Cell model implementations optimised for execution on modern CPUs

Source code for the implementations discussed in the paper "Resource-efficient use of modern processor architectures for numerically solving cardiac ionic cell models" by Kristian Gregorius Hustad and Xing Cai.

Dependencies

• CMake
• C and C++ compilers (see compiler-specific details below)

Building

build_dir="build"
rm -rf ${build_dir}
mkdir -p ${build_dir}
pushd ${build_dir}

cmake ../src
make -j8 # replace 8 by the number of parallel build jobs you want to run

popd

Fujitsu compiler

CMake 3.21 or newer is required for proper support of the Fujitsu compiler in Clang mode.

Controlling which compiler is used with CMake

• ARMclang: export CC=armclang CXX=armclang++
• Clang: export CC=clang CXX=clang++
• Fujitsu (Clang mode): export CC="fcc -Nclang" CXX="FCC -Nclang"
• GCC: export CC=gcc CXX=g++
• Intel: export CC=icc CXX=icpc

SIMD hints

We use ifdefs to determine how SIMD hints should be provided.

For Clang-based compilers (i.e. bog-standard clang, ArmClang, Fujitsu compiler in Clang mode), HINT_CLANG_SIMD should be enabled.

For other compilers (e.g. GCC, Intel), HINT_OMP_SIMD should be enabled. (Note that HINT_CLANG_SIMD is checked before HINT_OMP_SIMD, so there is no harm in leaving HINT_OMP_SIMD enabled when using a Clang-based compiler.) When using HINT_OMP_SIMD, -DVECTOR_LENGTH=8 can be passed to CMake to extend the hint with an additional clause simdlen(VECTOR_LENGTH). This is needed to ensure AVX-512 is used on Cascade Lake and Skylake (but not KNL) with the Intel compiler.

Target-specific options

Only one of the following options should be set.

• TARGET_CASCADE_LAKE: For the Cascade Lake CPUs (found on Oakbridge)
• TARGET_KNL: For the Knights Landing CPU (found on Oakforest)
• TARGET_SKYLAKE_256: For the Skylake client CPUs (without AVX-512)
• TARGET_SKYLAKE_512: For the Skylake (and later *lake) server CPUs (with AVX-512)

If none of these options are set, the compiler is set to target the native CPU architecture (if supported).

Note: The compiler used will affect which optimisation flags are passed by CMake.

Other options

• USE_EXPM1: Use the expm1() function to compute expressions on the form (exp(x) - 1).
• USE_FAST_MATH: Pass flags to the compiler that increase math performance, potentially at the cost of a slight loss of accuracy.
• USE_LTO: Use link-time optimisation (LTO) / interprocedural optimisation (IPO)
• VECTOR_LENGTH: Override the vector length. (It is safe to set this to a multiple of the hardware vector length.)

Running

We have provided scripts to run the same benchmarks that were used to generate the data listed in the following tables:

• Table 3: jobs/run_math_bench.sh
• Table 6: jobs/run_simd_singlethread.sh
• Table 7: jobs/run_simd_multithread.sh
• Table 8: jobs/run_ensemble.sh
• Table 10: jobs/run_TP06_lut.sh

These scripts take the build directory as an argument. Example usage:

bash jobs/run_math_TP06_lut.sh build