Re: 200 way issue?

preston@dawn.cs.rice.edu (Preston Briggs)
Fri, 1 Oct 1993 15:25:52 GMT

          From comp.compilers
Related articles
200 way issue? davidm@questor.rational.com (1993-09-29)
Re: 200 way issue? anton@mips.complang.tuwien.ac.at (1993-09-30)
Re: 200 way issue? pop@mtu.edu (1993-09-30)
Re: 200 way issue? grover@brahmand.Eng.Sun.COM (1993-09-30)
Re: 200 way issue? petersen@sp51.csrd.uiuc.edu (1993-09-30)
Re: 200 way issue? mac@coos.dartmouth.edu (1993-10-01)
Re: 200 way issue? preston@dawn.cs.rice.edu (1993-10-01)
Re: 200 way issue? daveg@thymus.synaptics.com (Dave Gillespie) (1993-10-04)
| List of all articles for this month |
Newsgroups: comp.compilers
From: preston@dawn.cs.rice.edu (Preston Briggs)
Keywords: performance, bibliography
Organization: Rice University, Houston
References: 93-09-142
Date: Fri, 1 Oct 1993 15:25:52 GMT

davidm@questor.rational.com (David Moore) writes:
>How many way issue can one actually use on real code?
>
>Suppose, for example, we took the spec benchmarks and optimized for an
>infinite issue machine. Now suppose we built a histogram of actual number
>of instructions issued per machine cycle. Has anyone published a paper on
>what this histogram would look like?


Papers like this tend to appear in ASPLOS.
For example:


@inproceedings{jouppi:89b,
    title="Available Instruction-Level Parallelism for Superscalar and
                                  Superpipelined Machines",
    author="Norman P. Jouppi and David W. Wall",
    pages="272--282",
    year=1989,
    address="Boston, Massachusetts",
    booktitle=asplosIII
}




@inproceedings{wall:91a,
    title="Limits of Instruction-Level Parallelism",
    author="David W. Wall",
    pages="176--189",
    address="Santa Clara, California",
    year=1991,
    booktitle=asplosIV
}




The hard parts about such papers (in general, no criticism of Jouppi
and Wall intended) are the assumptions about the machines and
optimization techniques. In any case, these two papers are certainly
a good starting point.


Preston Briggs
--

Post a followup to this message

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