In message <[EMAIL PROTECTED]>
Russell King - ARM Linux <[EMAIL PROTECTED]> wrote:
> Peter Naulls writes:
> > I tried out a Java benchmark, to test speed of FP - with this benchmark,
> > it appears to be nearly twice as fast, which is fairly impressive.
> >
> > Unfortunately, the benchmark uses exp() to calculate the final result,
> > which makes it result wrong, as it appears to be misbehaving:
>
> Under AcornFPE with the old a.out stuff:
>
> 54.598150
> 148.413159
> 403.428793
>
> NWFPE doesn't implement exp(), so the bug must be in glibc.
I'm afraid not (and I apologise for not checking this earlier) - things
work as expected under 2.4.1. Even though exp() is in glibc, it's
reasonable to assume that it relies on some FP behaviour provided by the
kernel.
I'm using 2.4.2-pre3-rmk1 to those who it's not clear.
Is there something wrong with the Fast FP patch listed on the website
in terms of the pre3 patch? My source lacks the help on this option, and
I'm able to select both NWFPE and FASTFPE. Disabling the latter causes
the built kernel to throw undefined instructions on boot.
hth, Peter
--
------------------------------------------------------------------------
Peter Naulls - [EMAIL PROTECTED]
Homepage - http://free.prohosting.com/~chocky/
Java for RISC OS and ARM - http://www.java.riscos.org.uk/
Debian Linux on RiscPCs - http://www.erble.freeserve.co.uk/
------------------------------------------------------------------------
_______________________________________________
http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm
Please visit the above address for information on this list.
