xref: /illumos-gate/usr/src/lib/libm/common/m9x/tgamma.c (revision 843c398e8904ed9d833d2af3103894f909fb4b52)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
24  */
25 /*
26  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #pragma weak __tgamma = tgamma
31 
32 /* INDENT OFF */
33 /*
34  * True gamma function
35  * double tgamma(double x)
36  *
37  * Error:
38  * ------
39  *  	Less that one ulp for both positive and negative arguments.
40  *
41  * Algorithm:
42  * ---------
43  *	A: For negative argument
44  *		(1) gamma(-n or -inf) is NaN
45  *		(2) Underflow Threshold
46  *		(3) Reduction to gamma(1+x)
47  *	B: For x between 1 and 2
48  * 	C: For x between 0 and 1
49  *	D: For x between 2 and 8
50  *	E: Overflow thresold {see over.c}
51  *	F: For overflow_threshold >= x >= 8
52  *
53  * Implementation details
54  * -----------------------
55  *							-pi
56  * (A) For negative argument, use gamma(-x) = ------------------------.
57  *                                            (sin(pi*x)*gamma(1+x))
58  *
59  *   (1) gamma(-n or -inf) is NaN with invalid signal by SUSv3 spec.
60  *	 (Ideally, gamma(-n) = 1/sinpi(n) = (-1)**(n+1) * inf.)
61  *
62  *   (2) Underflow Threshold. For each precision, there is a value T
63  *	such that when x>T and when x is not an integer, gamma(-x) will
64  *       always underflow. A table of the underflow threshold value is given
65  *	below. For proof, see file "under.c".
66  *
67  *	Precision	underflow threshold T =
68  *	----------------------------------------------------------------------
69  *	single	41.000041962					= 41  + 11 ULP
70  *		(machine format) 4224000B
71  *	double	183.000000000000312639				= 183 + 11 ULP
72  *		(machine format) 4066E000 0000000B
73  *	quad	1774.0000000000000000000000000000017749370	= 1774 + 9 ULP
74  *		(machine format) 4009BB80000000000000000000000009
75  *	----------------------------------------------------------------------
76  *
77  *   (3) Reduction to gamma(1+x).
78  *	Because of (1) and (2), we need only consider non-integral x
79  *	such that 0<x<T. Let k = [x] and z = x-[x]. Define
80  *                  sin(x*pi)                cos(x*pi)
81  *	kpsin(x) = --------- and kpcos(x) = --------- . Then
82  *                     pi                       pi
83  *                                    1
84  *		gamma(-x) = --------------------.
85  *		            -kpsin(x)*gamma(1+x)
86  *	Since x = k+z,
87  *                                                  k+1
88  *		-sin(x*pi) = -sin(k*pi+z*pi) = (-1)   *sin(z*pi),
89  *                               k+1
90  *	we have -kpsin(x) = (-1)   * kpsin(z).  We can further
91  *	reduce z to t by
92  *	   (I)   t = z	     when 0.00000     <= z < 0.31830...
93  *	   (II)  t = 0.5-z   when 0.31830...  <= z < 0.681690...
94  *	   (III) t = 1-z     when 0.681690... <= z < 1.00000
95  *	and correspondingly
96  *	   (I)   kpsin(z) = kpsin(t)  	... 0<= z < 0.3184
97  *	   (II)  kpsin(z) = kpcos(t) 	... |t|   < 0.182
98  *	   (III) kpsin(z) = kpsin(t) 	... 0<= t < 0.3184
99  *
100  *	Using a special Remez algorithm, we obtain the following polynomial
101  *	approximation for kpsin(t) for 0<=t<0.3184:
102  *
103  *	Computation note: in simulating higher precision arithmetic, kcpsin
104  *	return head = t and tail = ks[0]*t^3 + (...) to maintain extra bits.
105  *
106  *	Quad precision, remez error <= 2**(-129.74)
107  *                                   3            5                   27
108  *	    kpsin(t) = t + ks[0] * t  + ks[1] * t  + ... + ks[12] * t
109  *
110  *       ks[ 0] =  -1.64493406684822643647241516664602518705158902870e+0000
111  *       ks[ 1] =   8.11742425283353643637002772405874238094995726160e-0001
112  *       ks[ 2] =  -1.90751824122084213696472111835337366232282723933e-0001
113  *       ks[ 3] =   2.61478478176548005046532613563241288115395517084e-0002
114  *       ks[ 4] =  -2.34608103545582363750893072647117829448016479971e-0003
115  *       ks[ 5] =   1.48428793031071003684606647212534027556262040158e-0004
116  *       ks[ 6] =  -6.97587366165638046518462722252768122615952898698e-0006
117  *       ks[ 7] =   2.53121740413702536928659271747187500934840057929e-0007
118  *       ks[ 8] =  -7.30471182221385990397683641695766121301933621956e-0009
119  *       ks[ 9] =   1.71653847451163495739958249695549313987973589884e-0010
120  *       ks[10] =  -3.34813314714560776122245796929054813458341420565e-0012
121  *       ks[11] =   5.50724992262622033449487808306969135431411753047e-0014
122  *       ks[12] =  -7.67678132753577998601234393215802221104236979928e-0016
123  *
124  *	Double precision, Remez error <= 2**(-62.9)
125  *                                  3            5                  15
126  *	    kpsin(t) = t + ks[0] * t  + ks[1] * t  + ... + ks[6] * t
127  *
128  *       ks[0] =  -1.644934066848226406065691	(0x3ffa51a6 625307d3)
129  *       ks[1] =   8.11742425283341655883668741874008920850698590621e-0001
130  *       ks[2] =  -1.90751824120862873825597279118304943994042258291e-0001
131  *       ks[3] =   2.61478477632554278317289628332654539353521911570e-0002
132  *       ks[4] =  -2.34607978510202710377617190278735525354347705866e-0003
133  *       ks[5] =   1.48413292290051695897242899977121846763824221705e-0004
134  *       ks[6] =  -6.87730769637543488108688726777687262485357072242e-0006
135  *
136  *	Single precision, Remez error <= 2**(-34.09)
137  *                                  3            5                  9
138  *	    kpsin(t) = t + ks[0] * t  + ks[1] * t  + ... + ks[3] * t
139  *
140  *       ks[0] =  -1.64493404985645811354476665052005342839447790544e+0000
141  *       ks[1] =   8.11740794458351064092797249069438269367389272270e-0001
142  *       ks[2] =  -1.90703144603551216933075809162889536878854055202e-0001
143  *       ks[3] =   2.55742333994264563281155312271481108635575331201e-0002
144  *
145  *	Computation note: in simulating higher precision arithmetic, kcpsin
146  *	return head = t and tail = kc[0]*t^3 + (...) to maintain extra bits
147  *   	precision.
148  *
149  *	And for kpcos(t) for |t|< 0.183:
150  *
151  *	Quad precision, remez <= 2**(-122.48)
152  *                                     2            4                  22
153  *	    kpcos(t) = 1/pi +  pi/2 * t  + kc[2] * t + ... + kc[11] * t
154  *
155  *       kc[2] =   1.29192819501249250731151312779548918765320728489e+0000
156  *       kc[3] =  -4.25027339979557573976029596929319207009444090366e-0001
157  *       kc[4] =   7.49080661650990096109672954618317623888421628613e-0002
158  *       kc[5] =  -8.21458866111282287985539464173976555436050215120e-0003
159  *       kc[6] =   6.14202578809529228503205255165761204750211603402e-0004
160  *       kc[7] =  -3.33073432691149607007217330302595267179545908740e-0005
161  *       kc[8] =   1.36970959047832085796809745461530865597993680204e-0006
162  *       kc[9] =  -4.41780774262583514450246512727201806217271097336e-0008
163  *       kc[10]=   1.14741409212381858820016567664488123478660705759e-0009
164  *       kc[11]=  -2.44261236114707374558437500654381006300502749632e-0011
165  *
166  *	Double precision, remez < 2**(61.91)
167  *                                   2            4                  12
168  *	    kpcos(t) = 1/pi + pi/2 *t +  kc[2] * t  + ... + kc[6] * t
169  *
170  *       kc[2] =   1.29192819501230224953283586722575766189551966008e+0000
171  *       kc[3] =  -4.25027339940149518500158850753393173519732149213e-0001
172  *       kc[4] =   7.49080625187015312373925142219429422375556727752e-0002
173  *       kc[5] =  -8.21442040906099210866977352284054849051348692715e-0003
174  *       kc[6] =   6.10411356829515414575566564733632532333904115968e-0004
175  *
176  *	Single precision, remez < 2**(-30.13)
177  *                                       2                  6
178  *	    kpcos(t) = kc[0] +  kc[1] * t  + ... + kc[3] * t
179  *
180  *       kc[0] =   3.18309886183790671537767526745028724068919291480e-0001
181  *       kc[1] =  -1.57079581447762568199467875065854538626594937791e+0000
182  *       kc[2] =   1.29183528092558692844073004029568674027807393862e+0000
183  *       kc[3] =  -4.20232949771307685981015914425195471602739075537e-0001
184  *
185  *	Computation note: in simulating higher precision arithmetic, kcpcos
186  *	return head = 1/pi chopped, and tail = pi/2 *t^2 + (tail part of 1/pi
187  *	+ ...) to maintain extra bits precision. In particular, pi/2 * t^2
188  *	is calculated with great care.
189  *
190  *	Thus, the computation of gamma(-x), x>0, is:
191  *	Let k = int(x), z = x-k.
192  *	For z in (I)
193  *                                    k+1
194  *			          (-1)
195  * 		gamma(-x) = ------------------- ;
196  *		            kpsin(z)*gamma(1+x)
197  *
198  *	otherwise, for z in (II),
199  *                                      k+1
200  *			            (-1)
201  * 		gamma(-x) = ----------------------- ;
202  *			    kpcos(0.5-z)*gamma(1+x)
203  *
204  *	otherwise, for z in (III),
205  *                                      k+1
206  *			            (-1)
207  * 		gamma(-x) = --------------------- .
208  *		            kpsin(1-z)*gamma(1+x)
209  *
210  *	Thus, the computation of gamma(-x) reduced to the computation of
211  *	gamma(1+x) and kpsin(), kpcos().
