|
1594 |
uint64_t uiB64, uiB0;
|
|
|
1595 |
int signB;
|
|
|
1596 |
|
|
|
1597 |
aWPtr = (const uint64_t *) aPtr;
|
|
|
1598 |
bWPtr = (const uint64_t *) bPtr;
|
|
|
1599 |
uiA64 = aWPtr[indexWord( 2, 1 )];
|
|
|
1600 |
uiA0 = aWPtr[indexWord( 2, 0 )];
|
|
|
1601 |
signA = signF128UI64( uiA64 );
|
|
|
1602 |
uiB64 = bWPtr[indexWord( 2, 1 )];
|
|
|
1603 |
uiB0 = bWPtr[indexWord( 2, 0 )];
|
|
|
1604 |
signB = signF128UI64( uiB64 );
|
|
|
1605 |
|
|
|
1606 |
if ( signA == signB ) {
|
|
|
1607 |
*zPtr = softfloat_addMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1608 |
} else {
|
|
|
1609 |
*zPtr = softfloat_subMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1610 |
}
|
|
4999
|
1611 |
}
|
|
4992
|
1612 |
|
|
4999
|
1613 |
void
|
|
|
1614 |
f128M_sub( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1615 |
{
|
|
|
1616 |
const uint64_t *aWPtr, *bWPtr;
|
|
|
1617 |
uint_fast64_t uiA64, uiA0;
|
|
|
1618 |
bool signA;
|
|
|
1619 |
uint_fast64_t uiB64, uiB0;
|
|
|
1620 |
bool signB;
|
|
|
1621 |
|
|
|
1622 |
aWPtr = (const uint64_t *) aPtr;
|
|
|
1623 |
bWPtr = (const uint64_t *) bPtr;
|
|
|
1624 |
uiA64 = aWPtr[indexWord( 2, 1 )];
|
|
|
1625 |
uiA0 = aWPtr[indexWord( 2, 0 )];
|
|
|
1626 |
signA = signF128UI64( uiA64 );
|
|
|
1627 |
uiB64 = bWPtr[indexWord( 2, 1 )];
|
|
|
1628 |
uiB0 = bWPtr[indexWord( 2, 0 )];
|
|
|
1629 |
signB = signF128UI64( uiB64 );
|
|
|
1630 |
if ( signA == signB ) {
|
|
|
1631 |
*zPtr = softfloat_subMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1632 |
} else {
|
|
|
1633 |
*zPtr = softfloat_addMagsF128( uiA64, uiA0, uiB64, uiB0, signA );
|
|
|
1634 |
}
|
|
4992
|
1635 |
}
|
|
4999
|
1636 |
|
|
5000
|
1637 |
void
|
|
|
1638 |
f128M_mul( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
1639 |
{
|
|
|
1640 |
const uint32_t *aWPtr, *bWPtr;
|
|
|
1641 |
uint32_t *zWPtr;
|
|
|
1642 |
uint32_t uiA96;
|
|
|
1643 |
int32_t expA;
|
|
|
1644 |
uint32_t uiB96;
|
|
|
1645 |
int32_t expB;
|
|
|
1646 |
bool signZ;
|
|
|
1647 |
const uint32_t *ptr;
|
|
|
1648 |
uint32_t uiZ96, sigA[4];
|
|
|
1649 |
uint_fast8_t shiftDist;
|
|
|
1650 |
uint32_t sigB[4];
|
|
|
1651 |
int32_t expZ;
|
|
|
1652 |
uint32_t sigProd[8], *extSigZPtr;
|
|
|
1653 |
|
|
|
1654 |
/*------------------------------------------------------------------------
|
|
|
1655 |
*------------------------------------------------------------------------*/
|
|
|
1656 |
aWPtr = (const uint32_t *) aPtr;
|
|
|
1657 |
bWPtr = (const uint32_t *) bPtr;
|
|
|
1658 |
zWPtr = (uint32_t *) zPtr;
|
|
|
1659 |
/*------------------------------------------------------------------------
|
|
|
1660 |
*------------------------------------------------------------------------*/
|
|
|
1661 |
uiA96 = aWPtr[indexWordHi( 4 )];
|
|
|
1662 |
expA = expF128UI96( uiA96 );
|
|
|
1663 |
uiB96 = bWPtr[indexWordHi( 4 )];
|
|
|
1664 |
expB = expF128UI96( uiB96 );
|
|
|
1665 |
signZ = signF128UI96( uiA96 ) ^ signF128UI96( uiB96 );
|
|
|
1666 |
/*------------------------------------------------------------------------
|
|
|
1667 |
*------------------------------------------------------------------------*/
|
|
|
1668 |
if ( (expA == 0x7FFF) || (expB == 0x7FFF) ) {
|
|
|
1669 |
if ( softfloat_tryPropagateNaNF128M( aWPtr, bWPtr, zWPtr ) ) return;
|
|
|
1670 |
ptr = aWPtr;
|
|
|
1671 |
if ( ! expA ) goto possiblyInvalid;
|
|
|
1672 |
if ( ! expB ) {
|
|
|
1673 |
ptr = bWPtr;
|
|
|
1674 |
possiblyInvalid:
|
|
|
1675 |
if (
|
|
|
1676 |
! fracF128UI96( ptr[indexWordHi( 4 )] )
|
|
|
1677 |
&& ! (ptr[indexWord( 4, 2 )] | ptr[indexWord( 4, 1 )]
|
|
|
1678 |
| ptr[indexWord( 4, 0 )])
|
|
|
1679 |
) {
|
|
|
1680 |
softfloat_invalidF128M( zWPtr );
|
|
|
1681 |
return;
|
|
|
1682 |
}
|
|
|
1683 |
}
|
|
|
1684 |
uiZ96 = packToF128UI96( signZ, 0x7FFF, 0 );
|
|
|
1685 |
goto uiZ96;
|
|
|
1686 |
}
|
|
|
1687 |
/*------------------------------------------------------------------------
|
|
|
1688 |
*------------------------------------------------------------------------*/
|
|
|
1689 |
if ( expA ) {
|
|
|
1690 |
sigA[indexWordHi( 4 )] = fracF128UI96( uiA96 ) | 0x00010000;
|
|
|
1691 |
sigA[indexWord( 4, 2 )] = aWPtr[indexWord( 4, 2 )];
|
|
|
1692 |
sigA[indexWord( 4, 1 )] = aWPtr[indexWord( 4, 1 )];
|
|
|
1693 |
sigA[indexWord( 4, 0 )] = aWPtr[indexWord( 4, 0 )];
|
|
|
1694 |
} else {
|
|
|
1695 |
expA = softfloat_shiftNormSigF128M( aWPtr, 0, sigA );
|
|
|
1696 |
if ( expA == -128 ) goto zero;
|
|
|
1697 |
}
|
|
|
1698 |
if ( expB ) {
|
|
|
1699 |
sigB[indexWordHi( 4 )] = fracF128UI96( uiB96 ) | 0x00010000;
|
|
|
1700 |
sigB[indexWord( 4, 2 )] = bWPtr[indexWord( 4, 2 )];
|
|
|
1701 |
sigB[indexWord( 4, 1 )] = bWPtr[indexWord( 4, 1 )];
|
|
|
1702 |
sigB[indexWord( 4, 0 )] = bWPtr[indexWord( 4, 0 )];
|
|
|
1703 |
} else {
|
|
|
1704 |
expB = softfloat_shiftNormSigF128M( bWPtr, 0, sigB );
|
|
|
1705 |
if ( expB == -128 ) goto zero;
|
|
|
1706 |
}
|
|
|
1707 |
/*------------------------------------------------------------------------
|
|
|
1708 |
*------------------------------------------------------------------------*/
|
|
|
1709 |
expZ = expA + expB - 0x4000;
|
|
|
1710 |
softfloat_mul128MTo256M( sigA, sigB, sigProd );
|
|
|
1711 |
if (
|
|
|
1712 |
sigProd[indexWord( 8, 2 )]
|
|
|
1713 |
|| (sigProd[indexWord( 8, 1 )] | sigProd[indexWord( 8, 0 )])
|
|
|
1714 |
) {
|
|
|
1715 |
sigProd[indexWord( 8, 3 )] |= 1;
|
|
|
1716 |
}
|
|
|
1717 |
extSigZPtr = &sigProd[indexMultiwordHi( 8, 5 )];
|
|
|
1718 |
shiftDist = 16;
|
|
|
1719 |
if ( extSigZPtr[indexWordHi( 5 )] & 2 ) {
|
|
|
1720 |
++expZ;
|
|
|
1721 |
shiftDist = 15;
|
|
|
1722 |
}
|
|
|
1723 |
softfloat_shortShiftLeft160M( extSigZPtr, shiftDist, extSigZPtr );
|
|
|
1724 |
softfloat_roundPackMToF128M( signZ, expZ, extSigZPtr, zWPtr );
|
|
|
1725 |
return;
|
|
|
1726 |
/*------------------------------------------------------------------------
|
|
|
