Re: Why do we still assemble?

In 94-04-032 hbaker@netcom.com (Henry G. Baker) writes:
>[Other than institutional intertia, why compile to assembly code?]


Many retargetable compilers are written to generate code for an
abstraction of the real machine (I've heard Robert Henry call it the
``Compiler Writer's Virtual Machine'', I don't know if it's a common term
or not, but it should be :-). The CWVM contains a number of
simplifications of the real machine; the assembler and linker are expected
to mop up the spills.


Compiling for a CWVM fails for some of the reasons Henry mentions -- that
the compiler and assembler both do some of the same tasks. But for
``moderately optimizing'' compilers, the separation can be fairly good and
the machine descriptions can be very terse; sometimes a few hundred lines.
More detailed optimizations require a more detailed model of the machine,
though, so the advantages of a simple CWVM get lost.


A little off the topic, note that any retargetable compiler (whether it
generates assembly or machine code) is faced to some degree with the UNCOL
problem: either the abstract machine description omits some details and
thus loses some optimizations, or the abstract machine description comes
to be a union of all possible machine features.


%A M. E. Conway
%T A Proposal For An UNCOL
%J CACM
%V 1
%N 10
%D Oct. 1958
%P 5-8


(This is actually the inverse statement of the original UNCOL problem,
which needs an IR with all source language features, but one implies the
other.)


;-D on ( UNCOLing all cars ) Pardo
--

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