Intel Fortran 2013 compiler options

Intel Fortran 2013 compiler options

 

Hi, Dear Sir or Madam,

I am trying to test some new features from the latest Intel 2013 Fortran compiler. For example, I am testing the folloiwng options of 2013 Fortran compilers under Windows OS:  

/align:array64byte /O3

and add  " !dir$ vector aligned " to force vectorization  before some DO loop Fortran statements for array computation.

I did not get any improvements in performances of my Fortran Applications. Could you tell me the reasons and share your successful experiences with using "Align Data" vector aligned directive.

Thanks in advance.

Dingjun

 

 

 

5 帖子 / 0 全新
最新文章
如需更全面地了解编译器优化,请参阅优化注意事项

Could you share a bit of your code with us?

Nick

Nick

No. They are commercial codes. Sorry for that.

Steve Lionel (Intel)的头像

Have you enabled vectorization reports or /Qguide to see what the compiler says about vectorization? Perhaps alignment is not the gating factor.

Steve

Many of the new vectorization features are keyed to architecture, so you should use one more keyed to recent CPU, such as -arch:SSE4.1 (since Harpertown) or -arch:AVX (since Sandy Bridge).  array32byte should be sufficient.

!dir$ vector aligned is most likely to help with vectorization of loops containing conditionals, but it means that both true and false options have to be evaluated, followed by selection.

reductions are best done with Fortran intrinsics: sum, dot_product, maxloc, .... or by at least putting in the form required by !$omp simd reduction() directive (even if the directive isn't needed).

登陆并发表评论。