summaryrefslogtreecommitdiffstats
path: root/src/common/Cycles.cc
diff options
context:
space:
mode:
authorDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-21 11:54:28 +0000
committerDaniel Baumann <daniel.baumann@progress-linux.org>2024-04-21 11:54:28 +0000
commite6918187568dbd01842d8d1d2c808ce16a894239 (patch)
tree64f88b554b444a49f656b6c656111a145cbbaa28 /src/common/Cycles.cc
parentInitial commit. (diff)
downloadceph-e6918187568dbd01842d8d1d2c808ce16a894239.tar.xz
ceph-e6918187568dbd01842d8d1d2c808ce16a894239.zip
Adding upstream version 18.2.2.upstream/18.2.2
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/common/Cycles.cc')
-rw-r--r--src/common/Cycles.cc220
1 files changed, 220 insertions, 0 deletions
diff --git a/src/common/Cycles.cc b/src/common/Cycles.cc
new file mode 100644
index 000000000..2ebd24699
--- /dev/null
+++ b/src/common/Cycles.cc
@@ -0,0 +1,220 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2014 UnitedStack <haomai@unitedstack.com>
+ *
+ * Author: Haomai Wang <haomaiwang@gmail.com>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation. See file COPYING.
+ *
+ */
+/* Copyright (c) 2011-2014 Stanford University
+ *
+ * Permission to use, copy, modify, and distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR(S) DISCLAIM ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL AUTHORS BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include "debug.h"
+#include "Cycles.h"
+
+double Cycles::cycles_per_sec = 0;
+
+/**
+ * Perform once-only overall initialization for the Cycles class, such
+ * as calibrating the clock frequency. This method must be called
+ * before using the Cycles module.
+ *
+ * It is not initialized by default because the timing loops cause
+ * general process startup times to balloon
+ * (http://tracker.ceph.com/issues/15225).
+ */
+void Cycles::init()
+{
+ if (cycles_per_sec != 0)
+ return;
+
+ // Skip initialization if rtdsc is not implemented
+ if (rdtsc() == 0)
+ return;
+
+ // Compute the frequency of the fine-grained CPU timer: to do this,
+ // take parallel time readings using both rdtsc and gettimeofday.
+ // After 10ms have elapsed, take the ratio between these readings.
+
+ struct timeval start_time, stop_time;
+ uint64_t micros;
+ double old_cycles;
+
+ // There is one tricky aspect, which is that we could get interrupted
+ // between calling gettimeofday and reading the cycle counter, in which
+ // case we won't have corresponding readings. To handle this (unlikely)
+ // case, compute the overall result repeatedly, and wait until we get
+ // two successive calculations that are within 0.1% of each other.
+ old_cycles = 0;
+ while (1) {
+ if (gettimeofday(&start_time, NULL) != 0) {
+ ceph_abort_msg("couldn't read clock");
+ }
+ uint64_t start_cycles = rdtsc();
+ while (1) {
+ if (gettimeofday(&stop_time, NULL) != 0) {
+ ceph_abort_msg("couldn't read clock");
+ }
+ uint64_t stop_cycles = rdtsc();
+ micros = (stop_time.tv_usec - start_time.tv_usec) +
+ (stop_time.tv_sec - start_time.tv_sec)*1000000;
+ if (micros > 10000) {
+ cycles_per_sec = static_cast<double>(stop_cycles - start_cycles);
+ cycles_per_sec = 1000000.0*cycles_per_sec/ static_cast<double>(micros);
+ break;
+ }
+ }
+ double delta = cycles_per_sec/1000.0;
+ if ((old_cycles > (cycles_per_sec - delta)) &&
+ (old_cycles < (cycles_per_sec + delta))) {
+ return;
+ }
+ old_cycles = cycles_per_sec;
+ }
+}
+
+/**
+ * Return the number of CPU cycles per second.
+ */
+double Cycles::per_second()
+{
+ return get_cycles_per_sec();
+}
+
+/**
+ * Given an elapsed time measured in cycles, return a floating-point number
+ * giving the corresponding time in seconds.
+ * \param cycles
+ * Difference between the results of two calls to rdtsc.
+ * \param cycles_per_sec
+ * Optional parameter to specify the frequency of the counter that #cycles
+ * was taken from. Useful when converting a remote machine's tick counter
+ * to seconds. The default value of 0 will use the local processor's
+ * computed counter frequency.
