double rounding in x86

Chris Hinds uunet!wtkatz.sps.mot.com!chinds
Thu Aug 31 13:18:44 PDT 1995


> In a previous message, Chris Hinds (uunet!wtkatz.sps.mot.com!chindsauunet.uu.net)
> write:
>   >True statement for a 68681 or 68682. However, on the 68040 and 68060 the
>   >penalties for emulation of a single-precision or double-precision machine
>   >were eliminated. In other words, setting the rounding precision to single
>   >or double does not degrade performance of native instructions or emulated
>   >instructions.
> 
> According to the 68040 User's Manual, there is still a penalty in rounding to
> single or double precision on the 68040 in terms of performance.  Maybe this
> manual is mistaken.  (Mr. Hinds, can you verify this?)
> 
> - Sam Figueroa (figueroaacs.nyu.edu)
> 

The initial printing of the 68040 manual contained this error, copied from
the 68881/68882 manual. If you look in the M68040 manual (M68040UM/AD, 1992)
p. 9-11, the statement reflects my original posting.

Chris Hinds (chindsawtkatz.sps.mot.com)



More information about the Numeric-interest mailing list