1727 |
*------------------------------------------------------------------------*/
|
|
|
1728 |
zero:
|
|
|
1729 |
uiZ96 = packToF128UI96( signZ, 0, 0 );
|
|
|
1730 |
uiZ96:
|
|
|
1731 |
zWPtr[indexWordHi( 4 )] = uiZ96;
|
|
|
1732 |
zWPtr[indexWord( 4, 2 )] = 0;
|
|
|
1733 |
zWPtr[indexWord( 4, 1 )] = 0;
|
|
|
1734 |
zWPtr[indexWord( 4, 0 )] = 0;
|
|
|
1735 |
}
|
|
|
1736 |
|
|
5004
|
1737 |
float128_t
|
|
|
1738 |
f128_div( float128_t a, float128_t b )
|
|
|
1739 |
{
|
|
|
1740 |
union ui128_f128 uA;
|
|
|
1741 |
uint_fast64_t uiA64, uiA0;
|
|
|
1742 |
bool signA;
|
|
|
1743 |
int_fast32_t expA;
|
|
|
1744 |
struct uint128 sigA;
|
|
|
1745 |
union ui128_f128 uB;
|
|
|
1746 |
uint_fast64_t uiB64, uiB0;
|
|
|
1747 |
bool signB;
|
|
|
1748 |
int_fast32_t expB;
|
|
|
1749 |
struct uint128 sigB;
|
|
|
1750 |
bool signZ;
|
|
|
1751 |
struct exp32_sig128 normExpSig;
|
|
|
1752 |
int_fast32_t expZ;
|
|
|
1753 |
struct uint128 rem;
|
|
|
1754 |
uint_fast32_t recip32;
|
|
|
1755 |
int ix;
|
|
|
1756 |
uint_fast64_t q64;
|
|
|
1757 |
uint_fast32_t q;
|
|
|
1758 |
struct uint128 term;
|
|
|
1759 |
uint_fast32_t qs[3];
|
|
|
1760 |
uint_fast64_t sigZExtra;
|
|
|
1761 |
struct uint128 sigZ, uiZ;
|
|
|
1762 |
union ui128_f128 uZ;
|
|
|
1763 |
|
|
|
1764 |
/*------------------------------------------------------------------------
|
|
|
1765 |
*------------------------------------------------------------------------*/
|
|
|
1766 |
uA.f = a;
|
|
|
1767 |
uiA64 = uA.ui.v64;
|
|
|
1768 |
uiA0 = uA.ui.v0;
|
|
|
1769 |
signA = signF128UI64( uiA64 );
|
|
|
1770 |
expA = expF128UI64( uiA64 );
|
|
|
1771 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1772 |
sigA.v0 = uiA0;
|
|
|
1773 |
uB.f = b;
|
|
|
1774 |
uiB64 = uB.ui.v64;
|
|
|
1775 |
uiB0 = uB.ui.v0;
|
|
|
1776 |
signB = signF128UI64( uiB64 );
|
|
|
1777 |
expB = expF128UI64( uiB64 );
|
|
|
1778 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1779 |
sigB.v0 = uiB0;
|
|
|
1780 |
signZ = signA ^ signB;
|
|
|
1781 |
/*------------------------------------------------------------------------
|
|
|
1782 |
*------------------------------------------------------------------------*/
|
|
|
1783 |
if ( expA == 0x7FFF ) {
|
|
|
1784 |
if ( sigA.v64 | sigA.v0 ) goto propagateNaN;
|
|
|
1785 |
if ( expB == 0x7FFF ) {
|
|
|
1786 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1787 |
goto invalid;
|
|
|
1788 |
}
|
|
|
1789 |
goto infinity;
|
|
|
1790 |
}
|
|
|
1791 |
if ( expB == 0x7FFF ) {
|
|
|
1792 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1793 |
goto zero;
|
|
|
1794 |
}
|
|
|
1795 |
/*------------------------------------------------------------------------
|
|
|
1796 |
*------------------------------------------------------------------------*/
|
|
|
1797 |
if ( ! expB ) {
|
|
|
1798 |
if ( ! (sigB.v64 | sigB.v0) ) {
|
|
|
1799 |
if ( ! (expA | sigA.v64 | sigA.v0) ) goto invalid;
|
|
|
1800 |
softfloat_raiseFlags( softfloat_flag_infinite );
|
|
|
1801 |
goto infinity;
|
|
|
1802 |
}
|
|
|
1803 |
normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
|
|
|
1804 |
expB = normExpSig.exp;
|
|
|
1805 |
sigB = normExpSig.sig;
|
|
|
1806 |
}
|
|
|
1807 |
if ( ! expA ) {
|
|
|
1808 |
if ( ! (sigA.v64 | sigA.v0) ) goto zero;
|
|
|
1809 |
normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
|
|
|
1810 |
expA = normExpSig.exp;
|
|
|
1811 |
sigA = normExpSig.sig;
|
|
|
1812 |
}
|
|
|
1813 |
/*------------------------------------------------------------------------
|
|
|
1814 |
*------------------------------------------------------------------------*/
|
|
|
1815 |
expZ = expA - expB + 0x3FFE;
|
|
|
1816 |
sigA.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1817 |
sigB.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1818 |
rem = sigA;
|
|
|
1819 |
if ( softfloat_lt128( sigA.v64, sigA.v0, sigB.v64, sigB.v0 ) ) {
|
|
|
1820 |
--expZ;
|
|
|
1821 |
rem = softfloat_add128( sigA.v64, sigA.v0, sigA.v64, sigA.v0 );
|
|
|
1822 |
}
|
|
|
1823 |
recip32 = softfloat_approxRecip32_1( sigB.v64>>17 );
|
|
|
1824 |
ix = 3;
|
|
|
1825 |
for (;;) {
|
|
|
1826 |
q64 = (uint_fast64_t) (uint32_t) (rem.v64>>19) * recip32;
|
|
|
1827 |
q = (q64 + 0x80000000)>>32;
|
|
|
1828 |
--ix;
|
|
|
1829 |
if ( ix < 0 ) break;
|
|
|
1830 |
rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
|
|
|
1831 |
term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
|
|
|
1832 |
rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
|
|
|
1833 |
if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
|
|
1834 |
--q;
|
|
|
1835 |
rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
1836 |
}
|
|
|
1837 |
qs[ix] = q;
|
|
|
1838 |
}
|
|
|
1839 |
/*------------------------------------------------------------------------
|
|
|
1840 |
*------------------------------------------------------------------------*/
|
|
|
1841 |
if ( ((q + 1) & 7) < 2 ) {
|
|
|
1842 |
rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
|
|
|
1843 |
term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
|
|
|
1844 |
rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
|
|
|
1845 |
if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
|
|
1846 |
--q;
|
|
|
1847 |
rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
1848 |
} else if ( softfloat_le128( sigB.v64, sigB.v0, rem.v64, rem.v0 ) ) {
|
|
|
1849 |
++q;
|
|
|
1850 |
rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
1851 |
}
|
|
|
1852 |
if ( rem.v64 | rem.v0 ) q |= 1;
|
|
|
1853 |
}
|
|
|
1854 |
/*------------------------------------------------------------------------
|
|
|
1855 |
*------------------------------------------------------------------------*/
|
|
|
1856 |
sigZExtra = (uint64_t) ((uint_fast64_t) q<<60);
|
|
|
1857 |
term = softfloat_shortShiftLeft128( 0, qs[1], 54 );
|
|
|
1858 |
sigZ =
|
|
|
1859 |
softfloat_add128(
|
|
|
1860 |
