I've never checked if FRUIT works with OpenMP.
However probably FRUIT is not thread safe, because
* SAVE attribute is used in fruit.f90
* fixed I/O unit number is used
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Tried a tester code that uses OpenMP.
Tester code is:
modulesubs_test
!$ useomp_libusefruitimplicitnonecontainssubroutinetest_just_sumusesubs, only : just_suminteger :: resultcharacter(50) :: str
!$ompparallelwrite(str, '("thread", i3, " / ", i3)') omp_get_thread_num() + 1, omp_get_num_threads()
result = just_sum(2, 3)
callassert_equals(6, result, " it should fail, " // trim(str))
result = just_sum(4, 5)
callassert_equals(9, result, " it should success, " // trim(str))
!$ompendparallelendsubroutinetest_just_sumendmodulesubs_test
Tested code is:
modulesubsimplicitnonecontainsintegerfunctionjust_sum(a, b)
integer, intent(in) :: a, bjust_sum = a + bendfunctionjust_sumendmodulesubs
Each therad supposed to report one fail and one success.
Using two threads, result was OK. (log below)
Test module initialized
. : successful assert, F : failed assert
..running test: test_just_sum
F.F.
Un-satisfied spec:
-- just sum
Start of FRUIT summary:
Some tests failed!
-- Failed assertion messages:
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 1 / 2]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 2 / 2]
-- end of failed assertion messages.
Total asserts : 4
Successful : 2
Failed : 2
Successful rate: 50.00%
Successful asserts / total asserts : [ 2 / 4 ]
Successful cases / total cases : [ 0 / 1 ]
-- end of FRUIT summary
with 16 threads, message was corrupted.
Test module initialized
. : successful assert, F : failed assert
..running test: test_just_sum
F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.
Un-satisfied spec:
-- just sum
Start of FRUIT summary:
Some tests failed!
-- Failed assertion messages:
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 2 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 3 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 4 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 5 / 16]
[test_just_sum]:Expected; User message: [ it should fail, thread 1 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 6 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 7 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 13 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 11 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 8 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 10 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 15 / 16]
[test_just_sum]:Expected [6], Got [5] [5]; User message: [ it should fail, thread 12 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 9 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 14 / 16]
[test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 16 / 16]
-- end of failed assertion messages.
Total asserts : 32
Successful : 16
Failed : 16
Successful rate: 50.00%
Successful asserts / total asserts : [ 16 / 32 ]
Successful cases / total cases : [ 0 / 1 ]
-- end of FRUIT summary
So, fruit.f90 needs some more modification.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi istomoya,
Thanks for you providing test code with openMP. I do think message corrupted is caused by race condition. I also parallize my testing code with openMP directives and it works fine. I just use asserts from fruit to be parallelized and not let the whole framwork be parallelilzed. So I consider multi-threads doesn't impact asserts from fruit. What do you think of that?
module base
integer :: baseint
end module
module a
use base, ONLY: aint => baseint
end module
module b
use base, ONLY: bint => baseint
end module
module c
use a
use b
private
public :: aint, bint
end module
module launch_test
use fruit
implicit none
contains
subroutine launchHello
use c, only : aint, bint
aint = 3
bint = 5
write (, ) " ..running test: launchHello"
call assert_equals(aint, bint)
call assert_equals(aint+1, bint)
end subroutine
end module
module fruit_gen
use omp_lib
use launch_test
contains
subroutine hxlaunch_mimic
!$OMP PARALLEL DO
do i=1, 10
write (,) "thread id=",OMP_GET_THREAD_NUM()
call launchHello
end do
!$OMP END PARALLEL DO
end subroutine
subroutine test_all_tests
call set_unit_name('launch_test')
call run_test_case(hxlaunch_mimic, "launchHello")
end subroutine
end module
=======================================
Test module initialized
Hi istomoya,
I think you are right. When I increase theads to 100+, app gets a crash maybe.
..running test: launchHello
.F thread id= 20
..running test: launchHello
.F. thread id= 43
..running test: launchHello
.FF. ..running test: launchHello
. thread id= 33
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Backtrace for this error:
Backtrace for this error:
I think a easy way to fix this should add a mutex lock for subroutine failed_assert_action and add_success. But I'm not an expert in fortran. If possible, can you please help me on this. Thanks a lot.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi ismotoya,
I'm confused by test result. When I created 300 thread to launch a subrotine which contains assers from fruit. It get failed 1/20 approximately. Most of times it gets passed.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi istomoya,
Thanks for your effort on updating framework. I have a try on the latest version and run same app 100 times with 500 threads. Each run have 5 secondes sleep. Unfortunately it gets failed once.
