Installation

Installation#

aether is a header-only C++23 library: using it from C++ is a CMake find_package(aether CONFIG REQUIRED) against an installed tree, not a package manager step. A sealed Python payload, aether-dsc, ships the SAME headers for Python-only consumers that never touch a C++ toolchain directly — see Python API — aether_dsc.

Requirements#

  • CMake >= 3.20 and a C++23 compiler (GCC >= 12 or Clang >= 16).

  • CUDA mode additionally needs CUDA 12.6 or newer (tested with 12.6 and 13.0); nvcc accepts only host compilers up to its own ceiling (GCC 13 for CUDA 12.6), so point it at one with -DCMAKE_CUDA_HOST_COMPILER=<g++-13> if your default compiler is newer.

  • A GPU is OPTIONAL: build with AETHER_CPP_MODE=ON and the whole library — including every example on this site — runs on the CPU alone through its OpenMP backend.

From source (C++)#

PREFIX below is the install directory (inside a conda environment, use ${CONDA_PREFIX}):

# CUDA mode
cmake -DCMAKE_PREFIX_PATH=${PREFIX} -DAETHER_BUILD_TESTS=ON -B build .
# CPU-only mode (no CUDA toolchain needed)
cmake -DCMAKE_PREFIX_PATH=${PREFIX} -DAETHER_CPP_MODE=ON -DAETHER_BUILD_TESTS=ON -B build .

cmake --build build
cmake --install build --prefix ${PREFIX}

A consuming project then links against the installed package:

find_package(aether CONFIG REQUIRED)
target_link_libraries(your_target PRIVATE aether::aether)

aether-dsc (Python)#

aether-dsc carries a sealed, digest-named copy of the same headers for Python-only consumers — hawk’s NVRTC device-kernel compiler is the primary one, and any file-only host compiler can reach the same headers through Payload.serve(). Nothing in the blob is readable C++ source inside a wheel’s include tree; see Python API — aether_dsc.

# From a source checkout
pip install ./dsc

aether-dsc on PyPI is the sealed header payload that hawk compiles against; it is installed with hawk automatically (pip install raptor-hawk) and is not something C++ users install. aether itself is used through CMake, as above.

CUDA architectures#

AETHER_CUDA_ARCHS picks the architectures when CMAKE_CUDA_ARCHITECTURES is not given:

  • native (the default): the GPUs in this machine only, the fastest build. With no GPU it builds PTX for the oldest architecture the nvcc in use compiles (at least sm_60), which the driver compiles for any newer GPU at load.

  • all: every architecture from Pascal to Blackwell that the nvcc in use still compiles (CUDA 13 drops the ones below Turing).

  • one architecture or a list, such as 86 or "75;86": exactly those.

cmake -DAETHER_CUDA_ARCHS=all -B build .
cmake -DAETHER_CUDA_ARCHS="80;90" -B build .

CMAKE_CUDA_ARCHITECTURES, when given, still wins.

Verify#

Build with AETHER_BUILD_TESTS=ON (either mode above), then run the test-integrity gate directly — never the test binary bare, see tests/check_gate.sh’s own header comment:

tests/check_gate.sh cpp build/tests/aether_tests   # or: cuda