refactor: replace byte-level operations with word-level operations #692
Conversation
… cryptographic functions
|
@newpavlov please lmk if this is a step in the right direction at least |
|
is there any way to benchmark such that I can get stats for individual functions, so I can see where my bottleneck is, without wrapping every function inside a timing block |
|
cargo flamegraph is a common tool for this. You also could use |
There was a problem hiding this comment.
Looks like a good step forward even without noticeable performance improvements.
which I think is because we're still having to split the values up when doing the row rotation and substitution operations
Yeah, you should probably minimize number of conversion to and from bytes and operate over 64-bit integers whenever possible.
| @@ -147,4 +137,4 @@ impl SerializableState for KupynaLongVarCore { | |||
| } | |||
|
|
|||
| #[cfg(feature = "zeroize")] | |||
| impl ZeroizeOnDrop for KupynaLongVarCore {} | |||
| impl ZeroizeOnDrop for KupynaLongVarCore {} | |||
There was a problem hiding this comment.
Please keep the empty last line in all files.
|
|
||
| // println!("prev vector:="); | ||
| // for v in prev_vector.iter() { | ||
| // println!("{:016X?}", v); |
| for nu in 0..ROUNDS { | ||
| state = add_constant_plus(state, nu as usize); | ||
| state = apply_s_box(state); | ||
| state = rotate_rows(state); | ||
| state = mix_columns(state); | ||
| } | ||
| matrix_to_block(state) | ||
| state |
There was a problem hiding this comment.
It may be worth to change add_constant_plus, apply_s_box, rotate_rows, and mix_columns to work in-place, i.e. over &mut [u64; COLS].
| @@ -14,6 +14,7 @@ rust-version = "1.85" | |||
|
|
|||
| [dependencies] | |||
| digest = "=0.11.0-pre.10" | |||
| hex-literal = "1" | |||
There was a problem hiding this comment.
I don't think you need this dependency.
There was a problem hiding this comment.
I've set up a main.rs file to verify the code works as expected while making changes to the functions, this is the reason for both the hex-literal dependency and the print statements. will clean them all up once I'm done making large changes to the functions, it just helps to pinpoint error faster if some math goes wrong
# Conflicts: # kupyna/src/long.rs # kupyna/src/long_compress.rs # kupyna/src/short.rs # kupyna/src/short_compress.rs
finally got to this, I'm sorry for being so late ;-;
pre-refactor bench results:
post-refactor bench results:
not seeing any increase yet, which I think is because we're still having to split the values up when doing the row rotation and substitution operations. need to look into the math of that so I can handle it, will probably require a refactor in how our sbox and mds tables are stored. still, bulk of the work seems done