package ocaml-base-compiler

  1. Overview
  2. Docs
Package contains no libraries

Intermediate language used for tree-based analysis and optimization.

type call_kind =
  1. | Indirect
  2. | Direct of Closure_id.t

Whether the callee in a function application is known at compile time.

type const =
  1. | Int of int
  2. | Char of char
    (*

    Char is kept separate from Int to improve printing

    *)
  3. | Const_pointer of int
    (*

    Const_pointer is an immediate value of a type whose values may be boxed (typically a variant type with both constant and non-constant constructors).

    *)

Simple constants. ("Structured constants" are rewritten to invocations of Pmakeblock so that they easily take part in optimizations.)

type apply = {
  1. func : Variable.t;
  2. args : Variable.t list;
  3. kind : call_kind;
  4. dbg : Debuginfo.t;
  5. inline : Lambda.inline_attribute;
    (*

    Instructions from the source code as to whether the callee should be inlined.

    *)
  6. specialise : Lambda.specialise_attribute;
    (*

    Instructions from the source code as to whether the callee should be specialised.

    *)
}

The application of a function to a list of arguments.

type assign = {
  1. being_assigned : Mutable_variable.t;
  2. new_value : Variable.t;
}

The update of a mutable variable. Mutable variables are distinct from immutable variables in Flambda.

type send = {
  1. kind : Lambda.meth_kind;
  2. meth : Variable.t;
  3. obj : Variable.t;
  4. args : Variable.t list;
  5. dbg : Debuginfo.t;
}

The invocation of a method.

type project_closure = Projection.project_closure

For details on these types, see projection.mli.

type move_within_set_of_closures = Projection.move_within_set_of_closures
type project_var = Projection.project_var
type specialised_to = {
  1. var : Variable.t;
    (*

    The "outer variable".

    *)
  2. projection : Projection.t option;
    (*

    The projecting_from value (see projection.mli) of any projection must be another free variable or specialised argument (depending on whether this record type is involved in free_vars or specialised_args respectively) in the same set of closures. As such, this field describes a relation of projections between either the free_vars or the specialised_args.

    *)
}

See free_vars and specialised_args, below.

type t =
  1. | Var of Variable.t
  2. | Let of let_expr
  3. | Let_mutable of let_mutable
  4. | Let_rec of (Variable.t * named) list * t
    (*

    CR-someday lwhite: give Let_rec the same fields as Let.

    *)
  5. | Apply of apply
  6. | Send of send
  7. | Assign of assign
  8. | If_then_else of Variable.t * t * t
  9. | Switch of Variable.t * switch
  10. | String_switch of Variable.t * (string * t) list * t option
    (*

    Restrictions on Lambda.Lstringswitch also apply to String_switch.

    *)
  11. | Static_raise of Static_exception.t * Variable.t list
  12. | Static_catch of Static_exception.t * Variable.t list * t * t
  13. | Try_with of t * Variable.t * t
  14. | While of t * t
  15. | For of for_loop
  16. | Proved_unreachable

Flambda terms are partitioned in a pseudo-ANF manner; many terms are required to be let-bound. This in particular ensures there is always a variable name for an expression that may be lifted out (for example if it is found to be constant). Note: All bound variables in Flambda terms must be distinct. Flambda_invariants verifies this.

and named =
  1. | Symbol of Symbol.t
  2. | Const of const
  3. | Allocated_const of Allocated_const.t
  4. | Read_mutable of Mutable_variable.t
  5. | Read_symbol_field of Symbol.t * int
    (*

