From patchwork Tue Aug 15 09:43:35 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: dengjianbo X-Patchwork-Id: 1821308 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=) 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 (P-384) server-digest SHA384) (No client certificate requested) by legolas.ozlabs.org (Postfix) with ESMTPS id 4RQ5v53sGcz1yfS for ; Tue, 15 Aug 2023 19:44:01 +1000 (AEST) Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 8E26D385C6E5 for ; Tue, 15 Aug 2023 09:43:59 +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 BBE993857C5A for ; Tue, 15 Aug 2023 09:43:41 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org BBE993857C5A 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 _____8Cx7+vKSNtkHLYYAA--.49786S3; Tue, 15 Aug 2023 17:43:38 +0800 (CST) Received: from 5.5.5 (unknown [10.2.5.5]) by localhost.localdomain (Coremail) with SMTP id AQAAf8DxfSPJSNtkThBbAA--.45023S3; Tue, 15 Aug 2023 17:43:38 +0800 (CST) From: dengjianbo To: libc-alpha@sourceware.org Cc: adhemerval.zanella@linaro.org, xry111@xry111.site, caiyinyu@loongson.cn, xuchenghua@loongson.cn, huangpei@loongson.cn, dengjianbo Subject: [PATCH 1/2] Loongarch: Add ifunc support for strchr{aligned, lsx, lasx} and strchrnul{aligned, lsx, lasx} Date: Tue, 15 Aug 2023 17:43:35 +0800 Message-Id: <20230815094336.1722160-2-dengjianbo@loongson.cn> X-Mailer: git-send-email 2.31.1 In-Reply-To: <20230815094336.1722160-1-dengjianbo@loongson.cn> References: <20230815094336.1722160-1-dengjianbo@loongson.cn> MIME-Version: 1.0 X-CM-TRANSID: AQAAf8DxfSPJSNtkThBbAA--.45023S3 X-CM-SenderInfo: pghqwyxldqu0o6or00hjvr0hdfq/ X-Coremail-Antispam: 1Uk129KBj9fXoWfurWrCrW7uFWfArW5KFyDurX_yoW8uw18Co WSqF4DJrs2ka9xG395Crn7W3srWr1Sgr4jq3yjv3Z5Jr1xtry5GryFyasxKrZ7Grn5WFs5 Xa4xZr9xArZrGFn3l-sFpf9Il3svdjkaLaAFLSUrUUUUUb8apTn2vfkv8UJUUUU8wcxFpf 9Il3svdxBIdaVrn0xqx4xG64xvF2IEw4CE5I8CrVC2j2Jv73VFW2AGmfu7bjvjm3AaLaJ3 UjIYCTnIWjp_UUUY17kC6x804xWl14x267AKxVWUJVW8JwAFc2x0x2IEx4CE42xK8VAvwI 8IcIk0rVWrJVCq3wAFIxvE14AKwVWUGVWUXwA2ocxC64kIII0Yj41l84x0c7CEw4AK67xG Y2AK021l84ACjcxK6xIIjxv20xvE14v26r1I6r4UM28EF7xvwVC0I7IYx2IY6xkF7I0E14 v26r1j6r4UM28EF7xvwVC2z280aVAFwI0_Gr1j6F4UJwA2z4x0Y4vEx4A2jsIEc7CjxVAF wI0_Gr1j6F4UJwAS0I0E0xvYzxvE52x082IY62kv0487Mc804VCY07AIYIkI8VC2zVCFFI 0UMc02F40EFcxC0VAKzVAqx4xG6I80ewAv7VC0I7IYx2IY67AKxVWUXVWUAwAv7VC2z280 aVAFwI0_Jr0_Gr1lOx8S6xCaFVCjc4AY6r1j6r4UM4x0Y48IcxkI7VAKI48JMxAIw28Icx kI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMI8I3I0E5I8CrVAFwI0_Jr0_Jr4lx2Iq xVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUAVWUtwCIc40Y0x0EwIxGrwCI42 IY6xIIjxv20xvE14v26r1j6r1xMIIF0xvE2Ix0cI8IcVCY1x0267AKxVWUJVW8JwCI42IY 6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI0_Jr0_Gr1lIxAIcVC2z280aV CY1x0267AKxVWUJVW8JbIYCTnIWIevJa73UjIFyTuYvjxU7XTmDUUUU 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: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org Sender: "Libc-alpha" These implementations improve the time to run strchr{nul} microbenchmark in glibc as below: strchr-lasx reduces the runtime about 50%-83% strchr-lsx reduces the runtime about 30%-67% strchr-aligned reduces the runtime about 10%-20% strchrnul-lasx reduces the runtime about 50%-83% strchrnul-lsx reduces the runtime about 36%-65% strchrnul-aligned reduces the runtime about 6%-10% --- sysdeps/loongarch/lp64/multiarch/Makefile | 6 ++ .../lp64/multiarch/ifunc-impl-list.c | 16 +++ .../loongarch/lp64/multiarch/ifunc-strchr.h | 41 ++++++++ .../lp64/multiarch/ifunc-strchrnul.h | 41 ++++++++ .../loongarch/lp64/multiarch/strchr-aligned.S | 99 +++++++++++++++++++ .../loongarch/lp64/multiarch/strchr-lasx.S | 91 +++++++++++++++++ sysdeps/loongarch/lp64/multiarch/strchr-lsx.S | 73 ++++++++++++++ sysdeps/loongarch/lp64/multiarch/strchr.c | 36 +++++++ .../lp64/multiarch/strchrnul-aligned.S | 95 ++++++++++++++++++ .../loongarch/lp64/multiarch/strchrnul-lasx.S | 22 +++++ .../loongarch/lp64/multiarch/strchrnul-lsx.S | 22 +++++ sysdeps/loongarch/lp64/multiarch/strchrnul.c | 39 ++++++++ 12 files changed, 581 insertions(+) create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-strchr.h create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-strchrnul.h create mode 100644 sysdeps/loongarch/lp64/multiarch/strchr-aligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchr-lasx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchr-lsx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchr.c create mode 100644 