(uint_fast64_t) qs[2]<<19, ((uint_fast64_t) qs[0]<<25) + (q>>4),
|
|
|
1861 |
term.v64, term.v0
|
|
|
1862 |
);
|
|
|
1863 |
return
|
|
|
1864 |
softfloat_roundPackToF128( signZ, expZ, sigZ.v64, sigZ.v0, sigZExtra );
|
|
|
1865 |
/*------------------------------------------------------------------------
|
|
|
1866 |
*------------------------------------------------------------------------*/
|
|
|
1867 |
propagateNaN:
|
|
|
1868 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
1869 |
goto uiZ;
|
|
|
1870 |
/*------------------------------------------------------------------------
|
|
|
1871 |
*------------------------------------------------------------------------*/
|
|
|
1872 |
invalid:
|
|
|
1873 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
5014
|
1874 |
softfloat_commonNaNToF128M( (uint32_t*)&uiZ );
|
|
|
1875 |
// uiZ.v64 = defaultNaNF128UI64;
|
|
|
1876 |
// uiZ.v0 = defaultNaNF128UI0;
|
|
5004
|
1877 |
goto uiZ;
|
|
|
1878 |
/*------------------------------------------------------------------------
|
|
|
1879 |
*------------------------------------------------------------------------*/
|
|
|
1880 |
infinity:
|
|
|
1881 |
uiZ.v64 = packToF128UI64( signZ, 0x7FFF, 0 );
|
|
|
1882 |
goto uiZ0;
|
|
|
1883 |
/*------------------------------------------------------------------------
|
|
|
1884 |
*------------------------------------------------------------------------*/
|
|
|
1885 |
zero:
|
|
|
1886 |
uiZ.v64 = packToF128UI64( signZ, 0, 0 );
|
|
|
1887 |
uiZ0:
|
|
|
1888 |
uiZ.v0 = 0;
|
|
|
1889 |
uiZ:
|
|
|
1890 |
uZ.ui = uiZ;
|
|
|
1891 |
return uZ.f;
|
|
|
1892 |
}
|
|
|
1893 |
|
|
|
1894 |
static float128_t
|
|
|
1895 |
f128_rem( float128_t a, float128_t b )
|
|
|
1896 |
{
|
|
|
1897 |
union ui128_f128 uA;
|
|
|
1898 |
uint_fast64_t uiA64, uiA0;
|
|
|
1899 |
bool signA;
|
|
|
1900 |
int_fast32_t expA;
|
|
|
1901 |
struct uint128 sigA;
|
|
|
1902 |
union ui128_f128 uB;
|
|
|
1903 |
uint_fast64_t uiB64, uiB0;
|
|
|
1904 |
int_fast32_t expB;
|
|
|
1905 |
struct uint128 sigB;
|
|
|
1906 |
struct exp32_sig128 normExpSig;
|
|
|
1907 |
struct uint128 rem;
|
|
|
1908 |
int_fast32_t expDiff;
|
|
|
1909 |
uint_fast32_t q, recip32;
|
|
|
1910 |
uint_fast64_t q64;
|
|
|
1911 |
struct uint128 term, altRem, meanRem;
|
|
|
1912 |
bool signRem;
|
|
|
1913 |
struct uint128 uiZ;
|
|
|
1914 |
union ui128_f128 uZ;
|
|
|
1915 |
|
|
|
1916 |
/*------------------------------------------------------------------------
|
|
|
1917 |
*------------------------------------------------------------------------*/
|
|
|
1918 |
uA.f = a;
|
|
|
1919 |
uiA64 = uA.ui.v64;
|
|
|
1920 |
uiA0 = uA.ui.v0;
|
|
|
1921 |
signA = signF128UI64( uiA64 );
|
|
|
1922 |
expA = expF128UI64( uiA64 );
|
|
|
1923 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
1924 |
sigA.v0 = uiA0;
|
|
|
1925 |
uB.f = b;
|
|
|
1926 |
uiB64 = uB.ui.v64;
|
|
|
1927 |
uiB0 = uB.ui.v0;
|
|
|
1928 |
expB = expF128UI64( uiB64 );
|
|
|
1929 |
sigB.v64 = fracF128UI64( uiB64 );
|
|
|
1930 |
sigB.v0 = uiB0;
|
|
|
1931 |
/*------------------------------------------------------------------------
|
|
|
1932 |
*------------------------------------------------------------------------*/
|
|
|
1933 |
if ( expA == 0x7FFF ) {
|
|
|
1934 |
if (
|
|
|
1935 |
(sigA.v64 | sigA.v0) || ((expB == 0x7FFF) && (sigB.v64 | sigB.v0))
|
|
|
1936 |
) {
|
|
|
1937 |
goto propagateNaN;
|
|
|
1938 |
}
|
|
|
1939 |
goto invalid;
|
|
|
1940 |
}
|
|
|
1941 |
if ( expB == 0x7FFF ) {
|
|
|
1942 |
if ( sigB.v64 | sigB.v0 ) goto propagateNaN;
|
|
|
1943 |
return a;
|
|
|
1944 |
}
|
|
|
1945 |
/*------------------------------------------------------------------------
|
|
|
1946 |
*------------------------------------------------------------------------*/
|
|
|
1947 |
if ( ! expB ) {
|
|
|
1948 |
if ( ! (sigB.v64 | sigB.v0) ) goto invalid;
|
|
|
1949 |
normExpSig = softfloat_normSubnormalF128Sig( sigB.v64, sigB.v0 );
|
|
|
1950 |
expB = normExpSig.exp;
|
|
|
1951 |
sigB = normExpSig.sig;
|
|
|
1952 |
}
|
|
|
1953 |
if ( ! expA ) {
|
|
|
1954 |
if ( ! (sigA.v64 | sigA.v0) ) return a;
|
|
|
1955 |
normExpSig = softfloat_normSubnormalF128Sig( sigA.v64, sigA.v0 );
|
|
|
1956 |
expA = normExpSig.exp;
|
|
|
1957 |
sigA = normExpSig.sig;
|
|
|
1958 |
}
|
|
|
1959 |
/*------------------------------------------------------------------------
|
|
|
1960 |
*------------------------------------------------------------------------*/
|
|
|
1961 |
sigA.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1962 |
sigB.v64 |= UINT64_C( 0x0001000000000000 );
|
|
|
1963 |
rem = sigA;
|
|
|
1964 |
expDiff = expA - expB;
|
|
|
1965 |
if ( expDiff < 1 ) {
|
|
|
1966 |
if ( expDiff < -1 ) return a;
|
|
|
1967 |
if ( expDiff ) {
|
|
|
1968 |
--expB;
|
|
|
1969 |
sigB = softfloat_add128( sigB.v64, sigB.v0, sigB.v64, sigB.v0 );
|
|
|
1970 |
q = 0;
|
|
|
1971 |
} else {
|
|
|
1972 |
q = softfloat_le128( sigB.v64, sigB.v0, rem.v64, rem.v0 );
|
|
|
1973 |
if ( q ) {
|
|
|
1974 |
rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
1975 |
}
|
|
|
1976 |
}
|
|
|
1977 |
} else {
|
|
|
1978 |
recip32 = softfloat_approxRecip32_1( sigB.v64>>17 );
|
|
|
1979 |
expDiff -= 30;
|
|
|
1980 |
for (;;) {
|
|
|
1981 |
q64 = (uint_fast64_t) (uint32_t) (rem.v64>>19) * recip32;
|
|
|
1982 |
if ( expDiff < 0 ) break;
|
|
|
1983 |
q = (q64 + 0x80000000)>>32;
|
|
|
1984 |
rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, 29 );
|
|
|
1985 |
term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
|
|
|
1986 |
rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
|
|
|
1987 |
if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
|
|
1988 |
rem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
1989 |
}
|
|
|
1990 |
expDiff -= 29;
|
|
|
1991 |
}
|
|
|
1992 |
/*--------------------------------------------------------------------
|
|
|
1993 |
| (`expDiff' cannot be less than -29 here.)
