Simply just compiling my project with that allocator makes it 18% faster

From an average of 450ms down to 381ms, fastest was 374ms

$ hyperfine "make build-game" "make build-game-mimalloc"
Benchmark 1: make build-game
  Time (mean ± σ):     450.4 ms ±   4.6 ms    [User: 383.8 ms, System: 64.7 ms]
  Range (min … max):   443.9 ms … 458.9 ms    10 runs

Benchmark 2: make build-game-mimalloc
  Time (mean ± σ):     381.7 ms ±   6.7 ms    [User: 346.2 ms, System: 34.1 ms]
  Range (min … max):   374.7 ms … 393.4 ms    10 runs

Summary
  make build-game-mimalloc ran
    1.18 ± 0.02 times faster than make build-game

Try with your project, and share the result here, and perhaps we should make it the default if it's an improvement for everyone?

LD_PRELOAD=/usr/lib/libmimalloc.so dmd

I forgot to include the link of the project: https://github.com/microsoft/mimalloc/

On Tuesday, 26 March 2024 at 15:56:14 UTC, ryuukk_ wrote:

Yep, it's huge, definitely should be default for DMD releases.
LDC releases are built with mimalloc since ~2years.

-Johan

On Tuesday, 26 March 2024 at 18:23:07 UTC, Johan wrote:

Are you sure? i tried with LDC (1.37.0) and i get 2.3sec vs 2.0sec with LD_PRELOAD

On Tuesday, 26 March 2024 at 18:34:00 UTC, ryuukk_ wrote:

Quite sure. https://github.com/ldc-developers/ldc/blob/3a6f2cae331c5a84a2f0dd2f6d6e9e09dbfad86d/.github/actions/3-build-native/action.yml#L26

You can check whether your LDC has mimalloc included by setting environment variable MIMALLOC_SHOW_STATS=1 (shows statistics when the program terminates)
Don't know why it'd be 10% faster in your case with LD_PRELOAD...

-Johan

On Tuesday, 26 March 2024 at 15:56:14 UTC, ryuukk_ wrote:

I always wondered why do people report SW improvements as increase in speed, instead of reduction in time. Its not like we are 'driving' a compiler and need to watch its speedometer, I am pretty sure everyone are watching a clock instead.

After a bit of math, I realized that reporting the changes in "speed" always paints a more rosy picture than reporting the changes in time.

For example, a 25% reduction in time will be reported as 33% increase in speed, where as a 25% increase in time will be reported as only 20% worsening of speed.

Exaggeration is probably not the main reason. Or is it? cue Vsause music

Just a fun fact I wanted to share :-)

On Tuesday, 26 March 2024 at 15:56:14 UTC, ryuukk_ wrote:

Thanks for sharing! I tested it on one of my applications, and build time went down from 0.44s to 0.35s. Building all my apps and running all my unittests went from 11.6s to 9.95s, which is amazing for how little effort it took (sudo pacman -S mimalloc and export LD_PRELOAD=/usr/lib/libmimalloc.so).

On Tuesday, 26 March 2024 at 18:46:59 UTC, Johan wrote:

Perhaps this is affecting more than just LDC? Maybe the linker too?

-Steve