OpenCL Performance Evaluation on Modern Multicore CPUs

Algorithm 1

Vectorization on OpenCL versus OpenMP. The equivalent code in OpenCL is vectorizable while OpenMP code is not vectorizable.

/*OpenMP computation that doesn’t vectorize due to dependencies.*/

int main(){

 for (int  ; ; ++){

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

 }

/*Similar OpenCL kernel computation which vectorizes.*/

void VectorAdd (…, _ _global float  ^*dm_src, _ _global float  ^*dm_dst){

 for (int  ; ; ++){

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

  FMUL(_, _)

 }