mirror of
https://github.com/Mercury-Language/mercury.git
synced 2025-12-18 15:26:31 +00:00
Estimated hours taken: 220
Aditi update syntax, type and mode checking.
Change the hlds_goal for constructions in preparation for
structure reuse to avoid making multiple conflicting changes.
compiler/hlds_goal.m:
Merge `higher_order_call' and `class_method_call' into a single
`generic_call' goal type. This also has alternatives for the
various Aditi builtins for which type declarations can't
be written.
Remove the argument types field from higher-order/class method calls.
It wasn't used often, and wasn't updated by optimizations
such as inlining. The types can be obtained from the vartypes
field of the proc_info.
Add a `lambda_eval_method' field to lambda_goals.
Add a field to constructions to identify which RL code fragment should
be used for an top-down Aditi closure.
Add fields to constructions to hold structure reuse information.
This is currently ignored -- the changes to implement structure
reuse will be committed to the alias branch.
This is included here to avoid lots of CVS conflicts caused by
changing the definition of `hlds_goal' twice.
Add a field to `some' goals to specify whether the quantification
can be removed. This is used to make it easier to ensure that
indexes are used for updates.
Add a field to lambda_goals to describe whether the modes were
guessed by the compiler and may need fixing up after typechecking
works out the argument types.
Add predicate `hlds_goal__generic_call_id' to work out a call_id
for a generic call for use in error messages.
compiler/purity.m:
compiler/post_typecheck.m:
Fill in the modes of Aditi builtin calls and closure constructions.
This needs to know which are the `aditi__state' arguments, so
it must be done after typechecking.
compiler/prog_data.m:
Added `:- type sym_name_and_arity ---> sym_name/arity'.
Add a type `lambda_eval_method', which describes how a closure
is to be executed. The alternatives are normal Mercury execution,
bottom-up execution by Aditi and top-down execution by Aditi.
compiler/prog_out.m:
Add predicate `prog_out__write_sym_name_and_arity', which
replaces duplicated inline code in a few places.
compiler/hlds_data.m:
Add a `lambda_eval_method' field to `pred_const' cons_ids and
`pred_closure_tag' cons_tags.
compiler/hlds_pred.m:
Remove type `pred_call_id', replace it with type `simple_call_id',
which combines a `pred_or_func' and a `sym_name_and_arity'.
Add a type `call_id' which describes all the different types of call,
including normal calls, higher-order and class-method calls
and Aditi builtins.
Add `aditi_top_down' to the type `marker'.
Remove `aditi_interface' from type `marker'. Interfacing to
Aditi predicates is now handled by `generic_call' hlds_goals.
Add a type `rl_exprn_id' which identifies a predicate to
be executed top-down by Aditi.
Add a `maybe(rl_exprn_id)' field to type `proc_info'.
Add predicate `adjust_func_arity' to convert between the arity
of a function to its arity as a predicate.
Add predicates `get_state_args' and `get_state_args_det' to
extract the DCG state arguments from an argument list.
Add predicate `pred_info_get_call_id' to get a `simple_call_id'
for a predicate for use in error messages.
compiler/hlds_out.m:
Write the new representation for call_ids.
Add a predicate `hlds_out__write_call_arg_id' which
replaces similar code in mode_errors.m and typecheck.m.
compiler/prog_io_goal.m:
Add support for `aditi_bottom_up' and `aditi_top_down' annotations
on pred expressions.
compiler/prog_io_util.m:
compiler/prog_io_pragma.m:
Add predicates
- `prog_io_util:parse_name_and_arity' to parse `SymName/Arity'
(moved from prog_io_pragma.m).
- `prog_io_util:parse_pred_or_func_name_and_arity to parse
`pred SymName/Arity' or `func SymName/Arity'.
- `prog_io_util:parse_pred_or_func_and_args' to parse terms resembling
a clause head (moved from prog_io_pragma.m).
compiler/type_util.m:
Add support for `aditi_bottom_up' and `aditi_top_down' annotations
on higher-order types.
Add predicates `construct_higher_order_type',
`construct_higher_order_pred_type' and
`construct_higher_order_func_type' to avoid some code duplication.
compiler/mode_util.m:
Add predicate `unused_mode/1', which returns `builtin:unused'.
