[MITgcm-support] LOOP WAS VECTORIZED

[Martin Losch]

Hi Li,

as pointed out before, this is just a informational message by the compiler that a certain loop was vectorized. Vectorization usually improves performance, so seeing this message is usally good. Vectorization can change the results (at the truncation level), so sometimes you want to suppress vectorization, but I think in your case you should not worry about this. With "my" ifort, the vectorization is enabled by the "-x", and you can control the length of the messages by -vec-report[n], default for n=1, I guess -vec-report0 turns off the messages. See "man ifort" for more details on these compiler options.

M.

On Jul 20, 2010, at 5:22 AM, q li wrote:

> Hi all,
>  
> Recently I am using the MITgcm at a cluster with AMD64 cpu, together with ifort (9.1.040) and mpich (2.1.2). I received a message called "remark: LOOP WAS VECTORIZED". I googled "MITgcm remark: LOOP WAS VECTORIZED" and nothing feeds back. I am wondering if any MITgcmer had this problem. Then I keep searching and found someone said this is related to "optimization switches", then he suggest to add "-ftz -ip -ipo -parallel -axP " (seehttp://lists.apple.com/archives/fortran-dev/2006/Apr/msg00035.html). It does not show these "remark" messages immediately, but they show up together in the end of "make". I do the internal wave test in both way. The results "look" fine and exactly same.
>  
> My question is
> (1) Are these "remark" messages a warning or error? Will they influence my calculation?
>  
> (2) Any way to avoid it?
>  
> (3) Is any special setting for me to use Intel's ifort on a AMD64 cluster?
>  
> $ mpif77 -w95 -W0 -WB -convert big_endian -assume byterecl -fPIC  -O2 -align -xW -ip -c chksum_tiled.f
> chksum_tiled.f(722) : (col. 9) remark: LOOP WAS VECTORIZED.
> chksum_tiled.f(733) : (col. 9) remark: LOOP WAS VECTORIZED.
> $ mpif77 -w95 -W0 -WB -convert big_endian -assume byterecl -fPIC  -O2 -align -xW -ip -c -ftz -ip -ipo -parallel -axP chksum_tiled.f
> Regards,
>  
> Li
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support