#10397: Compiler performance regression 7.6 -> 7.8 in elimCommonBlocks -------------------------------------+------------------------------------- Reporter: TobyGoodwin | Owner: Type: bug | Status: merge Priority: normal | Milestone: 7.10.2 Component: Compiler | Version: 7.8.4 Resolution: | Keywords: Operating System: Unknown/Multiple | performance Type of failure: None/Unknown | Architecture: Blocked By: | Unknown/Multiple Related Tickets: | Test Case: see ticket | Blocking: | Differential Revisions: Phab:D892 -------------------------------------+------------------------------------- Comment (by nomeata): Replying to [comment:23 nomeata]:
I wonder if it is faster to hash the list of labels and group them using a single level IntMap instead of the current trie based on nested IntMap.... That’s worth a try.
no, it’s not. And profiling shows why: The time is still spent mostly in `eqBlockBodyWith`. With my refactoring, the number of calls to that function is minimal anyways, so the only way to improve the situation is * a better hash function (it currently does not take operators into account, and maybe it can also take “internal” functions into account) * no linear search by using a trie for blocks, similar how we do it for Core expressions. * lose precision. I’ll add some debug output to see how these blocks that cannot be merged look like. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/10397#comment:24 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler