glen/dyna3
1
0
Fork 0
Code Issues 9 Pull Requests 1 Actions Projects Releases 2 Wiki Activity

Write up Rust benchmark variants

Vectornaut 2024-08-13 20:30:58 +00:00
parent d06e8551ad
commit 34aa1eb66b
1 changed files with 10 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
Language-benchmarks.md

@ -14,6 +14,8 @@ To evaluate the performance cost, Aaron wrote a benchmark program in Rust and Ja
- Find the eigenvalues of $A,\;\ldots\;T^{R-1}A$. - Find the eigenvalues of $A,\;\ldots\;T^{R-1}A$.
To validate the computation, the benchmark program displays the eigenvalues of $T^r A$, with $r \in \{0, \ldots, R\}$ controlled by a slider. Displaying the eigenvalues isn't part of the benchmark computation, so it isn't timed. To validate the computation, the benchmark program displays the eigenvalues of $T^r A$, with $r \in \{0, \ldots, R\}$ controlled by a slider. Displaying the eigenvalues isn't part of the benchmark computation, so it isn't timed.
The language comparison benchmark uses 64-bit floating point matrices of size $N = 60$. Other variants of the benchmark, used to compare different design decisions within Rust, are described at the end.
## Running the benchmark ## Running the benchmark
### Rust ### Rust
- To build and run, call `trunk serve --release` from the `rust-benchmark` folder and go to the URL that Trunk is serving. - To build and run, call `trunk serve --release` from the `rust-benchmark` folder and go to the URL that Trunk is serving.
@ -48,4 +50,11 @@ The Rust version typically ran 6–11 times as fast as the Scala version, and it
### Chromium ### Chromium
The Rust version typically ran 5–7 times as fast as the Scala version, with comparable consistency. The Rust version typically ran 5–7 times as fast as the Scala version, with comparable consistency.
- Rust 80–90 ms - Rust 80–90 ms
- Scala: 520–590 ms - Scala: 520–590 ms
## Rust benchmark variants
### Low-precision variant
- For matrices of size $N = 50$, using 32-bit floating point instead of 64-bit made the computation about 15% faster (60 ms instead of 70 ms). However, for $N \ge 54$, the 32-bit floating point variant would hang indefinitely! Maybe the target precision doesn't change to accommodate the lower-precision data type?
### Statically sized variant
- For matrices of size $N = 60$, using statically sized matrices instead of dynamically sized ones made the computation about 10% *slower* (125–130 ms instead of 110–120 ms).
- For matrices of size $N = 50$, using statically sized matrices made the computation about 15% *slower* (80 ms instead of 70 ms).
- For matrices of size $N = 20$, statically and dynamically sized matrices gave comparable run times (12–15 ms).