Matthijs Kooijman [Thu, 6 Aug 2009 13:56:44 +0000 (15:56 +0200)]
Add letremoveunused transformation.
This transformation removes all unused let bindings.
Matthijs Kooijman [Thu, 6 Aug 2009 13:56:23 +0000 (15:56 +0200)]
Add changeif normalization helper function.
Matthijs Kooijman [Thu, 6 Aug 2009 13:56:04 +0000 (15:56 +0200)]
Add expr_uses_binders predicate.
Matthijs Kooijman [Thu, 6 Aug 2009 13:55:43 +0000 (15:55 +0200)]
Add {is,has}StateType predicates.
Matthijs Kooijman [Thu, 6 Aug 2009 13:47:19 +0000 (15:47 +0200)]
Let mkTupleShow support empty tuples.
Matthijs Kooijman [Thu, 6 Aug 2009 10:06:32 +0000 (12:06 +0200)]
Remove the unused "stateful" argument fomr makeVHDL*.
State will be explicitely specified in the Haskell sources, so no reason
anymore to pass the statefulness of a function around.
Matthijs Kooijman [Wed, 5 Aug 2009 15:10:41 +0000 (17:10 +0200)]
Merge branch 'master' of git://github.com/christiaanb/clash into cλash
* 'master' of git://github.com/christiaanb/clash:
Fix builtin functions (!),take and RangedWord
Conflicts:
cλash/CLasH/VHDL/Generate.hs
Matthijs Kooijman [Wed, 5 Aug 2009 15:04:48 +0000 (17:04 +0200)]
Restructure the "finder" functions.
Previously, there were three different functions for the top entity,
initial state and test input. Now, there is just a single one, which
guarantees that these things are properly linked together. This should no
also support generating multiple entities at the same time (though there
is no top level interface for this yet).
This change also makes the testbench generation optional. A bunch of
functions were moved from Utils to GhcTools, to prevent a dependency loop.
Matthijs Kooijman [Wed, 5 Aug 2009 15:04:06 +0000 (17:04 +0200)]
Add catMaybesM and concatM helper functions.
Matthijs Kooijman [Wed, 5 Aug 2009 15:00:47 +0000 (17:00 +0200)]
Use the right id for the testbench architecture.
Christiaan Baaij [Wed, 5 Aug 2009 14:17:17 +0000 (16:17 +0200)]
Fix builtin functions (!),take and RangedWord
Matthijs Kooijman [Wed, 5 Aug 2009 13:21:45 +0000 (15:21 +0200)]
Fix testbench again.
The testbench code has also been moved to a separate module.
Matthijs Kooijman [Wed, 5 Aug 2009 13:20:14 +0000 (15:20 +0200)]
Split of part of getNormalized into normalizeExpr.
Matthijs Kooijman [Wed, 5 Aug 2009 12:51:53 +0000 (14:51 +0200)]
Add unzipM helper function.
Matthijs Kooijman [Wed, 5 Aug 2009 12:05:45 +0000 (14:05 +0200)]
Slightly simplify getEntity.
Matthijs Kooijman [Wed, 5 Aug 2009 11:59:39 +0000 (13:59 +0200)]
Move around some helper functions.
This moves some stuff from NormalizeTools and NormalizeTypes into the
(new) BinderTools and TranslatorTypes. This also puts most of these
function in the TranslatorSession instead of the TransformMonad.
Matthijs Kooijman [Wed, 5 Aug 2009 10:36:38 +0000 (12:36 +0200)]
Remove compatability aliases for the old sessions.
Matthijs Kooijman [Wed, 5 Aug 2009 10:12:28 +0000 (12:12 +0200)]
Make vhdl generation and normalization lazy.
Previously, first all function would be (recursively) normalized, and then
VHDL would be generated. Now, functions are normalized when needed, and
recursion is done while generating VHDL (so we know exactly which
components we are instantiating).
This disables the testbench and TFVec constructor for now, I'll fix that
in the next commits.
This also moves some code around, to prevent loops between Generate and
VHDL (again...).
The VHDLSession and NormalizeSession have been removed, and replaced with
the (previously unused) TranslatorSession. There are a few backward
compatibility aliases in place, so the next commit will probably remove
these and do a bunch of trivial replaces all over the code.
Matthijs Kooijman [Mon, 3 Aug 2009 10:26:43 +0000 (12:26 +0200)]
Put a comment on UniqSupply generation back.
