Some improvements to the Rust version

[chancancode]

• Fix clippy warnings These are just linter warnings about unidiomatic/unnecessary code, should be pretty harmless to apply.

• Remove Node<T> generic This isn't supposed to change the compiled output (and at least on the stripped down version hosted on godbolt it didn't). Just making the Rust version match more 1:1 with the Zig version to remove some distractions.

• Cleanup benchmark script The main thing is that the purpose of bench_with_input is to automatically wrap the input in black_box, so it was needlessly double-wrapped.

• Make the two versions more identical The previous sum_rayon version has a extra check for (left && right) that made the code substantially different than the plain version, at least visually.

  This changes the implementations on both sides to match and also IMO a bit more idiomatic.

• Make sum a method This matches the Zig version a bit closer in the source code, and is IMO more idiomatic, but it shouldn't really change the compiled output.

I did make some (lousy) effort to make sure none of these obviously caused any regression (it may be slightly faster?), but I was just running it on my laptop with a ton of browser tabs open, etc, etc. So probably should double check that in your benchmark environment.

[judofyr]

Oh nice! I was actually thinking of running it through clippy, but forgot about it. These changes look sensible to me. I’m a bit busy this week, but hopefully in a few days I’ll be able to re-do the benchmark in the same environment and verify that the results are similar.

[judofyr]

I ran this again now and then I get the following diff:

# Rerun benchmark
(cd examples/rust-parallel-example && cargo bench)

# Regenerate CSV
rm bench/rayon-*.csv && make bench-rayon

# Diff
git diff

diff --git a/bench/rayon-tree-sum-1000.csv b/bench/rayon-tree-sum-1000.csv
index d9c615a..410a04e 100644
--- a/bench/rayon-tree-sum-1000.csv
+++ b/bench/rayon-tree-sum-1000.csv
@@ -1,7 +1,7 @@
-Baseline,1.5597891883229127
-Rayon 1 thread,23.51176644639185
-Rayon 2 threads,16.82203207978065
-Rayon 4 threads,14.939625885334602
-Rayon 8 threads,18.66874537774096
-Rayon 16 threads,24.183746144849145
-Rayon 32 threads,105.44453796411226
+Baseline,1.3574152155604338
+Rayon 1 thread,20.265803978522356
+Rayon 2 threads,15.612293062453892
+Rayon 4 threads,14.348052230794815
+Rayon 8 threads,16.42391594539397
+Rayon 16 threads,21.837912138899224
+Rayon 32 threads,96.13910653253916
diff --git a/bench/rayon-tree-sum-100M.csv b/bench/rayon-tree-sum-100M.csv
index b94de87..ef25279 100644
--- a/bench/rayon-tree-sum-100M.csv
+++ b/bench/rayon-tree-sum-100M.csv
@@ -1,7 +1,7 @@
-Baseline,7.4849190632000004
-Rayon 1 thread,22.9933083354
-Rayon 2 threads,11.7585642422
-Rayon 4 threads,5.898856567
-Rayon 8 threads,2.9989583844
-Rayon 16 threads,1.6433929402
-Rayon 32 threads,1.651134205
+Baseline,7.1566549926
+Rayon 1 thread,20.2334262024
+Rayon 2 threads,10.2145197118
+Rayon 4 threads,5.2211526282
+Rayon 8 threads,2.6398534263999998
+Rayon 16 threads,1.4336398069999998
+Rayon 32 threads,1.444313771

The machine appears to be slightly faster today (wonders of the cloud, maybe?), but the overall shape of the benchmark results still apply in this PR. Let's merge this!