Link hash table primes externally to prevent data duplication in binary

[Listwon]

I've started analyzing Godot binaries on Windows using SizeBench and on top of the duplicate data results were hash_table_size_primes and hash_table_size_primes_inv.

As visible above the tool shows around 168KB of wasted memory (tested on commit 36e943b scons target=template_release debug_symbols=yes)

This PR reduces build size by 547KB ( scons target=template_release).

File scoped const arrays are linked internally which means each time hashfuncs.h are included, those two arrays are not merged into single data point (which would be fine and expected for const, read only data). I've moved their initialization to hashfuncs.cpp and marked extern in hashfuncs.h. Thanks to @Malcolmnixon suggestion I've made it simpler by using inline constexpr from C++17 standard.

I've not benchmarked it yet. In theory it can improve the CPU cache, because all of the hash_maps and hash_sets reuse the data from the same address in memory (meaning - there is higher chance that it will be still present in cache when needed).

[fire]

I don't see how this functionally changes the code, so it seems like an decent optimization.

[Malcolmnixon]

That's a good catch.

Since C++17 is now supported and the guidelines promote using constexpr, you might want to see if keeping the table in the header and marking them constexpr (and possibly inline) works as well. In theory this should also merge multiple instances of the data across different translation units, but also expose the values as constants to the compiler allowing constant-folding if accessed discretely.

[Listwon]

I think I tried constexpr already and it didn't change anything in this case, but I'll try inline constexpr. As you said, in theory it should work the same as the solution with extern.

Edit: I've just confirmed it - constexpr gives the same output as const. inline constexpr reduces the size - it merges the data.

[Malcolmnixon]

This looks good. Nice change!

[akien-mga]

Looks great! Thanks for the initiative looking into optimizing binary size and memory usage, that's very welcome :)

[akien-mga]

Thanks!