If with 100 threads, all tests get passed.
33227 Program received signal SIGABRT: Process abort signal.
33228
33229 Backtrace for this error:
33230 ...........
33231 .........F.
33232 Program received signal SIGABRT: Process abort signal.
33233
33234 Backtrace for this error:
33235 FF.F.F.F.FFF.....F.FFFFFF.FF..FFFF.#0 0x7F886B344417
33236 ..#1 0x7F886B344A2E
33237 F#2 0x7F886A41F49F
33238 #3 0x7F886A41F425
33239 F.#4 0x7F886A422B8A
33240 F#5 0x7F886A45D39D
33241 F#6 F0x7F886A467B95
33242 ....#7 0x40CE7D in fruit_MOD_increase_message_stack_ at fruit.f90:0
33243 .#8 0x40BECC in fruit_MOD_failed_assert_action
33244 .F#9 0x408FE4 in fruit_MOD_assert_eq_int_
33245 F.#10 0x40FFDE in launch_test_MOD_launchhello
33246 F.F#11 0x410146 in __fruit_gen_MOD_hxlaunch_mimic._omp_fn.0 at access_spec.f90:0
33247 #12 0x7F886AE26865
33248 FFF#13 0x7F886A7B0E99
33249 F#14 0x7F886A4DD3FC
33250 F..F.F..
33251 F..
33252 Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
33253
33254 Backtrace for this error:
33255 ...FFFFFF.F....
33256 Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi istomoya,
I have another method to fix thread sync issue.
I replaced !#omp critical (xxxx) with !#omp single.
Then I created 3000 theads and each to run asserts 20 times. Test app doesn't get a crash this time. Thanks for your great effrot on this issue.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you for using FRUIT.
I afraid that using !#omp single ,number of assertions (both fails and successes) are reduced in the report.
Only assertations ran by one thread is counted and assertations called from other threads will not be counted.
In my case, only one error reported while 100 errors should have reported.
Well, I'll keep on trying to improve CVS version.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi istomoya,
I have a try with the latest version 3.4.0 and it seems work fine with 200 threads. I run test app several times and it doesn't get crashed. But if increasing threads to 300, some error message will output.
A minor bug maybe there is about error message output as below:
...FF thread id= 2955
..running test: anotherHello2
.F thread id= 2248
..running test: anotherHello1
...F
.F thread id= 2016
..running test: anotherHello
...F..F thread id= 2268
..running test: anotherHello
..F...FFF thread id= 1533
..running test: anotherHello4
.FF thread id= 1815
..running test: anotherHello2
.F thread id= 1692
..running test: anotherHello
.FFF thread id= 805
..running test: anotherHello4
F0n�pc�te
Last edit: fan xin 2015-11-23
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi istomoya,
There maybe still bugs in final summary. Please consider a test as below:
module launch_test
subroutine launchHello
use c, only : aint, bint
aint = 3
bint = 5
write (, ) " ..running test: launchHello"
call assert_equals(aint, bint)
call assert_equals(aint+1, bint)
end subroutine
end module
module fruit_gen
use omp_lib
use launch_test
contains
subroutine hxlaunch_mimic
integer :: tid
integer :: N=3000
call OMP_SET_NUM_THREADS(N)
!$OMP PARALLEL private(tid)
tid = OMP_GET_THREAD_NUM()
if (tid .eq. 0) then
print ,"number of threads=", OMP_GET_NUM_THREADS()
end if
write (,*) "thread id=",tid
call launchHello
!$OMP END PARALLEL
end subroutine
I Have launched 3000 threads and each to assert twice. I think total asserts should be 6000. But it report:
Total asserts : 5000
Successful : 3000
Failed : 2000
Successful rate: 60.00%
Successful asserts / total asserts : [ 3000 / 5000 ]
Successful cases / total cases : [ 0 / 0 ]
-- end of FRUIT summary
..STOP 1
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Number of errors is limited to 2000 (given as MAX_MAG_STACK_SIZE) in fruit.f90.
Whole test had been simply stopped (without caring about OpenMP) when this limit was reached.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I'm a newbie to use fruit for my testing work. I'd like to make sure with you that is fruit thread safe and works fine with OpenMP? Thanks.
I've never checked if FRUIT works with OpenMP.
However probably FRUIT is not thread safe, because
* SAVE attribute is used in fruit.f90
* fixed I/O unit number is used
Hi istomoya,
Thanks for your kindly help. Although I'm disappointed to hear that answer.
Tried a tester code that uses OpenMP.
Tester code is:
Tested code is:
Each therad supposed to report one fail and one success.