Add functions `aditi_di_mode/0', `aditi_ui_mode/0' and
`aditi_uo_mode/0' which return `in', `in', and `out', but will
be changed to return `di', `ui' and `uo' when alias tracking
is implemented.
compiler/goal_util.m:
Add predicate `goal_util__generic_call_vars' which returns
any arguments to a generic_call which are not in the argument list,
for example the closure passed to a higher-order call or
the typeclass_info for a class method call.
compiler/llds.m:
compiler/exprn_aux.m:
compiler/dupelim.m:
compiler/llds_out.m:
compiler/opt_debug.m:
Add builtin labels for the Aditi update operations.
compiler/hlds_module.m:
Add predicate predicate_table_search_pf_sym, used for finding
possible matches for a call with the wrong number of arguments.
compiler/intermod.m:
Don't write predicates which build `aditi_top_down' goals,
because there is currently no way to tell importing modules
which RL code fragment to use.
compiler/simplify.m:
Obey the `cannot_remove' field of explicit quantification goals.
compiler/make_hlds.m:
Parse Aditi updates.
Don't typecheck clauses for which syntax errors in Aditi updates
are found - this avoids spurious "undefined predicate `aditi_insert/3'"
errors.
Factor out some common code to handle terms of the form `Head :- Body'.
Factor out common code in the handling of pred and func expressions.
compiler/typecheck.m:
Typecheck Aditi builtins.
Allow the argument types of matching predicates to be adjusted
when typechecking the higher-order arguments of Aditi builtins.
Change `typecheck__resolve_pred_overloading' to take a list of
argument types rather than a `map(var, type)' and a list of
arguments to allow a transformation to be performed on the
argument types before passing them.
compiler/error_util.m:
Move the part of `report_error_num_args' which writes
"wrong number of arguments (<x>; expected <y>)" from
typecheck.m for use by make_hlds.m when reporting errors
for Aditi builtins.
compiler/modes.m:
compiler/unique_modes.m:
compiler/modecheck_call.m:
Modecheck Aditi builtins.
compiler/lambda.m:
Handle the markers for predicates introduced for
`aditi_top_down' and `aditi_bottom_up' lambda expressions.
compiler/polymorphism.m:
Add extra type_infos to `aditi_insert' calls
describing the tuple to insert.
compiler/call_gen.m:
Generate code for Aditi builtins.
compiler/unify_gen.m:
compiler/bytecode_gen.m:
Abort on `aditi_top_down' and `aditi_bottom_up' lambda
expressions - code generation for them is not yet implemented.
compiler/magic.m:
Use the `aditi_call' generic_call rather than create
a new procedure for each Aditi predicate called from C.
compiler/rl_out.pp:
compiler/rl_gen.m:
compiler/rl.m:
Move some utility code used by magic.m and call_gen.m into rl.m.
Remove an XXX comment about reference counting being not yet
implemented - Evan has fixed that.
library/ops.m:
compiler/mercury_to_mercury.m:
doc/transition_guide.texi:
Add unary prefix operators `aditi_bottom_up' and `aditi_top_down',
used as qualifiers on lambda expressions.
Add infix operator `==>' to separate the tuples in an
`aditi_modify' call.
compiler/follow_vars.m:
Thread a `map(prog_var, type)' through, needed because
type information is no longer held in higher-order call goals.
compiler/table_gen.m:
Use the `make_*_construction' predicates in hlds_goal.m
to construct constants.
compiler/*.m:
Trivial changes to add extra fields to hlds_goal structures.
doc/reference_manual.texi:
Document Aditi updates.
Use @samp{pragma base_relation} instead of
@samp{:- pragma base_relation} throughout the Aditi documentation
to be consistent with other parts of the reference manual.
tests/valid/Mmakefile:
tests/valid/aditi_update.m:
tests/valid/aditi.m:
Test case.
tests/valid/Mmakefile:
Remove some hard-coded --intermodule-optimization rules which are
no longer needed because `mmake depend' is now run in this directory.
tests/invalid/*.err_exp:
Fix expected output for changes in reporting of call_ids
in typecheck.m.
tests/invalid/Mmakefile
tests/invalid/aditi_update_errors.{m,err_exp}:
tests/invalid/aditi_update_mode_errors.{m,err_exp}:
Test error messages for Aditi updates.
tests/valid/aditi.m:
tests/invalid/aditi.m:
Cut down version of extras/aditi/aditi.m to provide basic declarations
for Aditi compilation such as `aditi__state' and the modes
`aditi_di', `aditi_uo' and `aditi_ui'. Installing extras/aditi/aditi.m
somewhere would remove the need for these.
