[PATCH v2] Add random locking benchmark
Carlos O'Donell
carlos@redhat.com
Mon Aug 22 21:13:46 GMT 2022
On 8/22/22 11:14, Wilco Dijkstra via Libc-alpha wrote:
> v2: rename as suggested, use random throughout
>
> Add a simple benchmark to measure the overhead of internal libc
> locks in the random() implementation on both single- and
> multi-threaded cases. This relies on the implementation of random
> using internal locks to access shared global data.
Suggest:
... This relies on the implementation of random
using internal locks to access shared global data, and that the runtime
enables locking once a single thread has been created.
~~~
I think this needs a few more explanatory comments. A lot of other people
look at our benchmarks and might scratch their heads at this one.
Post a v3 and I'll ACK that.
>
> ---
>
> diff --git a/benchtests/Makefile b/benchtests/Makefile
> index d99771be74b40f8afa3953f61c0721b19658d4b7..c413eac1d23568cb88bf22c6e50303e24ec10ea0 100644
> --- a/benchtests/Makefile
> +++ b/benchtests/Makefile
> @@ -236,6 +236,7 @@ hash-benchset := \
> stdlib-benchset := \
> arc4random \
> strtod \
> + random-lock \
OK.
> # stdlib-benchset
>
> stdio-common-benchset := sprintf
> diff --git a/benchtests/bench-random-lock.c b/benchtests/bench-random-lock.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..13c8e77cbd3538bb1a15f81def0a2c41a3763729
> --- /dev/null
> +++ b/benchtests/bench-random-lock.c
> @@ -0,0 +1,102 @@
> +/* Benchmark internal libc locking functions used in random.
OK.
> + Copyright (C) 2022 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 TEST_NAME "random-lock"
> +
> +#include <pthread.h>
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include "bench-timing.h"
> +#include "json-lib.h"
> +
We could run the benchmark for a fixed amount of time, rather than a
fixed number of iterations, see bench-skeleton.c and DURATION.
Suggest:
/* Arbitrarily pick 20M such that we can measure the time taken. */
> +#define NUM_ITERS 20000000
OK.
> +
> +json_ctx_t json_ctx;
> +
> +
> +/* Measure the overhead of __libc_lock_lock and __libc_lock_unlock by
> + calling random (). */
> +static void
> +bench_random_lock (size_t iters)
> +{
> + timing_t start, stop, total;
> +
> + srandom (0);
OK. Seed to 0.
Suggest:
/* Warmup. Arbitrarily pick 1/4 of iterations. */
> + for (int i = 0; i < iters / 4; i++)
> + (void) random ();
> +
> + TIMING_NOW (start);
> +
> + for (int i = 0; i < iters; i++)
> + (void) random ();
> +
> + TIMING_NOW (stop);
> +
> + TIMING_DIFF (total, start, stop);
> +
> + json_element_double (&json_ctx, (double) total / (double) iters);
> +}
> +
> +static void *
> +thread_start (void *p)
> +{
> + return p;
> +}
> +
> +int
> +do_bench (void)
> +{
> + json_init (&json_ctx, 0, stdout);
> +
> + json_document_begin (&json_ctx);
> +
> + json_attr_string (&json_ctx, "timing_type", TIMING_TYPE);
> + json_attr_object_begin (&json_ctx, "functions");
> + json_attr_object_begin (&json_ctx, "random");
> + json_attr_string (&json_ctx, "bench-variant", "single-threaded");
> + json_array_begin (&json_ctx, "results");
> +
> + /* Run benchmark single threaded. */
> + bench_random_lock (NUM_ITERS);
OK.
> +
> + json_array_end (&json_ctx);
> + json_attr_object_end (&json_ctx);
> +
> + json_attr_object_begin (&json_ctx, "random");
> + json_attr_string (&json_ctx, "bench-variant", "multi-threaded");
> + json_array_begin (&json_ctx, "results");
> +
> + pthread_t t;
Suggest:
/* We assume the runtimes decision to keep the library multi-threaded
once a single thread has launched, and not change back to
single-threaded. */
Provides a little more context to this section, even though
you mention 'SINGLE_THREAD_P == false' later.
> + pthread_create (&t, NULL, thread_start, NULL);
> + pthread_join (t, NULL);
OK. We are now in multi-threaded and will never switch back to the
single threaded case.
> +
> + /* Repeat benchmark now SINGLE_THREAD_P == false. */
> + bench_random_lock (NUM_ITERS);
> +
> + json_array_end (&json_ctx);
> + json_attr_object_end (&json_ctx);
> + json_attr_object_end (&json_ctx);
> + json_document_end (&json_ctx);
> + return 0;
> +}
> +
> +#define TEST_FUNCTION do_bench ()
> +
> +#include "../test-skeleton.c"
> +
>
>
>
>
--
Cheers,
Carlos.
More information about the Libc-alpha
mailing list