1
0
Fork 0

feat: add scx_lavd scheduler
Some checks failed
/ Build onc-kde:main-ver4a (push) Has been cancelled
/ Build onc-kde:main (push) Has been cancelled
/ Build onc-kde:main-nvidia (push) Has been cancelled
/ Build onc-kde:main-nvidia-ver4a (push) Has been cancelled

It seems to be the most energy efficient on my lunar lake laptop, having
the lowest low-utilization power draw
(scx_lavd intel_pstate active > intel pstate active > EAS intel pstate
passive), but there are some odd performance results, where only E-cores are used for
single-threaded workloads like "zstd -b1" and "pv /dev/urandom >
/dev/null". From my initial testing it seems to be way more convervative
with using P-cores for these specific workloads, and only uses them if
most of E-cores get overloaded, unlike EAS that engages P-cores when one
E-core is utilized above 80%.
This commit is contained in:
ver4a 2026-01-25 16:24:29 +01:00
parent e57fbda898
commit cadbd05c4a
2 changed files with 14 additions and 0 deletions

View file

@ -9,10 +9,23 @@ COPY ./mpv_options /build/mpv-build/mpv_options
RUN cd /build/mpv-build && ./rebuild -j$(nproc)
FROM quay.io/fedora-ostree-desktops/kinoite:43 AS scx-build
RUN dnf5 -y install --setopt install_weak_deps=false cargo git-core elfutils-libelf-devel clang rust-libseccomp-devel
RUN mkdir /build && cd /build && git clone https://github.com/sched-ext/scx
RUN cargo build --release -p scx_lavd
FROM quay.io/fedora-ostree-desktops/kinoite:43
COPY --from=mpv-build /build/mpv-build/mpv/build/mpv /bin/mpv
COPY --from=scx-build /build/scx/target/release/scx_lavd /usr/bin/scx_lavd
COPY --from=scx-build /build/scx/services/scx.service /usr/lib/systemd/system/scx.service
# dnf configuration has to be in effect during build
COPY etc/dnf /etc/dnf

View file

@ -0,0 +1 @@
SCX_SCHEDULER=scx_lavd