2 // Copyright (C) 2000, 2001, Intel Corporation
3 // All rights reserved.
5 // Contributed 2/2/2000 by John Harrison, Cristina Iordache, Ted Kubaska,
6 // Bob Norin, Shane Story, and Ping Tak Peter Tang of the Computational
7 // Software Lab, Intel Corporation.
9 // Redistribution and use in source and binary forms, with or without
10 // modification, are permitted provided that the following conditions are
13 // * Redistributions of source code must retain the above copyright
14 // notice, this list of conditions and the following disclaimer.
16 // * Redistributions in binary form must reproduce the above copyright
17 // notice, this list of conditions and the following disclaimer in the
18 // documentation and/or other materials provided with the distribution.
20 // * The name of Intel Corporation may not be used to endorse or promote
21 // products derived from this software without specific prior written
24 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
27 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
28 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
29 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
30 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
31 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
32 // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
33 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 // Intel Corporation is the author of this code, and requests that all
37 // problem reports or change requests be submitted to it directly at
38 // http://developer.intel.com/opensource.
41 //====================================================================
42 // 2/02/00 Initial version
43 // 3/02/00 New algorithm
44 // 4/04/00 Unwind support added
45 // 7/21/00 Fixed quotient=2^{24*m+23}*1.q1...q23 1 bug
46 // 8/15/00 Bundle added after call to __libm_error_support to properly
47 // set [the previously overwritten] GR_Parameter_RESULT.
48 //11/29/00 Set FR_Y to f9
51 //====================================================================
52 // long double remainderl(long double,long double);
54 // Overview of operation
55 //====================================================================
56 // remainder(a,b)=a-i*b,
57 // where i is an integer such that, if b!=0 and a is finite,
58 // |a/b-i|<=1/2. If |a/b-i|=1/2, i is even.
61 //====================================================================
62 // a). eliminate special cases
63 // b). if |a/b|<0.25 (first quotient estimate), return a
64 // c). use single precision divide algorithm to get quotient q
65 // rounded to 24 bits of precision
66 // d). calculate partial remainders (using both q and q-ulp);
67 // select one and RZ(a/b) based on the sign of |a|-|b|*q
68 // e). if the exponent difference (exponent(a)-exponent(b))
69 // is less than 24 (quotient estimate<2^{24}-2), use RZ(a/b)
70 // and sticky bits to round to integer; exit loop and
71 // calculate final remainder
72 // f). if exponent(a)-exponent(b)>=24, select new value of a as
73 // the partial remainder calculated using RZ(a/b);
77 //====================================================================
78 // a=+/- Inf, or b=+/-0: return NaN, call libm_error_support
79 // a=NaN or b=NaN: return NaN
81 #include "libm_support.h"
85 //====================================================================
86 // Predicate registers: p6-p14
87 // General registers: r2,r3,r28,r29,r32 (ar.pfs), r33-r39
88 // Floating point registers: f6-f15,f32
100 GR_Parameter_RESULT = r39
101 GR_Parameter_TAG = r40
118 .type __remainderl,@function
125 alloc r32=ar.pfs,1,4,4,0
147 cmp.eq p11,p10=r29,r0;;
150 // Y +-NAN, +-inf, +-0? p11
153 (p10) fclass.m p11,p10 = f9, 0xe7
156 // qnan snan inf norm unorm 0 -+
159 // X +-NAN, +-inf, ? p9
162 (p0) fclass.m.unc p9,p8 = f8, 0xe3
173 cmp.eq.unc p7,p0=r0,r0
176 frcpa.s1 f10,p6=f13,f14
179 // Y +-NAN, +-inf, +-0? p11
183 (p10) fclass.nm p11,p0 = f9, 0xff
187 // qnan snan inf norm unorm 0 -+
190 // X +-NAN, +-inf, ? p9
194 (p8) fclass.nm p9,p0 = f8, 0xff
199 (p9) br.cond.spnt L(FREM_X_NAN_INF)
200 (p11) br.cond.spnt L(FREM_Y_NAN_INF_ZERO)
204 // set D flag if a (f8) is denormal
