All this advice against "premature optimization" has created generations of programmers that don't understand how to use hardware efficiently.
Here's the problem: If you profile software that is 100x slower than it needs to be on every level, there are no obvious bottlenecks. Your whole program is just slow across the board, because you used tons of allocations, abstractions and indirections at every step of the way.
Rob Pike probably has never written a program where performance really mattered, because if he did, he would've found that you need to think about performance right from the beginning and all the way during development, because making bad decisions early on can force you to rewrite almost everything.
For instance, if you start writing a Go program with the mindset that you can just heap-allocate all the time, the garbage collector will eventually come back to bite you and the "bottleneck" will be your entire codebase.
Here's the problem: If you profile software that is 100x slower than it needs to be on every level, there are no obvious bottlenecks. Your whole program is just slow across the board, because you used tons of allocations, abstractions and indirections at every step of the way.
Rob Pike probably has never written a program where performance really mattered, because if he did, he would've found that you need to think about performance right from the beginning and all the way during development, because making bad decisions early on can force you to rewrite almost everything.
For instance, if you start writing a Go program with the mindset that you can just heap-allocate all the time, the garbage collector will eventually come back to bite you and the "bottleneck" will be your entire codebase.