Matthijs Kooijman [Mon, 3 Aug 2009 10:20:29 +0000 (12:20 +0200)]
Merge branch 'master' of git://github.com/christiaanb/clash into cλash
* 'master' of git://github.com/christiaanb/clash:
TFVec builtin should now completely work
Clean up imports in Generate.hs
Use createDirectoryIfMissing to create the vhdl directory, as it can create the parent directories too
Reflect API changes of clash in clash-nolibdir
Bring back listBind(ings) in Utils.hs by reorganising Translator.hs and GhcTools.hs
Fix a few comments so Haddock will complete
Further clean up Translator.hs (almost done now)
Cleanup Translator.hs
Fail again when we find a global function
Partially fixed TFVec builtin function. Still needs to be verified
Christiaan Baaij [Mon, 3 Aug 2009 07:51:55 +0000 (09:51 +0200)]
TFVec builtin should now completely work
Christiaan Baaij [Sun, 2 Aug 2009 19:17:21 +0000 (21:17 +0200)]
Clean up imports in Generate.hs
Christiaan Baaij [Sun, 2 Aug 2009 19:07:46 +0000 (21:07 +0200)]
Use createDirectoryIfMissing to create the vhdl directory, as it can create the parent directories too
Christiaan Baaij [Sun, 2 Aug 2009 18:51:29 +0000 (20:51 +0200)]
Reflect API changes of clash in clash-nolibdir
Christiaan Baaij [Sat, 1 Aug 2009 18:41:19 +0000 (20:41 +0200)]
Bring back listBind(ings) in Utils.hs by reorganising Translator.hs and GhcTools.hs
Christiaan Baaij [Fri, 31 Jul 2009 19:01:02 +0000 (21:01 +0200)]
Fix a few comments so Haddock will complete
Christiaan Baaij [Fri, 31 Jul 2009 18:46:17 +0000 (20:46 +0200)]
Further clean up Translator.hs (almost done now)
Christiaan Baaij [Fri, 31 Jul 2009 15:17:04 +0000 (17:17 +0200)]
Cleanup Translator.hs
Can be reduced in size even more by parameterizing the findcore functions
to work for both strings and annotations
Christiaan Baaij [Fri, 31 Jul 2009 15:15:54 +0000 (17:15 +0200)]
Fail again when we find a global function
Was only disabled to fix the builtin TFVec function
Christiaan Baaij [Fri, 31 Jul 2009 13:21:02 +0000 (15:21 +0200)]
Partially fixed TFVec builtin function. Still needs to be verified
Matthijs Kooijman [Fri, 31 Jul 2009 12:34:34 +0000 (14:34 +0200)]
Merge casevalsimpl and casewild into casesimpl.
Both transformations had a very similar structure, which are now merged.
Matthijs Kooijman [Fri, 31 Jul 2009 11:04:05 +0000 (13:04 +0200)]
Fix a small bug in the case wildening.
This bug could be triggered when the scrutinee was a datatype containing a
non-representable value. This is probably not translatable, but handle it
properly anyway.
Matthijs Kooijman [Fri, 31 Jul 2009 10:37:08 +0000 (12:37 +0200)]
Prevent conflicts with inlinenonrep in normalization.
This is a followup on
ebdc2024d7cf71: Let inlinefun inline everything non-representable.
which make inlinefun more general. A few other normalizations tried to
prevent loops with inline fun, but now work correctly for the (new)
inlinenonrep.
Matthijs Kooijman [Fri, 31 Jul 2009 09:26:59 +0000 (11:26 +0200)]
Merge branch 'master' of git://github.com/christiaanb/clash into cλash
* 'master' of git://github.com/christiaanb/clash:
Quick hack implementation of FSVec literals, needs to be fixed
We need the latest vhdl package
We now make a show function for all default datatypes.
Add automated testbench generation according to supplied test input
Conflicts:
cλash/CLasH/Translator.hs
Christiaan Baaij [Fri, 31 Jul 2009 07:19:57 +0000 (09:19 +0200)]
Quick hack implementation of FSVec literals, needs to be fixed
Christiaan Baaij [Wed, 29 Jul 2009 12:08:54 +0000 (14:08 +0200)]
We need the latest vhdl package
Christiaan Baaij [Wed, 29 Jul 2009 12:01:41 +0000 (14:01 +0200)]
We now make a show function for all default datatypes.
We need to figure out how to show custom datatypes in VHDL, for that
we will need a string representation for each constructor.
Christiaan Baaij [Tue, 28 Jul 2009 14:52:18 +0000 (16:52 +0200)]
Add automated testbench generation according to supplied test input
Will not compile in VHDL yet as we need to implement the VHDL show method first
Matthijs Kooijman [Tue, 28 Jul 2009 13:40:41 +0000 (15:40 +0200)]
Show type of binder in listBinding.
