Skip to content

fix(scx_lavd): Fix combinatorial state explosion in energy model generation causing absurd memory usage #3548

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

Open
BHLuotianyi wants to merge 2 commits into
base: main
Choose a base branch
from
+114 −119
Open

fix(scx_lavd): Fix combinatorial state explosion in energy model generation causing absurd memory usage #3548

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8f075c0
scx_lavd: avoid exponential EM CPU order search
BHLuotianyi Apr 30, 2026
4242ee5
scx_lavd: Fix exponential state explosion in DP energy model generation
BHLuotianyi Apr 30, 2026
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
233 changes: 114 additions & 119 deletions scheds/rust/scx_lavd/src/cpu_order.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,6 @@
// GNU General Public License version 2.

use anyhow::Result;
use combinations::Combinations;
use itertools::iproduct;
use scx_utils::CoreType;
use scx_utils::Cpumask;
Expand Down Expand Up @@ -471,12 +470,6 @@ struct EnergyModelOptimizer<'a> {
perf_cpu_order: RefCell<BTreeMap<usize, PerfCpuOrder>>,
}

#[derive(Debug, Clone, Eq, Hash, Ord, PartialOrd)]
struct PDS<'a> {
pd: &'a PerfDomain,
ps: &'a PerfState,
}

#[derive(Debug, Clone, Eq, Hash, Ord, PartialOrd)]
struct PDCpu<'a> {
pd: &'a PerfDomain, // performance domain
Expand Down Expand Up @@ -941,22 +934,80 @@ impl<'a> EnergyModelOptimizer<'a> {
}

fn gen_pds_combinations(&'a self, util: f32) -> Vec<PDSetInfo<'a>> {
let mut pdsi_vec = Vec::new();

let pds_set = self.gen_pds_set(util);
let n = pds_set.len();
for k in 1..n {
let pdss = pds_set.clone();
let pds_cmbs: Vec<_> = Combinations::new(pdss, k)
.map(|cmb| PDSetInfo::new(cmb.clone()))
.collect();
pdsi_vec.extend(pds_cmbs);
/*
* Build unique performance-domain / CPU-count combinations.
*
* The previous implementation expanded a performance domain with N
* CPUs into N identical PDS entries and then generated every subset of
* the expanded vector. That turns a 20 CPU system into almost 2^20
* intermediate combinations per utilization point, most of which
* collapse to the same "use K CPUs from this domain" result.
*
* Keep the same model, but generate those count choices directly and
* do dynamic programming across domains. For the same accumulated
* performance, a lower-power partial result always dominates a
* higher-power one because later domain choices add the same
* performance and power regardless of the earlier path.
*/
let mut states: BTreeMap<usize, HashSet<PDSetInfo<'a>>> = BTreeMap::new();
Self::insert_best_pdsi(&mut states, PDSetInfo::empty());

for (_, pd) in self.em.perf_doms.iter() {
let ps = pd.select_perf_state(util).unwrap();
let mut next: BTreeMap<usize, HashSet<PDSetInfo<'a>>> = BTreeMap::new();

for pdsi_set in states.values() {
for pdsi in pdsi_set {
for nr_cpus in 0..=pd.span.weight() {
let choice = PDSetInfo::from_pd_cpu_count(pd, ps, nr_cpus);
Self::insert_best_pdsi(&mut next, pdsi.combine(&choice));
}
}
}

states = next;
}

let pdsi = PDSetInfo::new(pds_set.clone());
pdsi_vec.push(pdsi);
states
.into_values()
.flat_map(|pdsi_set| pdsi_set.into_iter())
.filter(|pdsi| !pdsi.pdcpu_set.is_empty())
.collect()
}

pdsi_vec
fn insert_best_pdsi(
pdsi_map: &mut BTreeMap<usize, HashSet<PDSetInfo<'a>>>,
new_pdsi: PDSetInfo<'a>,
) -> bool {
match pdsi_map.get_mut(&new_pdsi.performance) {
Some(v) => {
let (pdsi_power, pdsi_pd_len) = {
let pdsi = v.iter().next().unwrap();
(pdsi.power, pdsi.pd_id_set.len())
};
if pdsi_power == new_pdsi.power {
if new_pdsi.pd_id_set.len() < pdsi_pd_len {
v.clear();
v.insert(new_pdsi);
true
} else {
false
}
} else if pdsi_power > new_pdsi.power {
v.clear();
v.insert(new_pdsi);
true
} else {
false
}
}
None => {
let mut v = HashSet::new();
v.insert(new_pdsi.clone());
pdsi_map.insert(new_pdsi.performance, v);
true
}
}
}

fn insert_pds_combinations(&self, new_pdsi_vec: &Vec<PDSetInfo<'a>>) -> bool {
Expand All @@ -973,13 +1024,19 @@ impl<'a> EnergyModelOptimizer<'a> {
// There are already PDSetInfo in the list.
Some(v) => {
let mut v = v.borrow_mut();
let pdsi = &v.iter().next().unwrap();
if pdsi.power == new_pdsi.power {
// If the power consumptions are the same, keep both.
if v.insert(new_pdsi.clone()) {
let (pdsi_power, pdsi_pd_len) = {
let pdsi = v.iter().next().unwrap();
(pdsi.power, pdsi.pd_id_set.len())
};
if pdsi_power == new_pdsi.power {
// If the power consumptions are the same, keep the one with fewer PDs
// to minimize leakage power and improve locality.
if new_pdsi.pd_id_set.len() < pdsi_pd_len {
v.clear();
v.insert(new_pdsi.clone());
found_new = true;
}
} else if pdsi.power > new_pdsi.power {
} else if pdsi_power > new_pdsi.power {
// If the new one takes less power, keep the new one.
v.clear();
v.insert(new_pdsi.clone());
Expand All @@ -998,50 +1055,6 @@ impl<'a> EnergyModelOptimizer<'a> {
}
found_new
}

/// Get a vector of (performance domain, performance state) to achieve
/// the given CPU utilization, @util.
fn gen_pds_set(&self, util: f32) -> Vec<PDS<'_>> {
let mut pds_set = vec![];
for (_, pd) in self.em.perf_doms.iter() {
let ps = pd.select_perf_state(util).unwrap();
let pds = PDS::new(pd, ps);
pds_set.push(pds);
}
self.expand_pds_set(&mut pds_set);
pds_set
}

/// Expand a PDS vector such that a performance domain with X CPUs
/// has N elements in the vector. This is purely for generating
/// combinations easy.
fn expand_pds_set(&self, pds_set: &mut Vec<PDS<'_>>) {
let mut xset = vec![];
// For a performance domain having nr_cpus, add nr_cpus-1 more
// PDS to make the PDS nr_cpus in the vector.
for pds in pds_set.iter() {
let nr_cpus = pds.pd.span.weight();
for _ in 1..nr_cpus {
xset.push(pds.clone());
}
}
pds_set.append(&mut xset);

// Sort the pds_set for easy comparison.
pds_set.sort();
}
}

impl<'a> PDS<'_> {
fn new(pd: &'a PerfDomain, ps: &'a PerfState) -> PDS<'a> {
PDS { pd, ps }
}
}

impl PartialEq for PDS<'_> {
fn eq(&self, other: &Self) -> bool {
self.pd == other.pd && self.ps == other.ps
}
}

impl<'a> PDCpu<'_> {
Expand All @@ -1056,64 +1069,46 @@ impl PartialEq for PDCpu<'_> {
}
}

impl fmt::Display for PDS<'_> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
write!(
f,
"pd:id:{}/pd:weight:{}/ps:cap:{}/ps:power:{}",
self.pd.id,
self.pd.span.weight(),
self.ps.performance,
self.ps.power,
)?;
Ok(())
impl<'a> PDSetInfo<'a> {
fn empty() -> PDSetInfo<'a> {
PDSetInfo {
performance: 0,
power: 0,
pdcpu_set: BTreeSet::new(),
pd_id_set: BTreeSet::new(),
}
}
}

impl<'a> PDSetInfo<'_> {
fn new(pds_set: Vec<PDS<'a>>) -> PDSetInfo<'a> {
// Create a pd_id_set and calculate performance and power.
let mut performance = 0;
let mut power = 0;
let mut pd_id_set: BTreeSet<usize> = BTreeSet::new();

for pds in pds_set.iter() {
performance += pds.ps.performance;
power += pds.ps.power;
pd_id_set.insert(pds.pd.id);
}
fn from_pd_cpu_count(pd: &'a PerfDomain, ps: &'a PerfState, nr_cpus: usize) -> PDSetInfo<'a> {
let mut pdcpu_set = BTreeSet::new();
let mut pd_id_set = BTreeSet::new();

// Create a pdcpu_set, so first gather the same PDS entries.
let mut pds_map: BTreeMap<PDS<'a>, RefCell<Vec<PDS<'a>>>> = BTreeMap::new();
for cpu_vid in 0..nr_cpus {
pdcpu_set.insert(PDCpu::new(pd, cpu_vid));
}

for pds in pds_set.iter() {
let v = pds_map.get(&pds);
match v {
Some(v) => {
let mut v = v.borrow_mut();
v.push(pds.clone());
}
None => {
let mut v: Vec<PDS<'a>> = Vec::new();
v.push(pds.clone());
pds_map.insert(pds.clone(), v.into());
}
}
if nr_cpus > 0 {
pd_id_set.insert(pd.id);
}
// Then assign cpu virtual ids to pdcpu_set.
let mut pdcpu_set: BTreeSet<PDCpu<'a>> = BTreeSet::new();
let pds_map = pds_map;

for (_, v) in pds_map.iter() {
for (cpu_vid, pds) in v.borrow().iter().enumerate() {
let pdcpu = PDCpu::new(pds.pd, cpu_vid);
pdcpu_set.insert(pdcpu);
}

PDSetInfo {
performance: ps.performance * nr_cpus,
power: ps.power * nr_cpus,
pdcpu_set,
pd_id_set,
}
}

fn combine(&self, other: &PDSetInfo<'a>) -> PDSetInfo<'a> {
let mut pdcpu_set = self.pdcpu_set.clone();
pdcpu_set.extend(other.pdcpu_set.iter().cloned());

let mut pd_id_set = self.pd_id_set.clone();
pd_id_set.extend(other.pd_id_set.iter().cloned());

PDSetInfo {
performance,
power,
performance: self.performance + other.performance,
power: self.power + other.power,
pdcpu_set,
pd_id_set,
}
Expand Down
Loading