Mercurial > hg > graal-jvmci-8
annotate src/share/vm/runtime/sharedRuntimeTrans.cpp @ 20286:631c3a4ea10c
8043301: Duplicate definitions in vm/runtime/sharedRuntimeTrans.cpp versus math.h in VS2013
Summary: Factor out definitions of copysignA and scalbnA into new file sharedRuntimeMath.hpp
Reviewed-by: kvn
author | lfoltan |
---|---|
date | Thu, 22 May 2014 11:36:23 -0400 |
parents | bdd155477289 |
children | a073be2ce5c2 |
rev | line source |
---|---|
0 | 1 /* |
20286
631c3a4ea10c
8043301: Duplicate definitions in vm/runtime/sharedRuntimeTrans.cpp versus math.h in VS2013
lfoltan
parents:
14411
diff
changeset
|
2 * Copyright (c) 2005, 2014, Oracle and/or its affiliates. All rights reserved. |
0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
0
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
0
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
0
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "prims/jni.h" | |
27 #include "runtime/interfaceSupport.hpp" | |
28 #include "runtime/sharedRuntime.hpp" | |
0 | 29 |
30 // This file contains copies of the fdlibm routines used by | |
31 // StrictMath. It turns out that it is almost always required to use | |
32 // these runtime routines; the Intel CPU doesn't meet the Java | |
33 // specification for sin/cos outside a certain limited argument range, | |
34 // and the SPARC CPU doesn't appear to have sin/cos instructions. It | |
35 // also turns out that avoiding the indirect call through function | |
36 // pointer out to libjava.so in SharedRuntime speeds these routines up | |
37 // by roughly 15% on both Win32/x86 and Solaris/SPARC. | |
38 | |
39 // Enabling optimizations in this file causes incorrect code to be | |
40 // generated; can not figure out how to turn down optimization for one | |
41 // file in the IDE on Windows | |
42 #ifdef WIN32 | |
43 # pragma optimize ( "", off ) | |
44 #endif | |
45 | |
20286
631c3a4ea10c
8043301: Duplicate definitions in vm/runtime/sharedRuntimeTrans.cpp versus math.h in VS2013
lfoltan
parents:
14411
diff
changeset
|
46 #include "runtime/sharedRuntimeMath.hpp" |
0 | 47 |
48 /* __ieee754_log(x) | |
49 * Return the logrithm of x | |
50 * | |
51 * Method : | |
52 * 1. Argument Reduction: find k and f such that | |
53 * x = 2^k * (1+f), | |
54 * where sqrt(2)/2 < 1+f < sqrt(2) . | |
55 * | |
56 * 2. Approximation of log(1+f). | |
57 * Let s = f/(2+f) ; based on log(1+f) = log(1+s) - log(1-s) | |
58 * = 2s + 2/3 s**3 + 2/5 s**5 + ....., | |
59 * = 2s + s*R | |
60 * We use a special Reme algorithm on [0,0.1716] to generate | |
61 * a polynomial of degree 14 to approximate R The maximum error | |
62 * of this polynomial approximation is bounded by 2**-58.45. In | |
63 * other words, | |
64 * 2 4 6 8 10 12 14 | |
65 * R(z) ~ Lg1*s +Lg2*s +Lg3*s +Lg4*s +Lg5*s +Lg6*s +Lg7*s | |
66 * (the values of Lg1 to Lg7 are listed in the program) | |
67 * and | |
68 * | 2 14 | -58.45 | |
69 * | Lg1*s +...+Lg7*s - R(z) | <= 2 | |
70 * | | | |
71 * Note that 2s = f - s*f = f - hfsq + s*hfsq, where hfsq = f*f/2. | |
72 * In order to guarantee error in log below 1ulp, we compute log | |
73 * by | |
74 * log(1+f) = f - s*(f - R) (if f is not too large) | |
