more, more on workstation speed, I'm still not satisfied ....

[Dan Hoey <hoey@nrl-aic.arpa>]

    Date: Fri, 14 Aug 87 10:15 EDT
    From: Daniel L. Weinreb <DLW@ALDERAAN.SCRC.Symbolics.COM>

        Date: Wed, 12 Aug 87 17:22 EDT
        From: Jeffrey Del Papa <dp@JASPER.PALLADIAN.COM>

	... compile times for 400k lines of code, times in hours (a
	summary for those who missed the prev)

          explorerII 16mb               1.5
          symbolics 3620 16mb           4.5
          explorerI 8mb                 7.2
          apollo 3000 8mb, lucid       36.

    I'd like to make sure everyone understands that these numbers
    should NOT be considered comparisons of processor speed.  The
    Symbolics compiler, the TI compiler, and the Apollo compiler are
    completely unrelated and very different programs.  These numbers
    reflect not only processor speed, but the speed of the compilers as
    well.

Now if for each pair (A,B), if you take the source of compiler A,
compile it on machine B, and take the binaries back to machine A to
compile this 400k lines of code, you might get sixteen numbers
(including the A=B cases, which may or may not be what is reported
above) that would tell you:

1. You can't get the source for compiler A,
2. Even if you can, it won't compile on machine B,
3. Even if it does, the binaries you get won't run on machine A,
4. But if they did, you would get an idea of how much optimization
   you get out of compiler A for typical programs like compiler B.

You just have to understand your benchmarks.

Dan Hoey