* loop unroll question
@ 2004-04-01 14:53 Revital Eres
0 siblings, 0 replies; 2+ messages in thread
From: Revital Eres @ 2004-04-01 14:53 UTC (permalink / raw)
To: rakdver, gcc
Hello,
For mainline on power-pc 4 with the option -O3 -funroll-loops,
the following loop is not unrolled, but in GCC 3.4 it is.
Also for mainline the bct is not inserted, but again in GCC 3.4
it is.
void
example_new(int a[],int b[],int c[], int j)
{
int i;
for(i=0 ; i<=j ; i++)
{
c[i]=a[i]+b[i];
}
return;
}
If the condition is changed to < (instead of <=) the unrolling is
done as expected and bct is inserted.
Thanks,
Revital
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: loop unroll question
[not found] <OF3FEBFE03.6A98C185-ONC2256E69.004E2096-C2256E69.004F31D6@il.ibm.com>
@ 2004-04-02 9:06 ` Zdenek Dvorak
0 siblings, 0 replies; 2+ messages in thread
From: Zdenek Dvorak @ 2004-04-02 9:06 UTC (permalink / raw)
To: Revital Eres; +Cc: gcc
Hello,
> For mainline on power-pc 4 with the option -O3 -funroll-loops,
> the following loop is not unrolled, but in GCC 3.4 it is.
> Also for mainline the bct is not inserted, but again in GCC 3.4
> it is.
>
> void
> example_new(int a[],int b[],int c[], int j)
> {
> int i;
>
> for(i=0 ; i<=j ; i++)
> {
> c[i]=a[i]+b[i];
> }
> return;
> }
as for inserting bct -- it was wrong to do this, since the loop could
be the infinite one (not really, since we work in a signed arithmetics
for that the overflow has an undefined behavior, but we cannot
verify this on rtl level).
As for unrolling, we are too conservative here; I will send a patch
for this soon.
Zdenek
> If the condition is changed to < (instead of <=) the unrolling is
> done as expected and bct is inserted.
>
> Thanks,
> Revital
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2004-04-02 9:06 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2004-04-01 14:53 loop unroll question Revital Eres
[not found] <OF3FEBFE03.6A98C185-ONC2256E69.004E2096-C2256E69.004F31D6@il.ibm.com>
2004-04-02 9:06 ` Zdenek Dvorak
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).