From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-yw1-x112e.google.com (mail-yw1-x112e.google.com [IPv6:2607:f8b0:4864:20::112e]) by sourceware.org (Postfix) with ESMTPS id 5E5A53858C53 for ; Thu, 14 Apr 2022 19:17:51 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 5E5A53858C53 Received: by mail-yw1-x112e.google.com with SMTP id 00721157ae682-2ef4a241cc5so58602637b3.2 for ; Thu, 14 Apr 2022 12:17:51 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=28v6BGqUMIBAiMbGsVBe4U4Lg310QxHcR7Ae1zVm22Y=; b=ZYVCgiutrhZ5o9hChsKPz/mrR//vcx2hZk/uqs+RXesZKdszSSn4MiVm34PUiSTW78 V9mFGxy3PvWbNuiTpYZyh82PK7M69b0J/W4N2I7+cvpSoTgzX9ZbvRC3pal1MQx5BAWp HGsn5ltn1L5f/guWhXucbaHFOZMkADGlGLEyT4u/Djdi0h1x12atqST0h/QjCCkCh/GR yiyjMzrt3KiONxDHJSKAdiafWKbgxN2wLeL3UnYAPpUnAkq2B8UxZbzFO6RzeNGgYmY2 A7ut8TasyO2oiyShyq3bNz0VmiUc3rwNjGmi1RMnUv3eKDZ+CbD4E0ti55e4zg2aZe56 OxkQ== X-Gm-Message-State: AOAM5319OHEXPuVs8+V1AnkBWw/nO7LaGtEhn7KcazKL/lDZjfooj7r9 5iItylWXNDDNnQyN9/LcEH/CdKsRbGWBkTVTblvW169P X-Google-Smtp-Source: ABdhPJxG/0OoAt8lOZKcrzeCA160T1BjJLiWjKY8tAEy7ymV0S0QVbk0U2J4GRxUtEjTQPPrCvTMYxbSUZuWPvu+Wsg= X-Received: by 2002:a81:1cc6:0:b0:2d5:427c:a23e with SMTP id c189-20020a811cc6000000b002d5427ca23emr3086744ywc.496.1649963870723; Thu, 14 Apr 2022 12:17:50 -0700 (PDT) MIME-Version: 1.0 References: <20220413202401.408267-1-adhemerval.zanella@linaro.org> <20220413202401.408267-4-adhemerval.zanella@linaro.org> In-Reply-To: <20220413202401.408267-4-adhemerval.zanella@linaro.org> From: Noah Goldstein Date: Thu, 14 Apr 2022 14:17:39 -0500 Message-ID: Subject: Re: [PATCH 3/7] benchtests: Add arc4random benchtest To: Adhemerval Zanella Cc: GNU C Library Content-Type: text/plain; charset="UTF-8" X-Spam-Status: No, score=-8.3 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, KAM_SHORT, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 14 Apr 2022 19:17:53 -0000 On Wed, Apr 13, 2022 at 3:26 PM Adhemerval Zanella via Libc-alpha wrote: > > It shows both throughput (total bytes obtained in the test duration) > and latecy for both arc4random and arc4random_buf with different > sizes. > > Checked on x86_64-linux-gnu, aarch64-linux, and powerpc64le-linux-gnu. > --- > benchtests/Makefile | 6 +- > benchtests/bench-arc4random.c | 243 ++++++++++++++++++++++++++++++++++ > 2 files changed, 248 insertions(+), 1 deletion(-) > create mode 100644 benchtests/bench-arc4random.c > > diff --git a/benchtests/Makefile b/benchtests/Makefile > index 8dfca592fd..50b96dd71f 100644 > --- a/benchtests/Makefile > +++ b/benchtests/Makefile > @@ -111,8 +111,12 @@ bench-string := \ > ffsll \ > # bench-string > > +bench-stdlib := \ > + arc4random \ > +# bench-stdlib > + > ifeq (${BENCHSET},) > -bench := $(bench-math) $(bench-pthread) $(bench-string) > +bench := $(bench-math) $(bench-pthread) $(bench-string) $(bench-stdlib) > else > bench := $(foreach B,$(filter bench-%,${BENCHSET}), ${${B}}) > endif > diff --git a/benchtests/bench-arc4random.c b/benchtests/bench-arc4random.c > new file mode 100644 > index 0000000000..9e2ba9ba34 > --- /dev/null > +++ b/benchtests/bench-arc4random.c > @@ -0,0 +1,243 @@ > +/* arc4random benchmarks. > + 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 > + . */ > + > +#include "bench-timing.h" > +#include "json-lib.h" > +#include > +#include > +#include > +#include > +#include > +#include > +#include > +#include > + > +static volatile uint32_t r; > +static volatile sig_atomic_t timer_finished; > + > +static void timer_callback (int unused) > +{ > + timer_finished = 1; > +} > + > +static const uint32_t sizes[] = { 0, 16, 32, 64, 128 }; > + > +static double > +bench_arc4random_throughput (void) > +{ > + /* Run for approximately DURATION seconds, and it does not matter who > + receive the signal (so not need to mask it on main thread). */ > + timer_finished = 0; > + timer_t timer = support_create_timer (DURATION, 0, false, timer_callback); > + > + uint64_t n = 0; > + > + while (1) > + { > + r = arc4random (); > + n++; > + > + if (timer_finished == 1) > + break; > + } > + > + support_delete_timer (timer); > + > + return (double) (n * sizeof (r)) / (double) DURATION; > +} > + > +static double > +bench_arc4random_latency (void) > +{ > + timing_t start, stop, cur; > + const size_t iters = 1024; > + > + TIMING_NOW (start); > + for (size_t i = 0; i < iters; i++) > + r = arc4random (); > + TIMING_NOW (stop); > + > + TIMING_DIFF (cur, start, stop); > + > + return (double) (cur) / (double) iters; > +} > + > +static double > +bench_arc4random_buf_throughput (size_t len) > +{ > + timer_finished = 0; > + timer_t timer = support_create_timer (DURATION, 0, false, timer_callback); > + > + uint8_t buf[len]; > + > + uint64_t n = 0; > + > + while (1) > + { > + arc4random_buf (buf, len); > + n++; > + > + if (timer_finished == 1) > + break; > + } > + > + support_delete_timer (timer); > + > + uint64_t total = (n * len); > + return (double) (total) / (double) DURATION; > +} > + > +static double > +bench_arc4random_buf_latency (size_t len) > +{ > + timing_t start, stop, cur; > + const size_t iters = 1024; > + > + uint8_t buf[len]; > + > + TIMING_NOW (start); > + for (size_t i = 0; i < iters; i++) > + arc4random_buf (buf, len); > + TIMING_NOW (stop); > + > + TIMING_DIFF (cur, start, stop); > + > + return (double) (cur) / (double) iters; > +} > + > +static void > +bench_singlethread (json_ctx_t *json_ctx) > +{ > + json_element_object_begin (json_ctx); > + > + json_array_begin (json_ctx, "throughput"); > + for (int i = 0; i < array_length (sizes); i++) > + if (sizes[i] == 0) > + json_element_double (json_ctx, bench_arc4random_throughput ()); > + else > + json_element_double (json_ctx, bench_arc4random_buf_throughput (sizes[i])); > + json_array_end (json_ctx); > + > + json_array_begin (json_ctx, "latency"); > + for (int i = 0; i < array_length (sizes); i++) > + if (sizes[i] == 0) > + json_element_double (json_ctx, bench_arc4random_latency ()); > + else > + json_element_double (json_ctx, bench_arc4random_buf_latency (sizes[i])); > + json_array_end (json_ctx); > + > + json_element_object_end (json_ctx); > +} > + > +struct thr_arc4random_arg > +{ > + double ret; > + uint32_t val; > +}; > + > +static void * > +thr_arc4random_throughput (void *closure) > +{ > + struct thr_arc4random_arg *arg = closure; > + arg->ret = arg->val == 0 ? bench_arc4random_throughput () > + : bench_arc4random_buf_throughput (arg->val); > + return NULL; > +} > + > +static void * > +thr_arc4random_latency (void *closure) > +{ > + struct thr_arc4random_arg *arg = closure; > + arg->ret = arg->val == 0 ? bench_arc4random_latency () > + : bench_arc4random_buf_latency (arg->val); > + return NULL; > +} I think the expectation is that the chacha calls will be cold, maybe it is worth adding a cache flush of sorts between calls. It may be some prefetching in the start will help the code in that case but would only be a regression with the hot in L1 benchmarks. Can wait though this V1 looks fine. > + > +static void > +bench_threaded (json_ctx_t *json_ctx) > +{ > + json_element_object_begin (json_ctx); > + > + json_array_begin (json_ctx, "throughput"); > + for (int i = 0; i < array_length (sizes); i++) > + { > + struct thr_arc4random_arg arg = { .val = sizes[i] }; > + pthread_t thr = xpthread_create (NULL, thr_arc4random_throughput, &arg); > + xpthread_join (thr); > + json_element_double (json_ctx, arg.ret); > + } > + json_array_end (json_ctx); > + > + json_array_begin (json_ctx, "latency"); > + for (int i = 0; i < array_length (sizes); i++) > + { > + struct thr_arc4random_arg arg = { .val = sizes[i] }; > + pthread_t thr = xpthread_create (NULL, thr_arc4random_latency, &arg); > + xpthread_join (thr); > + json_element_double (json_ctx, arg.ret); > + } > + json_array_end (json_ctx); > + > + json_element_object_end (json_ctx); > +} > + > +static void > +run_bench (json_ctx_t *json_ctx, const char *name, > + char *const*fnames, size_t fnameslen, > + void (*bench)(json_ctx_t *ctx)) > +{ > + json_attr_object_begin (json_ctx, name); > + json_array_begin (json_ctx, "functions"); > + for (int i = 0; i < fnameslen; i++) > + json_element_string (json_ctx, fnames[i]); > + json_array_end (json_ctx); > + > + json_array_begin (json_ctx, "results"); > + bench (json_ctx); > + json_array_end (json_ctx); > + json_attr_object_end (json_ctx); > +} > + > +static int > +do_test (void) > +{ > + char *fnames[array_length (sizes) + 1]; > + fnames[0] = (char *) "arc4random"; > + for (int i = 0; i < array_length (sizes); i++) > + fnames[i+1] = xasprintf ("arc4random_buf(%u)", sizes[i]); > + > + json_ctx_t json_ctx; > + json_init (&json_ctx, 0, stdout); > + > + json_document_begin (&json_ctx); > + json_attr_string (&json_ctx, "timing_type", TIMING_TYPE); > + > + run_bench (&json_ctx, "single-thread", fnames, array_length (fnames), > + bench_singlethread); > + run_bench (&json_ctx, "multi-thread", fnames, array_length (fnames), > + bench_threaded); > + > + json_document_end (&json_ctx); > + > + for (int i = 0; i < array_length (sizes); i++) > + free (fnames[i+1]); > + > + return 0; > +} > + > +#include > -- > 2.32.0 >