Merge pull request #732 from notoria/indexing

Implemented a different way to index clauses

README.md

@@ -226,6 +226,37 @@ arithmetic operators with the usual precedences,
 
 New operators can be defined using the `op` declaration.
 
+### First instantiated argument indexing
+
+Scryer Prolog indexes on the leftmost argument that is not a variable
+in all clauses of a predicate's definition. We call this strategy
+first *instantiated* argument indexing.
+
+A key motivation for first instantiated argument indexing is to enable
+indexing for meta-predicates such as `maplist/N` and `foldl/N`, where
+the first argument is a goal or partial goal that is a variable in the
+definition of these predicates and therefore cannot be used for
+indexing.
+
+For example, a natural definiton of `maplist/2` reads:
+
+```
+maplist(_, []).
+maplist(Goal, [L|Ls]) :-
+        call(Goal, L),
+        maplist(Goal, Ls).
+```
+
+In this case, first instantianted argument indexing automatically uses
+the *second* argument for indexing, and thus prevents choicepoints for
+calls with lists of fixed lengths (and deterministic goals).
+Conveniently, no auxiliary predicates with reorderd arguments are
+needed to benefit from indexing in such cases.
+
+Conventional first argument indexing naturally arises as a
+special case of this strategy, if the first argument is instantiated
+in any clause of a predicate's definition.
+
 ### Strings and partial strings
 
 In Scryer Prolog, the default value of the Prolog flag `double_quotes`

src/codegen.rs

@@ -1,3 +1,4 @@
+/// Code generation to WAM-like instructions.
 use crate::prolog_parser::ast::*;
 
 use crate::allocator::*;
@@ -857,21 +858,63 @@ impl<'a, TermMarker: Allocator<'a>> CodeGenerator<TermMarker> {
         Ok(code)
     }
 
-    fn split_predicate(clauses: &Vec<PredicateClause>) -> Vec<(usize, usize)> {
+    /// Returns the index of the first instantiated argument.
+    fn first_instantiated_index(clauses: &[PredicateClause]) -> Option<usize> {
+        let mut optimal_index = None;
+        let has_args = match clauses.first() {
+            Some(clause) => match clause.args() {
+                Some(args) => !args.is_empty(),
+                None => false,
+            },
+            None => false,
+        };
+        if !has_args {
+            return optimal_index;
+        }
+        for clause in clauses.iter() {
+            let args = clause.args().unwrap();
+            for (i, arg) in args.iter().enumerate() {
+                if let Some(optimal_index) = optimal_index {
+                    if i >= optimal_index {
+                        break;
+                    }
+                }
+                match **arg {
+                    Term::AnonVar | Term::Var(..) => (),
+                    _ => {
+                        match optimal_index {
+                            Some(ref mut optimal_i) => *optimal_i = i,
+                            None => optimal_index = Some(i),
+                        }
+                        break;
+                    }
+                }
+            }
+        }
+
+        optimal_index
+    }
+
+    fn split_predicate(clauses: &[PredicateClause], optimal_index: usize) -> Vec<(usize, usize)> {
         let mut subseqs = Vec::new();
         let mut left_index = 0;
 
-        for (right_index, clause) in clauses.iter().enumerate() {
+        if clauses.first().unwrap().args().is_some() {
-            match clause.first_arg() {
+            for (right_index, clause) in clauses.iter().enumerate() {
-                Some(&Term::Var(_, _)) | Some(&Term::AnonVar) => {
+                // Can unwrap safely.
-                    if left_index < right_index {
+                if let Some(arg) = clause.args().unwrap().iter().nth(optimal_index) {
-                        subseqs.push((left_index, right_index));
+                    match **arg {
-                    }
+                        Term::Var(..) | Term::AnonVar => {
+                            if left_index < right_index {
+                                subseqs.push((left_index, right_index));
+                            }
 
-                    subseqs.push((right_index, right_index + 1));
+                            subseqs.push((right_index, right_index + 1));
-                    left_index = right_index + 1;
+                            left_index = right_index + 1;
+                        }
+                        _ => (),
+                    }
                 }
-                _ => {}
             }
         }
 
@@ -901,6 +944,7 @@ impl<'a, TermMarker: Allocator<'a>> CodeGenerator<TermMarker> {
     fn compile_pred_subseq<'b: 'a>(
         &mut self,
         clauses: &'b [PredicateClause],
+        optimal_index: usize,
     ) -> Result<Code, ParserError> {
         let mut code_body = Vec::new();
         let mut code_offsets = CodeOffsets::new();
@@ -910,9 +954,9 @@ impl<'a, TermMarker: Allocator<'a>> CodeGenerator<TermMarker> {
         for (i, clause) in clauses.iter().enumerate() {
             self.marker.reset();
 
-            let mut clause_code = match clause {
+            let mut clause_code = match *clause {
-                &PredicateClause::Fact(ref fact, ..) => self.compile_fact(fact),
+                PredicateClause::Fact(ref fact, ..) => self.compile_fact(fact),
-                &PredicateClause::Rule(ref rule, ..) => self.compile_rule(rule)?,
+                PredicateClause::Rule(ref rule, ..) => self.compile_rule(rule)?,
             };
 
             if num_clauses > 1 {
@@ -925,17 +969,24 @@ impl<'a, TermMarker: Allocator<'a>> CodeGenerator<TermMarker> {
                 code_body.push(Line::Choice(choice));
             }
 
-            clause.first_arg().map(|arg| {
+            let arg = match clause.args() {
+                Some(args) => match args.iter().nth(optimal_index) {
+                    Some(term) => Some(term),
+                    None => None,
+                },
+                None => None,
+            };
+            if let Some(arg) = arg {
                 let index = code_body.len();
                 code_offsets.index_term(arg, index);
-            });
+            }
 
             code_body.append(&mut clause_code);
         }
 
         let mut code = Vec::new();
+        code_offsets.add_indices(&mut code, code_body, optimal_index + 1);
 
-        code_offsets.add_indices(&mut code, code_body);
         Ok(code)
     }
 
@@ -944,11 +995,16 @@ impl<'a, TermMarker: Allocator<'a>> CodeGenerator<TermMarker> {
         clauses: &'b Vec<PredicateClause>,
     ) -> Result<Code, ParserError> {
         let mut code = Vec::new();
-        let split_pred = Self::split_predicate(&clauses);
+        let optimal_index = match Self::first_instantiated_index(&clauses) {
+            Some(index) => index,
+            None => 0, // Default to first argument indexing.
+        };
+        let split_pred = Self::split_predicate(&clauses, optimal_index);
         let multi_seq = split_pred.len() > 1;
 
         for (l, r) in split_pred {
-            let mut code_segment = self.compile_pred_subseq(&clauses[l..r])?;
+            let mut code_segment =
+                self.compile_pred_subseq(&clauses[l..r], optimal_index)?;
 
             if multi_seq {
                 let choice = match l {

src/forms.rs

@@ -332,10 +332,16 @@ pub enum PredicateClause {
 }
 
 impl PredicateClause {
-    pub fn first_arg(&self) -> Option<&Term> {
+    // TODO: add this to `Term` in `prolog_parser` like `first_arg`.
-        match self {
+    pub fn args(&self) -> Option<&[Box<Term>]> {
-            &PredicateClause::Fact(ref term, ..) => term.first_arg(),
+        match *self {
-            &PredicateClause::Rule(ref rule, ..) => rule.head.1.first().map(|bt| bt.as_ref()),
+            PredicateClause::Fact(ref term, ..) => {
+                match term {
+                    Term::Clause(_, _, args, _) => Some(&args),
+                    _ => None,
+                }
+            },
+            PredicateClause::Rule(ref rule, ..) => Some(&rule.head.1),
         }
     }
 

src/indexing.rs

@@ -143,8 +143,8 @@ impl CodeOffsets {
         }
     }
 
-    pub fn index_term(&mut self, first_arg: &Term, index: usize) {
+    pub fn index_term(&mut self, optimal_arg: &Term, index: usize) {
-        match first_arg {
+        match optimal_arg {
             &Term::Clause(_, ref name, ref terms, _) => {
                 let code = self
                     .structures
@@ -237,12 +237,17 @@ impl CodeOffsets {
     fn switch_on_constant(
         con_ind: IndexMap<Constant, ThirdLevelIndex>,
         prelude: &mut CodeDeque,
+        optimal_index: usize,
     ) -> IntIndex {
         let con_ind = Self::second_level_index(con_ind, prelude);
 
         if con_ind.len() > 1 {
             let index = Self::flatten_index(con_ind, prelude.len());
-            let instr = IndexingInstruction::SwitchOnConstant(index.len(), index);
+            let instr = IndexingInstruction::SwitchOnConstant(
+                optimal_index,
+                index.len(),
+                index
+            );
 
             prelude.push_front(Line::from(instr));
 
@@ -259,12 +264,17 @@ impl CodeOffsets {
     fn switch_on_structure(
         str_ind: IndexMap<(ClauseName, usize), ThirdLevelIndex>,
         prelude: &mut CodeDeque,
+        optimal_index: usize,
     ) -> IntIndex {
         let str_ind = Self::second_level_index(str_ind, prelude);
 
         if str_ind.len() > 1 {
             let index = Self::flatten_index(str_ind, prelude.len());
-            let instr = IndexingInstruction::SwitchOnStructure(index.len(), index);
+            let instr = IndexingInstruction::SwitchOnStructure(
+                optimal_index,
+                index.len(),
+                index
+            );
 
             prelude.push_front(Line::from(instr));
 
@@ -325,7 +335,7 @@ impl CodeOffsets {
         }
     }
 
-    pub fn add_indices(self, code: &mut Code, mut code_body: Code) {
+    pub fn add_indices(self, code: &mut Code, mut code_body: Code, optimal_index: usize) {
         if self.no_indices() {
             *code = code_body;
             return;
@@ -334,8 +344,10 @@ impl CodeOffsets {
         let mut prelude = VecDeque::new();
 
         let lst_loc = Self::switch_on_list(self.lists, &mut prelude);
-        let str_loc = Self::switch_on_structure(self.structures, &mut prelude);
+        let str_loc =
-        let con_loc = Self::switch_on_constant(self.constants, &mut prelude);
+            Self::switch_on_structure(self.structures, &mut prelude, optimal_index);
+        let con_loc =
+            Self::switch_on_constant(self.constants, &mut prelude, optimal_index);
 
         let prelude_length = prelude.len();
 
@@ -355,8 +367,13 @@ impl CodeOffsets {
         let con_loc = Self::switch_on_con_offset_from(con_loc, prelude.len());
         let lst_loc = Self::switch_on_lst_offset_from(lst_loc, prelude.len());
 
-        let switch_instr =
+        let switch_instr = IndexingInstruction::SwitchOnTerm(
-            IndexingInstruction::SwitchOnTerm(prelude.len() + 1, con_loc, lst_loc, str_loc);
+            optimal_index,
+            prelude.len() + 1,
+            con_loc,
+            lst_loc,
+            str_loc
+        );
 
         prelude.push_front(Line::from(switch_instr));
 

src/instructions.rs

@@ -143,6 +143,7 @@ impl IndexedChoiceInstruction {
     }
 }
 
+/// A `Line` is an instruction (cf. page 98 of wambook).
 #[derive(Debug)]
 pub enum Line {
     Arithmetic(ArithmeticInstruction),
@@ -424,11 +425,13 @@ impl ControlInstruction {
     }
 }
 
+/// `IndexingInstruction` cf. page 110 of wambook.
 #[derive(Debug)]
 pub enum IndexingInstruction {
-    SwitchOnTerm(usize, usize, usize, usize),
+    // The first index is the optimal argument being indexed.
-    SwitchOnConstant(usize, IndexMap<Constant, usize>),
+    SwitchOnTerm(usize, usize, usize, usize, usize),
-    SwitchOnStructure(usize, IndexMap<(ClauseName, usize), usize>),
+    SwitchOnConstant(usize, usize, IndexMap<Constant, usize>),
+    SwitchOnStructure(usize, usize, IndexMap<(ClauseName, usize), usize>),
 }
 
 impl From<IndexingInstruction> for Line {
@@ -440,25 +443,26 @@ impl From<IndexingInstruction> for Line {
 impl IndexingInstruction {
     pub fn to_functor(&self) -> MachineStub {
         match self {
-            &IndexingInstruction::SwitchOnTerm(vars, constants, lists, structures) => {
+            &IndexingInstruction::SwitchOnTerm(arg, vars, constants, lists, structures) => {
                 functor!(
                     "switch_on_term",
-                    [integer(vars),
+                    [integer(arg),
+                     integer(vars),
                      integer(constants),
                      integer(lists),
                      integer(structures)]
                 )
             }
-            &IndexingInstruction::SwitchOnConstant(constants, _) => {
+            &IndexingInstruction::SwitchOnConstant(arg, constants, _) => {
                 functor!(
                     "switch_on_constant",
-                    [integer(constants)]
+                    [integer(arg), integer(constants)]
                 )
             }
-            &IndexingInstruction::SwitchOnStructure(structures, _) => {
+            &IndexingInstruction::SwitchOnStructure(arg, structures, _) => {
                 functor!(
                     "switch_on_structure",
-                    [integer(structures)]
+                    [integer(arg), integer(structures)]
                 )
             }
         }

src/machine/machine_state_impl.rs

@@ -1392,8 +1392,8 @@ impl MachineState {
     pub(super)
     fn execute_indexing_instr(&mut self, instr: &IndexingInstruction) {
         match instr {
-            &IndexingInstruction::SwitchOnTerm(v, c, l, s) => {
+            &IndexingInstruction::SwitchOnTerm(arg, v, c, l, s) => {
-                let addr = self[temp_v!(1)];
+                let addr = self[temp_v!(arg)];
                 let addr = self.store(self.deref(addr));
 
                 let offset = match addr {
@@ -1423,8 +1423,8 @@ impl MachineState {
                     o => self.p += o,
                 };
             }
-            &IndexingInstruction::SwitchOnConstant(_, ref hm) => {
+            &IndexingInstruction::SwitchOnConstant(arg, _, ref hm) => {
-                let addr = self[temp_v!(1)];
+                let addr = self[temp_v!(arg)];
                 let addr = self.store(self.deref(addr));
 
                 let offset =
@@ -1445,8 +1445,8 @@ impl MachineState {
                     o => self.p += o,
                 };
             }
-            &IndexingInstruction::SwitchOnStructure(_, ref hm) => {
+            &IndexingInstruction::SwitchOnStructure(arg, _, ref hm) => {
-                let a1 = self.registers[1];
+                let a1 = self.registers[arg];
                 let addr = self.store(self.deref(a1));
 
                 let offset = match addr {

src/write.rs

@@ -276,13 +276,13 @@ impl fmt::Display for ChoiceInstruction {
 impl fmt::Display for IndexingInstruction {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         match self {
-            &IndexingInstruction::SwitchOnTerm(v, c, l, s) => {
+            &IndexingInstruction::SwitchOnTerm(a, v, c, l, s) => {
-                write!(f, "switch_on_term {}, {}, {}, {}", v, c, l, s)
+                write!(f, "switch_on_term {}, {}, {}, {}, {}", a, v, c, l, s)
             }
-            &IndexingInstruction::SwitchOnConstant(num_cs, _) => {
+            &IndexingInstruction::SwitchOnConstant(_, num_cs, _) => {
                 write!(f, "switch_on_constant {}", num_cs)
             }
-            &IndexingInstruction::SwitchOnStructure(num_ss, _) => {
+            &IndexingInstruction::SwitchOnStructure(_, num_ss, _) => {
                 write!(f, "switch_on_structure {}", num_ss)
             }
         }