Return to tan.s CVS log

Up to [CSRG BSD Unix] / 43BSD / usr.lib / libm / VAX

Annotation of 43BSD/usr.lib/libm/VAX/tan.s, revision 1.1.1.1

1.1       root        1: # 
                      2: # Copyright (c) 1985 Regents of the University of California.
                      3: # 
                      4: # Use and reproduction of this software are granted  in  accordance  with
                      5: # the terms and conditions specified in  the  Berkeley  Software  License
                      6: # Agreement (in particular, this entails acknowledgement of the programs'
                      7: # source, and inclusion of this notice) with the additional understanding
                      8: # that  all  recipients  should regard themselves as participants  in  an
                      9: # ongoing  research  project and hence should  feel  obligated  to report
                     10: # their  experiences (good or bad) with these elementary function  codes,
                     11: # using "sendbug 4bsd-bugs@BERKELEY", to the authors.
                     12: #
                     13: 
                     14: # @(#)tan.s    1.1 (Berkeley) 8/21/85
                     15: 
                     16: #  This is the implementation of Peter Tang's double precision  
                     17: #  tangent for the VAX using Bob Corbett's argument reduction.
                     18: #  
                     19: #  Notes:
                     20: #       under 1,024,000 random arguments testing on [0,2*pi] 
                     21: #       tan() observed maximum error = 2.15 ulps
                     22: #
                     23: # double tan(arg)
                     24: # double arg;
                     25: # method: true range reduction to [-pi/4,pi/4], P. Tang  &  B. Corbett
                     26: # S. McDonald, April 4,  1985
                     27: #
                     28:        .globl  _tan
                     29:        .text
                     30:        .align  1
                     31: 
                     32: _tan:  .word   0xffc           # save r2-r11
                     33:        movq    4(ap),r0
                     34:        bicw3   $0x807f,r0,r2
                     35:        beql    1f              # if x is zero or reserved operand then return x
                     36: #
                     37: # Save the PSL's IV & FU bits on the stack.
                     38: #
                     39:        movpsl  r2
                     40:        bicw3   $0xff9f,r2,-(sp)
                     41: #
                     42: #  Clear the IV & FU bits.
                     43: #
                     44:        bicpsw  $0x0060
                     45:        jsb     libm$argred
                     46: #
                     47: #  At this point,
                     48: #         r0  contains the quadrant number, 0, 1, 2, or 3;
                     49: #      r2/r1  contains the reduced argument as a D-format number;
                     50: #         r3  contains a F-format extension to the reduced argument;
                     51: #
                     52: #  Save  r3/r0  so that we can call cosine after calling sine.
                     53: #
                     54:        movq    r2,-(sp)
                     55:        movq    r0,-(sp)
                     56: #
                     57: #  Call sine.  r4 = 0  implies sine.
                     58: #
                     59:        movl    $0,r4
                     60:        jsb     libm$sincos
                     61: #
                     62: #  Save  sin(x)  in  r11/r10 .
                     63: #
                     64:        movd    r0,r10
                     65: #
                     66: #  Call cosine.  r4 = 1  implies cosine.
                     67: #
                     68:        movq    (sp)+,r0
                     69:        movq    (sp)+,r2
                     70:        movl    $1,r4
                     71:        jsb     libm$sincos
                     72:        divd3   r0,r10,r0
                     73:        bispsw  (sp)+
                     74: 1:     ret

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.