75 * log(1+f) = f - (hfsq - s*(hfsq+R)). (better accuracy) | |
76 * | |
77 * 3. Finally, log(x) = k*ln2 + log(1+f). | |
78 * = k*ln2_hi+(f-(hfsq-(s*(hfsq+R)+k*ln2_lo))) | |
79 * Here ln2 is split into two floating point number: | |
80 * ln2_hi + ln2_lo, | |
81 * where n*ln2_hi is always exact for |n| < 2000. | |
82 * | |
83 * Special cases: | |
84 * log(x) is NaN with signal if x < 0 (including -INF) ; | |
85 * log(+INF) is +INF; log(0) is -INF with signal; | |
86 * log(NaN) is that NaN with no signal. | |
87 * | |
88 * Accuracy: | |
89 * according to an error analysis, the error is always less than | |
90 * 1 ulp (unit in the last place). | |
91 * | |
92 * Constants: | |
93 * The hexadecimal values are the intended ones for the following | |
94 * constants. The decimal values may be used, provided that the | |
95 * compiler will convert from decimal to binary accurately enough | |
96 * to produce the hexadecimal values shown. | |
97 */ | |
98 | |
99 static const double | |
100 ln2_hi = 6.93147180369123816490e-01, /* 3fe62e42 fee00000 */ | |
101 ln2_lo = 1.90821492927058770002e-10, /* 3dea39ef 35793c76 */ | |
102 Lg1 = 6.666666666666735130e-01, /* 3FE55555 55555593 */ | |
103 Lg2 = 3.999999999940941908e-01, /* 3FD99999 9997FA04 */ | |
104 Lg3 = 2.857142874366239149e-01, /* 3FD24924 94229359 */ | |
105 Lg4 = 2.222219843214978396e-01, /* 3FCC71C5 1D8E78AF */ | |
106 Lg5 = 1.818357216161805012e-01, /* 3FC74664 96CB03DE */ | |
107 Lg6 = 1.531383769920937332e-01, /* 3FC39A09 D078C69F */ | |
108 Lg7 = 1.479819860511658591e-01; /* 3FC2F112 DF3E5244 */ | |
109 | |
110 static double zero = 0.0; | |
111 | |
112 static double __ieee754_log(double x) { | |
113 double hfsq,f,s,z,R,w,t1,t2,dk; | |
114 int k,hx,i,j; | |
115 unsigned lx; | |
116 | |
117 hx = __HI(x); /* high word of x */ | |
118 lx = __LO(x); /* low word of x */ | |
119 | |
120 k=0; | |
121 if (hx < 0x00100000) { /* x < 2**-1022 */ | |
122 if (((hx&0x7fffffff)|lx)==0) | |
123 return -two54/zero; /* log(+-0)=-inf */ | |
124 if (hx<0) return (x-x)/zero; /* log(-#) = NaN */ | |
125 k -= 54; x *= two54; /* subnormal number, scale up x */ | |
126 hx = __HI(x); /* high word of x */ | |
127 } | |
128 if (hx >= 0x7ff00000) return x+x; | |
129 k += (hx>>20)-1023; | |
130 hx &= 0x000fffff; | |
131 i = (hx+0x95f64)&0x100000; | |
132 __HI(x) = hx|(i^0x3ff00000); /* normalize x or x/2 */ | |
133 k += (i>>20); | |
134 f = x-1.0; | |
135 if((0x000fffff&(2+hx))<3) { /* |f| < 2**-20 */ | |
136 if(f==zero) { | |
137 if (k==0) return zero; | |
138 else {dk=(double)k; return dk*ln2_hi+dk*ln2_lo;} | |
139 } | |
140 R = f*f*(0.5-0.33333333333333333*f); | |
141 if(k==0) return f-R; else {dk=(double)k; | |
142 return dk*ln2_hi-((R-dk*ln2_lo)-f);} | |
143 } | |
144 s = f/(2.0+f); | |
145 dk = (double)k; | |
146 z = s*s; | |
147 i = hx-0x6147a; | |
148 w = z*z; | |
149 j = 0x6b851-hx; | |
150 t1= w*(Lg2+w*(Lg4+w*Lg6)); | |
151 t2= z*(Lg1+w*(Lg3+w*(Lg5+w*Lg7))); | |
152 i |= j; | |
153 R = t2+t1; | |
154 if(i>0) { | |
155 hfsq=0.5*f*f; | |
156 if(k==0) return f-(hfsq-s*(hfsq+R)); else | |
157 return dk*ln2_hi-((hfsq-(s*(hfsq+R)+dk*ln2_lo))-f); | |
158 } else { | |
159 if(k==0) return f-s*(f-R); else | |
