CWLE improvements show up fault in AWLF F_intersect analysis
|Created on||Jan 07, 2010 15:30|
|Resolved on||Jan 07, 2010 16:49|
This one took a while to isolate... AWLFI inserts F_attachintersect data flow function calls to hold the results of partition-bounds/index-vector set computations for AWLF's use, in replacing a following _sel_VxA_( iv, producerWL). In today's adventure, the producerWL is a copy-WL, so all is well, at first (as in all adventures). Eventually, CWLE replaces the copy-WL by its predecessor (is there a better word for this?), which, in this case, happens to be a multi-op WL, producing two results. AWLF then comes along and blindly does the folding job, giving us an unpleasant goulash that also makes the _sel_VxA_ produce a non-scalar result. Things go downhill from there. I had a hard time finding a tiny example that fails.
To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information