Re: Auto vectorization

Anton Lokhmotov <al407@cam.ac.uk>
Fri, 23 May 2008 15:58:06 +0100

          From comp.compilers

Related articles
Auto vectorization javeria.abid@gmail.com (2008-05-15)
Re: Auto vectorization andreybokhanko@gmail.com (2008-05-20)
Re: Auto vectorization roland.leissa@googlemail.com (=?ISO-8859-1?Q?Roland_Lei=DFa?=) (2008-05-21)
Re: Auto vectorization al407@cam.ac.uk (Anton Lokhmotov) (2008-05-23)
| List of all articles for this month |

From: Anton Lokhmotov <al407@cam.ac.uk>
Newsgroups: comp.compilers
Date: Fri, 23 May 2008 15:58:06 +0100
Organization: Compilers Central
References: 08-05-061 08-05-082
Keywords: optimize, parallel
Posted-Date: 24 May 2008 16:50:34 EDT

> 1-- How you define the profitability of auto-vectorization phase? Is
> it just the speed up? If we do not get any speed up over scalar code
> then there is no need to do auto-parallelization.


Seems to be right. Since vector instructions typically do more work
than scalar ones, vector code is usually more *power* hungry. However,
if you achieve a considerable speed-up, the overall *energy*
consumption (power * time) can be less than that of scalar
code. (That's why vector instructions are so popular in embedded DSP
architectures.)


> 2--What are the phases or features in a compiler ( especially in the
> GCC) that control the quality of auto-vectorization?


Vectorization is (profitably) applicable to fairly specific code. Loop
restructuring transformations can massage code to a form more amenable
to vectorization.


If you are interested, I can send you a pdf of my PhD thesis on
programming and compiling for embedded SIMD architectures, which has a
survey chapter on automatic vectorization techniques.


Cheers,
Anton.


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.