sysdeps/loongarch/lp64/multiarch/strchrnul-aligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchrnul-lasx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchrnul-lsx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/strchrnul.c diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile index 76c506c966..110a8c5ca3 100644 --- a/sysdeps/loongarch/lp64/multiarch/Makefile +++ b/sysdeps/loongarch/lp64/multiarch/Makefile @@ -3,5 +3,11 @@ sysdep_routines += \ strlen-aligned \ strlen-lsx \ strlen-lasx \ + strchr-aligned \ + strchr-lsx \ + strchr-lasx \ + strchrnul-aligned \ + strchrnul-lsx \ + strchrnul-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 1a2a576fcd..c7164b45e6 100644 --- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c @@ -37,5 +37,21 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, #endif IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_aligned) ) + + IFUNC_IMPL (i, name, strchr, +#if !defined __loongarch_soft_float + IFUNC_IMPL_ADD (array, i, strchr, SUPPORT_LASX, __strchr_lasx) + IFUNC_IMPL_ADD (array, i, strchr, SUPPORT_LSX, __strchr_lsx) +#endif + IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_aligned) + ) + + IFUNC_IMPL (i, name, strchrnul, +#if !defined __loongarch_soft_float + IFUNC_IMPL_ADD (array, i, strchrnul, SUPPORT_LASX, __strchrnul_lasx) + IFUNC_IMPL_ADD (array, i, strchrnul, SUPPORT_LSX, __strchrnul_lsx) +#endif + IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_aligned) + ) return i; } diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-strchr.h b/sysdeps/loongarch/lp64/multiarch/ifunc-strchr.h new file mode 100644 index 0000000000..4494db79a5 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-strchr.h @@ -0,0 +1,41 @@ +/* Common definition for strchr 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/ifunc-strchrnul.h b/sysdeps/loongarch/lp64/multiarch/ifunc-strchrnul.h new file mode 100644 index 0000000000..8a92512064 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-strchrnul.h @@ -0,0 +1,41 @@ +/* Common definition for strchrnul 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/strchr-aligned.S b/sysdeps/loongarch/lp64/multiarch/strchr-aligned.S new file mode 100644 index 0000000000..5fb01806e4 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchr-aligned.S @@ -0,0 +1,99 @@ +/* Optimized strchr 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 STRCHR_NAME __strchr_aligned +#else +# define STRCHR_NAME strchr +#endif + +LEAF(STRCHR_NAME, 6) + slli.d t1, a0, 3 + bstrins.d a0, zero, 2, 0 + lu12i.w a2, 0x01010 + ld.d t2, a0, 0 + + ori a2, a2, 0x101 + andi a1, a1, 0xff + bstrins.d a2, a2, 63, 32 + li.w t0, -1 + + mul.d a1, a1, a2 + sll.d t0, t0, t1 + slli.d a3, a2, 7 + orn t2, t2, t0 + + sll.d t3, a1, t1 + xor t4, t2, t3 + sub.d a4, t2, a2 + sub.d a5, t4, a2 + + + andn a4, a4, t2 + andn a5, a5, t4 + or t0, a4, a5 + and t0, t0, a3 + + bnez t0, L(end) + addi.d a0, a0, 8 +L(loop): + ld.d t4, a0, 0 + xor t2, t4, a1 + + sub.d a4, t4, a2 + sub.d a5, t2, a2 + andn a4, a4, t4 + andn a5, a5, t2 + + or t0, a4, a5 + and t0, t0, a3 + bnez t0, L(end) + ld.d t4, a0, 8 + + + addi.d a0, a0, 16 + xor t2, t4, a1 + sub.d a4, t4, a2 + sub.d a5, t2, a2 + + andn a4, a4, t4 + andn a5, a5, t2 + or t0, a4, a5 + and t0, t0, a3 + + beqz t0, L(loop) + addi.d a0, a0, -8 +L(end): + and t0, a5, a3 + and t1, a4, a3 + + ctz.d t0, t0 + ctz.d t1, t1 + srli.w t2, t0, 3 + sltu t3, t1, t0 + + + add.d a0, a0, t2 + masknez a0, a0, t3 + jr ra +END(STRCHR_NAME) diff --git a/sysdeps/loongarch/lp64/multiarch/strchr-lasx.S b/sysdeps/loongarch/lp64/multiarch/strchr-lasx.S new file mode 100644 index 0000000000..254402daa5 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchr-lasx.S @@ -0,0 +1,91 @@ +/* Optimized strchr implementation using loongarch LASX SIMD 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 + +#ifndef AS_STRCHRNUL +# define STRCHR __strchr_lasx +#endif + +LEAF(STRCHR, 6) + andi t1, a0, 0x1f + bstrins.d a0, zero, 4, 0 + xvld xr0, a0, 0 + li.d t2, -1 + + xvreplgr2vr.b xr1, a1 + sll.d t1, t2, t1 + xvxor.v xr2, xr0, xr1 + xvmin.bu xr0, xr0, xr2 + + xvmsknz.b xr0, xr0 + xvpickve.w xr3, xr0, 4 + vilvl.h vr0, vr3, vr0 + movfr2gr.s t0, fa0 + + orn t0, t0, t1 + bne t0, t2, L(end) + addi.d a0, a0, 32 + nop + + +L(loop): + xvld xr0, a0, 0 + xvxor.v xr2, xr0, xr1 + xvmin.bu xr0, xr0, xr2 + xvsetanyeqz.b fcc0, xr0 + + bcnez fcc0, L(loop_end) + xvld xr0, a0, 32 + addi.d a0, a0, 64 + xvxor.v xr2, xr0, xr1 + + xvmin.bu xr0, xr0, xr2 + xvsetanyeqz.b fcc0, xr0 + bceqz fcc0, L(loop) + addi.d a0, a0, -32 + +L(loop_end): + xvmsknz.b xr0, xr0 + xvpickve.w xr1, xr0, 4 + vilvl.h vr0, vr1, vr0 + movfr2gr.s t0, fa0 + + +L(end): + cto.w t0, t0 + add.d a0, a0, t0 +#ifndef AS_STRCHRNUL + vreplgr2vr.b vr0, t0 + xvpermi.q xr3, xr2, 1 + + vshuf.b vr0, vr3, vr2, vr0 + vpickve2gr.bu t0, vr0, 0 + masknez a0, a0, t0 +#endif + jr ra + +END(STRCHR) + +libc_hidden_builtin_def(STRCHR) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/strchr-lsx.S b/sysdeps/loongarch/lp64/multiarch/strchr-lsx.S new file mode 100644 index 0000000000..dae98b0a55 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchr-lsx.S @@ -0,0 +1,73 @@ +/* Optimized strlen implementation using loongarch LSX SIMD 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 + +#ifndef AS_STRCHRNUL +# define STRCHR __strchr_lsx +#endif + +LEAF(STRCHR, 6) + andi t1, a0, 0xf + bstrins.d a0, zero, 3, 0 + vld vr0, a0, 0 + li.d t2, -1 + + vreplgr2vr.b vr1, a1 + sll.d t3, t2, t1 + vxor.v vr2, vr0, vr1 + vmin.bu vr0, vr0, vr2 + + vmsknz.b vr0, vr0 + movfr2gr.s t0, fa0 + ext.w.h t0, t0 + orn t0, t0, t3 + + beq t0, t2, L(loop) +L(found): + cto.w t0, t0 + add.d a0, a0, t0 +#ifndef AS_STRCHRNUL + vreplve.b vr2, vr2, t0 + vpickve2gr.bu t1, vr2, 0 + masknez a0, a0, t1 +#endif + jr ra + + +L(loop): + vld vr0, a0, 16 + addi.d a0, a0, 16 + vxor.v vr2, vr0, vr1 + vmin.bu vr0, vr0, vr2 + + vsetanyeqz.b fcc0, vr0 + bceqz fcc0, L(loop) + vmsknz.b vr0, vr0 + movfr2gr.s t0, fa0 + + b L(found) +END(STRCHR) + +libc_hidden_builtin_def (STRCHR) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/strchr.c b/sysdeps/loongarch/lp64/multiarch/strchr.c new file mode 100644 index 0000000000..404e97bded --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchr.c @@ -0,0 +1,36 @@ +/* Multiple versions of strchr. + 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 + . */ + +/* Define multiple versions only for the definition in libc. */ +#if IS_IN (libc) +# define strchr __redirect_strchr +# include +# undef strchr + +# define SYMBOL_NAME strchr +# include "ifunc-strchr.h" + +libc_ifunc_redirected (__redirect_strchr, strchr, IFUNC_SELECTOR ()); +weak_alias(strchr, index) +# ifdef SHARED +__hidden_ver1 (strchr, __GI_strchr, __redirect_strchr) + __attribute__ ((visibility ("hidden"))) __attribute_copy__ (strchr); +# endif + +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/strchrnul-aligned.S b/sysdeps/loongarch/lp64/multiarch/strchrnul-aligned.S new file mode 100644 index 0000000000..1c01a0232d --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchrnul-aligned.S @@ -0,0 +1,95 @@ +/* Optimized strchrnul 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 STRCHRNUL_NAME __strchrnul_aligned +#else +# define STRCHRNUL_NAME __strchrnul +#endif + +LEAF(STRCHRNUL_NAME, 6) + slli.d t1, a0, 3 + bstrins.d a0, zero, 2, 0 + lu12i.w a2, 0x01010 + ld.d t2, a0, 0 + + ori a2, a2, 0x101 + andi a1, a1, 0xff + bstrins.d a2, a2, 63, 32 + li.w t0, -1 + + mul.d a1, a1, a2 + sll.d t0, t0, t1 + slli.d a3, a2, 7 + orn t2, t2, t0 + + sll.d t3, a1, t1 + xor t4, t2, t3 + sub.d a4, t2, a2 + sub.d a5, t4, a2 + + + andn a4, a4, t2 + andn a5, a5, t4 + or t0, a4, a5 + and t0, t0, a3 + + bnez t0, L(end) + addi.d a0, a0, 8 +L(loop): + ld.d t4, a0, 0 + xor t2, t4, a1 + + sub.d a4, t4, a2 + sub.d a5, t2, a2 + andn a4, a4, t4 + andn a5, a5, t2 + + or t0, a4, a5 + and t0, t0, a3 + bnez t0, L(end) + ld.d t4, a0, 8 + + + addi.d a0, a0, 16 + xor t2, t4, a1 + sub.d a4, t4, a2 + sub.d a5, t2, a2 + + andn a4, a4, t4 + andn a5, a5, t2 + or t0, a4, a5 + and t0, t0, a3 + + beqz t0, L(loop) + addi.d a0, a0, -8 +L(end): + ctz.d t0, t0 + srli.w t0, t0, 3 + + + add.d a0, a0, t0 + jr ra +END(STRCHRNUL_NAME) + +libc_hidden_builtin_def (STRCHRNUL_NAME) diff --git a/sysdeps/loongarch/lp64/multiarch/strchrnul-lasx.S b/sysdeps/loongarch/lp64/multiarch/strchrnul-lasx.S new file mode 100644 index 0000000000..d45495e48f --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchrnul-lasx.S @@ -0,0 +1,22 @@ +/* Optimized strchrnul implementation using loongarch LASX SIMD 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 + . */ + +#define STRCHR __strchrnul_lasx +#define AS_STRCHRNUL +#include "strchr-lasx.S" diff --git a/sysdeps/loongarch/lp64/multiarch/strchrnul-lsx.S b/sysdeps/loongarch/lp64/multiarch/strchrnul-lsx.S new file mode 100644 index 0000000000..07d793ae5f --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchrnul-lsx.S @@ -0,0 +1,22 @@ +/* Optimized strchrnul implementation using loongarch LSX SIMD 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 + . */ + +#define STRCHR __strchrnul_lsx +#define AS_STRCHRNUL +#include "strchr-lsx.S" diff --git a/sysdeps/loongarch/lp64/multiarch/strchrnul.c