[PATCH v3] benchtests: Add pthread mutex trylock throughput test (BZ #33704)

[Sunil Pandey]

On Fri, Dec 12, 2025 at 4:12 PM H.J. Lu <hjl.tools@gmail.com> wrote:

> On Sat, Dec 13, 2025 at 8:03 AM Sunil K Pandey <sunil.k.pandey@intel.com>
> wrote:
> >
> > This benchmark measures number of successful updates per second
> > under high cache contention.
> >
> > Tested on x86_64.
> > ---
> >  benchtests/Makefile                           |   2 +
> >  .../bench-pthread-mutex-trylock-throughput.c  | 175 ++++++++++++++++++
> >  2 files changed, 177 insertions(+)
> >  create mode 100644 benchtests/bench-pthread-mutex-trylock-throughput.c
> >
> > diff --git a/benchtests/Makefile b/benchtests/Makefile
> > index 2ecdde857e..c3368a7603 100644
> > --- a/benchtests/Makefile
> > +++ b/benchtests/Makefile
> > @@ -175,6 +175,7 @@ bench-pthread := \
> >    pthread-locks \
> >    pthread-mutex-lock \
> >    pthread-mutex-trylock \
> > +  pthread-mutex-trylock-throughput \
> >    pthread-spin-lock \
> >    pthread-spin-trylock \
> >    pthread_once \
> > @@ -188,6 +189,7 @@ bench-resolv := \
> >
> >  LDLIBS-bench-pthread-mutex-lock += -lm
> >  LDLIBS-bench-pthread-mutex-trylock += -lm
> > +LDLIBS-bench-pthread-mutex-trylock-throughput += -lm
> >  LDLIBS-bench-pthread-spin-lock += -lm
> >  LDLIBS-bench-pthread-spin-trylock += -lm
> >
> > diff --git a/benchtests/bench-pthread-mutex-trylock-throughput.c
> b/benchtests/bench-pthread-mutex-trylock-throughput.c
> > new file mode 100644
> > index 0000000000..7b4908e5aa
> > --- /dev/null
> > +++ b/benchtests/bench-pthread-mutex-trylock-throughput.c
> > @@ -0,0 +1,175 @@
> > +/* Measure pthread trylock throughput under high cache contention.
> > +   Copyright (C) 2025 Free Software Foundation, Inc.
> > +   This file is part of the GNU C Library.
> > +
> > +   The GNU C Library is free software; you can redistribute it and/or
> > +   modify it under the terms of the GNU Lesser General Public
> > +   License as published by the Free Software Foundation; either
> > +   version 2.1 of the License, or (at your option) any later version.
> > +
> > +   The GNU C Library is distributed in the hope that it will be useful,
> > +   but WITHOUT ANY WARRANTY; without even the implied warranty of
> > +   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> > +   Lesser General Public License for more details.
> > +
> > +   You should have received a copy of the GNU Lesser General Public
> > +   License along with the GNU C Library; if not, see
> > +   <https://www.gnu.org/licenses/>.  */
> > +
> > +#define TEST_MAIN
> > +#define TIMEOUT (20 * 60)
> > +#define TEST_NAME "pthread-mutex-trylock-throughput"
> > +
> > +#include <stdio.h>
> > +#include <stdlib.h>
> > +#include <string.h>
> > +#include <unistd.h>
> > +#include <math.h>
> > +#include <pthread.h>
> > +#include <sys/time.h>
> > +#include <sys/sysinfo.h>
> > +#include "bench-timing.h"
> > +#include "bench-util.h"
> > +#include "json-lib.h"
> > +
> > +#define ITERS 10000000
> > +#define RUN_COUNT 10
> > +
> > +pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
> > +
> > +/* Shared counter only incremented by successful threads.  */
> > +int counter = 0;
> > +
> > +static void *
> > +worker (void *v)
> > +{
> > +  for (int i = 0; i < ITERS; i++)
> > +    if (pthread_mutex_trylock(&mutex) == 0)
> > +      {
> > +       counter++;
> > +       pthread_mutex_unlock(&mutex);
> > +      }
> > +  return NULL;
> > +}
> > +
> > +static void
> > +do_bench_one (const char *name, int num_threads, int nprocs, json_ctx_t
> *js)
> > +{
> > +  struct timeval ts, te;
> > +  double tsd, ted, td, mean, stdev;
> > +  int i, j;
> > +  pthread_t threads[num_threads];
>
> Also use malloc and free.
>
>
Sure, I will fix it in v4. Copied from existing pthread benchmark.

