use either llvm/clang or fujitsu with -Nclang option

[jdomke]

LLVM19 with threadpool is fastest; closely followed by FJ with -Nclang; OMP_WAIT_POLICY=ACTIVE gives some perf; only FJ trad mode with omp seems very "bad"

clang.fjomp.log
omp_thread len 32768, time 0.034168
clang.llvmomp.log
omp_thread len 32768, time 0.034435
clang.pool.log
threadpool len 32768, time 0.016248
fcc.nclang.omp.log
omp_thread len 32768, time 0.038099
fcc.nclang.pool.log
threadpool len 32768, time 0.017050
fcc.trad.omp.log
omp_thread len 32768, time 0.109991
fcc.trad.pool.log
threadpool len 32768, time 0.055810
result_omp.log
omp_thread len 32768, time 0.108256
result_threadpool.log
threadpool len 32768, time 0.056695