|
|
|
1994 |
*--------------------------------------------------------------------*/
|
|
|
1995 |
q = (uint32_t) (q64>>32)>>(~expDiff & 31);
|
|
|
1996 |
rem = softfloat_shortShiftLeft128( rem.v64, rem.v0, expDiff + 30 );
|
|
|
1997 |
term = softfloat_mul128By32( sigB.v64, sigB.v0, q );
|
|
|
1998 |
rem = softfloat_sub128( rem.v64, rem.v0, term.v64, term.v0 );
|
|
|
1999 |
if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
|
|
2000 |
altRem = softfloat_add128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
2001 |
goto selectRem;
|
|
|
2002 |
}
|
|
|
2003 |
}
|
|
|
2004 |
/*------------------------------------------------------------------------
|
|
|
2005 |
*------------------------------------------------------------------------*/
|
|
|
2006 |
do {
|
|
|
2007 |
altRem = rem;
|
|
|
2008 |
++q;
|
|
|
2009 |
rem = softfloat_sub128( rem.v64, rem.v0, sigB.v64, sigB.v0 );
|
|
|
2010 |
} while ( ! (rem.v64 & UINT64_C( 0x8000000000000000 )) );
|
|
|
2011 |
selectRem:
|
|
|
2012 |
meanRem = softfloat_add128( rem.v64, rem.v0, altRem.v64, altRem.v0 );
|
|
|
2013 |
if (
|
|
|
2014 |
(meanRem.v64 & UINT64_C( 0x8000000000000000 ))
|
|
|
2015 |
|| (! (meanRem.v64 | meanRem.v0) && (q & 1))
|
|
|
2016 |
) {
|
|
|
2017 |
rem = altRem;
|
|
|
2018 |
}
|
|
|
2019 |
signRem = signA;
|
|
|
2020 |
if ( rem.v64 & UINT64_C( 0x8000000000000000 ) ) {
|
|
|
2021 |
signRem = ! signRem;
|
|
|
2022 |
rem = softfloat_sub128( 0, 0, rem.v64, rem.v0 );
|
|
|
2023 |
}
|
|
|
2024 |
return softfloat_normRoundPackToF128( signRem, expB - 1, rem.v64, rem.v0 );
|
|
|
2025 |
/*------------------------------------------------------------------------
|
|
|
2026 |
*------------------------------------------------------------------------*/
|
|
|
2027 |
propagateNaN:
|
|
|
2028 |
uiZ = softfloat_propagateNaNF128UI( uiA64, uiA0, uiB64, uiB0 );
|
|
|
2029 |
goto uiZ;
|
|
|
2030 |
/*------------------------------------------------------------------------
|
|
|
2031 |
*------------------------------------------------------------------------*/
|
|
|
2032 |
invalid:
|
|
|
2033 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
5014
|
2034 |
softfloat_commonNaNToF128M( (uint32_t*)&uiZ );
|
|
|
2035 |
// uiZ.v64 = defaultNaNF128UI64;
|
|
|
2036 |
// uiZ.v0 = defaultNaNF128UI0;
|
|
5004
|
2037 |
uiZ:
|
|
|
2038 |
uZ.ui = uiZ;
|
|
|
2039 |
return uZ.f;
|
|
|
2040 |
}
|
|
|
2041 |
|
|
5007
|
2042 |
bool
|
|
|
2043 |
f128_isNan( float128_t a) {
|
|
|
2044 |
int_fast32_t expA;
|
|
|
2045 |
struct uint128 sigA;
|
|
|
2046 |
uint_fast64_t uiA64, uiA0;
|
|
|
2047 |
union ui128_f128 uA;
|
|
|
2048 |
|
|
|
2049 |
uA.f = a;
|
|
|
2050 |
uiA64 = uA.ui.v64;
|
|
|
2051 |
uiA0 = uA.ui.v0;
|
|
|
2052 |
return isNaNF128UI( uiA64, uiA0 );
|
|
|
2053 |
// expA = expF128UI64( uiA64 );
|
|
|
2054 |
// sigA.v64 = fracF128UI64( uiA64 );
|
|
|
2055 |
// sigA.v0 = uiA0;
|
|
|
2056 |
//
|
|
|
2057 |
// if ( expA == 0x7FFF ) {
|
|
|
2058 |
// if ( sigA.v64 | sigA.v0 ) return 1;
|
|
|
2059 |
// }
|
|
|
2060 |
// return 0;
|
|
|
2061 |
}
|
|
|
2062 |
|
|
|
2063 |
bool
|
|
|
2064 |
f128_isInf( float128_t a) {
|
|
|
2065 |
int_fast32_t expA;
|
|
|
2066 |
struct uint128 sigA;
|
|
|
2067 |
uint_fast64_t uiA64, uiA0;
|
|
|
2068 |
union ui128_f128 uA;
|
|
|
2069 |
|
|
|
2070 |
uA.f = a;
|
|
|
2071 |
uiA64 = uA.ui.v64;
|
|
|
2072 |
uiA0 = uA.ui.v0;
|
|
|
2073 |
expA = expF128UI64( uiA64 );
|
|
|
2074 |
sigA.v64 = fracF128UI64( uiA64 );
|
|
|
2075 |
sigA.v0 = uiA0;
|
|
|
2076 |
|
|
|
2077 |
if ( expA == 0x7FFF ) {
|
|
|
2078 |
if ( sigA.v64 == 0 ) return 1;
|
|
|
2079 |
}
|
|
|
2080 |
return 0;
|
|
|
2081 |
}
|
|
|
2082 |
|
|
5004
|
2083 |
static bool
|
|
|
2084 |
f128_eq( float128_t a, float128_t b )
|
|
|
2085 |
{
|
|
|
2086 |
union ui128_f128 uA;
|
|
|
2087 |
uint_fast64_t uiA64, uiA0;
|
|
|
2088 |
union ui128_f128 uB;
|
|
|
2089 |
uint_fast64_t uiB64, uiB0;
|
|
|
2090 |
|
|
|
2091 |
uA.f = a;
|
|
|
2092 |
uiA64 = uA.ui.v64;
|
|
|
2093 |
uiA0 = uA.ui.v0;
|
|
|
2094 |
uB.f = b;
|
|
|
2095 |
uiB64 = uB.ui.v64;
|
|
|
2096 |
uiB0 = uB.ui.v0;
|
|
|
2097 |
if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
|
|
|
2098 |
if (
|
|
|
2099 |
softfloat_isSigNaNF128UI( uiA64, uiA0 )
|
|
|
2100 |
|| softfloat_isSigNaNF128UI( uiB64, uiB0 )
|
|
|
2101 |
) {
|
|
|
2102 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
|
2103 |
}
|
|
5007
|
2104 |
return 0;
|
|
5004
|
2105 |
}
|
|
|
2106 |
return
|
|
|
2107 |
(uiA0 == uiB0)
|
|
|
2108 |
&& ( (uiA64 == uiB64)
|
|
|
2109 |
|| (! uiA0 && ! ((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF )))
|
|
|
2110 |
);
|
|
|
2111 |
}
|
|
|
2112 |
|
|
|
2113 |
static bool
|
|
|
2114 |
f128_lt( float128_t a, float128_t b )
|
|
|
2115 |
{
|
|
|
2116 |
union ui128_f128 uA;
|
|
|
2117 |
uint_fast64_t uiA64, uiA0;
|
|
|
2118 |
union ui128_f128 uB;
|
|
|
2119 |
uint_fast64_t uiB64, uiB0;
|
|
|
2120 |
bool signA, signB;
|
|
|
2121 |
|
|
|
2122 |
uA.f = a;
|
|
|
2123 |
uiA64 = uA.ui.v64;
|
|
|
2124 |
uiA0 = uA.ui.v0;
|
|
|
2125 |
uB.f = b;
|
|
|
2126 |
uiB64 = uB.ui.v64;
|
|
|
2127 |
uiB0 = uB.ui.v0;
|
|
|
2128 |
if ( isNaNF128UI( uiA64, uiA0 ) || isNaNF128UI( uiB64, uiB0 ) ) {
|
|
|
2129 |
softfloat_raiseFlags( softfloat_flag_invalid );
|
|
5007
|
2130 |
return 0;
|
|
5004
|
2131 |
}
|
|
|
2132 |
signA = signF128UI64( uiA64 );
|
|
|
2133 |
signB = signF128UI64( uiB64 );
|
|
|
2134 |
return
|
|
|
2135 |
(signA != signB)
|
|
|
2136 |
? signA
|
|
|
2137 |
&& (((uiA64 | uiB64) & UINT64_C( 0x7FFFFFFFFFFFFFFF ))
|
|
|
2138 |
| uiA0 | uiB0)
|
|
|
2139 |
: ((uiA64 != uiB64) || (uiA0 != uiB0))
|
|
|
2140 |
&& (signA ^ softfloat_lt128( uiA64, uiA0, uiB64, uiB0 ));
|
|
|
2141 |
}
|
|
|
2142 |
|
|
|
2143 |
static inline void
|
|
|
2144 |
f128M_div( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
2145 |
{
|
|
|
2146 |
*zPtr = f128_div( *aPtr, *bPtr );
|
|
|
2147 |
}
|
|
|
2148 |
|
|
|
2149 |
static inline void
|
|
|
2150 |
f128M_rem( const float128_t *aPtr, const float128_t *bPtr, float128_t *zPtr )
|
|
|
2151 |
{
|
|
|
2152 |
*zPtr = f128_rem( *aPtr, *bPtr );
|
|
|
2153 |
}
|
|
|
2154 |
|
|
|
2155 |
static inline bool
|
|
|
2156 |
f128M_eq( const float128_t *aPtr, const float128_t *bPtr )
|
|
|
2157 |
{
|
|
|
2158 |
return f128_eq( *aPtr, *bPtr );
|
|
|
2159 |
}
|
|
|
2160 |
|
|
|
2161 |
static inline bool
|
|
|
2162 |
f128M_lt( const float128_t *aPtr, const float128_t *bPtr )
|
|
|
2163 |
{
|
|
|
2164 |
return f128_lt( *aPtr, *bPtr );
|
|
|
2165 |
}
|
|
|
2166 |
|
|
4993
|
2167 |
#endif // SUPPORT_QUADFLOAT
|
|
4992
|
2168 |
|
|
|
2169 |
%}
|
|
|
2170 |
! !
