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

PCC: check facts on loaded and stored values, infer facts where needed, and add a basic vmctx/memory example. #7231

Merged
merged 5 commits into from
Oct 13, 2023
Merged

PCC: check facts on loaded and stored values, infer facts where needed, and add a basic vmctx/memory example. #7231

Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c5441a5
PCC: add checking of struct field types on loads and stores.
cfallin Oct 12, 2023
cf9578b
PCC: check that blockparams uphold facts appropriately.
cfallin Oct 12, 2023
e836e37
PCC: infer facts based on types' value ranges; add a mock vmctx example.
cfallin Oct 12, 2023
01239e0
Fix formatting.
cfallin Oct 12, 2023
a5433ec
Add two fixes required by new default facts:
cfallin Oct 13, 2023
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
9 changes: 9 additions & 0 deletions cranelift/codegen/src/ir/pcc.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,14 @@ impl<'a> FactContext<'a> {
// Reflexivity.
(l, r) if l == r => true,

// Any value on the LHS subsumes a minimal (always-true)
// fact about the max value of a given bitwidth on the
// RHS: e.g., no matter the value, the bottom 8 bits will
// always be <= 255.
(_, Fact::ValueMax { bit_width, max }) if *max == max_value_for_width(*bit_width) => {
true
}
Comment on lines +233 to +239
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think this is technically wrong but it feels like a footgun to allow a Mem fact to subsume a minimal ValueMax fact.

Would it be a problem to limit this to ValueMax LHSes? I guess that is what would happen if this case wasn't here, due to the match arm just below this one. Can you explain in more detail why we need this kind of cross-fact-kind subsumption then?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems that that narrower rule is indeed sufficient, at least for our test cases. The new rule is still necessary (not covered by the rule below) because of the remaining difference: it doesn't require matching bit-widths.

All of this is a bit of awkward fallout of the way default facts are working now though, so I'm going to see if I can try again to eliminate them and make subsume take the type instead. If not, I'll update as you suggest!


(
Fact::ValueMax {
bit_width: bw_lhs,
Expand Down Expand Up @@ -285,6 +293,7 @@ impl<'a> FactContext<'a> {
},
) if bw_lhs == bw_rhs && add_width >= *bw_lhs => {
let computed_max = lhs.checked_add(*rhs)?;
let computed_max = std::cmp::min(max_value_for_width(add_width), computed_max);
Some(Fact::ValueMax {
bit_width: *bw_lhs,
max: computed_max,
Expand Down