Related articles |
---|
Re: Why are integer and FP registers separated? stefan.monnier@epfl.ch (Stefan Monnier) (1995-08-11) |
Re: Why are integer and FP registers separated? meissner@cygnus.com (Michael Meissner) (1995-08-14) |
Re: Why are integer and FP registers separated? stefan.monnier@epfl.ch (Stefan Monnier) (1995-08-18) |
Re: Why are integer and FP registers separated? meissner@cygnus.com (Michael Meissner) (1995-08-15) |
Newsgroups: | comp.compilers |
From: | "Stefan Monnier" <stefan.monnier@epfl.ch> |
Keywords: | architecture, performance |
Organization: | Compilers Central |
References: | 95-08-083 |
Date: | Fri, 18 Aug 1995 14:19:45 GMT |
Date: Mon, 14 Aug 1995 18:02:15 +0200
Michael Meissner <meissner@cygnus.com> wrote:
> You could use implied tags, where floats are put in one set of pages,
> cons are put in another set (I forget what the generic name for this
> is -- Bag of Bits is what is coming to mind).
You probably mean BIBOP ("BIg Bags Of Pages" or somesuch)
Stefan
PS: it's definitely a possibility, but it means you need a pointer to the
float. The main problem was to find a scheme that allows to deal with
immediate floats. BIBOP is actually a tagging scheme that tags some bits
of your pointer, just like using the lower 2 bits. I'm not saying it's a bad
scheme (all tagging scheme are highly dependent on the specific case) but it's
not a solution to the "current" problem.
Stefan
--
Return to the
comp.compilers page.
Search the
comp.compilers archives again.