APEX logd3.sac slower with AWLF than with WLF, due to WLUR
|Created on||Mar 01, 2012 16:39|
|Resolved on||Jul 06, 2012 20:45|
|Attachments||logd3.sac, logd3.sac.boptuglf.awlf.17747_MODIFIED, logd3.sac.boptuglf.O3.17747_MODIFIED|
Created an attachment (id=857) source code to reproduce failure sac2c -V sac2c v1.00-beta (Haggis And Apple) developer rev 17747:MODIFIED linux-gnu_x86_64 (Wed Feb 29 17:00:50 EST 2012 by sac) The benchmark above runs significantly slower ( nearly 2X) with -doawlf -nowlf than with -noawlf -dowlf. The apparent cause is a failure to perform AWLF fast enough to beat WLUR to the punch: The AWLF version shows an unrolled (1-element) WL selecting one element from an otherwise unreferenced (and big) WL. In the WLF version, WLF apparently folds the two WLs before WLUR can take action. I do not see a swell, simple way to resolve this problem. One slightly unpleasant approach might be to extend AWLF to operate on an unrolled consumer-WL, as if it were a WL.
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information