Using two threads, result was OK. (log below)
Test module initialized . : successful assert, F : failed assert ..running test: test_just_sum F.F. Un-satisfied spec: -- just sum Start of FRUIT summary: Some tests failed! -- Failed assertion messages: [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 1 / 2] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 2 / 2] -- end of failed assertion messages. Total asserts : 4 Successful : 2 Failed : 2 Successful rate: 50.00% Successful asserts / total asserts : [ 2 / 4 ] Successful cases / total cases : [ 0 / 1 ] -- end of FRUIT summarywith 16 threads, message was corrupted.
Test module initialized . : successful assert, F : failed assert ..running test: test_just_sum F.F.F.F.F.F.F.F.F.F.F.F.F.F.F.F. Un-satisfied spec: -- just sum Start of FRUIT summary: Some tests failed! -- Failed assertion messages: [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 2 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 3 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 4 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 5 / 16] [test_just_sum]:Expected; User message: [ it should fail, thread 1 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 6 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 7 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 13 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 11 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 8 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 10 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 15 / 16] [test_just_sum]:Expected [6], Got [5] [5]; User message: [ it should fail, thread 12 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 9 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 14 / 16] [test_just_sum]:Expected [6], Got [5]; User message: [ it should fail, thread 16 / 16] -- end of failed assertion messages. Total asserts : 32 Successful : 16 Failed : 16 Successful rate: 50.00% Successful asserts / total asserts : [ 16 / 32 ] Successful cases / total cases : [ 0 / 1 ] -- end of FRUIT summarySo, fruit.f90 needs some more modification.
Hi istomoya,
Thanks for you providing test code with openMP. I do think message corrupted is caused by race condition. I also parallize my testing code with openMP directives and it works fine. I just use asserts from fruit to be parallelized and not let the whole framwork be parallelilzed. So I consider multi-threads doesn't impact asserts from fruit. What do you think of that?
module base
integer :: baseint
end module
module a
use base, ONLY: aint => baseint
end module
module b
use base, ONLY: bint => baseint
end module
module c
use a
use b
private
public :: aint, bint
end module
module launch_test
use fruit
implicit none
contains
subroutine launchHello
use c, only : aint, bint
aint = 3
bint = 5
write (, ) " ..running test: launchHello"
call assert_equals(aint, bint)
call assert_equals(aint+1, bint)
end subroutine
end module
module fruit_gen
use omp_lib
use launch_test
contains
subroutine hxlaunch_mimic
!$OMP PARALLEL DO
do i=1, 10
write (,) "thread id=",OMP_GET_THREAD_NUM()
call launchHello
end do
!$OMP END PARALLEL DO
end subroutine
subroutine test_all_tests call set_unit_name('launch_test') call run_test_case(hxlaunch_mimic, "launchHello") end subroutineend module
=======================================
Test module initialized
thread id= 4
..running test: launchHello
thread id= 1
..running test: launchHello
. thread id= 2
..running test: launchHello
. thread id= 0
..running test: launchHello
.. thread id= 7
..running test: launchHello
. thread id= 6
..running test: launchHello
. thread id= 5
..running test: launchHello
. thread id= 3
..running test: launchHello
.FFF thread id= 0
..running test: launchHello
.FF thread id= 1
..running test: launchHello
.FFFFF
Some tests failed!
-- Failed assertion messages:
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
[launchHello]:Expected [6], Got [5]
-- end of failed assertion messages.
Total asserts : 20
Successful : 10
Failed : 10
Successful rate: 50.00%
Successful asserts / total asserts : [ 10 / 20 ]
Successful cases / total cases : [ 0 / 1 ]
-- end of FRUIT summary
Hi istomoya,
I think you are right. When I increase theads to 100+, app gets a crash maybe.
..running test: launchHello
.F thread id= 20
..running test: launchHello
.F. thread id= 43
..running test: launchHello
.FF. ..running test: launchHello
. thread id= 33
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Backtrace for this error:
Backtrace for this error:
I think a easy way to fix this should add a mutex lock for subroutine failed_assert_action and add_success. But I'm not an expert in fortran. If possible, can you please help me on this. Thanks a lot.
Hi ismotoya,
I'm confused by test result. When I created 300 thread to launch a subrotine which contains assers from fruit. It get failed 1/20 approximately. Most of times it gets passed.
I tried adding several pairs of "critical" directives to fruit.f90.
Could you try fruit.f90 uploaded to CVS?
http://fortranxunit.cvs.sourceforge.net/viewvc/fortranxunit/fruit/src/fruit.f90?view=log
Hi istomoya,
Thanks for your effort on updating framework. I have a try on the latest version and run same app 100 times with 500 threads. Each run have 5 secondes sleep. Unfortunately it gets failed once.
