kwok@iris.ucdavis.edu (Conrad Kwok) (03/05/88)
[ This is the second posting of the same message because I believe
that the first posting is lost for some reason. Sorry if you have
already seen this ]
Recently I have purchased a 80287 for my 286 accelerator card.
After installation I run several old programs which do a lot of
floating point arithmetic using the emulation library. The library
knows the present of my 80287 and use it instead of software emulation now.
Very surprisingly, the error of the result is much much higher than before.
The error increase from less than 1e-7 to almost 1e-4.
Is there any explanation? It seems to me that they are using float instead
of double in all floating point operations when 80287 is used. If this is
true, is there any switch in MSC5 to force it to use double?
Please email the answer. I will summarize the answer if there are any
interesting responses.
-- Conrad Kwok
internet: kwok@iris.ucdavis.edu
csnet: kwok@ucd.csnet
csnet: kwok%iris.ucdavis.edu@csnet.relay
uucp: {ucbvax, uunet, ... }!ucdavis!iris!kwok