|
|
4991
|
2171 |
|
|
4983
|
2172 |
!QuadFloat class methodsFor:'documentation'!
|
|
|
2173 |
|
|
|
2174 |
documentation
|
|
|
2175 |
"
|
|
|
2176 |
QuadFloats represent rational numbers with limited precision
|
|
5007
|
2177 |
and are mapped to IEEE quadruple precision format (128bit),
|
|
|
2178 |
also called binary128.
|
|
|
2179 |
|
|
4983
|
2180 |
If the underlying cpu supports them natively, the machine format (long double) is
|
|
|
2181 |
used. Otherwise, a software emulation is done, which is much slower.
|
|
|
2182 |
Thus only use them, if you really need the additional precision;
|
|
|
2183 |
if not, use Float (which are doubles) or LongFloats which usually have IEEE extended precision (80bit).
|
|
|
2184 |
|
|
|
2185 |
QuadFloats give you definite 128 bit quadruple floats,
|
|
|
2186 |
thus, code using quadFloats is guaranteed to be portable from one architecture to another.
|
|
|
2187 |
|
|
|
2188 |
Representation:
|
|
4991
|
2189 |
128bit quadruple IEEE floats (16bytes);
|
|
|
2190 |
112 bit mantissa,
|
|
|
2191 |
16 bit exponent,
|
|
|
2192 |
34 decimal digits (approx.)
|
|
4983
|
2193 |
|
|
4992
|
2194 |
On Sparc CPUs, this is a native supported type (long double) and fast;
|
|
|
2195 |
on x86 CPUs, this is emulated and slow.
|
|
|
2196 |
|
|
4983
|
2197 |
Mixed mode arithmetic:
|
|
5007
|
2198 |
quadFloat op anyFloat -> quadFloat
|
|
|
2199 |
anyFloat op quadFloat -> quadFloat
|
|
4983
|
2200 |
|
|
|
2201 |
Range and precision of storage formats: see LimitedPrecisionReal >> documentation
|
|
|
2202 |
|
|
|
2203 |
[author:]
|
|
4991
|
2204 |
Claus Gittinger
|
|
4983
|
2205 |
|
|
|
2206 |
[see also:]
|
|
4991
|
2207 |
Number
|
|
|
2208 |
Float ShortFloat LongFloat Fraction FixedPoint Integer Complex
|
|
|
2209 |
FloatArray DoubleArray
|
|
|
2210 |
https://en.wikipedia.org/wiki/Extended_precision
|
|
4983
|
2211 |
"
|
|
|
2212 |
! !
|
|
|
2213 |
|
|
|
2214 |
!QuadFloat class methodsFor:'instance creation'!
|
|
|
2215 |
|
|
|
2216 |
basicNew
|
|
|
2217 |
"return a new quadFloat - here we return 0.0
|
|
|
2218 |
- QuadFloats are usually NOT created this way ...
|
|
|
2219 |
Its implemented here to allow things like binary store & load
|
|
|
2220 |
of quadFloats.
|
|
|
2221 |
(but it is not a good idea to store the bits of a float - the reader might have a
|
|
|
2222 |
totally different representation - so floats should be
|
|
|
2223 |
binary stored in a device independent format)."
|
|
|
2224 |
|
|
|
2225 |
%{ /* NOCONTEXT */
|
|
4993
|
2226 |
#ifdef SUPPORT_QUADFLOAT
|
|
4983
|
2227 |
OBJ newFloat;
|
|
4991
|
2228 |
|
|
|
2229 |
if (sizeof(long double) == sizeof(float128_t)) {
|
|
|
2230 |
__qMKLFLOAT(newFloat, 0.0); /* OBJECT ALLOCATION */
|
|
4983
|
2231 |
} else {
|
|
4991
|
2232 |
float128_t qf;
|
|
4992
|
2233 |
f64_to_f128M(0.0, &qf);
|
|
4991
|
2234 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
4983
|
2235 |
}
|
|
|
2236 |
RETURN (newFloat);
|
|
4993
|
2237 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2238 |
%}.
|
|
|
2239 |
self error:'QuadFloats not supported on this patform'
|
|
4983
|
2240 |
|
|
|
2241 |
"Created: / 06-06-2019 / 17:18:58 / Claus Gittinger"
|
|
|
2242 |
!
|
|
|
2243 |
|
|
|
2244 |
fromFloat:aFloat
|
|
|
2245 |
"return a new quadFloat, given a float value"
|
|
|
2246 |
|
|
|
2247 |
%{ /* NOCONTEXT */
|
|
4993
|
2248 |
#ifdef SUPPORT_QUADFLOAT
|
|
4983
|
2249 |
OBJ newFloat;
|
|
|
2250 |
|
|
|
2251 |
if (__isFloatLike(aFloat)) {
|
|
4992
|
2252 |
float64_t f = __floatVal(aFloat);
|
|
4991
|
2253 |
float128_t qf;
|
|
|
2254 |
|
|
4992
|
2255 |
f64_to_f128M(f, &qf);
|
|
4991
|
2256 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
2257 |
RETURN (newFloat);
|
|
4983
|
2258 |
}
|
|
4993
|
2259 |
#endif /* SUPPORT_QUADFLOAT */
|
|
4983
|
2260 |
%}.
|
|
4993
|
2261 |
aFloat isFloat ifTrue:[
|
|
4997
|
2262 |
self errorUnsupported
|
|
4993
|
2263 |
].
|
|
4997
|
2264 |
ArgumentError raise
|
|
4983
|
2265 |
|
|
|
2266 |
"
|
|
|
2267 |
QuadFloat fromFloat:123.0
|
|
|
2268 |
123.0 asQuadFloat
|
|
|
2269 |
123 asQuadFloat
|
|
|
2270 |
"
|
|
|
2271 |
|
|
|
2272 |
"Created: / 06-06-2019 / 18:01:03 / Claus Gittinger"
|
|
5011
|
2273 |
!
|
|
|
2274 |
|
|
5014
|
2275 |
fromLongFloat:aFloat
|
|
|
2276 |
"return a new quadFloat, given a long float value"
|
|
|
2277 |
|
|
|
2278 |
%{ /* NOCONTEXT */
|
|
|
2279 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2280 |
OBJ newFloat;
|
|
|
2281 |
union {
|
|
|
2282 |
LONGFLOAT_t lf; // is long double
|
|
|
2283 |
extFloat80_t ef; // is 80bit ext
|
|
|
2284 |
float128_t qf; // is 128bit quad
|
|
|
2285 |
} u;
|
|
|
2286 |
|
|
|
2287 |
if (__isLongFloat(aFloat)) {
|
|
|
2288 |
u.lf = __longFloatVal(aFloat);
|
|
|
2289 |
|
|
|
2290 |
if (sizeof(LONGFLOAT_t) == 16) {
|
|
|
2291 |
// longFloat is already 128 bits in size (sparc)
|
|
|
2292 |
__qMKQFLOAT(newFloat, u.qf); /* OBJECT ALLOCATION */
|
|
|
2293 |
RETURN (newFloat);
|
|
|
2294 |
}
|
|
|
2295 |
if (sizeof(LONGFLOAT_t) < 16) {
|
|
|
2296 |
// assume 80bit extended float format (amd64, x86_64)
|
|
|
2297 |
u.qf = extF80_to_f128( u.ef);
|
|
|
2298 |
__qMKQFLOAT(newFloat, u.qf); /* OBJECT ALLOCATION */
|
|
|
2299 |
RETURN (newFloat);
|
|
|
2300 |
}
|
|
|
2301 |
// fall into error case
|
|
|
2302 |
}
|
|
|
2303 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2304 |
%}.
|
|
|
2305 |
aFloat isLongFloat ifTrue:[
|
|
|
2306 |
self errorUnsupported
|
|
|
2307 |
].
