From patchwork Mon Aug 28 07:26:46 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: dengjianbo X-Patchwork-Id: 1826628 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@legolas.ozlabs.org Authentication-Results: legolas.ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=sourceware.org (client-ip=2620:52:3:1:0:246e:9693:128c; helo=server2.sourceware.org; envelope-from=libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org; receiver=patchwork.ozlabs.org) Received: from server2.sourceware.org (server2.sourceware.org [IPv6:2620:52:3:1:0:246e:9693:128c]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (secp384r1) server-digest SHA384) (No client certificate requested) by legolas.ozlabs.org (Postfix) with ESMTPS id 4RZ2FD3pP0z1yfX for ; Mon, 28 Aug 2023 17:27:12 +1000 (AEST) Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 94D293858002 for ; Mon, 28 Aug 2023 07:27:10 +0000 (GMT) X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail.loongson.cn (mail.loongson.cn [114.242.206.163]) by sourceware.org (Postfix) with ESMTP id CCF573858D35 for ; Mon, 28 Aug 2023 07:26:56 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org CCF573858D35 Authentication-Results: sourceware.org; dmarc=none (p=none dis=none) header.from=loongson.cn Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=loongson.cn Received: from loongson.cn (unknown [10.2.5.5]) by gateway (Coremail) with SMTP id _____8BxIvA+TOxkYnUcAA--.57276S3; Mon, 28 Aug 2023 15:26:54 +0800 (CST) Received: from 5.5.5 (unknown [10.2.5.5]) by localhost.localdomain (Coremail) with SMTP id AQAAf8Dx4eQ9TOxkBYBlAA--.49174S3; Mon, 28 Aug 2023 15:26:54 +0800 (CST) From: dengjianbo To: libc-alpha@sourceware.org Subject: [PATCH 1/6] LoongArch: Add ifunc support for rawmemchr{aligned, lsx, lasx} Date: Mon, 28 Aug 2023 15:26:46 +0800 Message-Id: <20230828072651.3085034-2-dengjianbo@loongson.cn> X-Mailer: git-send-email 2.31.1 In-Reply-To: <20230828072651.3085034-1-dengjianbo@loongson.cn> References: <20230828072651.3085034-1-dengjianbo@loongson.cn> MIME-Version: 1.0 X-CM-TRANSID: AQAAf8Dx4eQ9TOxkBYBlAA--.49174S3 X-CM-SenderInfo: pghqwyxldqu0o6or00hjvr0hdfq/ X-Coremail-Antispam: 1Uk129KBj9fXoW3ZrW7KFy7Xr4kKr1rWFy8JFc_yoW8JFyUZo WftF4DXrs2krs8KrZ8CrsrX39ruF1fKr1jv3yYva1rJry8trW7CFWfCwnIkFsrZrn5WrWr XasrX3sxJrWxGFn3l-sFpf9Il3svdjkaLaAFLSUrUUUUbb8apTn2vfkv8UJUUUU8wcxFpf 9Il3svdxBIdaVrn0xqx4xG64xvF2IEw4CE5I8CrVC2j2Jv73VFW2AGmfu7bjvjm3AaLaJ3 UjIYCTnIWjp_UUUYI7kC6x804xWl14x267AKxVWUJVW8JwAFc2x0x2IEx4CE42xK8VAvwI 8IcIk0rVWrJVCq3wAFIxvE14AKwVWUGVWUXwA2ocxC64kIII0Yj41l84x0c7CEw4AK67xG Y2AK021l84ACjcxK6xIIjxv20xvE14v26r4j6ryUM28EF7xvwVC0I7IYx2IY6xkF7I0E14 v26r4j6F4UM28EF7xvwVC2z280aVAFwI0_Gr1j6F4UJwA2z4x0Y4vEx4A2jsIEc7CjxVAF wI0_Gr1j6F4UJwAS0I0E0xvYzxvE52x082IY62kv0487Mc804VCY07AIYIkI8VC2zVCFFI 0UMc02F40EFcxC0VAKzVAqx4xG6I80ewAv7VC0I7IYx2IY67AKxVWUXVWUAwAv7VC2z280 aVAFwI0_Jr0_Gr1lOx8S6xCaFVCjc4AY6r1j6r4UM4x0Y48IcxkI7VAKI48JMxAIw28Icx kI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMxCIbckI1I0E14v26r126r1DMI8I3I0E 5I8CrVAFwI0_Jr0_Jr4lx2IqxVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUAV WUtwCIc40Y0x0EwIxGrwCI42IY6xIIjxv20xvE14v26r1j6r1xMIIF0xvE2Ix0cI8IcVCY 1x0267AKxVWUJVW8JwCI42IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI 0_Jr0_Gr1lIxAIcVC2z280aVCY1x0267AKxVWUJVW8JbIYCTnIWIevJa73UjIFyTuYvjxU 7XTmDUUUU X-Spam-Status: No, score=-10.8 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_STATUS, KAM_SHORT, KAM_STOCKGEN, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.30 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: caiyinyu@loongson.cn, xuchenghua@loongson.cn, huangpei@loongson.cn, dengjianbo Errors-To: libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org Sender: "Libc-alpha" According to glibc rawmemchr microbenchmark, A few cases tested with char '\0' experience performance degradation due to the lasx and lsx versions don't handle the '\0' separately. Overall, rawmemchr-lasx implementation could reduce the runtime about 40%-80%, rawmemchr-lsx implementation could reduce the runtime about 40%-66%, rawmemchr-aligned implementation could reduce the runtime about 20%-40%. --- sysdeps/loongarch/lp64/multiarch/Makefile | 3 + .../lp64/multiarch/ifunc-impl-list.c | 8 ++ .../lp64/multiarch/ifunc-rawmemchr.h | 40 ++++++ .../lp64/multiarch/rawmemchr-aligned.S | 124 ++++++++++++++++++ .../loongarch/lp64/multiarch/rawmemchr-lasx.S | 82 ++++++++++++ .../loongarch/lp64/multiarch/rawmemchr-lsx.S | 71 ++++++++++ sysdeps/loongarch/lp64/multiarch/rawmemchr.c | 37 ++++++ 7 files changed, 365 insertions(+) create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-rawmemchr.h create mode 100644 sysdeps/loongarch/lp64/multiarch/rawmemchr-aligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/rawmemchr-lasx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/rawmemchr-lsx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/rawmemchr.c diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile index 5d7ae7ae73..64416b025a 100644 --- a/sysdeps/loongarch/lp64/multiarch/Makefile +++ b/sysdeps/loongarch/lp64/multiarch/Makefile @@ -21,5 +21,8 @@ sysdep_routines += \ memmove-unaligned \ memmove-lsx \ memmove-lasx \ + rawmemchr-aligned \ + rawmemchr-lsx \ + rawmemchr-lasx \ # sysdep_routines endif diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c index c8ba87bd81..3db9af1460 100644 --- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c @@ -94,5 +94,13 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_aligned) ) + IFUNC_IMPL (i, name, rawmemchr, +#if !defined __loongarch_soft_float + IFUNC_IMPL_ADD (array, i, rawmemchr, SUPPORT_LASX, __rawmemchr_lasx) + IFUNC_IMPL_ADD (array, i, rawmemchr, SUPPORT_LSX, __rawmemchr_lsx) +#endif + IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_aligned) + ) + return i; } diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-rawmemchr.h b/sysdeps/loongarch/lp64/multiarch/ifunc-rawmemchr.h new file mode 100644 index 0000000000..a7bb4cf9ea --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-rawmemchr.h @@ -0,0 +1,40 @@ +/* Common definition for rawmemchr ifunc selections. + All versions must be listed in ifunc-impl-list.c. + Copyright (C) 