Support hardware counters (instructions, cache misses, ...) (#983)

* Add hardware counters to JSON exports

* Clarify Linux hardware counter setup

src/benchmark/executor.rs

@@ -62,6 +62,7 @@ fn run_command_and_measure_common(
     command_input_policy: &CommandInputPolicy,
     command_output_policy: &CommandOutputPolicy,
     command_name: &str,
+    collect_hardware_counters: bool,
 ) -> Result<Measurement> {
     let stdin = command_input_policy.get_stdin()?;
     let (stdout, stderr) = command_output_policy.get_stdout_stderr()?;
@@ -76,7 +77,7 @@ fn run_command_and_measure_common(
         command.env("HYPERFINE_ITERATION", value);
     }
 
-    let measurement = execute_and_measure(command)
+    let measurement = execute_and_measure(command, collect_hardware_counters)
         .with_context(|| format!("Failed to run command '{command_name}'"))?;
 
     if !measurement.exit_status.success() {
@@ -144,6 +145,7 @@ impl Executor for RawExecutor<'_> {
             &self.options.command_input_policy,
             output_policy,
             &command.get_command_line(),
+            matches!(iteration, BenchmarkIteration::Benchmark(_)),
         )
     }
 
@@ -199,6 +201,7 @@ impl Executor for ShellExecutor<'_> {
             &self.options.command_input_policy,
             output_policy,
             &command.get_command_line(),
+            false,
         )?;
 
         // Subtract shell spawning time
@@ -278,6 +281,7 @@ impl Executor for ShellExecutor<'_> {
             time_user: measurements.time_user_mean(),
             time_system: measurements.time_system_mean(),
             memory_peak_resident: None,
+            hardware_counters: Default::default(),
             exit_status: ExitStatus::default(),
         });
 
@@ -336,6 +340,7 @@ impl Executor for MockExecutor {
             time_user: Time::zero(),
             time_system: Time::zero(),
             memory_peak_resident: Some(Information::zero()),
+            hardware_counters: Default::default(),
             exit_status,
         })
     }

src/benchmark/measurement.rs

@@ -29,6 +29,47 @@ where
     }
 }
 
+fn serialize_optional_count<S>(value: &Option<u64>, serializer: S) -> Result<S::Ok, S::Error>
+where
+    S: serde::Serializer,
+{
+    #[derive(Serialize)]
+    struct Count {
+        value: u64,
+    }
+
+    value.map(|value| Count { value }).serialize(serializer)
+}
+
+#[derive(Debug, Clone, Default, Serialize, PartialEq)]
+pub struct HardwareCounters {
+    #[serde(
+        skip_serializing_if = "Option::is_none",
+        serialize_with = "serialize_optional_count"
+    )]
+    pub cpu_cycles: Option<u64>,
+    #[serde(
+        skip_serializing_if = "Option::is_none",
+        serialize_with = "serialize_optional_count"
+    )]
+    pub instructions: Option<u64>,
+    #[serde(
+        skip_serializing_if = "Option::is_none",
+        serialize_with = "serialize_optional_count"
+    )]
+    pub cache_references: Option<u64>,
+    #[serde(
+        skip_serializing_if = "Option::is_none",
+        serialize_with = "serialize_optional_count"
+    )]
+    pub cache_misses: Option<u64>,
+    #[serde(
+        skip_serializing_if = "Option::is_none",
+        serialize_with = "serialize_optional_count"
+    )]
+    pub branch_misses: Option<u64>,
+}
+
 /// Performance metric measurements and exit code for a single run
 #[derive(Debug, Clone, Default, Serialize, PartialEq)]
 pub struct Measurement {
@@ -51,6 +92,9 @@ pub struct Measurement {
     )]
     pub memory_peak_resident: Option<Information>,
 
+    #[serde(flatten)]
+    pub hardware_counters: HardwareCounters,
+
     // The exit status of the process
     #[serde(rename = "exit_code", serialize_with = "serialize_exit_status")]
     pub exit_status: ExitStatus,