213 // q0 = a * y0 in f15
214 (p6) fma.s1 f12=f13,f10,f0
219 // e0 = 1 - b * y0 in f7
220 (p6) fnma.s1 f7=f14,f10,f1
231 fma.s1 f15=f12,f7,f12
237 // e1 = e0 * e0 + E in f7
238 (p6) fma.s1 f7=f7,f7,f32
242 (p7) getf.exp r29=f12
251 // q2 = q1 + e1 * q1 in f11
252 (p6) fma.s.s1 f11=f7,f15,f15
257 // q2 = q1 + e1 * q1 in f6
258 (p6) fma.s1 f6=f7,f15,f15
265 // q<1/4 ? (i.e. expon< -2)
266 (p7) cmp.gt p7,p0=r28,r29
273 // if |a/b|<1/4, set D flag before returning
274 (p7) fma.s0 f9=f9,f0,f8
279 // can be combined with bundle above if sign of 0 or
280 // FTZ enabled are not important
281 (p7) fmerge.s f8=f8,f9
282 // return if |a|<4*|b| (estimated quotient < 1/4)
283 (p7) br.ret.spnt b0;;
288 // set f8 to current a value | sign
294 // last step ? (q<2^{23})
295 fcmp.lt.unc.s1 p0,p12=f6,f12
301 fnma.s1 f6=f14,f11,f13
306 // q'=q-q*(1.25*2^{-24}) (q'=q-ulp)
307 fnma.s.s1 f15=f11,f15,f11
312 cmp.eq p11,p14=r2,r28
316 .pred.rel "mutex",p11,p14
319 // if exp_q=2^23, then r=a-b*2^{23}
320 (p11) fnma.s1 f13=f12,f14,f13
326 (p14) fnma.s1 f13=f14,f15,f13
331 // r>0 iff q=RZ(a/b) and inexact
332 fcmp.gt.unc.s1 p8,p0=f6,f0
336 // r<0 iff q'=RZ(a/b) and inexact
337 (p14) fcmp.lt.unc.s1 p9,p10=f6,f0
341 .pred.rel "mutex",p8,p9
344 // (p8) Q=q+(last iteration ? sticky bits:0)
345 // i.e. Q=q+q*x (x=2^{-32} or 0)
346 (p8) fma.s1 f11=f11,f7,f11
350 // (p9) Q=q'+(last iteration ? sticky bits:0)
351 // i.e. Q=q'+q'*x (x=2^{-32} or 0)
352 (p9) fma.s1 f11=f15,f7,f15
358 // (p9) set r=r2 (new a, if not last iteration)
361 (p12) br.cond.sptk L(remloop24);;
390 // This can be removed if sign of 0 is not important
391 // get remainder using sf1
392 fnma.s1 f12=f9,f11,f8
404 // This can be removed if sign of 0 is not important
405 fcmp.eq.unc.s1 p8,p0=f12,f0
410 // if f8=0, set sign correctly
411 // This can be removed if sign of 0 is not important
412 (p8) fmerge.s f8=f7,f8
429 fcmp.eq.unc.s1 p11,p0=f10,f0
436 (p11) br.cond.spnt L(FREM_Y_ZERO);;
439 // X infinity? Return QNAN indefinite
442 (p0) fclass.m.unc p8,p0 = f8, 0x23
445 // X infinity? Return QNAN indefinite
448 (p0) fclass.m.unc p11,p0 = f8, 0x23
454 (p8) fclass.m.unc p0,p8=f9,0xc3
459 // also set Denormal flag if necessary
460 (p8) fnma.s0 f9=f9,f1,f9
465 (p8) frcpa.s0 f8,p7 = f8,f8
483 (p11) br.cond.spnt L(EXP_ERROR_RETURN);;
492 L(FREM_Y_NAN_INF_ZERO):
496 (p0) fclass.m.unc p7,p0 = f9, 0x23
503 (p7) br.ret.spnt b0 ;;
509 (p0) fclass.m.unc p9,p10 = f9, 0xc3
514 (p10) fclass.nm p9,p0 = f9, 0xff
521 (p9) br.ret.spnt b0 ;;
525 // Y zero? Must be zero at this point
526 // because it is the only choice left.
527 // Return QNAN indefinite
532 (p0) fclass.m.unc p9,p10 = f8, 0xc3
537 (p10) fclass.nm p9,p10 = f8, 0xff
543 (p9) frcpa f11,p7=f8,f0
548 (p10) frcpa f11,p7 = f0,f0
554 (p0) fmerge.s f10 = f8, f8
560 (p0) fma f8=f11,f1,f0
567 (p0) mov GR_Parameter_TAG = 123
569 (p0) br.sptk __libm_error_region;;
573 ASM_SIZE_DIRECTIVE(remainderl)
575 ASM_SIZE_DIRECTIVE(__remainderl)
578 .proc __libm_error_region
582 add GR_Parameter_Y=-32,sp // Parameter 2 value
584 .save ar.pfs,GR_SAVE_PFS
585 mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
589 add sp=-64,sp // Create new stack
591 mov GR_SAVE_GP=gp // Save gp
594 stfe [GR_Parameter_Y] = FR_Y,16 // Save Parameter 2 on stack
595 add GR_Parameter_X = 16,sp // Parameter 1 address
597 mov GR_SAVE_B0=b0 // Save b0
601 stfe [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
602 add GR_Parameter_RESULT = 0,GR_Parameter_Y
603 nop.b 0 // Parameter 3 address
606 stfe [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
607 add GR_Parameter_Y = -16,GR_Parameter_Y
608 br.call.sptk b0=__libm_error_support# // Call error handling function
613 add GR_Parameter_RESULT = 48,sp
616 ldfe f8 = [GR_Parameter_RESULT] // Get return result off stack
618 add sp = 64,sp // Restore stack pointer
619 mov b0 = GR_SAVE_B0 // Restore return address
622 mov gp = GR_SAVE_GP // Restore gp
623 mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
624 br.ret.sptk b0 // Return
627 .endp __libm_error_region
628 ASM_SIZE_DIRECTIVE(__libm_error_region)
631 .type __libm_error_support#,@function
632 .global __libm_error_support#