160 return dk*ln2_hi-((s*(f-R)-dk*ln2_lo)-f); | |
161 } | |
162 } | |
163 | |
164 JRT_LEAF(jdouble, SharedRuntime::dlog(jdouble x)) | |
165 return __ieee754_log(x); | |
166 JRT_END | |
167 | |
168 /* __ieee754_log10(x) | |
169 * Return the base 10 logarithm of x | |
170 * | |
171 * Method : | |
172 * Let log10_2hi = leading 40 bits of log10(2) and | |
173 * log10_2lo = log10(2) - log10_2hi, | |
174 * ivln10 = 1/log(10) rounded. | |
175 * Then | |
176 * n = ilogb(x), | |
177 * if(n<0) n = n+1; | |
178 * x = scalbn(x,-n); | |
179 * log10(x) := n*log10_2hi + (n*log10_2lo + ivln10*log(x)) | |
180 * | |
181 * Note 1: | |
182 * To guarantee log10(10**n)=n, where 10**n is normal, the rounding | |
183 * mode must set to Round-to-Nearest. | |
184 * Note 2: | |
185 * [1/log(10)] rounded to 53 bits has error .198 ulps; | |
186 * log10 is monotonic at all binary break points. | |
187 * | |
188 * Special cases: | |
189 * log10(x) is NaN with signal if x < 0; | |
190 * log10(+INF) is +INF with no signal; log10(0) is -INF with signal; | |
191 * log10(NaN) is that NaN with no signal; | |
192 * log10(10**N) = N for N=0,1,...,22. | |
193 * | |
194 * Constants: | |
195 * The hexadecimal values are the intended ones for the following constants. | |
196 * The decimal values may be used, provided that the compiler will convert | |
197 * from decimal to binary accurately enough to produce the hexadecimal values | |
198 * shown. | |
199 */ | |
200 | |
201 static const double | |
202 ivln10 = 4.34294481903251816668e-01, /* 0x3FDBCB7B, 0x1526E50E */ | |
203 log10_2hi = 3.01029995663611771306e-01, /* 0x3FD34413, 0x509F6000 */ | |
204 log10_2lo = 3.69423907715893078616e-13; /* 0x3D59FEF3, 0x11F12B36 */ | |
205 | |
206 static double __ieee754_log10(double x) { | |
207 double y,z; | |
208 int i,k,hx; | |
209 unsigned lx; | |
210 | |
211 hx = __HI(x); /* high word of x */ | |
212 lx = __LO(x); /* low word of x */ | |
213 | |
214 k=0; | |
215 if (hx < 0x00100000) { /* x < 2**-1022 */ | |
216 if (((hx&0x7fffffff)|lx)==0) | |
217 return -two54/zero; /* log(+-0)=-inf */ | |
218 if (hx<0) return (x-x)/zero; /* log(-#) = NaN */ | |
219 k -= 54; x *= two54; /* subnormal number, scale up x */ | |
220 hx = __HI(x); /* high word of x */ | |
221 } | |
222 if (hx >= 0x7ff00000) return x+x; | |
223 k += (hx>>20)-1023; | |
224 i = ((unsigned)k&0x80000000)>>31; | |
225 hx = (hx&0x000fffff)|((0x3ff-i)<<20); | |
226 y = (double)(k+i); | |
227 __HI(x) = hx; | |
228 z = y*log10_2lo + ivln10*__ieee754_log(x); | |
229 return z+y*log10_2hi; | |
230 } | |
231 | |
232 JRT_LEAF(jdouble, SharedRuntime::dlog10(jdouble x)) | |
233 return __ieee754_log10(x); | |
234 JRT_END | |
235 | |
236 | |
237 /* __ieee754_exp(x) | |
238 * Returns the exponential of x. | |
239 * | |
240 * Method | |
241 * 1. Argument reduction: | |
242 * Reduce x to an r so that |r| <= 0.5*ln2 ~ 0.34658. | |
243 * Given x, find r and integer k such that | |
244 * | |
245 * x = k*ln2 + r, |r| <= 0.5*ln2. | |
246 * | |
247 * Here r will be represented as r = hi-lo for better | |
248 * accuracy. | |
249 * | |
250 * 2. Approximation of exp(r) by a special rational function on | |
