How do we best make sure {Int,Word,Float,Double} to {Int,Word,Float,Double} conversions stay efficient
Hi, I've been tracking down a few (unrelated) performance bugs related to conversions between primitive types. What these issues had in common is that some rule failed to fire and the conversion went via Integer, killing performance. How do we best write a test that make sure that fromIntegral conversions between these types don't regress? Is it possible to test the output of the simplifier or do we have to do something less direct, like writing a loop that does such conversions and make sure that the allocation stay low? -- Johan
On Fri, Dec 14, 2012 at 12:53:36PM -0800, Johan Tibell wrote:
I've been tracking down a few (unrelated) performance bugs related to conversions between primitive types. What these issues had in common is that some rule failed to fire and the conversion went via Integer, killing performance. How do we best write a test that make sure that fromIntegral conversions between these types don't regress? Is it possible to test the output of the simplifier or do we have to do something less direct, like writing a loop that does such conversions and make sure that the allocation stay low?
We have some tests looking at the -ddumjp-simpl output, e.g. integerConstantFolding. I'm not sure if we have one for type conversions OTTOMH. Thanks Ian
A related question I have is that I've some code that will map the singleton Nats to Ints, and last time I looked into this/ had a chat on the ghc-users list, it sounded like sometimes having Integer values constructed in between are unavoidable. Is that still the case with post 7.6.1 ghc? (And or, how might I be able to help?) On Friday, December 14, 2012, Ian Lynagh wrote:
On Fri, Dec 14, 2012 at 12:53:36PM -0800, Johan Tibell wrote:
I've been tracking down a few (unrelated) performance bugs related to conversions between primitive types. What these issues had in common is that some rule failed to fire and the conversion went via Integer, killing performance. How do we best write a test that make sure that fromIntegral conversions between these types don't regress? Is it possible to test the output of the simplifier or do we have to do something less direct, like writing a loop that does such conversions and make sure that the allocation stay low?
We have some tests looking at the -ddumjp-simpl output, e.g. integerConstantFolding. I'm not sure if we have one for type conversions OTTOMH.
Thanks Ian
_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org javascript:; http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
Hi Carter, On Fri, Dec 14, 2012 at 04:34:29PM -0500, Carter Schonwald wrote:
A related question I have is that I've some code that will map the singleton Nats to Ints, and last time I looked into this/ had a chat on the ghc-users list, it sounded like sometimes having Integer values constructed in between are unavoidable. Is that still the case with post 7.6.1 ghc? (And or, how might I be able to help?)
I don't know what a singleton Nat is, but if you mean Word then that's what Johan's been working on. Thanks Ian
On Fri, Dec 14, 2012 at 10:47 PM, Ian Lynagh wrote:
A related question I have is that I've some code that will map the singleton Nats to Ints, and last time I looked into this/ had a chat on
On Fri, Dec 14, 2012 at 04:34:29PM -0500, Carter Schonwald wrote: the
ghc-users list, it sounded like sometimes having Integer values constructed in between are unavoidable. Is that still the case with post 7.6.1 ghc? (And or, how might I be able to help?)
I don't know what a singleton Nat is, but if you mean Word then that's what Johan's been working on.
I'm guessing Carter means Sing Nat: http://www.haskell.org/ghc/docs/latest/html/libraries/base-4.6.0.0/src/GHC-T... Regards, Sean
Yes, thanks Sean for helping clarify!
yes, The TypeLits Nat
I had a discussion on the list with Iavor and others a few months back
http://www.haskell.org/pipermail/glasgow-haskell-users/2012-August/022789.ht...
and there were some examples and dialogue beyond that with Iavor and SPJ,
but it was off list.
point being: for Word or Int sized instances of Nat singleton, theres a
number of interesting ways having that information used to statically
specialize a numerical computation with some constants is handy, and last
time I asked it sounded like the story for this wasn't fully worked out.
