sys7.1-doc-wip/OS/FPUEmulation/Hyperbolic.a
2019-07-27 22:37:48 +08:00

554 lines
12 KiB
Plaintext

;
; File: Hyperbolic.a
;
; Contains: Routines to emulate hyperbolic functions
;
; Originally Written by: Motorola Inc.
; Adapted to Apple/MPW: Jon Okada
;
; Copyright: © 1990, 1991 by Apple Computer, Inc., all rights reserved.
;
; This file is used in these builds: Mac32
;
; Change History (most recent first):
;
; <2> 3/30/91 BG Rolling in Jon Okada's latest changes.
; <1> 12/14/90 BG First checked into TERROR/BBS.
; hyperbolic.a
; Based upon Motorola files 'satanh.sa', 'scosh.sa', 'ssinh.sa', and 'stanh.sa'.
; CHANGE LOG:
; 04 Jan 91 JPO Moved constants T1, T2, and TWO16380 (used in scosh/ssinh)
; to file 'constants.a'. Renamed constant BOUNDS1 (used
; in stanh) to BNDTANH.
;
; satanh
*
* satanh.sa 3.1 12/10/90
*
* The entry point satanh computes the inverse
* hyperbolic tangent of
* an input argument; satanhd does the same except for denormalized
* input.
*
* Input: Double-extended number X in location pointed to
* by address register a0.
*
* Output: The value arctanh(X) returned in floating-point register Fp0.
*
* Accuracy and Monotonicity: The returned result is within 3 ulps in
* 64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
* result is subsequently rounded to double precision. The
* result is provably monotonic in double precision.
*
* Speed: The program satanh takes approximately 270 cycles.
*
* Algorithm:
*
* ATANH
* 1. If |X| >= 1, go to 3.
*
* 2. (|X| < 1) Calculate atanh(X) by
* sgn := sign(X)
* y := |X|
* z := 2y/(1-y)
* atanh(X) := sgn * (1/2) * logp1(z)
* Exit.
*
* 3. If |X| > 1, go to 5.
*
* 4. (|X| = 1) Generate infinity with an appropriate sign and
* divide-by-zero by
* sgn := sign(X)
* atan(X) := sgn / (+0).
* Exit.
*
* 5. (|X| > 1) Generate an invalid operation by 0 * infinity.
* Exit.
*
* Copyright (C) Motorola, Inc. 1990
* All Rights Reserved
*
* THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
* The copyright notice above does not evidence any
* actual or intended publication of such source code.
* satanh IDNT 2,1 Motorola 040 Floating Point Software Package
satanhd:
*--ATANH(X) = X FOR DENORMALIZED X
bra t_extdnrm
satanh:
move.l (a0),d0
move.w 4(a0),d0
ANDI.L #$7FFFFFFF,D0
CMPI.L #$3FFF8000,D0
BGE.B ATANHBIG
*--THIS IS THE USUAL CASE, |X| < 1
*--Y = |X|, Z = 2Y/(1-Y), ATANH(X) = SIGN(X) * (1/2) * LOG1P(Z).
FABS.X (a0),FP0 ;...Y = |X|
FMOVE.X FP0,FP1
FNEG.X FP1 ;...-Y
FADD.X FP0,FP0 ;...2Y
FADD.S #"$3F800000",FP1 ;...1-Y
FDIV.X FP1,FP0 ;...2Y/(1-Y)
move.l (a0),d0
ANDI.L #$80000000,D0
ORI.L #$3F000000,D0 ;...SIGN(X)*HALF
move.l d0,-(sp)
fmovem.x fp0,(a0) ;...overwrite input
move.l d1,-(sp)
clr.l d1
bsr slognp1 ;...LOG1P(Z)
fmove.l (sp)+,fpcr
FMUL.S (sp)+,FP0
bra t_frcinx
ATANHBIG:
FABS.X (a0),FP0 ;...|X|
FCMP.S #"$3F800000",FP0
fbgt t_operr
bra t_dz
; scosh
*
* scosh.sa 3.1 12/10/90
*
* The entry point sCosh computes the hyperbolic cosine of
* an input argument; sCoshd does the same except for denormalized
* input.
*
* Input: Double-extended number X in location pointed to
* by address register a0.
*
* Output: The value cosh(X) returned in floating-point register Fp0.
*
* Accuracy and Monotonicity: The returned result is within 3 ulps in
* 64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
* result is subsequently rounded to double precision. The
* result is provably monotonic in double precision.
*
* Speed: The program sCOSH takes approximately 250 cycles.