Matthijs Kooijman [Tue, 28 Jul 2009 13:38:30 +0000 (15:38 +0200)]
Make listBind use listBinding.
Matthijs Kooijman [Tue, 28 Jul 2009 13:26:58 +0000 (15:26 +0200)]
Extend CoreShow for TyCons.
This still does not completely show TyCons, since they are quite complex,
but at least distinguishes the various constructors and some useful info.
Christiaan Baaij [Thu, 23 Jul 2009 21:18:35 +0000 (23:18 +0200)]
Remove NoGenerics pragma, as the ghc bug 3391 is now fixed
Christiaan Baaij [Thu, 23 Jul 2009 19:14:19 +0000 (21:14 +0200)]
Add a clash wrapper that fills in the libdir from GHC.Paths
Christiaan Baaij [Thu, 23 Jul 2009 16:03:54 +0000 (18:03 +0200)]
Also add NoGenerics pragma to the other types.hs files
Christiaan Baaij [Thu, 23 Jul 2009 15:59:24 +0000 (17:59 +0200)]
Fix core show
Christiaan Baaij [Thu, 23 Jul 2009 15:22:47 +0000 (17:22 +0200)]
Currently Generics support is broken in combination with deriveAccessors
Christiaan Baaij [Wed, 22 Jul 2009 11:38:20 +0000 (13:38 +0200)]
Add licence to top directory
Christiaan Baaij [Wed, 22 Jul 2009 11:33:44 +0000 (13:33 +0200)]
Removed need for GHC.Paths, some functions however require a top libdir
Christiaan Baaij [Wed, 22 Jul 2009 07:37:59 +0000 (09:37 +0200)]
Update .cabal file so that a default build is expected to be in the GHC tree
Christiaan Baaij [Wed, 22 Jul 2009 07:31:30 +0000 (09:31 +0200)]
Make sure clash is build using stage2 ghc when build in ghc build tree
Christiaan Baaij [Wed, 22 Jul 2009 07:27:03 +0000 (09:27 +0200)]
Changed name of cabal file so it will work in the GHC build tree
Christiaan Baaij [Mon, 20 Jul 2009 12:01:08 +0000 (14:01 +0200)]
Changed build-depends to only those libraries required to build clash
tfvec, tfp and th-lift-ng are however required to translate any haskell
file to VHDL. They will be included in the clash package
Christiaan Baaij [Thu, 16 Jul 2009 13:25:10 +0000 (15:25 +0200)]
We now support Annotations to indicate top-level entity and initial state
Christiaan Baaij [Wed, 15 Jul 2009 18:14:06 +0000 (20:14 +0200)]
No need for Main.hs now that clash is a library
Christiaan Baaij [Wed, 15 Jul 2009 18:11:44 +0000 (20:11 +0200)]
Adepted the modules to their new structure
Christiaan Baaij [Wed, 15 Jul 2009 17:29:53 +0000 (19:29 +0200)]
Moved clash to it's own library directory, and started on library structure
Christiaan Baaij [Wed, 15 Jul 2009 15:41:13 +0000 (17:41 +0200)]
Moved to new GHC API (6.11). Also use vhdl package for the VHDL AST
Matthijs Kooijman [Tue, 14 Jul 2009 13:36:39 +0000 (15:36 +0200)]
Remove the (now unused) is_wild predicate.
Matthijs Kooijman [Tue, 14 Jul 2009 13:31:56 +0000 (15:31 +0200)]
Don't create selector cases for unused binders.
Previously, no selector cases were created for wild binders, using the
hacky is_wild predicate. Now, this checks the free variables of the case
value instead.
Matthijs Kooijman [Tue, 14 Jul 2009 11:00:13 +0000 (13:00 +0200)]
Merge git://github.com/darchon/clash into cλash
* git://github.com/darchon/clash:
Added resize function
Christiaan Baaij [Tue, 14 Jul 2009 08:28:03 +0000 (10:28 +0200)]
Added resize function
Matthijs Kooijman [Mon, 13 Jul 2009 13:50:23 +0000 (15:50 +0200)]
Remove some unused code.
Matthijs Kooijman [Mon, 13 Jul 2009 13:50:07 +0000 (15:50 +0200)]
Use noSrcSpan instead of a dummy SrcSpan.
Matthijs Kooijman [Mon, 13 Jul 2009 13:33:37 +0000 (15:33 +0200)]
Simplify eitherCoreOrExprArgs.
