[SYSTEMDS-2698] Initial cuda codegen functionality

• Type: New Feature
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: SystemDS 2.1
• Fix Version/s: SystemDS 2.1
• Component/s: Compiler, NativeRuntime
• Labels: CUDA GPU code-generation
• External issue URL: https://github.com/apache/systemds/pull/1092
• Epic Link: spoof-cuda
• Assignee: Mark Dokter
• Reporter: Mark Dokter
• Due: 21/Oct/20
• Created: 20/Oct/20 19:47
• Updated: 09/Nov/20 11:17
• Resolved: 09/Nov/20 11:17

Description

Cleaned up/improved pull request for initial CUDA codegen. This includes code cleanup and grouping the changes into a few commits (original PR had 60+ commits and was hard to review)

In terms of functionality, this PR will include:

• CUDA code reorganization
• JNI parts to launch generated CUDA kernels
• SPOOF compiler extensions
• CPlan templates
• Runtime instruction for dense input
• Code generation for the SpoofCellwise template

Issue Links

• is blocked by: SYSTEMDS-2692 move cuda codebase