212  *
213  * (B) For x between 1 and 2.  We break [1,2] into three parts:
214  *	GT1 = [1.0000, 1.2845]
215  * 	GT2 = [1.2844, 1.6374]
216  * 	GT3 = [1.6373, 2.0000]
217  *
218  *    For x in GTi, i=1,2,3, let
219  * 	z1  =  1.134861805732790769689793935774652917006
220  *	gz1 = gamma(z1)  =   0.9382046279096824494097535615803269576988
221  *	tz1 = gamma'(z1) =  -0.3517214357852935791015625000000000000000
222  *
223  *	z2  =  1.461632144968362341262659542325721328468e+0000
224  *	gz2 = gamma(z2)  = 0.8856031944108887002788159005825887332080
225  *	tz2 = gamma'(z2) = 0.00
226  *
227  *	z3  =  1.819773101100500601787868704921606996312e+0000
228  *	gz3 = gamma(z3)  = 0.9367814114636523216188468970808378497426
229  *	tz3 = gamma'(z3) = 0.2805306315422058105468750000000000000000
230  *
231  *    and
232  *	y = x-zi	... for extra precision, write y = y.h + y.l
233  *    Then
234  *	gamma(x) = gzi + tzi*(y.h+y.l) + y*y*Ri(y),
235  *		 = gzi.h + (tzi*y.h + ((tzi*y.l+gzi.l) +  y*y*Ri(y)))
236  *		 = gy.h + gy.l
237  *    where
238  *	(I) For double precision
239  *
240  *		Ri(y) = Pi(y)/Qi(y), i=1,2,3;
241  *
242  *		P1(y) = p1[0] + p1[1]*y + ... + p1[4]*y^4
243  *		Q1(y) = q1[0] + q1[1]*y + ... + q1[5]*y^5
244  *
245  *		P2(y) = p2[0] + p2[1]*y + ... + p2[3]*y^3
246  *		Q2(y) = q2[0] + q2[1]*y + ... + q2[6]*y^6
247  *
248  *		P3(y) = p3[0] + p3[1]*y + ... + p3[4]*y^4
249  *		Q3(y) = q3[0] + q3[1]*y + ... + q3[5]*y^5
250  *
251  *		Remez precision of Ri(y):
252  *		|gamma(x)-(gzi+tzi*y) - y*y*Ri(y)|  <= 2**-62.3	... for i = 1
253  *					            <= 2**-59.4	... for i = 2
254  *					            <= 2**-62.1	... for i = 3
255  *
256  *	(II) For quad precision
257  *
258  *		Ri(y) = Pi(y)/Qi(y), i=1,2,3;
259  *
260  *		P1(y) = p1[0] + p1[1]*y + ... + p1[9]*y^9
261  *		Q1(y) = q1[0] + q1[1]*y + ... + q1[8]*y^8
262  *
263  *		P2(y) = p2[0] + p2[1]*y + ... + p2[9]*y^9
264  *		Q2(y) = q2[0] + q2[1]*y + ... + q2[9]*y^9
265  *
266  *		P3(y) = p3[0] + p3[1]*y + ... + p3[9]*y^9
267  *		Q3(y) = q3[0] + q3[1]*y + ... + q3[9]*y^9
268  *
269  *		Remez precision of Ri(y):
270  *		|gamma(x)-(gzi+tzi*y) - y*y*Ri(y)|  <= 2**-118.2 ... for i = 1
271  *					            <= 2**-126.8 ... for i = 2
272  *					            <= 2**-119.5 ... for i = 3
273  *
274  *	(III) For single precision
275  *
276  *		Ri(y) = Pi(y), i=1,2,3;
277  *
278  *		P1(y) = p1[0] + p1[1]*y + ... + p1[5]*y^5
279  *
280  *		P2(y) = p2[0] + p2[1]*y + ... + p2[5]*y^5
281  *
282  *		P3(y) = p3[0] + p3[1]*y + ... + p3[4]*y^4
283  *
284  *		Remez precision of Ri(y):
285  *		|gamma(x)-(gzi+tzi*y) - y*y*Ri(y)|  <= 2**-30.8	... for i = 1
286  *					            <= 2**-31.6	... for i = 2
287  *					            <= 2**-29.5	... for i = 3
288  *
289  *    Notes. (1) GTi and zi are choosen to balance the interval width and
290  *		minimize the distant between gamma(x) and the tangent line at
291  *		zi. In particular, we have
292  *		|gamma(x)-(gzi+tzi*(x-zi))|  <=   0.01436... for x in [1,z2]
293  *					     <=   0.01265... for x in [z2,2]
294  *
295  *           (2) zi are slightly adjusted so that tzi=gamma'(zi) is very
296  *		close to a single precision value.
297  *
298  *    Coefficents: Single precision
299  *	i= 1:
300  *       P1[0] =   7.09087253435088360271451613398019280077561279443e-0001
301  *       P1[1] =  -5.17229560788652108545141978238701790105241761089e-0001
302  *       P1[2] =   5.23403394528150789405825222323770647162337764327e-0001
303  *       P1[3] =  -4.54586308717075010784041566069480411732634814899e-0001
304  *       P1[4] =   4.20596490915239085459964590559256913498190955233e-0001
305  *	P1[5] =  -3.57307589712377520978332185838241458642142185789e-0001
306  *
307  *	i = 2:
308  *       p2[0] =   4.28486983980295198166056119223984284434264344578e-0001
309  *       p2[1] =  -1.30704539487709138528680121627899735386650103914e-0001
310  *       p2[2] =   1.60856285038051955072861219352655851542955430871e-0001
311  *       p2[3] =  -9.22285161346010583774458802067371182158937943507e-0002
312  *       p2[4] =   7.19240511767225260740890292605070595560626179357e-0002
313  *       p2[5] =  -4.88158265593355093703112238534484636193260459574e-0002
314  *
315  *	i = 3
316  *       p3[0] =   3.82409531118807759081121479786092134814808872880e-0001
317  *       p3[1] =   2.65309888180188647956400403013495759365167853426e-0002
318  *       p3[2] =   8.06815109775079171923561169415370309376296739835e-0002
319  *       p3[3] =  -1.54821591666137613928840890835174351674007764799e-0002
320  *       p3[4] =   1.76308239242717268530498313416899188157165183405e-0002
321  *
322  *    Coefficents: Double precision
323  * 	i = 1:
324  *       p1[0]   =   0.70908683619977797008004927192814648151397705078125000
325  *       p1[1]   =   1.71987061393048558089579513384356441668351720061e-0001
326  *       p1[2]   =  -3.19273345791990970293320316122813960527705450671e-0002
327  *       p1[3]   =   8.36172645419110036267169600390549973563534476989e-0003
328  *       p1[4]   =   1.13745336648572838333152213474277971244629758101e-0003
329  *	 q1[0]   =   1.0
330  *       q1[1]   =   9.71980217826032937526460731778472389791321968082e-0001
331  *       q1[2]   =  -7.43576743326756176594084137256042653497087666030e-0002
332  *       q1[3]   =  -1.19345944932265559769719470515102012246995255372e-0001
333  *       q1[4]   =   1.59913445751425002620935120470781382215050284762e-0002
334  *	 q1[5]   =   1.12601136853374984566572691306402321911547550783e-0003
335  * 	i = 2:
336  *       p2[0]   =   0.42848681585558601181418225678498856723308563232421875
337  *       p2[1]   =   6.53596762668970816023718845105667418483122103629e-0002
338  *       p2[2]   =  -6.97280829631212931321050770925128264272768936731e-0003
339  *       p2[3]   =   6.46342359021981718947208605674813260166116632899e-0003
340  *	 q2[0]   =   1.0
341  *       q2[1]   =   4.57572620560506047062553957454062012327519313936e-0001
342  *       q2[2]   =  -2.52182594886075452859655003407796103083422572036e-0001
343  *       q2[3]   =  -1.82970945407778594681348166040103197178711552827e-0002
344  *       q2[4]   =   2.43574726993169566475227642128830141304953840502e-0002
345  *       q2[5]   =  -5.20390406466942525358645957564897411258667085501e-0003
346  *       q2[6]   =   4.79520251383279837635552431988023256031951133885e-0004
347  * 	i = 3:
348  *	 p3[0]   =   0.382409479734567459008331979930517263710498809814453125
349  *       p3[1]   =   1.42876048697668161599069814043449301572928034140e-0001
350  *       p3[2]   =   3.42157571052250536817923866013561760785748899071e-0003
351  *       p3[3]   =  -5.01542621710067521405087887856991700987709272937e-0004
352  *       p3[4]   =   8.89285814866740910123834688163838287618332122670e-0004
353  *	 q3[0]   =   1.0
354  *       q3[1]   =   3.04253086629444201002215640948957897906299633168e-0001
355  *       q3[2]   =  -2.23162407379999477282555672834881213873185520006e-0001
356  *       q3[3]   =  -1.05060867741952065921809811933670131427552903636e-0002
357  *       q3[4]   =   1.70511763916186982473301861980856352005926669320e-0002
358  *       q3[5]   =  -2.12950201683609187927899416700094630764182477464e-0003
359  *
360  *    Note that all pi0 are exact in double, which is obtained by a
361  *    special Remez Algorithm.
362  *
363  *    Coefficents: Quad precision
364  * 	i = 1:
365  *       p1[0] =   0.709086836199777919037185741507610124611513720557
366  *       p1[1] =   4.45754781206489035827915969367354835667391606951e-0001
367  *       p1[2] =   3.21049298735832382311662273882632210062918153852e-0002
368  *       p1[3] =  -5.71296796342106617651765245858289197369688864350e-0003
369  *       p1[4] =   6.04666892891998977081619174969855831606965352773e-0003
370  *       p1[5] =   8.99106186996888711939627812174765258822658645168e-0004
371  *       p1[6] =  -6.96496846144407741431207008527018441810175568949e-0005
372  *       p1[7] =   1.52597046118984020814225409300131445070213882429e-0005
373  *       p1[8] =   5.68521076168495673844711465407432189190681541547e-0007
374  *       p1[9] =   3.30749673519634895220582062520286565610418952979e-0008
375  *       q1[0] =   1.0+0000
376  *       q1[1] =   1.35806511721671070408570853537257079579490650668e+0000
377  *       q1[2] =   2.97567810153429553405327140096063086994072952961e-0001
378  *       q1[3] =  -1.52956835982588571502954372821681851681118097870e-0001
379  *       q1[4] =  -2.88248519561420109768781615289082053597954521218e-0002
380  *       q1[5] =   1.03475311719937405219789948456313936302378395955e-0002
381  *       q1[6] =   4.12310203243891222368965360124391297374822742313e-0004
382  *       q1[7] =  -3.12653708152290867248931925120380729518332507388e-0004
383  *       q1[8] =   2.36672170850409745237358105667757760527014332458e-0005
384  *
385  * 	i = 2:
386  *       p2[0] =   0.428486815855585429730209907810650616737756697477
387  *       p2[1] =   2.63622124067885222919192651151581541943362617352e-0001
388  *       p2[2] =   3.85520683670028865731877276741390421744971446855e-0002
389  *       p2[3] =   3.05065978278128549958897133190295325258023525862e-0003
390  *       p2[4] =   2.48232934951723128892080415054084339152450445081e-0003
391  *       p2[5] =   3.67092777065632360693313762221411547741550105407e-0004
392  *       p2[6] =   3.81228045616085789674530902563145250532194518946e-0006
393  *       p2[7] =   4.61677225867087554059531455133839175822537617677e-0006
394  *       p2[8] =   2.18209052385703200438239200991201916609364872993e-0007
395  *       p2[9] =   1.00490538985245846460006244065624754421022542454e-0008
396  *       q2[0] =   1.0
397  *       q2[1] =   9.20276350207639290567783725273128544224570775056e-0001
398  *       q2[2] =  -4.79533683654165107448020515733883781138947771495e-0003
399  *       q2[3] =  -1.24538337585899300494444600248687901947684291683e-0001
400  *       q2[4] =   4.49866050763472358547524708431719114204535491412e-0003
401  *       q2[5] =   7.20715455697920560621638325356292640604078591907e-0003
402  *       q2[6] =  -8.68513169029126780280798337091982780598228096116e-0004
403  *       q2[7] =  -1.25104431629401181525027098222745544809974229874e-0004
404  *       q2[8] =   3.10558344839000038489191304550998047521253437464e-0005
405  *       q2[9] =  -1.76829227852852176018537139573609433652506765712e-0006
406  *
407  *	i = 3
408  *       p3[0] =   0.3824094797345675048502747661075355640070439388902
409  *       p3[1] =   3.42198093076618495415854906335908427159833377774e-0001
410  *       p3[2] =   9.63828189500585568303961406863153237440702754858e-0002
411  *       p3[3] =   8.76069421042696384852462044188520252156846768667e-0003
412  *       p3[4] =   1.86477890389161491224872014149309015261897537488e-0003
413  *       p3[5] =   8.16871354540309895879974742853701311541286944191e-0004
414  *       p3[6] =   6.83783483674600322518695090864659381650125625216e-0005
415  *       p3[7] =  -1.10168269719261574708565935172719209272190828456e-0006
416  *       p3[8] =   9.66243228508380420159234853278906717065629721016e-0007
417  *       p3[9] =   2.31858885579177250541163820671121664974334728142e-0008
418  *       q3[0] =   1.0
419  *       q3[1] =   8.25479821168813634632437430090376252512793067339e-0001
420  *       q3[2] =  -1.62251363073937769739639623669295110346015576320e-0002
421  *       q3[3] =  -1.10621286905916732758745130629426559691187579852e-0001
422  *       q3[4] =   3.48309693970985612644446415789230015515365291459e-0003
423  *       q3[5] =   6.73553737487488333032431261131289672347043401328e-0003
424  *       q3[6] =  -7.63222008393372630162743587811004613050245128051e-0004
425  *       q3[7] =  -1.35792670669190631476784768961953711773073251336e-0004
426  *       q3[8] =   3.19610150954223587006220730065608156460205690618e-0005
427  *       q3[9] =  -1.82096553862822346610109522015129585693354348322e-0006
428  *
429  * (C) For x between 0 and 1.
430  *     Let P stand for the number of significant bits in the working precision.
431  *                      -P                            1
432  *    (1)For 0 <= x <= 2   , gamma(x) is computed by --- rounded to nearest.
433  *                                                    x
434  *       The error is bound by 0.739 ulp(gamma(x)) in IEEE double precision.
435  *	Proof.
