T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Cranelift: remove slow invariant validation in cfg(fuzzing) from MachBuffer. (#4038)

Browse Source 
Following the merge of regalloc2 support, this became slower because we
are stricter about the critical-edge invariant, generating a separate
edge block for every out-edge even if two or more out-edges go to the
same successor (this is significant in cases of `br_table` with many
entries having the same target block, for example).

Many of those edge blocks are empty and end up collapsed by the
MachBuffer, which leads to a large set of aliased labels.

The invariant validation will dutifully iterate over all the data
structures at every step, validating all of our conditions. But this
gets way slower in the new context, to the point that we'll probably
have some fuzz timeouts.

This was pointed out in [1] but I missed removing this in #3989. Given
that `MachBuffer` has been around for nearly two years now, has been
fuzzed continuously with the invariant validation for that time, and
also has a correctness proof in the comments, it's probably reasonable
to remove this high (recently increased) cost from the fuzzing-specific
compilation configuration.

[1]
https://github.com/bytecodealliance/wasmtime/pull/3989#discussion_r847712263
This commit is contained in:
Chris Fallin
2022-04-15 07:04:02 -07:00
committed by GitHub
parent cf533a8041
commit 7cf5f05830
1 changed files with 0 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
cranelift/codegen/src/machinst/buffer.rs
View File

@@ -326,62 +326,6 @@ impl<I: VCodeInst> MachBuffer<I> {
}
}
/// Debug-only: check invariants of labels and branch-records described
/// under "Branch-optimization Correctness" above.
///
/// These invariants are checked at branch-simplification
/// time. Note that they may be temporarily violated at other
/// times, e.g. after calling `add_{cond,uncond}_branch()` and
/// before emitting branch bytes.
fn check_label_branch_invariants(&self) {
if !cfg!(fuzzing) {
return;
}
let cur_off = self.cur_offset();
// Check that every entry in latest_branches has *correct*
// labels_at_this_branch lists. We do not check completeness because
// that would require building a reverse index, which is too slow even
// for a debug invariant check.
let mut last_end = 0;
for b in &self.latest_branches {
debug_assert!(b.start < b.end);
debug_assert!(b.end <= cur_off);
debug_assert!(b.start >= last_end);
last_end = b.end;
for &l in &b.labels_at_this_branch {
debug_assert_eq!(self.resolve_label_offset(l), b.start);
debug_assert_eq!(self.label_aliases[l.0 as usize], UNKNOWN_LABEL);
}
}
// Check that every label is unresolved, or resolved at or
// before cur_offset. If at cur_offset, must be in
// `labels_at_tail`. We skip labels that are aliased to
// others already.
for (i, &off) in self.label_offsets.iter().enumerate() {
let label = MachLabel(i as u32);
if self.label_aliases[i] != UNKNOWN_LABEL {
continue;
}
debug_assert!(off == UNKNOWN_LABEL_OFFSET || off <= cur_off);
if off == cur_off {
debug_assert!(
self.labels_at_tail_off == cur_off && self.labels_at_tail.contains(&label)
);
}
}
// Check that every label in `labels_at_tail_off` is precise, i.e.,
// resolves to the cur offset.
debug_assert!(self.labels_at_tail_off <= cur_off);
if self.labels_at_tail_off == cur_off {
for &l in &self.labels_at_tail {
debug_assert_eq!(self.resolve_label_offset(l), cur_off);
debug_assert_eq!(self.label_aliases[l.0 as usize], UNKNOWN_LABEL);
}
}
}
/// Current offset from start of buffer.
pub fn cur_offset(&self) -> CodeOffset {
self.data.len() as CodeOffset
@@ -536,7 +480,6 @@ impl<I: VCodeInst> MachBuffer<I> {
// offset and added it to the list (which contains all labels at the
// current offset).
self.check_label_branch_invariants();
self.optimize_branches();
// Post-invariant: by `optimize_branches()` (see argument there).