build_libsmm generate small1

[Joe Greenstone]

Hi,

On Friday, 13 September 2013 17:15:48 UTC+9:30, Alfio Lazzaro wrote:

> Hi,
> is the tiny1 phase generating the correct output files:
>
> tiny_gen_optimal_dnn_linux.gnu.out
>
> ? You should see a list of entries like:
>
> 1 1 1    5   1   1   1    2.490000       0.080
> 1 1 2    4   1   1   1    3.210003       0.125
>
 
Yes, the entries look fine.
 
 

> Then, the tiny1 phase doesn't link to BLAS, while small1 does.
> I wonder if you are linking to the correct mkl libraries.
> Could you take a look inside the directory
>
> run_small_*
>
> and see if the source files are produced? 
>
 
Yes, the source files are produced.
 
 

> Then you can compile by hand one of this file and see if it works when 
> executed.
> For instance, you can do something like:
>
> cd <small directory>
> gfortran -O2 -funroll-loops -ffast-math -ftree-vectorize -march=native 
> -fno-inline-functions small_find_1_1_1.f90 -o 
> output_linux.gnu/small_find_1_1_1.x -L/mkl/lib/intel64 -Wl,--start-group 
> -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group 
> cd <output_linux.gnu>
> ./small_find_1_1_1.x
>
 
 
To avoid the question of gfortran/MKL compatibility (though Intel 11.0 is 
said to be suitable for gnu 4.4.0 and later), I have switched to Intel/MKL 
for tiny1, small1 and the test described further below. The small1 output 
is:
./output_linux.intel/small_find_9_32_23.x > 
output_linux.intel/small_find_9_32_23.out
./output_linux.intel/small_find_9_32_24.x > 
output_linux.intel/small_find_9_32_24.out
./output_linux.intel/small_find_9_32_26.x > 
output_linux.intel/small_find_9_32_26.out
./output_linux.intel/small_find_9_32_32.x > 
output_linux.intel/small_find_9_32_32.out
/bin/sh: line 1: 16024 Segmentation fault      (core dumped) 
./output_linux.intel/small_find_9_32_32.x > 
output_linux.intel/small_find_9_32_32.out
make: *** [output_linux.intel/small_find_9_32_32.out] Error 139
make: *** Waiting for unfinished jobs....
 
Compiling manually using:
ifort -O2 -funroll-loops -vec-report2 -warn -xHost -fno-inline-functions 
-nogen-interfaces -openmp -nolib-inline -static-intel -mkl=
sequential small_find_9_32_32.f90 -o output_linux.intel/small_find_9_32_32.x
gives no error message (just messages suggesting that more vectorization 
opportunities exist).
cd output_linux.intel
./small_find_9_32_32.x
gives a segmentation fault without other output.
 
Do you still have Intel v11 there?
 
Thanking you for your help.
 
 
 
 
 

>  
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20130913/4808a957/attachment.htm>