[SYSTEMDS-1547] Cache-conscious post processing tsmm

• Type: Task
• Status: Closed
• Priority: Major
• Resolution: Done
• Fix Version/s: SystemML 0.14
• Assignee: Matthias Boehm
• Reporter: Matthias Boehm
• Created: 19/Apr/17 22:56
• Updated: 09/Aug/17 20:43
• Resolved: 09/Aug/17 20:43

Description

For scenarios like tsmm outer products t(X) %*% X, where X is a row vector, the post-processing in terms of (1) copying the upper to lower triangle, and (2) nnz recomputation significantly contribute to the execution time.

This task aims at a cache-conscious copy of upper to lower triangles similar to dense-dense transpose as well as fusing the nnz recompuation into either the computation or copy operation.