b/sysdeps/loongarch/lp64/multiarch/strchrnul.c new file mode 100644 index 0000000000..f3b8296ee8 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/strchrnul.c @@ -0,0 +1,39 @@ +/* Multiple versions of strchrnul. + 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 + . */ + +/* Define multiple versions only for the definition in libc. */ + +#if IS_IN (libc) +# define strchrnul __redirect_strchrnul +# define __strchrnul __redirect___strchrnul +# include +# undef __strchrnul +# undef strchrnul + +# define SYMBOL_NAME strchrnul +# include "ifunc-strchrnul.h" + +libc_ifunc_redirected (__redirect_strchrnul, __strchrnul, + IFUNC_SELECTOR ()); +weak_alias (__strchrnul, strchrnul) +# ifdef SHARED +__hidden_ver1 (__strchrnul, __GI___strchrnul, __redirect_strchrnul) + __attribute__((visibility ("hidden"))) __attribute_copy__ (strchrnul); +# endif +#endif From patchwork Tue Aug 15 09:43:36 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: dengjianbo X-Patchwork-Id: 1821309 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=8.43.85.97; helo=server2.sourceware.org; envelope-from=libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org; receiver=) Received: from server2.sourceware.org (server2.sourceware.org [8.43.85.97]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature ECDSA (P-384) server-digest SHA384) (No client certificate requested) by legolas.ozlabs.org (Postfix) with ESMTPS id 4RQ5vS2H40z1yfS for ; Tue, 15 Aug 2023 19:44:20 +1000 (AEST) Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 487BF385C6F5 for ; Tue, 15 Aug 2023 09:44:18 +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 B4B8E3858028 for ; Tue, 15 Aug 2023 09:43:41 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org B4B8E3858028 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 _____8CxtPDMSNtkILYYAA--.51522S3; Tue, 15 Aug 2023 17:43:40 +0800 (CST) Received: from 5.5.5 (unknown [10.2.5.5]) by localhost.localdomain (Coremail) with SMTP id AQAAf8DxfSPJSNtkThBbAA--.45023S4; Tue, 15 Aug 2023 17:43:39 +0800 (CST) From: dengjianbo To: libc-alpha@sourceware.org Cc: adhemerval.zanella@linaro.org, xry111@xry111.site, caiyinyu@loongson.cn, xuchenghua@loongson.cn, huangpei@loongson.cn, dengjianbo Subject: [PATCH 2/2] Loongarch: Add ifunc support for memcpy{aligned, unaligned, lsx, lasx} and memmove{aligned, unaligned, lsx, lasx} Date: Tue, 15 Aug 2023 17:43:36 +0800 Message-Id: <20230815094336.1722160-3-dengjianbo@loongson.cn> X-Mailer: git-send-email 2.31.1 In-Reply-To: <20230815094336.1722160-1-dengjianbo@loongson.cn> References: <20230815094336.1722160-1-dengjianbo@loongson.cn> MIME-Version: 1.0 X-CM-TRANSID: AQAAf8DxfSPJSNtkThBbAA--.45023S4 X-CM-SenderInfo: pghqwyxldqu0o6or00hjvr0hdfq/ X-Coremail-Antispam: 1Uk129KBj9fXoWDZr15WF15trWDGw45GF47ZFc_yoW7GrW7Wo W5AFZFqwn7G3yUJFW7Cw13J3s7WFWfGr1jqw4rAa18tryrKF47Jr98CF4q9347trn5uFsr Xas7ZFnxC397KFnrl-sFpf9Il3svdjkaLaAFLSUrUUUUUb8apTn2vfkv8UJUUUU8wcxFpf 9Il3svdxBIdaVrn0xqx4xG64xvF2IEw4CE5I8CrVC2j2Jv73VFW2AGmfu7bjvjm3AaLaJ3 UjIYCTnIWjp_UUUYI7kC6x804xWl14x267AKxVWUJVW8JwAFc2x0x2IEx4CE42xK8VAvwI 8IcIk0rVWrJVCq3wAFIxvE14AKwVWUJVWUGwA2ocxC64kIII0Yj41l84x0c7CEw4AK67xG Y2AK021l84ACjcxK6xIIjxv20xvE14v26r1I6r4UM28EF7xvwVC0I7IYx2IY6xkF7I0E14 v26r1j6r4UM28EF7xvwVC2z280aVAFwI0_Gr1j6F4UJwA2z4x0Y4vEx4A2jsIEc7CjxVAF wI0_Gr1j6F4UJwAS0I0E0xvYzxvE52x082IY62kv0487Mc804VCY07AIYIkI8VC2zVCFFI 0UMc02F40EFcxC0VAKzVAqx4xG6I80ewAv7VC0I7IYx2IY67AKxVWUAVWUtwAv7VC2z280 aVAFwI0_Jr0_Gr1lOx8S6xCaFVCjc4AY6r1j6r4UM4x0Y48IcxkI7VAKI48JMxAIw28Icx kI7VAKI48JMxC20s026xCaFVCjc4AY6r1j6r4UMxCIbckI1I0E14v26r1Y6r17MI8I3I0E 5I8CrVAFwI0_Jr0_Jr4lx2IqxVCjr7xvwVAFwI0_JrI_JrWlx4CE17CEb7AF67AKxVWUAV WUtwCIc40Y0x0EwIxGrwCI42IY6xIIjxv20xvE14v26r1I6r4UMIIF0xvE2Ix0cI8IcVCY 1x0267AKxVWUJVW8JwCI42IY6xAIw20EY4v20xvaj40_Jr0_JF4lIxAIcVC2z280aVAFwI 0_Jr0_Gr1lIxAIcVC2z280aVCY1x0267AKxVWUJVW8JbIYCTnIWIevJa73UjIFyTuYvjxU wMKuUUUUU X-Spam-Status: No, score=-9.3 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_STATUS, KAM_SHORT, KAM_STOCKGEN, SCC_10_SHORT_WORD_LINES, SCC_20_SHORT_WORD_LINES, SCC_35_SHORT_WORD_LINES, 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: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: libc-alpha-bounces+incoming=patchwork.ozlabs.org@sourceware.org Sender: "Libc-alpha" These implementations improve the time to copy data in the glibc microbenchmark as below: memcpy-lasx reduces the runtime about 8%-76% memcpy-lsx reduces the runtime about 8%-72% memcpy-unaligned reduces the runtime of unaligned data copying up to 40% memcpy-aligned reduece the