251 * the interval [0,0.34658]: | |
252 * Write | |
253 * R(r**2) = r*(exp(r)+1)/(exp(r)-1) = 2 + r*r/6 - r**4/360 + ... | |
254 * We use a special Reme algorithm on [0,0.34658] to generate | |
255 * a polynomial of degree 5 to approximate R. The maximum error | |
256 * of this polynomial approximation is bounded by 2**-59. In | |
257 * other words, | |
258 * R(z) ~ 2.0 + P1*z + P2*z**2 + P3*z**3 + P4*z**4 + P5*z**5 | |
259 * (where z=r*r, and the values of P1 to P5 are listed below) | |
260 * and | |
261 * | 5 | -59 | |
262 * | 2.0+P1*z+...+P5*z - R(z) | <= 2 | |
263 * | | | |
264 * The computation of exp(r) thus becomes | |
265 * 2*r | |
266 * exp(r) = 1 + ------- | |
267 * R - r | |
268 * r*R1(r) | |
269 * = 1 + r + ----------- (for better accuracy) | |
270 * 2 - R1(r) | |
271 * where | |
272 * 2 4 10 | |
273 * R1(r) = r - (P1*r + P2*r + ... + P5*r ). | |
274 * | |
275 * 3. Scale back to obtain exp(x): | |
276 * From step 1, we have | |
277 * exp(x) = 2^k * exp(r) | |
278 * | |
279 * Special cases: | |
280 * exp(INF) is INF, exp(NaN) is NaN; | |
281 * exp(-INF) is 0, and | |
282 * for finite argument, only exp(0)=1 is exact. | |
283 * | |
284 * Accuracy: | |
285 * according to an error analysis, the error is always less than | |
286 * 1 ulp (unit in the last place). | |
287 * | |
288 * Misc. info. | |
289 * For IEEE double | |
290 * if x > 7.09782712893383973096e+02 then exp(x) overflow | |
291 * if x < -7.45133219101941108420e+02 then exp(x) underflow | |
292 * | |
293 * Constants: | |
294 * The hexadecimal values are the intended ones for the following | |
295 * constants. The decimal values may be used, provided that the | |
296 * compiler will convert from decimal to binary accurately enough | |
297 * to produce the hexadecimal values shown. | |
298 */ | |
299 | |
300 static const double | |
301 one = 1.0, | |
302 halF[2] = {0.5,-0.5,}, | |
303 twom1000= 9.33263618503218878990e-302, /* 2**-1000=0x01700000,0*/ | |
304 o_threshold= 7.09782712893383973096e+02, /* 0x40862E42, 0xFEFA39EF */ | |
305 u_threshold= -7.45133219101941108420e+02, /* 0xc0874910, 0xD52D3051 */ | |
306 ln2HI[2] ={ 6.93147180369123816490e-01, /* 0x3fe62e42, 0xfee00000 */ | |
307 -6.93147180369123816490e-01,},/* 0xbfe62e42, 0xfee00000 */ | |
308 ln2LO[2] ={ 1.90821492927058770002e-10, /* 0x3dea39ef, 0x35793c76 */ | |
309 -1.90821492927058770002e-10,},/* 0xbdea39ef, 0x35793c76 */ | |
310 invln2 = 1.44269504088896338700e+00, /* 0x3ff71547, 0x652b82fe */ | |
311 P1 = 1.66666666666666019037e-01, /* 0x3FC55555, 0x5555553E */ | |
312 P2 = -2.77777777770155933842e-03, /* 0xBF66C16C, 0x16BEBD93 */ | |
313 P3 = 6.61375632143793436117e-05, /* 0x3F11566A, 0xAF25DE2C */ | |
314 P4 = -1.65339022054652515390e-06, /* 0xBEBBBD41, 0xC5D26BF1 */ | |
315 P5 = 4.13813679705723846039e-08; /* 0x3E663769, 0x72BEA4D0 */ | |
316 | |
317 static double __ieee754_exp(double x) { | |
318 double y,hi=0,lo=0,c,t; | |
319 int k=0,xsb; | |
320 unsigned hx; | |
321 | |
322 hx = __HI(x); /* high word of x */ | |
323 xsb = (hx>>31)&1; /* sign bit of x */ | |
324 hx &= 0x7fffffff; /* high word of |x| */ | |
325 | |
326 /* filter out non-finite argument */ | |
327 if(hx >= 0x40862E42) { /* if |x|>=709.78... */ | |
