home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!warwick!pavo.csi.cam.ac.uk!cam-cl!rf
- From: rf@cl.cam.ac.uk (Robin Fairbairns)
- Newsgroups: comp.arch
- Subject: Re: Are 64 Int or FP registers useful?
- Keywords: Chip Real State, Register Usage
- Message-ID: <1992Sep8.140635.12078@infodev.cam.ac.uk>
- Date: 8 Sep 92 14:06:35 GMT
- References: <1992Sep7.091904.2626@newsroom.bsc.no>
- Sender: news@infodev.cam.ac.uk (USENET news)
- Reply-To: rf@cl.cam.ac.uk (Robin Fairbairns)
- Organization: U of Cambridge, England
- Lines: 27
- Nntp-Posting-Host: ely.cl.cam.ac.uk
-
-
- In article <1992Sep7.091904.2626@newsroom.bsc.no>, izahi@bsc.no (Raul
- Izahi Lopez Hernandez) writes:
- |> Can somebody provide me with comments or references to pros and cons of
- |> having a large register set on-chip. Lets say 64 Int and 64 FP
- registers be it
- |> 32-bit or 64-bit.
- |> Specifically is there a study that shows how many registers typical
- |> programs would like to have (i.e. unlimited registers as resources, like
- |> the studies on unlimited number of execution units)?
- |> This is for a microprocessor.
-
- I can't offer a study, and I can't say anything about microprocessors.
- However... We used to have here a (mainframe) machine that offered
- user-mode programs free use of 93 registers (2 of which always read as
- zero, one of which was the program counter, and one of which was a
- register indexer - see that other thread...).
-
- The compilers for that machine _never_ made cogent use of this
- enourmous number of registers, except by way of the trick that one
- regularly used in assembler programs of loading "useful constants"
- into a set of registers at the start. High-level language register
- dumps had a certain monotony about them...
- --
- Robin (dump Brian Kay) Fairbairns rf@cl.cam.ac.uk
- U of Cambridge Computer Lab, Pembroke St, Cambridge CB2 3QG, UK
- [Finally back in work - what a relief!]
-