*
* Algorithm:
*
* COSH
* 1. If |X| > 16380 log2, go to 3.
*
* 2. (|X| <= 16380 log2) Cosh(X) is obtained by the formulae
* y = |X|, z = exp(Y), and
* cosh(X) = (1/2)*( z + 1/z ).
* Exit.
*
* 3. (|X| > 16380 log2). If |X| > 16480 log2, go to 5.
*
* 4. (16380 log2 < |X| <= 16480 log2)
* cosh(X) = sign(X) * exp(|X|)/2.
* However, invoking exp(|X|) may cause premature overflow.
* Thus, we calculate sinh(X) as follows:
* Y := |X|
* Fact := 2**(16380)
* Y' := Y - 16381 log2
* cosh(X) := Fact * exp(Y').
* Exit.
*
* 5. (|X| > 16480 log2) sinh(X) must overflow. Return
* Huge*Huge to generate overflow and an infinity with
* the appropriate sign. Huge is the largest finite number in
* extended format. Exit.
*
*
* Copyright (C) Motorola, Inc. 1990
* All Rights Reserved
*
* THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
* The copyright notice above does not evidence any
* actual or intended publication of such source code.
* SCOSH IDNT 2,1 Motorola 040 Floating Point Software Package
scoshd:
*--COSH(X) = 1 FOR DENORMALIZED X
FMOVE.S #"$3F800000",FP0
FMOVE.L d1,FPCR
FADD.S #"$00800000",FP0
bra t_frcinx
scosh:
FMOVE.X (a0),FP0 ;...LOAD INPUT
move.l (a0),d0
move.w 4(a0),d0
ANDI.L #$7FFFFFFF,d0
CMPI.L #$400CB167,d0
BGT.B COSHBIG
*--THIS IS THE USUAL CASE, |X| < 16380 LOG2
*--COSH(X) = (1/2) * ( EXP(X) + 1/EXP(X) )
FABS.X FP0 ;...|X|
move.l d1,-(sp)
clr.l d1
fmovem.x fp0,(a0) ;pass parameter to setox
bsr setox ;...FP0 IS EXP(|X|)
FMUL.S #"$3F000000",FP0 ;...(1/2)EXP(|X|)
move.l (sp)+,d1
FMOVE.S #"$3E800000",FP1 ;...(1/4)
FDIV.X FP0,FP1 ;...1/(2 EXP(|X|))
FMOVE.L d1,FPCR
FADD.X fp1,FP0
bra t_frcinx
COSHBIG:
CMPI.L #$400CB2B3,d0
BGT.B COSHHUGE
FABS.X FP0
FSUB.D T1(pc),FP0 ; ...(|X|-16381LOG2_LEAD)
FSUB.D T2(pc),FP0 ; ...|X| - 16381 LOG2, ACCURATE
move.l d1,-(sp)
clr.l d1
fmovem.x fp0,(a0)
bsr setox
fmove.l (sp)+,fpcr
FMUL.X TWO16380(pc),FP0
bra t_frcinx
COSHHUGE:
fmove.l #0,fpsr ;clr N bit if set by source
bclr.b #7,(a0) ;always return positive value
fmovem.x (a0),fp0
bra t_ovfl
; ssinh
*
* ssinh.sa 3.1 12/10/90
*
* The entry point sSinh computes the hyperbolic sine of
* an input argument; sSinhd does the same except for denormalized
* input.
*
* Input: Double-extended number X in location pointed to
* by address register a0.
*
* Output: The value sinh(X) returned in floating-point register Fp0.
*
* Accuracy and Monotonicity: The returned result is within 3 ulps in
* 64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
* result is subsequently rounded to double precision. The
* result is provably monotonic in double precision.
*
* Speed: The program sSINH takes approximately 280 cycles.
*
* Algorithm:
*
* SINH
* 1. If |X| > 16380 log2, go to 3.
*
* 2. (|X| <= 16380 log2) Sinh(X) is obtained by the formulae
* y = |X|, sgn = sign(X), and z = expm1(Y),
* sinh(X) = sgn*(1/2)*( z + z/(1+z) ).
* Exit.
*
* 3. If |X| > 16480 log2, go to 5.
*
* 4. (16380 log2 < |X| <= 16480 log2)
* sinh(X) = sign(X) * exp(|X|)/2.
* However, invoking exp(|X|) may cause premature overflow.
* Thus, we calculate sinh(X) as follows:
* Y := |X|
* sgn := sign(X)
* sgnFact := sgn * 2**(16380)
* Y' := Y - 16381 log2
* sinh(X) := sgnFact * exp(Y').