    During the lifting of let bindings to program constructions after closure conversion, we generate symbols and their corresponding definitions (which may or may not be constant), together with field accesses to such symbols. We would like it to be the case that such field accesses are simplified to the relevant component of the symbol concerned. (The rationale is to generate efficient code and share constants as expected: see e.g. tests/asmcomp/staticalloc.ml.) The components of the symbol would be identified by other symbols. This sort of access pattern is feasible because the top-level structure of symbols is statically allocated and fixed at compile time. It may seem that Prim (Pfield, ...) expressions could be used to perform the field accesses. However for simplicity, to avoid having to keep track of properties of individual fields of blocks, Inconstant_idents never deems a Prim (Pfield, ...) expression to be constant. This would in general prevent field accesses to symbols from being simplified in the way we would like, since Lift_constants would not assign new symbols (i.e. the things we would like to simplify to) to the various projections from the symbols in question. To circumvent this problem we use Read_symbol_field when generating projections from the top level of symbols. Owing to the properties of symbols described above, such expressions may be eligible for declaration as constant by Inconstant_idents (and thus themselves lifted to another symbol), without any further complication. Read_symbol_field may only be used when the definition of the symbol is in scope in the program. For external unresolved symbols, Pfield may still be used; it will be changed to Read_symbol_field by Inline_and_simplify when (and if) the symbol is imported.

    *)
  6. | Set_of_closures of set_of_closures
  7. | Project_closure of project_closure
  8. | Move_within_set_of_closures of move_within_set_of_closures
  9. | Project_var of project_var
  10. | Prim of Clambda_primitives.primitive * Variable.t list * Debuginfo.t
  11. | Expr of t
    (*

    ANF escape hatch.

    *)

Values of type named will always be let-bound to a Variable.t.

and let_expr = private {
  1. var : Variable.t;
  2. defining_expr : named;
  3. body : t;
  4. free_vars_of_defining_expr : Variable.Set.t;
    (*

    A cache of the free variables in the defining expression of the let.

    *)
  5. free_vars_of_body : Variable.Set.t;
    (*

    A cache of the free variables of the body of the let. This is an important optimization.

    *)
}
and let_mutable = {
  1. var : Mutable_variable.t;
  2. initial_value : Variable.t;
  3. contents_kind : Lambda.value_kind;
  4. body : t;
}
and set_of_closures = private {
  1. function_decls : function_declarations;
  2. free_vars : specialised_to Variable.Map.t;
    (*

    Mapping from all variables free in the body of the function_decls to variables in scope at the definition point of the set_of_closures. The domain of this map is sometimes known as the "variables bound by the closure".

    *)
  3. specialised_args : specialised_to Variable.Map.t;
    (*

    Parameters whose corresponding arguments are known to always alias a particular value. These are the only parameters that may, during Inline_and_simplify, have non-unknown approximations.

    An argument may only be specialised to a variable in the scope of the corresponding set of closures declaration. Usually, that variable itself also appears in the position of the specialised argument at all call sites of the function. However it may also be the case (for example in code generated as a result of Augment_specialised_args) that the various call sites of such a function have differing variables in the position of the specialised argument. This is permissible *so long as it is certain they all alias the same value*. Great care must be taken in transformations that result in this situation since there are no invariant checks for correctness.

    As an example, supposing all call sites of f are represented here: let x = ... in let f a b c = ... in let y = ... in f x y 1; f x y 1 the specialised arguments of f can (but does not necessarily) contain the association a -> x, but cannot contain b -> y because f is not in the scope of y. If f were the recursive function let rec f a b c = f a 1 2 in, a -> x would still be a valid specialised argument because all recursive calls maintain the invariant.

    This information is used for optimization purposes, if such a binding is known, it is possible to specialise the body of the function according to its parameter. This is usually introduced when specialising a recursive function, for instance. let rec map f = function | [] -> [] | h :: t -> f h :: map f t let map_succ l = let succ x = x + 1 in map succ l map can be duplicated in map_succ to be specialised for the argument f. This will result in let map_succ l = let succ x = x + 1 in let rec map f = function | [] -> [] | h :: t -> f h :: map f t in map succ l with map having f -> succ in its specialised_args field.

    Specialised argument information for arguments that are used must never be erased. This ensures that specialised arguments whose approximations describe closures maintain those approximations, which is essential to transport the closure freshening information to the point of use (e.g. a Project_var from such an argument).

    *)
  4. direct_call_surrogates : Variable.t Variable.Map.t;
    (*

    If direct_call_surrogates maps fun_var1 to fun_var2 then direct calls to fun_var1 should be redirected to fun_var2. This is used to reduce the overhead of transformations that introduce wrapper functions (which will be inlined at direct call sites, but will penalise indirect call sites). direct_call_surrogates may not be transitively closed.

    *)
}

The representation of a set of function declarations (possibly mutually recursive). Such a set encapsulates the declarations themselves, information about their defining environment, and information used specifically for optimization. Before a function can be applied it must be "projected" from a set of closures to yield a "closure". This is done using Project_closure (see above). Given a closure, not only can it be applied, but information about its defining environment can be retrieved (using Project_var, see above). At runtime, a set_of_closures corresponds to an OCaml value with tag Closure_tag (possibly with inline Infix_tag(s)). As an optimization, an operation (Move_within_set_of_closures) is provided (see above) which enables one closure within a set to be located given another closure in the same set. This avoids keeping a pointer to the whole set of closures alive when compiling, for example, mutually-recursive functions.

and function_declarations = private {
  1. is_classic_mode : bool;
    (*

    Indicates whether this function_declarations was compiled with -Oclassic.

    *)
  2. set_of_closures_id : Set_of_closures_id.t;
    (*

    An identifier (unique across all Flambda trees currently in memory) of the set of closures associated with this set of function declarations.

    *)
  3. set_of_closures_origin : Set_of_closures_origin.t;
    (*

    An identifier of the original set of closures on which this set of function declarations is based. Used to prevent different specialisations of the same functions from being inlined/specialised within each other.

    *)
  4. funs : function_declaration Variable.Map.t;
    (*

    The function(s) defined by the set of function declarations. The keys of this map are often referred to in the code as "fun_var"s.

    *)
}
and function_declaration = private {
  1. closure_origin : Closure_origin.t;
  2. params : Parameter.t list;
  3. body : t;
  4. free_variables : Variable.Set.t;
    (*

    All variables free in the *body* of the function. For example, a variable that is bound as one of the function's parameters will still be included in this set. This field is present as an optimization.

    *)
  5. free_symbols : Symbol.Set.t;
    (*

    All symbols that occur in the function's body. (Symbols can never be bound in a function's body; the only thing that binds symbols is the program constructions below.)

    *)
  6. stub : bool;
    (*

    A stub function is a generated function used to prepare arguments or return values to allow indirect calls to functions with a special calling convention. For instance indirect calls to tuplified functions must go through a stub. Stubs will be unconditionally inlined.

    *)
  7. dbg : Debuginfo.t;
    (*

    Debug info for the function declaration.

    *)
  8. inline : Lambda.inline_attribute;
    (*

    Inlining requirements from the source code.

    *)
  9. specialise : Lambda.specialise_attribute;
    (*

    Specialising requirements from the source code.

    *)
  10. is_a_functor : bool;
    (*

    Whether the function is known definitively to be a functor.

    *)
}
and switch = {
  1. numconsts : Numbers.Int.Set.t;
    (*

    Integer cases

    *)
  2. consts : (int * t) list;
    (*

    Integer cases

    *)
  3. numblocks : Numbers.Int.Set.t;
    (*

    Number of tag block cases

    *)
  4. blocks : (int * t) list;
    (*

    Tag block cases

    *)
  5. failaction : t option;
    (*

    Action to take if none matched

    *)
}

Equivalent to the similar type in Lambda.

and for_loop = {
  1. bound_var : Variable.t;
  2. from_value : Variable.t;
  3. to_value : Variable.t;
  4. direction : Asttypes.direction_flag;
  5. body : t;
}

Equivalent to the similar type in Lambda.

and constant_defining_value =
  1. | Allocated_const of Allocated_const.t
    (*

    A single constant. These are never "simple constants" (type const) but instead more complicated constructions.

    *)
  2. | Block of Tag.t * constant_defining_value_block_field list
    (*

    A pre-allocated block full of constants (either simple constants or references to other constants, see below).

    *)
  3. | Set_of_closures of set_of_closures
    (*

    A closed (and thus constant) set of closures. (That is to say, free_vars must be empty.)

    *)
  4. | Project_closure of Symbol.t * Closure_id.t
    (*

    Selection of one closure from a constant set of closures. Analogous to the equivalent operation on expressions.

