up

lbm_cl.py

@@ -25,8 +25,8 @@ _n = 4 * _m
 _ivplot = 0
 
 # grid size
-_nx = 1024
-_ny = 1024
+_nx = 256
+_ny = 256
 
 _Lx = 1
 _Ly = 1
@@ -113,6 +113,7 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
     event = prg.init_sol(queue, (nx * ny, ), (32, ), fnp1_gpu)
     event.wait()
 
+    queue.finish()
     # number of animation frames
     nbplots = 100
     itermax = int(np.floor(Tmax / dt))
@@ -133,9 +134,10 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
     while t < Tmax:
         t = t + dt
         #event = prg.time_step(queue, (nx * ny, ), (32, ), wn_gpu, wnp1_gpu)
-        event = prg.time_step(queue, (nx * ny, ), (64, ), fn_gpu, fnp1_gpu)
+        event = prg.time_step(queue, (nx * ny, ), (32, ), fn_gpu, fnp1_gpu)
         #event = prg.time_step(queue, (nx * ny, ), (32, ), wn_gpu, wnp1_gpu, wait_for = [event])
         event.wait()
+        queue.finish()
         elapsed += 1e-9 * (event.profile.end - event.profile.start)
         # exchange buffer references for avoiding a copy
         fn_gpu, fnp1_gpu = fnp1_gpu, fn_gpu

lbm_kernels.cl

@@ -26,6 +26,7 @@ __constant int dir[4][2] = { {-1, 0}, {1, 0},
 __constant real ds[4] = { _DY, _DY, _DX, _DX };
 
 
+
 // physical flux of the hyperbolic system
 /* void flux_phy(const real *w, const real* vnorm, real* flux){ */
 /*   real vdotn = _VX * vnorm[0] + _VY * vnorm[1]; */
@@ -74,7 +75,7 @@ void flux_phy(const real *W, const real *vn, real *flux) {
 void w2f(const real *w, real *f){
   for(int d = 0; d < 4; d++){
     real flux[_M];
-    real vnorm[2] = {dir[d][0], dir[d][1]};
+    real vnorm[2] = {(real) dir[d][0], (real) dir[d][1]};
     flux_phy(w, vnorm, flux);   
     for(int iv = 0; iv < _M; iv++){
       f[d * _M + iv] = w[iv] / 4 + flux[iv] / 2 / _LAMBDA;