Re: Folk Theorem: Assemblers are superior to Compilers

elliottm@csulb.edu (Mike Elliott) writes:
> Folk Theorems
>[assembler code is 1) faster and 2) smaller than compiler code]


I should know better than to jump into such a flame-inducing question as
this, but here goes:


1) Sometimes compilers are better than humans (Assembly):
      Compilers do all their optimizations all the time.
      Compilers don't make mistakes (although compiler writers do :-).
      Compilers can take the time and effort to look over the entire program,
          or at least really big chunks of it.


      Common examples: Compiling Fortran, C/C++,etc to a modern RISC (Mips,
                                        Alpha, RS/6000, etc).


2) Sometimes humans (Assembly) are better than compilers:
      Compilers only know what compiler writers teach 'em;
          other humans may know that for this problem instance an otherwise
          illegal transformation is legal.
      Various compiler optimizations are geared toward regular register and
          instruction sets; compiler writers have not learned how to handle well
          the irregular register sets of some machines.
      Compilers solve a particular problem: translating a HLL to machine code.
          Sometimes humans want to solve a different problem (i.e., knowing the
          EXACT number of cycles a particular stretch of code takes so we can
          count on it being done before the next interrupt).


      Common examples: squeezing the last drop of performance out of a Cray,
                                        making an i860 do anything with it's pipeline, writing
                                        hard-real-time code.


Cliff
--
cliffc@cs.rice.edu -- Massively Scalar Compiler Group, Rice University
--

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