Skip to content

Use dedicated arm64 runners #2

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

Merged
bfierz merged 3 commits into from
Sep 11, 2026
Merged

Use dedicated arm64 runners #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c610f73
Use Windows/Linux Arm64 runners directly
bfierz Sep 11, 2026
eef6502
Add MaxOS AppleSilicon runner
bfierz Sep 11, 2026
398de82
Use explicit list of valid configurations
bfierz Sep 11, 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
117 changes: 68 additions & 49 deletions .github/workflows/build.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,47 +14,62 @@ jobs:
fail-fast: false

matrix:
os: [ubuntu-latest, windows-latest]
build_type: [Release]
c_compiler: [gcc-12, clang-18, cl, emcc]
# One leg per WEAVER_VECTORIZE backend, so every op-codegen change is
# actually compiled (and, where the runner's CPU allows it, run)
# against all of ref/SSE/AVX/AVX2/AVX512 rather than just the SSE 4.2
# default. NEON is not covered here: none of the hosted runners below
# are ARM.
vectorize: ['Generic', 'SSE 4.2', 'AVX', 'AVX 2', 'AVX 512']
# Every valid (os, compiler, vectorize) combination is listed
# directly, rather than built from a cross product pruned down with
# exclude: — each host only supports one arch-appropriate
# compiler/backend pairing (x86 hosts: SSE/AVX/AVX2/AVX512; arm64
# hosts: NEON; only x64 Linux also gets emcc), so a full cross
# product needed far more exclude rules than valid combinations.
# Adding a new host, compiler, or backend is just adding rows here.
include:
- os: windows-latest
c_compiler: cl
- os: ubuntu-latest
c_compiler: gcc-12
cxx_compiler: g++-12
- os: ubuntu-latest
c_compiler: clang-18
cxx_compiler: clang++-18
exclude:
- os: windows-latest
c_compiler: gcc-12
- os: windows-latest
c_compiler: clang-18
- os: windows-latest
c_compiler: emcc
- os: ubuntu-latest
c_compiler: cl
# Emscripten's headers only emulate up to SSE 4.2 over wasm32 SIMD128;
# there is no __m256/__m512 support to target AVX/AVX2/AVX512 with.
- c_compiler: emcc
vectorize: 'AVX'
- c_compiler: emcc
vectorize: 'AVX 2'
- c_compiler: emcc
vectorize: 'AVX 512'
# ubuntu-latest (x64)
- { os: ubuntu-latest, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: Generic, build_type: Release }
- { os: ubuntu-latest, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: 'SSE 4.2', build_type: Release }
- { os: ubuntu-latest, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: AVX, build_type: Release }
- { os: ubuntu-latest, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: 'AVX 2', build_type: Release }
- { os: ubuntu-latest, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: 'AVX 512', build_type: Release }
- { os: ubuntu-latest, c_compiler: clang-18, cxx_compiler: clang++-18, vectorize: Generic, build_type: Release }
- { os: ubuntu-latest, c_compiler: clang-18, cxx_compiler: clang++-18, vectorize: 'SSE 4.2', build_type: Release }
- { os: ubuntu-latest, c_compiler: clang-18, cxx_compiler: clang++-18, vectorize: AVX, build_type: Release }
- { os: ubuntu-latest, c_compiler: clang-18, cxx_compiler: clang++-18, vectorize: 'AVX 2', build_type: Release }
- { os: ubuntu-latest, c_compiler: clang-18, cxx_compiler: clang++-18, vectorize: 'AVX 512', build_type: Release }
# Emscripten's headers only emulate up to SSE 4.2 over wasm32
# SIMD128; no __m256/__m512 to target AVX/AVX2/AVX512 with.
- { os: ubuntu-latest, c_compiler: emcc, vectorize: Generic, build_type: Release }
- { os: ubuntu-latest, c_compiler: emcc, vectorize: 'SSE 4.2', build_type: Release }

# windows-latest (x64)
- { os: windows-latest, c_compiler: cl, vectorize: Generic, build_type: Release }
- { os: windows-latest, c_compiler: cl, vectorize: 'SSE 4.2', build_type: Release }
- { os: windows-latest, c_compiler: cl, vectorize: AVX, build_type: Release }
- { os: windows-latest, c_compiler: cl, vectorize: 'AVX 2', build_type: Release }
- { os: windows-latest, c_compiler: cl, vectorize: 'AVX 512', build_type: Release }

# ubuntu-22.04-arm (native GitHub-hosted arm64 runner). Uses
# clang-15, not clang-18: that's what's actually preinstalled on
# this image, unlike its x64 counterpart.
- { os: ubuntu-22.04-arm, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: Generic, build_type: Release }
- { os: ubuntu-22.04-arm, c_compiler: gcc-12, cxx_compiler: g++-12, vectorize: NEON, build_type: Release }
- { os: ubuntu-22.04-arm, c_compiler: clang-15, cxx_compiler: clang++-15, vectorize: Generic, build_type: Release }
- { os: ubuntu-22.04-arm, c_compiler: clang-15, cxx_compiler: clang++-15, vectorize: NEON, build_type: Release }

# windows-11-arm (native GitHub-hosted arm64 runner)
- { os: windows-11-arm, c_compiler: cl, vectorize: Generic, build_type: Release }
- { os: windows-11-arm, c_compiler: cl, vectorize: NEON, build_type: Release }

