On 1 Mar 2021, at 6:29 AM, Zhang, Hong <hzhang@mcs.anl.gov> wrote:

Pierre,

This is a bug in MatMatMultSymbolic_MPIAIJ_MPIDense() during optimization of block column size of B. Run your code with

'-matmatmult_Bbn 1', the infinite loop should not occur.

I'll try to figure out a fix tomorrow.

Hong

---

From: Zhang, Hong <hzhang@mcs.anl.gov>
Sent: Sunday, February 28, 2021 11:05 PM
To: Pierre Jolivet <pierre@joliv.et>; For users of the development version of PETSc <petsc-dev@mcs.anl.gov>; Zhang, Hong <hzhang@mcs.anl.gov>
Subject: Re: [petsc-dev] Infinite loop in A*B

 

The infinite loop in MatMatMultNumeric_MPIAIJ_MPIDense()

for (i=0; i<BN; i+=n) {

}

is cause by n=contents->workB->cmap->n=0 (line 590 in mpimatmatmult.c)

Hong

---

From: petsc-dev <petsc-dev-bounces@mcs.anl.gov> on behalf of Zhang, Hong via petsc-dev <petsc-dev@mcs.anl.gov>
Sent: Sunday, February 28, 2021 10:33 PM
To: Pierre Jolivet <pierre@joliv.et>; For users of the development version of PETSc <petsc-dev@mcs.anl.gov>
Subject: Re: [petsc-dev] Infinite loop in A*B

 

I can reproduce the hang with 

mpiexec -n 2 ./matmatmult

It seems in an infinite loop of calling MatDensePlaceArray() from 

#0  MatDensePlaceArray (mat=0xda5c50, array=0xd15e60)

    at /home/hongsu/soft/petsc/src/mat/impls/dense/mpi/mpidense.c:2047

#1  0x00007fa0d13bf4f7 in MatDenseGetSubMatrix_SeqDense (A=0xcfb2b0, cbegin=0,

    cend=0, v=0xd90370)

    at /home/hongsu/soft/petsc/src/mat/impls/dense/seq/dense.c:2997

#2  0x00007fa0d13c574e in MatDenseGetSubMatrix (A=0xcfb2b0, cbegin=0, cend=0,

    v=0xd90370) at /home/hongsu/soft/petsc/src/mat/impls/dense/seq/dense.c:3371

#3  0x00007fa0d13db5ce in MatDenseGetSubMatrix_MPIDense (A=0xca5250, cbegin=0,

    cend=0, v=0x7ffe87d41de0)

    at /home/hongsu/soft/petsc/src/mat/impls/dense/mpi/mpidense.c:1835

#4  0x00007fa0d13c574e in MatDenseGetSubMatrix (A=0xca5250, cbegin=0, cend=0,

    v=0x7ffe87d41de0)

    at /home/hongsu/soft/petsc/src/mat/impls/dense/seq/dense.c:3371

#5  0x00007fa0d179c2fa in MatMatMultNumeric_MPIAIJ_MPIDense (A=0xc55490,

    B=0xca5250, C=0xd282b0)

    at /home/hongsu/soft/petsc/src/mat/impls/aij/mpi/mpimatmatmult.c:593

#6  0x00007fa0d1181331 in MatProductNumeric_AB (mat=0xd282b0)

    at /home/hongsu/soft/petsc/src/mat/interface/matproduct.c:567

#7  0x00007fa0d1182c14 in MatProductNumeric (mat=0xd282b0)

    at /home/hongsu/soft/petsc/src/mat/interface/matproduct.c:679

#8  0x00007fa0d115ef69 in MatProduct_Private (A=0xc55490, B=0xca5250,

    scall=MAT_INITIAL_MATRIX, fill=-2, ptype=MATPRODUCT_AB, C=0x7ffe87d42018)

    at /home/hongsu/soft/petsc/src/mat/interface/matrix.c:9405

---Type <return> to continue, or q <return> to quit---

#9  0x00007fa0d115f274 in MatMatMult (A=0xc55490, B=0xca5250, scall=MAT_INITIAL_MATRIX, fill=-2,

    C=0x7ffe87d42018) at /home/hongsu/soft/petsc/src/mat/interface/matrix.c:9445

#10 0x000000000040130a in main (argc=2, argv=0x7ffe87d42108) at ex1.c:20

I'll try to figure out what is going on. If anyone has a clue, please help. The above stack comes from 'release' branch.

Hong

---

From: petsc-dev <petsc-dev-bounces@mcs.anl.gov> on behalf of Pierre Jolivet <pierre@joliv.et>
Sent: Sunday, February 28, 2021 4:17 PM
To: For users of the development version of PETSc <petsc-dev@mcs.anl.gov>
Subject: [petsc-dev] Infinite loop in A*B

 

Hello,
The following MWE loops indefinitely for MPI_Comm_size in {2; 3}.
Nothing fancy, just MatAIJ and MatDense.
The problem is either in MatMPIDenseScatter() or MatMatMultSymbolic_MPIAIJ_MPIDense(), I believe, so if someone familiar with those routines can figure out a hot fix, I’m all ears.
I could of course switch to a MatMult(), but the same infinite loop happens in another more complex code with
A = rows=8, cols=35212
B = rows=35212, cols=9
So I’ll need a fix eventually.

Thanks,
Pierre