From patchwork Wed Sep 22 01:05:58 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Naohiro Tamura X-Patchwork-Id: 45265 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 05EA53858402 for ; Wed, 22 Sep 2021 01:06:39 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 05EA53858402 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1632272799; bh=yhqNQbM5p/+5UkvR+o7pCH6wBnm6vl+0g0NNQPGcXdY=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=O/YLCKwx2SVTbNb92RMMzeI+ChRYbont+Lyk4I2ctvDyl4MVWQDcAhKa2N7iORhjo +p8cAEdmNNiYAsHynORhlfXDW7MG6x0MYoVpOp40rHqq5TAPfVYtBplUhsr+4UJG67 lCqkmDUYcQ8eGJMmDTgOY0/MJyS2rGsgpkoRhcb8= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from esa1.hc1455-7.c3s2.iphmx.com (esa1.hc1455-7.c3s2.iphmx.com [207.54.90.47]) by sourceware.org (Postfix) with ESMTPS id D77D93858D39 for ; Wed, 22 Sep 2021 01:06:16 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org D77D93858D39 IronPort-SDR: ew6D0PSIqk2c74VNc2ut3RfGooVAoArJc7vpfhEqx/ufT0Uo6VOlzX6CGj7y7Q/IRe9+nCH9xf 5vB8duVaesyJMqklYJSsWPw/lFqFEmgqzF+OWz3lQvkLX+YRhyXtLg5KPkqHa6vohiEA/BUjlL plE99jUsInLwnii0WtDrmSRjZWsgaGWM1S4DLQMwDXSGd0Y00X2m7cCeirVT6S8oMV6L3X2iq6 yRH1Sw/OxiYWghKnhkbjeW1kW5S4+Hdby6m1teqLp7SW9EXN7MqA1xoVQe9yWyGZ3V8PIvvpXu OOZB9FenH6q03721wAooDQJr X-IronPort-AV: E=McAfee;i="6200,9189,10114"; a="46040154" X-IronPort-AV: E=Sophos;i="5.85,312,1624287600"; d="scan'208";a="46040154" Received: from unknown (HELO yto-r4.gw.nic.fujitsu.com) ([218.44.52.220]) by esa1.hc1455-7.c3s2.iphmx.com with ESMTP; 22 Sep 2021 10:06:15 +0900 Received: from yto-m2.gw.nic.fujitsu.com (yto-nat-yto-m2.gw.nic.fujitsu.com [192.168.83.65]) by yto-r4.gw.nic.fujitsu.com (Postfix) with ESMTP id 2AA586CCA9 for ; Wed, 22 Sep 2021 10:06:14 +0900 (JST) Received: from m3051.s.css.fujitsu.com (m3051.s.css.fujitsu.com [10.134.21.209]) by yto-m2.gw.nic.fujitsu.com (Postfix) with ESMTP id 6E16C9B0B7 for ; Wed, 22 Sep 2021 10:06:13 +0900 (JST) Received: from bionic.lxd (unknown [10.126.53.116]) by m3051.s.css.fujitsu.com (Postfix) with ESMTP id 5CAED9B; Wed, 22 Sep 2021 10:06:13 +0900 (JST) To: Wilco Dijkstra , "'Lucas A . M . Magalhaes'" , Noah Goldstein , libc-alpha@sourceware.org Subject: [PATCH v5] benchtests: Add memset zero fill benchtest Date: Wed, 22 Sep 2021 01:05:58 +0000 Message-Id: <20210922010558.1145487-1-naohirot@fujitsu.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: References: X-TM-AS-GCONF: 00 X-Spam-Status: No, score=-12.2 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_NONE, KAM_DMARC_STATUS, KAM_LAZY_DOMAIN_SECURITY, KAM_SHORT, SPF_HELO_PASS, SPF_NONE, TXREP 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-Patchwork-Original-From: Naohiro Tamura via Libc-alpha From: Naohiro Tamura Reply-To: Naohiro Tamura Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" Memset takes 0 as the second parameter in most cases. However, we cannot measure the zero fill performance by bench-memset.c, bench-memset-large.c and bench-memset-walk.c precisely. X86_64 micro-architecture has some zero-over-zero optimization, and AArch64 micro-architecture also has some optimization for DC ZVA instruction. This patch provides bench-memset-zerofill.c which is suitable to analyze the zero fill performance by comparing among 4 patterns, zero-over-zero, zero-over-one, one-over-zero and one-over-one through L1, L2 and L3 caches. The following commands are examples to analyze a JSON output, bench-memset-zerofill.out, by 'jq' and 'plot_strings.py'. 1) compare zero-over-zero performance $ cat bench-memset-zerofill.out | \ jq -r ' .functions.memset."bench-variant"="zerofill-0o0" | del(.functions.memset.results[] | select(.char1 != 0 or .char2 != 0)) ' | \ plot_strings.py -l -p thru -v - 2) compare zero paformance $ cat bench-memset-zerofill.out | \ jq -r ' .functions.memset."bench-variant"="zerofill-zero" | del(.functions.memset.results[] | select(.char2 != 0)) ' | \ plot_strings.py -l -p thru -v - 3) compare nonzero paformance $ cat bench-memset-zerofill.out | \ jq -r ' .functions.memset."bench-variant"="zerofill-nonzero" | del(.functions.memset.results[] | select(.char2 == 0)) ' | \ plot_strings.py -l -p thru -v - Reviewed-by: Lucas A. M. Magalhaes Reviewed-by: Noah Goldstein Reviewed-by: Wilco Dijkstra --- benchtests/Makefile | 2 +- benchtests/bench-memset-zerofill.c | 140 +++++++++++++++++++++++++++++ 2 files changed, 141 insertions(+), 1 deletion(-) create mode 100644 benchtests/bench-memset-zerofill.c diff --git a/benchtests/Makefile b/benchtests/Makefile index 1530939a8ce8..21b95c736190 100644 --- a/benchtests/Makefile +++ b/benchtests/Makefile @@ -53,7 +53,7 @@ string-benchset := memccpy memchr memcmp memcpy memmem memmove \ strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \ strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok \ strcoll memcpy-large memcpy-random memmove-large memset-large \ - memcpy-walk memset-walk memmove-walk + memcpy-walk memset-walk memmove-walk memset-zerofill # Build and run locale-dependent benchmarks only if we're building natively. ifeq (no,$(cross-compiling)) diff --git a/benchtests/bench-memset-zerofill.c b/benchtests/bench-memset-zerofill.c new file mode 100644 index 000000000000..2e146e157e25 --- /dev/null +++ b/benchtests/bench-memset-zerofill.c @@ -0,0 +1,140 @@ +/* Measure memset functions with zero fill data. + Copyright (C) 2021 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 + . */ + +#define TEST_MAIN +#define TEST_NAME "memset" +#define START_SIZE (16 * 1024) +#define MIN_PAGE_SIZE (getpagesize () + 16 * 1024 * 1024) +#define TIMEOUT (20 * 60) +#include "bench-string.h" + +#include "json-lib.h" + +void *generic_memset (void *, int, size_t); +typedef void *(*proto_t) (void *, int, size_t); + +IMPL (MEMSET, 1) +IMPL (generic_memset, 0) + +static void +__attribute__((noinline, noclone)) +do_one_test (json_ctx_t *json_ctx, impl_t *impl, CHAR *s, + int c1 __attribute ((unused)), int c2 __attribute ((unused)), + size_t n) +{ + size_t i, iters = 32; + timing_t start, stop, cur, latency = 0; + + CALL (impl, s, c2, n); // warm up + + for (i = 0; i < iters; i++) + { + memset (s, c1, n); // alternation + + TIMING_NOW (start); + + CALL (impl, s, c2, n); + + TIMING_NOW (stop); + TIMING_DIFF (cur, start, stop); + TIMING_ACCUM (latency, cur); + } + + json_element_double (json_ctx, (double) latency / (double) iters); +} + +static void +do_test (json_ctx_t *json_ctx, size_t align, int c1, int c2, size_t len) +{ + align &= getpagesize () - 1; + if ((align + len) * sizeof (CHAR) > page_size) + return; + + json_element_object_begin (json_ctx); + json_attr_uint (json_ctx, "length", len); + json_attr_uint (json_ctx, "alignment", align); + json_attr_int (json_ctx, "char1", c1); + json_attr_int (json_ctx, "char2", c2); + json_array_begin (json_ctx, "timings"); + + FOR_EACH_IMPL (impl, 0) + { + do_one_test (json_ctx, impl, (CHAR *) (buf1) + align, c1, c2, len); + alloc_bufs (); + } + + json_array_end (json_ctx); + json_element_object_end (json_ctx); +} + +int +test_main (void) +{ + json_ctx_t json_ctx; + size_t i; + int c1, c2; + + test_init (); + + 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, TEST_NAME); + json_attr_string (&json_ctx, "bench-variant", "zerofill"); + + json_array_begin (&json_ctx, "ifuncs"); + FOR_EACH_IMPL (impl, 0) + json_element_string (&json_ctx, impl->name); + json_array_end (&json_ctx); + + json_array_begin (&json_ctx, "results"); + + // To analyze zero fill performance by comparing among the following 4 + // patterns through L1, L2 and L3 caches. + // - zero-over-zero: c1=0, c2=0 + // - zero-over-one: c1=0, c2=1 + // - one-over-zero: c1=1, c2=0 + // - one-over-one: c1=1, c2=1 + for (c1 = 0; c1 < 2; c1++) + for (c2 = 0; c2 < 2; c2++) + for (i = START_SIZE; i <= MIN_PAGE_SIZE; i <<= 1) + { + do_test (&json_ctx, 0, c1, c2, i); + do_test (&json_ctx, 3, c1, c2, i); + } + + json_array_end (&json_ctx); + json_attr_object_end (&json_ctx); + json_attr_object_end (&json_ctx); + json_document_end (&json_ctx); + + return ret; +} + +#include + +#define libc_hidden_builtin_def(X) +#define libc_hidden_def(X) +#define libc_hidden_weak(X) +#define weak_alias(X,Y) +#undef MEMSET +#define MEMSET generic_memset +#include