Wait on address default

Cargo.toml

@@ -13,7 +13,7 @@ readme = "README.md"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [workspace]
-members = ["snmalloc-sys"]
+members = ["snmalloc-sys", "xtask"]
 
 [dependencies]
 snmalloc-sys = { version = "0.3.8", path = "snmalloc-sys", default-features = false }
@@ -34,3 +34,9 @@ notls = ["snmalloc-sys/notls"]
 stats = ["snmalloc-sys/stats"]
 usewait-on-address = ["snmalloc-sys/usewait-on-address"]
 libc-api = ["snmalloc-sys/libc-api"]
+tracing = ["snmalloc-sys/tracing"]
+fuzzing = ["snmalloc-sys/fuzzing"]
+vendored-stl = ["snmalloc-sys/vendored-stl"]
+check-loads = ["snmalloc-sys/check-loads"]
+pageid = ["snmalloc-sys/pageid"]
+gwp-asan = ["snmalloc-sys/gwp-asan"]

README.md

@@ -20,18 +20,12 @@ the [`snmalloc` paper](https://github.com/microsoft/snmalloc/blob/master/snmallo
 are listed at
 [bench_suite](https://github.com/SchrodingerZhu/bench_suite). There are three features defined in this crate:
 
-- `debug`: Enable the `Debug` mode in `snmalloc`.
-- ~~`1mib`: Use the `1mib` chunk configuration. From `0.2.17`, this is set as a default feature~~ (removed since 0.3.0)
-- ~~`16mib`: Use the `16mib` chunk configuration.~~ (removed since 0.3.0)
-- ~~`cache-friendly`: Make the allocator more cache friendly (setting `CACHE_FRIENDLY_OFFSET` to `64` in building the
-  library).~~ (removed since 0.3.0)
+- `debug`: Enable the `Debug` mode in `snmalloc`. This is also automatically enabled if Cargo's `DEBUG` environment variable is set to `true`.
 - `native-cpu`: Optimize `snmalloc` for the native CPU of the host machine. (this is not a default behavior
   since `0.2.14`)
 - `qemu`: Workaround `madvise` problem of QEMU environment
-- ~~`stats`: Enable statistics~~ (removed since 0.3.0)
 - `local_dynamic_tls`: Workaround cannot allocate memory in static tls block
 - `build_cc`: Use of cc crate instead of cmake (cmake still default) as builder (more platform agnostic)
-- ~~`usecxx20`: Enable C++20 standard if available~~ (removed since 0.3.0)
 - `usecxx17`: Use C++17 standard
 - `check`: Enable extra checks to improve security, see upstream [security docs](https://github.com/microsoft/snmalloc/tree/main/docs/security).
   Note that the `memcpy` protection is not enabled in Rust.
@@ -40,6 +34,25 @@ are listed at
 - `notls`: Enables to be loaded dynamically, thus disable tls.
 - `stats`: Enables allocation statistics.
 - `libc-api`: Enables libc API backed by snmalloc.
+- `usewait-on-address`: Enable `WaitOnAddress` support on Windows (enabled by default).
+- `tracing`: Enable structured tracing/logging.
+- `fuzzing`: Enable fuzzing support.
+- `vendored-stl`: Use self-vendored STL.
+- `check-loads`: Enable check loads feature.
+- `pageid`: Enable page ID feature.
+- `gwp-asan`: Enable GWP-ASan integration. Requires `SNMALLOC_GWP_ASAN_INCLUDE_PATH` and `SNMALLOC_GWP_ASAN_LIBRARY_PATH`.
+
+## Build Configuration
+
+The build script ensures architectural alignment between the Rust profile and the underlying `snmalloc` allocator:
+
+### Environment Variables
+The following environment variables are automatically detected and propagated:
+- `DEBUG`: Synchronizes the `snmalloc` build type with the Cargo profile. If `true`, `snmalloc` is built in `Debug` mode.
+- `OPT_LEVEL`: Propagated to the C++ compiler to ensure optimization parity between Rust and C++ components.
+
+### Windows CRT Consistency
+On Windows, the build script enforces static CRT linking (`/MT` or `/MTd`) across both `cc` and `cmake` builders. This prevents linker errors and ensures consistency when `snmalloc` is used as a global allocator.
 
 **To get the crates compiled, you need to choose either `1mib` or `16mib` to determine the chunk configuration**
 

examples/bench_contention.rs

@@ -0,0 +1,76 @@
+use std::sync::mpsc::channel;
+use std::thread;
+use std::time::Instant;
+use std::alloc::Layout;
+
+#[global_allocator]
+static ALLOC: snmalloc_rs::SnMalloc = snmalloc_rs::SnMalloc;
+
+const BLOCK_SIZE: usize = 64;
+const ITERATIONS: usize = 1_000_000;
+
+struct Ptr(*mut u8);
+unsafe impl Send for Ptr {}
+
+fn main() {
+    let thread_count = std::thread::available_parallelism().map(|n| n.get()).unwrap_or(4);
+    println!("Running contention benchmark with {} threads, {} iterations per thread", thread_count, ITERATIONS);
+
+    // Use std::sync::Barrier
+    let barrier = std::sync::Arc::new(std::sync::Barrier::new(thread_count + 1));
+
+    let mut senders = Vec::new();
+    let mut receivers = Vec::new();
+
+    // Create a ring topology channels
+    for _ in 0..thread_count {
+        let (tx, rx) = channel::<Ptr>();
+        senders.push(tx);
+        receivers.push(Some(rx));
+    }
+
+    let mut handles = Vec::new();
+
+    // Start timing from here, but actual work starts after barrier
+    let _start = Instant::now();
+
+    for i in 0..thread_count {
+        let barrier = barrier.clone();
+        // Thread i sends to (i + 1) % N
+        let tx = senders[(i + 1) % thread_count].clone();
+        // Thread i receives from i
+        let rx = receivers[i].take().unwrap();
+
+        handles.push(thread::spawn(move || {
+            // Pre-allocate some items to fill the pipe
+            let layout = Layout::from_size_align(BLOCK_SIZE, 8).unwrap();
+
+            barrier.wait(); // Synchronize start
+
+            for _ in 0..ITERATIONS {
+                // 1. Allocate a new block
+                let ptr = unsafe { std::alloc::alloc(layout) };
+
+                // 2. Send to next neighbor (who will free it)
+                tx.send(Ptr(ptr)).unwrap();
+
+                // 3. Receive from prev neighbor (who allocated it)
+                let received = rx.recv().unwrap();
+
+                // 4. Free the received block
+                unsafe { std::alloc::dealloc(received.0, layout) };
+            }
+        }));
+    }
+
+    barrier.wait(); // Start timing
+    let loop_start = Instant::now();
+
+    for h in handles {
+        h.join().unwrap();
+    }
+
+    let duration = loop_start.elapsed();
+    println!("Benchmark completed in {:.2?}", duration);
+    println!("Throughput: {:.2} Mops/sec", (thread_count * ITERATIONS) as f64 / duration.as_secs_f64() / 1_000_000.0);
+}

snmalloc-sys/Cargo.toml

@@ -17,7 +17,7 @@ cc = { version = "1.0", optional = true }
 cmake = { version = "0.1", optional = true }
 
 [features]
-default = ["build_cmake"]
+default = ["build_cmake", "usewait-on-address"]
 build_cc = ["cc"]
 build_cmake = ["cmake"]
 qemu = []
@@ -33,3 +33,9 @@ notls = []
 stats = []
 usewait-on-address = []
 libc-api = []
+tracing = []
+fuzzing = []
+vendored-stl = []
+check-loads = []
+pageid = []
+gwp-asan = []