599 lines
21 KiB
Mathematica
599 lines
21 KiB
Mathematica
%-----------------------------------------------------------------------------%
|
|
% Copyright (C) 1997-1999 The University of Melbourne.
|
|
% This file may only be copied under the terms of the GNU General
|
|
% Public License - see the file COPYING in the Mercury distribution.
|
|
%-----------------------------------------------------------------------------%
|
|
%
|
|
% term_util.m
|
|
% Main author: crs.
|
|
%
|
|
% This module:
|
|
%
|
|
% - defines the types used by termination analysis
|
|
% - defines predicates for computing functor norms
|
|
% - defines some utility predicates
|
|
%
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
:- module term_util.
|
|
|
|
:- interface.
|
|
|
|
:- import_module term_errors, prog_data.
|
|
:- import_module hlds_module, hlds_pred, hlds_data, hlds_goal.
|
|
|
|
:- import_module std_util, bool, int, list, map, bag.
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
%
|
|
% The types `arg_size_info' and `termination_info' hold information
|
|
% about procedures which is used for termination analysis.
|
|
% These types are stored as fields in the HLDS proc_info.
|
|
% For cross-module analysis, the information is written out as
|
|
% `pragma termination_info(...)' declarations in the
|
|
% `.opt' and `.trans_opt' files. The module prog_data.m defines
|
|
% types similar to these two (but without the `list(term_errors__error)')
|
|
% which are used when parsing `termination_info' pragmas.
|
|
%
|
|
|
|
% The arg size info defines an upper bound on the difference
|
|
% between the sizes of the output arguments of a procedure and the sizes
|
|
% of the input arguments:
|
|
%
|
|
% | input arguments | + constant >= | output arguments |
|
|
%
|
|
% where | | represents a semilinear norm.
|
|
|
|
:- type arg_size_info
|
|
---> finite(int, list(bool))
|
|
% The termination constant is a finite integer.
|
|
% The list of bool has a 1:1 correspondence
|
|
% with the input arguments of the procedure.
|
|
% It stores whether the argument contributes
|
|
% to the size of the output arguments.
|
|
|
|
; infinite(list(term_errors__error)).
|
|
% There is no finite integer for which the
|
|
% above equation is true. The argument says
|
|
% why the analysis failed to find a finite
|
|
% constant.
|
|
|
|
:- type termination_info
|
|
---> cannot_loop % This procedure terminates for all
|
|
% possible inputs.
|
|
|
|
; can_loop(list(term_errors__error)).
|
|
% The analysis could not prove that the
|
|
% procedure terminates.
|
|
|
|
% The type `used_args' holds a mapping which specifies for each procedure
|
|
% which of its arguments are used.
|
|
|
|
:- type used_args == map(pred_proc_id, list(bool)).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
% We use semilinear norms (denoted by ||) to compute the sizes of terms.
|
|
% These have the form
|
|
%
|
|
% | f(t1, ... tn) | = weight(f) + sum of | ti |
|
|
% where i is an element of a set I, and I is a subset of {1, ... n}
|
|
%
|
|
% We currently support four kinds of semilinear norms.
|
|
|
|
:- type functor_info
|
|
---> simple % All non-constant functors have weight 1,
|
|
% while constants have weight 0.
|
|
% Use the size of all subterms (I = {1, ..., n}.
|
|
|
|
; total % All functors have weight = arity of the functor.
|
|
% Use the size of all subterms (I = {1, ..., n}.
|
|
|
|
; use_map(weight_table)
|
|
% The weight of each functor is given by the table.
|
|
% Use the size of all subterms (I = {1, ..., n}.
|
|
|
|
; use_map_and_args(weight_table).
|
|
% The weight of each functor is given by the table,
|
|
% and so is the set of arguments of the functor whose
|
|
% size should be counted (I is given by the table
|
|
% entry of the functor).
|
|
|
|
:- type unify_info == pair(map(prog_var, type), functor_info).
|
|
|
|
:- type weight_info ---> weight(int, list(bool)).
|
|
:- type weight_table == map(pair(type_id, cons_id), weight_info).
|
|
|
|
:- pred find_weights(module_info::in, weight_table::out) is det.
|
|
|
|
% This predicate is computes the weight of a functor and the set of arguments
|
|
% of that functor whose sizes should be counted towards the size of the whole
|
|
% term.
|
|
|
|
:- pred functor_norm(functor_info::in, type_id::in, cons_id::in,
|
|
module_info::in, int::out, list(prog_var)::in, list(prog_var)::out,
|
|
list(uni_mode)::in, list(uni_mode)::out) is det.
|
|
|
|
:- type pass_info
|
|
---> pass_info(
|
|
functor_info,
|
|
int, % Max number of errors to gather.
|
|
int % Max number of paths to analyze.
|
|
).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
% This predicate partitions the arguments of a call into a list of input
|
|
% variables and a list of output variables,
|
|
|
|
:- pred partition_call_args(module_info::in, list(mode)::in, list(prog_var)::in,
|
|
bag(prog_var)::out, bag(prog_var)::out) is det.
