[Q][SOLVED] open64 for abinit?

[bjeon]

Hi, I am trying to install abinit using open64 compiler, with AMD processors.
Has anyone made with open64? It seems that abinit configure script doesn't recognize openf95.

I used as:

export FC=/home/xx/open64-4.2.4/bin/openf95
export F90=/home/xx/open64-4.2.4/bin/openf95
export CC=/home/xx/open64-4.2.4/bin/opencc
export CPP='/home/xx/open64-4.2.4/bin/opencc -E'
export CXX=/home/xx/open64-4.2.4/bin/openCC
./configure --prefix=/home/xx/abinit_serial


But there are error messages such as:
openf95-855 openf95: ERROR DEFS_DATATYPES, File = defs_datatypes.F90, Line = 81, Column = 8
The compiler has detected errors in module "DEFS_DATATYPES". No module information file will be created for this module.


Any comment will be greatly appreciated.

ByoungSeon

[pouillon]

If the compiler doesn't say what these errors consist in, I'm afraid we'll not be able to help you much.

I don't remember why I didn't put Open64 support in the Abinit build system, but seeing this error message, I guess it is probably because it was unable to build the source code.

[jbeuken]

Hi ByoungSeon,

for information, I successed to compile sequential ABINIT 6.9.1 (patched by me (not published))
with the AMD version of open64 ( 4.2.5.1 )
( http://developer.amd.com/tools/open64/P ... fault.aspx ) under a HP/DL185 ( 2x OctoOpteron )

the result of tests are pretty good...

Code: Select all

6.9.1-private/r694
======================================================
Tests SEQ start at 15:25 and done after     2685s 
 test built_in OK
========================================================
     Serie  #tests  #succes  #passed  #failed  #missing 
========================================================
   atompaw     1        1       0       0          0
    bigdft    13       13       0       0          0
   etsf_io     9        9       0       0          0
      fast    27       27       0       0          0
     libxc    13       12       1       0          0
       seq    35       26       8       1          0
 tutoplugs     4        0       4       0          0
tutorespfn    46       39       7       0          0
  tutorial    53       40      13       0          0
   unitary     4        4       0       0          0
        v1    96       96       0       0          0
        v2    95       88       7       0          0
        v3    93       86       7       0          0
        v4    94       79      15       0          0
        v5    99       74      24       1          0
        v6    88       64      24       0          0
 wannier90     3        0       3       0          0
========================================================

but the problem with the seq test is so serious that I decided not to publish the source changes to support open64...

a new version of open64 was released ( 4.2.5.2 ) : I will try to compile abinit 6.10.x with it...

to be followed...

jmb

[bjeon]

Hi, thanks for the comment.
Seems that some of source codes need to be modified.
Anyway, I hope that new compiler may solve the issue - even though I tried before with 6.8.1.
If you post the result when you're done, it will be great help.


Best regards,

ByoungSeon

Hi ByoungSeon,

for information, I successed to compile sequential ABINIT 6.9.1 (patched by me (not published))
with the AMD version of open64 ( 4.2.5.1 )
( http://developer.amd.com/tools/open64/P ... fault.aspx ) under a HP/DL185 ( 2x OctoOpteron )

but the problem with the seq test is so serious that I decided not to publish the source changes to support open64...

a new version of open64 was released ( 4.2.5.2 ) : I will try to compile abinit 6.10.x with it...

to be followed...

jmb

[bjeon]

Using abinit 6.12.1 and open64 5.0, abinit could have been installed.
Thanks abinit developers for the effort.

Also I leave the configure commands I used as references.

ByoungSeon

******** install openmpi -> then install abinit 6.12.1
1) install openmpi 1.4.5 using open64 5.0
$ ./configure --prefix=some_dir_for_openmpi CC=/open64_dir/bin/opencc CXX=/open64_dir/bin/openCC F77=/open64_dir/bin/openf90 FC=/open64_dir/bin/openf90
$ make
$ sudo make install

2) install abinit 6.12.1 using installed openmpi
$./configure --prefix=some_dir_for_abinit --enable-mpi --with-mpi-prefix=some_dir_for_openmpi
$ make
$ sudo make install
$ cd tests
$ make tests

[jbeuken]

Hi,

my experience with open64 5.0 from open64.org
( better under Xeon / CentOS 6.2 than open64 4.2.5 from AMD !?! )

I use openmpi 1.4.4 and my .ac file for configure of abinit is :

Code: Select all

enable_mpi="yes"
enable_mpi_io="no"
with_mpi_prefix="/usr/local/openmpi_open64"
enable_64bit_flags="yes"
enable_gw_dpc="yes"

remark : the mpiio test Q4 failed ( as for the pathscale compiler ) then I put "enable_mpi_io="no""

all the tests passed except one : v6#t48 -> segfault !?!

After a painful research of the problem, the workaround is the to add a line of code in "src/32_util/ys.F90"

Code: Select all

...
mone = -one
 ys_val = czero

! strange workaround for open64 5.0 to prevent segfault with v6#t48
! add by jmb 1/3/12
 write(std_out,*)'ys :  ys_val = ',ys_val

 if (lp == ll) then
...

regards

jmb

[pouillon]

Does the segfault on v6#t48 still happen if you recompile the file without optimizations?

[jbeuken]

Hi Yann,

Does the segfault on v6#t48 still happen if you recompile the file without optimizations?

if I compile only "32_util/ysF90" with -O0 ( and without the workaround ), the test succeeds !

I try to modify the code inside ys.F90 ( for ex : replace the case construction,... ) without success

jmb

[pouillon]

Then this proves that it comes from a bug in the compiler.