On 2019-04-15 15:10, Claes Redestad wrote:
This reduces first-time bootstrap times for the targeted patterns from ~20ms to be lost in the noise on my machine (so ~0.5ms), while being peak performance neutral and not regressing bootstrap characteristics on more complex patterns.
Correction: I accidentally had some tuning flags applied to one set of runs but not the other. In a more fair apple-to-apples re-run the first- time bootstrap cost drops from ~30ms to ~9ms. The 20ms drop was correct in absolute terms, but there's still a measurable cost overhead of initializing SCF itself (this RFE helps amortize those costs in many instances) /Claes