Compilers — Magus02 HPC

Latest revision as of 06:40, 20 April 2026

📍 Magus02 Knowledge Base → HPC User → Compilers

A compiler translates human-readable source code (C, C++, Fortran) into executable binary programs. Choosing the right compiler and flags can significantly affect application performance on Magus02.

General

Compilers are loaded via the module system. Never invoke compilers directly without loading the appropriate module first.

A basic compilation workflow:

# 1. Load the compiler module
$ module load intel/oneapi

# 2. Compile your source code
$ icc hello_world.c -o hello_world

# 3. Run the resulting binary
$ ./hello_world
Hello, World!

Optimization Flags

Always compile with optimization enabled for production runs. Common optimization levels:

Flag	Effect
`-O0`	No optimization. Fastest compile; use when debugging.
`-O1`	Basic optimizations.
`-O2`	Recommended default for production — good balance of speed and compile time.
`-O3`	Aggressive optimization, including vectorization. May change floating-point behavior.
`-march=native`	Optimize specifically for the current CPU architecture (AMD EPYC / Intel Xeon on Magus02).

Intel Compiler (icc / ifort)

The Intel compiler is optimized for Intel and compatible AMD microprocessors and often produces the fastest binaries for numerical codes.

Load the module:

$ module load intel/oneapi

Compile C, C++, or Fortran:

$ icc   source.c   -O2 -o my_program      # C
$ icpc  source.cpp -O2 -o my_program      # C++
$ ifort source.f90 -O2 -o my_program      # Fortran 90

GNU Compiler Collection (gcc / gfortran)

GCC is a free, open-source compiler available on all platforms. Reliable and widely tested.

Load the module:

$ module load gcc/11.2.0

Compile:

$ gcc      source.c   -O2 -o my_program   # C
$ g++      source.cpp -O2 -o my_program   # C++
$ gfortran source.f90 -O2 -o my_program   # Fortran

LLVM / Clang

The LLVM project provides modern, modular compiler infrastructure. Useful for C/C++ and when targeting specific optimization analyses.

$ module load llvm
$ clang   source.c   -O2 -o my_program   # C
$ clang++ source.cpp -O2 -o my_program   # C++

PGI / NVIDIA Compilers

PGI compilers (now NVIDIA HPC SDK) are particularly useful for GPU programming with OpenACC pragmas.

$ module load nvhpc
$ pgcc   source.c   -O2 -o my_program
$ pgfort source.f90 -O2 -o my_program

Compiling Multi-file Projects

For programs split across multiple source files, use a linker or a build system:

# Compile with multiple source files:
$ icc file1.c file2.c file3.c -O2 -o my_program

# Or use make (for larger projects with a Makefile):
$ module load intel/oneapi
$ make

References

↑ Back to Knowledge Base · Maintained by HPC Administration · wiki.snu.edu.in