This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Loop optimisation
- To: egcs at cygnus dot com
- Subject: Loop optimisation
- From: "Michael P. Hayes" <michaelh at ongaonga dot chch dot cri dot nz>
- Date: Wed, 5 Nov 1997 16:35:29 +1300
I've noticed that GCC poorly optimises the following loops:
void foo(const float *src, float *dst, int size, int offset)
{
int i;
for (i = 0; i < size; i++)
dst[i + offset] = src[i];
}
where the loop RTL of interest after CSE is
(set (reg:SI 43) (plus:SI (reg/v:SI 40) (reg/v:SI 42)))
(set (reg:SI 44) (plus:SI (reg/v:SI 37) (reg:SI 43)))
(set (mem/s:SF (reg:SI 44)) (reg:SF 45))
where 40 is i, the induction variable pseudo
42 is offset
37 is &dst
Ideally, the redundant address calculations should be hoisted out of
the loop. Is this within the scope of the current loop optimiser?
Michael.