[patch] vectorizer cost-model: fixes and powerpc testcases

[Dorit Nuzman]

"Jagasia, Harsha" <harsha.jagasia@amd.com> wrote on 18/06/2007 21:32:46:

> Hi Dorit,
>
> >Harsha, I had to change the expected dg-final check in one of the i386
> >tests due to the changes above. There maybe some things need to be
> modified
> >for the x86_64 tests too - could you take a look at that?
>
> I did a quick non-bootstrap build. I had to change the expected dg-final
> check on the same x86_64 test as the i386 test (patch attached)
>
> ChangeLogs:
> * gcc.dg/vect/costmodel/x86_64/costmodel-vect-68.c: Now vectorized.
>
> I will bootstrap and do a full regression testing with your patch on
> x86_64 and update the list in a short while.
>
> I also noticed that you had added some sub tests based on vect-76.c to
> ppc. I thought vect-76 gave similar coverage as vect-33 and hence I did
> not include vect-76. Do you think otherwise i.e is it worthwhile to add
> the vect-76 tests to x86_64 and i386 as well?
>

I just remembered from our off-list discussions that vect-76 behaved
differently on Altivec vs. SSE, because it has an unaligned load, which is
handled differently with Altivec/SSE, whereas in the vect-33 testcase we
store a constant, so there's no loads there.

thanks for testing on x86_64 and for the testcase fix,

dorit

> Thanks,
> Harsha
> [attachment "costmodelfixes.june18.txt" deleted by Dorit
Nuzman/Haifa/IBM]