Related articles |
---|
A question about Computer Architecture rsherry8@comcast.net (Robert Sherry) (2003-01-29) |
Re: A question about Computer Architecture david.thompson1@worldnet.att.net (David Thompson) (2003-02-06) |
From: | "David Thompson" <david.thompson1@worldnet.att.net> |
Newsgroups: | comp.compilers |
Date: | 6 Feb 2003 00:07:53 -0500 |
Organization: | AT&T Worldnet |
References: | 03-01-173 |
Keywords: | architecture, arithmetic |
Posted-Date: | 06 Feb 2003 00:07:53 EST |
Robert Sherry <rsherry8@comcast.net> wrote :
....
> I would like to know if there are any architectures other then the
> Pentium and the X86 family that do floating point computations using a
> floating point stack, and if these architectures give the user the
> ability to use operands loading on the stack but are not top element
> or next to the top element like the Pentium does.
....
> [I'd be surprised if there were any live stack architectures left
> other than the x87 and the Burroughs A series, and that register
> stacks are not popular these days because they make it harder to keep
> common subexpressions around and it's a pain (on the x87) to be sure
> you don't overflow the stack. -John]
The Tandem^WCompaq^WHP NonStop 1 or 2, which used
8x16bit registers mostly as a stack but with some direct access
for both integer and floating, still runs in emulation on their later
MIPS-based TNS-R systems (at quite a few sites I am aware of)
if you count that as live. Yes, their compilers (in TNS mode)
don't do much CSE, but spilling and refilling the stack doesn't
seem to be much of a problem; it probably would in assembly,
but that isn't really supported.
Of course, the sorts of applications Tandems are bought for generally
involve little if any floating, and anyone who chose Tandem for
numbercrunching needs their head replaced.
--
- David.Thompson 1 now at worldnet.att.net
Return to the
comp.compilers page.
Search the
comp.compilers archives again.