|
|
|
|
% Given a list of variables from a unification, this predicate divides the
|
|
% list into a bag of input variables, and a bag of output variables.
|
|
|
|
:- pred split_unification_vars(list(prog_var)::in, list(uni_mode)::in,
|
|
module_info::in, bag(prog_var)::out, bag(prog_var)::out) is det.
|
|
|
|
% Used to create lists of boolean values, which are used for used_args.
|
|
% make_bool_list(HeadVars, BoolIn, BoolOut) creates a bool list which is
|
|
% (length(HeadVars) - length(BoolIn)) `no' followed by BoolIn. This is
|
|
% used to set the used args for compiler generated predicates. The no's
|
|
% at the start are because the Type infos are not used. length(BoolIn)
|
|
% should equal the arity of the predicate, and the difference in length
|
|
% between the arity of the procedure and the arity of the predicate is
|
|
% the number of type infos.
|
|
|
|
:- pred term_util__make_bool_list(list(_T)::in, list(bool)::in,
|
|
list(bool)::out) is det.
|
|
|
|
% Removes variables from the InVarBag that are not used in the call.
|
|
% remove_unused_args(InVarBag0, VarList, BoolList, InVarBag)
|
|
% VarList and BoolList are corresponding lists. Any variable in VarList
|
|
% that has a `no' in the corresponding place in the BoolList is removed
|
|
% from InVarBag.
|
|
|
|
:- pred remove_unused_args(bag(prog_var), list(prog_var), list(bool),
|
|
bag(prog_var)).
|
|
:- mode remove_unused_args(in, in, in, out) is det.
|
|
|
|
% This predicate sets the argument size info of a given a list of procedures.
|
|
|
|
:- pred set_pred_proc_ids_arg_size_info(list(pred_proc_id)::in,
|
|
arg_size_info::in, module_info::in, module_info::out) is det.
|
|
|
|
% This predicate sets the termination info of a given a list of procedures.
|
|
|
|
:- pred set_pred_proc_ids_termination_info(list(pred_proc_id)::in,
|
|
termination_info::in, module_info::in, module_info::out) is det.
|
|
|
|
:- pred lookup_proc_termination_info(module_info::in, pred_proc_id::in,
|
|
maybe(termination_info)::out) is det.
|
|
|
|
:- pred lookup_proc_arg_size_info(module_info::in, pred_proc_id::in,
|
|
maybe(arg_size_info)::out) is det.
|
|
|
|
% Succeeds if one or more variables in the list are higher order.
|
|
|
|
:- pred horder_vars(list(prog_var), map(prog_var, type)).
|
|
:- mode horder_vars(in, in) is semidet.
|
|
|
|
% Succeeds if all values of the given type are zero size (for all norms).
|
|
|
|
:- pred zero_size_type(type, module_info).
|
|
:- mode zero_size_type(in, in) is semidet.
|
|
|
|
:- pred get_context_from_scc(list(pred_proc_id)::in, module_info::in,
|
|
prog_context::out) is det.
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
% Convert a prog_data__pragma_termination_info into a
|
|
% term_util__termination_info, by adding the appropriate context.
|
|
|
|
:- pred add_context_to_termination_info(maybe(pragma_termination_info),
|
|
prog_context, maybe(termination_info)).
|
|
:- mode add_context_to_termination_info(in, in, out) is det.
|
|
|
|
% Convert a prog_data__pragma_arg_size_info into a
|
|
% term_util__arg_size_info, by adding the appropriate context.
|
|
|
|
:- pred add_context_to_arg_size_info(maybe(pragma_arg_size_info),
|
|
prog_context, maybe(arg_size_info)).
|
|
:- mode add_context_to_arg_size_info(in, in, out) is det.
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
:- implementation.
|
|
|
|
:- import_module inst_match, prog_out, mode_util, type_util.
|
|
:- import_module globals, options.
|
|
|
|
:- import_module assoc_list, require.
