On my implementation, when writing to the same ssbo locations from multiple workgroups of one
glDispatchCompute command via
atomicAdd, it seems to work correctly (results match my expectations) and i don’t use the
I would interpret in this case that these operations happen within “one stage” and therefore it works and also matches the spec.
It would be very kind and helpful to confirm this behaviour to be in line with the spec (also, enforced by the spec!), as i also have my difficulties to correctly interpret the specification documents.
Thank you very much,