Steen Schmidt Posted April 2, 2009 Report Share Posted April 2, 2009 Hi. Since so much focus is on multicore and parallelism today, isn't it time for a true parallel For-loop structure? I know the compiler reputedly has some capability for unrolling unrelated cycles in a For-loop, but I still haven't seen any real-life cases of tangible performance gains from this. I often encounter For-loops with a dynamic number of cycles, but where each cycle is unrelated to the others (initialization of N instruments for instance, where N is only known at runtime). I'd like to see an option in the context menu of a For-loop saying something like "Run in parallel if possible". This would defeat shift register capability of course, among other things. The compiler obviously already has the ability to determine cycle-to-cycle relations. Cheers, Steen Quote Link to comment
Aristos Queue Posted April 2, 2009 Report Share Posted April 2, 2009 It would be reasonable to assume that LV R&D might be working on something like that. Quote Link to comment
shoneill Posted April 2, 2009 Report Share Posted April 2, 2009 QUOTE (Aristos Queue @ Apr 1 2009, 02:17 PM) It would be reasonable to assume that LV R&D might be working on something like that. Can you reply again to this message tomorrow with the same content so that we know it isn't an april fool wind-up? This would be cool! Shane. Quote Link to comment
TobyD Posted April 2, 2009 Report Share Posted April 2, 2009 QUOTE (Steen Schmidt @ Apr 1 2009, 02:30 AM) Since so much focus is on multicore and parallelism today, isn't it time for a true parallel For-loop structure? I know the compiler reputedly has some capability for unrolling unrelated cycles in a For-loop, but I still haven't seen any real-life cases of tangible performance gains from this. I often encounter For-loops with a dynamic number of cycles, but where each cycle is unrelated to the others (initialization of N instruments for instance, where N is only known at runtime). I'd like to see an option in the context menu of a For-loop saying something like "Run in parallel if possible". This would defeat shift register capability of course, among other things. I think now would be a good time to recommend that you sign up for the http://www.ni.com/beta' target="_blank">LabVIEW beta program. :ninja: Quote Link to comment
ragglefrock Posted April 3, 2009 Report Share Posted April 3, 2009 Really? I personally think a serialized for loop is a much cooler feature to have. After all, it's unparalleled. Quote Link to comment
Aristos Queue Posted April 3, 2009 Report Share Posted April 3, 2009 QUOTE (shoneill @ Apr 1 2009, 08:33 AM) Can you reply again to this message tomorrow with the same content so that we know it isn't an april fool wind-up? It would not be unreasonable to assume that we might be working on something like that. Quote Link to comment
shoneill Posted April 3, 2009 Report Share Posted April 3, 2009 QUOTE (Aristos Queue @ Apr 2 2009, 10:27 PM) It would not be unreasonable to assume that we might be working on something like that. Nice of you to tolerate my (essentially tongue-in-cheek) request. :worship: I'll be looking forward to this.... Thanks AQ. Shane. Quote Link to comment
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.