- Fuse final update position functions into a single more efficient kernel - Refactor gpu_hist with a more explicit ellpack matrix representation
- Fuse final update position functions into a single more efficient kernel - Refactor gpu_hist with a more explicit ellpack matrix representation