Goal is to reduce OpenCL kernel recompilations.
Currently viewport renders are still set to use 64 closures as this seems to be faster and we don't want to cause a performance regression there. Needs to be investigated.
Details
Details
- Reviewers
Sergey Sharybin (sergey) Brecht Van Lommel (brecht) - Commits
- rBS087331c495b0: Cycles: Replace __MAX_CLOSURE__ build option with runtime integrator variable
rCfb39e26c242e: Replace __MAX_CLOSURE__ build option with runtime integrator variable
rB087331c495b0: Cycles: Replace __MAX_CLOSURE__ build option with runtime integrator variable
Diff Detail
Diff Detail
Event Timeline
Comment Actions
Seems generally fine. But it's quite odd that only the viewport would have its performance affected? Are we sure the regular render isn't affected as well?
| intern/cycles/kernel/split/kernel_split_data.h | ||
|---|---|---|
| 35 | Maybe use this to make the lines a little shorter in this and the next function: uint64_t closure_size = sizeof(ShaderClosure) * (kernel_data.integrator.max_closures-1); | |
Comment Actions
Final render is actually a few percent faster with this, I don't have an explanation.
Comment Actions
Looks good to me then.
Also suggesting a small optimization that wasn't possible before.
| intern/cycles/kernel/split/kernel_branched.h | ||
|---|---|---|
| 37–39 | Could we replace kernel_data.integrator.max_closures with branched_state->sd.num_closure here and in the function below, to make it a little faster? | |