Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

fix(mem2reg): Handle aliases better when setting a known value for a load #5959

Merged
merged 2 commits into from
Sep 6, 2024
Merged

fix(mem2reg): Handle aliases better when setting a known value for a load #5959

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
c43b9e6
handle aliases better when setting a known value for a load
vezenovm Sep 6, 2024
fbf0638
merge w/ master
vezenovm Sep 6, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 7 additions & 7 deletions compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,7 +56,7 @@
//!
//! Repeating this algorithm for each block in the function in program order should result in
//! optimizing out most known loads. However, identifying all aliases correctly has been proven
//! undecidable in general (Landi, 1992). So this pass will not always optimize out all loads

Check warning on line 59 in compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs

View workflow job for this annotation

GitHub Actions / Code 

Unknown word (Landi)
//! that could theoretically be optimized out. This pass can be performed at any time in the
//! SSA optimization pipeline, although it will be more successful the simpler the program's CFG is.
//! This pass is currently performed several times to enable other passes - most notably being
Expand Down Expand Up @@ -111,7 +111,7 @@
/// Load and Store instructions that should be removed at the end of the pass.
///
/// We avoid removing individual instructions as we go since removing elements
/// from the middle of Vecs many times will be slower than a single call to `retain`.

Check warning on line 114 in compiler/noirc_evaluator/src/ssa/opt/mem2reg.rs

View workflow job for this annotation

GitHub Actions / Code 

Unknown word (Vecs)
instructions_to_remove: BTreeSet<InstructionId>,

/// Track a value's last load across all blocks.
Expand Down Expand Up @@ -286,19 +286,19 @@
} else {
references.mark_value_used(address, self.inserter.function);

let expression = if let Some(expression) = references.expressions.get(&result) {
expression.clone()
} else {
references.expressions.insert(result, Expression::Other(result));
Expression::Other(result)
};
if let Some(aliases) = references.aliases.get_mut(&expression) {
let expression =
references.expressions.entry(result).or_insert(Expression::Other(result));
// Make sure this load result is marked an alias to itself
if let Some(aliases) = references.aliases.get_mut(expression) {
// If we have an alias set, add to the set
aliases.insert(result);
} else {
// Otherwise, create a new alias set containing just the load result
references
.aliases
.insert(Expression::Other(result), AliasSet::known(result));
}
// Mark that we know a load result is equivalent to the address of a load.
references.set_known_value(result, address);

self.last_loads.insert(address, (instruction, block_id));
Expand Down
Loading