* Exit.
*
* 5. (|X| > 16480 log2) sinh(X) must overflow. Return
* sign(X)*Huge*Huge to generate overflow and an infinity with
* the appropriate sign. Huge is the largest finite number in
* extended format. Exit.
*
* Copyright (C) Motorola, Inc. 1990
* All Rights Reserved
*
* THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
* The copyright notice above does not evidence any
* actual or intended publication of such source code.
* SSINH IDNT 2,1 Motorola 040 Floating Point Software Package
ssinhd:
*--SINH(X) = X FOR DENORMALIZED X
bra t_extdnrm
ssinh:
FMOVE.x (a0),FP0 ...LOAD INPUT
move.l (a0),d0
move.w 4(a0),d0
move.l d0,a1 ;save a copy of original (compacted) operand
AND.L #$7FFFFFFF,D0
CMP.L #$400CB167,D0
BGT.B SINHBIG
*--THIS IS THE USUAL CASE, |X| < 16380 LOG2
*--Y = |X|, Z = EXPM1(Y), SINH(X) = SIGN(X)*(1/2)*( Z + Z/(1+Z) )
FABS.X FP0 ...Y = |X|
movem.l a1/d1,-(sp)
fmovem.x fp0,(a0)
clr.l d1
bsr setoxm1 ...FP0 IS Z = EXPM1(Y)
fmove.l #0,fpcr
movem.l (sp)+,a1/d1
FMOVE.X FP0,FP1
FADD.S #"$3F800000",FP1 ...1+Z
FMOVE.X FP0,-(sp)
FDIV.X FP1,FP0 ...Z/(1+Z)
MOVE.L a1,d0
AND.L #$80000000,D0
OR.L #$3F000000,D0
FADD.X (sp)+,FP0
MOVE.L D0,-(sp)
fmove.l d1,fpcr
fmul.s (sp)+,fp0 ;last fp inst - possible exceptions set
bra t_frcinx
SINHBIG:
cmp.l #$400CB2B3,D0
bgt t_ovfl
FABS.X FP0
FSUB.D T1(pc),FP0 ...(|X|-16381LOG2_LEAD)
move.l #0,-(sp)
move.l #$80000000,-(sp)
move.l a1,d0
AND.L #$80000000,D0
OR.L #$7FFB0000,D0
MOVE.L D0,-(sp) ...EXTENDED FMT
FSUB.D T2(pc),FP0 ...|X| - 16381 LOG2, ACCURATE
move.l d1,-(sp)
clr.l d1
fmovem.x fp0,(a0)
bsr setox
fmove.l (sp)+,fpcr
fmul.x (sp)+,fp0 ;possible exception
bra t_frcinx
; stanh
*
* stanh.sa 3.1 12/10/90
*
* The entry point sTanh computes the hyperbolic tangent of
* an input argument; sTanhd does the same except for denormalized
* input.
*
* Input: Double-extended number X in location pointed to
* by address register a0.
*
* Output: The value tanh(X) returned in floating-point register Fp0.
*
* Accuracy and Monotonicity: The returned result is within 3 ulps in
* 64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
* result is subsequently rounded to double precision. The
* result is provably monotonic in double precision.
*
* Speed: The program stanh takes approximately 270 cycles.
*
* Algorithm:
*
* TANH
* 1. If |X| >= (5/2) log2 or |X| <= 2**(-40), go to 3.
*
* 2. (2**(-40) < |X| < (5/2) log2) Calculate tanh(X) by
* sgn := sign(X), y := 2|X|, z := expm1(Y), and
* tanh(X) = sgn*( z/(2+z) ).
* Exit.
*
* 3. (|X| <= 2**(-40) or |X| >= (5/2) log2). If |X| < 1,
* go to 7.
*
* 4. (|X| >= (5/2) log2) If |X| >= 50 log2, go to 6.
*
* 5. ((5/2) log2 <= |X| < 50 log2) Calculate tanh(X) by
* sgn := sign(X), y := 2|X|, z := exp(Y),
* tanh(X) = sgn - [ sgn*2/(1+z) ].
* Exit.
*
* 6. (|X| >= 50 log2) Tanh(X) = +-1 (round to nearest). Thus, we
* calculate Tanh(X) by
* sgn := sign(X), Tiny := 2**(-126),
* tanh(X) := sgn - sgn*Tiny.
* Exit.
*
* 7. (|X| < 2**(-40)). Tanh(X) = X. Exit.
