the LCAS and C

[Nelson H.F. Beebe]

Paul Eggert writes:

>> ...
>> With respect to C and IEEE floating point, the major problems with the
>> Language Compatible Arithmetic Standard (LCAS) are notification and
>> documentation of translation.  The first problem is the most severe,
>> because the LCAS is incompatible with the IEEE 754 default.
>> ...

The IBM RS/6000 IEEE 754 floating-point implementation is non-stop; no
exceptions are taken when Inf, NaN, or underflow are generated.
The only way to isolate the occurrence of one is to bracket statements
with calls to a library routine that tests a flags register for bits
indicating exceptional conditions.  A portion of the relevant man page
is appended to this message.


>> ...
>> Section B.3 of the LCAS claims that `char' is always smaller than
>> `int', but the C standard does not support this claim, and permits
>> `char' to be the same size as `int'.
>> ...

An indeed, this has been the case.  On the DEC-20, PCC-20 stores char,
short, int, and long as 36-bit fullword quantities.  It seems likely
that other word-oriented architectures may have done the same.  KCC-20
uses 9-bit bytes for char, and tight loops over char* arrays run a few
times faster with PCC than with KCC, because word-oriented accesses
faster than byte-oriented ones on that machine (even though it DOES
have byte access instructions).

========================================================================

Here is part of the output of "man fp_invalid_op" on the IBM RS/6000
(AIX 3.1); the point noted above is marked by the !! prefix:

>> ...
>>    Purpose
>> 
>>       Tests to see if a floating-point exception has occurred.
>> 
>>    Syntax
>> 
>>       #include<float.h>
>>       #include<fpxcp.h>
>> 
>>       int fp_invalid_op()
>>       int fp_divbyzero()
>>       int fp_overflow()
>>       int fp_underflow()
>>       int fp_inexact()
>>       int fp_any_xcp()
>>       int fp_iop_snan()
>>       int fp_iop_infsinf()
>>       int fp_iop_infdinf()
>>       int fp_iop_zrdzr()
>>       int fp_iop_infmzr()
>>       int fp_iop_invcmp()
>> 
>>     Description
>> 
!!>>       The RISC System/6000 currently does not generate an interrupt
!!>>   for  floating-point  exceptions.  Therefore, the common method of
!!>>   catching  the  signal  SIGFPE  and  calling  an  appropriate trap
!!>>   handler to identify the floating-point exception  is not support-
!!>>   ed.
>> 
>>       These subroutines aid in determining when an exception has
>>   occurred and the exception type.  These subroutines can be called
>>   explicitly around blocks of code  that may cause a floating-point
>>   exception.
>> 
>>     Return Values
>> 
>>       The fp_invalid_op subroutine returns 1 if a floating-point
>>   invalid operation exception status flag is set.   Otherwise, 0 is
>>   returned.
>> 
>>       The fp_divbyzero subroutine returns 1 if a floating-point
>>   divide by zero exception status flag is set.  Otherwise, 0 is re-
>>   turned.
>> 
>>       The fp_overflow subroutine returns 1 if a floating-point
>>   overflow exception status flag is set.  Otherwise, 0 is returned.
>> 
>>       The fp_underflow subroutine returns 1 if a floating-point un-
>>   derflow exception status flag is set.  Otherwise, 0 is returned.
>> 
>>       The fp_inexact subroutine returns 1 if a floating-point inex-
>>   act exception status flag is set.  Otherwise, 0 is returned.
>> 
>>       The fp_any_xcp subroutine returns 1 if a floating-point in-
>>   valid operation,  divide by zero, overflow, underflow, or inexact
>>   exception status flag is set.  Otherwise, 0 is returned.
>> 
>>       The following routines are available for the AIX for RISC
>>   System/6000 platform only:
>> 
>>       The fp_iop_snan subroutine returns 1 if a floating-point in-
>>   valid operation exception  status flag is set due to a signalling
>>   NaN (NaNS).  Otherwise, 0 is returned.
>> 
>>       The fp_iop_infsinf subroutine returns 1 if a floating-point
>>   invalid operation exception status flag is set due  to a INF-INF.
>>   Otherwise, 0 is returned.
>> 
>>       The fp_iop_infdinf subroutine returns 1 if a floating-point
>>   invalid  operation exception status flag is set due to a INF/INF.
>>   Otherwise, 0 is returned.
>> 
>>       The fp_iop_zrdzr subroutine returns 1 if a floating-point in-
>>   valid  operation exception status flag is set due to  a  0.0/0.0.
>>   Otherwise, 0 is returned.
>> 
>>       The fp_iop_infxzr subroutine returns 1 if a floating-point
>>   invalid operation  exception status flag is set due to a INF*0.0.
>>   Otherwise, 0 is returned.
>> 
>>       The fp_iop_invcmp subroutine returns 1 if a floating-point
>>   invalid operation exception status flag is set due to  a  compare
>>   involving a NaN.  Otherwise, 0 is returned.
>> 
>> ...


========================================================================
Nelson H.F. Beebe
Center for Scientific Computing
Department of Mathematics
220 South Physics Building
University of Utah
Salt Lake City, UT 84112
 Tel: (801) 581-5254
 FAX: (801) 581-4148
 Internet: beebeamath.utah.edu
========================================================================