Efficient CSR-Based Sparse Matrix-Vector Multiplication on GPU

Algorithm 4

Main procedure of generating row blocks.

Input: , , , ;

Output: , ;

()    ; ; ; ;

()    for to do

      //Compute non-zeros and the total rows

()        +=  ;

()      ++;

()      if ∥

       (  &&  ) then

       //This row fills up SHARED_SIZE or threads per block

()       ; ++; ; ;

()      else if then

       //This row is an extra one that is excluded

()       ; ++; ; ; −−;

()      end

() done

      //Extra case

() if   !=   then

()    ;

() else

()    −−;

() end