Re: Summary of "Languages that give 'hints' to the compiler"

wg@cs.tu-berlin.de (Wolfgang Grieskamp) writes:


>Joe Hummel <jhummel@cy4.ICS.UCI.EDU> writes:
>
>> The big gotcha of course: what if the programmer supplies info, and he/she
>>is wrong?
>
>I do not see much differences to that the programmer assumes some
>properties of her code for the validity of handcrafted
>optimizations. Its even better, since the assumptions are stated
>explicitely.
>


More importantly, if the compiler does the optimising given some
user specified description of the expected behaviour, you can expect
the compiler to produce correct code (though maybe inefficient),
whereas if the programmer implements the optimisation, he might
introduce errors into an otherwise correct program.


I think that the general priciple of letting the compiler do the work
is a good one. It means that the compiler writer debugs the program
once, and then it works for _all_ programs. I think I saw someone
mention the use of profiling feedback - I think this should become
an extremely important component of compilation environments in the
future since it gives the compiler oppertunities to make usage based
optimisations without requiring guesswork from the programmer.


Thomas
--
Thomas Conway conway@cs.mu.oz.au


--

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