328 if(hx>=0x7ff00000) { | |
329 if(((hx&0xfffff)|__LO(x))!=0) | |
330 return x+x; /* NaN */ | |
331 else return (xsb==0)? x:0.0; /* exp(+-inf)={inf,0} */ | |
332 } | |
333 if(x > o_threshold) return hugeX*hugeX; /* overflow */ | |
334 if(x < u_threshold) return twom1000*twom1000; /* underflow */ | |
335 } | |
336 | |
337 /* argument reduction */ | |
338 if(hx > 0x3fd62e42) { /* if |x| > 0.5 ln2 */ | |
339 if(hx < 0x3FF0A2B2) { /* and |x| < 1.5 ln2 */ | |
340 hi = x-ln2HI[xsb]; lo=ln2LO[xsb]; k = 1-xsb-xsb; | |
341 } else { | |
342 k = (int)(invln2*x+halF[xsb]); | |
343 t = k; | |
344 hi = x - t*ln2HI[0]; /* t*ln2HI is exact here */ | |
345 lo = t*ln2LO[0]; | |
346 } | |
347 x = hi - lo; | |
348 } | |
349 else if(hx < 0x3e300000) { /* when |x|<2**-28 */ | |
350 if(hugeX+x>one) return one+x;/* trigger inexact */ | |
351 } | |
352 else k = 0; | |
353 | |
354 /* x is now in primary range */ | |
355 t = x*x; | |
356 c = x - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); | |
357 if(k==0) return one-((x*c)/(c-2.0)-x); | |
358 else y = one-((lo-(x*c)/(2.0-c))-hi); | |
359 if(k >= -1021) { | |
360 __HI(y) += (k<<20); /* add k to y's exponent */ | |
361 return y; | |
362 } else { | |
363 __HI(y) += ((k+1000)<<20);/* add k to y's exponent */ | |
364 return y*twom1000; | |
365 } | |
366 } | |
367 | |
368 JRT_LEAF(jdouble, SharedRuntime::dexp(jdouble x)) | |
369 return __ieee754_exp(x); | |
370 JRT_END | |
371 | |
372 /* __ieee754_pow(x,y) return x**y | |
373 * | |
374 * n | |
375 * Method: Let x = 2 * (1+f) | |
376 * 1. Compute and return log2(x) in two pieces: | |
377 * log2(x) = w1 + w2, | |
378 * where w1 has 53-24 = 29 bit trailing zeros. | |
379 * 2. Perform y*log2(x) = n+y' by simulating muti-precision | |
380 * arithmetic, where |y'|<=0.5. | |
381 * 3. Return x**y = 2**n*exp(y'*log2) | |
382 * | |
383 * Special cases: | |
384 * 1. (anything) ** 0 is 1 | |
385 * 2. (anything) ** 1 is itself | |
386 * 3. (anything) ** NAN is NAN | |
387 * 4. NAN ** (anything except 0) is NAN | |
388 * 5. +-(|x| > 1) ** +INF is +INF | |
389 * 6. +-(|x| > 1) ** -INF is +0 | |
390 * 7. +-(|x| < 1) ** +INF is +0 | |
391 * 8. +-(|x| < 1) ** -INF is +INF | |
392 * 9. +-1 ** +-INF is NAN | |
393 * 10. +0 ** (+anything except 0, NAN) is +0 | |
394 * 11. -0 ** (+anything except 0, NAN, odd integer) is +0 | |
395 * 12. +0 ** (-anything except 0, NAN) is +INF | |
396 * 13. -0 ** (-anything except 0, NAN, odd integer) is +INF | |
397 * 14. -0 ** (odd integer) = -( +0 ** (odd integer) ) | |
398 * 15. +INF ** (+anything except 0,NAN) is +INF | |
399 * 16. +INF ** (-anything except 0,NAN) is +0 | |
400 * 17. -INF ** (anything) = -0 ** (-anything) | |
401 * 18. (-anything) ** (integer) is (-1)**(integer)*(+anything**integer) | |
402 * 19. (-anything except 0 and inf) ** (non-integer) is NAN | |
403 * | |
404 * Accuracy: | |
405 * pow(x,y) returns x**y nearly rounded. In particular | |
406 * pow(integer,integer) | |
407 * always returns the correct integer provided it is | |
408 * representable. | |
409 * | |
410 * Constants : | |
411 * The hexadecimal values are the intended ones for the following | |
