This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: thanks for the memory work
- From: Andrew MacLeod <amacleod at redhat dot com>
- To: Jeff Law <law at redhat dot com>
- Cc: Steven Bosscher <s dot bosscher at student dot tudelft dot nl>, gcc mailing list <gcc at gcc dot gnu dot org>
- Date: 25 Nov 2003 22:49:40 -0500
- Subject: Re: thanks for the memory work
- References: <200311260305.hAQ35SnC006894@speedy.slc.redhat.com>
On Tue, 2003-11-25 at 22:05, law@redhat.com wrote:
> In message <200311260055.30356.s.bosscher@student.tudelft.nl>, Steven Bosscher
> writes:
>
> >The vast majority of stmts also only have at most 2 use operands, so there we
> >could do without varrays as well.
> But the distribution isn't nearly as skewed as with output operands. My
> gut tells me that if we tried to go down this path for uses that ultimately
> we'd get something that is significantly slower.
>
Use operands are read a lot more too I think. And if you think about it,
you have to check an awful lot... if you have a MODIFY_EXPR, and if the
RHS matches a certain class of binary operators, then you still have to
check if the LHS or the RHS is an SSA_NAME or not, so figuring out where
use(0) is maybe take quite a bit of calculation. I do not think this is
the direction to go. I would expect it to be measurably slower.
Once its abstracted out, its easy for someone to play with tho I
suppose.
Andrew