    *)

Like a subset of Flambda.named, except that instead of Variable.ts we have Symbol.ts, and everything is a constant (i.e. with a fixed value known at compile time). Values of this type describe constants that will be directly assigned to symbols in the object file (see below).

and constant_defining_value_block_field =
  1. | Symbol of Symbol.t
  2. | Const of const
type expr = t
type program_body =
  1. | Let_symbol of Symbol.t * constant_defining_value * program_body
    (*

    Define the given symbol to have the given constant value.

    *)
  2. | Let_rec_symbol of (Symbol.t * constant_defining_value) list * program_body
    (*

    As for Let_symbol, but recursive. This is needed to treat examples like this, where a constant set of closures is lifted to toplevel:

    let rec f x = f x

    After lifting this produces (in pseudo-Flambda):

    Let_rec_symbol set_of_closures_symbol = (Set_of_closures f x -> let applied_function = Symbol f_closure in Apply (applied_function, x) ) and f_closure = Project_closure (set_of_closures_symbol, f)

    Use of Let_rec_symbol, by virtue of the special handling in Inline_and_simplify.define_let_rec_symbol_approx, enables the approximation of the set of closures to be present in order to correctly simplify the Project_closure construction. (See Inline_and_simplify.simplify_project_closure for that part.)

    *)
  3. | Initialize_symbol of Symbol.t * Tag.t * t list * program_body
    (*

    Define the given symbol as a constant block of the given size and tag; but with a possibly non-constant initializer. The initializer will be executed at most once (from the entry point of the compilation unit).

    *)
  4. | Effect of t * program_body
    (*

    Cause the given expression, which may have a side effect, to be executed. The resulting value is discarded. Effect constructions are never re-ordered.

    *)
  5. | End of Symbol.t
    (*

    End accepts the root symbol: the only symbol that can never be eliminated.

    *)

A "program" is the contents of one compilation unit. It describes the various values that are assigned to symbols (and in some cases fields of such symbols) in the object file. As such, it is closely related to the compilation of toplevel modules.

type program = {
  1. imported_symbols : Symbol.Set.t;
  2. program_body : program_body;
}
val free_variables : ?ignore_uses_as_callee:unit -> ?ignore_uses_as_argument:unit -> ?ignore_uses_in_project_var:unit -> t -> Variable.Set.t

Compute the free variables of a term. (This is O(1) for Lets). If ignore_uses_as_callee, all free variables inside Apply expressions are ignored. Likewise ignore_uses_in_project_var for Project_var expressions.

val free_variables_named : ?ignore_uses_in_project_var:unit -> named -> Variable.Set.t

Compute the free variables of a named expression.

val used_variables : ?ignore_uses_as_callee:unit -> ?ignore_uses_as_argument:unit -> ?ignore_uses_in_project_var:unit -> t -> Variable.Set.t

Compute _all_ variables occurring inside an expression.

val used_variables_named : ?ignore_uses_in_project_var:unit -> named -> Variable.Set.t

Compute _all_ variables occurring inside a named expression.

val free_symbols : expr -> Symbol.Set.t
val free_symbols_named : named -> Symbol.Set.t
val free_symbols_program : program -> Symbol.Set.t
val fold_lets_option : t -> init:'a -> for_defining_expr:('a -> Variable.t -> named -> 'a * Variable.t * named) -> for_last_body:('a -> t -> t * 'b) -> filter_defining_expr: ('b -> Variable.t -> named -> Variable.Set.t -> 'b * Variable.t * named option) -> t * 'b

Used to avoid exceeding the stack limit when handling expressions with multiple consecutive nested Let-expressions. This saves rewriting large simplification functions in CPS. This function provides for the rewriting or elimination of expressions during the fold.

val map_lets : t -> for_defining_expr:(Variable.t -> named -> named) -> for_last_body:(t -> t) -> after_rebuild:(t -> t) -> t

Like fold_lets_option, but just a map.

val iter_lets : t -> for_defining_expr:(Variable.t -> named -> unit) -> for_last_body:(t -> unit) -> for_each_let:(t -> unit) -> unit

Like map_lets, but just an iterator.

val create_let : Variable.t -> named -> t -> t

Creates a Let expression. (This computes the free variables of the defining expression and the body.)

val map_defining_expr_of_let : let_expr -> f:(named -> named) -> t

Apply the specified function f to the defining expression of the given Let-expression, returning a new Let.

module With_free_variables : sig ... end

A module for the manipulation of terms where the recomputation of free variable sets is to be kept to a minimum.

val create_function_declaration : params:Parameter.t list -> body:t -> stub:bool -> dbg:Debuginfo.t -> inline:Lambda.inline_attribute -> specialise:Lambda.specialise_attribute -> is_a_functor:bool -> closure_origin:Closure_origin.t -> function_declaration

Create a function declaration. This calculates the free variables and symbols occurring in the specified body.

val update_function_declaration : function_declaration -> params:Parameter.t list -> body:t -> function_declaration

Create a function declaration based on another function declaration

val create_function_declarations : is_classic_mode:bool -> funs:function_declaration Variable.Map.t -> function_declarations

Create a set of function declarations given the individual declarations.

val create_function_declarations_with_origin : is_classic_mode:bool -> funs:function_declaration Variable.Map.t -> set_of_closures_origin:Set_of_closures_origin.t -> function_declarations

Create a set of function declarations with a given set of closures origin.

val update_body_of_function_declaration : function_declaration -> body:expr -> function_declaration

Change only the code of a function declaration.

val update_function_decl's_params_and_body : function_declaration -> params:Parameter.t list -> body:expr -> function_declaration

Change only the code and parameters of a function declaration.

Create a set of function declarations based on another set of function declarations.

val create_function_declarations_with_closures_origin : is_classic_mode:bool -> funs:function_declaration Variable.Map.t -> set_of_closures_origin:Set_of_closures_origin.t -> function_declarations
val create_set_of_closures : function_decls:function_declarations -> free_vars:specialised_to Variable.Map.t -> specialised_args:specialised_to Variable.Map.t -> direct_call_surrogates:Variable.t Variable.Map.t -> set_of_closures

Create a set of closures. Checks are made to ensure that free_vars and specialised_args are reasonable.

Given a function declaration, find which of its parameters (if any) are used in the body.

type maybe_named =
  1. | Is_expr of t
  2. | Is_named of named
val iter_general : toplevel:bool -> (t -> unit) -> (named -> unit) -> maybe_named -> unit

This function is designed for the internal use of Flambda_iterators. See that module for iterators to be used over Flambda terms.

val print : Format.formatter -> t -> unit
val print_named : Format.formatter -> named -> unit
val print_program : Format.formatter -> program -> unit
val print_const : Format.formatter -> const -> unit
val print_constant_defining_value : Format.formatter -> constant_defining_value -> unit
val print_function_declaration : Format.formatter -> (Variable.t * function_declaration) -> unit
val print_function_declarations : Format.formatter -> function_declarations -> unit
val print_project_closure : Format.formatter -> project_closure -> unit
val print_move_within_set_of_closures : Format.formatter -> move_within_set_of_closures -> unit
val print_project_var : Format.formatter -> project_var -> unit
val print_set_of_closures : Format.formatter -> set_of_closures -> unit
val print_specialised_to : Format.formatter -> specialised_to -> unit
val equal_call_kind : call_kind -> call_kind -> bool
val equal_specialised_to : specialised_to -> specialised_to -> bool
val compare_const : const -> const -> int
val compare_project_var : project_var -> project_var -> int
val compare_move_within_set_of_closures : move_within_set_of_closures -> move_within_set_of_closures -> int
val compare_project_closure : project_closure -> project_closure -> int
OCaml

Innovation. Community. Security.

GitHub Discord Twitter Peertube RSS
About
Changelog Releases Industrial Users Academic Users Why OCaml
Ecosystem
Packages Community Events OCaml Planet Jobs
Resources
Install OCaml Get Started Platform Tools Language Manual Standard Library API Books Exercises Papers OCaml Playground Logo
Policies
Carbon Footprint Governance Privacy Code of Conduct