Skip Navigation
Search
Vectorization Flags
To enable vectorization it is important to set specific flags for the compiler. There are also flags for causing the compiler to generate messages on what it did vectorize. Here the most important flags are given.
Cray |
Arm |
GNU |
Nvidia |
||
| Flags |
|||||
| Pre-23 CPE |
CPE 23 and later: (not applicable for Fortran) |
default | default | default | |
| Optimization | -O3 | -O3 |
-O3 or -Ofast | -O3 or -Ofast | -fast (check documentation) |
| Vectorization | -h vector3 | Automatic (if -O3 or -O2 flag is set) | -mcpu=a64fx -armpl | -mcpu=a64fx | -tp=native (on a64fx node) |
| Vectorization report | -h msgs | -Rpass=loop-vectorize | -Rpass=loop-vectorize | -fopt-info-vec | -Minfo=vect |
| Report on missed optimization | -h negmsgs | -Rpass-analysis=loop-vectorize | -Rpass-analysis=loop-vectorize | -fopt-info-vec-missed | -Mneginfo |
| OpenMP | -h omp | -fopenmp | -fopenmp | -fopenmp |
-mp -acc for OpenACC |
| Debugging | -G 2 | -ggdb |
-ggdb | -ggdb | -g or -gopt (see documentation) |
| Large memory | -h pic | -mcmodel=large | -mcmodel=large | -mcmodel=large | -mcmodel=large |
| Module | CPE/version | CPE/23.02(or newer) |
arm-modules/version | gcc/version | nvidia/nvhpc-nompi/version |
| Compiler names | |||||
| C | cc |
armclang | gcc | nvc | |
| C++ | CC |
armclang++ | g++ | nvc++ | |
| Fortran | ftn |
armflang | gfortran | nvfortran | |