More than 16?

Am wondering, what ports have pushed past the 16 floating point number thing? Any that push to 24, 32 or even 64 out there?

You could redefine fixed_t to be a 64-bit int. Then you'd get 48.16 precision.

You should make sure that angle_t remains 32-bit because it relies on overflows. Also, when reading data structures (from WADs and stuff) you'd still need to use 32-bit ints.

If it was upped to 32 or past 32, wouldn't that mean rewriting how all the important tidbits (level structure) would be handled under that?

What benefits would 64 bit integers have for the Doom engine if it was done?

Mr. Chris said:

What benefits would 64 bit integers have for the Doom engine if it was done?

You could make larger levels than 32768 units wide.

Not exactly clear which floating point this is about.

DoomLegacy is mostly 32 bit fixed point.

There is a 64 bit binary (see alpha5 downloads) that uses 64 bit integers. Because of this many data structures use types from stdint (uint32_t etc). Reading the wad and external interfaces were critical, but if did not affect most of the calculations. A few times a simple type change would generate artifacts (usually due to needing a sign,
where it was not obvious one was needed).

The OpenGL rendering uses float, with double for calculations.
Of course our OpenGL has some gaps which might be improved by using double instead.

Still have trouble with a software render overflow somewhere in the code on large maps, like longdays.

Doomsday uses 64-bit double-precision floating point and the old 32768 map size limit (and others) no longer applies. Our node builder uses this representation natively, also, meaning there is no quantization or "map unit snapping" going on. In libdeng2 we don't use fixed point math at all, for anything.