1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
//! Time-stamp counter (TSC) timer
//!
//! This library provides a time-stamp counter (TSC) based timer for micro
//! benchmarking.
//!
//! # Example
//!
//! ```
//! # #![feature(test)]
//! # extern crate test;
//! # use self::test::black_box;
//! # use tsc::*;
//! // The function we want to time:
//! pub fn fibonacci(n: u64) -> u64 {
//!     match n {
//!         0 | 1 => 1,
//!         n => fibonacci(n - 1) + fibonacci(n - 2),
//!     }
//! }
//!
//! // Non-invariant TSCs might produce unreliable results:
//! assert!(has_invariant_tsc(), "The TSC is not invariant!");
//!
//! let (duration, result) =
//!     Duration::span(|| black_box(fibonacci(black_box(8))));
//!
//! assert_eq!(result, 34);
//!
//! println!("Reference cycle count: {} cycles.", duration.cycles());
//! // On my machine prints: "Reference cycle count: 951 cycles."
//! ```
//!
//! # Notes
//!
//! * The TSC runs at a different frequency than the CPU clock frequency, so
//! the cycles reported here are "reference cycles" and not real CPU clock
//! cycles.
//!
//! * If the TSC is not _invariant_ (Nehalem-and-later) the measurements might
//! not be very accurate due to turbo boost, speed-step, power management, etc.
//!
//! * Converting "reference cycles" to time (e.g., nanoseconds) is, in general,
//! not possible to do reliably in user-space.
//!
//! * One might want to disable preemption and hard interrupts before timing
//! (see [How to Benchmark Code Execution Times on Intel® IA-32 and IA-64
//! Instruction Set Architectures][intel_bench_paper]) to further improve the
//! accuracy of the measurements.
//!
//! # References
//!
//! * [How to Benchmark Code Execution Times on Intel® IA-32 and IA-64
//! Instruction Set Architectures][intel_bench_paper]
//! * [Pitfalls of TSC usage][pitfalls_tsc]
//! * [Time Stamps Counters][tinola_blog]
//! * [SO answer to "Get CPU cycle count"][so_cpu_cycles]
//! * [So answer to "Using Time stamp counter to get the time
//! stamp"][so_time_stamp]
//!
//! [intel_bench_paper]:
//! https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/ia-32-ia-64-benchmark-code-execution-paper.pdf
//! [pitfalls_tsc]: http://oliveryang.net/2015/09/pitfalls-of-TSC-usage/
//! [tinola_blog]: http://blog.tinola.com/?e=54
//! [so_cpu_cycles]: https://stackoverflow.com/a/51907627/1422197
//! [so_time_stamp]: https://stackoverflow.com/a/42490374/1422197
#![feature(asm, stdsimd, test)]
#![cfg_attr(not(test), no_std)]

#[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
compile_error!(
    "The TSC crate only supports the \"x86\" and \"x86_64\" architectures"
);

extern crate test;

use core::ops;

#[cfg(target_arch = "x86")]
use core::arch::x86 as arch;
#[cfg(target_arch = "x86_64")]
use core::arch::x86_64 as arch;

/// Returns true if the CPU has an invariant TSC.
///
/// Without an invariant TSC, the timings reported by this library might be
/// unreliable.
pub fn has_invariant_tsc() -> bool {
    use self::arch::{has_cpuid, CpuidResult, __cpuid};

    // The invariant TSC is advertised in the CPUID.80000007H:EDX[8] bit.

    // CPU doesn't have a CPUID instruction => too old to have an invariant
    // TSC.
    if !has_cpuid() {
        return false;
    }

    // Obtain the largest basic CPUID leaf supported by the CPUID
    let CpuidResult { eax: max_basic_leaf, .. } = unsafe { __cpuid(0_u32) };

    // Earlier Intel 486 => too old to have an invariant TSC.
    if max_basic_leaf < 1 {
        return false;
    }

    // Obtain the largest extended CPUID leaf supported by the CPUID
    let CpuidResult { eax: max_extended_leaf, .. } =
        unsafe { __cpuid(0x8000_0000_u32) };

    // CPU doesn't have "Advanced Power Management Information" => too old to
    // have an invariant TSC.
    if max_extended_leaf < 7 {
        return false;
    }

    let CpuidResult { edx, .. } = unsafe { __cpuid(0x8000_0007_u32) };

    // Test CPUID.80000007H:EDX[8], if the bit is set, the CPU has an
    // invariant TSC
    edx & (1 << 8) != 0
}

/// Start time instant
pub struct Start(u64);

impl Start {
    /// Start measurement
    pub fn now() -> Self {
        unsafe {
            let _ = arch::__cpuid(0);
            Start(core::mem::transmute(arch::_rdtsc()))
        }
    }
}

/// Stop time instant
pub struct Stop(u64);

impl Stop {
    /// Stop measurement
    pub fn now() -> Self {
        unsafe {
            let mut core: u32 = 0;
            let r = arch::__rdtscp(&mut core as *mut _) as u64;
            let _ = arch::__cpuid(0);
            Stop(r)
        }
    }
}

/// Duration between two time instants
#[derive(Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub struct Duration(u64);

impl Duration {
    /// Number of clock cycles
    pub fn cycles(self) -> u64 {
        self.0
    }

    /// Returns a tuple of the execution duration of the function `f` and its
    /// result.
    pub fn span<R, F: Fn() -> R>(f: F) -> (Self, R) {
        let start = Start::now();
        let result = f();
        let stop = Stop::now();
        let measurement_duration = stop - start;
        let measurement_overhead = Self::span_overhead();
        assert!(measurement_overhead <= measurement_duration);
        let f_duration = measurement_duration - measurement_overhead;
        (f_duration, result)
    }

    /// Returns the overhead that is intrinsic to [`span`].
    ///
    /// That is, how many cycles does it take to time a no-op.
    fn span_overhead() -> Self {
        let start = Start::now();
        let _result = test::black_box(0);
        let stop = Stop::now();
        stop - start
    }
}

impl ops::Sub<Start> for Stop {
    type Output = Duration;
    fn sub(self, start: Start) -> Duration {
        debug_assert!(
            self.0 > start.0,
            "stop time instant happened after start time instant"
        );
        Duration(self.0 - start.0)
    }
}

impl ops::Sub<Duration> for Duration {
    type Output = Self;
    fn sub(self, other: Self) -> Self {
        debug_assert!(self.0 > other.0, "subtracting durations overflows");
        Duration(self.0 - other.0)
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    #[test]
    fn start_stop() {
        let start = Start::now();
        let stop = Stop::now();
        let dur1 = stop - start;

        fn foo() -> i32 {
            2_i32.pow(test::black_box(8))
        }

        let (dur2, r) = Duration::span(foo);
        assert_eq!(r, 256);
        assert!(dur2.cycles() > dur1.cycles());
    }

    #[test]
    fn print_span_overhead() {
        println!(
            "span overhead: {} cycles",
            Duration::span_overhead().cycles()
        );
    }

    #[test]
    fn invariant_tsc() {
        assert!(has_invariant_tsc(), "the cpu does not have an invariant TSC");
    }
}