4 files changed, 63 insertions, 13 deletions

diff --git a/crates/ra_cli/Cargo.toml b/crates/ra_cli/Cargo.toml
index 970cb1f82..9c2ebdad2 100644
--- a/crates/ra_cli/Cargo.toml
+++ b/crates/ra_cli/Cargo.toml
@@ -18,5 +18,5 @@ ra_db = { path = "../ra_db" }
 
 [dependencies.ra_prof]
 path = "../ra_prof"
-# features = [ "cpuprofiler" ]
+# features = [ "cpu_profiler" ]
 # features = [ "jemalloc" ]
diff --git a/crates/ra_prof/Cargo.toml b/crates/ra_prof/Cargo.toml
index 84dcc9813..d35adeeeb 100644
--- a/crates/ra_prof/Cargo.toml
+++ b/crates/ra_prof/Cargo.toml
@@ -9,10 +9,9 @@ publish = false
 once_cell = "0.2.0"
 itertools = "0.8.0"
 backtrace = "0.3.28"
-cpuprofiler = { version = "0.0.3", optional = true }
 jemallocator = { version = "0.3.2", optional = true }
 jemalloc-ctl = { version = "0.3.2", optional = true }
 
-
 [features]
 jemalloc = [ "jemallocator", "jemalloc-ctl" ]
+cpu_profiler = []
diff --git a/crates/ra_prof/src/google_cpu_profiler.rs b/crates/ra_prof/src/google_cpu_profiler.rs
new file mode 100644
index 000000000..db865c65b
--- /dev/null
+++ b/crates/ra_prof/src/google_cpu_profiler.rs
@@ -0,0 +1,39 @@
+//! https://github.com/gperftools/gperftools
+
+use std::{
+    ffi::CString,
+    os::raw::c_char,
+    path::Path,
+    sync::atomic::{AtomicUsize, Ordering},
+};
+
+#[link(name = "profiler")]
+#[allow(non_snake_case)]
+extern "C" {
+    fn ProfilerStart(fname: *const c_char) -> i32;
+    fn ProfilerStop();
+}
+
+static PROFILER_STATE: AtomicUsize = AtomicUsize::new(OFF);
+const OFF: usize = 0;
+const ON: usize = 1;
+const PENDING: usize = 2;
+
+pub fn start(path: &Path) {
+    if PROFILER_STATE.compare_and_swap(OFF, PENDING, Ordering::SeqCst) != OFF {
+        panic!("profiler already started");
+    }
+    let path = CString::new(path.display().to_string()).unwrap();
+    if unsafe { ProfilerStart(path.as_ptr()) } == 0 {
+        panic!("profiler failed to start")
+    }
+    assert!(PROFILER_STATE.compare_and_swap(PENDING, ON, Ordering::SeqCst) == PENDING);
+}
+
+pub fn stop() {
+    if PROFILER_STATE.compare_and_swap(ON, PENDING, Ordering::SeqCst) != ON {
+        panic!("profiler is not started")
+    }
+    unsafe { ProfilerStop() };
+    assert!(PROFILER_STATE.compare_and_swap(PENDING, OFF, Ordering::SeqCst) == PENDING);
+}
diff --git a/crates/ra_prof/src/lib.rs b/crates/ra_prof/src/lib.rs
index 6d44fef33..d32a289be 100644
--- a/crates/ra_prof/src/lib.rs
+++ b/crates/ra_prof/src/lib.rs
@@ -1,4 +1,6 @@
 mod memory_usage;
+#[cfg(feature = "cpu_profiler")]
+mod google_cpu_profiler;
 
 use std::{
     cell::RefCell,
@@ -268,25 +270,35 @@ impl Drop for Scope {
     }
 }
 
-/// A wrapper around https://github.com/AtheMathmo/cpuprofiler
+/// A wrapper around google_cpu_profiler.
 ///
-/// It can be used to capture sampling profiles of sections of code.
-/// It is not exactly out-of-the-box, as it relies on gperftools.
-/// See the docs for the crate for more!
+/// Usage:
+/// 1. Install gpref_tools (https://github.com/gperftools/gperftools), probably packaged with your Linux distro.
+/// 2. Build with `cpu_profiler` feature.
+/// 3. Tun the code, the *raw* output would be in the `./out.profile` file.
+/// 4. Install pprof for visualization (https://github.com/google/pprof).
+/// 5. Use something like `pprof -svg target/release/ra_cli ./out.profile` to see the results.
+///
+/// For example, here's how I run profiling on NixOS:
+///
+/// ```bash
+/// $ nix-shell -p gperftools --run \
+///     'cargo run --release -p ra_cli -- parse < ~/projects/rustbench/parser.rs > /dev/null'
+/// ```
 #[derive(Debug)]
 pub struct CpuProfiler {
     _private: (),
 }
 
 pub fn cpu_profiler() -> CpuProfiler {
-    #[cfg(feature = "cpuprofiler")]
+    #[cfg(feature = "cpu_profiler")]
     {
-        cpuprofiler::PROFILER.lock().unwrap().start("./out.profile").unwrap();
+        google_cpu_profiler::start("./out.profile".as_ref())
     }
 
-    #[cfg(not(feature = "cpuprofiler"))]
+    #[cfg(not(feature = "cpu_profiler"))]
     {
-        eprintln!("cpuprofiler feature is disabled")
+        eprintln!("cpu_profiler feature is disabled")
     }
 
     CpuProfiler { _private: () }
@@ -294,9 +306,9 @@ pub fn cpu_profiler() -> CpuProfiler {
 
 impl Drop for CpuProfiler {
     fn drop(&mut self) {
-        #[cfg(feature = "cpuprofiler")]
+        #[cfg(feature = "cpu_profiler")]
         {
-            cpuprofiler::PROFILER.lock().unwrap().stop().unwrap();
+            google_cpu_profiler::stop()
         }
     }
 }