For rendering, it is useful to have normal vectors. One option is to create another compute shader to recalculate the normal vectors after the positions are updated. For example, we might execute the preceding compute shader 1,000 times, dispatch the other compute shader once to update the normals, and then render the cloth.

Additionally, we may be able to achieve better performance with the use of local shared data within the work group. In the preceding implementation, the position of each particle is read a maximum of eight times. Each read can be costly in terms of execution time. It is faster to read from memory that is closer to the GPU. One way to achieve this is to read data into local shared memory once, and then read from the shared memory for subsequent reads. In the next recipe, we'll see an example of how this is done. It would be straightforward to update this recipe in a similar way.