Re: IEEE Decimal Float on Itanium
- From: "Tom Linden" <tom@xxxxxxxxxxxxxxxxx>
- Date: Sat, 09 Dec 2006 05:58:45 -0800
On Fri, 08 Dec 2006 19:36:54 -0800, Arne Vajhøj <arne@xxxxxxxxxx> wrote:
Tom Linden wrote:I presume that is IEEE (proposed) float decimal?On Fri, 08 Dec 2006 12:06:31 -0800, Stephen Hoffman <Hoff@xxxxxxxxxxxxxxxxxxxxxxxxxxx> wrote:John Reagan wrote:Most definitely. Currency calculations. Many (and I do mean MANY) applications incorrectly use binary floating point to manipulate currency. They often find out too late that 1.23 + 4.56 may not equal 5.79 (at least exactly). COBOL on Alpha and I64 have lots of software to emulate that decimal abilities on the VAX. Having it back in hardware might allow COBOL applications to see a performance increase.
The question should better have been "do you have something in particular here requiring integer values larger than, say, quadwords?" And if so, are the numbers of quadword (or larger) operations required going to see a particular benefit from the addition of hardware and/or software?We had a bit of a discussion on this topic in comp.lang.pl1 and I think
There are certainly organizations that are dealing with vast sums of money, though most also have the budgets needed for customized hardware.
it was somewhat evenly divided on the value of this. The group I fell
into was of the opinion that scaled fixed decimal (which on VAX and Alpha
gives us 31 digits) was adequated for the task. The other group that it
was good because it would obviate FIXED OVER FLOW EXCEPTIONS, FOFL in PL/I
ecxception handling parlance.
As it turns out, studying the IBM papers on this, they use an encoding
techniques which gives them two additional digits of accuracy to 33, but
when the computations are performed they convert to packed decimal, bcd
format. My own feeling is that if they add this as a new data type in
C, for example, it would make C viable to be used in financial applications.
My god, have we sunk that low:-)
decimal is actually a builtin type in the newest "big" language: C#.
Well the code generator would have to be taught to emit all those new
IBM could use C# via Mono on Linux running on zSeries hardware
instructions
But Cobol or Java seems more likely on zSeries (BigDecimal is
a class in Java, but nothing prevents the JVM from JIT'ing
it to something hardware supported).
Arne
--
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/
.
- Follow-Ups:
- Re: IEEE Decimal Float on Itanium
- From: Arne Vajhøj
- Re: IEEE Decimal Float on Itanium
- References:
- IEEE Decimal Float on Itanium
- From: Tom Linden
- Re: IEEE Decimal Float on Itanium
- From: Chip Coldwell
- Re: IEEE Decimal Float on Itanium
- From: Stephen Hoffman
- Re: IEEE Decimal Float on Itanium
- From: John Reagan
- Re: IEEE Decimal Float on Itanium
- From: Stephen Hoffman
- Re: IEEE Decimal Float on Itanium
- From: Tom Linden
- Re: IEEE Decimal Float on Itanium
- From: Arne Vajhøj
- IEEE Decimal Float on Itanium
- Prev by Date: Re: Novice's questions
- Next by Date: Re: Novice's questions
- Previous by thread: Re: IEEE Decimal Float on Itanium
- Next by thread: Re: IEEE Decimal Float on Itanium
- Index(es):
Relevant Pages
|
Loading