If with 100 threads, all tests get passed.
33227 Program received signal SIGABRT: Process abort signal.
33228
33229 Backtrace for this error:
33230 ...........
33231 .........F.
33232 Program received signal SIGABRT: Process abort signal.
33233
33234 Backtrace for this error:
33235 FF.F.F.F.FFF.....F.FFFFFF.FF..FFFF.#0 0x7F886B344417
33236 ..#1 0x7F886B344A2E
33237 F#2 0x7F886A41F49F
33238 #3 0x7F886A41F425
33239 F.#4 0x7F886A422B8A
33240 F#5 0x7F886A45D39D
33241 F#6 F0x7F886A467B95
33242 ....#7 0x40CE7D in fruit_MOD_increase_message_stack_ at fruit.f90:0
33243 .#8 0x40BECC in fruit_MOD_failed_assert_action
33244 .F#9 0x408FE4 in fruit_MOD_assert_eq_int_
33245 F.#10 0x40FFDE in launch_test_MOD_launchhello
33246 F.F#11 0x410146 in __fruit_gen_MOD_hxlaunch_mimic._omp_fn.0 at access_spec.f90:0
33247 #12 0x7F886AE26865
33248 FFF#13 0x7F886A7B0E99
33249 F#14 0x7F886A4DD3FC
33250 F..F.F..
33251 F..
33252 Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
33253
33254 Backtrace for this error:
33255 ...FFFFFF.F....
33256 Program received signal SIGSEGV: Segmentation fault - invalid memory reference.
Hi istomoya,
I have another method to fix thread sync issue.
I replaced !#omp critical (xxxx) with !#omp single.
Then I created 3000 theads and each to run asserts 20 times. Test app doesn't get a crash this time. Thanks for your great effrot on this issue.
Thank you for using FRUIT.
I afraid that using !#omp single ,number of assertions (both fails and successes) are reduced in the report.
Only assertations ran by one thread is counted and assertations called from other threads will not be counted.
In my case, only one error reported while 100 errors should have reported.
Well, I'll keep on trying to improve CVS version.
Hi istomoya,
You are right. I can reproduce this issue on my side. Do you mean there are bugs in framework still? Are there any chances to fix that?
I found a bug in fruit.f90 and fixed version is
http://fortranxunit.cvs.sourceforge.net/viewvc/fortranxunit/fruit/src/fruit.f90?view=log
as Revision 1.60.
Hi istomoya,
I have a try with the latest version 3.4.0 and it seems work fine with 200 threads. I run test app several times and it doesn't get crashed. But if increasing threads to 300, some error message will output.
A minor bug maybe there is about error message output as below:
...FF thread id= 2955
..running test: anotherHello2
.F thread id= 2248
..running test: anotherHello1
...F
.F thread id= 2016
..running test: anotherHello
...F..F thread id= 2268
..running test: anotherHello
..F...FFF thread id= 1533
..running test: anotherHello4
.FF thread id= 1815
..running test: anotherHello2
.F thread id= 1692
..running test: anotherHello
.FFF thread id= 805
..running test: anotherHello4
F0n�pc�te
Last edit: fan xin 2015-11-23
Hi istomoya,
There maybe still bugs in final summary. Please consider a test as below:
module launch_test
subroutine launchHello
use c, only : aint, bint
aint = 3
bint = 5
write (, ) " ..running test: launchHello"
call assert_equals(aint, bint)
call assert_equals(aint+1, bint)
end subroutine
end module
module fruit_gen
use omp_lib
use launch_test
contains
subroutine hxlaunch_mimic
integer :: tid
integer :: N=3000
call OMP_SET_NUM_THREADS(N)
!$OMP PARALLEL private(tid)
tid = OMP_GET_THREAD_NUM()
if (tid .eq. 0) then
print ,"number of threads=", OMP_GET_NUM_THREADS()
end if
write (,*) "thread id=",tid
call launchHello
!$OMP END PARALLEL
end subroutine
I Have launched 3000 threads and each to assert twice. I think total asserts should be 6000. But it report:
Total asserts : 5000
Successful : 3000
Failed : 2000
Successful rate: 60.00%
Successful asserts / total asserts : [ 3000 / 5000 ]
Successful cases / total cases : [ 0 / 0 ]
-- end of FRUIT summary
..STOP 1
Number of errors is limited to 2000 (given as MAX_MAG_STACK_SIZE) in fruit.f90.
Whole test had been simply stopped (without caring about OpenMP) when this limit was reached.
Hi istomoya,
You make great sense to me. So I think framework works fine with my tests. Thanks for your great effort!