Re: vectorization in icc

kf@iki.fi wrote:


: I've been experimenting with the Intel C/C++ compiler for Linux, and in
: particular, with the automatic vectorization.


: I have the following piece of code (all the arrays are of type char):
: [...]


: Both work just fine, but the vectorized code is significantly slower!
: I certainly expected the vectorized code to be much faster.


Don't worry! I experienced something similar when porting FFTW-GEL (a
special purpose compiler for FFTs) to the Intel Pentium 4. See
http://www.complang.tuwien.ac.at/skral/fftwgel.html for details.


: What's going on? Are the sse2 instructions really so slow compared to
: the standard integer instructions? If so, what's the point of the
: vectorization anyways?


I would like to note two things:
    (1) Usually, Intel does not properly support some ISA extension
            until some time passes (or until AMD does it first ;-)


    (2) Vectorized code is very sensitive to alignment issues. Make sure
            that the character arrays you are using are 128-bit aligned.


Best Regards,
Stefan
--
Stefan Kral http://www.complang.tuwien.ac.at/skral/

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