runtime of unaligned data copying up to 25% memmove-lasx reduces the runtime about 20%-73% memmove-lsx reduces the runtime about 50% memmove-unaligned reduces the runtime of unaligned data moving up to 40% memmove-aligned reduces the runtime of unaligned data moving up to 25% --- sysdeps/loongarch/lp64/multiarch/Makefile | 5 + .../lp64/multiarch/ifunc-impl-list.c | 18 + sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h | 45 + .../loongarch/lp64/multiarch/memcpy-aligned.S | 783 ++++++++++++++++++ .../loongarch/lp64/multiarch/memcpy-lasx.S | 20 + sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S | 20 + .../lp64/multiarch/memcpy-unaligned.S | 247 ++++++ sysdeps/loongarch/lp64/multiarch/memcpy.c | 37 + .../lp64/multiarch/memmove-aligned.S | 20 + .../loongarch/lp64/multiarch/memmove-lasx.S | 287 +++++++ .../loongarch/lp64/multiarch/memmove-lsx.S | 534 ++++++++++++ .../lp64/multiarch/memmove-unaligned.S | 380 +++++++++ sysdeps/loongarch/lp64/multiarch/memmove.c | 38 + 13 files changed, 2434 insertions(+) create mode 100644 sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memcpy.c create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-aligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-lasx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-lsx.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S create mode 100644 sysdeps/loongarch/lp64/multiarch/memmove.c diff --git a/sysdeps/loongarch/lp64/multiarch/Makefile b/sysdeps/loongarch/lp64/multiarch/Makefile index 110a8c5ca3..afa5104174 100644 --- a/sysdeps/loongarch/lp64/multiarch/Makefile +++ b/sysdeps/loongarch/lp64/multiarch/Makefile @@ -9,5 +9,10 @@ sysdep_routines += \ strchrnul-aligned \ strchrnul-lsx \ strchrnul-lasx \ + memcpy-aligned \ + memcpy-unaligned \ + memmove-unaligned \ + memmove-lsx \ + memmove-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 c7164b45e6..ab99cbeb9d 100644 --- a/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-impl-list.c @@ -53,5 +53,23 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, #endif IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_aligned) ) + + IFUNC_IMPL (i, name, memcpy, +#if !defined __loongarch_soft_float + IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_LASX, __memcpy_lasx) + IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_LSX, __memcpy_lsx) +#endif + IFUNC_IMPL_ADD (array, i, memcpy, SUPPORT_UAL, __memcpy_unaligned) + IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_aligned) + ) + IFUNC_IMPL (i, name, memmove, +#if !defined __loongarch_soft_float + IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_LASX, __memmove_lasx) + IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_LSX, __memmove_lsx) +#endif + IFUNC_IMPL_ADD (array, i, memmove, SUPPORT_UAL, __memmove_unaligned) + IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_aligned) + ) + return i; } diff --git a/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h b/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h new file mode 100644 index 0000000000..3be67da611 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/ifunc-lasx.h @@ -0,0 +1,45 @@ +/* Common definition for ifunc selection implementation. + 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; +extern __typeof (REDIRECT_NAME) OPTIMIZE (unaligned) 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 + if (SUPPORT_UAL) + return OPTIMIZE (unaligned); + else + return OPTIMIZE (aligned); +} diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S b/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S new file mode 100644 index 0000000000..299dd49ce1 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memcpy-aligned.S @@ -0,0 +1,783 @@ +/* Optimized memcpy_aligned 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 MEMCPY_NAME __memcpy_aligned +# define MEMMOVE_NAME __memmove_aligned +#else +# define MEMCPY_NAME memcpy +# define MEMMOVE_NAME memmove +#endif + +#define LD_64(reg, n) \ + ld.d t0, reg, n; \ + ld.d t1, reg, n + 8; \ + ld.d t2, reg, n + 16; \ + ld.d t3, reg, n + 24; \ + ld.d t4, reg, n + 32; \ + ld.d t5, reg, n + 40; \ + ld.d t6, reg, n + 48; \ + ld.d t7, reg, n + 56; + +#define ST_64(reg, n) \ + st.d t0, reg, n; \ + st.d t1, reg, n + 8; \ + st.d t2, reg, n + 16; \ + st.d t3, reg, n + 24; \ + st.d t4, reg, n + 32; \ + st.d t5, reg, n + 40; \ + st.d t6, reg, n + 48; \ + st.d t7, reg, n + 56; + +LEAF(MEMMOVE_NAME, 6) + sub.d t0, a0, a1 + bltu t0, a2, L(copy_back) +END(MEMMOVE_NAME) + +LEAF_NO_ALIGN(MEMCPY_NAME) + srai.d a3, a2, 4 + beqz a3, L(short_data) + + move a4, a0 + andi a5, a0, 0x7 + andi a6, a1, 0x7 + li.d t8, 8 + beqz a5, L(check_align) + + sub.d t2, t8, a5 + sub.d a2, a2, t2 + pcaddi t1, 20 + slli.d t3, t2, 3 + + add.d a1, a1, t2 + sub.d t1, t1, t3 + add.d a4, a4, t2 + jr t1 + +L(al7): + ld.b t0, a1, -7 + st.b t0, a4, -7 +L(al6): + ld.b t0, a1, -6 + st.b t0, a4, -6 +L(al5): + ld.b t0, a1, -5 + st.b t0, a4, -5 +L(al4): + ld.b t0, a1, -4 + st.b t0, a4, -4 +L(al3): + ld.b t0, a1, -3 + st.b t0, a4, -3 +L(al2): + ld.b t0, a1, -2 + st.b t0, a4, -2 +L(al1): + ld.b t0, a1, -1 + st.b t0, a4, -1 + +L(check_align): + bne a5, a6, L(unalign) + srai.d a3, a2, 4 + beqz a3, L(al_less_16bytes) + andi a3, a2, 0x3f + + beq a3, a2, L(al_less_64bytes) + sub.d t0, a2, a3 + move a2, a3 + add.d a5, a1, t0 + +L(loop_64bytes): + LD_64(a1, 0) + addi.d a1, a1, 64 + ST_64(a4, 0) + + addi.d a4, a4, 64 + bne a1, a5, L(loop_64bytes) + +L(al_less_64bytes): + srai.d a3, a2, 5 + beqz a3, L(al_less_32bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + addi.d a1, a1, 32 + addi.d a2, a2, -32 + + st.d t0, a4, 0 + st.d t1, a4, 8 + st.d t2, a4, 16 + st.d t3, a4, 24 + + addi.d a4, a4, 32 + +L(al_less_32bytes): + srai.d a3, a2, 4 + beqz a3, L(al_less_16bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + addi.d a1, a1, 16 + addi.d a2, a2, -16 + + st.d t0, a4, 0 + st.d t1, a4, 8 + addi.d a4, a4, 16 + +L(al_less_16bytes): + srai.d a3, a2, 3 + beqz a3, L(al_less_8bytes) + + ld.d t0, a1, 0 + addi.d a1, a1, 8 + addi.d a2, a2, -8 + st.d t0, a4, 0 + addi.d a4, a4, 8 + +L(al_less_8bytes): + srai.d a3, a2, 2 + beqz a3, L(al_less_4bytes) + + ld.w t0, a1, 0 + addi.d a1, a1, 4 + addi.d a2, a2, -4 + st.w t0, a4, 0 + addi.d a4, a4, 4 + +L(al_less_4bytes): + srai.d a3, a2, 1 + beqz a3, L(al_less_2bytes) + + ld.h t0, a1, 0 + addi.d a1, a1, 2 + addi.d a2, a2, -2 + st.h t0, a4, 0 + addi.d a4, a4, 2 + +L(al_less_2bytes): + beqz a2, L(al_less_1byte) + + ld.b t0, a1, 0 + st.b t0, a4, 0 + +L(al_less_1byte): + jr ra + +L(unalign): + andi a5, a1, 0x7 + bstrins.d a1, zero, 2, 0 + sub.d t8, t8, a5 + slli.d a5, a5, 3 + + ld.d t0, a1, 0 + addi.d a1, a1, 8 + slli.d a6, t8, 3 + srl.d a7, t0, a5 + + srai.d a3, a2, 4 + beqz a3, L(un_less_16bytes) + andi a3, a2, 0x3f + beq a3, a2, L(un_less_64bytes) + + sub.d t0, a2, a3 + move a2, a3 + add.d a3, a1, t0 + +L(un_long_bytes): + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + srl.d t4, t0, a5 + sll.d t0, t0, a6 + srl.d t5, t1, a5 + sll.d t1, t1, a6 + + srl.d t6, t2, a5 + sll.d t2, t2, a6 + srl.d t7, t3, a5 + sll.d t3, t3, a6 + + or t0, a7, t0 + or t1, t4, t1 + or t2, t5, t2 + or t3, t6, t3 + + ld.d t4, a1, 32 + ld.d t5, a1, 40 + ld.d t6, a1, 48 + ld.d a7, a1, 56 + + st.d t0, a4, 0 + st.d t1, a4, 8 + st.d t2, a4, 16 + st.d t3, a4, 24 + + addi.d a1, a1, 64 + + srl.d t0, t4, a5 + sll.d t4, t4, a6 + srl.d t1, t5, a5 + sll.d t5, t5, a6 + + srl.d t2, t6, a5 + sll.d t6, t6, a6 + sll.d t3, a7, a6 + srl.d a7, a7, a5 + + or t4, t7, t4 + or t5, t0, t5 + or t6, t1, t6 + or t3, t2, t3 + + st.d t4, a4, 32 + st.d t5, a4, 40 + st.d t6, a4, 48 + st.d t3, a4, 56 + + addi.d a4, a4, 64 + bne a3, a1, L(un_long_bytes) + +L(un_less_64bytes): + srai.d a3, a2, 5 + beqz a3, L(un_less_32bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + addi.d a1, a1, 32 + addi.d a2, a2, -32 + + srl.d t4, t0, a5 + sll.d t0, t0, a6 + srl.d t5, t1, a5 + sll.d t1, t1, a6 + + srl.d t6, t2, a5 + sll.d t2, t2, a6 + or t0, a7, t0 + srl.d a7, t3, a5 + sll.d t3, t3, a6 + + or t1, t4, t1 + or t2, t5, t2 + or t3, t6, t3 + + st.d t0, a4, 0 + st.d t1, a4, 8 + st.d t2, a4, 16 + st.d t3, a4, 24 + + addi.d a4, a4, 32 + +L(un_less_32bytes): + srai.d a3, a2, 4 + beqz a3, L(un_less_16bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + addi.d a1, a1, 16 + addi.d a2, a2, -16 + + srl.d t2, t0, a5 + sll.d t3, t0, a6 + sll.d t4, t1, a6 + or t3, a7, t3 + or t4, t2, t4 + + srl.d a7, t1, a5 + st.d t3, a4, 0 + st.d t4, a4, 8 + addi.d a4, a4, 16 + +L(un_less_16bytes): + srai.d a3, a2, 3 + beqz a3, L(un_less_8bytes) + + ld.d t0, a1, 0 + addi.d a1, a1, 8 + addi.d a2, a2, -8 + sll.d t1, t0, a6 + + or t2, a7, t1 + srl.d a7, t0, a5 + st.d t2, a4, 0 + addi.d a4, a4, 8 + +L(un_less_8bytes): + beqz a2, L(un_less_1byte) + bge t8, a2, 1f + + ld.d t0, a1, 0 + sll.d t0, t0, a6 + or a7, a7, t0 + +1: + srai.d a3, a2, 2 + beqz a3, L(un_less_4bytes) + + addi.d a2, a2, -4 + st.w a7, a4, 0 + addi.d a4, a4, 4 + srai.d a7, a7, 32 + +L(un_less_4bytes): + srai.d a3, a2, 1 + beqz a3, L(un_less_2bytes) + + addi.d a2, a2, -2 + st.h a7, a4, 0 + addi.d a4, a4, 2 + srai.d a7, a7, 16 + +L(un_less_2bytes): + beqz