Compiler writers vs. architects

In light of the RISC addressing mode discussion on comp.compilers, I couldn't
resist posting a legend I heard some time ago, supposedly about one of the
successors to the VAX 780 (the 8000 series perhaps?). The claim was that the
compiler writers were doing a series of compiler improvements and used
benchmark statistics from the 780 to avoid the slow instructions. The new
computer architects used the same statistics to speed up the slow
instructions! (One could substitute "addressing mode" for "slow instruction"
and get a similar story, though that was not how I heard it.)


In my opinion, things like this do not happen because DEC is stupid or people
there are stupid. Rather, it happens because of the way we have fragmented
computer science into subspecialties. One of the great lessons of RISC is that
there is important synergy between different areas, in this case architecture
and compilers. But this point can be made more broadly: we will get better
solutions to overall systems problems if we consider whole systems rather than
single pieces. This is not easy, but I think it is where the future lies.
Yours in philosophizing ....
--
J. Eliot B. Moss, Assistant Professor
Department of Computer and Information Science
Lederle Graduate Research Center
University of Massachusetts
Amherst, MA 01003
(413) 545-4206; Moss@cs.umass.edu

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