cuda-flags.file

### FILE cuda-flags
# define the CUDA compilation flags in a way that can be shared by SCRAM-based and regular tools

# on X86 and Power, build support for Pascal, Volta and Turing
%ifarch x86_64 ppc64le
%define cuda_arch 60 70 75
%endif

# on ARM, build support for Volta, Xavier SoC and Turing
%ifarch aarch64
%define cuda_arch 70 72 75
%endif


# LLVM style for listing the supported CUDA compute architectures
%define llvm_cuda_arch %(echo $(for ARCH in %cuda_arch; do echo "$ARCH"; done) | sed -e"s/ /,/g")

# C++ standard to use for building host and device code with nvcc
%define nvcc_stdcxx -std=c++17

# generate optimised code
%define cuda_flags_0 -O3

# generate debugging information for device code
%define cuda_flags_1 --generate-line-info --source-in-ptx --display-error-number

# imply __host__, __device__ attributes in constexpr functions
%define cuda_flags_2 --expt-relaxed-constexpr

# allow __host__, __device__ attributes in lambda declaration
%define cuda_flags_3 --extended-lambda

# build support for the various compute architectures
%define cuda_flags_4 %(echo $(for ARCH in %cuda_arch; do echo "-gencode arch=compute_$ARCH,code=sm_$ARCH"; done)) -Wno-deprecated-gpu-targets

# disable warnings about attributes on defaulted methods
%define cuda_flags_5 -Xcudafe --diag_suppress=esa_on_defaulted_function_ignored

# link the CUDA runtime shared library
%define cuda_flags_6 --cudart shared

# collect all CUDA flags
%define nvcc_cuda_flags %{nvcc_stdcxx} %{cuda_flags_0} %{cuda_flags_1} %{cuda_flags_2} %{cuda_flags_3} %{cuda_flags_4} %{cuda_flags_5} %{cuda_flags_6}