Hacktoberfest 2026: los issues que los mantenedores marcaron para octubre, abiertos y aptos para principiantes. Explorar issues de Hacktoberfest

linux-aarch64 CPython ships without PGO (cross-compiled build skips --enable-optimizations); interpreter-bound code measurably slower on Graviton

Abierto
  • ¿Algo no está bien?
#955 1 comentario 0 reacciones 0 asignados Ver en GitHub

Los mantenedores suelen responder en 1 día

Nadie ha tomado este issue todavía.

Evaluación

Dificultad
4/5
Tiempo estimado
3-5 días
Aptitud para principiantes
38/100
Tipo de issue
Nueva funcionalidad
Claridad
Bastante claro
Estado de actividad
Activo
Stack tecnológico
aws, linux, python, shell
Área
build-system, developer-experience, performance

Línea de trabajo

Start with recipe/build_base.sh around the conditional that adds --with-lto and --enable-optimizations, and inspect how _OPTIMIZED and CONDA_BUILD_CROSS_COMPILATION are set. Trace the generated CONFIG_ARGS for native arm64 and cross-compiled builds, then evaluate native arm64 CI or an emulated profile step. Done means identifying a supported PGO/LTO build path and verifying the resulting configure arguments and interpreter performance.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

Observation. sysconfig.get_config_var("CONFIG_ARGS") for conda-forge CPython 3.12.13, linux-aarch64 (built Mar 5 2026, GCC 14.3.0) contains --build=x86_64-conda-linux-gnu --host=aarch64-conda-linux-gnu and neither --enable-optimizations nor --with-lto. CPython 3.12.11, linux-64 (Jun 4 2025) has --enable-optimizations --with-lto=full.

Why. recipe/build_base.sh on main, around L281–285:

if [[ ${_OPTIMIZED} == yes ]]; then
  _common_configure_args+=(--with-lto=full)
  if [[ "$CONDA_BUILD_CROSS_COMPILATION" != "1" ]]; then
    _common_configure_args+=(--enable-optimizations)

linux-aarch64 is cross-compiled, so PGO is always skipped there. I haven't traced why --with-lto is also missing from the 3.12.13 aarch64 build's CONFIG_ARGS.

Measured impact (confounded, stated as such). We ran the same pure-Python inner loop (MAGeCK's count: gzip line reading plus dict lookups) on the same bytes on AWS, in each platform's conda-forge interpreter:

Graviton4 (aarch64, no PGO) Intel Xeon 6975P-C (x86, PGO+LTO) AMD EPYC 9R45 (x86, PGO+LTO)
gzip lines/s 2.56 M 3.85 M 5.02 M
guide lookups/s 0.60 M 1.19 M 1.60 M

This mixes core differences, a patch-version difference, and the build difference, so it doesn't isolate PGO's share. (PGO+LTO speedups on interpreter-bound code are commonly cited as 10–20%; I haven't measured that here.) On Graviton, arm64 has no published PGO build to compare against at all, since it's skipped by construction.

Question. Is a native (non-cross) linux-aarch64 build with PGO feasible now that native arm64 CI runners exist, or is there a way to run the profile step under emulation? Every Python-heavy bioinformatics or science tool on linux-aarch64 inherits this.

Found while verifying real workloads on AWS for scientific-codes-cookbook (measurements/mageck-real).

Lenguaje dominante
Shell
Estrellas
47
Forks
115
Merge medio
5 d 11 h
PR fusionados (30 d)
19

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de conda-forge/python-feedstock

Todos los issues de conda-forge/python-feedstock

Issues similares

Más issues de Shell/Bash

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.