Optimized Jumps in High-level Languages

[Mike Lodman]

We are currently looking for a high level language that
contains constructs which optimize jumps at the machine
code level. We are also interested in modifying "C" to
include such constructs.

As an example,

if (A>B) /*mostly*/ { }
else { }

or

if (A>B) { }
else /*mostly*/ { }

with the /*mostly*/ indicating which flow the programmer
thinks would occur more often.

We may then optimize the code to use the various delayed
and immediate jumps which the chip supports to get the
best results most of the time, without going to the
assembly level of programming.

I would appreciate any help or suggestions on this
subject.

Michael Lodman
Advanced Development and Systems Architecture
NCR Corporation E&M San Diego
(619) 485-3335

ARPA : ncr-sd!lod...@sdcsvax.edu
CompuServe : 76367,1745

[Ed Segall]

This sounds like trace scheduling, where the compiler uses information
about the likelihood of the direction of branches to optimize use
of pipelines, parallelism (e.g. in a microcoded machine), and perhaps
other performance enhancers. John Fisher, who was at Yale until a
couple years ago, wrote a trace scheduling compiler for VLIW machines
(Very Long Instruction Word - i.e. parallel machines controlled with
lots of bits, such as horizontally microprogrammed machines). He wrote
a book on it called something like "Bulldog - a Trace Scheduling
Compiler for VLIW Machines," which is published by MIT Press. There
are also Yale technical reports on the work. I've also seen a
reference to trace scheduling in VLSI design. I don't remember the
book, but I'm almost certain that it is a Springer-Verlag monograph,
and that it has an introduction Edsger Dijkstra (sp?). Or maybe the
intro was by Hoare. Anyway, it was somebody famous for being very
insistant about formality. I think it was Dijkstra, and I think the
work was done in a Northern European country, which would fit. How's
that for free association?

Have fun,

Ed Segall