Efficient CSR-Based Sparse Matrix-Vector Multiplication on GPU

Algorithm 5

Multiple scalar-style reduction.

()    ;

()    ;

    //Perform a multiple scalar-style reduction from temp_

()    ;

()    ;   & ();

()    if then

     //Perform a partial reduction from temp_

()     ;

()     ;

()     ;

()     for to with   +=   do

()     +=  ;

()  done

()  ;

()  ();

     //Perform a warp reduction from bVAL_s

()  if   &&  

()    += ;

()   ();

()  if   &&    >=  16

()    += ; ();

()  if   &&    >= 8

()    += ; ();

()  if   &&  

()    += ; ();

()  if   &&  

()   ;

() end