Return to README CVS log

Up to [CSRG BSD Unix] / 43BSDReno / lib / libm / mc68881

Annotation of 43BSDReno/lib/libm/mc68881/README, revision 1.1.1.1

1.1       root        1: These routines were produced by someone who is not a great authority on
                      2: floating point, and may not be entirely correct.  Where possible I tested
                      3: the special cases for routines.
                      4: 
                      5: The directory testieee contains test programs for IEEE-format machines.
                      6: I took a stab at making them work on the vax, but gave up as dealing with
                      7: exceptions (e.g. underflow, overflow, reserved operand) was just too tedious.
                      8: 
                      9: Note: it is possible to build a library with MACHINE=ieee but a couple
                     10: of warnings:
                     11: 
                     12:        Be careful when compiling floor.c.  These routines rely on
                     13:        certain variables being only double precision.  If these
                     14:        variables are placed in 68881 registers, they will be extended
                     15:        precision and the routines will produce incorrect results.
                     16:        Unless your compiler does its own register allocation, this
                     17:        is not likely to be a problem as none of the variables in
                     18:        question are declared "register".  If you are using GCC
                     19:        you can specify -ffloat-store to avoid this problem.
                     20: 
                     21:        The C version of drem() in ieee/support.c appears to compute
                     22:        the incorrect results for drem(+-1, +-2).  It yields 1 when
                     23:        it should be -1 and -1 when it should be 1.  "should be" is
                     24:        based on what the VAX version yields and by cranking through
                     25:        the formula.
                     26: 
                     27: If you do build using MACHINE=ieee and run the tests in testieee you
                     28: will note that some routines return errors:
                     29: 
                     30:        floor/ceil/rint report that they got 0 when expecting -0.
                     31:        Don't really know which is correct, is floor(-0) == 0 or -0?
                     32:        For C it shouldn't really matter since 0 is the same as -0
                     33:        in comparisons.
                     34: 
                     35:        scalb(-1, -2100) returns 0 instead of -0.  2 ** -2100 is
                     36:        effectively 0 but -anything * 0 == -0 according to the 68881.
                     37:        Similarly for scalb(-pi, 2100).  It returns INF instead
                     38:        of -INF.  2 ** 2100 is effectively INF but -anything * INF
                     39:        is -INF.  What is correct?
                     40: 
                     41:        drem(+-1, +-2) fails as mentioned above.  This is a real error.
                     42: 
                     43: ----
                     44: Mike Hibler
                     45: U of Utah CS Dept.
                     46: [email protected]

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