Re: [9fans] floating point

Re: [9fans] floating point

[Andrew Simmons]

 > Shortish, sensible, benchmarks would be a appreciated.

Not exactly a benchmark, and it may be neither short nor sensible, but I
bodged up a version of Kahan's Paranoia to run under Plan 9 a while back.
The summary was as follows:

No failures, defects nor flaws have been discovered.
Rounding appears to conform to the proposed IEEE standard P754,
except for possibly Double Rounding during Gradual Underflow.
The arithmetic diagnosed appears to be Excellent!

The exception was a "denormalized operand" exception in the search for
overflow threshold.

Re: [9fans] floating point

[Bruce Ellis]

excellent - tho not a benchmark it is a very rigorous program that
i've used before in compiler development.  please send.

brucee
----- Original Message -----
From: "Andrew Simmons" <andrew.simmons@monitorbm.co.nz>
To: <9fans@cse.psu.edu>
Sent: Friday, February 13, 2004 9:16 AM
Subject: Re: [9fans] floating point


> > Shortish, sensible, benchmarks would be a appreciated.
>
> Not exactly a benchmark, and it may be neither short nor sensible, but I
> bodged up a version of Kahan's Paranoia to run under Plan 9 a while back.
> The summary was as follows:
>
> No failures, defects nor flaws have been discovered.
> Rounding appears to conform to the proposed IEEE standard P754,
> except for possibly Double Rounding during Gradual Underflow.
> The arithmetic diagnosed appears to be Excellent!
>
> The exception was a "denormalized operand" exception in the search for
> overflow threshold.

Re: [9fans] floating point

[Bruce Ellis]

this may have originally been the case but a lot of the code in com64.c
never gets reached because of earlier 'optimizations'.  maybe "messy"
was too harsh.

brucee
----- Original Message -----
From: "David Presotto" <presotto@closedmind.org>
To: <9fans@cse.psu.edu>
Sent: Tuesday, February 10, 2004 1:06 PM
Subject: Re: [9fans] floating point


> It's not messy, just inefficient.  Everything is done by calls to routines that
> take an operator and operands and then call more routines to do the work.  It
> would be nice if adds and subtracts and shifts are done inline and only mul/div
> is done the hard way.

Re: [9fans] floating point

[YAMANASHI Takeshi]

> i've often used povray as a decent benchmark. it's not the
> best renderer (as duff will tell all who ask), but it
> compiles reasonalby and mades good use of floating point.

Adding to povray, it would be great if we have HPL Linpack
benchmark running on Plan 9.  Please don't tell me whether
it is a good benchmak or not. :)
--

Re: [9fans] floating point

[David Presotto]

It's not messy, just inefficient.  Everything is done by calls to routines that
take an operator and operands and then call more routines to do the work.  It
would be nice if adds and subtracts and shifts are done inline and only mul/div
is done the hard way.

Re: [9fans] floating point

[Bruce Ellis]

that's what i was thinking, tho i wasn't aware it was so messy.

brucee
----- Original Message -----
From: "David Presotto" <presotto@closedmind.org>
To: <9fans@cse.psu.edu>
Sent: Tuesday, February 10, 2004 12:44 PM
Subject: Re: [9fans] floating point


> Fixing the vlong stuff would be nice too, if you have the extra time...

Re: [9fans] floating point

[David Presotto]

[-- Attachment #1: Type: text/plain, Size: 71 bytes --]

Fixing the vlong stuff would be nice too, if you have the extra time...

[-- Attachment #2: Type: message/rfc822, Size: 2415 bytes --]

From: "Bruce Ellis" <brucee@chunder.com>
To: <9fans@cse.psu.edu>
Subject: [9fans] floating point
Date: Tue, 10 Feb 2004 10:43:12 +1100
Message-ID: <01be01c3ef66$e0349170$8201a8c0@cc77109e>

Any p9 x86 users out there do a lot of floating point stuff?
I'm thinking of tinkering a bit more with 8c and 8l.
I can think of a few candidate benchmarks, like ghostscript,
but i don't want to recompile it all the time!
Shortish, sensible, benchmarks would be a appreciated.
(Or comments on any of the benchmarks floating around
the web.)

brucee

Re: [9fans] floating point

[a]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="UTF-8", Size: 388 bytes --]

i've often used povray as a decent benchmark. it's not the
best renderer (as duff will tell all who ask), but it
compiles reasonalby and mades good use of floating point.
and models are cross-platform. it's been a good while
since i've compiled it, but i distinctly remember it 
being not nearly as arduous as ghostscript.
ã‚
it also has the advantage of being a real app.
ã‚¢

[9fans] floating point

[Bruce Ellis]

Any p9 x86 users out there do a lot of floating point stuff?
I'm thinking of tinkering a bit more with 8c and 8l.
I can think of a few candidate benchmarks, like ghostscript,
but i don't want to recompile it all the time!
Shortish, sensible, benchmarks would be a appreciated.
(Or comments on any of the benchmarks floating around
the web.)

brucee