412 * constants. The decimal values may be used, provided that the | |
413 * compiler will convert from decimal to binary accurately enough | |
414 * to produce the hexadecimal values shown. | |
415 */ | |
416 | |
417 static const double | |
418 bp[] = {1.0, 1.5,}, | |
419 dp_h[] = { 0.0, 5.84962487220764160156e-01,}, /* 0x3FE2B803, 0x40000000 */ | |
420 dp_l[] = { 0.0, 1.35003920212974897128e-08,}, /* 0x3E4CFDEB, 0x43CFD006 */ | |
421 zeroX = 0.0, | |
422 two = 2.0, | |
423 two53 = 9007199254740992.0, /* 0x43400000, 0x00000000 */ | |
424 /* poly coefs for (3/2)*(log(x)-2s-2/3*s**3 */ | |
425 L1X = 5.99999999999994648725e-01, /* 0x3FE33333, 0x33333303 */ | |
426 L2X = 4.28571428578550184252e-01, /* 0x3FDB6DB6, 0xDB6FABFF */ | |
427 L3X = 3.33333329818377432918e-01, /* 0x3FD55555, 0x518F264D */ | |
428 L4X = 2.72728123808534006489e-01, /* 0x3FD17460, 0xA91D4101 */ | |
429 L5X = 2.30660745775561754067e-01, /* 0x3FCD864A, 0x93C9DB65 */ | |
430 L6X = 2.06975017800338417784e-01, /* 0x3FCA7E28, 0x4A454EEF */ | |
431 lg2 = 6.93147180559945286227e-01, /* 0x3FE62E42, 0xFEFA39EF */ | |
432 lg2_h = 6.93147182464599609375e-01, /* 0x3FE62E43, 0x00000000 */ | |
433 lg2_l = -1.90465429995776804525e-09, /* 0xBE205C61, 0x0CA86C39 */ | |
434 ovt = 8.0085662595372944372e-0017, /* -(1024-log2(ovfl+.5ulp)) */ | |
435 cp = 9.61796693925975554329e-01, /* 0x3FEEC709, 0xDC3A03FD =2/(3ln2) */ | |
436 cp_h = 9.61796700954437255859e-01, /* 0x3FEEC709, 0xE0000000 =(float)cp */ | |
437 cp_l = -7.02846165095275826516e-09, /* 0xBE3E2FE0, 0x145B01F5 =tail of cp_h*/ | |
438 ivln2 = 1.44269504088896338700e+00, /* 0x3FF71547, 0x652B82FE =1/ln2 */ | |
439 ivln2_h = 1.44269502162933349609e+00, /* 0x3FF71547, 0x60000000 =24b 1/ln2*/ | |
440 ivln2_l = 1.92596299112661746887e-08; /* 0x3E54AE0B, 0xF85DDF44 =1/ln2 tail*/ | |
441 | |
442 double __ieee754_pow(double x, double y) { | |
443 double z,ax,z_h,z_l,p_h,p_l; | |
444 double y1,t1,t2,r,s,t,u,v,w; | |
445 int i0,i1,i,j,k,yisint,n; | |
446 int hx,hy,ix,iy; | |
447 unsigned lx,ly; | |
448 | |
449 i0 = ((*(int*)&one)>>29)^1; i1=1-i0; | |
450 hx = __HI(x); lx = __LO(x); | |
451 hy = __HI(y); ly = __LO(y); | |
452 ix = hx&0x7fffffff; iy = hy&0x7fffffff; | |
453 | |
454 /* y==zero: x**0 = 1 */ | |
455 if((iy|ly)==0) return one; | |
456 | |
457 /* +-NaN return x+y */ | |
458 if(ix > 0x7ff00000 || ((ix==0x7ff00000)&&(lx!=0)) || | |
459 iy > 0x7ff00000 || ((iy==0x7ff00000)&&(ly!=0))) | |
460 return x+y; | |
461 | |
462 /* determine if y is an odd int when x < 0 | |
463 * yisint = 0 ... y is not an integer | |
464 * yisint = 1 ... y is an odd int | |
465 * yisint = 2 ... y is an even int | |
466 */ | |
467 yisint = 0; | |
468 if(hx<0) { | |
469 if(iy>=0x43400000) yisint = 2; /* even integer y */ | |
470 else if(iy>=0x3ff00000) { | |
471 k = (iy>>20)-0x3ff; /* exponent */ | |
472 if(k>20) { | |
473 j = ly>>(52-k); | |
474 if((unsigned)(j<<(52-k))==ly) yisint = 2-(j&1); | |
475 } else if(ly==0) { | |
476 j = iy>>(20-k); | |
477 if((j<<(20-k))==iy) yisint = 2-(j&1); | |
478 } | |
479 } | |
480 } | |
481 | |
482 /* special value of y */ | |
483 if(ly==0) { | |
484 if (iy==0x7ff00000) { /* y is +-inf */ | |
485 if(((ix-0x3ff00000)|lx)==0) | |