436  *                1                       2
437  *	Since  --------  ~  x + 0.577...*x  - ...,  we have, for small x,
438  *              gamma(x)
439  *           1                    1
440  *	----------- < gamma(x) < --- and
441  *      x(1+0.578x)               x
442  *              1                 1           1
443  *	  0 <  --- - gamma(x) <= ---  -  ----------- < 0.578
444  *              x                 x      x(1+0.578x)
445  *                                     1       1                        -P
446  * 	The error is thus bounded by --- ulp(---) + 0.578. Since x <= 2   ,
447  *                                     2       x
448  *       1      P       1           P                                      1
449  *	--- >= 2 , ulp(---) >= ulp(2  ) >= 2. Thus 0.578=0.289*2<=0.289ulp(-)
450  *       x              x                                                  x
451  *       Thus
452  *                             1                                 1
453  *		| gamma(x) - [---] rounded | <= (0.5+0.289)*ulp(---).
454  *			       x	                         x
455  *                         -P                              1
456  *	Note that for x<= 2  , it is easy to see that ulp(---)=ulp(gamma(x))
457  *                                                         x
458  *                            n                             1
459  *	except only when x = 2 , (n<= -53). In such cases, --- is exact
460  *                                                          x
461  * 	and therefore the error is bounded by
462  *                         1
463  *		0.298*ulp(---) = 0.298*2*ulp(gamma(x)) = 0.578ulp(gamma(x)).
464  *                         x
465  *	Thus we conclude that the error in gamma is less than 0.739 ulp.
466  *
467  *    (2)Otherwise, for x in GTi-1 (see B), let y = x-(zi-1). From (B) we obtain
468  *                                                          gamma(1+x)
469  *	gamma(1+x) = gy.h + gy.l,  then compute gamma(x) by -----------.
470  *                                                               x
471  *                                                          gy.h
472  *	Implementaion note. Write x = x.h+x.l, and Let th = ----- chopped to
473  *                                                            x
474  *	20 bits, then
475  *                                gy.h+gy.l
476  *		gamma(x) = th + (----------  - th )
477  *                                    x
478  *                               1
479  *			 = th + ---*(gy.h-th*x.h+gy.l-th*x.l)
480  *	                         x
481  *
482  * (D) For x between 2 and 8. Let n = 1+x chopped to an integer. Then
483  *
484  *               gamma(x)=(x-1)*(x-2)*...*(x-n)*gamma(x-n)
485  *
486  *     Since x-n is between 1 and 2, we can apply (B) to compute gamma(x).
487  *
488  *     Implementation detail. The computation of (x-1)(x-2)...(x-n) in simulated
489  *     higher precision arithmetic can be somewhat optimized.  For example, in
490  *     computing (x-1)*(x-2)*(x-3)*(x-4), if we compute (x-1)*(x-4) = z.h+z.l,
491  *     then (x-2)(x-3) = z.h+2+z.l readily. In below, we list the expression
492  *     of the formula to compute gamma(x).
493  *
494  *     Assume x-n is in GTi (i=1,2, or 3, see B for detail). Let y = x - n - zi.
495  *     By (B) we have gamma(x-n) = gy.h+gy.l. If x = x.h+x.l, then we have
496  *      n=1 (x in [2,3]):
497  *	 gamma(x) = (x-1)*gamma(x-1) = (x-1)*(gy.h+gy.l)
498  *                 = [(x.h-1)+x.l]*(gy.h+gy.l)
499  *      n=2 (x in [3,4]):
500  *        gamma(x) = (x-1)(x-2)*gamma(x-2) = (x-1)*(x-2)*(gy.h+gy.l)
501  *                 = ((x.h-2)+x.l)*((x.h-1)+x.l)*(gy.h+gy.l)
502  *                 = [x.h*(x.h-3)+2+x.l*(x+(x.h-3))]*(gy.h+gy.l)
503  *      n=3 (x in [4,5])
504  *	 gamma(x) = (x-1)(x-2)(x-3)*(gy.h+gy.l)
505  *                 = (x.h*(x.h-3)+2+x.l*(x+(x.h-3)))*[((x.h-3)+x.l)(gy.h+gy.l)]
506  *      n=4 (x in [5,6])
507  *	 gamma(x) = [(x-1)(x-4)]*[(x-2)(x-3)]*(gy.h+gy.l)
508  *                 = [(x.h*(x.h-5)+4+x.l(x+(x.h-5)))]*[(x-2)*(x-3)]*(gy.h+gy.l)
509  *                 = (y.h+y.l)*(y.h+1+y.l)*(gy.h+gy.l)
510  *      n=5 (x in [6,7])
511  *	 gamma(x) = [(x-1)(x-4)]*[(x-2)(x-3)]*[(x-5)*(gy.h+gy.l)]
512  *      n=6 (x in [7,8])
513  *	 gamma(x) = [(x-1)(x-6)]*[(x-2)(x-5)]*[(x-3)(x-4)]*(gy.h+gy.l)]
514  *		  = [(y.h+y.l)(y.h+4+y.l)][(y.h+6+y.l)(gy.h+gy.l)]
515  *
516  * (E)Overflow Thresold. For x > Overflow thresold of gamma,
517  *    return huge*huge (overflow).
518  *
519  *    By checking whether lgamma(x) >= 2**{128,1024,16384}, one can
520  *    determine the overflow threshold for x in single, double, and
521  *    quad precision. See over.c for details.
522  *
523  *    The overflow threshold of gamma(x) are
524  *
525  *    single: x = 3.5040096283e+01
526  *              = 0x420C290F (IEEE single)
527  *    double: x = 1.71624376956302711505e+02
528  *              = 0x406573FAE561F647 (IEEE double)
529  *    quad:   x = 1.7555483429044629170038892160702032034177e+03
530  *              = 0x4009B6E3180CD66A5C4206F128BA77F4  (quad)
531  *
532  * (F)For overflow_threshold >= x >= 8, we use asymptotic approximation.
533  *    (1) Stirling's formula
534  *
535  *      log(G(x)) ~= (x-.5)*(log(x)-1) + .5(log(2*pi)-1) + (1/x)*P(1/(x*x))
536  *		  = L1 + L2 + L3,
537  *    where
538  *		L1(x) = (x-.5)*(log(x)-1),
539  *		L2    = .5(log(2pi)-1) = 0.41893853....,
540  *		L3(x) = (1/x)P(1/(x*x)),
541  *
542  *    The range of L1,L2, and L3 are as follows:
543  *
544  *	------------------------------------------------------------------
545  *  	Range(L1) =  (single) [8.09..,88.30..]	 =[2** 3.01..,2**  6.46..]
546  *                   (double) [8.09..,709.3..]   =[2** 3.01..,2**  9.47..]
547  *		     (quad)   [8.09..,11356.10..]=[2** 3.01..,2** 13.47..]
548  *  	Range(L2) = 0.41893853.....
549  *	Range(L3) = [0.0104...., 0.00048....]	 =[2**-6.58..,2**-11.02..]
550  *	------------------------------------------------------------------
551  *
552  *    Gamma(x) is then computed by exp(L1+L2+L3).
553  *
554  *    (2) Error analysis of (F):
555  *    --------------------------
556  *    The error in Gamma(x) depends on the error inherited in the computation
557  *    of L= L1+L2+L3. Let L' be the computed value of L. The absolute error
558  *    in L' is t = L-L'. Since exp(L') = exp(L-t) = exp(L)*exp(t) ~
559  *    (1+t)*exp(L), the relative error in exp(L') is approximately t.
560  *
561  *    To guarantee the relatively accuracy in exp(L'), we would like
562  *    |t| < 2**(-P-5) where P denotes for the number of significant bits
563  *    of the working precision. Consequently, each of the L1,L2, and L3
564  *    must be computed with absolute error bounded by 2**(-P-5) in absolute
565  *    value.
566  *
567  *    Since L2 is a constant, it can be pre-computed to the desired accuracy.
568  *    Also |L3| < 2**-6; therefore, it suffices to compute L3 with the
569  *    working precision.  That is,
570  *	L3(x) approxmiate log(G(x))-(x-.5)(log(x)-1)-.5(log(2pi)-1)
571  *    to a precision bounded by 2**(-P-5).
572  *
573  *                                   2**(-6)
574  *			    _________V___________________
575  *		L1(x):	   |_________|___________________|
576  *			           __ ________________________
577  *		L2:	          |__|________________________|
578  *			              __________________________
579  *         +    L3(x):               |__________________________|
580  *                       -------------------------------------------
581  *                         [leading] + [Trailing]
582  *
583  *    For L1(x)=(x-0.5)*(log(x)-1), we need ilogb(L1(x))+5 extra bits for
584  *    both multiplicants to guarantee L1(x)'s absolute error is bounded by
585  *    2**(-P-5) in absolute value. Here ilogb(y) is defined to be the unbias
586  *    binary exponent of y in IEEE format.  We can get x-0.5 to the desire
587  *    accuracy easily. It remains to compute log(x)-1 with ilogb(L1(x))+5
588  *    extra bits accracy. Note that the range of L1 is 88.30.., 709.3.., and
589  *    11356.10... for single, double, and quadruple precision, we have
590  *
591  *                           single     double      quadruple
592  *                         ------------------------------------
593  *	ilogb(L1(x))+5 <=     11	  14	       18
594  *                         ------------------------------------
595  *
596  *    (3) Table Driven Method for log(x)-1:
597  *    --------------------------------------
598  *    Let x = 2**n * y, where 1 <= y < 2. Let Z={z(i),i=1,...,m}
599  *    be a set of predetermined evenly distributed floating point numbers
600  *    in [1, 2]. Let z(j) be the closest one to y, then
601  *	log(x)-1 = n*log(2)-1  +  log(y)
602  *		 = n*log(2)-1  +  log(z(j)*y/z(j))
603  *		 = n*log(2)-1  +  log(z(j))  +  log(y/z(j))
604  *		 = T1(n)       +  T2(j)      +  T3,
605  *
606  *    where T1(n) = n*log(2)-1 and T2(j) = log(z(j)). Both T1 and T2 can be
607  *    pre-calculated and be looked-up in a table. Note that 8 <= x < 1756
608  *    implies 3<=n<=10 implies 1.079.. < T1(n) < 6.931.
609  *
610  *
611  *                     y-z(i)          y       1+s
612  *    For T3, let s = --------; then ----- =  ----- and
613  *                     y+z(i)         z(i)     1-s
614  *                1+s           2   3    2   5
615  *    	T3 = log(-----) = 2s + --- s  + --- s  + ....
616  *                1-s           3        5
617  *
618  *    Suppose the first term 2s is compute in extra precision. The
619  *    dominating error in T3 would then be the rounding error of the
620  *    second term 2/3*s**3. To force the rounding bounded by
621  *    the required accuracy, we have
622  *        single:  |2/3*s**3| < 2**-11   == > |s|<0.09014...
623  *        double:  |2/3*s**3| < 2**-14   == > |s|<0.04507...
624  *        quad  :  |2/3*s**3| < 2**-18   == > |s|<0.01788... = 2**(-5.80..)
625  *
626  *    Base on this analysis, we choose Z = {z(i)|z(i)=1+i/64+1/128, 0<=i<=63}.
627  *    For any y in [1,2), let j = [64*y] chopped to integer, then z(j) is
628  *    the closest to y, and it is not difficult to see that |s| < 2**(-8).
629  *    Please note that the polynomial approximation of T3 must be accurate
630  *        -24-11   -35    -53-14    -67         -113-18   -131
631  *    to 2       =2   ,  2       = 2   ,  and  2        =2
632  *    for single, double, and quadruple precision respectively.
633  *
634  *    Inplementation notes.
635  *    (1) Table look-up entries for T1(n) and T2(j), as well as the calculation
636  *        of the leading term 2s in T3,  are broken up into leading and trailing
637  *        part such that (leading part)* 2**24 will always be an integer. That
638  *        will guarantee the addition of the leading parts will be exact.
639  *
640  *                                   2**(-24)
641  *			    _________V___________________
642  *		T1(n):	   |_________|___________________|
643  *			      _______ ______________________
644  *		T2(j):	     |_______|______________________|
645  *			         ____ _______________________
646  *		2s:	        |____|_______________________|
647  *			             __________________________
648  *         +    T3(s)-2s:           |__________________________|
649  *                       -------------------------------------------
650  *                         [leading] + [Trailing]
651  *
652  *    (2) How to compute 2s accurately.
