config.yml

# Options are nvidia, hip_nvidia, and hip_amd. Options are defined at https://latticeqcd.github.io/SIMULATeQCD/01_gettingStarted/gettingStarted.html#
# nvidia - Builds the source just using CUDA. See https://latticeqcd.github.io/SIMULATeQCD/01_gettingStarted/gettingStarted.html#building-source-with-cuda
# Build the source with HIP for Nvidia Platforms. See https://latticeqcd.github.io/SIMULATeQCD/01_gettingStarted/gettingStarted.html#building-source-with-hip-for-nvidia-platforms-experimental
# Build the source with HIP for AMD Platforms. See https://latticeqcd.github.io/SIMULATeQCD/01_gettingStarted/gettingStarted.html#building-source-with-hip-for-amd-platforms-experimental
# Note: This automatically sets the CMake variable BACKEND.
# TODO - ONLY nvidia IS IMPLEMENTED. AMD IS STILL A WORK IN PROGRESS.
PROFILE: nvidia

# This controls the number of threads that the build will spawn. It is used as
# follows, `make -j ${CORES}` If left undefined, it will use all cores available
# on the system.
CORES: 

# USE_GPU_AWARE_MPI boolean option that enables or disables the use of GPU-aware MPI libraries when
# building an MPI-enabled application with CUDA. When USE_GPU_AWARE_MPI is set to ON, CMake will
# attempt to find and link against a GPU-aware MPI library, which is a version of the MPI library
# that is optimized for communication between CPUs and GPUs. This can result in better performance
# when running MPI applications that use CUDA/HIP. See https://developer.nvidia.com/blog/introduction-cuda-aware-mpi/
# Options are ON or OFF
USE_GPU_AWARE_MPI: "ON"

# This sets the version of the OS *in the container*. This does not need to match
# the target computer. The primary reason for changing this is to allow the use
# of older verisons of the CUDA toolkit. Older toolkits are only available on
# older operating systems. This can be ignored when using the hip_amd profile.
# You will generally want to leave this at latest. This tells the code to grab
# the latest version available. You can also set it to RHEL's major version number.
# Ex: 7, 8, 9.
RHEL_VERSION: latest

# This is the CUDA version you want to use for the build. See https://developer.nvidia.com/cuda-toolkit-archive
# for a list of versions. If you set it to latest, the code will check that link for the highest
# version and use it.
CUDA_VERSION: latest

# USE_GPU_P2P enables or disables the use of GPU peer-to-peer (P2P) communication when building an
# application with CUDA. Nvidia's proprietary GPU interconnect is called NVLink. GPU P2P communication
# allows for direct data transfer between two GPUs without the need to transfer the data to the host
# system first. This can result in faster data transfer and reduced CPU overhead, especially for
# multi-GPU systems. Options are ON or OFF
USE_GPU_P2P: "ON"

# Enable or disable the use of the HIP programming language for AMD GPUs. Options are ON or OFF
# WARNING: This is not currently defined and does not do anything!
# TODO - THIS IS NOT YET IMPLEMENTED
USE_HIP_AMD: "OFF" 

################################################################################################
#                                          For CUDA                                            #
################################################################################################
# This is the CUDA architecture for which you want to generate device code. 
# It is defined here: https://cmake.org/cmake/help/latest/prop_tgt/CUDA_ARCHITECTURES.html.
# Each CUDA Toolkit has a list of supported architectures. This post is
# helpful: https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
# See https://en.wikipedia.org/wiki/CUDA#GPUs_supported for a list of compatibilities
# Current list:
# Fermi: 20
# Kepler: 30, 35, 37
# Maxwell: 50, 52, 53
# Pascal: 60, 61, 62
# Volta: 70, 72
# Turing: 75
# Ampere: 80, 86, 87
# Ada: 89
# Hopper: 90
################################################################################################
#                                          For AMD                                             #
################################################################################################
# AMD Vega 10: gfx900
# AMD Vega 7nm: gfx906
# AMD Instinct® MI100 accelerator: gfx908
# Aldebaran: gfx90a
# Navi10: gfx1010
# Navi12: gfx1011
# Navi14: gfx1012
# Sienna Cichlid: gfx1030
# Navy Flounder: gfx1031
# See https://rocmdocs.amd.com/projects/ROCgdb/en/hybrid/gdb/doc/gdb/AMD-GPU.html
ARCHITECTURE: "70"

# The build target. If you specify nothing, it will build all targets. You can get a list of possible build
# targets by running `./simulateqcd list`
TARGET:

# These are added to the end of the cmake line. Any other options you would like to add to cmake you can add here
ADDITIONAL_CMAKE_OPTIONS: 

# These are added to the end of the make line. Any other options you would like to add to make you can add here
ADDITIONAL_MAKE_OPTIONS:

# The directory you want the build to be copied to after the build completes. It will create a folder called build
# in this directory.
OUTPUT_DIRECTORY: ./