... | ... | @@ -60,7 +60,7 @@ Here is what I do: |
|
|
|
|
|
- In both build trees, edit `build.mk` as described above. That makes the libraries generate ticky-ticky information.
|
|
|
|
|
|
- Run `nofib` in both trees and use `nofib-analyse` to compare, as described in [Building/RunningNoFib](building/running-no-fib).
|
|
|
- Run `nofib` in both trees and use `nofib-analyse` to compare, as described in [Building/RunningNoFib](building/running-nofib).
|
|
|
|
|
|
- Find one where the performance changes in the wrong direction, say `x2n1`. It is often best, at least at first, to use allocation as a proxy for execution time; allocation has the huge merit of being repeatable and measurable at a fine grain.
|
|
|
|
... | ... | |