653  *        (A) Compute v = 2s to the working precision. If |v| < 2**(-18),
654  *            stop.
655  *        (B) chopped v to 2**(-24): v = ((int)(v*2**24))/2**24
656  *	 (C) 2s = v + (2s - v), where
657  *                        1
658  *		2s - v = --- * (2(y-z) - v*(y+z) )
659  *                       y+z
660  *                         1
661  *                      = --- * ( [2(y-z) - v*(y+z)_h ]  - v*(y+z)_l  )
662  *                        y+z
663  *           where (y+z)_h = (y+z) rounded to 24 bits by (double)(float),
664  *	    and (y+z)_l = ((z+z)-(y+z)_h)+(y-z).  Note the the quantity
665  *	    in [] is exact.
666  *                                                      2         4
667  *    (3) Remez approximation for (T3(s)-2s)/s = T3[0]*s + T3[1]*s + ...:
668  *	 Single precision: 1 term (compute in double precision arithmetic)
669  *	    T3(s) = 2s + S1*s^3, S1 = 0.6666717231848518054693623697539230
670  *	    Remez error: |T3(s)/s - (2s+S1*s^3)| < 2**(-35.87)
671  *	 Double precision: 3 terms, Remez error is bounded by 2**(-72.40),
672  *	    see "tgamma_log"
673  *	 Quad precision: 7 terms, Remez error is bounded by 2**(-136.54),
674  *	    see "tgammal_log"
675  *
676  *   The computation of 0.5*(ln(2pi)-1):
677  *   	0.5*(ln(2pi)-1) =  0.4189385332046727417803297364056176398614...
678  *	split 0.5*(ln(2pi)-1) to hln2pi_h + hln2pi_l, where hln2pi_h is the
679  *	leading 21 bits of the constant.
680  *	    hln2pi_h= 0.4189383983612060546875
681  *	    hln2pi_l= 1.348434666870928297364056176398612173648e-07
682  *
683  *   The computation of 1/x*P(1/x^2) = log(G(x))-(x-.5)(ln(x)-1)-(.5ln(2pi)-1):
684  *	Let s = 1/x <= 1/8 < 0.125. We have
685  *	quad precision
686  *	    |GP(s) - s*P(s^2)| <= 2**(-120.6), where
687  *			       3      5            39
688  *	    GP(s) = GP0*s+GP1*s +GP2*s +... +GP19*s    ,
689  *       GP0  =   0.083333333333333333333333333333333172839171301
690  *			hex 0x3ffe5555 55555555 55555555 55555548
691  *       GP1  =  -2.77777777777777777777777777492501211999399424104e-0003
692  *       GP2  =   7.93650793650793650793635650541638236350020883243e-0004
693  *       GP3  =  -5.95238095238095238057299772679324503339241961704e-0004
694  *       GP4  =   8.41750841750841696138422987977683524926142600321e-0004
695  *       GP5  =  -1.91752691752686682825032547823699662178842123308e-0003
696  *       GP6  =   6.41025641022403480921891559356473451161279359322e-0003
697  *       GP7  =  -2.95506535798414019189819587455577003732808185071e-0002
698  *       GP8  =   1.79644367229970031486079180060923073476568732136e-0001
699  *       GP9  =  -1.39243086487274662174562872567057200255649290646e+0000
700  *       GP10 =   1.34025874044417962188677816477842265259608269775e+0001
701  *       GP11 =  -1.56803713480127469414495545399982508700748274318e+0002
702  *       GP12 =   2.18739841656201561694927630335099313968924493891e+0003
703  *       GP13 =  -3.55249848644100338419187038090925410976237921269e+0004
704  *       GP14 =   6.43464880437835286216768959439484376449179576452e+0005
705  *       GP15 =  -1.20459154385577014992600342782821389605893904624e+0007
706  *       GP16 =   2.09263249637351298563934942349749718491071093210e+0008
707  *       GP17 =  -2.96247483183169219343745316433899599834685703457e+0009
708  *       GP18 =   2.88984933605896033154727626086506756972327292981e+0010
709  *       GP19 =  -1.40960434146030007732838382416230610302678063984e+0011
710  *
711  *       double precision
712  *	    |GP(s) - s*P(s^2)| <= 2**(-63.5), where
713  *			       3      5      7      9      11      13      15
714  *	    GP(s) = GP0*s+GP1*s +GP2*s +GP3*s +GP4*s +GP5*s  +GP6*s  +GP7*s  ,
715  *
716  *		GP0=  0.0833333333333333287074040640618477 (3FB55555 55555555)
717  *		GP1= -2.77777777776649355200565611114627670089130772843e-0003
718  *		GP2=  7.93650787486083724805476194170211775784158551509e-0004
719  *		GP3= -5.95236628558314928757811419580281294593903582971e-0004
720  *		GP4=  8.41566473999853451983137162780427812781178932540e-0004
721  *		GP5= -1.90424776670441373564512942038926168175921303212e-0003
722  *		GP6=  5.84933161530949666312333949534482303007354299178e-0003
723  *		GP7= -1.59453228931082030262124832506144392496561694550e-0002
724  *       single precision
725  *	    |GP(s) - s*P(s^2)| <= 2**(-37.78), where
726  *			       3      5
727  *	    GP(s) = GP0*s+GP1*s +GP2*s
728  *        GP0 =   8.33333330959694065245736888749042811909994573178e-0002
729  *        GP1 =  -2.77765545601667179767706600890361535225507762168e-0003
730  *        GP2 =   7.77830853479775281781085278324621033523037489883e-0004
731  *
732  *
733  *	Implementation note:
734  *	z = (1/x), z2 = z*z, z4 = z2*z2;
735  *	p = z*(GP0+z2*(GP1+....+z2*GP7))
736  *	  = z*(GP0+(z4*(GP2+z4*(GP4+z4*GP6))+z2*(GP1+z4*(GP3+z4*(GP5+z4*GP7)))))
737  *
738  *   Adding everything up:
739  *	t = rr.h*ww.h+hln2pi_h      		... exact
740  *	w = (hln2pi_l + ((x-0.5)*ww.l+rr.l*ww.h)) + p
741  *
742  *   Computing exp(t+w):
743  *	s = t+w; write s = (n+j/32)*ln2+r, |r|<=(1/64)*ln2, then
744  *	exp(s) = 2**n * (2**(j/32) + 2**(j/32)*expm1(r)), where
745  *	expm1(r) = r + Et1*r^2 + Et2*r^3 + ... + Et5*r^6, and
746  *	2**(j/32) is obtained by table look-up S[j]+S_trail[j].
747  *	Remez error bound:
748  *	|exp(r) - (1+r+Et1*r^2+...+Et5*r^6)| <= 2^(-63).
749  */
750 
751 #include "libm.h"
752 
753 #define	__HI(x)	((int *) &x)[HIWORD]
754 #define	__LO(x)	((unsigned *) &x)[LOWORD]
755 
756 struct Double {
757 	double h;
758 	double l;
759 };
760 
761 /* Hex value of GP0 shoule be 3FB55555 55555555 */
762 static const double c[] = {
763 	+1.0,
764 	+2.0,
765 	+0.5,
766 	+1.0e-300,
767 	+6.66666666666666740682e-01,				/* A1=T3[0] */
768 	+3.99999999955626478023093908674902212920e-01,		/* A2=T3[1] */
769 	+2.85720221533145659809237398709372330980e-01,		/* A3=T3[2] */
770 	+0.0833333333333333287074040640618477,			/* GP[0] */
771 	-2.77777777776649355200565611114627670089130772843e-03,
772 	+7.93650787486083724805476194170211775784158551509e-04,
773 	-5.95236628558314928757811419580281294593903582971e-04,
774 	+8.41566473999853451983137162780427812781178932540e-04,
775 	-1.90424776670441373564512942038926168175921303212e-03,
776 	+5.84933161530949666312333949534482303007354299178e-03,
777 	-1.59453228931082030262124832506144392496561694550e-02,
778 	+4.18937683105468750000e-01,				/* hln2pi_h */
779 	+8.50099203991780279640e-07,				/* hln2pi_l */
780 	+4.18938533204672741744150788368695779923320328369e-01,	/* hln2pi */
781 	+2.16608493865351192653e-02,				/* ln2_32hi */
782 	+5.96317165397058656257e-12,				/* ln2_32lo */
783 	+4.61662413084468283841e+01,				/* invln2_32 */
784 	+5.0000000000000000000e-1,				/* Et1 */
785 	+1.66666666665223585560605991943703896196054020060e-01,	/* Et2 */
786 	+4.16666666665895103520154073534275286743788421687e-02,	/* Et3 */
787 	+8.33336844093536520775865096538773197505523826029e-03,	/* Et4 */
788 	+1.38889201930843436040204096950052984793587640227e-03,	/* Et5 */
789 };
790 
791 #define	one	  c[0]
792 #define	two	  c[1]
793 #define	half	  c[2]
794 #define	tiny	  c[3]
795 #define	A1	  c[4]
796 #define	A2	  c[5]
797 #define	A3	  c[6]
798 #define	GP0	  c[7]
799 #define	GP1	  c[8]
800 #define	GP2	  c[9]
801 #define	GP3	  c[10]
802 #define	GP4	  c[11]
803 #define	GP5	  c[12]
804 #define	GP6	  c[13]
805 #define	GP7	  c[14]
806 #define	hln2pi_h  c[15]
807 #define	hln2pi_l  c[16]
808 #define	hln2pi	  c[17]
809 #define	ln2_32hi  c[18]
810 #define	ln2_32lo  c[19]
811 #define	invln2_32 c[20]
812 #define	Et1	  c[21]
813 #define	Et2	  c[22]
814 #define	Et3	  c[23]
815 #define	Et4	  c[24]
816 #define	Et5	  c[25]
817 
818 /*
819  * double precision coefficients for computing log(x)-1 in tgamma.
820  *  See "algorithm" for details
821  *
822  *  log(x) - 1 = T1(n) + T2(j) + T3(s), where x = 2**n * y,  1<=y<2,
823  *  j=[64*y], z[j]=1+j/64+1/128, s = (y-z[j])/(y+z[j]), and
824  *       T1(n) = T1[2n,2n+1] = n*log(2)-1,
825  *       T2(j) = T2[2j,2j+1] = log(z[j]),
826  *       T3(s) = 2s + T3[0]s^3 + T3[1]s^5 + T3[2]s^7
827  *	       = 2s + A1*s^3 + A2*s^5 + A3*s^7  (see const A1,A2,A3)
828  *  Note
829  *  (1) the leading entries are truncated to 24 binary point.
