Implement MPI_Gatherv Wrapper for String Array as DataType and add bindings for MPI_LOGICAL and MPI_CHARACTER #126
Conversation
|
lfortran is failing for the |
|
I locally get this after this diff aditya-trivedi tests gatherv_char ≢ ?1 ~1 git diff
diff --git a/tests/gatherv_2.f90 b/tests/gatherv_2.f90
index 8092207..3125f87 100644
--- a/tests/gatherv_2.f90
+++ b/tests/gatherv_2.f90
@@ -28,7 +28,7 @@ program gatherv_pfunit_1
numEntries = rank + 1
allocate(sendBuffer(numEntries))
do i = 1, numEntries
- write(sendBuffer(i), '(A,I0)') 'proc', rank ! Dummy words: "proc0", "proc1", etc.
+ sendBuffer(i) = 'proc' // trim(adjustl(int2str(rank))) ! Dummy words: "proc0", "proc1", etc.
end do
! Allocate receive buffers on root
LFortran Output Details#################################
Using FC=lfortran --cpp compiler
Using CC=gcc compiler
################################
Received argument(s). Will only compile/run: gatherv_2.f90
Compiling gatherv_2...
Running gatherv_2 with 1 MPI ranks...
MPI_Gatherv pFUnit test passed on root
Test gatherv_2 with 1 MPI ranks PASSED!
Running gatherv_2 with 2 MPI ranks...
malloc(): corrupted top size
[Observer:30822] *** Process received signal ***
[Observer:30822] Signal: Aborted (6)
[Observer:30822] Signal code: (-6)
[Observer:30822] [ 0] /lib/x86_64-linux-gnu/libc.so.6(+0x45330)[0x7d7579445330]
[Observer:30822] [ 1] /lib/x86_64-linux-gnu/libc.so.6(pthread_kill+0x11c)[0x7d757949eb2c]
[Observer:30822] [ 2] /lib/x86_64-linux-gnu/libc.so.6(gsignal+0x1e)[0x7d757944527e]
[Observer:30822] [ 3] /lib/x86_64-linux-gnu/libc.so.6(abort+0xdf)[0x7d75794288ff]
[Observer:30822] [ 4] /lib/x86_64-linux-gnu/libc.so.6(+0x297b6)[0x7d75794297b6]
[Observer:30822] [ 5] /lib/x86_64-linux-gnu/libc.so.6(+0xa8ff5)[0x7d75794a8ff5]
[Observer:30822] [ 6] /lib/x86_64-linux-gnu/libc.so.6(+0xac2fc)[0x7d75794ac2fc]
[Observer:30822] [ 7] /lib/x86_64-linux-gnu/libc.so.6(malloc+0xa4)[0x7d75794ad6f4]
[Observer:30822] [ 8] /home/aditya-trivedi/thats_me/main/lfortran/src/bin/../runtime/liblfortran_runtime.so.0(_lfortran_str_item+0x7f)[0x7d7579c09ebb]
[Observer:30822] [ 9] ./gatherv_2(+0x291c)[0x5f0ce8c3591c]
[Observer:30822] [10] ./gatherv_2(+0x2af0)[0x5f0ce8c35af0]
[Observer:30822] [11] ./gatherv_2(+0x3fd9)[0x5f0ce8c36fd9]
[Observer:30822] [12] /lib/x86_64-linux-gnu/libc.so.6(+0x2a1ca)[0x7d757942a1ca]
[Observer:30822] [13] /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0x8b)[0x7d757942a28b]
[Observer:30822] [14] ./gatherv_2(+0x2209)[0x5f0ce8c35209]
[Observer:30822] *** End of error message ***
--------------------------------------------------------------------------
prterun noticed that process rank 0 with PID 30822 on node Observer exited on
signal 6 (Aborted).
--------------------------------------------------------------------------
Test gatherv_2 with 2 MPI ranks FAILED!
aditya-trivedi tests gatherv_char ≢ ?1 ~1 echo $?
1
While GFortran passes the test Output#################################
Using FC=gfortran -cpp compiler
Using CC=gcc compiler
################################
Received argument(s). Will only compile/run: gatherv_2.f90
Compiling gatherv_2...
Running gatherv_2 with 1 MPI ranks...
MPI_Gatherv pFUnit test passed on root
Test gatherv_2 with 1 MPI ranks PASSED!
Running gatherv_2 with 2 MPI ranks...
MPI_Gatherv pFUnit test passed on root
Test gatherv_2 with 2 MPI ranks PASSED!
Running gatherv_2 with 4 MPI ranks...
MPI_Gatherv pFUnit test passed on root
Test gatherv_2 with 4 MPI ranks PASSED!
... Running standalone tests took 5 seconds ... |
|
Try to figure out some workaround that works in LFortran. If not easy, then we need to fix LFortran first. |
|
I think we might need to fix LFortran first To mimic it i must use MPI to make it use 2 process at a time and if i use it , then the exact point of failure is kinda lost. Or I'm not able to get it either way I may need help here |
|
Let me look into this today. |
|
I’ve raised an issue on the fortran_mpi repository to extract a minimal reproducible example (MRE): #128. We still need to report this to LFortran — I haven’t had the time today, but I’ll do that later. I suspect the issue is related to Fortran intrinsics like write, adjustl, or trim. In the meantime, I’ve added a new test case for MPI_Gatherv and pushed it as the latest commit in this PR. If the tests pass, I’ll go ahead and merge it. Fortunately, this test doesn’t use the intrinsics that caused the previous failure, so it should be more stable. |
|
The tests pass, I'm merging this. |
|
Thanks @gxyd !! |
Towards #121