perf(server): make tiles encoding parallel with rayon

This can help a lot wall-clock time, but depends on CPU.

rfx_enc                 time:   [9.7885 ms 10.123 ms 10.439 ms]
                        change: [-80.484% -79.847% -79.208%] (p = 0.00 < 0.05)
                        Performance has improved.

Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>

Author: elmarco

Cargo.lock

@@ -16,7 +16,9 @@ doctest = true
 test = false
 
 [features]
+default = ["rayon"]
 helper = ["dep:x509-cert", "dep:rustls-pemfile"]
+rayon = ["dep:rayon"]
 
 # Internal (PRIVATE!) features used to aid testing.
 # Don't rely on these whatsoever. They may disappear at any time.
@@ -42,6 +44,7 @@ ironrdp-rdpsnd.workspace = true
 tracing.workspace = true
 x509-cert = { version = "0.2.5", optional = true }
 rustls-pemfile = { version = "2.2.0", optional = true }
+rayon = { version = "1.10.0", optional = true }
 
 [dev-dependencies]
 tokio = { version = "1", features = ["sync"] }

crates/ironrdp-server/Cargo.toml

@@ -144,9 +144,16 @@ impl<'a> UpdateEncoder<'a> {
     }
 
     fn encode(&self, data: &'a mut UpdateEncoderData) -> EncodeResult<Vec<rfx::Tile<'a>>> {
+        #[cfg(feature = "rayon")]
+        use rayon::prelude::*;
+
         let (tiles_x, tiles_y) = self.tiles_xy();
 
+        #[cfg(not(feature = "rayon"))]
         let chunks = data.0.chunks_mut(64 * 64 * 3);
+        #[cfg(feature = "rayon")]
+        let chunks = data.0.par_chunks_mut(64 * 64 * 3);
+
         let tiles: Vec<_> = (0..tiles_y).flat_map(|y| (0..tiles_x).map(move |x| (x, y))).collect();
 
         chunks