source: CIVL/examples/experimental/pathfinder2.cvl

/**
 * This file is modified from the original pathfinder_cuda.cvl 
 * program by the following way:
 * First, GPU/GPU_BLOCK/GPU_THREAD no longer access global variables of 
 * int type, but use their value (by parameter passing) instead.
 * Second, the process that spawns GPU_BLOCK/GPU_THREAD
 * doen't reuse itself to run GPU_BLOCK/GPU_THREAD.
 */

#include<civlc.cvh>

//pyramid height of 0 does nothing, of 1 is normal stencil, 2 is where
//the algorithm even kicks in
#define pyramid_height 2

//This is how many iterations of the loop (minus 1). If less than
//the pyramid_height, then the pyramid_height does nothing.
#define rows 3

//I think this is the minimum this can be to be useful
//(the number of cells)
#define cols 2


//BLOCK_SIZE must satisfy (BLOCK_SIZE > pyramid_height*2)
#define BLOCK_SIZE 5
#define HALO 1

int borderCols     = (pyramid_height)*HALO;
int smallBlockCol  = BLOCK_SIZE - (pyramid_height) * HALO * 2;
int blockCols      = cols/smallBlockCol+((cols%smallBlockCol==0)?0:1);

int result[cols];

//GPU MEMORY (couldn't do it scoped because need references to memory
//on host side)
int gpuResult[2][cols];
int gpuWall[rows*cols - cols];

//$input int wall[rows][cols];

#define IN_RANGE(x, min, max)   ((x)>=(min) && (x)<=(max))
#define CLAMP_RANGE(x, min, max) x = (x<(min)) ? min : ((x>(max)) ? max : x )
#define MIN(a, b) ((a)<=(b) ? (a) : (b))

//Not implemented yet.
//void __syncthreads() {
//}

void GPU(int iteration, int src, int dst, int startStep, int blocks, int threads, int gBorderCols) {
                
        void GPU_BLOCK(int bx, int bIteration, int bBorderCols, int src, int dst, int startStep){
            //shared memory
            int prev[BLOCK_SIZE];
            int result[BLOCK_SIZE];
                
            void GPU_THREAD(int tx, int tbx, int tIteration, int tBorderCols, int src, int dst, int startStep, int* prev, int* result){ 
                int small_block_cols = BLOCK_SIZE-tIteration*HALO*2;
                int blkX = small_block_cols*tbx-tBorderCols;
                int blkXmax = blkX+BLOCK_SIZE-1;
                
                // calculate the global thread coordination
                int xidx = blkX+tx;
                
                // effective range within this block that falls within
                // the valid range of the input data
                // used to rule out computation outside the boundary.
                int validXmin = (blkX < 0) ? -blkX : 0;
                int validXmax = (blkXmax > cols-1) ? BLOCK_SIZE-1-(blkXmax-cols+1) : BLOCK_SIZE-1;
                
                int W = tx-1;
                int E = tx+1;
                
                W = (W < validXmin) ? validXmin : W;
                E = (E > validXmax) ? validXmax : E;

                int isValid = IN_RANGE(tx, validXmin, validXmax);    
                if(IN_RANGE(xidx, 0, cols-1)){
                        prev[tx] = gpuResult[src][xidx];
                }

                int computed;
                
                for (int i=0; i<tIteration; i++){
                        computed = 0;
                        if( IN_RANGE(tx, i+1, BLOCK_SIZE-i-2) && isValid) {
                                computed = 1;
                                int left = prev[W];
                                int up = prev[tx];
                                int right = prev[E];
                                int shortest = MIN(left, up);
                                shortest = MIN(shortest, right);
                                int index = cols*(startStep+i)+xidx;
                                result[tx] = shortest + gpuWall[index];
                        }
                        
                        //Break not implemented yet...
                        /* if(i==iteration-1)  */
                        /*   break;  */
                        if(computed != 0)        //Assign the computation range
                                prev[tx]= result[tx];
                        //__syncthreads();         // [Ronny] Added sync to avoid race on prev Aug. 14 201
                }
                
                // update the global memory
                // after the last iteration, only threads coordinated within the
                // small block perform the calculation and switch on ``computed''
                if (computed != 0) {
                        gpuResult[dst][xidx]=result[tx];
                }
            }

            $proc thread_procs[BLOCK_SIZE];
          
            //Launch the threads per block
            for (int tp = 0; tp < BLOCK_SIZE; tp++) {
              thread_procs[tp] = $spawn GPU_THREAD(tp, bx, bIteration, bBorderCols, src, dst, startStep, prev, result);
            }
            
            for (int tp = 0; tp < BLOCK_SIZE; tp++) {
              $wait thread_procs[tp];
            }
        }
        
        $proc block_procs[blocks];
        
        //Launch the blocks
        for (int b = 0; b < blocks; b++) {
                block_procs[b] = $spawn GPU_BLOCK(b, iteration, gBorderCols, src,  dst, startStep);
        }
        
        for (int b = 0; b < blocks; b++) {
                $wait block_procs[b];
        }    
}

void calc_path() {
        int src = 1, dst = 0;
        for (int t = 0; t < rows-1; t+=pyramid_height) {
                int temp = src;
                src = dst;
                dst = temp;
                GPU(MIN(pyramid_height, rows-t-1), src, dst,t, blockCols, BLOCK_SIZE, borderCols);
        }    
}

void main() {
        $assert(BLOCK_SIZE > pyramid_height*2);
        calc_path();
}