830  *      See Remezpak/sun/tgamma_log_64.c
831  *  (2) Remez error for T3(s) is bounded by 2**(-72.4)
832  *      See mpremez/work/Log/tgamma_log_4_outr2
833  */
834 
835 static const double T1[] = {
836 	-1.00000000000000000000e+00,	/* 0xBFF00000 0x00000000 */
837 	+0.00000000000000000000e+00,	/* 0x00000000 0x00000000 */
838 	-3.06852817535400390625e-01,	/* 0xBFD3A37A 0x00000000 */
839 	-1.90465429995776763166e-09,	/* 0xBE205C61 0x0CA86C38 */
840 	+3.86294305324554443359e-01,	/* 0x3FD8B90B 0xC0000000 */
841 	+5.57953361754750897367e-08,	/* 0x3E6DF473 0xDE6AF279 */
842 	+1.07944148778915405273e+00,	/* 0x3FF14564 0x70000000 */
843 	+5.38906818755173187963e-08,	/* 0x3E6CEEAD 0xCDA06BB5 */
844 	+1.77258867025375366211e+00,	/* 0x3FFC5C85 0xF0000000 */
845 	+5.19860275755595544734e-08,	/* 0x3E6BE8E7 0xBCD5E4F2 */
846 	+2.46573585271835327148e+00,	/* 0x4003B9D3 0xB8000000 */
847 	+5.00813732756017835330e-08,	/* 0x3E6AE321 0xAC0B5E2E */
848 	+3.15888303518295288086e+00,	/* 0x40094564 0x78000000 */
849 	+4.81767189756440192100e-08,	/* 0x3E69DD5B 0x9B40D76B */
850 	+3.85203021764755249023e+00,	/* 0x400ED0F5 0x38000000 */
851 	+4.62720646756862482697e-08,	/* 0x3E68D795 0x8A7650A7 */
852 	+4.54517740011215209961e+00,	/* 0x40122E42 0xFC000000 */
853 	+4.43674103757284839467e-08,	/* 0x3E67D1CF 0x79ABC9E4 */
854 	+5.23832458257675170898e+00,	/* 0x4014F40B 0x5C000000 */
855 	+4.24627560757707130063e-08,	/* 0x3E66CC09 0x68E14320 */
856 	+5.93147176504135131836e+00,	/* 0x4017B9D3 0xBC000000 */
857 	+4.05581017758129486834e-08,	/* 0x3E65C643 0x5816BC5D */
858 };
859 
860 static const double T2[] = {
861 	+7.78210163116455078125e-03,	/* 0x3F7FE020 0x00000000 */
862 	+3.88108903981662140884e-08,	/* 0x3E64D620 0xCF11F86F */
863 	+2.31670141220092773438e-02,	/* 0x3F97B918 0x00000000 */
864 	+4.51595251008850513740e-08,	/* 0x3E683EAD 0x88D54940 */
865 	+3.83188128471374511719e-02,	/* 0x3FA39E86 0x00000000 */
866 	+5.14549991480218823411e-08,	/* 0x3E6B9FEB 0xD5FA9016 */
867 	+5.32444715499877929688e-02,	/* 0x3FAB42DC 0x00000000 */
868 	+4.29688244898971182165e-08,	/* 0x3E671197 0x1BEC28D1 */
869 	+6.79506063461303710938e-02,	/* 0x3FB16536 0x00000000 */
870 	+5.55623773783008185114e-08,	/* 0x3E6DD46F 0x5C1D0C4C */
871 	+8.24436545372009277344e-02,	/* 0x3FB51B07 0x00000000 */
872 	+1.46738736635337847313e-08,	/* 0x3E4F830C 0x1FB493C7 */
873 	+9.67295765876770019531e-02,	/* 0x3FB8C345 0x00000000 */
874 	+4.98708741103424492282e-08,	/* 0x3E6AC633 0x641EB597 */
875 	+1.10814332962036132812e-01,	/* 0x3FBC5E54 0x00000000 */
876 	+3.33782539813823062226e-08,	/* 0x3E61EB78 0xE862BAC3 */
877 	+1.24703466892242431641e-01,	/* 0x3FBFEC91 0x00000000 */
878 	+1.16087148042227818450e-08,	/* 0x3E48EDF5 0x5D551729 */
879 	+1.38402283191680908203e-01,	/* 0x3FC1B72A 0x80000000 */
880 	+3.96674382274822001957e-08,	/* 0x3E654BD9 0xE80A4181 */
881 	+1.51916027069091796875e-01,	/* 0x3FC371FC 0x00000000 */
882 	+1.49567501781968021494e-08,	/* 0x3E500F47 0xBA1DE6CB */
883 	+1.65249526500701904297e-01,	/* 0x3FC526E5 0x80000000 */
884 	+4.63946052585787334062e-08,	/* 0x3E68E86D 0x0DE8B900 */
885 	+1.78407609462738037109e-01,	/* 0x3FC6D60F 0x80000000 */
886 	+4.80100802600100279538e-08,	/* 0x3E69C674 0x8723551E */
887 	+1.91394805908203125000e-01,	/* 0x3FC87FA0 0x00000000 */
888 	+4.70914263296092971436e-08,	/* 0x3E694832 0x44240802 */
889 	+2.04215526580810546875e-01,	/* 0x3FCA23BC 0x00000000 */
890 	+1.48478803446288209001e-08,	/* 0x3E4FE2B5 0x63193712 */
891 	+2.16873884201049804688e-01,	/* 0x3FCBC286 0x00000000 */
892 	+5.40995645549315919488e-08,	/* 0x3E6D0B63 0x358A7E74 */
893 	+2.29374051094055175781e-01,	/* 0x3FCD5C21 0x00000000 */
894 	+4.99707906542102284117e-08,	/* 0x3E6AD3EE 0xE456E443 */
895 	+2.41719901561737060547e-01,	/* 0x3FCEF0AD 0x80000000 */
896 	+3.53254081075974352804e-08,	/* 0x3E62F716 0x4D948638 */
897 	+2.53915190696716308594e-01,	/* 0x3FD04025 0x80000000 */
898 	+1.92842471355435739091e-08,	/* 0x3E54B4D0 0x40DAE27C */
899 	+2.65963494777679443359e-01,	/* 0x3FD1058B 0xC0000000 */
900 	+5.37194584979797487125e-08,	/* 0x3E6CD725 0x6A8C4FD0 */
901 	+2.77868449687957763672e-01,	/* 0x3FD1C898 0xC0000000 */
902 	+1.31549854251447496506e-09,	/* 0x3E16999F 0xAFBC68E7 */
903 	+2.89633274078369140625e-01,	/* 0x3FD2895A 0x00000000 */
904 	+1.85046735362538929911e-08,	/* 0x3E53DE86 0xA35EB493 */
905 	+3.01261305809020996094e-01,	/* 0x3FD347DD 0x80000000 */
906 	+2.47691407849191245052e-08,	/* 0x3E5A987D 0x54D64567 */
907 	+3.12755703926086425781e-01,	/* 0x3FD40430 0x80000000 */
908 	+6.07781046260499658610e-09,	/* 0x3E3A1A9F 0x8EF4304A */
909 	+3.24119448661804199219e-01,	/* 0x3FD4BE5F 0x80000000 */
910 	+1.99924077768719198045e-08,	/* 0x3E557778 0xA0DB4C99 */
911 	+3.35355520248413085938e-01,	/* 0x3FD57677 0x00000000 */
912 	+2.16727247443196802771e-08,	/* 0x3E57455A 0x6C549AB7 */
913 	+3.46466720104217529297e-01,	/* 0x3FD62C82 0xC0000000 */
914 	+4.72419910516215900493e-08,	/* 0x3E695CE3 0xCA97B7B0 */
915 	+3.57455849647521972656e-01,	/* 0x3FD6E08E 0x80000000 */
916 	+3.92742818015697624778e-08,	/* 0x3E6515D0 0xF1C609CA */
917 	+3.68325531482696533203e-01,	/* 0x3FD792A5 0x40000000 */
918 	+2.96760111198451042238e-08,	/* 0x3E5FDD47 0xA27C15DA */
919 	+3.79078328609466552734e-01,	/* 0x3FD842D1 0xC0000000 */
920 	+2.43255029056564770289e-08,	/* 0x3E5A1E8B 0x17493B14 */
921 	+3.89716744422912597656e-01,	/* 0x3FD8F11E 0x80000000 */
922 	+6.71711261571421332726e-09,	/* 0x3E3CD98B 0x1DF85DA7 */
923 	+4.00243163108825683594e-01,	/* 0x3FD99D95 0x80000000 */
924 	+1.01818702333557515008e-09,	/* 0x3E117E08 0xACBA92EF */
925 	+4.10659909248352050781e-01,	/* 0x3FDA4840 0x80000000 */
926 	+1.57369163351530571459e-08,	/* 0x3E50E5BB 0x0A2BFCA7 */
927 	+4.20969247817993164062e-01,	/* 0x3FDAF129 0x00000000 */
928 	+4.68261364720663662040e-08,	/* 0x3E6923BC 0x358899C2 */
929 	+4.31173443794250488281e-01,	/* 0x3FDB9858 0x80000000 */
930 	+2.10241208525779214510e-08,	/* 0x3E569310 0xFB598FB1 */
931 	+4.41274523735046386719e-01,	/* 0x3FDC3DD7 0x80000000 */
932 	+3.70698288427707487748e-08,	/* 0x3E63E6D6 0xA6B9D9E1 */
933 	+4.51274633407592773438e-01,	/* 0x3FDCE1AF 0x00000000 */
934 	+1.07318658117071930723e-08,	/* 0x3E470BE7 0xD6F6FA58 */
935 	+4.61175680160522460938e-01,	/* 0x3FDD83E7 0x00000000 */
936 	+3.49616477054305011286e-08,	/* 0x3E62C517 0x9F2828AE */
937 	+4.70979690551757812500e-01,	/* 0x3FDE2488 0x00000000 */
938 	+2.46670332000468969567e-08,	/* 0x3E5A7C6C 0x261CBD8F */
939 	+4.80688512325286865234e-01,	/* 0x3FDEC399 0xC0000000 */
940 	+1.70204650424422423704e-08,	/* 0x3E52468C 0xC0175CEE */
941 	+4.90303933620452880859e-01,	/* 0x3FDF6123 0xC0000000 */
942 	+5.44247409572909703749e-08,	/* 0x3E6D3814 0x5630A2B6 */
943 	+4.99827861785888671875e-01,	/* 0x3FDFFD2E 0x00000000 */
944 	+7.77056065794633071345e-09,	/* 0x3E40AFE9 0x30AB2FA0 */
945 	+5.09261846542358398438e-01,	/* 0x3FE04BDF 0x80000000 */
946 	+5.52474495483665749052e-08,	/* 0x3E6DA926 0xD265FCC1 */
947 	+5.18607735633850097656e-01,	/* 0x3FE0986F 0x40000000 */
948 	+2.85741955344967264536e-08,	/* 0x3E5EAE6A 0x41723FB5 */
949 	+5.27867078781127929688e-01,	/* 0x3FE0E449 0x80000000 */
950 	+1.08397144554263914271e-08,	/* 0x3E474732 0x2FDBAB97 */
951 	+5.37041425704956054688e-01,	/* 0x3FE12F71 0x80000000 */
952 	+4.01919275998792285777e-08,	/* 0x3E6593EF 0xBC530123 */
953 	+5.46132385730743408203e-01,	/* 0x3FE179EA 0xA0000000 */
954 	+5.18673922421792693237e-08,	/* 0x3E6BD899 0xA0BFC60E */
955 	+5.55141448974609375000e-01,	/* 0x3FE1C3B8 0x00000000 */
956 	+5.85658922177154808539e-08,	/* 0x3E6F713C 0x24BC94F9 */
957 	+5.64070105552673339844e-01,	/* 0x3FE20CDC 0xC0000000 */
958 	+3.27321296262276338905e-08,	/* 0x3E6192AB 0x6D93503D */
959 	+5.72919726371765136719e-01,	/* 0x3FE2555B 0xC0000000 */
960 	+2.71900203723740076878e-08,	/* 0x3E5D31EF 0x96780876 */
961 	+5.81691682338714599609e-01,	/* 0x3FE29D37 0xE0000000 */
962 	+5.72959078829112371070e-08,	/* 0x3E6EC2B0 0x8AC85CD7 */
963 	+5.90387403964996337891e-01,	/* 0x3FE2E474 0x20000000 */
964 	+4.26371800367512948470e-08,	/* 0x3E66E402 0x68405422 */
965 	+5.99008142948150634766e-01,	/* 0x3FE32B13 0x20000000 */
966 	+4.66979327646159769249e-08,	/* 0x3E69121D 0x71320557 */
967 	+6.07555210590362548828e-01,	/* 0x3FE37117 0xA0000000 */
968 	+3.96341792466729582847e-08,	/* 0x3E654747 0xB5C5DD02 */
969 	+6.16029858589172363281e-01,	/* 0x3FE3B684 0x40000000 */
970 	+1.86263416563663175432e-08,	/* 0x3E53FFF8 0x455F1DBE */
971 	+6.24433279037475585938e-01,	/* 0x3FE3FB5B 0x80000000 */
972 	+8.97441791510503832111e-09,	/* 0x3E4345BD 0x096D3A75 */
973 	+6.32766664028167724609e-01,	/* 0x3FE43F9F 0xE0000000 */
974 	+5.54287010493641158796e-09,	/* 0x3E37CE73 0x3BD393DD */
975 	+6.41031146049499511719e-01,	/* 0x3FE48353 0xC0000000 */
976 	+3.33714317793368531132e-08,	/* 0x3E61EA88 0xDF73D5E9 */
977 	+6.49227917194366455078e-01,	/* 0x3FE4C679 0xA0000000 */
978 	+2.94307433638127158696e-08,	/* 0x3E5F99DC 0x7362D1DA */
979 	+6.57358050346374511719e-01,	/* 0x3FE50913 0xC0000000 */
980 	+2.23619855184231409785e-08,	/* 0x3E5802D0 0xD6979675 */
981 	+6.65422618389129638672e-01,	/* 0x3FE54B24 0x60000000 */
982 	+1.41559608102782173188e-08,	/* 0x3E4E6652 0x5EA4550A */
983 	+6.73422634601593017578e-01,	/* 0x3FE58CAD 0xA0000000 */
984 	+4.06105737027198329700e-08,	/* 0x3E65CD79 0x893092F2 */
985 	+6.81359171867370605469e-01,	/* 0x3FE5CDB1 0xC0000000 */
986 	+5.29405324634793230630e-08,	/* 0x3E6C6C17 0x648CF6E4 */
987 	+6.89233243465423583984e-01,	/* 0x3FE60E32 0xE0000000 */
988 	+3.77733853963405370102e-08,	/* 0x3E644788 0xD8CA7C89 */
989 };
990 
991 /* S[j],S_trail[j] = 2**(j/32.) for the final computation of exp(t+w) */
992 static const double S[] = {
993 	+1.00000000000000000000e+00,	/* 3FF0000000000000 */
