These benchmarks serve as justifications for some of my choices in some small pieces of code throughout my library.

Unless someone can find more efficient ways to perform these actions, I will not accept modifications to these bits of code.
And especially not if "mUh CleAN coDe IS BeTTer" is waived around.

Test Machine:

Windows 11 22H2
Ryzen 5 3600
16 GB RAM @ 2133 MHz

More test machines will be added once the hashmaps are done.
There should be Win32 and Linux x64, and ARM64 Linux.

Test Conditions:

Using RELEASE target.
Test is ran 2 times before to prevent any caching and prefetch BS.
Ran through a non-admin CLI.

Programming Conventions:

Assembly-level optimisations can be done but aren't prioritised.
Intrinsics aren't allowed in generic tests.
- They are often CPU/architecture dependant.
- They should be done in their own parallel test.
We're using a trash value to prevent zero-code optimisations.
- Slow down the tests by one bitwise AND instruction for each iteration.
- Prevents having cumbersome compiler directives and linker options to prevent it.
Non-removal of zero-code is checked through a quick look in Ghidra.
Optimisations through CPU ISA extensions isn't done yet.
- Requires more complex setup.
- Currently only interested in results that apply to the library.

CRC32 Implementations

Compares the performances of multiple CRC32B variants on a 512 MiB block of random data.

Variant	Time (s)	Hash rate (MiB/s)	Hash rate (Mib/s)
Original - Branchless	5.84	87.701268	701.610140
Original - Condition	3.43	149.358226	1194.865811
Reduced - Branchless	5.90	86.750254	694.002033
Reduced - Look-up	1.08	474.513438	3796.107507

Source Code (bm_crc32_implementations.c)

Calculating powers of 2

Compares the performances between the 2 following pieces of code:

pow(2, exponent)
2 << (exponent - 1)

Results:

Iterations: 100.000.000
Bitwise: 1421 ms
Pow(): 3780 ms

Remarks:

Pow takes 266.01% of the time Bitwise takes.
Bitwise takes 37.59% of the time Pow takes.
Results, and bitwise method, cannot be applied to powers of any number who isn't itself a power of 2.
Can't they actually...

Limitations:

Cannot be done with exponent == n if n is our integer bit size/width.
The exponent cannot be negative.

Source Code (bm_powers_of_two.c)