|
|
|
2308 |
ArgumentError raise
|
|
|
2309 |
|
|
|
2310 |
"
|
|
|
2311 |
QuadFloat fromLongFloat:123.0 asLongFloat
|
|
|
2312 |
"
|
|
|
2313 |
!
|
|
|
2314 |
|
|
5011
|
2315 |
fromInteger:anInteger
|
|
|
2316 |
"return a new quadFloat, given an integer value"
|
|
|
2317 |
|
|
|
2318 |
%{ /* NOCONTEXT */
|
|
|
2319 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2320 |
OBJ newFloat;
|
|
|
2321 |
|
|
|
2322 |
if (__isSmallInteger(anInteger)) {
|
|
|
2323 |
INT iVal = __intVal(anInteger);
|
|
|
2324 |
float128_t qf;
|
|
|
2325 |
|
|
5014
|
2326 |
#if __POINTER_SIZE__ == 4
|
|
|
2327 |
i32_to_f128M( iVal, &qf );
|
|
|
2328 |
else
|
|
|
2329 |
i64_to_f128M( iVal, &qf );
|
|
|
2330 |
#endif
|
|
5011
|
2331 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
2332 |
RETURN (newFloat);
|
|
|
2333 |
}
|
|
|
2334 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2335 |
%}.
|
|
|
2336 |
^ super fromInteger:anInteger
|
|
|
2337 |
|
|
|
2338 |
"
|
|
|
2339 |
QuadFloat fromInteger:123
|
|
|
2340 |
123 asQuadFloat
|
|
|
2341 |
"
|
|
|
2342 |
!
|
|
|
2343 |
|
|
|
2344 |
fromShortFloat:aShortFloat
|
|
|
2345 |
"return a new quadFloat, given a float value"
|
|
|
2346 |
|
|
5013
|
2347 |
^ self fromFloat:(aShortFloat asFloat)
|
|
5011
|
2348 |
|
|
|
2349 |
"
|
|
|
2350 |
QuadFloat fromShortFloat:123.0 asShortFloat
|
|
|
2351 |
"
|
|
|
2352 |
|
|
|
2353 |
"Created: / 08-06-2019 / 03:28:37 / Claus Gittinger"
|
|
4983
|
2354 |
! !
|
|
|
2355 |
|
|
|
2356 |
!QuadFloat class methodsFor:'coercing & converting'!
|
|
|
2357 |
|
|
|
2358 |
coerce:aNumber
|
|
|
2359 |
"convert the argument aNumber into an instance of the receiver's class and return it."
|
|
|
2360 |
|
|
|
2361 |
^ aNumber asQuadFloat.
|
|
|
2362 |
|
|
|
2363 |
"Created: / 06-06-2019 / 16:51:01 / Claus Gittinger"
|
|
|
2364 |
! !
|
|
|
2365 |
|
|
|
2366 |
!QuadFloat class methodsFor:'constants'!
|
|
|
2367 |
|
|
|
2368 |
NaN
|
|
5011
|
2369 |
"return a quadFloat which represents not-a-Number (i.e. an invalid number)"
|
|
|
2370 |
|
|
|
2371 |
|nan|
|
|
4983
|
2372 |
|
|
|
2373 |
NaN isNil ifTrue:[
|
|
5011
|
2374 |
%{ /* NOCONTEXT */
|
|
|
2375 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2376 |
{
|
|
|
2377 |
OBJ newFloat;
|
|
|
2378 |
float128_t qf;
|
|
|
2379 |
|
|
5014
|
2380 |
softfloat_commonNaNToF128M( (uint32_t*)(&qf) );
|
|
5011
|
2381 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
2382 |
nan = newFloat;
|
|
|
2383 |
}
|
|
|
2384 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2385 |
%}.
|
|
|
2386 |
NaN := nan
|
|
4983
|
2387 |
].
|
|
|
2388 |
^ NaN
|
|
|
2389 |
!
|
|
|
2390 |
|
|
|
2391 |
e
|
|
|
2392 |
"return the constant e as quadFloat"
|
|
|
2393 |
|
|
|
2394 |
E isNil ifTrue:[
|
|
4991
|
2395 |
"/ eDigits has enough digits for 128bit IEEE quads
|
|
|
2396 |
"/ do not use as a literal constant here - we cannot depend on the underlying C-compiler here...
|
|
|
2397 |
E := self readFrom:(Number eDigits)
|
|
4983
|
2398 |
].
|
|
|
2399 |
^ E
|
|
|
2400 |
|
|
|
2401 |
"Created: / 06-06-2019 / 17:01:54 / Claus Gittinger"
|
|
|
2402 |
!
|
|
|
2403 |
|
|
5012
|
2404 |
negativeInfinity
|
|
|
2405 |
"return a quadFloat which represents -INF"
|
|
|
2406 |
|
|
|
2407 |
|inf|
|
|
|
2408 |
|
|
|
2409 |
NegativeInfinity isNil ifTrue:[
|
|
|
2410 |
%{ /* NOCONTEXT */
|
|
|
2411 |
#ifdef SUPPORT_QUADFLOAT
|
|
5013
|
2412 |
{
|
|
|
2413 |
OBJ newFloat;
|
|
|
2414 |
struct uint128 uiZ;
|
|
|
2415 |
union ui128_f128 uZ;
|
|
|
2416 |
float128_t qf;
|
|
5012
|
2417 |
|
|
5013
|
2418 |
uiZ.v64 = packToF128UI64( 1, 0x7FFF, 0 );
|
|
|
2419 |
uiZ.v0 = 0;
|
|
|
2420 |
uZ.ui = uiZ;
|
|
|
2421 |
qf = uZ.f;
|
|
|
2422 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
2423 |
inf = newFloat;
|
|
|
2424 |
}
|
|
5012
|
2425 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2426 |
%}.
|
|
5013
|
2427 |
NegativeInfinity := inf
|
|
5012
|
2428 |
].
|
|
|
2429 |
^ NegativeInfinity
|
|
|
2430 |
|
|
|
2431 |
"Created: / 08-06-2019 / 14:05:50 / Claus Gittinger"
|
|
|
2432 |
!
|
|
|
2433 |
|
|
4983
|
2434 |
pi
|
|
|
2435 |
"return the constant pi as quadFloat"
|
|
|
2436 |
|
|
|
2437 |
Pi isNil ifTrue:[
|
|
4991
|
2438 |
"/ piDigits has enough digits for 128bit IEEE quads
|
|
|
2439 |
"/ do not use as a literal constant here - we cannot depend on the underlying C-compiler here...
|
|
|
2440 |
Pi := self readFrom:(Number piDigits)
|
|
4983
|
2441 |
].
|
|
|
2442 |
^ Pi
|
|
|
2443 |
|
|
|
2444 |
"Created: / 06-06-2019 / 17:09:51 / Claus Gittinger"
|
|
5006
|
2445 |
!
|
|
|
2446 |
|
|
5014
|
2447 |
infinity
|
|
5012
|
2448 |
"return a quadFloat which represents +INF"
|
|
|
2449 |
|
|
|
2450 |
|inf|
|
|
|
2451 |
|
|
|
2452 |
PositiveInfinity isNil ifTrue:[
|
|
|
2453 |
%{ /* NOCONTEXT */
|
|
|
2454 |
#ifdef SUPPORT_QUADFLOAT
|
|
5013
|
2455 |
{
|
|
|
2456 |
OBJ newFloat;
|
|
|
2457 |
struct uint128 uiZ;
|
|
|
2458 |
union ui128_f128 uZ;
|
|
|
2459 |
float128_t qf;
|
|
5012
|
2460 |
|
|
5013
|
2461 |
uiZ.v64 = packToF128UI64( 0, 0x7FFF, 0 );
|
|
|
2462 |
uiZ.v0 = 0;
|
|
|
2463 |
uZ.ui = uiZ;
|
|
|
2464 |
qf = uZ.f;
|
|
|
2465 |
__qMKQFLOAT(newFloat, qf); /* OBJECT ALLOCATION */
|
|
|
2466 |
inf = newFloat;
|
|
|
2467 |
}
|
|
5012
|
2468 |
#endif /* SUPPORT_QUADFLOAT */
|
|
|
2469 |
%}.
|
|
5013
|
2470 |
PositiveInfinity := inf
|
|
5012
|
2471 |
].
|
|
|
2472 |
^ PositiveInfinity
|
|
|
2473 |
|
|
|
2474 |
"Created: / 08-06-2019 / 14:05:26 / Claus Gittinger"
|
|
|
2475 |
!