Christiaan Baaij [Mon, 13 Jul 2009 13:25:53 +0000 (15:25 +0200)]
Pulled 'varToVHDLExpr' into the TypeSession monad
Christiaan Baaij [Mon, 13 Jul 2009 10:04:22 +0000 (12:04 +0200)]
No longer any need to explicitly load module interface in 'toCore'
We have to HscEnv from the original HaskellToCore compilation step
that already includes all the module interfaces we need.
Christiaan Baaij [Mon, 13 Jul 2009 10:01:00 +0000 (12:01 +0200)]
Merge branch 'cλash' of git.stderr.nl/matthijs/projects/master-project
* 'cλash' of http://git.stderr.nl/matthijs/projects/master-project:
Generate proper VHDL for top level bindings with no arguments.
Use is_local_var for limiting appsimpl and letsimpl.
Add newline at the end of file.
Add is_local_var predicate.
Santize comment dashes position.
Normalize all used global binders.
Matthijs Kooijman [Mon, 13 Jul 2009 09:57:25 +0000 (11:57 +0200)]
Generate proper VHDL for top level bindings with no arguments.
Previously, a = b bindings would always generate an unconditional
assignment. Now, they use genApplication to generate VHDL, and
genApplications knows how to generate unconditional assignments when b is
a local identifier, and a component instantiation when b is a top level
binder.
Matthijs Kooijman [Mon, 13 Jul 2009 09:55:48 +0000 (11:55 +0200)]
Use is_local_var for limiting appsimpl and letsimpl.
Previously, appsimple and letsimpl would refrain from simplifying any Var
expression, since they are already simple. Now, they only refrain from
simplifying local variables, so references to top-level bindings will
still be simplified. This enables proper normalization of top level
bindings without arguments.
Matthijs Kooijman [Mon, 13 Jul 2009 09:54:58 +0000 (11:54 +0200)]
Add newline at the end of file.
Matthijs Kooijman [Mon, 13 Jul 2009 09:52:44 +0000 (11:52 +0200)]
Add is_local_var predicate.
Christiaan Baaij [Mon, 13 Jul 2009 09:22:43 +0000 (11:22 +0200)]
Only save normalized tfp integers in the tfp-cache
Christiaan Baaij [Mon, 13 Jul 2009 09:08:10 +0000 (11:08 +0200)]
Use HscEnv from initial HsToCore transformation when translating tfp to int
Matthijs Kooijman [Mon, 13 Jul 2009 07:59:58 +0000 (09:59 +0200)]
Santize comment dashes position.
This allows us to comment some lines in and out without having to remember
the exact position of the comment dashes (to prevent unneeded changes in
git).
Matthijs Kooijman [Mon, 13 Jul 2009 07:43:31 +0000 (09:43 +0200)]
Normalize all used global binders.
Previously, only function-typed binders were normalized (not necessarily
limited to global ones!).
Christiaan Baaij [Sat, 11 Jul 2009 20:23:59 +0000 (22:23 +0200)]
Great speed-up in type generation
Now just normalize tfp ints that are used as keys for the type map
Before these were translated to integers.
Save a map caches tfp ints, and their correspding integer literal
Christiaan Baaij [Thu, 9 Jul 2009 14:22:24 +0000 (16:22 +0200)]
Correctly handle negate for unsigned integers
Christiaan Baaij [Wed, 8 Jul 2009 14:44:01 +0000 (16:44 +0200)]
Partly fixed implementation for integer literals.
But it still needs alot of fixing
Christiaan Baaij [Wed, 8 Jul 2009 13:21:41 +0000 (15:21 +0200)]
VERY Ad-hoc support of literals.
Current implementation will probably fail in a lot of places
Christiaan Baaij [Wed, 8 Jul 2009 12:20:10 +0000 (14:20 +0200)]
Added support for SizedInts and cleaned up some function related to SizedWords
Christiaan Baaij [Tue, 7 Jul 2009 12:39:16 +0000 (14:39 +0200)]
Started adding numeric operations
Christiaan Baaij [Mon, 6 Jul 2009 14:57:41 +0000 (16:57 +0200)]
Caching converted tfp integers to speedup translation
Christiaan Baaij [Fri, 3 Jul 2009 19:49:06 +0000 (21:49 +0200)]
Updated cabal file to reflect change in TFVec, and the need for the filepath library
Christiaan Baaij [Fri, 3 Jul 2009 19:42:19 +0000 (21:42 +0200)]
Merge branch 'cλash' of git.stderr.nl/matthijs/projects/master-project
* 'cλash' of http://git.stderr.nl/matthijs/projects/master-project:
Let inlinefun inline everything non-representable.