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
% Calculate the weight to be assigned to each function symbol for the
|
|
% use_map and use_map_and_args semilinear norms.
|
|
%
|
|
% Given a type definition such as
|
|
%
|
|
% :- type t(Tk) ---> f1(a11, ... a1n1) where n1 is the arity of f1
|
|
% ; ...
|
|
% ; fm(am1, ... amnm) where nm is the arity of fm
|
|
%
|
|
% we check, for each aij, whether its type is recursive (i.e. it is t with
|
|
% type variable arguments that are a permutation of Tk). The weight info
|
|
% we compute for each functor will have a boolean list that has a `yes'
|
|
% for each recursive argument and a `no' for each nonrecursive argument.
|
|
% The weight to be assigned to the functor is the number of nonrecursive
|
|
% arguments, except that we assign a weight of at least 1 to all functors
|
|
% which are not constants.
|
|
|
|
find_weights(ModuleInfo, Weights) :-
|
|
module_info_types(ModuleInfo, TypeTable),
|
|
map__to_assoc_list(TypeTable, TypeList),
|
|
map__init(Weights0),
|
|
find_weights_for_type_list(TypeList, Weights0, Weights).
|
|
|
|
:- pred find_weights_for_type_list(assoc_list(type_id, hlds_type_defn)::in,
|
|
weight_table::in, weight_table::out) is det.
|
|
|
|
find_weights_for_type_list([], Weights, Weights).
|
|
find_weights_for_type_list([TypeId - TypeDefn | TypeList], Weights0, Weights) :-
|
|
find_weights_for_type(TypeId, TypeDefn, Weights0, Weights1),
|
|
find_weights_for_type_list(TypeList, Weights1, Weights).
|
|
|
|
:- pred find_weights_for_type(type_id::in, hlds_type_defn::in,
|
|
weight_table::in, weight_table::out) is det.
|
|
|
|
find_weights_for_type(TypeId, TypeDefn, Weights0, Weights) :-
|
|
hlds_data__get_type_defn_body(TypeDefn, TypeBody),
|
|
(
|
|
TypeBody = du_type(Constructors, _, _, _),
|
|
hlds_data__get_type_defn_tparams(TypeDefn, TypeParams),
|
|
find_weights_for_cons_list(Constructors, TypeId, TypeParams,
|
|
Weights0, Weights)
|
|
;
|
|
TypeBody = uu_type(_),
|
|
error("undiscriminated union types not yet implemented")
|
|
;
|
|
% This type does not introduce any functors
|
|
TypeBody = eqv_type(_),
|
|
Weights = Weights0
|
|
;
|
|
% This type may introduce some functors,
|
|
% but we will never see them in this analysis
|
|
TypeBody = abstract_type,
|
|
Weights = Weights0
|
|
).
|
|
|
|
:- pred find_weights_for_cons_list(list(constructor)::in,
|
|
type_id::in, list(type_param)::in,
|
|
weight_table::in, weight_table::out) is det.
|
|
|
|
find_weights_for_cons_list([], _, _, Weights, Weights).
|
|
find_weights_for_cons_list([Constructor | Constructors], TypeId, Params,
|
|
Weights0, Weights) :-
|
|
find_weights_for_cons(Constructor, TypeId, Params, Weights0, Weights1),
|
|
find_weights_for_cons_list(Constructors, TypeId, Params,
|
|
Weights1, Weights).
|
|
|
|
:- pred find_weights_for_cons(constructor::in,
|
|
type_id::in, list(type_param)::in,
|
|
weight_table::in, weight_table::out) is det.
|
|
|
|
find_weights_for_cons(Ctor, TypeId, Params, Weights0, Weights) :-
|
|
% XXX should we do something about ExistQVars here?
|
|
Ctor = ctor(_ExistQVars, _Constraints, SymName, Args),
|
|
list__length(Args, Arity),
|
|
( Arity > 0 ->
|
|
find_and_count_nonrec_args(Args, TypeId, Params,
|
|
NumNonRec, ArgInfos0),
|
|
( NumNonRec = 0 ->
|
|
Weight = 1,
|
|
list__duplicate(Arity, yes, ArgInfos)
|
|
;
|
|
Weight = NumNonRec,
|
|
ArgInfos = ArgInfos0
|
|
),
|
|
WeightInfo = weight(Weight, ArgInfos)
|
|
;
|
|
WeightInfo = weight(0, [])
|
|
),
|
|
ConsId = cons(SymName, Arity),
|
|
map__det_insert(Weights0, TypeId - ConsId, WeightInfo, Weights).
|
|
|
|
:- pred find_and_count_nonrec_args(list(constructor_arg)::in,
|
|
type_id::in, list(type_param)::in,
|
|
int::out, list(bool)::out) is det.
|
|
|
|
find_and_count_nonrec_args([], _, _, 0, []).
|
|
find_and_count_nonrec_args([Arg | Args], Id, Params, NonRecArgs, ArgInfo) :-
|
|
find_and_count_nonrec_args(Args, Id, Params, NonRecArgs0, ArgInfo0),
|
|
( is_arg_recursive(Arg, Id, Params) ->
|
|
NonRecArgs = NonRecArgs0,
|
|
ArgInfo = [yes | ArgInfo0]
|
|
;
|
|
NonRecArgs is NonRecArgs0 + 1,
|
|
ArgInfo = [no | ArgInfo0]
|
|
).