2023 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 +#include + +#if !defined __loongarch_soft_float +extern __typeof (REDIRECT_NAME) OPTIMIZE (lasx) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (lsx) attribute_hidden; +#endif +extern __typeof (REDIRECT_NAME) OPTIMIZE (aligned) attribute_hidden; + +static inline void * +IFUNC_SELECTOR (void) +{ +#if !defined __loongarch_soft_float + if (SUPPORT_LASX) + return OPTIMIZE (lasx); + else if (SUPPORT_LSX) + return OPTIMIZE (lsx); + else +#endif + return OPTIMIZE (aligned); +} diff --git a/sysdeps/loongarch/lp64/multiarch/rawmemchr-aligned.S b/sysdeps/loongarch/lp64/multiarch/rawmemchr-aligned.S new file mode 100644 index 0000000000..9c7155ae82 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/rawmemchr-aligned.S @@ -0,0 +1,124 @@ +/* Optimized rawmemchr implementation using basic LoongArch instructions. + Copyright (C) 2023 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 +#include +#include + +#if IS_IN (libc) +# define RAWMEMCHR_NAME __rawmemchr_aligned +#else +# define RAWMEMCHR_NAME __rawmemchr +#endif + +LEAF(RAWMEMCHR_NAME, 6) + andi t1, a0, 0x7 + bstrins.d a0, zero, 2, 0 + lu12i.w a2, 0x01010 + bstrins.d a1, a1, 15, 8 + + ld.d t0, a0, 0 + slli.d t1, t1, 3 + ori a2, a2, 0x101 + bstrins.d a1, a1, 31, 16 + + li.w t8, -1 + bstrins.d a1, a1, 63, 32 + bstrins.d a2, a2, 63, 32 + sll.d t2, t8, t1 + + sll.d t3, a1, t1 + orn t0, t0, t2 + slli.d a3, a2, 7 + beqz a1, L(find_zero) + + xor t0, t0, t3 + sub.d t1, t0, a2 + andn t2, a3, t0 + and t3, t1, t2 + + bnez t3, L(count_pos) + addi.d a0, a0, 8 + +L(loop): + ld.d t0, a0, 0 + xor t0, t0, a1 + + sub.d t1, t0, a2 + andn t2, a3, t0 + and t3, t1, t2 + bnez t3, L(count_pos) + + ld.d t0, a0, 8 + addi.d a0, a0, 16 + xor t0, t0, a1 + sub.d t1, t0, a2 + + andn t2, a3, t0 + and t3, t1, t2 + beqz t3, L(loop) + addi.d a0, a0, -8 +L(count_pos): + ctz.d t0, t3 + srli.d t0, t0, 3 + add.d a0, a0, t0 + jr ra + +L(loop_7bit): + ld.d t0, a0, 0 +L(find_zero): + sub.d t1, t0, a2 + and t2, t1, a3 + bnez t2, L(more_check) + + ld.d t0, a0, 8 + addi.d a0, a0, 16 + sub.d t1, t0, a2 + and t2, t1, a3 + + beqz t2, L(loop_7bit) + addi.d a0, a0, -8 + +L(more_check): + andn t2, a3, t0 + and t3, t1, t2 + bnez t3, L(count_pos) + addi.d a0, a0, 8 + +L(loop_8bit): + ld.d t0, a0, 0 + + sub.d t1, t0, a2 + andn t2, a3, t0 + and t3, t1, t2 + bnez t3, L(count_pos) + + ld.d t0, a0, 8 + addi.d a0, a0, 16 + sub.d t1, t0, a2 + + andn t2, a3, t0 + and t3, t1, t2 + beqz t3, L(loop_8bit) + + addi.d a0, a0, -8 + b L(count_pos) + +END(RAWMEMCHR_NAME) + +libc_hidden_builtin_def (__rawmemchr) diff --git a/sysdeps/loongarch/lp64/multiarch/rawmemchr-lasx.S b/sysdeps/loongarch/lp64/multiarch/rawmemchr-lasx.S new file mode 100644 index 0000000000..be2eb59dbe --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/rawmemchr-lasx.S @@ -0,0 +1,82 @@ +/* Optimized rawmemchr implementation using LoongArch LASX instructions. + Copyright (C) 2023 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 +#include +#include + +#if IS_IN (libc) && !defined __loongarch_soft_float + +# define