> +
> > +  /* Array to hold throughput(#updates/second) result of each run.  We
> do
> > +     RUN_COUNT + 2 runs so that we can discard the highest and lowest as
> > +     outliers.  */
> > +  double throughput_count[RUN_COUNT + 2];
> > +
> > +  for (i = 0; i < RUN_COUNT + 2; i++)
> > +    {
> > +      /* Initialize counter before each run.  */
> > +      counter = 0;
> > +      gettimeofday (&ts, NULL);
> > +
> > +      for (j = 0; j < num_threads; j++)
> > +       pthread_create (&threads[j], NULL, worker, NULL);
> > +
> > +      for (j = 0; j < num_threads; j++)
> > +       pthread_join (threads[j], NULL);
> > +      gettimeofday (&te, NULL);
> > +      tsd = ts.tv_sec + ts.tv_usec / 1000000.0;
> > +      ted = te.tv_sec + te.tv_usec / 1000000.0;
> > +      td = ted - tsd;
> > +
> > +      /* Number of successful updates per second.  */
> > +      throughput_count[i] = counter / td;
> > +    }
> > +
> > +  /* Sort the results so we can discard the largest and smallest
> > +     throughput as outliers.  */
> > +  for (i = 0; i < RUN_COUNT + 1; i++)
> > +    for (j = i + 1; j < RUN_COUNT + 2; j++)
> > +      if (throughput_count[i] > throughput_count[j])
> > +       {
> > +         double temp = throughput_count[i];
> > +         throughput_count[i] = throughput_count[j];
> > +         throughput_count[j] = temp;
> > +       }
> > +
> > +  /* Calculate mean and standard deviation.  */
> > +  mean = 0;
> > +  for (i = 1; i < RUN_COUNT + 1; i++)
> > +    mean += throughput_count[i];
> > +  mean /= RUN_COUNT;
> > +
> > +  stdev = 0.0;
> > +  for (i = 1; i < RUN_COUNT + 1; i++)
> > +    {
> > +      double s = throughput_count[i] - mean;
> > +      stdev += s * s;
> > +    }
> > +  stdev = sqrt (stdev / (RUN_COUNT - 1));
> > +
> > +  char buf[256];
> > +  snprintf (buf, sizeof buf, "%s,threads=%d,HW threads=%d", name,
> num_threads,
> > +           nprocs);
> > +
> > +  json_attr_object_begin (js, buf);
> > +
> > +  json_attr_uint (js, "mean", mean);
> > +  json_attr_uint (js, "stdev", stdev);
> > +  json_attr_uint (js, "min-outlier", throughput_count[0]);
> > +  json_attr_uint (js, "min", throughput_count[1]);
> > +  json_attr_uint (js, "max", throughput_count[RUN_COUNT]);
> > +  json_attr_uint (js, "max-outlier", throughput_count[RUN_COUNT + 1]);
> > +
> > +  json_attr_object_end (js);
> > +}
> > +
> > +int
> > +do_bench (void)
> > +{
> > +  int rv = 0;
> > +  json_ctx_t json_ctx;
> > +  int th_num, th_conf, nprocs;
> > +  char name[128];
> > +
> > +  json_init (&json_ctx, 2, stdout);
> > +  json_attr_object_begin (&json_ctx, TEST_NAME);
> > +
> > +  /* The thread config begins from 2, and increases by 2x until nprocs.
> > +     We also want to test nprocs and over-saturation case 2*nprocs and
> > +     4*nprocs.  */
> > +  nprocs = get_nprocs ();
> > +
> > +  /* Allocate [nprocs + 2] thread config array.  */
> > +  int *threads = malloc ((nprocs + 2) * sizeof (int));
>
> Shouldn't it be (nprocs + 3) since you add 3 more below?
>
>
But it is first testing for  th_num < nproc, so nproc+2 will be sufficient
for all configurations.


> > +
> > +  th_num = 2;
> > +  for (th_conf = 0; th_num < nprocs; th_conf++)
> > +    {
> > +      threads[th_conf] = th_num;
> > +      th_num <<= 1;
> > +    }
> > +  threads[th_conf++] = nprocs;
> > +  threads[th_conf++] = nprocs * 2;
> > +  threads[th_conf++] = nprocs * 4;
> > +
> > +  snprintf (name, sizeof name, "type=throughput(#updates/second)");
> > +
> > +  for (int i = 0; i < th_conf; i++)
> > +    {
> > +      th_num = threads[i];
> > +      do_bench_one (name, th_num, nprocs, &json_ctx);
> > +    }
> > +
> > +  free (threads);
> > +
> > +  json_attr_object_end (&json_ctx);
> > +
> > +  return rv;
> > +}
> > +
> > +#define TEST_FUNCTION do_bench ()
> > +
> > +#include "../test-skeleton.c"
> > --
> > 2.51.0
> >
>
>
> --
> H.J.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20251212/4bf3eaf1/attachment-0001.htm>