[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: ieee compliance

On Mon, Sep 18, 2000 at 08:32:48PM -0400, R Clint Whaley wrote:
> First, does anyone know of the arithmetic done by SSE/SSE2/3DNow! is
> ieee compliant or not?  I particularly wonder about overflow, underflow,
> nan handling, etc.  Once we get this stuff in an official release, it's

A quote from the "Intel  Architecture Optimization Reference Manual"
for 1999:

"Streaming SIMD Extensions fully support the IEEE Standard 754 for
Binary Floating-Point Architecture."

Furthermore, the LDMXCSR (Load Streaming SIMD Extension
Control/Status) instruction allows you to set the behaviour for
over/underflow. I think the default is 754 compliance, but some bits
allow you to change the behaviour to be faster but non-compliant.

> definitely a question we'll have to answer, and I thought if any of you guys
> actually doing the work knew the answer I might save myself the effort of
> figuring it out . . .
Hope this clears it up for the intel stuff.


-Doug  -- http://beaker.anu.edu.au, Ph:(02) 6279-8608, Fax:(02) 6279-8651
Glib's Fourth Law of Unreliability : Investment in reliability will 
increase until it exceeds the probable cost of errors, or until someone 
insists on getting some work done.