kludge@[EMAIL PROTECTED]
(Scott Dorsey) writes:
> I don't think compiler authors should be allowed to even look at
> benchmarks. Too much looking at benchmarks results in writing compilers
> that compile benchmarks really well, at the expense of actual running
code.
Well, you can try to click your heels together, and see if it helps.
The problem is that compiler writers need to test their compilers, and
if you don't have an official benchmark, you'll get unofficial ones -
at worst, you get a compiler that only can compile itself. :-)
So if you are interested in performance, you really should care about
benchmarks as well.
SPEC isn't perfect, but at least it is an attempt at a large and
comprehensive suite, and my impression is that they try hard to
represent a large set of problem domains and workloads.
-k
--
If I haven't seen further, it is by standing in the footprints of giants


|