Select better default thread count on Apple Silicon (M1)

2 files changed, 15 insertions, 2 deletions

diff --git a/src/main.rs b/src/main.rs
index 9313c26..9f1be5e 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -13,7 +13,7 @@ fn main() -> Result<()> {
 
     let opt: options::Args = options::Args::parse_from(wild::args_os());
     let walk_options = dua::WalkOptions {
-        threads: opt.threads.unwrap_or(0),
+        threads: opt.threads,
         byte_format: opt.format.map(Into::into).unwrap_or(ByteFormat::Metric),
         apparent_size: opt.apparent_size,
         count_hard_links: opt.count_hard_links,
diff --git a/src/options.rs b/src/options.rs
index 9d8ab76..dc9a355 100644
--- a/src/options.rs
+++ b/src/options.rs
@@ -61,7 +61,20 @@ pub struct Args {
     /// The amount of threads to use. Defaults to the amount of logical processors.
     /// Set to 1 to use only a single thread.
     #[clap(short = 't', long = "threads")]
-    pub threads: Option<usize>,
+    // On macos with apple silicon, the IO subsystem is entirely different and one thread can mostly max it out.
+    // Thus using more threads just burns energy unnecessarily.
+    // It's notable that `du` is very fast even on a single core and more power efficient than dua with a single core.
+    // The default of '4' seems related to the amount of performance cores present in the system.
+    #[cfg_attr(
+        all(target_os = "macos", target_arch = "aarch64"),
+        clap(default_value = "4")
+    )]
+    // On everything else, it's usually a good idea to use as many threads as possible for noticeable speedups.
+    #[cfg_attr(
+        not(all(target_os = "macos", target_arch = "aarch64")),
+        clap(default_value = "0")
+    )]
+    pub threads: usize,
 
     /// The format with which to print byte counts.
     /// Metric - uses 1000 as base (default)