[wwwdocs] Vectorizer's updates to 4.3.0 changes

Gerald Pfeifer gerald@pfeifer.com
Fri Mar 14 13:21:00 GMT 2008


On Wed, 12 Mar 2008, Ira Rosen wrote:
> Is this better?

Thanks for your patch, Ira!  I was also going to give it a try, but
you're certainly much more of an expert in this area. 

I understand that there were some further comments, but the perfect
is the enemyof the good, and we can always make updates later on, so
let's proceed with your patch with two minor changes:

> +     <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.

How about making this "has to be enabled as well" here?

> +     <li>The vectorizer was enhanced to support vectorization of
> outer-loops,
> +         intra-iteration parallelism (loop-aware SLP), vectorization of
> strided,
> +         accesses and loops with multiple data-types. Run-time dependence
> +         testing using loop-versioning was added. Cost model, turned on by
> +         <code>-fvect-cost-model</code>, was developed.

Should this be "outer loops", "dependency testing", "loop versioning" and 
"The cost model" here?

Thanks again for stepping up to address this, Ira!

Gerald



More information about the Gcc-patches mailing list