Are we really having this debate again?

I think what it boils down to is that the tricks that allow a good assembly
programmer to outdo a compiler frequently become difficult and
unmaintainable as the size of the project increases.  Meanwhile, the tricks
that a compiler can use don't affect maintainabilty (as long as it does them
RIGHT), so it eventually pulls ahead.  In other words, as your program
becomes bigger, you'd be foolish not to implement some of the same things
that make compiled code slower (ie, a standard calling convention with
register saving), and equally foolish to implement some of the speedups
that a compiler can get away with.

I think even an average assembly programmer can outdo a compiler on a small
piece of code.  Usually, that's all that is necessary.

BillW

(and do you want to know how many magabytes of assembler I have that only
 runs on unobtainium CPUs, and I'd dearly LOVE to have be more portable!)