Re: SGI files for chapter 11
- From: "Tom Linden" <tom@xxxxxxxxxx>
- Date: Mon, 08 May 2006 18:47:32 -0700
On Mon, 08 May 2006 17:43:45 -0700, Bill Todd <billtodd@xxxxxxxxxxxxx> wrote:
Dan Foster wrote:
...
While x86-64 is clearly better than the VAX for the registers, it may
not be sufficiently significantly better.
I'll suggest that it need not be any better at all. 16 registers are more than sufficient for common register-use patterns, and in the less-common cases where they are not sufficient to eliminate the need for re-use the cost of moving register contents back and forth occasionally to a fast L1 D cache is hardly a show-stopper (especially when executing on an OoO processor).
For coloring algorithms 16 is a good number. If you have a large number of registers
a good strategy is to organize them into context sets as the Nord-10 did
(35 years ago) corresponding to priority levels.
I bet it *could* be made to work on x86 if given sufficient interest and
money, but would guess it'd still be rather painful.
Perhaps for code written in assembler; otherwise, compilers take care of the problem (and even Bliss has an x86 compiler already, IIRC - just not one that HP was willing to support externally for Oracle).
- bill
.
- References:
- SGI files for chapter 11
- From: Andrew
- Re: SGI files for chapter 11
- From: Andrew
- Re: SGI files for chapter 11
- From: Malcolm Dunnett
- Re: SGI files for chapter 11
- From: Bill Gunshannon
- Re: SGI files for chapter 11
- From: Dan Foster
- Re: SGI files for chapter 11
- From: Bill Todd
- SGI files for chapter 11
- Prev by Date: Re: SGI files for chapter 11
- Next by Date: Re: This post will self-destruct in 10secs (Was Re: X windows vulnerability)
- Previous by thread: Re: SGI files for chapter 11
- Next by thread: Re: SGI files for chapter 11
- Index(es):