concurrent write access in OpenCL kernel

[guernika]

hello to all,

I was giving a look at the OpenCL soft-body solver SolvePositions.cl,
and I noticed the last two lines:

g_vertexPositions[node0] = (float4)(position0, 0.f);
g_vertexPositions[node1] = (float4)(position1, 0.f);

Am I wrong or they could give a concurrent write access?
Or is OpenCL preventing this?

Best regards

Francesco

[Erwin Coumans]

The kernel will be processed over many work items in parallel, writing to different g_vertexPositions.

The batching makes sure that there are no work items that share the same node.
Thanks,
Erwin

[guernika]

Hello and thanks for your reply,

you're right...I didn't notice the generateBatches method

Regards

Francesco