Hi,
(a) DRD tool: in Sec. 8.1 (Overview) of the Valgrind User Manual
(release 3.13.0 15 June 2017), it is indicated that the DRD Valgrind
tool "is a Valgrind tool for detecting errors in multithreaded C and
C++ programs". Then, Fortran programs are excluded? If so, why?
(b) Helgrind tool: in Sec. 7.5 (Hints and Tips for Effective Use of Helgrind)
of the same manual, lists 9 tips and suggestions. The false positives
explained in point 8 due to standard I/O are (almost) not shown in simple
programs, but they tend to appear in rather large programs. For example,
the following (reduced) simple test:
program simple_test
use omp_lib
implicit none
integer :: ih, im, nh, np
im = 0
!$omp parallel private (ih) shared (nh,np) reduction (max: im)
ih = omp_get_thread_num()
nh = omp_get_num_threads()
np = omp_get_num_procs()
im = max (im,ih)
!$omp end parallel
write (*,*) np, nh, im
end program simple_test
compiled with:
$ gfortran --version
GNU Fortran (GCC) 8.0.1 20180122 (experimental)
on an Intel(R) Core(TM) i7-3930K CPU @ 3.20GHz machine (6 cores)
with Fedora 27:
$ cat /proc/version
Linux version 4.14.13-300.fc27.x86_64
(mockbu...@bkernel01.phx2.fedoraproject.org)
(gcc version 7.2.1 20170915 (Red Hat 7.2.1-2) (GCC))
#1 SMP Thu Jan 11 04:00:01 UTC 2018
using valgrind-3.13.0 and:
$ export OMP_NUM_THREADS=2
three tests are performed:
Test 1/3:
$ valgrind --tool=memcheck --leak-check=full --show-leak-kinds=all ...
it gives:
Conditional jump or move depends on uninitialised value(s)
at 0x505EA5B: write_decimal.constprop.10 (write.c:808)
by 0x505EE13: write_integer (write.c:1351)
by 0x505FCBD: list_formatted_write_scalar (write.c:1865)
by 0x5060994: _gfortrani_list_formatted_write (write.c:1943)
by 0x4009D6: MAIN__ (valtest-002.F90:19)
definitely lost: 0 bytes in 0 blocks
indirectly lost: 0 bytes in 0 blocks
possibly lost: 288 bytes in 1 blocks
still reachable: 2,000 bytes in 4 blocks
suppressed: 0 bytes in 0 blocks
Test 2/3):
$ valgrind --tool=helgrind --free-is-write=yes --track-lockorders=no
--check-stack-refs=no --read-var-info=yes
--sim-hints=no-nptl-pthread-stackcache --fair-sched=try ...
it gives:
Possible data race with a previous ...
This conflicts with a previous ...
ERROR SUMMARY: 9 errors from 9 contexts (suppressed: 1 from 1)
Test 3/3 (forgetting comment (a)):
$ valgrind --tool=drd --read-var-info=yes --first-race-only=yes ...
it gives:
Conflicting store by thread ...
ERROR SUMMARY: 5 errors from 5 contexts (suppressed: 3 from 2)
Then,
(i) It would be interesting to avoid (bogus?) messages like as:
"Conditional jump or move depends on uninitialised value(s)"
in Fortran programs that use "free-format" read/write, e.g.
line (valtest-002.F90:19) of Test 1, or "Possible data race"
(Test 2) and "Conflicting store by thread" (Test 3) in
OpenMP programs compiled/linked with the gcc/fortran flag
"-fopenmp" (which implies "-pthread");
(ii) In addition to the tips and suggestions given in the user
manual, could there be other means to mitigate these (bogus?)
messages?
Thanks in advance.
Regards.
Jorge D'Elia.
--
CIMEC (UNL-CONICET), http://www.cimec.org.ar/
Predio CONICET-Santa Fe, Colec. Ruta Nac. 168,
Paraje El Pozo, 3000, Santa Fe, ARGENTINA.
Tel +54-342-4511594/95 ext 7062, fax: +54-342-4511169
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Valgrind-users mailing list
Valgrind-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/valgrind-users
