[SYSTEMDS-2061] Performance features ultra-sparse operations

• Type: Epic
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: SystemML 1.1
• Epic Name: Performance features ultra-sparse operations
• Assignee: Unassigned
• Reporter: Matthias Boehm
• Created: 08/Jan/18 05:43
• Updated: 11/Sep/20 09:47
• Resolved: 11/Sep/20 09:47

Description

This epic aims to address a number of performance shortcomings regarding ultra-sparse operations.

Sub-Tasks

• Performance sparse-sparse transpose: Closed
• Performance ultra-sparse data generation: Closed
• Performance sparse-sparse reshape: Closed
• Improved caching decision ultra-sparse matrices with invalid cp dims: Closed
• Upper bound nnz propagation over spark instructions: Closed
• Ultra-sparse cpmm empty block pruning: Closed
• Performance ultra-sparse reshape, especially colwise vector-matrix: Closed
• Set spark cpmm preferred join-multiply parallelism: Closed