mirror of
https://gitlab.bixilon.de/bixilon/minosoft.git
synced 2025-08-03 12:06:08 -04:00

Since it is on demand only, things got way faster and disabling it is kind of removing a feature by default
74 lines
2.9 KiB
Markdown
74 lines
2.9 KiB
Markdown
# Why Minosoft is faster than Minecraft
|
|
|
|
Yes, it is true. There are a lot of reasons, I want to explain some of them, or at least bring the idea nearer
|
|
|
|
## Kotlin over Java
|
|
|
|
Sounds like magic and it is not magic. If you use kotlin, you use a different style of programming
|
|
and that indeed can make it faster (e.g. `map[key]?.let` vs `if map.containsKey(key)) map.get(key) …`
|
|
Also language features help out here (e.g. removing function call layers with `inline`).
|
|
Delegates can cache values (especially maps) and make observers light (at least from code perspective)
|
|
|
|
## Dirty hacks
|
|
|
|
Sometimes I do/allow dirty hacks (no proper opengl buffer unbinding, disabling bloated features like biome noise).
|
|
Also some stuff just gets implemented half way and some bugs (e.g. transparency) are a side effect, but they are not major.
|
|
|
|
|
|
## Code simplicity
|
|
|
|
Code should be simple. I try to write everything as simple as possible, not like Minecraft.
|
|
Minecraft has a lot of pieces that are not understandable. Normally simple codes is faster.
|
|
|
|
## Profiling
|
|
|
|
From time to time I am profiling minosoft and analyzing the current code sequence. It sometimes is pretty
|
|
obvious why things are slow, sometimes it takes longer, but in either way I am doing changes and then
|
|
I profile them again until it is fast enough.
|
|
|
|
## Mutable objects & memory
|
|
|
|
You can profile memory in two ways:
|
|
|
|
### Usage of memory
|
|
|
|
Sounds for most daus bad, but your system memory has 1 job: getting used. With that principle in mind I can
|
|
cache a lot of things (like 3d biomes) or block states. That often makes things WAY faster.
|
|
I am using as much memory as needed to improve performance.
|
|
|
|
### Allocation rate
|
|
|
|
The higher the allocation rate, the more (useless) memory bandwidth is used and the garbage collector needs to work.
|
|
Despite loving final (i.e. immutable) things, I am reusing objects in all performance critical parts (especially in physics and rendering).
|
|
That makes things a lot faster.
|
|
|
|
## Being lighter
|
|
|
|
Meant in the meaning that a lot of stuff is not yet implemented. Should be a bad thing?
|
|
|
|
## Doing things different
|
|
|
|
Sometimes I am implementing things completely different from minecraft, but the result is the same
|
|
(e.g. beds, signs not as block entity)
|
|
|
|
## Modern opengl
|
|
|
|
Minosoft does everything with shaders and vbos (vertex buffer object).
|
|
Minecraft often still uses legacy opengl (i.e. pushing matrices, ...).
|
|
|
|
## Multithreading
|
|
|
|
Most modern cpus have 8 cores. Minecraft does most important stuff on one thread (one cpu core).
|
|
Minosoft is mostly not aware of "static threads". That means that all operations are executed
|
|
in parallel and timed that almost all operations can make use of modern cpus.
|
|
|
|
# Why it could be slower
|
|
|
|
## Multiple versions
|
|
|
|
Checking stuff for every version is expensive. Only one version is easy :)
|
|
|
|
## PixLyzer
|
|
|
|
All data is dynamic and "liable". That needs to be corrected (e.g. maps over arrays for registry id mapping)
|