*
* Copyright (C) Motorola, Inc. 1990
* All Rights Reserved
*
* THIS IS UNPUBLISHED PROPRIETARY SOURCE CODE OF MOTOROLA
* The copyright notice above does not evidence any
* actual or intended publication of such source code.
* STANH IDNT 2,1 Motorola 040 Floating Point Software Package
X equ FP_SCR5
XDCARE equ X+2
XFRAC equ X+4
SGN equ L_SCR3
V equ FP_SCR6
BNDTANH DC.L $3FD78000,$3FFFDDCE ; 2^(-40), (5/2)LOG2 - label changed <1/4/91, JPO>
stanhd:
*--TANH(X) = X FOR DENORMALIZED X
bra t_extdnrm
stanh:
FMOVE.X (a0),FP0 ...LOAD INPUT
FMOVE.X FP0,X(a6)
move.l (a0),d0
move.w 4(a0),d0
MOVE.L D0,X(a6)
AND.L #$7FFFFFFF,D0
CMP2.L BNDTANH(pc),D0 ...2**(-40) < |X| < (5/2)LOG2 ?
BCS.B TANHBORS
*--THIS IS THE USUAL CASE
*--Y = 2|X|, Z = EXPM1(Y), TANH(X) = SIGN(X) * Z / (Z+2).
MOVE.L X(a6),D0
MOVE.L D0,SGN(a6)
AND.L #$7FFF0000,D0
ADD.L #$00010000,D0 ...EXPONENT OF 2|X|
MOVE.L D0,X(a6)
AND.L #$80000000,SGN(a6)
FMOVE.X X(a6),FP0 ...FP0 IS Y = 2|X|
move.l d1,-(a7)
clr.l d1
fmovem.x fp0,(a0)
bsr setoxm1 ...FP0 IS Z = EXPM1(Y)
move.l (a7)+,d1
FMOVE.X FP0,FP1
FADD.S #"$40000000",FP1 ...Z+2
MOVE.L SGN(a6),D0
FMOVE.X FP1,V(a6)
EOR.L D0,V(a6)
FMOVE.L d1,FPCR ;restore users exceptions
FDIV.X V(a6),FP0
bra t_frcinx
TANHBORS:
CMP.L #$3FFF8000,D0
BLT.B TANHSM
CMP.L #$40048AA1,D0
BGT.W TANHHUGE
*-- (5/2) LOG2 < |X| < 50 LOG2,
*--TANH(X) = 1 - (2/[EXP(2X)+1]). LET Y = 2|X|, SGN = SIGN(X),
*--TANH(X) = SGN - SGN*2/[EXP(Y)+1].
MOVE.L X(a6),D0
MOVE.L D0,SGN(a6)
AND.L #$7FFF0000,D0
ADD.L #$00010000,D0 ...EXPO OF 2|X|
MOVE.L D0,X(a6) ...Y = 2|X|
AND.L #$80000000,SGN(a6)
MOVE.L SGN(a6),D0
FMOVE.X X(a6),FP0 ...Y = 2|X|
move.l d1,-(a7)
clr.l d1
fmovem.x fp0,(a0)
bsr setox ...FP0 IS EXP(Y)
move.l (a7)+,d1
move.l SGN(a6),d0
FADD.S #"$3F800000",FP0 ...EXP(Y)+1
EOR.L #$C0000000,D0 ...-SIGN(X)*2
FMOVE.S d0,FP1 ...-SIGN(X)*2 IN SGL FMT
FDIV.X FP0,FP1 ...-SIGN(X)2 / [EXP(Y)+1 ]
MOVE.L SGN(a6),D0
OR.L #$3F800000,D0 ...SGN
FMOVE.S d0,FP0 ...SGN IN SGL FMT
FMOVE.L d1,FPCR ;restore users exceptions
FADD.X fp1,FP0
bra t_frcinx
TANHSM:
MOVE.W #$0000,XDCARE(a6)
FMOVE.L d1,FPCR ;restore users exceptions
FMOVE.X X(a6),FP0 ;last inst - possible exception set
bra t_frcinx
TANHHUGE:
*---RETURN SGN(X) - SGN(X)EPS
MOVE.L X(a6),D0
AND.L #$80000000,D0
OR.L #$3F800000,D0
FMOVE.S d0,FP0
AND.L #$80000000,D0
EOR.L #$80800000,D0 ...-SIGN(X)*EPS
FMOVE.L d1,FPCR ;restore users exceptions
FADD.S d0,FP0
bra t_frcinx