994 	+1.02189714865411662714e+00,	/* 3FF059B0D3158574 */
995 	+1.04427378242741375480e+00,	/* 3FF0B5586CF9890F */
996 	+1.06714040067682369717e+00,	/* 3FF11301D0125B51 */
997 	+1.09050773266525768967e+00,	/* 3FF172B83C7D517B */
998 	+1.11438674259589243221e+00,	/* 3FF1D4873168B9AA */
999 	+1.13878863475669156458e+00,	/* 3FF2387A6E756238 */
1000 	+1.16372485877757747552e+00,	/* 3FF29E9DF51FDEE1 */
1001 	+1.18920711500272102690e+00,	/* 3FF306FE0A31B715 */
1002 	+1.21524735998046895524e+00,	/* 3FF371A7373AA9CB */
1003 	+1.24185781207348400201e+00,	/* 3FF3DEA64C123422 */
1004 	+1.26905095719173321989e+00,	/* 3FF44E086061892D */
1005 	+1.29683955465100964055e+00,	/* 3FF4BFDAD5362A27 */
1006 	+1.32523664315974132322e+00,	/* 3FF5342B569D4F82 */
1007 	+1.35425554693689265129e+00,	/* 3FF5AB07DD485429 */
1008 	+1.38390988196383202258e+00,	/* 3FF6247EB03A5585 */
1009 	+1.41421356237309514547e+00,	/* 3FF6A09E667F3BCD */
1010 	+1.44518080697704665027e+00,	/* 3FF71F75E8EC5F74 */
1011 	+1.47682614593949934623e+00,	/* 3FF7A11473EB0187 */
1012 	+1.50916442759342284141e+00,	/* 3FF82589994CCE13 */
1013 	+1.54221082540794074411e+00,	/* 3FF8ACE5422AA0DB */
1014 	+1.57598084510788649659e+00,	/* 3FF93737B0CDC5E5 */
1015 	+1.61049033194925428347e+00,	/* 3FF9C49182A3F090 */
1016 	+1.64575547815396494578e+00,	/* 3FFA5503B23E255D */
1017 	+1.68179283050742900407e+00,	/* 3FFAE89F995AD3AD */
1018 	+1.71861929812247793414e+00,	/* 3FFB7F76F2FB5E47 */
1019 	+1.75625216037329945351e+00,	/* 3FFC199BDD85529C */
1020 	+1.79470907500310716820e+00,	/* 3FFCB720DCEF9069 */
1021 	+1.83400808640934243066e+00,	/* 3FFD5818DCFBA487 */
1022 	+1.87416763411029996256e+00,	/* 3FFDFC97337B9B5F */
1023 	+1.91520656139714740007e+00,	/* 3FFEA4AFA2A490DA */
1024 	+1.95714412417540017941e+00,	/* 3FFF50765B6E4540 */
1025 };
1026 
1027 static const double S_trail[] = {
1028 	+0.00000000000000000000e+00,
1029 	+5.10922502897344389359e-17,	/* 3C8D73E2A475B465 */
1030 	+8.55188970553796365958e-17,	/* 3C98A62E4ADC610A */
1031 	-7.89985396684158212226e-17,	/* BC96C51039449B3A */
1032 	-3.04678207981247114697e-17,	/* BC819041B9D78A76 */
1033 	+1.04102784568455709549e-16,	/* 3C9E016E00A2643C */
1034 	+8.91281267602540777782e-17,	/* 3C99B07EB6C70573 */
1035 	+3.82920483692409349872e-17,	/* 3C8612E8AFAD1255 */
1036 	+3.98201523146564611098e-17,	/* 3C86F46AD23182E4 */
1037 	-7.71263069268148813091e-17,	/* BC963AEABF42EAE2 */
1038 	+4.65802759183693679123e-17,	/* 3C8ADA0911F09EBC */
1039 	+2.66793213134218609523e-18,	/* 3C489B7A04EF80D0 */
1040 	+2.53825027948883149593e-17,	/* 3C7D4397AFEC42E2 */
1041 	-2.85873121003886075697e-17,	/* BC807ABE1DB13CAC */
1042 	+7.70094837980298946162e-17,	/* 3C96324C054647AD */
1043 	-6.77051165879478628716e-17,	/* BC9383C17E40B497 */
1044 	-9.66729331345291345105e-17,	/* BC9BDD3413B26456 */
1045 	-3.02375813499398731940e-17,	/* BC816E4786887A99 */
1046 	-3.48399455689279579579e-17,	/* BC841577EE04992F */
1047 	-1.01645532775429503911e-16,	/* BC9D4C1DD41532D8 */
1048 	+7.94983480969762085616e-17,	/* 3C96E9F156864B27 */
1049 	-1.01369164712783039808e-17,	/* BC675FC781B57EBC */
1050 	+2.47071925697978878522e-17,	/* 3C7C7C46B071F2BE */
1051 	-1.01256799136747726038e-16,	/* BC9D2F6EDB8D41E1 */
1052 	+8.19901002058149652013e-17,	/* 3C97A1CD345DCC81 */
1053 	-1.85138041826311098821e-17,	/* BC75584F7E54AC3B */
1054 	+2.96014069544887330703e-17,	/* 3C811065895048DD */
1055 	+1.82274584279120867698e-17,	/* 3C7503CBD1E949DB */
1056 	+3.28310722424562658722e-17,	/* 3C82ED02D75B3706 */
1057 	-6.12276341300414256164e-17,	/* BC91A5CD4F184B5C */
1058 	-1.06199460561959626376e-16,	/* BC9E9C23179C2893 */
1059 	+8.96076779103666776760e-17,	/* 3C99D3E12DD8A18B */
1060 };
1061 
1062 /* Primary interval GTi() */
1063 static const double cr[] = {
1064 /* p1, q1 */
1065 	+0.70908683619977797008004927192814648151397705078125000,
1066 	+1.71987061393048558089579513384356441668351720061e-0001,
1067 	-3.19273345791990970293320316122813960527705450671e-0002,
1068 	+8.36172645419110036267169600390549973563534476989e-0003,
1069 	+1.13745336648572838333152213474277971244629758101e-0003,
1070 	+1.0,
1071 	+9.71980217826032937526460731778472389791321968082e-0001,
1072 	-7.43576743326756176594084137256042653497087666030e-0002,
1073 	-1.19345944932265559769719470515102012246995255372e-0001,
1074 	+1.59913445751425002620935120470781382215050284762e-0002,
1075 	+1.12601136853374984566572691306402321911547550783e-0003,
1076 /* p2, q2 */
1077 	+0.42848681585558601181418225678498856723308563232421875,
1078 	+6.53596762668970816023718845105667418483122103629e-0002,
1079 	-6.97280829631212931321050770925128264272768936731e-0003,
1080 	+6.46342359021981718947208605674813260166116632899e-0003,
1081 	+1.0,
1082 	+4.57572620560506047062553957454062012327519313936e-0001,
1083 	-2.52182594886075452859655003407796103083422572036e-0001,
1084 	-1.82970945407778594681348166040103197178711552827e-0002,
1085 	+2.43574726993169566475227642128830141304953840502e-0002,
1086 	-5.20390406466942525358645957564897411258667085501e-0003,
1087 	+4.79520251383279837635552431988023256031951133885e-0004,
1088 /* p3, q3 */
1089 	+0.382409479734567459008331979930517263710498809814453125,
1090 	+1.42876048697668161599069814043449301572928034140e-0001,
1091 	+3.42157571052250536817923866013561760785748899071e-0003,
1092 	-5.01542621710067521405087887856991700987709272937e-0004,
1093 	+8.89285814866740910123834688163838287618332122670e-0004,
1094 	+1.0,
1095 	+3.04253086629444201002215640948957897906299633168e-0001,
1096 	-2.23162407379999477282555672834881213873185520006e-0001,
1097 	-1.05060867741952065921809811933670131427552903636e-0002,
1098 	+1.70511763916186982473301861980856352005926669320e-0002,
1099 	-2.12950201683609187927899416700094630764182477464e-0003,
1100 };
1101 
1102 #define	P10   cr[0]
1103 #define	P11   cr[1]
1104 #define	P12   cr[2]
1105 #define	P13   cr[3]
1106 #define	P14   cr[4]
1107 #define	Q10   cr[5]
1108 #define	Q11   cr[6]
1109 #define	Q12   cr[7]
1110 #define	Q13   cr[8]
1111 #define	Q14   cr[9]
1112 #define	Q15   cr[10]
1113 #define	P20   cr[11]
1114 #define	P21   cr[12]
1115 #define	P22   cr[13]
1116 #define	P23   cr[14]
1117 #define	Q20   cr[15]
1118 #define	Q21   cr[16]
1119 #define	Q22   cr[17]
1120 #define	Q23   cr[18]
1121 #define	Q24   cr[19]
1122 #define	Q25   cr[20]
1123 #define	Q26   cr[21]
1124 #define	P30   cr[22]
1125 #define	P31   cr[23]
1126 #define	P32   cr[24]
1127 #define	P33   cr[25]
1128 #define	P34   cr[26]
1129 #define	Q30   cr[27]
1130 #define	Q31   cr[28]
1131 #define	Q32   cr[29]
1132 #define	Q33   cr[30]
1133 #define	Q34   cr[31]
1134 #define	Q35   cr[32]
1135 
1136 static const double
1137 	GZ1_h = +0.938204627909682398190,
1138 	GZ1_l = +5.121952600248205157935e-17,
1139 	GZ2_h = +0.885603194410888749921,
1140 	GZ2_l = -4.964236872556339810692e-17,
1141 	GZ3_h = +0.936781411463652347038,
1142 	GZ3_l = -2.541923110834479415023e-17,
1143 	TZ1 = -0.3517214357852935791015625,
1144 	TZ3 = +0.280530631542205810546875;
1145 /* INDENT ON */
1146 
1147 /* compute gamma(y=yh+yl) for y in GT1 = [1.0000, 1.2845] */
1148 /* assume yh got 20 significant bits */
1149 static struct Double
1150 GT1(double yh, double yl) {
1151 	double t3, t4, y, z;
1152 	struct Double r;
1153 
1154 	y = yh + yl;
1155 	z = y * y;
1156 	t3 = (z * (P10 + y * ((P11 + y * P12) + z * (P13 + y * P14)))) /
1157 		(Q10 + y * ((Q11 + y * Q12) + z * ((Q13 + Q14 * y) + z * Q15)));
1158 	t3 += (TZ1 * yl + GZ1_l);
1159 	t4 = TZ1 * yh;
1160 	r.h = (double) ((float) (t4 + GZ1_h + t3));
1161 	t3 += (t4 - (r.h - GZ1_h));
1162 	r.l = t3;
1163 	return (r);
1164 }
1165 
1166 /* compute gamma(y=yh+yl) for y in GT2 = [1.2844, 1.6374] */
1167 /* assume yh got 20 significant bits */
1168 static struct Double
1169 GT2(double yh, double yl) {
1170 	double t3, y, z;
1171 	struct Double r;
1172 
1173 	y = yh + yl;
1174 	z = y * y;
1175 	t3 = (z * (P20 + y * P21 + z * (P22 + y * P23))) /
1176 		(Q20 + (y * ((Q21 + Q22 * y) + z * Q23) +
1177 		(z * z) * ((Q24 + Q25 * y) + z * Q26))) + GZ2_l;
1178 	r.h = (double) ((float) (GZ2_h + t3));
1179 	r.l = t3 - (r.h - GZ2_h);
1180 	return (r);
1181 }
1182 
1183 /* compute gamma(y=yh+yl) for y in GT3 = [1.6373, 2.0000] */
1184 /* assume yh got 20 significant bits */
1185 static struct Double
1186 GT3(double yh, double yl) {
1187 	double t3, t4, y, z;
1188 	struct Double r;
1189 
1190 	y = yh + yl;
1191 	z = y * y;
1192 	t3 = (z * (P30 + y * ((P31 + y * P32) + z * (P33 + y * P34)))) /
1193 		(Q30 + y * ((Q31 + y * Q32) + z * ((Q33 + Q34 * y) + z * Q35)));
1194 	t3 += (TZ3 * yl + GZ3_l);
1195 	t4 = TZ3 * yh;
1196 	r.h = (double) ((float) (t4 + GZ3_h + t3));
1197 	t3 += (t4 - (r.h - GZ3_h));
1198 	r.l = t3;
1199 	return (r);
1200 }
1201 
1202 /* INDENT OFF */
1203 /*
1204  * return tgamma(x) scaled by 2**-m for 8<x<=171.62... using Stirling's formula
1205  *     log(G(x)) ~= (x-.5)*(log(x)-1) + .5(log(2*pi)-1) + (1/x)*P(1/(x*x))
1206  *                = L1 + L2 + L3,
1207  */
1208 /* INDENT ON */
1209 static struct Double
1210 large_gam(double x, int *m) {
1211 	double z, t1, t2, t3, z2, t5, w, y, u, r, z4, v, t24 = 16777216.0,
1212 		p24 = 1.0 / 16777216.0;
1213 	int n2, j2, k, ix, j;
1214 	unsigned lx;
1215 	struct Double zz;
1216 	double u2, ss_h, ss_l, r_h, w_h, w_l, t4;
1217 
1218 /* INDENT OFF */
1219 /*
1220  * compute ss = ss.h+ss.l = log(x)-1 (see tgamma_log.h for details)
1221  *
1222  *  log(x) - 1 = T1(n) + T2(j) + T3(s), where x = 2**n * y,  1<=y<2,
1223  *  j=[64*y], z[j]=1+j/64+1/128, s = (y-z[j])/(y+z[j]), and
1224  *       T1(n) = T1[2n,2n+1] = n*log(2)-1,
1225  *       T2(j) = T2[2j,2j+1] = log(z[j]),
1226  *       T3(s) = 2s + A1[0]s^3 + A2[1]s^5 + A3[2]s^7
1227  *  Note
1228  *  (1) the leading entries are truncated to 24 binary point.
