brian@utradio.UUCP (07/21/87)
I apologize if this is in the wrong newsgroup.
We are running a Sun 3/180 with a floating point accelerator (Weitek chipset).
Our machine is mainly for running a large radio astronomy computing package,
which has some very floating point bound subroutines. This package was
developed on other machines, and the floating point subroutines are generic,
they have been in no way optimized, other than with the -O compiler switch,
for the sun/fpa.
My question: Is it possible to realize significant code speed ups (say > 20%)
by playing tricks with the fpa? What are the most fruitful avenues to explore?
Can the chips be managed to do some pipelined operations?
If you have optimized codes I'd be very interested in taking a look at them.
A hot fft would be extremely useful, and would earn you our undying gratitude!
Please respond by mail, and I will summarize to the net.
--
Brian Glendenning UUCP - {allegra,decvax,pyramid}!utgpu!radio!brian
brian@radio.toronto.edu BITNET - glendenn@utorphys.bitnet