|
|
|
2476 |
|
|
5006
|
2477 |
unity
|
|
|
2478 |
"return the neutral element for multiplication (1.0) as QuadFloat"
|
|
|
2479 |
|
|
|
2480 |
QuadFloatOne isNil ifTrue:[
|
|
5007
|
2481 |
QuadFloatOne := 1.0 asQuadFloat.
|
|
5006
|
2482 |
].
|
|
|
2483 |
^ QuadFloatOne
|
|
|
2484 |
|
|
|
2485 |
"Created: / 07-06-2019 / 03:26:38 / Claus Gittinger"
|
|
5008
|
2486 |
!
|
|
|
2487 |
|
|
|
2488 |
zero
|
|
|
2489 |
"return the neutral element for addition (0.0) as QuadFloat"
|
|
|
2490 |
|
|
|
2491 |
QuadFloatZero isNil ifTrue:[
|
|
5011
|
2492 |
QuadFloatZero := 0.0 asQuadFloat
|
|
5008
|
2493 |
].
|
|
|
2494 |
^ QuadFloatZero
|
|
|
2495 |
|
|
|
2496 |
"Created: / 07-06-2019 / 09:22:56 / Claus Gittinger"
|
|
4983
|
2497 |
! !
|
|
|
2498 |
|
|
5003
|
2499 |
!QuadFloat class methodsFor:'error reportng'!
|
|
4994
|
2500 |
|
|
4997
|
2501 |
errorUnsupported
|
|
|
2502 |
self error:'QuadFloats not supported on this patform'
|
|
|
2503 |
|
|
5003
|
2504 |
"Created: / 07-06-2019 / 02:44:39 / Claus Gittinger"
|
|
4994
|
2505 |
! !
|
|
|
2506 |
|
|
5005
|
2507 |
!QuadFloat class methodsFor:'queries'!
|
|
|
2508 |
|
|
|
2509 |
numBitsInMantissa
|
|
|
2510 |
"answer the number of bits in the mantissa
|
|
|
2511 |
the hidden bit is not counted here:
|
|
|
2512 |
|
|
|
2513 |
This is an 128bit quadfloat,
|
|
5007
|
2514 |
where 112 bits are available in the mantissa:
|
|
|
2515 |
seeeeeee eeeeeeee mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm mmmmmmmm...
|
|
5005
|
2516 |
"
|
|
|
2517 |
|
|
|
2518 |
^ 112
|
|
|
2519 |
|
|
|
2520 |
"
|
|
|
2521 |
1.0 class numBitsInMantissa
|
|
|
2522 |
1.0 asShortFloat class numBitsInMantissa
|
|
|
2523 |
1.0 asLongFloat class numBitsInMantissa
|
|
|
2524 |
1.0 asQuadFloat class numBitsInMantissa
|
|
|
2525 |
"
|
|
|
2526 |
|
|
|
2527 |
"Created: / 07-06-2019 / 03:24:20 / Claus Gittinger"
|
|
|
2528 |
! !
|
|
|
2529 |
|
|
4994
|
2530 |
!QuadFloat methodsFor:'arithmetic'!
|
|
|
2531 |
|
|
|
2532 |
* aNumber
|
|
|
2533 |
"return the product of the receiver and the argument."
|
|
|
2534 |
|
|
5011
|
2535 |
thisContext isRecursive ifTrue:[self halt].
|
|
4994
|
2536 |
^ aNumber productFromQuadFloat:self
|
|
|
2537 |
!
|
|
|
2538 |
|
|
|
2539 |
+ aNumber
|
|
|
2540 |
"return the sum of the receiver and the argument, aNumber"
|
|
|
2541 |
|
|
|
2542 |
^ aNumber sumFromQuadFloat:self
|
|
|
2543 |
!
|
|
|
2544 |
|
|
|
2545 |
- aNumber
|
|
|
2546 |
"return the difference of the receiver and the argument, aNumber"
|
|
|
2547 |
|
|
|
2548 |
^ aNumber differenceFromQuadFloat:self
|
|
|
2549 |
!
|
|
|
2550 |
|
|
|
2551 |
/ aNumber
|
|
|
2552 |
"return the quotient of the receiver and the argument, aNumber"
|
|
|
2553 |
|
|
|
2554 |
aNumber isZero ifTrue:[
|
|
|
2555 |
"
|
|
|
2556 |
No, you shalt not divide by zero
|
|
|
2557 |
"
|
|
|
2558 |
^ ZeroDivide raiseRequestWith:thisContext.
|
|
|
2559 |
].
|
|
|
2560 |
^ aNumber quotientFromQuadFloat:self
|
|
|
2561 |
!
|
|
|
2562 |
|
|
|
2563 |
rem: aNumber
|
|
|
2564 |
"return the floating point remainder of the receiver and the argument, aNumber"
|
|
|
2565 |
|
|
|
2566 |
aNumber isZero ifTrue:[
|
|
|
2567 |
"
|
|
|
2568 |
No, you shalt not divide by zero
|
|
|
2569 |
"
|
|
|
2570 |
^ ZeroDivide raiseRequestWith:thisContext.
|
|
|
2571 |
].
|
|
|
2572 |
^ aNumber remainderFromLongFloat:self
|
|
|
2573 |
! !
|
|
|
2574 |
|
|
5008
|
2575 |
!QuadFloat methodsFor:'coercing & converting'!
|
|
|
2576 |
|
|
|
2577 |
generality
|
|
|
2578 |
"return the generality value - see ArithmeticValue>>retry:coercing:"
|
|
|
2579 |
|
|
|
2580 |
^ 93
|
|
|
2581 |
|
|
|
2582 |
"Created: / 07-06-2019 / 09:30:58 / Claus Gittinger"
|
|
|
2583 |
! !
|
|
|
2584 |
|
|
|
2585 |
!QuadFloat methodsFor:'comparing'!
|
|
|
2586 |
|
|
|
2587 |
< aNumber
|
|
|
2588 |
"return true, if the argument is greater"
|
|
|
2589 |
|
|
|
2590 |
^ aNumber lessFromQuadFloat:self
|
|
|
2591 |
|
|
|
2592 |
"Created: / 07-06-2019 / 09:25:47 / Claus Gittinger"
|
|
|
2593 |
!
|
|
|
2594 |
|
|
|
2595 |
= aNumber
|
|
|
2596 |
"return true, if the argument represents the same numeric value
|
|
|
2597 |
as the receiver, false otherwise"
|
|
|
2598 |
|
|
|
2599 |
^ aNumber equalFromQuadFloat:self
|
|
|
2600 |
|
|
|
2601 |
"Created: / 07-06-2019 / 09:25:27 / Claus Gittinger"
|
|
|
2602 |
!
|
|
|
2603 |
|
|
|
2604 |
hash
|
|
|
2605 |
"return a number for hashing; redefined, since floats compare
|
|
|
2606 |
by numeric value (i.e. 3.0 = 3), therefore 3.0 hash must be the same
|
|
|
2607 |
as 3 hash."
|
|
|
2608 |
|
|
|
2609 |
|i|
|
|
|
2610 |
|
|
|
2611 |
(self >= SmallInteger minVal and:[self <= SmallInteger maxVal]) ifTrue:[
|
|
5011
|
2612 |
i := self asInteger.
|
|
|
2613 |
self = i ifTrue:[
|
|
|
2614 |
^ i hash
|
|
|
2615 |
].
|
|
5008
|
2616 |
].
|
|
|
2617 |
|
|
|
2618 |
^ self asFloat hash
|
|
|
2619 |
|
|
|
2620 |
"
|
|
|
2621 |
1.2345 hash
|
|
|
2622 |
1.2345 asShortFloat hash
|
|
|
2623 |
1.2345 asLongFloat hash
|
|
|
2624 |
1.2345 asQuadFloat hash
|
|
5011
|
2625 |
|
|
5008
|
2626 |
1.0 hash
|
|
|
2627 |
1.0 asShortFloat hash
|
|
|
2628 |
1.0 asLongFloat hash
|
|
|
2629 |
1.0 asQuadFloat hash
|
|
5011
|
2630 |
|
|
5008
|
2631 |
0.5 asShortFloat hash
|
|
|
2632 |
0.5 asShortFloat hash
|
|
|
2633 |
0.5 asLongFloat hash
|
|
|
2634 |
0.5 asQuadFloat hash
|
|
|
2635 |
|
|
|
2636 |
0.25 asShortFloat hash
|
|
|
2637 |
0.25 asShortFloat hash
|
|
|
2638 |
0.25 asLongFloat hash
|
|
|
2639 |
0.25 asQuadFloat hash
|
|
|
2640 |
"
|
|
|
2641 |
|
|
|
2642 |
"Created: / 07-06-2019 / 09:28:07 / Claus Gittinger"
|
|
|
2643 |
! !
