This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCSE memory usage
- To: egcs at egcs dot cygnus dot com
- Subject: Re: GCSE memory usage
- From: Gabriel Dos_Reis <Gabriel dot Dos_Reis at sophia dot inria dot fr>
- Date: 26 Oct 1999 01:06:58 +0200
- Cc: law at cygnus dot com
- Organization: I.N.R.I.A Sophia-Antipolis (France)
- References: <199910252001.WAA05776@quatramaran.ens.fr>
- Reply-To: egcs at egcs dot cygnus dot com
Marc Espie <espie@quatramaran.ens.fr> writes:
| In article <8591.940819795@upchuck> you write:
| >I disagree, unless you can find some way that doesn't severely impact normal
| >code. If we can't find an implementation that does both reasonably, then we
| >should drop in some heuristics to simply not perform such expensive
| >optimizations on absurd code. Really. 15kbbs, 264k regs. That's absurd.
|
| This is not absurb code at all. This is the code you get if you try to
| migrate old Fortran-like computations to advanced expression-templates
| libraries.
|
| Do state that you don't want to support such code, but don't label it as
| `absurb'. Such code is going to get more and more common, as modern C++
| libraries spread... unless of course, enough buggy compilers go belly up
| on them.
|
| Right now, expression templates are about the only way to get `expert'
| optimizations on high-level C++ code that make it feasible to reach a better
| level of optimization than Fortran 90 code.
Not to mention that part of libstdc++-v3 uses intensive
expression-template like code. Not just because GCC is unable to
provide reasonable support for such programs they should be labelled
'absurd'.
-- Gaby