|
|
1.1 root 1: # Copyright (c) 1987 Regents of the University of California.
2: # All rights reserved.
3: #
4: # Redistribution and use in source and binary forms are permitted
5: # provided that the above copyright notice and this paragraph are
6: # duplicated in all such forms and that any documentation,
7: # advertising materials, and other materials related to such
8: # distribution and use acknowledge that the software was developed
9: # by the University of California, Berkeley. The name of the
10: # University may not be used to endorse or promote products derived
11: # from this software without specific prior written permission.
12: # THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
13: # IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
14: # WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
15: #
16: # All recipients should regard themselves as participants in an ongoing
17: # research project and hence should feel obligated to report their
18: # experiences (good or bad) with these elementary function codes, using
19: # the sendbug(8) program, to the authors.
20: #
21: # @(#)cbrt.s 5.4 (Berkeley) 6/30/88
22: #
23: .data
24: .align 2
25: _sccsid:
26: .asciz "@(#)cbrt.s 5.4 (ucb.elefunt) 6/30/88"
27:
28: # double cbrt(double arg)
29: # W. Kahan, 10/13/80. revised 1/13/84 for keeping sign symmetry
30: # Re-coded in tahoe assembly language by Z. Alex Liu (7/13/87)
31: # Max error less than 0.667 ULPs _if_ +,-,*,/ were all correctly rounded...
32: .globl _cbrt
33: .globl _d_cbrt
34: .globl _dcbrt_
35: .text
36: .align 2
37: _cbrt:
38: _d_cbrt:
39: .word 0x01fc # save r2-r8
40: movl 4(fp),r0 # r0:r1 = x
41: movl 8(fp),r1
42: brb 1f
43: _dcbrt_:
44: .word 0x01fc # save r2-r8
45: movl 4(fp),r8
46: movl (r8),r0
47: movl 4(r8),r1 # r0:r1 = x
48:
49: 1: andl3 $0x7f800000,r0,r2 # biased exponent of x
50: beql return # dcbrt(0)=0 dcbrt(res)=res. operand
51: andl3 $0x80000000,r0,r8 # r8 has sign(x)
52: xorl2 r8,r0 # r0 is abs(x)
53: movl r0,r2 # r2 has abs(x)
54: divl2 $3,r2 # rough dcbrt with bias/3
55: addl2 B,r2 # restore bias, diminish fraction
56: ldf r2 # acc = |q|=|dcbrt| to 5 bits
57: mulf r2 # acc = qq
58: divf r0 # acc = qq/|x|
59: mulf r2 # acc = qqq/|x|
60: addf C # acc = C+qqq/|x|
61: stf r3 # r3 = s = C+qqq/|x|
62: ldf D # acc = D
63: divf r3 # acc = D/s
64: addf E # acc = E+D/s
65: addf r3 # acc = s+E+D/s
66: stf r3 # r3 = s+E+D/s
67: ldf F # acc = F
68: divf r3 # acc = F/(s+E+D/s)
69: addf G # acc = G+F/(s+E+D/s)
70: mulf r2 # acc = q*(G+F/(s+E+D/s)) = new q to 23 bits
71: stf r2 # r2 = q*(G+F/(s+E+D/s)) = new q to 23 bits
72: clrl r3 # r2:r3 = q as double float
73: ldd r2 # acc = q as double float
74: muld r2 # acc = qq exactly
75: std r4 # r4:r5 = qq exactly
76: ldd r0 # acc = |x|
77: divd r4 # acc = |x|/(q*q) rounded
78: std r0 # r0:r1 = |x|/(q*q) rounded
79: subd r2 # acc = |x|/(q*q)-q exactly
80: std r6 # r6:r7 = |x|/(q*q)-q exactly
81: movl r2,r4
82: clrl r5 # r4:r5 = q as double float
83: addl2 $0x800000,r4 # r4:r5 = 2*q
84: ldd r4 # acc = 2*q
85: addd r0 # acc = 2*q+|x|/(q*q)
86: std r4 # r4:r5 = 2*q+|x|/(q*q)
87: ldd r6 # acc = |x|/(q*q)-q
88: divd r4 # acc = (|x|/(q*q)-q)/(2*q+|x|/(q*q))
89: muld r2 # acc = q*(|x|/(q*q)-q)/(2*q+|x|/(q*q))
90: addd r2 # acc = q+q*(|x|/(q*q)-q)/(2*q+|x|/(q*q))
91: std r0 # r0:r1 = |result|
92: orl2 r8,r0 # restore the sign bit
93: return: ret # error less than 0.667ULPs?
94:
95: .data
96: .align 2
97: B : .long 721142941 #(86-0.03306235651)*(2^23)
98: .align 2
99: C: .long 0x400af8b0 #.float 0f0.5428571429 # 19/35
100: .align 2
101: D: .long 0xc0348ef1 #.float 0f-0.7053061224 # -864/1225
102: .align 2
103: E: .long 0x40b50750 #.float 0f1.414285714 # 99/70
104: .align 2
105: F: .long 0x40cdb6db #.float 0f1.607142857 # 45/28
106: .align 2
107: G: .long 0x3fb6db6e #.float 0f0.3571428571 # 5/14
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.