|
|
|
2644 |
|
|
5003
|
2645 |
!QuadFloat methodsFor:'double dispatching'!
|
|
|
2646 |
|
|
|
2647 |
differenceFromQuadFloat:aQuadFloat
|
|
|
2648 |
%{
|
|
|
2649 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2650 |
OBJ newFloat;
|
|
|
2651 |
float128_t result, myVal, argVal;
|
|
|
2652 |
|
|
|
2653 |
myVal = __quadFloatVal(self);
|
|
|
2654 |
argVal = __quadFloatVal(aQuadFloat);
|
|
|
2655 |
f128M_sub( &myVal, &argVal, &result );
|
|
|
2656 |
__qMKQFLOAT(newFloat, result);
|
|
|
2657 |
RETURN ( newFloat );
|
|
|
2658 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2659 |
%}.
|
|
|
2660 |
self errorUnsupported
|
|
|
2661 |
!
|
|
|
2662 |
|
|
5005
|
2663 |
equalFromQuadFloat:aQuadFloat
|
|
|
2664 |
"sent when aQuadFloat does not know how to compare agaist the receiver, self"
|
|
5003
|
2665 |
|
|
|
2666 |
%{
|
|
|
2667 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2668 |
OBJ newFloat;
|
|
|
2669 |
float128_t result, myVal, argVal;
|
|
|
2670 |
|
|
|
2671 |
myVal = __quadFloatVal(self);
|
|
|
2672 |
argVal = __quadFloatVal(aQuadFloat);
|
|
5011
|
2673 |
RETURN (f128M_eq( &argVal, &myVal ) ? true : false);
|
|
5003
|
2674 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2675 |
%}.
|
|
|
2676 |
self errorUnsupported
|
|
5011
|
2677 |
|
|
|
2678 |
"Modified: / 08-06-2019 / 13:31:48 / Claus Gittinger"
|
|
5004
|
2679 |
!
|
|
|
2680 |
|
|
5005
|
2681 |
lessFromQuadFloat:aQuadFloat
|
|
|
2682 |
"sent when aQuadFloat does not know how to compare agaist the receiver, self"
|
|
|
2683 |
|
|
|
2684 |
%{
|
|
|
2685 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2686 |
OBJ newFloat;
|
|
|
2687 |
float128_t result, myVal, argVal;
|
|
|
2688 |
|
|
|
2689 |
myVal = __quadFloatVal(self);
|
|
|
2690 |
argVal = __quadFloatVal(aQuadFloat);
|
|
5011
|
2691 |
RETURN (f128M_lt( &argVal, &myVal ) ? true : false);
|
|
5005
|
2692 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2693 |
%}.
|
|
|
2694 |
self errorUnsupported
|
|
|
2695 |
!
|
|
|
2696 |
|
|
5004
|
2697 |
productFromQuadFloat:aQuadFloat
|
|
|
2698 |
"sent when aQuadFloat does not know how to multiply the receiver, self"
|
|
|
2699 |
|
|
|
2700 |
%{
|
|
|
2701 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2702 |
OBJ newFloat;
|
|
|
2703 |
float128_t result, myVal, argVal;
|
|
|
2704 |
|
|
|
2705 |
myVal = __quadFloatVal(self);
|
|
|
2706 |
argVal = __quadFloatVal(aQuadFloat);
|
|
|
2707 |
f128M_mul( &myVal, &argVal, &result );
|
|
|
2708 |
__qMKQFLOAT(newFloat, result);
|
|
|
2709 |
RETURN ( newFloat );
|
|
|
2710 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2711 |
%}.
|
|
|
2712 |
self errorUnsupported
|
|
|
2713 |
!
|
|
|
2714 |
|
|
|
2715 |
quotientFromQuadFloat:aQuadFloat
|
|
|
2716 |
"sent when aQuadFloat does not know how to multiply the receiver, self"
|
|
|
2717 |
|
|
|
2718 |
%{
|
|
|
2719 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2720 |
OBJ newFloat;
|
|
|
2721 |
float128_t result, myVal, argVal;
|
|
|
2722 |
|
|
|
2723 |
myVal = __quadFloatVal(self);
|
|
|
2724 |
argVal = __quadFloatVal(aQuadFloat);
|
|
|
2725 |
f128M_div( &myVal, &argVal, &result );
|
|
|
2726 |
__qMKQFLOAT(newFloat, result);
|
|
|
2727 |
RETURN ( newFloat );
|
|
|
2728 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2729 |
%}.
|
|
|
2730 |
self errorUnsupported
|
|
|
2731 |
!
|
|
|
2732 |
|
|
5005
|
2733 |
sumFromQuadFloat:aQuadFloat
|
|
|
2734 |
"sent when aQuadFloat does not know how to add the receiver, self"
|
|
5004
|
2735 |
|
|
|
2736 |
%{
|
|
|
2737 |
#ifdef SUPPORT_QUADFLOAT
|
|
|
2738 |
OBJ newFloat;
|
|
|
2739 |
float128_t result, myVal, argVal;
|
|
|
2740 |
|
|
|
2741 |
myVal = __quadFloatVal(self);
|
|
|
2742 |
argVal = __quadFloatVal(aQuadFloat);
|
|
5005
|
2743 |
f128M_add( &myVal, &argVal, &result );
|
|
|
2744 |
__qMKQFLOAT(newFloat, result);
|
|
5004
|
2745 |
RETURN ( newFloat );
|
|
|
2746 |
#endif // SUPPORT_QUADFLOAT
|
|
|
2747 |
%}.
|
|
|
2748 |
self errorUnsupported
|
|
5003
|
2749 |
! !
|
|
|
2750 |
|
|
|
2751 |
!QuadFloat methodsFor:'error reportng'!
|
|
|
2752 |
|
|
|
2753 |
errorUnsupported
|
|
|
2754 |
self class errorUnsupported
|
|
|
2755 |
|
|
|
2756 |
"Modified: / 07-06-2019 / 02:44:51 / Claus Gittinger"
|
|
|
2757 |
! !
|
|
|
2758 |
|
|
5011
|
2759 |
!QuadFloat methodsFor:'queries'!
|
|
|
2760 |
|
|
|
2761 |
isFinite
|
|
|
2762 |
"return true, if the receiver is a finite float
|
|
|
2763 |
i.e. not NaN and not infinite."
|
|
|
2764 |
|
|
|
2765 |
%{ /* NOCONTEXT */
|
|
|
2766 |
float128_t myVal;
|
|
|
2767 |
|
|
|
2768 |
myVal = __quadFloatVal(self);
|
|
|
2769 |
RETURN (f128_isInf(myVal) ? true : false)
|
|
|
2770 |
%}.
|
|
|
2771 |
|
|
|
2772 |
"
|
|
|
2773 |
1.0 isFinite
|
|
|
2774 |
self NaN isFinite
|
|
|
2775 |
self infinity isFinite
|
|
|
2776 |
(0.0 uncheckedDivide: 0.0) isFinite
|
|
|
2777 |
(1.0 uncheckedDivide: 0.0) isFinite
|
|
|
2778 |
"
|
|
|
2779 |
!
|
|
|
2780 |
|
|
|
2781 |
isNaN
|
|
|
2782 |
"return true, if the receiver is a finite float
|
|
|
2783 |
i.e. not NaN and not infinite."
|
|
|
2784 |
|
|
|
2785 |
%{ /* NOCONTEXT */
|
|
|
2786 |
float128_t myVal;
|
|
|
2787 |
|
|
|
2788 |
myVal = __quadFloatVal(self);
|
|
|
2789 |
RETURN (f128_isNan(myVal) ? true : false)
|
|
|
2790 |
%}.
|
|
|
2791 |
|
|
|
2792 |
"
|
|
|
2793 |
1.0 isFinite
|
|
|
2794 |
self NaN isFinite
|
|
|
2795 |
self infinity isFinite
|
|
|
2796 |
(0.0 uncheckedDivide: 0.0) isFinite
|
|
|
2797 |
(1.0 uncheckedDivide: 0.0) isFinite
|
|
|
2798 |
"
|
|
|
2799 |
! !
|
|
|
2800 |
|
|
4983
|
2801 |
!QuadFloat class methodsFor:'documentation'!
|
|
|
2802 |
|
|
|
2803 |
version_CVS
|
|
|
2804 |
^ '$Header$'
|
|
|
2805 |
! !
|