NWChem TCE CCSD(T) loop-driven kernels for performance optimization experiments
make TARGET=? -j8
Set TARGET to one of the following:
- CRAY-INTEL
- CRAY-GCC
- CRAY (Cray compilers)
- TUKEY-INTEL
- TUKEY-GCC
- KAYLA (GCC on ARM)
- BGQ (BGXLC)
- POWER7-AIX
- POWER7 (XLC)
- POWER7-GCC
- MAC-CLANG
- MAC (GCC)
- INTEL
- MIC (Intel)
Or just define your own new TARGET using the existing ones as a template.
Jeffs-MacBook-Pro:src jhammond$ OMP_NUM_THREADS=8 ./test_ccsd_t.x 16
testing NWChem CCSD(T) kernels on 8 threads with tilesize 16
MEMCPY gigabytes/s of your processor is 5.540682
DAXPY GF/s of your processor is 1.307957
DGER GF/s of your processor is 2.109818
DGEMM (k=t) GF/s of your processor is 20.871256
DGEMM (k=t^2) GF/s of your processor is 43.129097
randomized initialization took 0.001193 s
STARTING FORTRAN OPENMP KERNELS
0: sd_t_s1_1 time = 0.045883 s GF/s = 0.731305
0: sd_t_s1_2 time = 0.016226 s GF/s = 2.067936
0: sd_t_s1_3 time = 0.015528 s GF/s = 2.160904
0: sd_t_s1_4 time = 0.018180 s GF/s = 1.845689
0: sd_t_s1_5 time = 0.017175 s GF/s = 1.953683
0: sd_t_s1_6 time = 0.015956 s GF/s = 2.102945
0: sd_t_s1_7 time = 0.016046 s GF/s = 2.091134
0: sd_t_s1_8 time = 0.017487 s GF/s = 1.918816
0: sd_t_s1_9 time = 0.016471 s GF/s = 2.037194
...
Just read the first few lines of main() to see what arguments,
if any, these programs take. The Fortran tests have hard-coded
dimensions that must be edited by hand prior to compilation.