Re: Is register stack compilers' friend?

cliffc@ami.sps.mot.com (Cliff Click)
Sun, 5 Nov 1995 01:30:30 GMT

          From comp.compilers
Related articles
Is register stack compilers' friend? jaidi@technet.sg (1995-10-31)
Re: Is register stack compilers' friend? hbaker@netcom.com (1995-11-04)
Re: Is register stack compilers' friend? cliffc@ami.sps.mot.com (1995-11-05)
Re: Is register stack compilers' friend? stevec@pact.srf.ac.uk (1995-11-06)
Is register stack compilers' friend? dave@occl-cam.demon.co.uk (Dave Lloyd) (1995-11-06)
Re: Is register stack compilers' friend? paulb@pablo.taligent.com (1995-11-09)
Re: Is register stack compilers' friend? jeremy@suede.sw.oz.au (1995-11-13)
| List of all articles for this month |
Newsgroups: comp.compilers
From: cliffc@ami.sps.mot.com (Cliff Click)
Keywords: architecture, optimize
Organization: none
References: 95-11-026
Date: Sun, 5 Nov 1995 01:30:30 GMT

jaidi@technet.sg (Nor Jaidi) writes:


> After reading a book on Transputer architecture I got curious on how well
> compilers generate code for it. ...


The stack on the transputer is tiny (3 deep), doesn't have any under
or overflow support and is flushed by branches and other control flow.
Hence it's only useful for very local or temporary calculations.
Instead the transputer has a workspace pointer, with quick access to
anything within a 4 BIT offset: 16 values. These 16 values become
your "registers" and you need to allocate them efficiently. You
"spill" by using a slower 8-bit offset. The later transputers can
issue several such loads & stores per cycle, making them very similar
to registers in efficiency.


Cliff
--
Cliff Click Compiler Researcher & Designer
RISC Software, Motorola PowerPC Compilers
cliffc@risc.sps.mot.com (512) 891-7240
--

Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.