#538: Enable CSE in McLachlan -----------------------------------+---------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Keywords: | -----------------------------------+---------------------------------------- Kranc supports CSE, which has the potential to reduce code size and improve performance. We should enable it in McLachlan.
#538: Enable CSE in McLachlan ------------------------------------+--------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Resolution: | Keywords: ------------------------------------+---------------------------------------
Comment (by knarf):
What out of http://en.wikipedia.org/wiki/CSE is the CSE you mean? :)
#538: Enable CSE in McLachlan ------------------------------------+--------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Resolution: | Keywords: ------------------------------------+---------------------------------------
Comment (by eschnett):
The one activated by the UseCSE->True flag, http://en.wikipedia.org/wiki/Common_subexpression_elimination.
#538: Enable CSE in McLachlan ------------------------------------+--------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Resolution: | Keywords: ------------------------------------+---------------------------------------
Comment (by hinder):
Does CSE lead to improved performance on some systems? I believe Barry found that it gave improvements for !WeylScal4 but not for !McLachlan, but I might be misremembering.
#538: Enable CSE in McLachlan ------------------------------------+--------------------------------------- Reporter: eschnett | Owner: Type: enhancement | Status: new Priority: minor | Milestone: Component: EinsteinToolkit thorn | Version: Resolution: | Keywords: ------------------------------------+---------------------------------------
Comment (by barry.wardell):
Yes, when I was testing it, I found that it actually slowed !McLachlan down a bit, but did give a speedup for !WeylScal4. This was before a lot of the recent speed improvements in !McLachlan so it would be worth checking again. I don't think we should just blindly enable it, though.
The benefits of CSE might also depend on e.g. whether Jacobians are enabled or not, how aggressive the CSE is (this can be tuned), etc.
trac@lists.einsteintoolkit.org