Use -Z threads=0 option in config_fast_builds.toml

[doonv]

Objective

Improve compile times.

Solution

The -Z threads=0 option has been introduced into nightly rust somewhat recently, and was showcased in this rust-lang article. This option improves multithreading in rust and speeds up compilation. I added this option to config_fast_builds.toml so others can also use this option to improve compile times.

-Z threads=0 automatically uses the same amount of threads as the amount of threads in your system (See rustc source code).

Benchmarks

Disclaimer: This section was written before I knew of -Z threads=0, so it uses -Z threads=8 instead.

I compiled bevy with/without the -Z threads=8 and saw about a 7% improvement in compliation times on my Kubuntu system with a 13th Gen Intel® Core™ i5-13400.

Also the compile times go down over time, probably because I had other things running in the background.

Without -Z threads=8

• 42.33s
• 40.90s
• 38.27s
• 38.07s
• 37.17s
• 37.67s
• 36.63s
• 37.24s

Average: 38.535

With -Z threads=8

• 36.77s
• 39.50s
• 38.86s
• 35.61s
• 34.37s
• 34.32s
• 34.44s
• 33.74s

Average: 35.95125

[doonv]

Can any MacOS users test if this works well? It might reduce performance on MacOS or even not work at all.

[BD103]

After testing it on my M1 Mac, this option actually seems to slow down the compilation process. I built with the --timings option to generate the following charts. The time increases from 58.4 seconds to 63.6 seconds.

Before:

After:

---

Edit: Redid timings so the only difference between "Before" and "After" was commenting out -Zthreads=0.

[doonv]

Is this an average or just one test?

[BD103]

That was just one test. Here's my results running cargo clean && cargo +nightly build with and without -Zthreads=0 commented out.

Before

• 1:03s
• 1:06s
• 1:11s
• 1:12s
• 1:08s
• 0:59.4s
• 1:03s

Average: 1:06.05s

After

• 1:05s
• 1:11s
• 1:13s
• 1:20s
• 1:07s
• 1:11s
• 1:12s

Average: 1:11.29s

[doonv]

Windows users should also test this out

[Bcompartment]

I suggest to:

-Z threads= u128::MAX

With this we can compile instantly, I guess ....

But seriously talking, why 8 ? Is there any way to automatically set it to max (best) number possible for given CPU?

[doonv]

why 8 ?

That's the recommended amount.

Is there any way to automatically set it to max (best) number possible for given CPU?

After reading some rustc source code, I realised that threads=0 uses the amount of threads in your system (16 for me). And it performs similarly.