record.txt

module load papi
mkdir build
cd build/
module load cmake
module load intel-compiler
cmake ..
make -j 64

// 以下是用作分析用的 最好是结合intel-compiler，intel-compiler 会有个分析说哪些地方没有vectorized
// The opengl_main is the one with visualization whilst kernel_main just prints the energies.
module load intel-advisor
advixe-cl --collect survey --project-dir ./adv_results --search-dir src:=../ --search-dir bin:=./ -- ./kernel_vect_omp
advixe-gui &

// roofline 分析
qsub -I -P c07 -q normal -l walltime=00:15:00,mem=190GB,jobfs=100GB,ncpus=48
module load intel-advisor
export ADVIXE_EXPERIMENTAL=int_roofline
// 这里如果有openmp用就加上
export OMP_NUM_THREADS=<num_threads>
advixe-cl --collect survey --project-dir ./adv_results --search-dir src:=../ --search-dir bin:=./ -- ./kernel_opt
advixe-cl --collect tripcounts -flop -enable-cache-simulation --project-dir ./adv_results --search-dir src:=../ --search-dir bin:=./ -- ./kernel_opt
advixe-gui &

sqrt root pow symmetric() 会比较花费时间

time ./kernel_main -n 1000
time ./kernel_opt -n 1000
time ./kernel_sse -n 1000
time ./kernel_vect_omp -n 1000
time ./kernel_omp -n 1000