|
|
|
|
:- pred is_arg_recursive(constructor_arg::in,
|
|
type_id::in, list(type_param)::in) is semidet.
|
|
|
|
is_arg_recursive(Arg, Id, Params) :-
|
|
Arg = _Name - ArgType,
|
|
type_to_type_id(ArgType, ArgTypeId, ArgTypeParams),
|
|
Id = ArgTypeId,
|
|
list__perm(Params, ArgTypeParams).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
% Although the module info is not used in either of these norms, it could
|
|
% be needed for other norms, so it should not be removed.
|
|
|
|
functor_norm(simple, _, ConsId, _, Int, Args, Args, Modes, Modes) :-
|
|
(
|
|
ConsId = cons(_, Arity),
|
|
Arity \= 0
|
|
->
|
|
Int = 1
|
|
;
|
|
Int = 0
|
|
).
|
|
functor_norm(total, _, ConsId, _Module, Int, Args, Args, Modes, Modes) :-
|
|
( ConsId = cons(_, Arity) ->
|
|
Int = Arity
|
|
;
|
|
Int = 0
|
|
).
|
|
functor_norm(use_map(WeightMap), TypeId, ConsId, _Module, Int,
|
|
Args, Args, Modes, Modes) :-
|
|
( map__search(WeightMap, TypeId - ConsId, WeightInfo) ->
|
|
WeightInfo = weight(Int, _)
|
|
;
|
|
Int = 0
|
|
).
|
|
functor_norm(use_map_and_args(WeightMap), TypeId, ConsId, _Module, Int,
|
|
Args0, Args, Modes0, Modes) :-
|
|
( map__search(WeightMap, TypeId - ConsId, WeightInfo) ->
|
|
WeightInfo = weight(Int, UseArgList),
|
|
(
|
|
functor_norm_filter_args(UseArgList, Args0, Args1,
|
|
Modes0, Modes1)
|
|
->
|
|
Modes = Modes1,
|
|
Args = Args1
|
|
;
|
|
error("Unmatched lists in functor_norm_filter_args.")
|
|
)
|
|
;
|
|
Int = 0,
|
|
Modes = Modes0,
|
|
Args = Args0
|
|
).
|
|
|
|
% This predicate will fail if the length of the input lists are not matched.
|
|
|
|
:- pred functor_norm_filter_args(list(bool), list(prog_var), list(prog_var),
|
|
list(uni_mode), list(uni_mode)).
|
|
:- mode functor_norm_filter_args(in, in, out, in, out) is semidet.
|
|
|
|
functor_norm_filter_args([], [], [], [], []).
|
|
functor_norm_filter_args([yes | Bools], [Arg0 | Args0], [Arg0 | Args],
|
|
[Mode0 | Modes0], [Mode0 | Modes]) :-
|
|
functor_norm_filter_args(Bools, Args0, Args, Modes0, Modes).
|
|
functor_norm_filter_args([no | Bools], [_Arg0 | Args0], Args,
|
|
[_Mode0 | Modes0], Modes) :-
|
|
functor_norm_filter_args(Bools, Args0, Args, Modes0, Modes).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
partition_call_args(Module, ArgModes, Args, InVarsBag, OutVarsBag) :-
|
|
partition_call_args_2(Module, ArgModes, Args, InVars, OutVars),
|
|
bag__from_list(InVars, InVarsBag),
|
|
bag__from_list(OutVars, OutVarsBag).
|
|
|
|
:- pred partition_call_args_2(module_info::in, list(mode)::in,
|
|
list(prog_var)::in, list(prog_var)::out, list(prog_var)::out) is det.
|
|
|
|
partition_call_args_2(_, [], [], [], []).
|
|
partition_call_args_2(_, [], [_ | _], _, _) :-
|
|
error("Unmatched variables in term_util:partition_call_args").
|
|
partition_call_args_2(_, [_ | _], [], _, _) :-
|
|
error("Unmatched variables in term_util__partition_call_args").
|
|
partition_call_args_2(ModuleInfo, [ArgMode | ArgModes], [Arg | Args],
|
|
InputArgs, OutputArgs) :-
|
|
partition_call_args_2(ModuleInfo, ArgModes, Args,
|
|
InputArgs1, OutputArgs1),
|
|
( mode_is_input(ModuleInfo, ArgMode) ->
|
|
InputArgs = [Arg | InputArgs1],
|
|
OutputArgs = OutputArgs1
|
|
; mode_is_output(ModuleInfo, ArgMode) ->
|
|
InputArgs = InputArgs1,
|
|
OutputArgs = [Arg | OutputArgs1]
|
|
;
|
|
InputArgs = InputArgs1,
|
|
OutputArgs = OutputArgs1
|
|
).