486 return y - y; /* inf**+-1 is NaN */ | |
487 else if (ix >= 0x3ff00000)/* (|x|>1)**+-inf = inf,0 */ | |
488 return (hy>=0)? y: zeroX; | |
489 else /* (|x|<1)**-,+inf = inf,0 */ | |
490 return (hy<0)?-y: zeroX; | |
491 } | |
492 if(iy==0x3ff00000) { /* y is +-1 */ | |
493 if(hy<0) return one/x; else return x; | |
494 } | |
495 if(hy==0x40000000) return x*x; /* y is 2 */ | |
496 if(hy==0x3fe00000) { /* y is 0.5 */ | |
497 if(hx>=0) /* x >= +0 */ | |
498 return sqrt(x); | |
499 } | |
500 } | |
501 | |
502 ax = fabsd(x); | |
503 /* special value of x */ | |
504 if(lx==0) { | |
505 if(ix==0x7ff00000||ix==0||ix==0x3ff00000){ | |
506 z = ax; /*x is +-0,+-inf,+-1*/ | |
507 if(hy<0) z = one/z; /* z = (1/|x|) */ | |
508 if(hx<0) { | |
509 if(((ix-0x3ff00000)|yisint)==0) { | |
1681
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
510 #ifdef CAN_USE_NAN_DEFINE |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
511 z = NAN; |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
512 #else |
0 | 513 z = (z-z)/(z-z); /* (-1)**non-int is NaN */ |
1681
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
514 #endif |
0 | 515 } else if(yisint==1) |
516 z = -1.0*z; /* (x<0)**odd = -(|x|**odd) */ | |
517 } | |
518 return z; | |
519 } | |
520 } | |
521 | |
522 n = (hx>>31)+1; | |
523 | |
524 /* (x<0)**(non-int) is NaN */ | |
1681
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
525 if((n|yisint)==0) |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
526 #ifdef CAN_USE_NAN_DEFINE |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
527 return NAN; |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
528 #else |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
529 return (x-x)/(x-x); |
126ea7725993
6953477: Increase portability and flexibility of building Hotspot
bobv
parents:
1552
diff
changeset
|
530 #endif |
0 | 531 |
532 s = one; /* s (sign of result -ve**odd) = -1 else = 1 */ | |
533 if((n|(yisint-1))==0) s = -one;/* (-ve)**(odd int) */ | |
534 | |
535 /* |y| is huge */ | |
536 if(iy>0x41e00000) { /* if |y| > 2**31 */ | |
537 if(iy>0x43f00000){ /* if |y| > 2**64, must o/uflow */ | |
538 if(ix<=0x3fefffff) return (hy<0)? hugeX*hugeX:tiny*tiny; | |
539 if(ix>=0x3ff00000) return (hy>0)? hugeX*hugeX:tiny*tiny; | |
540 } | |
541 /* over/underflow if x is not close to one */ | |
542 if(ix<0x3fefffff) return (hy<0)? s*hugeX*hugeX:s*tiny*tiny; | |
543 if(ix>0x3ff00000) return (hy>0)? s*hugeX*hugeX:s*tiny*tiny; | |
544 /* now |1-x| is tiny <= 2**-20, suffice to compute | |
545 log(x) by x-x^2/2+x^3/3-x^4/4 */ | |
546 t = ax-one; /* t has 20 trailing zeros */ | |
547 w = (t*t)*(0.5-t*(0.3333333333333333333333-t*0.25)); | |
548 u = ivln2_h*t; /* ivln2_h has 21 sig. bits */ | |
549 v = t*ivln2_l-w*ivln2; | |
550 t1 = u+v; | |
551 __LO(t1) = 0; | |
552 t2 = v-(t1-u); | |
553 } else { | |
554 double ss,s2,s_h,s_l,t_h,t_l; | |
555 n = 0; | |
556 /* take care subnormal number */ | |
557 if(ix<0x00100000) | |
558 {ax *= two53; n -= 53; ix = __HI(ax); } | |
559 n += ((ix)>>20)-0x3ff; | |
560 j = ix&0x000fffff; | |
561 /* determine interval */ | |
562 ix = j|0x3ff00000; /* normalize ix */ | |
563 if(j<=0x3988E) k=0; /* |x|<sqrt(3/2) */ | |