1229  *  (2) Remez error for T3(s) is bounded by 2**(-72.4)
1230  *                                   2**(-24)
1231  *                           _________V___________________
1232  *               T1(n):     |_________|___________________|
1233  *                             _______ ______________________
1234  *               T2(j):       |_______|______________________|
1235  *                                ____ _______________________
1236  *               2s:             |____|_______________________|
1237  *                                    __________________________
1238  *          +    T3(s)-2s:           |__________________________|
1239  *                       -------------------------------------------
1240  *                          [leading] + [Trailing]
1241  */
1242 /* INDENT ON */
1243 	ix = __HI(x);
1244 	lx = __LO(x);
1245 	n2 = (ix >> 20) - 0x3ff;	/* exponent of x, range:3-7 */
1246 	n2 += n2;			/* 2n */
1247 	ix = (ix & 0x000fffff) | 0x3ff00000;	/* y = scale x to [1,2] */
1248 	__HI(y) = ix;
1249 	__LO(y) = lx;
1250 	__HI(z) = (ix & 0xffffc000) | 0x2000;	/* z[j]=1+j/64+1/128 */
1251 	__LO(z) = 0;
1252 	j2 = (ix >> 13) & 0x7e;	/* 2j */
1253 	t1 = y + z;
1254 	t2 = y - z;
1255 	r = one / t1;
1256 	t1 = (double) ((float) t1);
1257 	u = r * t2;		/* u = (y-z)/(y+z) */
1258 	t4 = T2[j2 + 1] + T1[n2 + 1];
1259 	z2 = u * u;
1260 	k = __HI(u) & 0x7fffffff;
1261 	t3 = T2[j2] + T1[n2];
1262 	if ((k >> 20) < 0x3ec) {	/* |u|<2**-19 */
1263 		t2 = t4 + u * ((two + z2 * A1) + (z2 * z2) * (A2 + z2 * A3));
1264 	} else {
1265 		t5 = t4 + u * (z2 * A1 + (z2 * z2) * (A2 + z2 * A3));
1266 		u2 = u + u;
1267 		v = (double) ((int) (u2 * t24)) * p24;
1268 		t2 = t5 + r * ((two * t2 - v * t1) - v * (y - (t1 - z)));
1269 		t3 += v;
1270 	}
1271 	ss_h = (double) ((float) (t2 + t3));
1272 	ss_l = t2 - (ss_h - t3);
1273 
1274 	/*
1275 	 * compute ww = (x-.5)*(log(x)-1) + .5*(log(2pi)-1) + 1/x*(P(1/x^2)))
1276 	 * where ss = log(x) - 1 in already in extra precision
1277 	 */
1278 	z = one / x;
1279 	r = x - half;
1280 	r_h = (double) ((float) r);
1281 	w_h = r_h * ss_h + hln2pi_h;
1282 	z2 = z * z;
1283 	w = (r - r_h) * ss_h + r * ss_l;
1284 	z4 = z2 * z2;
1285 	t1 = z2 * (GP1 + z4 * (GP3 + z4 * (GP5 + z4 * GP7)));
1286 	t2 = z4 * (GP2 + z4 * (GP4 + z4 * GP6));
1287 	t1 += t2;
1288 	w += hln2pi_l;
1289 	w_l = z * (GP0 + t1) + w;
1290 	k = (int) ((w_h + w_l) * invln2_32 + half);
1291 
1292 	/* compute the exponential of w_h+w_l */
1293 	j = k & 0x1f;
1294 	*m = (k >> 5);
1295 	t3 = (double) k;
1296 
1297 	/* perform w - k*ln2_32 (represent as w_h - w_l) */
1298 	t1 = w_h - t3 * ln2_32hi;
1299 	t2 = t3 * ln2_32lo;
1300 	w = w_l - t2;
1301 	w_h = t1 + w_l;
1302 	w_l = t2 - (w_l - (w_h - t1));
1303 
1304 	/* compute exp(w_h+w_l) */
1305 	z = w_h - w_l;
1306 	z2 = z * z;
1307 	t1 = z2 * (Et1 + z2 * (Et3 + z2 * Et5));
1308 	t2 = z2 * (Et2 + z2 * Et4);
1309 	t3 = w_h - (w_l - (t1 + z * t2));
1310 	zz.l = S_trail[j] * (one + t3) + S[j] * t3;
1311 	zz.h = S[j];
1312 	return (zz);
1313 }
1314 
1315 /* INDENT OFF */
1316 /*
1317  * kpsin(x)= sin(pi*x)/pi
1318  *                 3        5        7        9        11        13        15
1319  *	= x+ks[0]*x +ks[1]*x +ks[2]*x +ks[3]*x +ks[4]*x  +ks[5]*x  +ks[6]*x
1320  */
1321 static const double ks[] = {
1322 	-1.64493406684822640606569,
1323 	+8.11742425283341655883668741874008920850698590621e-0001,
1324 	-1.90751824120862873825597279118304943994042258291e-0001,
1325 	+2.61478477632554278317289628332654539353521911570e-0002,
1326 	-2.34607978510202710377617190278735525354347705866e-0003,
1327 	+1.48413292290051695897242899977121846763824221705e-0004,
1328 	-6.87730769637543488108688726777687262485357072242e-0006,
1329 };
1330 /* INDENT ON */
1331 
1332 /* assume x is not tiny and positive */
1333 static struct Double
1334 kpsin(double x) {
1335 	double z, t1, t2, t3, t4;
1336 	struct Double xx;
1337 
1338 	z = x * x;
1339 	xx.h = x;
1340 	t1 = z * x;
1341 	t2 = z * z;
1342 	t4 = t1 * ks[0];
1343 	t3 = (t1 * z) * ((ks[1] + z * ks[2] + t2 * ks[3]) + (z * t2) *
1344 		(ks[4] + z * ks[5] + t2 * ks[6]));
1345 	xx.l = t4 + t3;
1346 	return (xx);
1347 }
1348 
1349 /* INDENT OFF */
1350 /*
1351  * kpcos(x)= cos(pi*x)/pi
1352  *                     2        4        6        8        10        12
1353  *	= 1/pi +kc[0]*x +kc[1]*x +kc[2]*x +kc[3]*x +kc[4]*x  +kc[5]*x
1354  */
1355 
1356 static const double one_pi_h = 0.318309886183790635705292970,
1357 		one_pi_l = 3.583247455607534006714276420e-17;
1358 static const double npi_2_h = -1.5625,
1359 		npi_2_l = -0.00829632679489661923132169163975055099555883223;
1360 static const double kc[] = {
1361 	-1.57079632679489661923132169163975055099555883223e+0000,
1362 	+1.29192819501230224953283586722575766189551966008e+0000,
1363 	-4.25027339940149518500158850753393173519732149213e-0001,
1364 	+7.49080625187015312373925142219429422375556727752e-0002,
1365 	-8.21442040906099210866977352284054849051348692715e-0003,
1366 	+6.10411356829515414575566564733632532333904115968e-0004,
1367 };
1368 /* INDENT ON */
1369 
1370 /* assume x is not tiny and positive */
1371 static struct Double
1372 kpcos(double x) {
1373 	double z, t1, t2, t3, t4, x4, x8;
1374 	struct Double xx;
1375 
1376 	z = x * x;
1377 	xx.h = one_pi_h;
1378 	t1 = (double) ((float) x);
1379 	x4 = z * z;
1380 	t2 = npi_2_l * z + npi_2_h * (x + t1) * (x - t1);
1381 	t3 = one_pi_l + x4 * ((kc[1] + z * kc[2]) + x4 * (kc[3] + z *
1382 		kc[4] + x4 * kc[5]));
1383 	t4 = t1 * t1;	/* 48 bits mantissa */
1384 	x8 = t2 + t3;
1385 	t4 *= npi_2_h;	/* npi_2_h is 5 bits const. The product is exact */
1386 	xx.l = x8 + t4;	/* that will minimized the rounding error in xx.l */
1387 	return (xx);
1388 }
1389 
1390 /* INDENT OFF */
1391 static const double
1392 	/* 0.134861805732790769689793935774652917006 */
1393 	t0z1   =  0.1348618057327907737708,
1394 	t0z1_l = -4.0810077708578299022531e-18,
1395 	/* 0.461632144968362341262659542325721328468 */
1396 	t0z2   =  0.4616321449683623567850,
1397 	t0z2_l = -1.5522348162858676890521e-17,
1398 	/* 0.819773101100500601787868704921606996312 */
1399 	t0z3   =  0.8197731011005006118708,
1400 	t0z3_l = -1.0082945122487103498325e-17;
1401 	/* 1.134861805732790769689793935774652917006 */
1402 /* INDENT ON */
1403 
1404 /* gamma(x+i) for 0 <= x < 1  */
1405 static struct Double
1406 gam_n(int i, double x) {
1407 	struct Double rr = {0.0L, 0.0L}, yy;
1408 	double r1, r2, t2, z, xh, xl, yh, yl, zh, z1, z2, zl, x5, wh, wl;
1409 
1410 	/* compute yy = gamma(x+1) */
1411 	if (x > 0.2845) {
1412 		if (x > 0.6374) {
1413 			r1 = x - t0z3;
1414 			r2 = (double) ((float) (r1 - t0z3_l));
1415 			t2 = r1 - r2;
1416 			yy = GT3(r2, t2 - t0z3_l);
1417 		} else {
1418 			r1 = x - t0z2;
1419 			r2 = (double) ((float) (r1 - t0z2_l));
1420 			t2 = r1 - r2;
1421 			yy = GT2(r2, t2 - t0z2_l);
1422 		}
1423 	} else {
1424 		r1 = x - t0z1;
1425 		r2 = (double) ((float) (r1 - t0z1_l));
1426 		t2 = r1 - r2;
1427 		yy = GT1(r2, t2 - t0z1_l);
1428 	}
1429 
1430 	/* compute gamma(x+i) = (x+i-1)*...*(x+1)*yy, 0<i<8 */
1431 	switch (i) {
1432 	case 0:		/* yy/x */
1433 		r1 = one / x;
1434 		xh = (double) ((float) x);	/* x is not tiny */
1435 		rr.h = (double) ((float) ((yy.h + yy.l) * r1));
1436 		rr.l = r1 * (yy.h - rr.h * xh) -
1437 			((r1 * rr.h) * (x - xh) - r1 * yy.l);
1438 		break;
1439 	case 1:		/* yy */
1440 		rr.h = yy.h;
1441 		rr.l = yy.l;
1442 		break;