# macos-26 (native GitHub-hosted Apple Silicon runner). Xcode's
# clang isn't a versioned llvm.org build like clang-18/clang-15,
# so it gets its own unversioned c_compiler value.
- { os: macos-26, c_compiler: clang, cxx_compiler: clang++, vectorize: Generic, build_type: Release }
- { os: macos-26, c_compiler: clang, cxx_compiler: clang++, vectorize: NEON, build_type: Release }

steps:
- uses: actions/checkout@v3

- name: Setup build toolchain configuration
if: "${{ matrix.c_compiler == 'gcc-12' || matrix.c_compiler == 'clang-18' }}"
# Every non-MSVC, non-emcc leg (gcc-12/clang-18/clang) sets cxx_compiler
# via its include entry above; cl and emcc don't need CC/CXX overridden.
if: "${{ matrix.cxx_compiler != '' }}"
run: |
echo "CC=${{ matrix.c_compiler }}" >> $GITHUB_ENV
echo "CXX=${{ matrix.cxx_compiler }}" >> $GITHUB_ENV
Expand Down Expand Up @@ -91,8 +106,12 @@ jobs:
uses: actions/cache@v3
with:
path: ~/.conan2
key: conan2-${{ runner.os }}-${{ matrix.c_compiler }}-${{ hashFiles('**/conanfile.*') }}
restore-keys: conan2-${{ runner.os }}-${{ matrix.c_compiler }}-
# Keyed on matrix.os rather than runner.os: ubuntu-latest and
# ubuntu-22.04-arm both report runner.os Linux but build
# architecture-specific binaries, so runner.os alone would collide
# x64 and arm64 caches under the same gcc-12/clang-18 compiler name.
key: conan2-${{ matrix.os }}-${{ matrix.c_compiler }}-${{ hashFiles('**/conanfile.*') }}
restore-keys: conan2-${{ matrix.os }}-${{ matrix.c_compiler }}-

- name: Detect default Conan profile
if: steps.conan-cache.outputs.cache-hit != 'true'
Expand All @@ -119,16 +138,18 @@ jobs:
# Build your program with the given configuration. Note that --config is needed because the default Windows generator is a multi-config generator (Visual Studio generator).
run: cmake --build ${{ steps.strings.outputs.build-output-dir }} --config ${{ matrix.build_type }}

- name: Test (Linux)
if: "${{ runner.os == 'Linux' && matrix.c_compiler != 'emcc' }}"
- name: Test (Linux/macOS)
if: "${{ (runner.os == 'Linux' || runner.os == 'macOS') && matrix.c_compiler != 'emcc' }}"
working-directory: ${{ steps.strings.outputs.build-output-dir }}
# Execute tests defined by the CMake configuration. Note that --build-config is needed because the default Windows generator is a multi-config generator (Visual Studio generator).
# See https://cmake.org/cmake/help/latest/manual/ctest.1.html for more detail
#
# AVX-512 is special-cased: GitHub-hosted runners don't guarantee AVX-512
# CPU support even though the build above always compiles it (that's what
# catches cross-backend interface bugs). If this runner's CPU lacks it,
# skip running rather than fail the job on an unrelated SIGILL.
# skip running rather than fail the job on an unrelated SIGILL. Only
# reachable on Linux (macOS legs are NEON-only, see the vectorize/os
# excludes above), so checking /proc/cpuinfo unconditionally is safe.
run: |
if [ "${{ matrix.vectorize }}" = "AVX 512" ]; then
if ! grep -qE '\bavx512f\b' /proc/cpuinfo || ! grep -qE '\bavx512vl\b' /proc/cpuinfo || ! grep -qE '\bavx512dq\b' /proc/cpuinfo; then
Expand Down Expand Up @@ -161,11 +182,13 @@ jobs:
# See https://cmake.org/cmake/help/latest/manual/ctest.1.html for more detail
run: node weaver_tests.js

# No GitHub-hosted runner is ARM, so the NEON backend is cross-compiled here
# on an x86 runner and its tests are run through QEMU user-mode emulation
# instead. CMAKE_CROSSCOMPILING_EMULATOR (set explicitly below, since Conan
# doesn't know about QEMU) makes both `cmake --build` and `ctest` transparently
# exec test binaries as e.g. `qemu-arm -L /usr/arm-linux-gnueabihf <binary>`.
# GitHub-hosted arm64 runners (ubuntu-22.04-arm, windows-11-arm) cover
# aarch64 natively in the main build matrix above, but there's no 32-bit ARM
# hosted runner, so armv7hf is still cross-compiled here on an x86 runner
# and its tests run through QEMU user-mode emulation instead.
# CMAKE_CROSSCOMPILING_EMULATOR (set explicitly below, since Conan doesn't
# know about QEMU) makes both `cmake --build` and `ctest` transparently exec
# test binaries as e.g. `qemu-arm -L /usr/arm-linux-gnueabihf <binary>`.
build-arm:
runs-on: ubuntu-latest

Expand All @@ -178,10 +201,6 @@ jobs:
triple: arm-linux-gnueabihf
qemu: qemu-arm
profile: cmake-conan/arm_profile
- arch: aarch64
triple: aarch64-linux-gnu
qemu: qemu-aarch64
profile: cmake-conan/aarch64_profile

steps:
- uses: actions/checkout@v3
Expand Down
11 changes: 0 additions & 11 deletions cmake-conan/aarch64_profile
View file
Open in desktop

This file was deleted.

Loading