(systemds) branch main updated: [SYSTEMDS-3769] Fix sparse-dense matrix multiplication edge cases

[mboehm7]

This is an automated email from the ASF dual-hosted git repository.

mboehm7 pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/systemds.git


The following commit(s) were added to refs/heads/main by this push:
     new 3e1eeb534d [SYSTEMDS-3769] Fix sparse-dense matrix multiplication edge 
cases
3e1eeb534d is described below

commit 3e1eeb534d18a11fef634bbb2c42c78b45683400
Author: Matthias Boehm <mboe...@gmail.com>
AuthorDate: Fri Sep 13 15:36:23 2024 +0200

    [SYSTEMDS-3769] Fix sparse-dense matrix multiplication edge cases
    
    This patch fixes index-out-of-bounds issues of our LibMatrixMult for
    parallel sparse-dense operations with 13x8K mm 8Kx1 matrices which
    uses a matrix-vector multiplication kernel that does not support pm2
    parallelization (where we parallelize over the rhs matrix).
---
 src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git 
a/src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java 
b/src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java
index 3ea882abe3..061c09ea3d 100644
--- a/src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java
+++ b/src/main/java/org/apache/sysds/runtime/matrix/data/LibMatrixMult.java
@@ -4332,7 +4332,7 @@ public class LibMatrixMult
                //parallelize over rows in rhs matrix if number of rows in 
lhs/output is very small
                double jvmMem = InfrastructureAnalyzer.getLocalMaxMemory();
                return (m1.rlen==1 && !(m1.isUltraSparse()||m2.isUltraSparse()))
-                       || (m1.rlen<=16 && m2.rlen > m1.rlen 
+                       || (m1.rlen<=16 && m2.rlen > m1.rlen && (!m1.sparse | 
m2.clen > 1)
                           && ( !m1.isUltraSparse() && !(m1.sparse & m2.sparse) 
) //dense-dense / sparse-dense / dense-sparse
                           && (long)k * 8 * m1.rlen * m2.clen < 
Math.max(MEM_OVERHEAD_THRESHOLD,0.01*jvmMem) );
        }