+ * \return
+ * The time in seconds corresponding to cycles.
+ */
+double Cycles::to_seconds(uint64_t cycles, double cycles_per_sec)
+{
+ if (cycles_per_sec == 0)
+ cycles_per_sec = get_cycles_per_sec();
+ return static_cast<double>(cycles)/cycles_per_sec;
+}
+
+/**
+ * Given a time in seconds, return the number of cycles that it
+ * corresponds to.
+ * \param seconds
+ * Time in seconds.
+ * \param cycles_per_sec
+ * Optional parameter to specify the frequency of the counter that #cycles
+ * was taken from. Useful when converting a remote machine's tick counter
+ * to seconds. The default value of 0 will use the local processor's
+ * computed counter frequency.
+ * \return
+ * The approximate number of cycles corresponding to #seconds.
+ */
+uint64_t Cycles::from_seconds(double seconds, double cycles_per_sec)
+{
+ if (cycles_per_sec == 0)
+ cycles_per_sec = get_cycles_per_sec();
+ return (uint64_t) (seconds*cycles_per_sec + 0.5);
+}
+
+/**
+ * Given an elapsed time measured in cycles, return an integer
+ * giving the corresponding time in microseconds. Note: to_seconds()
+ * is faster than this method.
+ * \param cycles
+ * Difference between the results of two calls to rdtsc.
+ * \param cycles_per_sec
+ * Optional parameter to specify the frequency of the counter that #cycles
+ * was taken from. Useful when converting a remote machine's tick counter
+ * to seconds. The default value of 0 will use the local processor's
+ * computed counter frequency.
+ * \return
+ * The time in microseconds corresponding to cycles (rounded).
+ */
+uint64_t Cycles::to_microseconds(uint64_t cycles, double cycles_per_sec)
+{
+ return to_nanoseconds(cycles, cycles_per_sec) / 1000;
+}
+
+/**
+ * Given an elapsed time measured in cycles, return an integer
+ * giving the corresponding time in nanoseconds. Note: to_seconds()
+ * is faster than this method.
+ * \param cycles
+ * Difference between the results of two calls to rdtsc.
+ * \param cycles_per_sec
+ * Optional parameter to specify the frequency of the counter that #cycles
+ * was taken from. Useful when converting a remote machine's tick counter
+ * to seconds. The default value of 0 will use the local processor's
+ * computed counter frequency.
+ * \return
+ * The time in nanoseconds corresponding to cycles (rounded).
+ */
+uint64_t Cycles::to_nanoseconds(uint64_t cycles, double cycles_per_sec)
+{
+ if (cycles_per_sec == 0)
+ cycles_per_sec = get_cycles_per_sec();
+ return (uint64_t) (1e09*static_cast<double>(cycles)/cycles_per_sec + 0.5);
+}
+
+/**
+ * Given a number of nanoseconds, return an approximate number of
+ * cycles for an equivalent time length.
+ * \param ns
+ * Number of nanoseconds.
+ * \param cycles_per_sec
+ * Optional parameter to specify the frequency of the counter that #cycles
+ * was taken from. Useful when converting a remote machine's tick counter
+ * to seconds. The default value of 0 will use the local processor's
+ * computed counter frequency.
+ * \return
+ * The approximate number of cycles for the same time length.
+ */
+uint64_t
+Cycles::from_nanoseconds(uint64_t ns, double cycles_per_sec)
+{
+ if (cycles_per_sec == 0)
+ cycles_per_sec = get_cycles_per_sec();
+ return (uint64_t) (static_cast<double>(ns)*cycles_per_sec/1e09 + 0.5);
+}
+
+/**
+ * Busy wait for a given number of microseconds.
+ * Callers should use this method in most reasonable cases as opposed to
+ * usleep for accurate measurements. Calling usleep may put the the processor
+ * in a low power mode/sleep state which reduces the clock frequency.
+ * So, each time the process/thread wakes up from usleep, it takes some time
+ * to ramp up to maximum frequency. Thus meausrements often incur higher
+ * latencies.
+ * \param us
+ * Number of microseconds.
+ */
+void
+Cycles::sleep(uint64_t us)
+{
+ uint64_t stop = Cycles::rdtsc() + Cycles::from_nanoseconds(1000*us);
+ while (Cycles::rdtsc() < stop);
+}