[LLVMdev] some superoptimizer results (original) (raw)
John Regehr regehr at cs.utah.edu
Wed Jul 22 21:14:42 PDT 2015
- Previous message: [LLVMdev] some superoptimizer results
- Next message: [LLVMdev] some superoptimizer results
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Are you taking into account critical path length?
No, but of course that't not hard. There are a lot of things that could be taken into account in the cost functions, but I was hoping to keep it really simple, so that the results would not be opaque.
case all 3 instructions are dependent. So the former case can execute in 2 cycles while the latter takes 3. Modern OoO chips do in fact exploit this kind of thing.
Sure. One way to deal with this would be to use the HW as its own model: run a lot of code fragments on real chips under a benchmark harness and then use the results to train a cost model.
John
- Previous message: [LLVMdev] some superoptimizer results
- Next message: [LLVMdev] some superoptimizer results
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]