564 else if(j<0xBB67A) k=1; /* |x|<sqrt(3) */ | |
565 else {k=0;n+=1;ix -= 0x00100000;} | |
566 __HI(ax) = ix; | |
567 | |
568 /* compute ss = s_h+s_l = (x-1)/(x+1) or (x-1.5)/(x+1.5) */ | |
569 u = ax-bp[k]; /* bp[0]=1.0, bp[1]=1.5 */ | |
570 v = one/(ax+bp[k]); | |
571 ss = u*v; | |
572 s_h = ss; | |
573 __LO(s_h) = 0; | |
574 /* t_h=ax+bp[k] High */ | |
575 t_h = zeroX; | |
576 __HI(t_h)=((ix>>1)|0x20000000)+0x00080000+(k<<18); | |
577 t_l = ax - (t_h-bp[k]); | |
578 s_l = v*((u-s_h*t_h)-s_h*t_l); | |
579 /* compute log(ax) */ | |
580 s2 = ss*ss; | |
581 r = s2*s2*(L1X+s2*(L2X+s2*(L3X+s2*(L4X+s2*(L5X+s2*L6X))))); | |
582 r += s_l*(s_h+ss); | |
583 s2 = s_h*s_h; | |
584 t_h = 3.0+s2+r; | |
585 __LO(t_h) = 0; | |
586 t_l = r-((t_h-3.0)-s2); | |
587 /* u+v = ss*(1+...) */ | |
588 u = s_h*t_h; | |
589 v = s_l*t_h+t_l*ss; | |
590 /* 2/(3log2)*(ss+...) */ | |
591 p_h = u+v; | |
592 __LO(p_h) = 0; | |
593 p_l = v-(p_h-u); | |
594 z_h = cp_h*p_h; /* cp_h+cp_l = 2/(3*log2) */ | |
595 z_l = cp_l*p_h+p_l*cp+dp_l[k]; | |
596 /* log2(ax) = (ss+..)*2/(3*log2) = n + dp_h + z_h + z_l */ | |
597 t = (double)n; | |
598 t1 = (((z_h+z_l)+dp_h[k])+t); | |
599 __LO(t1) = 0; | |
600 t2 = z_l-(((t1-t)-dp_h[k])-z_h); | |
601 } | |
602 | |
603 /* split up y into y1+y2 and compute (y1+y2)*(t1+t2) */ | |
604 y1 = y; | |
605 __LO(y1) = 0; | |
606 p_l = (y-y1)*t1+y*t2; | |
607 p_h = y1*t1; | |
608 z = p_l+p_h; | |
609 j = __HI(z); | |
610 i = __LO(z); | |
611 if (j>=0x40900000) { /* z >= 1024 */ | |
612 if(((j-0x40900000)|i)!=0) /* if z > 1024 */ | |
613 return s*hugeX*hugeX; /* overflow */ | |
614 else { | |
615 if(p_l+ovt>z-p_h) return s*hugeX*hugeX; /* overflow */ | |
616 } | |
617 } else if((j&0x7fffffff)>=0x4090cc00 ) { /* z <= -1075 */ | |
618 if(((j-0xc090cc00)|i)!=0) /* z < -1075 */ | |
619 return s*tiny*tiny; /* underflow */ | |
620 else { | |
621 if(p_l<=z-p_h) return s*tiny*tiny; /* underflow */ | |
622 } | |
623 } | |
624 /* | |
625 * compute 2**(p_h+p_l) | |
626 */ | |
627 i = j&0x7fffffff; | |
628 k = (i>>20)-0x3ff; | |
629 n = 0; | |
630 if(i>0x3fe00000) { /* if |z| > 0.5, set n = [z+0.5] */ | |
631 n = j+(0x00100000>>(k+1)); | |
632 k = ((n&0x7fffffff)>>20)-0x3ff; /* new k for n */ | |
633 t = zeroX; | |
634 __HI(t) = (n&~(0x000fffff>>k)); | |
635 n = ((n&0x000fffff)|0x00100000)>>(20-k); | |
636 if(j<0) n = -n; | |
637 p_h -= t; | |
638 } | |
639 t = p_l+p_h; | |
640 __LO(t) = 0; | |
641 u = t*lg2_h; | |
642 v = (p_l-(t-p_h))*lg2+t*lg2_l; | |
643 z = u+v; | |
644 w = v-(z-u); | |
645 t = z*z; | |
646 t1 = z - t*(P1+t*(P2+t*(P3+t*(P4+t*P5)))); | |
647 r = (z*t1)/(t1-two)-(w+z*w); | |
648 z = one-(r-z); | |
649 j = __HI(z); | |
650 j += (n<<20); | |
20286
631c3a4ea10c
8043301: Duplicate definitions in vm/runtime/sharedRuntimeTrans.cpp versus math.h in VS2013
lfoltan
parents:
14411
diff
changeset
|
651 if((j>>20)<=0) z = scalbnA(z,n); /* subnormal output */ |
0 | 652 else __HI(z) += (n<<20); |
653 return s*z; | |
654 } | |
655 | |
656 | |
657 JRT_LEAF(jdouble, SharedRuntime::dpow(jdouble x, jdouble y)) | |
658 return __ieee754_pow(x, y); | |
659 JRT_END | |
660 | |
661 #ifdef WIN32 | |
662 # pragma optimize ( "", on ) | |
663 #endif |