tech-userlevel archive

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

Re: Support for non-default rounding modes in the softfloat layer



Martin Husemann <martin%duskware.de@localhost> writes:

> On Tue, Apr 30, 2013 at 01:28:47PM +0200, ????? ???????????? wrote:
>> A lot of scientific software uses non-default rounding modes as
>> optimisation short cut. but then this is software which runs on high
>> end hardware a la SPARC T5, SPARC64, NEC SX9 or the very latest
>> IA32/AMD64 hardware.
>
> Ok, but the basic questions are:
>
>  - does it switch between non standard rounding modes often?

Does the standard specify a default, or is that up to the
implementation?

>  - does it use different rounding modes in different threads?

My guess from limited experience is that a single big program has a
planned rounding mode and it just gets set once for the program.

> But, what I wonder most:
>
>  - would the per-thread rounding mode penalty even matter if doing softfloat?

Probably not.  But can you point to something in standards or existing
practice that makes per-thread rounding mode make any sense at all?
This discussion is the first I've heard of the concept.

Attachment: pgpRSZbeyTR7S.pgp
Description: PGP signature



Home | Main Index | Thread Index | Old Index