- jhc - Haskell.org

Frabjous Day: Jhc now has the beginnings of region support for stack allocations
by John Meacham 06 Sep '09

06 Sep '09

My most recent patch enables a feature that has been a long time coming, it performs an analysis that allows it to allocate and free memory on the stack, It is extremely basic as the moment as I want to test it in stages, some tests seem to not be affected much, others show some dramatic benefits, mandelbrot used 50% less memory and was 30% faster for instance. it is currently just a simple fixpoint calculation, though a type inference based version will perform better. John -- John Meacham - ⑆repetae.net⑆john⑈ - http://notanumber.net/

2 3

darcs patch: enable storage analysis to enable alloca... (and 4 more)
by John Meacham 06 Sep '09

06 Sep '09

Sat Sep 5 04:26:13 PDT 2009 John Meacham <john(a)repetae.net> * enable storage analysis to enable allocation on the stack Sat Sep 5 04:26:20 PDT 2009 John Meacham <john(a)repetae.net> * bump version number Sat Sep 5 15:47:14 PDT 2009 John Meacham <john(a)repetae.net> * allow deadcode and node analysis to see constant partial applications that are arguments to a store Sat Sep 5 16:11:29 PDT 2009 John Meacham <john(a)repetae.net> * add 'constraints' test from nobench to regression tests Sat Sep 5 17:13:29 PDT 2009 John Meacham <john(a)repetae.net> * have the ho cache names depend on the compiler version, to avoid cache poisoning by fixed bugs

1 0

darcs patch: try to abstract IO a bit (and 1 more)
by David Roundy 05 Sep '09

05 Sep '09

Hi John, Here's a fix to my IO abstraction patch. I'll copy the text from my previous submission below for your convenience. The previous patch was missing a change that I made to Prim.hs, not realizing that it was autogenerated from names.txt. So I'm also sending a second patch adding extra comments indicating that Prim.hs shouldn't be edited. I feel like this is a good general policy: to clearly mark autogenerated files, so naive users won't accidentally edit them and fail to notice that darcs doesn't record those changes. David Here is the previous explanation of why I'm interested in changing the type of IO: This is an attempt to move towards the point where I could try implementing concurrent IO. I am not sure whether this approach will work at all, so I'll appreciate your feedback. I'll probably send a separate email with more details on what I'm thinking. This patch is essentially a cleanup, which moves the newtype of IO into Jhc.IO and avoids exporting its constructor. The hope is that I can later change the type of IO, perhaps to something like newtype OldIO a = IO (World__ -> (# World__, a #)) newtype IO a = IO { runIO :: (a -> ThreadState -> OldIO ()) -- ^ a continuation -> ThreadState -- ^ the current state -> OldIO () } -- ^ the combined computation although I'd also want to figure out how to work exceptions into the picture. Anyhow, I think this patch is reasonable on its own. My greatest uncertainty at the moment due to my lack of understanding of how you deal with FFI calls. i.e. how is the code for a delcaration like foreign import ccall "stdio.h jhc_utf8_putc" c_putc :: Ptr CFile -> Int -> IO () generated? I imagine that some work will have to happen in the core compiler in order for this import to generate a call to Jhc.IO.io (which is what I'd like). Sat Sep 5 12:38:02 EDT 2009 David Roundy <droundy(a)darcs.net> * try to abstract IO a bit Sat Sep 5 12:38:28 EDT 2009 David Roundy <roundyd(a)physics.oregonstate.edu> * add warnings not to edit auto-generated Prim.hs file.

1 0

darcs patch: clean up fixIO
by David Roundy 05 Sep '09

05 Sep '09

This is just a formatting fix, along with removal of overlapping definitions. In its previous format, I found this function totally incomprehensible. David Sat Sep 5 09:04:34 EDT 2009 David Roundy <droundy(a)darcs.net> * clean up fixIO

1 0

darcs patch: fix instance of U2U primitive not being ... (and 5 more)
by John Meacham 05 Sep '09

05 Sep '09

Fri Sep 4 21:21:49 PDT 2009 John Meacham <john(a)repetae.net> * fix instance of U2U primitive not being processed properly Fri Sep 4 21:22:39 PDT 2009 John Meacham <john(a)repetae.net> * clean ups Fri Sep 4 23:16:45 PDT 2009 John Meacham <john(a)repetae.net> * fix strict newtypes bug Fri Sep 4 23:17:22 PDT 2009 John Meacham <john(a)repetae.net> * use strict writer monad as suggested by Taral Fri Sep 4 23:32:31 PDT 2009 John Meacham <john(a)repetae.net> * fix desugaring inside of list comprehensions Sat Sep 5 00:02:41 PDT 2009 John Meacham <john(a)repetae.net> * allow C generator to handle unknown values properly