1443 	case 2:		/* (x+1)*yy */
1444 		z = x + one;	/* may not be exact */
1445 		zh = (double) ((float) z);
1446 		rr.h = zh * yy.h;
1447 		rr.l = z * yy.l + (x - (zh - one)) * yy.h;
1448 		break;
1449 	case 3:		/* (x+2)*(x+1)*yy */
1450 		z1 = x + one;
1451 		z2 = x + 2.0;
1452 		z = z1 * z2;
1453 		xh = (double) ((float) z);
1454 		zh = (double) ((float) z1);
1455 		xl = (x - (zh - one)) * (z2 + zh) - (xh - zh * (zh + one));
1456 		rr.h = xh * yy.h;
1457 		rr.l = z * yy.l + xl * yy.h;
1458 		break;
1459 
1460 	case 4:		/* (x+1)*(x+3)*(x+2)*yy */
1461 		z1 = x + 2.0;
1462 		z2 = (x + one) * (x + 3.0);
1463 		zh = z1;
1464 		__LO(zh) = 0;
1465 		__HI(zh) &= 0xfffffff8;	/* zh 18 bits mantissa */
1466 		zl = x - (zh - 2.0);
1467 		z = z1 * z2;
1468 		xh = (double) ((float) z);
1469 		xl = zl * (z2 + zh * (z1 + zh)) - (xh - zh * (zh * zh - one));
1470 		rr.h = xh * yy.h;
1471 		rr.l = z * yy.l + xl * yy.h;
1472 		break;
1473 	case 5:		/* ((x+1)*(x+4)*(x+2)*(x+3))*yy */
1474 		z1 = x + 2.0;
1475 		z2 = x + 3.0;
1476 		z = z1 * z2;
1477 		zh = (double) ((float) z1);
1478 		yh = (double) ((float) z);
1479 		yl = (x - (zh - 2.0)) * (z2 + zh) - (yh - zh * (zh + one));
1480 		z2 = z - 2.0;
1481 		z *= z2;
1482 		xh = (double) ((float) z);
1483 		xl = yl * (z2 + yh) - (xh - yh * (yh - 2.0));
1484 		rr.h = xh * yy.h;
1485 		rr.l = z * yy.l + xl * yy.h;
1486 		break;
1487 	case 6:		/* ((x+1)*(x+2)*(x+3)*(x+4)*(x+5))*yy */
1488 		z1 = x + 2.0;
1489 		z2 = x + 3.0;
1490 		z = z1 * z2;
1491 		zh = (double) ((float) z1);
1492 		yh = (double) ((float) z);
1493 		z1 = x - (zh - 2.0);
1494 		yl = z1 * (z2 + zh) - (yh - zh * (zh + one));
1495 		z2 = z - 2.0;
1496 		x5 = x + 5.0;
1497 		z *= z2;
1498 		xh = (double) ((float) z);
1499 		zh += 3.0;
1500 		xl = yl * (z2 + yh) - (xh - yh * (yh - 2.0));
1501 						/* xh+xl=(x+1)*...*(x+4) */
1502 		/* wh+wl=(x+5)*yy */
1503 		wh = (double) ((float) (x5 * (yy.h + yy.l)));
1504 		wl = (z1 * yy.h + x5 * yy.l) - (wh - zh * yy.h);
1505 		rr.h = wh * xh;
1506 		rr.l = z * wl + xl * wh;
1507 		break;
1508 	case 7:		/* ((x+1)*(x+2)*(x+3)*(x+4)*(x+5)*(x+6))*yy */
1509 		z1 = x + 3.0;
1510 		z2 = x + 4.0;
1511 		z = z2 * z1;
1512 		zh = (double) ((float) z1);
1513 		yh = (double) ((float) z);	/* yh+yl = (x+3)(x+4) */
1514 		yl = (x - (zh - 3.0)) * (z2 + zh) - (yh - (zh * (zh + one)));
1515 		z1 = x + 6.0;
1516 		z2 = z - 2.0;	/* z2 = (x+2)*(x+5) */
1517 		z *= z2;
1518 		xh = (double) ((float) z);
1519 		xl = yl * (z2 + yh) - (xh - yh * (yh - 2.0));
1520 						/* xh+xl=(x+2)*...*(x+5) */
1521 		/* wh+wl=(x+1)(x+6)*yy */
1522 		z2 -= 4.0;	/* z2 = (x+1)(x+6) */
1523 		wh = (double) ((float) (z2 * (yy.h + yy.l)));
1524 		wl = (z2 * yy.l + yl * yy.h) - (wh - (yh - 6.0) * yy.h);
1525 		rr.h = wh * xh;
1526 		rr.l = z * wl + xl * wh;
1527 	}
1528 	return (rr);
1529 }
1530 
1531 double
1532 tgamma(double x) {
1533 	struct Double ss, ww;
1534 	double t, t1, t2, t3, t4, t5, w, y, z, z1, z2, z3, z5;
1535 	int i, j, k, m, ix, hx, xk;
1536 	unsigned lx;
1537 
1538 	hx = __HI(x);
1539 	lx = __LO(x);
1540 	ix = hx & 0x7fffffff;
1541 	y = x;
1542 
1543 	if (ix < 0x3ca00000)
1544 		return (one / x);	/* |x| < 2**-53 */
1545 	if (ix >= 0x7ff00000)
1546 			/* +Inf -> +Inf, -Inf or NaN -> NaN */
1547 		return (x * ((hx < 0)? 0.0 : x));
1548 	if (hx > 0x406573fa ||	/* x > 171.62... overflow to +inf */
1549 	    (hx == 0x406573fa && lx > 0xE561F647)) {
1550 		z = x / tiny;
1551 		return (z * z);
1552 	}
1553 	if (hx >= 0x40200000) {	/* x >= 8 */
1554 		ww = large_gam(x, &m);
1555 		w = ww.h + ww.l;
1556 		__HI(w) += m << 20;
1557 		return (w);
1558 	}
1559 	if (hx > 0) {		/* 0 < x < 8 */
1560 		i = (int) x;
1561 		ww = gam_n(i, x - (double) i);
1562 		return (ww.h + ww.l);
1563 	}
1564 
1565 	/* negative x */
1566 	/* INDENT OFF */
1567 	/*
1568 	 * compute: xk =
1569 	 *	-2 ... x is an even int (-inf is even)
1570 	 *	-1 ... x is an odd int
1571 	 *	+0 ... x is not an int but chopped to an even int
1572 	 *	+1 ... x is not an int but chopped to an odd int
1573 	 */
1574 	/* INDENT ON */
1575 	xk = 0;
1576 	if (ix >= 0x43300000) {
1577 		if (ix >= 0x43400000)
1578 			xk = -2;
1579 		else
1580 			xk = -2 + (lx & 1);
1581 	} else if (ix >= 0x3ff00000) {
1582 		k = (ix >> 20) - 0x3ff;
1583 		if (k > 20) {
1584 			j = lx >> (52 - k);
1585 			if ((j << (52 - k)) == lx)
1586 				xk = -2 + (j & 1);
1587 			else
1588 				xk = j & 1;
1589 		} else {
1590 			j = ix >> (20 - k);
1591 			if ((j << (20 - k)) == ix && lx == 0)
1592 				xk = -2 + (j & 1);
1593 			else
1594 				xk = j & 1;
1595 		}
1596 	}
1597 	if (xk < 0)
1598 		/* ideally gamma(-n)= (-1)**(n+1) * inf, but c99 expect NaN */
1599 		return ((x - x) / (x - x));		/* 0/0 = NaN */
1600 
1601 
1602 	/* negative underflow thresold */
1603 	if (ix > 0x4066e000 || (ix == 0x4066e000 && lx > 11)) {
1604 		/* x < -183.0 - 11ulp */
1605 		z = tiny / x;
1606 		if (xk == 1)
1607 			z = -z;
1608 		return (z * tiny);
1609 	}
1610 
1611 	/* now compute gamma(x) by  -1/((sin(pi*y)/pi)*gamma(1+y)), y = -x */
1612 
1613 	/*
1614 	 * First compute ss = -sin(pi*y)/pi , so that
1615 	 * gamma(x) = 1/(ss*gamma(1+y))
1616 	 */
1617 	y = -x;
1618 	j = (int) y;
1619 	z = y - (double) j;
1620 	if (z > 0.3183098861837906715377675)
1621 		if (z > 0.6816901138162093284622325)
1622 			ss = kpsin(one - z);
1623 		else
1624 			ss = kpcos(0.5 - z);
1625 	else
1626 		ss = kpsin(z);
1627 	if (xk == 0) {
1628 		ss.h = -ss.h;
1629 		ss.l = -ss.l;
1630 	}
1631 
1632 	/* Then compute ww = gamma(1+y), note that result scale to 2**m */
1633 	m = 0;
1634 	if (j < 7) {
1635 		ww = gam_n(j + 1, z);
1636 	} else {
1637 		w = y + one;
1638 		if ((lx & 1) == 0) {	/* y+1 exact (note that y<184) */
1639 			ww = large_gam(w, &m);
1640 		} else {
1641 			t = w - one;
1642 			if (t == y) {	/* y+one exact */
1643 				ww = large_gam(w, &m);
1644 			} else {	/* use y*gamma(y) */
1645 				if (j == 7)
1646 					ww = gam_n(j, z);
1647 				else
1648 					ww = large_gam(y, &m);
1649 				t4 = ww.h + ww.l;
1650 				t1 = (double) ((float) y);
1651 				t2 = (double) ((float) t4);
1652 						/* t4 will not be too large */
1653 				ww.l = y * (ww.l - (t2 - ww.h)) + (y - t1) * t2;
1654 				ww.h = t1 * t2;
1655 			}
1656 		}
1657 	}
1658 
1659 	/* compute 1/(ss*ww) */
1660 	t3 = ss.h + ss.l;
1661 	t4 = ww.h + ww.l;
1662 	t1 = (double) ((float) t3);
1663 	t2 = (double) ((float) t4);
1664 	z1 = ss.l - (t1 - ss.h);	/* (t1,z1) = ss */
1665 	z2 = ww.l - (t2 - ww.h);	/* (t2,z2) = ww */
1666 	t3 = t3 * t4;			/* t3 = ss*ww */
1667 	z3 = one / t3;			/* z3 = 1/(ss*ww) */
1668 	t5 = t1 * t2;
1669 	z5 = z1 * t4 + t1 * z2;		/* (t5,z5) = ss*ww */
1670 	t1 = (double) ((float) t3);	/* (t1,z1) = ss*ww */
1671 	z1 = z5 - (t1 - t5);
1672 	t2 = (double) ((float) z3);	/* leading 1/(ss*ww) */
1673 	z2 = z3 * (t2 * z1 - (one - t2 * t1));
1674 	z = t2 - z2;
1675 
1676 	/* check whether z*2**-m underflow */
1677 	if (m != 0) {
1678 		hx = __HI(z);
1679 		i = hx & 0x80000000;
1680 		ix = hx ^ i;
1681 		j = ix >> 20;
1682 		if (j > m) {
1683 			ix -= m << 20;
1684 			__HI(z) = ix ^ i;
1685 		} else if ((m - j) > 52) {
1686 			/* underflow */
1687 			if (xk == 0)
1688 				z = -tiny * tiny;
1689 			else
1690 				z = tiny * tiny;
1691 		} else {
1692 			/* subnormal */
1693 			m -= 60;
1694 			t = one;
1695 			__HI(t) -= 60 << 20;
1696 			ix -= m << 20;
1697 			__HI(z) = ix ^ i;
1698 			z *= t;
1699 		}
1700 	}
1701 	return (z);
1702 }
1703