this case works on nvidia but not with a larger mesh

lbm_cl.py

@@ -25,8 +25,8 @@ _n = 4 * _m
 _ivplot = 0
 
 # grid size
-_nx = 256
-_ny = 256
+_nx = 1024
+_ny = 1024
 
 _Lx = 1
 _Ly = 1
@@ -43,7 +43,7 @@ _Tmax = 2.
 def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
               Tmax = _Tmax,
               Lx = _Lx, Ly = _Ly,
-              animate = False,
+              animate = True,
               interactive=True,
               precision="single"):
 
@@ -94,8 +94,7 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
     mf = cl.mem_flags
 
     # compile OpenCL C program
-    prg = cl.Program(ctx, source).build(options = "-cl-strict-aliasing  \
-                                                  -cl-fast-relaxed-math")
+    prg = cl.Program(ctx, source).build(options = "")
 
     # create OpenCL buffers
     fn_gpu   = cl.Buffer(ctx, mf.READ_WRITE,
@@ -111,6 +110,9 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
     event = prg.init_sol(queue, (nx * ny, ), (32, ), fn_gpu)
     event.wait()
 
+    event = prg.init_sol(queue, (nx * ny, ), (32, ), fnp1_gpu)
+    event.wait()
+
     # number of animation frames
     nbplots = 100
     itermax = int(np.floor(Tmax / dt))
@@ -130,7 +132,6 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
     print("start OpenCL computations...")
     while t < Tmax:
         t = t + dt
-        iter = iter + 1
         #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, ), wn_gpu, wnp1_gpu, wait_for = [event])
@@ -153,6 +154,8 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
         else:
             print(title, end='\r')
 
+        iter = iter + 1
+
     # copy OpenCL data to CPU and return the results
     cl.enqueue_copy(queue, fn_cpu, fn_gpu).wait()
 
@@ -162,7 +165,7 @@ def solve_ocl(m = _m, n = _n, nx = _nx, ny = _ny,
 
 if __name__ == '__main__':
     # gpu solve
-    wplot_gpu = solve_ocl(animate=False)
+    wplot_gpu = solve_ocl()
     #print(np.sum(wplot_gpu[:, _ivplot, :, :],axis=0))
     plt.clf()
     #plt.imshow(np.sum(wplot_gpu[:, _ivplot, :, :],axis=0), vmin=0, vmax=1)

lbm_kernels.cl

 
-#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+//#pragma OPENCL EXTENSION cl_khr_fp64 : enable
 
 #define real _real_