https://github.com/redodo/ranger
Rust solution of bloomon challenge using SIMD
https://github.com/redodo/ranger
Last synced: over 1 year ago
JSON representation
Rust solution of bloomon challenge using SIMD
- Host: GitHub
- URL: https://github.com/redodo/ranger
- Owner: redodo
- Created: 2023-01-21T12:04:21.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2023-02-05T22:44:49.000Z (over 3 years ago)
- Last Synced: 2025-01-30T23:16:07.455Z (over 1 year ago)
- Language: Rust
- Homepage:
- Size: 1.37 MB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Rust solution of bloomon challenge using SIMD
This solution uses 512-bit SIMD vectors to store and compare stem stock to stem
requirements per design. The vectors that are used have 32 lanes of `u16` values. This
allows for every possible species (a to z) to have a maximum of 65535 of stock.
The feasibility of a design can be checked in a couple of operations, regardless of the
amount of stems in a design:
1. Take the maximum amount of stems we can take: `stems = simd_min(stock, design.max_stems)`
2. Check if we have enough stems: `reduce_sum(stems) >= design.total`
3. Check if we at least one of each needed stem: `all(simd_ge(stems, design.min_stems))`
After this we have enough stems. However, since we took the maximum amount possible per
stem, we need to make sure that we return the excess amount of stems we took. This is
done over a simple iteration where stems are returned to the stock until the amount
required by the design is met.
Besides this a number of other optimizations are implemented:
- On input of a stem, only designs with that stem are checked.
- Total stems in stock is separately kept, and designs are skipped when the total stems
in stock does not meet the design requirements.
- The implicit minimum of a species in a design is increased from 1 where possible. For
example, *AL5a5* requires at least 5 times *a*, and *BL5a1b5* requires at least 4
times *a*.
- The maximum species in a design is decreased where possible. For example, *AL10a5*
can not have more than 5 times *a*, and *BL5a1b5* can not have more than 4 times *a*.
- The maximum possible stems per species in any design is used to skip bouquet creation
if a stem is input that already exceeds the stock by this amount.
## How to run it
Make a release build:
```bash
cargo build --release
```
Run it:
```bash
target/release/ranger
```
## Comparison to other solutions
*These benchmarks were run on an AMD Ryzen 7 1700.*
[redodo/cstem](https://github.com/redodo/cstem) (C):
```bash
$ multitime -n100 -q -s0 -i "cat samples/1m.txt" ./cstem
===> multitime results
1: -i "cat samples/1m.txt" -q ./cstem
Mean Std.Dev. Min Median Max
real 0.097 0.012 0.079 0.091 0.116
user 0.095 0.012 0.075 0.090 0.116
sys 0.002 0.002 0.000 0.000 0.007
```
redodo/ranger (Rust):
```bash
$ multitime -n100 -q -s0 -i "cat samples/1m.txt" ./ranger
===> multitime results
1: -i "cat samples/1m.txt" -q ./ranger
Mean Std.Dev. Min Median Max
real 0.124 0.013 0.107 0.118 0.145
user 0.103 0.013 0.073 0.102 0.130
sys 0.021 0.006 0.003 0.020 0.036
```
[edelooff/carrange](https://github.com/edelooff/carrange) (C++):
```bash
$ multitime -n100 -q -s0 -i "cat samples/1m.txt" ./composer
===> multitime results
1: -i "cat samples/1m.txt" -q ./composer
Mean Std.Dev. Min Median Max
real 0.177 0.013 0.161 0.171 0.214
user 0.157 0.014 0.131 0.153 0.200
sys 0.020 0.007 0.003 0.020 0.037
```
[Gradecak/rs-bouquets](https://github.com/Gradecak/rs-bouquets) (Rust):
```bash
$ multitime -n100 -q -s0 -i "cat samples/1m.txt" ./rs-bouquets
===> multitime results
1: -i "cat samples/1m.txt" -q ./rs-bouquets
Mean Std.Dev. Min Median Max
real 0.360 0.015 0.338 0.356 0.396
user 0.338 0.017 0.311 0.336 0.379
sys 0.021 0.008 0.003 0.020 0.043
```
## Benchmark of reference solution
This gets its own section because compiled languages are in a league of their own.
[reference solution](https://github.com/bloomon/code-challenge-verifier/blob/master/reference.py) (Python 3.10):
```bash
$ multitime -n10 -q -s0 -i "cat samples/1m.txt" python3.10 reference.py
===> multitime results
1: -i "cat samples/1m.txt" -q python3.10 reference.py
Mean Std.Dev. Min Median Max
real 20.417 0.122 20.154 20.424 20.570
user 20.339 0.136 20.088 20.333 20.522
sys 0.031 0.029 0.006 0.021 0.111
```
Curiously, the reference solution was slower in Python 3.11 🤔
[reference solution](https://github.com/bloomon/code-challenge-verifier/blob/master/reference.py) (Python 3.11):
```bash
$ multitime -n10 -q -s0 -i "cat samples/1m.txt" python3.11 reference.py
===> multitime results
1: -i "cat samples/1m.txt" -q python3.11 reference.py
Mean Std.Dev. Min Median Max
real 26.699 0.333 26.163 26.691 27.391
user 26.613 0.348 26.117 26.619 27.332
sys 0.026 0.026 0.000 0.016 0.091
```