|
|
|
|
% For these next two predicates (split_unification_vars and
|
|
% partition_call_args) there is a problem of what needs to be done for
|
|
% partially instantiated data structures. The correct answer is that the
|
|
% system shoud use a norm such that the size of the uninstantiated parts of
|
|
% a partially instantiated structure have no effect on the size of the data
|
|
% structure according to the norm. For example when finding the size of a
|
|
% list-skeleton, list-length norm should be used. Therefore, the size of
|
|
% any term must be given by
|
|
% sizeof(term) = constant + sum of the size of each
|
|
% (possibly partly) instantiated subterm.
|
|
% It is probably easiest to implement this by modifying term_weights.
|
|
% The current implementation does not correctly handle partially
|
|
% instantiated data structures.
|
|
|
|
split_unification_vars([], Modes, _ModuleInfo, Vars, Vars) :-
|
|
bag__init(Vars),
|
|
( Modes = [] ->
|
|
true
|
|
;
|
|
error("term_util:split_unification_vars: Unmatched Variables")
|
|
).
|
|
split_unification_vars([Arg | Args], Modes, ModuleInfo,
|
|
InVars, OutVars):-
|
|
( Modes = [UniMode | UniModes] ->
|
|
split_unification_vars(Args, UniModes, ModuleInfo,
|
|
InVars0, OutVars0),
|
|
UniMode = ((_VarInit - ArgInit) -> (_VarFinal - ArgFinal)),
|
|
( % if
|
|
inst_is_bound(ModuleInfo, ArgInit)
|
|
->
|
|
% Variable is an input variable
|
|
bag__insert(InVars0, Arg, InVars),
|
|
OutVars = OutVars0
|
|
; % else if
|
|
inst_is_free(ModuleInfo, ArgInit),
|
|
inst_is_bound(ModuleInfo, ArgFinal)
|
|
->
|
|
% Variable is an output variable
|
|
InVars = InVars0,
|
|
bag__insert(OutVars0, Arg, OutVars)
|
|
; % else
|
|
InVars = InVars0,
|
|
OutVars = OutVars0
|
|
)
|
|
;
|
|
error("term_util__split_unification_vars: Unmatched Variables")
|
|
).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
term_util__make_bool_list(HeadVars0, Bools, Out) :-
|
|
list__length(Bools, Arity),
|
|
( list__drop(Arity, HeadVars0, HeadVars1) ->
|
|
HeadVars = HeadVars1
|
|
;
|
|
error("Unmatched variables in term_util:make_bool_list")
|
|
),
|
|
term_util__make_bool_list_2(HeadVars, Bools, Out).
|
|
|
|
:- pred term_util__make_bool_list_2(list(_T), list(bool), list(bool)).
|
|
:- mode term_util__make_bool_list_2(in, in, out) is det.
|
|
|
|
term_util__make_bool_list_2([], Bools, Bools).
|
|
term_util__make_bool_list_2([ _ | Vars ], Bools, [no | Out]) :-
|
|
term_util__make_bool_list_2(Vars, Bools, Out).
|
|
|
|
remove_unused_args(Vars, [], [], Vars).
|
|
remove_unused_args(Vars, [], [_X | _Xs], Vars) :-
|
|
error("Unmatched variables in term_util:remove_unused_args").
|
|
remove_unused_args(Vars, [_X | _Xs], [], Vars) :-
|
|
error("Unmatched variables in term_util__remove_unused_args").
|
|
remove_unused_args(Vars0, [ Arg | Args ], [ UsedVar | UsedVars ], Vars) :-
|
|
( UsedVar = yes ->
|
|
% The variable is used, so leave it
|
|
remove_unused_args(Vars0, Args, UsedVars, Vars)
|
|
;
|
|
% The variable is not used in producing output vars, so
|
|
% dont include it as an input variable.
|
|
bag__delete(Vars0, Arg, Vars1),
|
|
remove_unused_args(Vars1, Args, UsedVars, Vars)
|
|
).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
set_pred_proc_ids_arg_size_info([], _ArgSize, Module, Module).
