Le mardi 07 janvier 2014 à 12:28 -0500, Stefan Karpinski a écrit : > You actually can when you use a tuple because the number of elements > in a tuple is part of its type, so type inference can figure out how > many elements a tuple has and eliminate the overhead. In this case, > you're splatting an array and it's length is not something that can be > inferred, so lots of work has to be done at run time to do the > dispatch. This could maybe be optimized but I really think it's better > to write the code differently. > No, I'm using a tuple in the first table() function (in the branch where usena=false, which is the one I tested for this comparison). The el tuple is built using ntuple(n, ...), where n is the length of the varargs. Couldn't this be optimized more efficiently?
Else, what's the recommended pattern? Computing the linear index as I do in table3()? Regards
