Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
This reverts commit 23a5e30. Tests showed that the new 2x2 scheme was usually slower or only marginally faster (<1%) than the previous 3minus scheme both in Linux i7 desktop and Macbook Air i5 laptop, and therefore we revert to the previous 3minus scheme. This is strange, because profiling compiled code with operf hinted that 2x2 should be faster except for sparse data (bryceveg). Here microbenchmarks from Linux: > mb2x2 Unit: milliseconds expr min lq mean median uq max neval simulate(nullmodel(sipoo,"quasiswap"),100) 402 442 462 464 480 537 100 simulate(nullmodel(BCI,"quasiswap"),100) 6803 7461 7659 7726 7915 8483 100 simulate(nullmodel(bryceveg,"quasiswap"),100) 3301 3550 3682 3674 3805 4059 100 > mb3minus Unit: milliseconds expr min lq mean median uq max neval simulate(nullmodel(sipoo,"quasiswap"),100) 358 404 422 422 440 642 100 simulate(nullmodel(BCI,"quasiswap"),100) 7101 7517 7689 7686 7830 8382 100 simulate(nullmodel(bryceveg,"quasiswap"),100) 3007 3088 3137 3124 3168 3462 100
- Loading branch information