1 0

darcs patch: add a lot of new regression tests, many... (and 4 more)
by John Meacham 05 Sep '09

05 Sep '09

Tue Sep 1 17:20:17 PDT 2009 John Meacham <john(a)repetae.net> * add a lot of new regression tests, many bugs that the lhc developers found, make regress.prl check the 'mustfail' status of tests Tue Sep 1 17:20:58 PDT 2009 John Meacham <john(a)repetae.net> * add mockbuild.sh to repo Fri Sep 4 18:06:50 PDT 2009 John Meacham <john(a)repetae.net> * bug updates Fri Sep 4 18:57:18 PDT 2009 John Meacham <john(a)repetae.net> * start adding storage analysis support Fri Sep 4 21:21:31 PDT 2009 John Meacham <john(a)repetae.net> * add regression test for strict newtype bug

1 0

bug report: "Alias still
by David Roundy 05 Sep '09

05 Sep '09

Hi all, I'm getting the following compile error: Converting to Grin... Found 2 CAFs to convert to constants, 0 of which are recursive. Recursive Constant theMain$10 v-1282411543 &(CControl.CoolConcurrent.Thread &(CInt# 0) &(P3_theMain$8)) theMain$11 v-1297615895 &(CControl.CoolConcurrent.TS 0 0 &(CControl.CoolConcurrent.Thread &(CInt# 0) &(P3_theMain$8)) &(CQ.Q &[] &[])) CAFS Alias still exists: ELit (Control.CoolConcurrent.TID (EVar (x51::ELit (Jhc.Prim.Int::ESort *)))::ELit (Jhc.Prim.Int::ESort *)) You can (hopefully?) reproduce this error by darcs getting my repository and compiling the demo: darcs get http://abridgegame.org/concurrency jhc demo.hs I believe this is vanilla Haskell code that ought to compile just fine. It does hide parts of the Prelude, but I don't see how that could be involved. For the curious, I started experimenting with a pure-Haskell implementation of concurrency. It looks feasible, although I haven't yet implemented any blocking IO using non-blocking primitives (which is often the whole point of concurrency). Of course, once the FFI is involved, we may want to involve the compiler, since any FFI call may block--and thus might be advisable to run in a separate thread. But that's not an issue for any of my applications. -- David Roundy

3 6

darcs patch: fix System.CPUTime.
by David Roundy 05 Sep '09

05 Sep '09

Hi John, This patch runs into a problem with cross-compiling that I don't know if you have a solution to. The (portable) clock function returns a clock_t, which may be different sizes on different systems. I used hsc2hs to get the size (since Foreign.C.Types is just a stub), but this will fail on cross-compilation. What is the recommended approach? Do we need two copies of the libraries, or is the data from targets.ini somehow available to us, or a way to generate clock_t directly in the C code? David Fri Sep 4 07:20:20 EDT 2009 David Roundy <roundyd(a)physics.oregonstate.edu> * fix System.CPUTime.

3 3

Initial profiling results
by Taral 05 Sep '09

05 Sep '09

Looks like we have two big leaked memory producers: freeMetaVarsEnv/tiNonRecImpl/tiImplGroups/tcBindGroup/tiProgram/tiModules/doModules’/typeCheckGraph/parseFiles/buildLibrary/main freeMetaVarsEnv/tiExpl/tcBindGroup/tiProgram/tiModules/doModules’/typeCheckGraph/parseFiles/buildLibrary/main Any ideas? I'll keep digging. -- Taral <taralx(a)gmail.com> "Please let me know if there's any further trouble I can give you." -- Unknown

2 3

Success with strictness
by Taral 05 Sep '09

05 Sep '09

--- old-jhc/src/FrontEnd/Tc/Monad.hs 2009-09-04 16:27:03.000000000 -0700 +++ new-jhc/src/FrontEnd/Tc/Monad.hs 2009-09-04 16:27:04.000000000 -0700 @@ -50,7 +50,7 @@ import Control.Monad.Error import Control.Monad.Reader -import Control.Monad.Writer +import Control.Monad.Writer.Strict import qualified Data.Traversable as T import qualified Data.Foldable as T import qualified Data.Sequence as Seq Allows compilation of containers with 750m max heap. Takes a couple tries, though, which implies we still may have some leaks. -- Taral <taralx(a)gmail.com> "Please let me know if there's any further trouble I can give you." -- Unknown

2 5