-
Notifications
You must be signed in to change notification settings - Fork 162
Home
KaiGai Kohei edited this page Apr 3, 2017
·
29 revisions
- Pass-down Limit clause on ForeignScan/CustomScan (returned with feedback; v10)
- Callbacks to reclaim statistics of background-worker on DSM (v10 accepted)
- Long varlena support (works in progress)
- Matrix data type support
- Callbacks when all_visible is set/clear
- Based on PostgreSQL v10
- In-memory columnar cache
- GpuPreAgg + GpuJoin + GpuScan by mono-kernel
- PCI-E topology consideration on SSD-to-GPU P2P DMA
- Long varlena based Matrix support (PL/CUDA)
- Fully utilization of unified device memory on Pascal model.
- Minimum requirement shall be pulled up to CC6.0 or later.
- CUDA Infrastructure Renew
- CUDA/GPU Management Server using BGworker (done)
- Portable Pointer on dynamic shared memory segment (done)
- Update of GpuScan (done)
- Update of GpuJoin (done)
- Update of GpuPreAgg (done)
- Update of GpuSort; only valuable when LIMIT is used together (postponed)
- Needs PostgreSQL v10 infrastructure
- Based on PostgreSQL v9.6
- CPU+GPU Hybrid Parallelism (done)
- Support of the new planner structure (done)
- SSD-to-GPU Direct Block Loading (done)
- BRIN index support on GpuScan
- Incremental improvement around optimizer stuff
- ...and others
- Based on PostgreSQL v9.5
- Dynamic Parallelism Adoption (done)
- CPU Fallback of GpuJoin (done)
- GpuSort reworks (done)
- LIKE operator support (done)
- Official Documentation (done)
- Regression Tests (postpone)
- RPM Package (done)
- COALESCE, GREATEST, LEAST operators (done)
- ScalarArrayOp operators (done)
- PL/CUDA support and Matrix/Vector data type (done)
- Under construction