Sequential - Start

• cblas not available?
  > sudo apt-get install libatlas-dev libatlas-base-dev
• GNU compliers (optimization)
• PGI compilers by PGI, see trial version.
• Intel compilers: list, non-commercial, useful options and pragmas,

Profiling:

• GCC_:           make cache
  > g++  -g *.cpp
> g++ *.o  -o main.GCC_                   # [no '-pg' in link command]
  > valgrind --tool=callgrind --simulate-cache=yes ./main.GCC_
> kcachegrind `ls -1tr  callgrind.out.* |tail -1`
• ICC_:            make cache
  > icpc -g *.cpp
> icpc *.o -o main.ICC_
> amplxe-gui &
  • error: "set /proc/sys/kernel/yama/ptrace_scope to 0", see help.
    > echo "0"|sudo tee /proc/sys/kernel/yama/ptrace_scope
    after each reboot or once
    edit file /etc/sysctl.d/10-ptrace.conf and set variable kernel.yama.ptrace_scope variable to 0 .
• PGI_:           make cache
• do not use option  -pg  in compile flags as well as in link flags
• Trace your heap:
  > heaptrack ./main.GCC_
  > heaptrack_gui heaptrack.main.GCC_.<pid>.gz

Original code for inner product:

double scalar(const int N, const double x[], const double y[])
{
 double sum = 0.0;
 for (int i=0; i<N; ++i)
 {
    sum += x[i]*y[i];
 }
 return sum;
}

int main()
{
  ...
  double s = scalar(n,a,b);
  ... 
}

• g++ -O2 skalar.cpp
          Preprocessor symbol: __GNUC__
• icpc -O2  skalar.cpp
          Preprocessor symbol: __INTEL_COMPILER
• pgc++ -fast skalar.cpp
          Preprocessor symbol: __PGI
• clang++ -O3 skalar.cpp 
          Preprocessor symbol: __clang__

Performance improvements:

• loop unrolling (Compiler Opt., H2)
• tiling (H1)
• restrict keyword (H1, H2):    __restrict__ ,
• prefetching (H1):  __builtin_prefetch ,
• data alignment (H1, H2):
• SIMD and Intel Advisor: Parallel Universe Issue 22, 2015
• processor specific options, see also output of  more /proc/cpuinfo