|
|
set_pred_proc_ids_arg_size_info([PPId | PPIds], ArgSize, Module0, Module) :-
|
|
PPId = proc(PredId, ProcId),
|
|
module_info_preds(Module0, PredTable0),
|
|
map__lookup(PredTable0, PredId, PredInfo0),
|
|
pred_info_procedures(PredInfo0, ProcTable0),
|
|
map__lookup(ProcTable0, ProcId, ProcInfo0),
|
|
|
|
proc_info_set_maybe_arg_size_info(ProcInfo0, yes(ArgSize), ProcInfo),
|
|
|
|
map__det_update(ProcTable0, ProcId, ProcInfo, ProcTable),
|
|
pred_info_set_procedures(PredInfo0, ProcTable, PredInfo),
|
|
map__det_update(PredTable0, PredId, PredInfo, PredTable),
|
|
module_info_set_preds(Module0, PredTable, Module1),
|
|
set_pred_proc_ids_arg_size_info(PPIds, ArgSize, Module1, Module).
|
|
|
|
set_pred_proc_ids_termination_info([], _Termination, Module, Module).
|
|
set_pred_proc_ids_termination_info([PPId | PPIds], Termination,
|
|
Module0, Module) :-
|
|
PPId = proc(PredId, ProcId),
|
|
module_info_preds(Module0, PredTable0),
|
|
map__lookup(PredTable0, PredId, PredInfo0),
|
|
pred_info_procedures(PredInfo0, ProcTable0),
|
|
map__lookup(ProcTable0, ProcId, ProcInfo0),
|
|
|
|
proc_info_set_maybe_termination_info(ProcInfo0, yes(Termination),
|
|
ProcInfo),
|
|
|
|
map__det_update(ProcTable0, ProcId, ProcInfo, ProcTable),
|
|
pred_info_set_procedures(PredInfo0, ProcTable, PredInfo),
|
|
map__det_update(PredTable0, PredId, PredInfo, PredTable),
|
|
module_info_set_preds(Module0, PredTable, Module1),
|
|
set_pred_proc_ids_termination_info(PPIds, Termination,
|
|
Module1, Module).
|
|
|
|
lookup_proc_termination_info(Module, PredProcId, MaybeTermination) :-
|
|
PredProcId = proc(PredId, ProcId),
|
|
module_info_pred_proc_info(Module, PredId, ProcId, _, ProcInfo),
|
|
proc_info_get_maybe_termination_info(ProcInfo, MaybeTermination).
|
|
|
|
lookup_proc_arg_size_info(Module, PredProcId, MaybeArgSize) :-
|
|
PredProcId = proc(PredId, ProcId),
|
|
module_info_pred_proc_info(Module, PredId, ProcId, _, ProcInfo),
|
|
proc_info_get_maybe_arg_size_info(ProcInfo, MaybeArgSize).
|
|
|
|
horder_vars([Arg | Args], VarType) :-
|
|
(
|
|
map__lookup(VarType, Arg, Type),
|
|
type_is_higher_order(Type, _, _, _)
|
|
;
|
|
horder_vars(Args, VarType)
|
|
).
|
|
|
|
zero_size_type(Type, Module) :-
|
|
classify_type(Type, Module, TypeCategory),
|
|
zero_size_type_category(TypeCategory, Type, Module, yes).
|
|
|
|
:- pred zero_size_type_category(builtin_type, type, module_info, bool).
|
|
:- mode zero_size_type_category(in, in, in, out) is det.
|
|
|
|
zero_size_type_category(int_type, _, _, yes).
|
|
zero_size_type_category(char_type, _, _, yes).
|
|
zero_size_type_category(str_type, _, _, yes).
|
|
zero_size_type_category(float_type, _, _, yes).
|
|
zero_size_type_category(pred_type, _, _, no).
|
|
zero_size_type_category(enum_type, _, _, yes).
|
|
zero_size_type_category(polymorphic_type, _, _, no).
|
|
zero_size_type_category(user_type, _, _, no).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
get_context_from_scc(SCC, Module, Context) :-
|
|
( SCC = [proc(PredId, _) | _] ->
|
|
module_info_pred_info(Module, PredId, PredInfo),
|
|
pred_info_context(PredInfo, Context)
|
|
;
|
|
error("Empty SCC in pass 2 of termination analysis")
|
|
).
|
|
|
|
%-----------------------------------------------------------------------------%
|
|
|
|
add_context_to_termination_info(no, _, no).
|
|
add_context_to_termination_info(yes(cannot_loop), _, yes(cannot_loop)).
|
|
add_context_to_termination_info(yes(can_loop), Context,
|
|
yes(can_loop([Context - imported_pred]))).
|
|
|
|
add_context_to_arg_size_info(no, _, no).
|
|
add_context_to_arg_size_info(yes(finite(A, B)), _, yes(finite(A, B))).
|
|
add_context_to_arg_size_info(yes(infinite), Context,
|
|
yes(infinite([Context - imported_pred]))).
|
|
|
|
%-----------------------------------------------------------------------------%
|