#8256: adding locality levels to prefetch# and friends -------------------------------------+------------------------------------ Reporter: carter | Owner: carter Type: task | Status: patch Priority: normal | Milestone: 7.8.1 Component: Compiler | Version: 7.7 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: #8252 ,#8107 -------------------------------------+------------------------------------ Comment (by carter): its worth noting that the "right" benchmark for prefetch would be something like having two large arrays, one of which is indices into the other array, chosen randomly. the benchmark would be comparing a naive random lookup vs a prefetched one. (the idea being that the test fails if prefetch isn't faster than naive). i'll cook up this example some other time. but would be the only sane way to benchmark if prefetch is working or not! (because prefetch on any valid memory address will not change the naive perf correctness). NB: the one way prefetching could interact with correctness in an unrelated to performance way, would be in relation to all those various side channel timing based security issues. proper mitigation would probably require ghc also supporting streaming reads and writes, in addition to prefetch. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8256#comment:23 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler