rust/src/librustc/middle/dataflow.rs

605 lines
22 KiB
Rust
Raw Normal View History

// Copyright 2012-2014 The Rust Project Developers. See the COPYRIGHT
2013-03-15 14:24:24 -05:00
// file at the top-level directory of this distribution and at
// http://rust-lang.org/COPYRIGHT.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
/*!
* A module for propagating forward dataflow information. The analysis
* assumes that the items to be propagated can be represented as bits
* and thus uses bitvectors. Your job is simply to specify the so-called
* GEN and KILL bits for each expression.
*/
use middle::cfg;
use middle::cfg::CFGIndex;
use middle::ty;
2013-11-11 00:46:32 -06:00
use std::io;
use std::uint;
2013-03-15 14:24:24 -05:00
use syntax::ast;
use syntax::ast_util::IdRange;
use syntax::visit;
2013-03-15 14:24:24 -05:00
use syntax::print::{pp, pprust};
use util::nodemap::NodeMap;
2013-03-15 14:24:24 -05:00
#[deriving(Show)]
pub enum EntryOrExit { Entry, Exit }
2013-07-02 14:47:32 -05:00
#[deriving(Clone)]
pub struct DataFlowContext<'a, 'tcx: 'a, O> {
tcx: &'a ty::ctxt<'tcx>,
2013-03-15 14:24:24 -05:00
/// a name for the analysis using this dataflow instance
analysis_name: &'static str,
2013-03-15 14:24:24 -05:00
/// the data flow operator
oper: O,
2013-03-15 14:24:24 -05:00
/// number of bits to propagate per id
bits_per_id: uint,
2013-03-15 14:24:24 -05:00
/// number of words we will use to store bits_per_id.
/// equal to bits_per_id/uint::BITS rounded up.
words_per_id: uint,
2013-03-15 14:24:24 -05:00
// mapping from node to cfg node index
// FIXME (#6298): Shouldn't this go with CFG?
nodeid_to_index: NodeMap<CFGIndex>,
// Bit sets per cfg node. The following three fields (`gens`, `kills`,
2013-03-15 14:24:24 -05:00
// and `on_entry`) all have the same structure. For each id in
// `id_range`, there is a range of words equal to `words_per_id`.
// So, to access the bits for any given id, you take a slice of
// the full vector (see the method `compute_id_range()`).
/// bits generated as we exit the cfg node. Updated by `add_gen()`.
gens: Vec<uint>,
2013-03-15 14:24:24 -05:00
/// bits killed as we exit the cfg node. Updated by `add_kill()`.
kills: Vec<uint>,
2013-03-15 14:24:24 -05:00
/// bits that are valid on entry to the cfg node. Updated by
2013-03-15 14:24:24 -05:00
/// `propagate()`.
on_entry: Vec<uint>,
}
2013-03-15 14:24:24 -05:00
pub trait BitwiseOperator {
/// Joins two predecessor bits together, typically either `|` or `&`
fn join(&self, succ: uint, pred: uint) -> uint;
}
2013-03-15 14:24:24 -05:00
/// Parameterization for the precise form of data flow that is used.
pub trait DataFlowOperator : BitwiseOperator {
2013-03-15 14:24:24 -05:00
/// Specifies the initial value for each bit in the `on_entry` set
fn initial_value(&self) -> bool;
}
struct PropagationContext<'a, 'b: 'a, 'tcx: 'b, O: 'a> {
dfcx: &'a mut DataFlowContext<'b, 'tcx, O>,
changed: bool
}
fn to_cfgidx_or_die(id: ast::NodeId, index: &NodeMap<CFGIndex>) -> CFGIndex {
let opt_cfgindex = index.find(&id).map(|&i|i);
opt_cfgindex.unwrap_or_else(|| {
fail!("nodeid_to_index does not have entry for NodeId {}", id);
})
}
impl<'a, 'tcx, O:DataFlowOperator> DataFlowContext<'a, 'tcx, O> {
fn has_bitset_for_nodeid(&self, n: ast::NodeId) -> bool {
assert!(n != ast::DUMMY_NODE_ID);
self.nodeid_to_index.contains_key(&n)
}
}
2013-03-15 14:24:24 -05:00
impl<'a, 'tcx, O:DataFlowOperator> pprust::PpAnn for DataFlowContext<'a, 'tcx, O> {
2014-03-16 13:58:11 -05:00
fn pre(&self,
ps: &mut pprust::State,
2014-03-16 13:58:11 -05:00
node: pprust::AnnNode) -> io::IoResult<()> {
let id = match node {
pprust::NodeIdent(_) | pprust::NodeName(_) => 0,
2014-03-16 13:58:11 -05:00
pprust::NodeExpr(expr) => expr.id,
pprust::NodeBlock(blk) => blk.id,
pprust::NodeItem(_) => 0,
pprust::NodePat(pat) => pat.id
};
if self.has_bitset_for_nodeid(id) {
assert!(self.bits_per_id > 0);
let cfgidx = to_cfgidx_or_die(id, &self.nodeid_to_index);
let (start, end) = self.compute_id_range(cfgidx);
let on_entry = self.on_entry.slice(start, end);
let entry_str = bits_to_string(on_entry);
let gens = self.gens.slice(start, end);
let gens_str = if gens.iter().any(|&u| u != 0) {
format!(" gen: {}", bits_to_string(gens))
} else {
"".to_string()
};
let kills = self.kills.slice(start, end);
let kills_str = if kills.iter().any(|&u| u != 0) {
format!(" kill: {}", bits_to_string(kills))
} else {
"".to_string()
};
try!(ps.synth_comment(format!("id {}: {}{}{}", id, entry_str,
gens_str, kills_str)));
try!(pp::space(&mut ps.s));
}
2014-01-29 20:42:19 -06:00
Ok(())
}
}
fn build_nodeid_to_index(decl: Option<&ast::FnDecl>,
cfg: &cfg::CFG) -> NodeMap<CFGIndex> {
let mut index = NodeMap::new();
// FIXME (#6298): Would it be better to fold formals from decl
// into cfg itself? i.e. introduce a fn-based flow-graph in
// addition to the current block-based flow-graph, rather than
// have to put traversals like this here?
match decl {
None => {}
Some(decl) => add_entries_from_fn_decl(&mut index, decl, cfg.entry)
}
cfg.graph.each_node(|node_idx, node| {
if node.data.id != ast::DUMMY_NODE_ID {
index.insert(node.data.id, node_idx);
}
true
});
return index;
fn add_entries_from_fn_decl(index: &mut NodeMap<CFGIndex>,
decl: &ast::FnDecl,
entry: CFGIndex) {
//! add mappings from the ast nodes for the formal bindings to
//! the entry-node in the graph.
struct Formals<'a> {
entry: CFGIndex,
index: &'a mut NodeMap<CFGIndex>,
}
let mut formals = Formals { entry: entry, index: index };
visit::walk_fn_decl(&mut formals, decl);
impl<'a, 'v> visit::Visitor<'v> for Formals<'a> {
fn visit_pat(&mut self, p: &ast::Pat) {
self.index.insert(p.id, self.entry);
visit::walk_pat(self, p)
}
}
}
}
impl<'a, 'tcx, O:DataFlowOperator> DataFlowContext<'a, 'tcx, O> {
pub fn new(tcx: &'a ty::ctxt<'tcx>,
analysis_name: &'static str,
decl: Option<&ast::FnDecl>,
cfg: &cfg::CFG,
2013-03-15 14:24:24 -05:00
oper: O,
id_range: IdRange,
bits_per_id: uint) -> DataFlowContext<'a, 'tcx, O> {
let words_per_id = (bits_per_id + uint::BITS - 1) / uint::BITS;
let num_nodes = cfg.graph.all_nodes().len();
2013-03-15 14:24:24 -05:00
debug!("DataFlowContext::new(analysis_name: {:s}, id_range={:?}, \
bits_per_id={:?}, words_per_id={:?}) \
num_nodes: {}",
analysis_name, id_range, bits_per_id, words_per_id,
num_nodes);
let entry = if oper.initial_value() { uint::MAX } else {0};
2013-03-15 14:24:24 -05:00
let gens = Vec::from_elem(num_nodes * words_per_id, 0);
let kills = Vec::from_elem(num_nodes * words_per_id, 0);
let on_entry = Vec::from_elem(num_nodes * words_per_id, entry);
2013-03-15 14:24:24 -05:00
let nodeid_to_index = build_nodeid_to_index(decl, cfg);
2013-03-15 14:24:24 -05:00
DataFlowContext {
tcx: tcx,
analysis_name: analysis_name,
2013-03-15 14:24:24 -05:00
words_per_id: words_per_id,
nodeid_to_index: nodeid_to_index,
2013-03-15 14:24:24 -05:00
bits_per_id: bits_per_id,
oper: oper,
gens: gens,
kills: kills,
on_entry: on_entry
}
}
pub fn add_gen(&mut self, id: ast::NodeId, bit: uint) {
2013-03-15 14:24:24 -05:00
//! Indicates that `id` generates `bit`
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
debug!("{:s} add_gen(id={:?}, bit={:?})",
self.analysis_name, id, bit);
assert!(self.nodeid_to_index.contains_key(&id));
assert!(self.bits_per_id > 0);
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
let cfgidx = to_cfgidx_or_die(id, &self.nodeid_to_index);
let (start, end) = self.compute_id_range(cfgidx);
2014-09-14 22:27:36 -05:00
let gens = self.gens.slice_mut(start, end);
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
set_bit(gens, bit);
2013-03-15 14:24:24 -05:00
}
pub fn add_kill(&mut self, id: ast::NodeId, bit: uint) {
2013-03-15 14:24:24 -05:00
//! Indicates that `id` kills `bit`
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
debug!("{:s} add_kill(id={:?}, bit={:?})",
self.analysis_name, id, bit);
assert!(self.nodeid_to_index.contains_key(&id));
assert!(self.bits_per_id > 0);
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
let cfgidx = to_cfgidx_or_die(id, &self.nodeid_to_index);
let (start, end) = self.compute_id_range(cfgidx);
2014-09-14 22:27:36 -05:00
let kills = self.kills.slice_mut(start, end);
Ensure dataflow of a proc never looks at blocks from closed-over context. Details: in a program like: ``` type T = proc(int) -> int; /* 4 */ pub fn outer(captured /* pat 16 */: T) -> T { (proc(x /* pat 23 */) { ((captured /* 29 */).foo((x /* 30 */)) /* 28 */) } /* block 27 */ /* 20 */) } /* block 19 */ /* 12 */ ``` the `captured` arg is moved from the outer fn into the inner proc (id=20). The old dataflow analysis for flowed_move_data_moves, when looking at the inner proc, would attempt to add a kill bit for `captured` at the end of its scope; the problem is that it thought the end of the `captured` arg's scope was the outer fn (id=12), even though at that point in the analysis, the `captured` arg's scope should now be restricted to the proc itself (id=20). This patch fixes handling of upvars so that dataflow of a fn/proc should never attempts to add a gen or kill bit to any NodeId outside of the current fn/proc. It accomplishes this by adding an `LpUpvar` variant to `borrowck::LoanPath`, so for cases like `captured` above will carry both their original `var_id`, as before, as well as the `NodeId` for the closure that is capturing them. As a drive-by fix to another occurrence of a similar bug that nikomatsakis pointed out to me earlier, this also fixes `gather_loans::compute_kill_scope` so that it computes the kill scope of the `captured` arg to be block 27; that is, the block for the proc itself (id=20). (This is an updated version that generalizes the new loan path variant to cover all upvars, and thus renamed the variant from `LpCopiedUpvar` to just `LpUpvar`.)
2014-06-13 10:19:29 -05:00
set_bit(kills, bit);
2013-03-15 14:24:24 -05:00
}
fn apply_gen_kill(&self, cfgidx: CFGIndex, bits: &mut [uint]) {
//! Applies the gen and kill sets for `cfgidx` to `bits`
debug!("{:s} apply_gen_kill(cfgidx={}, bits={}) [before]",
self.analysis_name, cfgidx, mut_bits_to_string(bits));
assert!(self.bits_per_id > 0);
2013-03-15 14:24:24 -05:00
let (start, end) = self.compute_id_range(cfgidx);
let gens = self.gens.slice(start, end);
bitwise(bits, gens, &Union);
let kills = self.kills.slice(start, end);
bitwise(bits, kills, &Subtract);
debug!("{:s} apply_gen_kill(cfgidx={}, bits={}) [after]",
self.analysis_name, cfgidx, mut_bits_to_string(bits));
}
fn compute_id_range(&self, cfgidx: CFGIndex) -> (uint, uint) {
let n = cfgidx.node_id();
let start = n * self.words_per_id;
2013-03-15 14:24:24 -05:00
let end = start + self.words_per_id;
assert!(start < self.gens.len());
assert!(end <= self.gens.len());
assert!(self.gens.len() == self.kills.len());
assert!(self.gens.len() == self.on_entry.len());
2013-03-15 14:24:24 -05:00
(start, end)
}
pub fn each_bit_on_entry(&self,
id: ast::NodeId,
f: |uint| -> bool)
-> bool {
//! Iterates through each bit that is set on entry to `id`.
//! Only useful after `propagate()` has been called.
if !self.has_bitset_for_nodeid(id) {
return true;
}
let cfgidx = to_cfgidx_or_die(id, &self.nodeid_to_index);
self.each_bit_for_node(Entry, cfgidx, f)
}
pub fn each_bit_for_node(&self,
e: EntryOrExit,
cfgidx: CFGIndex,
f: |uint| -> bool)
-> bool {
//! Iterates through each bit that is set on entry/exit to `cfgidx`.
//! Only useful after `propagate()` has been called.
if self.bits_per_id == 0 {
// Skip the surprisingly common degenerate case. (Note
// compute_id_range requires self.words_per_id > 0.)
return true;
}
let (start, end) = self.compute_id_range(cfgidx);
let on_entry = self.on_entry.slice(start, end);
let temp_bits;
let slice = match e {
Entry => on_entry,
Exit => {
let mut t = on_entry.to_vec();
self.apply_gen_kill(cfgidx, t.as_mut_slice());
temp_bits = t;
temp_bits.as_slice()
}
};
debug!("{:s} each_bit_for_node({}, cfgidx={}) bits={}",
self.analysis_name, e, cfgidx, bits_to_string(slice));
self.each_bit(slice, f)
}
pub fn each_gen_bit(&self, id: ast::NodeId, f: |uint| -> bool)
-> bool {
//! Iterates through each bit in the gen set for `id`.
if !self.has_bitset_for_nodeid(id) {
return true;
}
if self.bits_per_id == 0 {
// Skip the surprisingly common degenerate case. (Note
// compute_id_range requires self.words_per_id > 0.)
return true;
}
let cfgidx = to_cfgidx_or_die(id, &self.nodeid_to_index);
let (start, end) = self.compute_id_range(cfgidx);
let gens = self.gens.slice(start, end);
debug!("{:s} each_gen_bit(id={:?}, gens={})",
self.analysis_name, id, bits_to_string(gens));
self.each_bit(gens, f)
}
fn each_bit(&self, words: &[uint], f: |uint| -> bool) -> bool {
2013-05-03 12:08:08 -05:00
//! Helper for iterating over the bits in a bit set.
//! Returns false on the first call to `f` that returns false;
//! if all calls to `f` return true, then returns true.
2013-05-03 12:08:08 -05:00
for (word_index, &word) in words.iter().enumerate() {
2013-05-03 12:08:08 -05:00
if word != 0 {
let base_index = word_index * uint::BITS;
for offset in range(0u, uint::BITS) {
2013-05-03 12:08:08 -05:00
let bit = 1 << offset;
if (word & bit) != 0 {
// NB: we round up the total number of bits
// that we store in any given bit set so that
// it is an even multiple of uint::BITS. This
2013-05-03 12:08:08 -05:00
// means that there may be some stray bits at
// the end that do not correspond to any
// actual value. So before we callback, check
// whether the bit_index is greater than the
// actual value the user specified and stop
// iterating if so.
let bit_index = base_index + offset;
if bit_index >= self.bits_per_id {
return true;
} else if !f(bit_index) {
return false;
}
}
}
}
}
return true;
}
pub fn add_kills_from_flow_exits(&mut self, cfg: &cfg::CFG) {
//! Whenever you have a `break` or `continue` statement, flow
//! exits through any number of enclosing scopes on its way to
//! the new destination. This function infers the kill bits of
//! those control operators based on the kill bits associated
//! with those scopes.
//!
//! This is usually called (if it is called at all), after
//! all add_gen and add_kill calls, but before propagate.
debug!("{:s} add_kills_from_flow_exits", self.analysis_name);
if self.bits_per_id == 0 {
// Skip the surprisingly common degenerate case. (Note
// compute_id_range requires self.words_per_id > 0.)
return;
}
cfg.graph.each_edge(|_edge_index, edge| {
let flow_exit = edge.source();
let (start, end) = self.compute_id_range(flow_exit);
let mut orig_kills = self.kills.slice(start, end).to_vec();
let mut changed = false;
for &node_id in edge.data.exiting_scopes.iter() {
let opt_cfg_idx = self.nodeid_to_index.find(&node_id).map(|&i|i);
match opt_cfg_idx {
Some(cfg_idx) => {
let (start, end) = self.compute_id_range(cfg_idx);
let kills = self.kills.slice(start, end);
if bitwise(orig_kills.as_mut_slice(), kills, &Union) {
changed = true;
}
}
None => {
debug!("{:s} add_kills_from_flow_exits flow_exit={} \
no cfg_idx for exiting_scope={:?}",
self.analysis_name, flow_exit, node_id);
}
}
}
if changed {
2014-09-14 22:27:36 -05:00
let bits = self.kills.slice_mut(start, end);
debug!("{:s} add_kills_from_flow_exits flow_exit={} bits={} [before]",
self.analysis_name, flow_exit, mut_bits_to_string(bits));
bits.copy_from(orig_kills.as_slice());
debug!("{:s} add_kills_from_flow_exits flow_exit={} bits={} [after]",
self.analysis_name, flow_exit, mut_bits_to_string(bits));
}
true
});
}
2013-03-15 14:24:24 -05:00
}
impl<'a, 'tcx, O:DataFlowOperator+Clone+'static> DataFlowContext<'a, 'tcx, O> {
// ^^^^^^^^^^^^^ only needed for pretty printing
pub fn propagate(&mut self, cfg: &cfg::CFG, blk: &ast::Block) {
2013-03-15 14:24:24 -05:00
//! Performs the data flow analysis.
if self.bits_per_id == 0 {
// Optimize the surprisingly common degenerate case.
return;
}
2013-07-02 14:47:32 -05:00
{
let words_per_id = self.words_per_id;
2013-07-02 14:47:32 -05:00
let mut propcx = PropagationContext {
2014-03-05 21:07:47 -06:00
dfcx: &mut *self,
2013-07-02 14:47:32 -05:00
changed: true
};
2013-03-15 14:24:24 -05:00
let mut temp = Vec::from_elem(words_per_id, 0u);
2013-07-02 14:47:32 -05:00
while propcx.changed {
propcx.changed = false;
propcx.reset(temp.as_mut_slice());
propcx.walk_cfg(cfg, temp.as_mut_slice());
2013-07-02 14:47:32 -05:00
}
2013-03-15 14:24:24 -05:00
}
debug!("Dataflow result for {:s}:", self.analysis_name);
debug!("{}", {
2014-04-25 03:08:02 -05:00
self.pretty_print_to(box io::stderr(), blk).unwrap();
2013-03-15 14:24:24 -05:00
""
});
}
fn pretty_print_to(&self, wr: Box<io::Writer+'static>,
2014-01-29 20:42:19 -06:00
blk: &ast::Block) -> io::IoResult<()> {
let mut ps = pprust::rust_printer_annotated(wr, self);
2014-03-16 13:58:11 -05:00
try!(ps.cbox(pprust::indent_unit));
try!(ps.ibox(0u));
try!(ps.print_block(blk));
pp::eof(&mut ps.s)
2013-03-15 14:24:24 -05:00
}
}
impl<'a, 'b, 'tcx, O:DataFlowOperator> PropagationContext<'a, 'b, 'tcx, O> {
fn walk_cfg(&mut self,
cfg: &cfg::CFG,
in_out: &mut [uint]) {
debug!("DataFlowContext::walk_cfg(in_out={}) {:s}",
bits_to_string(in_out), self.dfcx.analysis_name);
assert!(self.dfcx.bits_per_id > 0);
cfg.graph.each_node(|node_index, node| {
debug!("DataFlowContext::walk_cfg idx={} id={} begin in_out={}",
node_index, node.data.id, bits_to_string(in_out));
let (start, end) = self.dfcx.compute_id_range(node_index);
// Initialize local bitvector with state on-entry.
in_out.copy_from(self.dfcx.on_entry.slice(start, end));
// Compute state on-exit by applying transfer function to
// state on-entry.
self.dfcx.apply_gen_kill(node_index, in_out);
// Propagate state on-exit from node into its successors.
self.propagate_bits_into_graph_successors_of(in_out, cfg, node_index);
true // continue to next node
});
2013-03-15 14:24:24 -05:00
}
fn reset(&mut self, bits: &mut [uint]) {
let e = if self.dfcx.oper.initial_value() {uint::MAX} else {0};
2014-09-14 22:27:36 -05:00
for b in bits.iter_mut() {
*b = e;
}
2013-03-15 14:24:24 -05:00
}
fn propagate_bits_into_graph_successors_of(&mut self,
pred_bits: &[uint],
cfg: &cfg::CFG,
cfgidx: CFGIndex) {
cfg.graph.each_outgoing_edge(cfgidx, |_e_idx, edge| {
self.propagate_bits_into_entry_set_for(pred_bits, edge);
true
});
2013-03-15 14:24:24 -05:00
}
fn propagate_bits_into_entry_set_for(&mut self,
pred_bits: &[uint],
edge: &cfg::CFGEdge) {
let source = edge.source();
let cfgidx = edge.target();
debug!("{:s} propagate_bits_into_entry_set_for(pred_bits={}, {} to {})",
self.dfcx.analysis_name, bits_to_string(pred_bits), source, cfgidx);
assert!(self.dfcx.bits_per_id > 0);
let (start, end) = self.dfcx.compute_id_range(cfgidx);
let changed = {
// (scoping mutable borrow of self.dfcx.on_entry)
2014-09-14 22:27:36 -05:00
let on_entry = self.dfcx.on_entry.slice_mut(start, end);
bitwise(on_entry, pred_bits, &self.dfcx.oper)
2013-03-15 14:24:24 -05:00
};
if changed {
debug!("{:s} changed entry set for {:?} to {}",
self.dfcx.analysis_name, cfgidx,
bits_to_string(self.dfcx.on_entry.slice(start, end)));
2013-03-15 14:24:24 -05:00
self.changed = true;
}
}
}
fn mut_bits_to_string(words: &mut [uint]) -> String {
bits_to_string(words)
2013-03-15 14:24:24 -05:00
}
fn bits_to_string(words: &[uint]) -> String {
let mut result = String::new();
2013-03-15 14:24:24 -05:00
let mut sep = '[';
// Note: this is a little endian printout of bytes.
for &word in words.iter() {
2013-03-15 14:24:24 -05:00
let mut v = word;
for _ in range(0u, uint::BYTES) {
result.push_char(sep);
result.push_str(format!("{:02x}", v & 0xFF).as_slice());
2013-03-15 14:24:24 -05:00
v >>= 8;
sep = '-';
}
}
result.push_char(']');
return result
2013-03-15 14:24:24 -05:00
}
#[inline]
fn bitwise<Op:BitwiseOperator>(out_vec: &mut [uint],
in_vec: &[uint],
op: &Op) -> bool {
2013-03-15 14:24:24 -05:00
assert_eq!(out_vec.len(), in_vec.len());
let mut changed = false;
2014-09-14 22:27:36 -05:00
for (out_elt, in_elt) in out_vec.iter_mut().zip(in_vec.iter()) {
let old_val = *out_elt;
let new_val = op.join(old_val, *in_elt);
*out_elt = new_val;
changed |= old_val != new_val;
2013-03-15 14:24:24 -05:00
}
changed
2013-03-15 14:24:24 -05:00
}
fn set_bit(words: &mut [uint], bit: uint) -> bool {
debug!("set_bit: words={} bit={}",
mut_bits_to_string(words), bit_str(bit));
let word = bit / uint::BITS;
let bit_in_word = bit % uint::BITS;
2013-03-15 14:24:24 -05:00
let bit_mask = 1 << bit_in_word;
debug!("word={} bit_in_word={} bit_mask={}", word, bit_in_word, word);
2013-03-15 14:24:24 -05:00
let oldv = words[word];
let newv = oldv | bit_mask;
words[word] = newv;
oldv != newv
}
fn bit_str(bit: uint) -> String {
2013-03-15 14:24:24 -05:00
let byte = bit >> 8;
let lobits = 1u << (bit & 0xFF);
format!("[{}:{}-{:02x}]", bit, byte, lobits)
2013-03-15 14:24:24 -05:00
}
struct Union;
impl BitwiseOperator for Union {
fn join(&self, a: uint, b: uint) -> uint { a | b }
}
struct Subtract;
impl BitwiseOperator for Subtract {
fn join(&self, a: uint, b: uint) -> uint { a & !b }
}