RAWMEMCHR __rawmemchr_lasx + +LEAF(RAWMEMCHR, 6) + move a2, a0 + bstrins.d a0, zero, 5, 0 + xvld xr0, a0, 0 + xvld xr1, a0, 32 + + xvreplgr2vr.b xr2, a1 + xvseq.b xr0, xr0, xr2 + xvseq.b xr1, xr1, xr2 + xvmsknz.b xr0, xr0 + + xvmsknz.b xr1, xr1 + xvpickve.w xr3, xr0, 4 + xvpickve.w xr4, xr1, 4 + vilvl.h vr0, vr3, vr0 + + vilvl.h vr1, vr4, vr1 + vilvl.w vr0, vr1, vr0 + movfr2gr.d t0, fa0 + sra.d t0, t0, a2 + + + beqz t0, L(loop) + ctz.d t0, t0 + add.d a0, a2, t0 + jr ra + +L(loop): + xvld xr0, a0, 64 + xvld xr1, a0, 96 + addi.d a0, a0, 64 + xvseq.b xr0, xr0, xr2 + + xvseq.b xr1, xr1, xr2 + xvmax.bu xr3, xr0, xr1 + xvseteqz.v fcc0, xr3 + bcnez fcc0, L(loop) + + xvmsknz.b xr0, xr0 + xvmsknz.b xr1, xr1 + xvpickve.w xr3, xr0, 4 + xvpickve.w xr4, xr1, 4 + + + vilvl.h vr0, vr3, vr0 + vilvl.h vr1, vr4, vr1 + vilvl.w vr0, vr1, vr0 + movfr2gr.d t0, fa0 + + ctz.d t0, t0 + add.d a0, a0, t0 + jr ra +END(RAWMEMCHR) + +libc_hidden_builtin_def (RAWMEMCHR) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/rawmemchr-lsx.S b/sysdeps/loongarch/lp64/multiarch/rawmemchr-lsx.S new file mode 100644 index 0000000000..2f6fe024dc --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/rawmemchr-lsx.S @@ -0,0 +1,71 @@ +/* Optimized rawmemchr implementation using LoongArch LSX instructions. + Copyright (C) 2023 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 +#include +#include + +#if IS_IN (libc) && !defined __loongarch_soft_float + +# define RAWMEMCHR __rawmemchr_lsx + +LEAF(RAWMEMCHR, 6) + move a2, a0 + bstrins.d a0, zero, 4, 0 + vld vr0, a0, 0 + vld vr1, a0, 16 + + vreplgr2vr.b vr2, a1 + vseq.b vr0, vr0, vr2 + vseq.b vr1, vr1, vr2 + vmsknz.b vr0, vr0 + + vmsknz.b vr1, vr1 + vilvl.h vr0, vr1, vr0 + movfr2gr.s t0, fa0 + sra.w t0, t0, a2 + + beqz t0, L(loop) + ctz.w t0, t0 + add.d a0, a2, t0 + jr ra + + +L(loop): + vld vr0, a0, 32 + vld vr1, a0, 48 + addi.d a0, a0, 32 + vseq.b vr0, vr0, vr2 + + vseq.b vr1, vr1, vr2 + vmax.bu vr3, vr0, vr1 + vseteqz.v fcc0, vr3 + bcnez fcc0, L(loop) + + vmsknz.b vr0, vr0 + vmsknz.b vr1, vr1 + vilvl.h vr0, vr1, vr0 + movfr2gr.s t0, fa0 + + ctz.w t0, t0 + add.d a0, a0, t0 + jr ra +END(RAWMEMCHR) + +libc_hidden_builtin_def (RAWMEMCHR) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/rawmemchr.c b/sysdeps/loongarch/lp64/multiarch/rawmemchr.c new file mode 100644 index 0000000000..89c7ffff8f --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/rawmemchr.c @@ -0,0 +1,37 @@ +/* Multiple versions of rawmemchr. + All versions must be listed in ifunc-impl-list.c. + Copyright (C) 2023 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 + . */ + +#if IS_IN (libc) +# define rawmemchr __redirect_rawmemchr +# define __rawmemchr __redirect___rawmemchr +# include +# undef rawmemchr +# undef __rawmemchr + +# define SYMBOL_NAME rawmemchr +# include "ifunc-rawmemchr.h" + +libc_ifunc_redirected (__redirect_rawmemchr, __rawmemchr, + IFUNC_SELECTOR ()); +weak_alias (__rawmemchr, rawmemchr) +# ifdef SHARED +__hidden_ver1 (__rawmemchr, __GI___rawmemchr, __redirect___rawmemchr) + __attribute__((visibility ("hidden"))); +# endif +#endif