#367: Infinite loops can hang Concurrent Haskell
------------------------------------------+---------------------------------
Reporter: simonpj | Owner: ezyang
Type: bug | Status: new
Priority: lowest | Milestone: _|_
Component: Compiler | Version: 6.4.1
Resolution: None | Keywords: scheduler
allocation
Os: Unknown/Multiple | Architecture: Unknown/Multiple
Failure: Incorrect result at runtime | Difficulty: Unknown
Testcase: | Blockedby:
Blocking: | Related:
------------------------------------------+---------------------------------
Comment(by simonmar):
Replying to [comment:20 ezyang]:
> Replying to [comment:19 simonmar]:
> > You could improve code size by omitting the `HpAlloc = 0` assignment
(perhaps making sure that it is initialized to zero in `LOAD_THREAD_STATE`
or something).
>
> Fascinatingly enough, this doesn't help all that much, since instruction
alignments adds in nops to fill in the space savings.
But we don't actually align the heap-check failure branch, so I'm
confused. Can you post the asm code you're seeing?
> > Another alternative is to use `SpLim` instead of `HpLim` to trigger
the interrupt, on the grounds that there are more stack checks than heap
checks. We would have to put `SpLim` in a memory location instead of a
register, but we could move `HpLim` into a register.
>
> To be clear, this is changing globally how preemption would work, since
prior to this patch we were zeroing HpLim to trigger a yield. But it
should otherwise work. I'll chase up some stats here too. (If SpLim is
checked more often, won't we pay a performance cost for having it in a
memory location?)
Maybe, but it's worth measuring I think.
> > Something else we could do is add a flag on every top-level function
to say whether it is non-allocating (rather like the `NoCafRefs` flag),
and we could use that to optimise away many of the extra checks.
>
> I don't quite understand what this means: isn't alloc = 0 in the heap
check just this information?
I had in mind making it a transitive property - a function would get the
alloc flag if it is guaranteed to allocate within a bounded time, so then
any callers don't need a yield check.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/367#comment:22>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
