This is the mail archive of the
mailing list for the GCC project.
Re: [wwwdocs] Vectorizer's updates to 4.3.0 changes
- From: Jakub Jelinek <jakub at redhat dot com>
- To: Dominique Dhumieres <dominiq at lps dot ens dot fr>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Wed, 12 Mar 2008 04:47:02 -0400
- Subject: Re: [wwwdocs] Vectorizer's updates to 4.3.0 changes
- References: <20080312084122.9DAC943E17@mailhost.lps.ens.fr>
- Reply-to: Jakub Jelinek <jakub at redhat dot com>
On Wed, Mar 12, 2008 at 09:41:22AM +0100, Dominique Dhumieres wrote:
> > > + <code>-O3</code>. In order to generate code for a SIMD extension, it
> > > + has to be specified as well: use <code>-maltivec</code> for PowerPC
> > > + platforms and <code>-msse/-msse2</code> for i?86 and x86_64.
> > You should not need -msse/-msse2 for x86-64 at all since that is the
> > default. If you do not this, I would consider it a bug.
> I find this part extremely confusing. If -O3 enable vectorization, why does not it
> set the relevant -m* to make it effective? If not, I would also consider it a bug.
Because on i?86 and PowerPC -msse2 resp. -maltivec isn't available on all
CPUs? -O3 is not an option that should change the slected instruction set.
If on i?86 -msse2 isn't given (nor -march= for CPUs that support SSE2), the
vectorizer vectorizes what it can given the available instruction set.