C "optimization"

[Jim Pantaleone]

In article <403 at ihdev.UUCP> pdg at ihdev.UUCP (55224-P. D. Guthrie) writes:
>I would be interested to know if there have been any articles published
>on DEC's optimization methods.

I remember reading a volume from Digital Press a few years ago on the
subject of the creation of DEC's first PL/1 compiler. It wasn't bad at
all - a bit of "Soul of a New Machine" mixed in with blood and guts only
a compiler phreak could love. Interesting discussion of how they chose/
developed/cooked up the optimizations that made it into the final
product. [Your local DEC people should have a pub. catalogue somewhere.]

Point is, rev 1 of the C compiler produced code that had well-optimized
code mixed in with plug-ugly. And it looked eerily familia. Familiar. I
think they hacked the PL/1!

The C code still isn't as beauteous as the PL/1 (DO-loops where the 
index variable can get redefined if that works out better) or Fortran 
(machine NOP's interspersed to optimize memory cycling), but the latest 
version (2.1?) does actually produce correct, good, code. As far
as I know.  And the screen debugger is (most of the time) at least as 
much a boon to mankind as the compiler... now, why aren't ALL the system 
symbols available in ".H" file form?

I wish there were a mechanism whereby those that were willing to pay
for it could get really good software for Unix too. Distributed program-
ming in the middle of the night is not really conducive to projects
that require major manpower...  Lord! my Pcc was dumb when I first
got it! Ugly too.  

But it's true DEC would never have let me at the source... but it's
also true I wouldn't have needed to...


Garry Wiegand     (Cornell Engineering // Flying Moose Graphics)
garry%geology at cu-arpa.cs.cornell.edu  (arpa)

(Tho' it may not sound like it, I have no official connection to DEC.)