2 | satanh.sa 3.3 12/19/90
4 | The entry point satanh computes the inverse
5 | hyperbolic tangent of
6 | an input argument; satanhd does the same except for denormalized
9 | Input: Double-extended number X in location pointed to
10 | by address register a0.
12 | Output: The value arctanh(X) returned in floating-point register Fp0.
14 | Accuracy and Monotonicity: The returned result is within 3 ulps in
15 | 64 significant bit, i.e. within 0.5001 ulp to 53 bits if the
16 | result is subsequently rounded to double precision. The
17 | result is provably monotonic in double precision.
19 | Speed: The program satanh takes approximately 270 cycles.
24 | 1. If |X| >= 1, go to 3.
26 | 2. (|X| < 1) Calculate atanh(X) by
30 | atanh(X) := sgn * (1/2) * logp1(z)
33 | 3. If |X| > 1, go to 5.
35 | 4. (|X| = 1) Generate infinity with an appropriate sign and
38 | atan(X) := sgn / (+0).
41 | 5. (|X| > 1) Generate an invalid operation by 0 * infinity.
45 | Copyright (C) Motorola, Inc. 1990
48 | For details on the license for this file, please see the
49 | file, README, in this same directory.
51 |satanh idnt 2,1 | Motorola 040 Floating Point Software Package
63 |--ATANH(X) = X FOR DENORMALIZED X
75 |--THIS IS THE USUAL CASE, |X| < 1
76 |--Y = |X|, Z = 2Y/(1-Y), ATANH(X) = SIGN(X) * (1/2) * LOG1P(Z).
78 fabsx (%a0),%fp0 | ...Y = |X|
81 faddx %fp0,%fp0 | ...2Y
82 fadds #0x3F800000,%fp1 | ...1-Y
83 fdivx %fp1,%fp0 | ...2Y/(1-Y)
86 oril #0x3F000000,%d0 | ...SIGN(X)*HALF
89 fmovemx %fp0-%fp0,(%a0) | ...overwrite input
92 bsr slognp1 | ...LOG1P(Z)
98 fabsx (%a0),%fp0 | ...|X|
99 fcmps #0x3F800000,%fp0