C Language Features Used in This Course

This documents summarize the C language features frequently employed in this course.

Compiler

gcc

We choose the gcc C language compiler from GCC (GNU Compiler Collection) as our official compiler. This is the default C compiler on Linux (including the Windows subsystem of Linux, WSL). Use other compilers at your own risk.

Warning

The gcc provided with the MinGW package under Windows is not totally compatible to the gcc we use. Use it at your own risk.

Warning

The gcc from the XCode Command-line tool on Mac OS is an alias of clang, which is the LLVM C compiler. It has some known differences with the OpenMP for gcc covered in class.

Compilation Command

A typical gcc compilation command looks like:

gcc <flags> <arguments>

  • <flags> are the compilation flags

  • <arguments> are the source files (.c) and libraries (.o)

Useful Compilation Flags

  • -O specifies the compilation optimization level

    • -O0 reduced optimization, default level, use as baseline

    • -O2

    • -O3 recommended

    • -Ofast with non-standard-compliant optimizations

  • -I specifies the path to include headers

    • No space between -I and the path!

    • -I. means to look for headers in the current directory

  • -l specifies the libraries to use

    • -lpthread POSIX thread library

  • -f specifies flags

    • -fpic Position dependent

    • -fopenmp enable OpenMP support

  • -m specifies the architecture

    • -march=native use the native architecture

    • -m64 use 64-bit architecture

    • -msse4.2 use SSE4.2 instruction set

  • -std specifies the C language standard

    • -std=gnu90

    • -std=gnu11

    • -std=c99 -pedantic

    • -std=c11 -pedantic

Performance Optimization in C Language

  • Compilation optimization

    • -O2

    • -O3

    • -Ofast

    • -ffast-math

  • Selected Optimization Approaches

    • Loop unrolling

    • Cache awareness

      • Optimize the memory access pattern for better cache performance

      • Loop interchange

      • Loop tiling

    • Vectorization (hardware architecture dependent)

      • Intel MMX, SSE, AVX

  • Memory operations

    Block memory operations are usually faster than element-wise operations. The strategy is to use block memory operations whenever possible. The following functions are frequently used in this course:

    • Memory allocation/deallocation

      • malloc and free

      • calloc and free

      • aligned_alloc and free (C11)

    • Memory block copy/set

      • memcpy

      • memset