I hope I'm explaining clearly, if not I'll be happy to try to elaborate
further
-Carter
On Fri, Dec 14, 2012 at 5:12 PM, Sean Leather
On Fri, Dec 14, 2012 at 10:47 PM, Ian Lynagh wrote:
On Fri, Dec 14, 2012 at 04:34:29PM -0500, Carter Schonwald wrote:
A related question I have is that I've some code that will map the singleton Nats to Ints, and last time I looked into this/ had a chat on the ghc-users list, it sounded like sometimes having Integer values constructed in between are unavoidable. Is that still the case with post 7.6.1 ghc? (And or, how might I be able to help?)
I don't know what a singleton Nat is, but if you mean Word then that's what Johan's been working on.
I'm guessing Carter means Sing Nat:
http://www.haskell.org/ghc/docs/latest/html/libraries/base-4.6.0.0/src/GHC-T...
Regards, Sean
_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
Best thing to do is to produce a concrete example, showing the code you think is sub-optimal Simon From: glasgow-haskell-users-bounces@haskell.org [mailto:glasgow-haskell-users-bounces@haskell.org] On Behalf Of Sean Leather Sent: 14 December 2012 22:13 To: GHC Users List; Ian Lynagh Subject: Re: How do we best make sure {Int,Word,Float,Double} to {Int,Word,Float,Double} conversions stay efficient On Fri, Dec 14, 2012 at 10:47 PM, Ian Lynagh wrote: On Fri, Dec 14, 2012 at 04:34:29PM -0500, Carter Schonwald wrote:
A related question I have is that I've some code that will map the singleton Nats to Ints, and last time I looked into this/ had a chat on the ghc-users list, it sounded like sometimes having Integer values constructed in between are unavoidable. Is that still the case with post 7.6.1 ghc? (And or, how might I be able to help?) I don't know what a singleton Nat is, but if you mean Word then that's what Johan's been working on.
I'm guessing Carter means Sing Nat: http://www.haskell.org/ghc/docs/latest/html/libraries/base-4.6.0.0/src/GHC-T... Regards, Sean
I believe the testsuite now has support for inspecting assembly
output. Not sure if that would be best for your use case, but
something to consider.
On 14 December 2012 12:53, Johan Tibell
Hi,
I've been tracking down a few (unrelated) performance bugs related to conversions between primitive types. What these issues had in common is that some rule failed to fire and the conversion went via Integer, killing performance. How do we best write a test that make sure that fromIntegral conversions between these types don't regress? Is it possible to test the output of the simplifier or do we have to do something less direct, like writing a loop that does such conversions and make sure that the allocation stay low?
-- Johan
_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
The sure-fire way is to make a loop that doesn't allocate if the rules fire; after all that's the ultimate goal. Then you can put it in tests/perf/should_run. doing -ddump-simpl and greping for stuff that should/should-not be there is another alternative we use in places. Simon | -----Original Message----- | From: glasgow-haskell-users-bounces@haskell.org [mailto:glasgow-haskell- | users-bounces@haskell.org] On Behalf Of Johan Tibell | Sent: 14 December 2012 20:54 | To: glasgow-haskell-users | Subject: How do we best make sure {Int,Word,Float,Double} to | {Int,Word,Float,Double} conversions stay efficient | | Hi, | | I've been tracking down a few (unrelated) performance bugs related to | conversions between primitive types. What these issues had in common is | that some rule failed to fire and the conversion went via Integer, | killing performance. How do we best write a test that make sure that | fromIntegral conversions between these types don't regress? Is it | possible to test the output of the simplifier or do we have to do | something less direct, like writing a loop that does such conversions | and make sure that the allocation stay low? | | -- Johan | | _______________________________________________ | Glasgow-haskell-users mailing list | Glasgow-haskell-users@haskell.org | http://www.haskell.org/mailman/listinfo/glasgow-haskell-users
participants (7)
-
Carter Schonwald -
David Terei -
Ian Lynagh -
Ian Lynagh -
Johan Tibell -
Sean Leather -
Simon Peyton-Jones