OpenCL advection-remesh operator test segfaults on the Intel OpenCL platform

Testing 1D DirectionalAdvection_Euler_L2_1: inplace=True dtype=float32 shape=(10,), cfl=0.62 
 >Implementation PYTHON: ....
 >Implementation OPENCL_CODEGEN:
Kernel <directional_inplace_remesh_1d__lambda_2_1__float__2p__2s__full> was successfully vectorized (8)
...
Segmentation fault

added bug label

added critical label

Possibly linked to issue #6 (closed).

Invalid reads and writes confirmed by oclgrind:

Invalid read of size 4 at global memory address 0x1000000000030
	Kernel: directional_cached_advection_Euler_f1p_1g
	Entity: Global(0,0,0) Local(0,0,0) Group(0,0,0)
	  %25 = load float, float addrspace(1)* %arrayidx, align 4, !dbg !149
	At line 184 (column 43) of input.cl:
	  const float tmp = (float)(src0[GID]);

changed milestone to %Working local advection-remesh-transpose on all devices.

closed via commit 897ea3a1

reopened

removed critical label

The issue is still there for nparticles>1.

This should be fixed by now, reopen the issue if needed.

closed