On 09/26/2013 02:31 PM, Paul Berry wrote: [snip] > However, there are two things which I expected to see in this patch > series but I didn't: > > - Do atomic_uints need similar handling to > ir_sampler_replacement_visitor in opt_function_inlining? > > - Do we need to update schedule_node::set_latency_gen7() with an > estimate of how long it takes to execute an atomic operation? Currently > it looks like it's assuming 14 cycles, which seems almost certainly wrong.
If I recall correctly, atomic operations are considered barriers for the purposes of scheduling, so I don't think their latency really matters. It would be nice to add it eventually (either Matt or Eric should have the shader_time tricks needed to determine how many cycles it takes). _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
