CaLRS: A Critical-Aware Shared LLC Request Scheduling Algorithm on GPGPU

Algorithm 1

CaLRS scheduling algorithm.

queue[5] contain 5 LLC priority queue

cur_mreq is current memory request

prio stands for a memory request’s priority

prioMap[5] is the mapping table for queue priority rotation

CF-prio[33] =

       [−1,  0,  1,  1,  1,  2,  2,  2,  2,  3,  3,  3,  3,  3,  3,  3,  3,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4]

       is the CF to priority mapping table

Procedure  CaLRS(cur_mreq, queue[5])

     Insertion:

         prio = CF-prio[cur_mreq->CF]; //change CF to priority

         while  queue[prioMap[prio]].full() && prio  <  4

                prio++;

         end while

         if  prio  =  4

         Set_LLC_Queue(STALL);

         else

               queue[prioMap[prio]].push_back(cur_mreq);

     Issue:

         int = 0;

         while  queue[prioMap[i]].empty()

                 ++;

         if    <  6

                cur_mreq = queue[prioMap[]].pop_front();

                ISSUE(cur_mreq);

     Rotation:

         if  queue[prioMap[]].empty()

                for  int     =  0;   <  5; ++

                        prioMap[] = (prioMap[]  +  4)%5;

                if  Get_LLC_Queue() = STALL

                        Set_LLC_Queue(UNSTALL);