Efficient CSR-Based Sparse Matrix-Vector Multiplication on GPU

Algorithm 3

Improved PCSR.

Input: , , , , ;

Output: ;

()    ;

()    ; ;

()    ;

()    ;

()    ;

()    ;

    //Assemble into shared memory

()    for to with   +=   do

()       ;

()    done

() ();

    //Perform a scalar-style reduction from temp_

() ;

() if then

()    ;

()    ;

()    ;

()    for to do

()      +=  ;

()    done

()    ;

() end