[SYSTEMDS-2827] Rowwise template for CUDA codegen

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: SystemDS 2.1
• Epic Link: spoof-cuda
• Assignee: Mark Dokter
• Reporter: Mark Dokter
• Created: 04/Feb/21 23:22
• Updated: 26/Feb/21 12:38
• Resolved: 26/Feb/21 12:38

Description

Make the rowwise SPOOF template available in CUDA codegen. This also includes sparse access (issue #2826) because the rowwise template contains separate methods for dense and sparse. Also the plethora of vector primitives needs to be supported (plus a temporary memory ringbuffer)