RFR(S): 8188151: "assert(entry_for(catch_pco, handler_bcis->at(i), scope_depth)->pco() == handler_pcos->at(i))" failure with C1

Roland Westrelin rwestrel at redhat.com
Mon Oct 2 09:40:34 UTC 2017


http://cr.openjdk.java.net/~roland/8188151/webrev.00/

When Compilation::generate_exception_handler_table() walks the exception
handler information to populate the exception handler table, it has some
logic that removes duplicate handlers for one particular throwing pc and
it is wrong AFAICT.

That code iterates over already processed (handler_bci, scope_count,
entry_pco) triples stored in GrowableArrays bcis, scope_depths, pcos and
looks for entries for which handler_bci, scope_count are identical to
the current one. It does that by looking for an entry with same
handler_bci in the bcis array and then checks whether scope_count
matches too. The list of triples could be something like:

1: (13, 0, ..)
2: (13, 1, ..)

and the next triple to be process: (13, 1, ..) which is a duplicate of
2. That logic looks for a handler with bci 13, finds entry 1 which
doesn't have scope count 1. And concludes that there no duplicate
entry. It would need to look at the following entry too. Given scope
counts are sorted in increasing order, rather that iterate over the list
of triples from the start, looking for duplicates fromt the end of the
list fixes that problem.

Roland.


More information about the hotspot-compiler-dev mailing list