[C++] Possible to reduce object code generated in compute/kernels/take.cc?

According to

https://gist.github.com/wesm/90f73d050a81cbff6772aea2203cdf93

take.cc is our largest piece of object code in the codebase. This is a pretty important function but I wonder if it's possible to make the implementation "leaner" than it is currently to reduce generated code, without sacrificing performance.