Merge performance enhancements enabling feasible XLA compilation
This merge request addresses the issue of unfeasibly slow XLA compilation. Speed (V100) has increased from 1.2it/s to 22it/s giving a 18x speedup.
This merge request addresses the issue of unfeasibly slow XLA compilation. Speed (V100) has increased from 1.2it/s to 22it/s giving a 18x speedup.