On Thu, May 11, 2017, at 09:12, David Turner wrote:
Could anyone help, e.g. by pointing me at the bit in the Core that is allocating within the main loop?
GHC has a -ticky flag that tracks precisely where allocations are happening. It's quite low level (you may have to stare at the STG in addition to the Core to interpret the results) but I've found it incredibly useful when trying to understand performance swings in GHC's own benchmarks. https://ghc.haskell.org/trac/ghc/wiki/Debugging/TickyTicky You can enable it per module, which is nice break from the rebuild-the-world -prof, but the flip side IIRC is that if the allocations are happening in a module that wasn't compiled with -ticky (e.g. somewhere inside Data.Array) you might not see them. Hope this helps! Eric