Re: High Level Language vs Assembly

> francis.doyle@donovandata.com wrote:
> > Hi,
> > When I took a compiler class in Grad School, my professor was
> > adamant about the efficiency of code produced by new high level
> > language compilers. He stated that a human assembly language writer
> > was no longer able to compete with these compilers...


It seems to me that the question is meaningless without some notion of
costs. How long is this assembly programmer allowed to hack upon the
code? (Longer than it takes for the hardware folks to double the
speed of the CPU?) For that matter, how much are you willing to pay
for a super-smart assembly programmer with 20 years experience
optimizing code for your particular machine? How much are you willing
to pay to maintain the code? Is it likely to change? When the code
needs to be changed, how much time is the assembly programmer allowed
to spend re-optimizing it?


In short: it boils down to economics.


Is the programmer allowed to hack upon the output of the compiler? We
know that optimizing compilers do not produce "optimal" code in all
cases. So surely one can print out the compiler output, and just find
one little instruction to eliminate (in many cases), to win the
contest?


Is the compiler writer allowed to look at the expert's code, and train
the compiler to produce that code? That's how the HAL/S compiler for
the space shuttle was done at Intermetrics: there were multiple rounds
of letting assembly programmers try to beat the compiler, followed by
compiler writers trying to beat the assembly programmers -- each
seeing the output of the other's previous round.


- Bob

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