Unify typeprop and funprop into argprop.
Extract only representable arguments.
Add predicates for testing representability of types.
Christiaan Baaij [Fri, 3 Jul 2009 19:41:04 +0000 (21:41 +0200)]
Made Bit an instance of Lift so it can be used in vectorTH
vectorTH works now, but you will need the latest version of TFVec
Matthijs Kooijman [Fri, 3 Jul 2009 17:32:10 +0000 (19:32 +0200)]
Let inlinefun inline everything non-representable.
This changes its name to inlinenonrep and makes it inline everything that
is not representable (which helps in particular for dictionary
construction expressions...).
I've not yet changed the conditions in various other transforms, so if
something silly would happen like having a dictionary as the value or
scrutinee of a case of let expression, things might go into an infinite
loop.
Also, since HTypes are generated very often now, things are really
starting to slow down...
Matthijs Kooijman [Fri, 3 Jul 2009 17:09:32 +0000 (19:09 +0200)]
Unify typeprop and funprop into argprop.
This also generalizes the conditions a bit by propagating anything that is
not representable.
Matthijs Kooijman [Fri, 3 Jul 2009 16:35:02 +0000 (18:35 +0200)]
Extract only representable arguments.
Previously, anything but types and functions was extracted. This
definition is slightly more general, but it should not matter for the
programs we work with.
Matthijs Kooijman [Fri, 3 Jul 2009 16:33:46 +0000 (18:33 +0200)]
Add predicates for testing representability of types.
These predicates try to build a VHDL type from a Core Type, to see if it
will be representable in hardware.
Matthijs Kooijman [Fri, 3 Jul 2009 11:40:43 +0000 (13:40 +0200)]
Let mkHType also return errors using Either.
Matthijs Kooijman [Fri, 3 Jul 2009 11:27:57 +0000 (13:27 +0200)]
Merge git://github.com/darchon/clash into cλash
* git://github.com/darchon/clash:
Keys for typemap can now deal with vector lengths based on type operators
Conflicts:
Generate.hs
VHDLTools.hs
Matthijs Kooijman [Fri, 3 Jul 2009 11:16:55 +0000 (13:16 +0200)]
Put a TypeState in TransformState.
This will allow transforms to test if a Type is representable in VHDL.
Matthijs Kooijman [Fri, 3 Jul 2009 11:10:05 +0000 (13:10 +0200)]
Split off the type related VHDLState variables.
These are now put in a substate of VHDLState, named TypeState. This allows
a TypeState to be put into TransformState later on as well.
Matthijs Kooijman [Fri, 3 Jul 2009 10:28:55 +0000 (12:28 +0200)]
Make vhdl_ty and friends return errors with Either.
Previously, error messages were passed forward. However, this does not
allow a type lookup to gracefully fail, which will be needed for
normalization soon. Now, error messages are returned using an Either
value, so they can be ignored at top level if needed.
Christiaan Baaij [Fri, 3 Jul 2009 09:48:02 +0000 (11:48 +0200)]
Keys for typemap can now deal with vector lengths based on type operators
Implementation is very slow, evalCoreExpr has to be run every time we
lookup and insert a new vector.
Matthijs Kooijman [Fri, 3 Jul 2009 09:33:03 +0000 (11:33 +0200)]
Stop extracting dataconstructor arguments.
Previously, dataconstructor arguments that had no arguments themselves
(eg. False, or High), would be extracted from an argument list, even
though they were just simple variable references. Since such a datacon can
be translated to a VHDL literal and VHDL supports mapping expressions (at
least constant expressions) to a port, there is no need for extraction
anymore.
Matthijs Kooijman [Thu, 2 Jul 2009 14:47:18 +0000 (16:47 +0200)]
Merge git://github.com/darchon/clash into cλash
* git://github.com/darchon/clash:
We now output VHDL types in the correct order
Removed the need for a special vector-type map.
Added builtin functions: concat, reverse, iterate, iteraten, generate and generaten
Christiaan Baaij [Thu, 2 Jul 2009 14:44:58 +0000 (16:44 +0200)]
We now output VHDL types in the correct order
Christiaan Baaij [Thu, 2 Jul 2009 13:19:07 +0000 (15:19 +0200)]
Removed the need for a special vector-type map.
Vector types are stripped of their length and then used as key.
Note that is of course an invalid haskell type, and is only meant
for internal use.
Matthijs Kooijman [Wed, 1 Jul 2009 15:00:49 +0000 (17:00 +0200)]
Add TODO regarding removal of is_wild.