a2, L(un_less_1byte) + st.b a7, a4, 0 + +L(un_less_1byte): + jr ra + +L(short_data): + pcaddi t1, 36 + slli.d t2, a2, 3 + add.d a4, a0, a2 + sub.d t1, t1, t2 + add.d a1, a1, a2 + jr t1 + +L(short_15_bytes): + ld.b t0, a1, -15 + st.b t0, a4, -15 +L(short_14_bytes): + ld.b t0, a1, -14 + st.b t0, a4, -14 +L(short_13_bytes): + ld.b t0, a1, -13 + st.b t0, a4, -13 +L(short_12_bytes): + ld.b t0, a1, -12 + st.b t0, a4, -12 +L(short_11_bytes): + ld.b t0, a1, -11 + st.b t0, a4, -11 +L(short_10_bytes): + ld.b t0, a1, -10 + st.b t0, a4, -10 +L(short_9_bytes): + ld.b t0, a1, -9 + st.b t0, a4, -9 +L(short_8_bytes): + ld.b t0, a1, -8 + st.b t0, a4, -8 +L(short_7_bytes): + ld.b t0, a1, -7 + st.b t0, a4, -7 +L(short_6_bytes): + ld.b t0, a1, -6 + st.b t0, a4, -6 +L(short_5_bytes): + ld.b t0, a1, -5 + st.b t0, a4, -5 +L(short_4_bytes): + ld.b t0, a1, -4 + st.b t0, a4, -4 +L(short_3_bytes): + ld.b t0, a1, -3 + st.b t0, a4, -3 +L(short_2_bytes): + ld.b t0, a1, -2 + st.b t0, a4, -2 +L(short_1_bytes): + ld.b t0, a1, -1 + st.b t0, a4, -1 + jr ra + +L(copy_back): + srai.d a3, a2, 4 + beqz a3, L(back_short_data) + + add.d a4, a0, a2 + add.d a1, a1, a2 + + andi a5, a4, 0x7 + andi a6, a1, 0x7 + beqz a5, L(back_check_align) + + sub.d a2, a2, a5 + sub.d a1, a1, a5 + sub.d a4, a4, a5 + + pcaddi t1, 18 + slli.d t3, a5, 3 + sub.d t1, t1, t3 + jr t1 + + ld.b t0, a1, 6 + st.b t0, a4, 6 + ld.b t0, a1, 5 + st.b t0, a4, 5 + ld.b t0, a1, 4 + st.b t0, a4, 4 + ld.b t0, a1, 3 + st.b t0, a4, 3 + ld.b t0, a1, 2 + st.b t0, a4, 2 + ld.b t0, a1, 1 + st.b t0, a4, 1 + ld.b t0, a1, 0 + st.b t0, a4, 0 + +L(back_check_align): + bne a5, a6, L(back_unalign) + + srai.d a3, a2, 4 + beqz a3, L(back_less_16bytes) + + andi a3, a2, 0x3f + beq a3, a2, L(back_less_64bytes) + + sub.d t0, a2, a3 + move a2, a3 + sub.d a5, a1, t0 + +L(back_loop_64bytes): + LD_64(a1, -64) + addi.d a1, a1, -64 + ST_64(a4, -64) + + addi.d a4, a4, -64 + bne a1, a5, L(back_loop_64bytes) + +L(back_less_64bytes): + srai.d a3, a2, 5 + beqz a3, L(back_less_32bytes) + + ld.d t0, a1, -32 + ld.d t1, a1, -24 + ld.d t2, a1, -16 + ld.d t3, a1, -8 + + addi.d a1, a1, -32 + addi.d a2, a2, -32 + + st.d t0, a4, -32 + st.d t1, a4, -24 + st.d t2, a4, -16 + st.d t3, a4, -8 + + addi.d a4, a4, -32 + +L(back_less_32bytes): + srai.d a3, a2, 4 + beqz a3, L(back_less_16bytes) + + ld.d t0, a1, -16 + ld.d t1, a1, -8 + + addi.d a2, a2, -16 + addi.d a1, a1, -16 + + st.d t0, a4, -16 + st.d t1, a4, -8 + addi.d a4, a4, -16 + +L(back_less_16bytes): + srai.d a3, a2, 3 + beqz a3, L(back_less_8bytes) + + ld.d t0, a1, -8 + addi.d a2, a2, -8 + addi.d a1, a1, -8 + + st.d t0, a4, -8 + addi.d a4, a4, -8 + +L(back_less_8bytes): + srai.d a3, a2, 2 + beqz a3, L(back_less_4bytes) + + ld.w t0, a1, -4 + addi.d a2, a2, -4 + addi.d a1, a1, -4 + + st.w t0, a4, -4 + addi.d a4, a4, -4 + +L(back_less_4bytes): + srai.d a3, a2, 1 + beqz a3, L(back_less_2bytes) + + ld.h t0, a1, -2 + addi.d a2, a2, -2 + addi.d a1, a1, -2 + + st.h t0, a4, -2 + addi.d a4, a4, -2 + +L(back_less_2bytes): + beqz a2, L(back_less_1byte) + + ld.b t0, a1, -1 + st.b t0, a4, -1 + +L(back_less_1byte): + jr ra + +L(back_unalign): + andi t8, a1, 0x7 + bstrins.d a1, zero, 2, 0 + + sub.d a6, zero, t8 + + ld.d t0, a1, 0 + slli.d a6, a6, 3 + slli.d a5, t8, 3 + sll.d a7, t0, a6 + + srai.d a3, a2, 4 + beqz a3, L(back_un_less_16bytes) + + andi a3, a2, 0x3f + beq a3, a2, L(back_un_less_64bytes) + + sub.d t0, a2, a3 + move a2, a3 + sub.d a3, a1, t0 + +L(back_un_long_bytes): + ld.d t0, a1, -8 + ld.d t1, a1, -16 + ld.d t2, a1, -24 + ld.d t3, a1, -32 + + sll.d t4, t0, a6 + srl.d t0, t0, a5 + + sll.d t5, t1, a6 + srl.d t1, t1, a5 + + sll.d t6, t2, a6 + srl.d t2, t2, a5 + + sll.d t7, t3, a6 + srl.d t3, t3, a5 + + or t0, t0, a7 + or t1, t1, t4 + or t2, t2, t5 + or t3, t3, t6 + + ld.d t4, a1, -40 + ld.d t5, a1, -48 + ld.d t6, a1, -56 + ld.d a7, a1, -64 + st.d t0, a4, -8 + st.d t1, a4, -16 + st.d t2, a4, -24 + st.d t3, a4, -32 + + addi.d a1, a1, -64 + + sll.d t0, t4, a6 + srl.d t4, t4, a5 + + sll.d t1, t5, a6 + srl.d t5, t5, a5 + + sll.d t2, t6, a6 + srl.d t6, t6, a5 + + srl.d t3, a7, a5 + sll.d a7, a7, a6 + + or t4, t7, t4 + or t5, t0, t5 + or t6, t1, t6 + or t3, t2, t3 + + st.d t4, a4, -40 + st.d t5, a4, -48 + st.d t6, a4, -56 + st.d t3, a4, -64 + + addi.d a4, a4, -64 + bne a3, a1, L(back_un_long_bytes) + +L(back_un_less_64bytes): + srai.d a3, a2, 5 + beqz a3, L(back_un_less_32bytes) + + ld.d t0, a1, -8 + ld.d t1, a1, -16 + ld.d t2, a1, -24 + ld.d t3, a1, -32 + + addi.d a1, a1, -32 + addi.d a2, a2, -32 + + sll.d t4, t0, a6 + srl.d t0, t0, a5 + + sll.d t5, t1, a6 + srl.d t1, t1, a5 + + sll.d t6, t2, a6 + srl.d t2, t2, a5 + + or t0, a7, t0 + + sll.d a7, t3, a6 + srl.d t3, t3, a5 + + or t1, t4, t1 + or t2, t5, t2 + or t3, t6, t3 + + st.d t0, a4, -8 + st.d t1, a4, -16 + st.d t2, a4, -24 + st.d t3, a4, -32 + + addi.d a4, a4, -32 + +L(back_un_less_32bytes): + srai.d a3, a2, 4 + beqz a3, L(back_un_less_16bytes) + + ld.d t0, a1, -8 + ld.d t1, a1, -16 + + addi.d a1, a1, -16 + addi.d a2, a2, -16 + + sll.d t2, t0, a6 + srl.d t3, t0, a5 + + srl.d t4, t1, a5 + or t3, a7, t3 + or t4, t2, t4 + sll.d a7, t1, a6 + + st.d t3, a4, -8 + st.d t4, a4, -16 + + addi.d a4, a4, -16 + +L(back_un_less_16bytes): + srai.d a3, a2, 3 + beqz a3, L(back_un_less_8bytes) + + ld.d t0, a1, -8 + + addi.d a1, a1, -8 + addi.d a2, a2, -8 + + srl.d t1, t0, a5 + or t2, a7, t1 + sll.d a7, t0, a6 + + st.d t2, a4, -8 + addi.d a4, a4, -8 + +L(back_un_less_8bytes): + beqz a2, L(back_end) + bge t8, a2, 1f + + ld.d t0, a1, -8 + srl.d t0, t0, a5 + or a7, a7, t0 + +1: + srai.d a3, a2, 2 + beqz a3, L(back_un_less_4bytes) + + srai.d t0, a7, 32 + addi.d a2, a2, -4 + st.w t0, a4, -4 + addi.d a4, a4, -4 + slli.d a7, a7, 32 + +L(back_un_less_4bytes): + srai.d a3, a2, 1 + beqz a3, L(back_un_less_2bytes) + srai.d t0, a7, 48 + addi.d a2, a2, -2 + st.h t0, a4, -2 + addi.d a4, a4, -2 + slli.d a7, a7, 16 +L(back_un_less_2bytes): + beqz a2, L(back_un_less_1byte) + srai.d t0, a7, 56 + st.b t0, a4, -1 +L(back_un_less_1byte): + jr ra + +L(back_short_data): + pcaddi t1, 34 + slli.d t2, a2, 3 + sub.d t1, t1, t2 + jr t1 + + ld.b t0, a1, 14 + st.b t0, a0, 14 + ld.b t0, a1, 13 + st.b t0, a0, 13 + ld.b t0, a1, 12 + st.b t0, a0, 12 + ld.b t0, a1, 11 + st.b t0, a0, 11 + ld.b t0, a1, 10 + st.b t0, a0, 10 + ld.b t0, a1, 9 + st.b t0, a0, 9 + ld.b t0, a1, 8 + st.b t0, a0, 8 + ld.b t0, a1, 7 + st.b t0, a0, 7 + ld.b t0, a1, 6 + st.b t0, a0, 6 + ld.b t0, a1, 5 + st.b t0, a0, 5 + ld.b t0, a1, 4 + st.b t0, a0, 4 + ld.b t0, a1, 3 + st.b t0, a0, 3 + ld.b t0, a1, 2 + st.b t0, a0, 2 + ld.b t0, a1, 1 + st.b t0, a0, 1 + ld.b t0, a1, 0 + st.b t0, a0, 0 +L(back_end): + jr ra + +END(MEMCPY_NAME) + +libc_hidden_builtin_def (MEMMOVE_NAME) +libc_hidden_builtin_def (MEMCPY_NAME) diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S b/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S new file mode 100644 index 0000000000..4aae5bf831 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memcpy-lasx.S @@ -0,0 +1,20 @@ +/* Optimized memcpy 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 + . */ + +/* memcpy is part of memmove.S */ diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S b/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S new file mode 100644 index 0000000000..6ebbe7a2c7 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memcpy-lsx.S @@ -0,0 +1,20 @@ +/* Optimized memcpy 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 + . */ + +/* memcpy is part of memmove.S */ diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S b/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S new file mode 100644 index 0000000000..78c4d0a948 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memcpy-unaligned.S @@ -0,0 +1,247 @@ +/* Optimized unaligned memcpy 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 MEMCPY_NAME __memcpy_unaligned + +# define LD_64(reg, n) \ + ld.d t0, reg, n; \ + ld.d t1, reg, n + 8; \ + ld.d t2, reg, n + 16; \ + ld.d t3, reg, n + 24; \ + ld.d t4, reg, n + 32; \ + ld.d t5, reg, n + 40; \ + ld.d t6, reg, n + 48; \ + ld.d t7, reg, n + 56; + +# define ST_64(reg, n) \ + st.d t0, reg, n; \ + st.d t1, reg, n + 8; \ + st.d t2, reg, n + 16; \ + st.d t3, reg, n + 24; \ + st.d t4, reg, n + 32; \ + st.d t5, reg, n + 40; \ + st.d t6, reg, n + 48; \ + st.d t7, reg, n + 56; + +LEAF(MEMCPY_NAME, 3) + add.d a4, a1, a2 + add.d a3, a0, a2 + li.w a6, 16 + bge a6, a2, L(less_16bytes) + + li.w a6, 128 + blt a6, a2, L(long_bytes) + li.w a6, 64 + blt a6, a2, L(more_64bytes) + + li.w a6, 32 + blt a6, a2, L(more_32bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a4, -16 + ld.d t3, a4, -8 + + st.d t0, a0, 0 + st.d t1, a0, 8 + st.d t2, a3, -16 + st.d t3, a3, -8 + jr ra + +L(more_64bytes): + srli.d t8, a0, 3 + slli.d t8, t8, 3 + addi.d t8, t8, 0x8 + sub.d a7, a0, t8 + + ld.d t0, a1, 0 + sub.d a1, a1, a7 + st.d t0, a0, 0 + add.d a7, a7, a2 + addi.d a7, a7, -0x20 + +L(loop_32): + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + st.d t0, t8, 0 + st.d t1, t8, 8 + st.d t2, t8, 16 + st.d t3, t8, 24 + + addi.d t8, t8, 0x20 + addi.d a1, a1, 0x20 + addi.d a7, a7, -0x20 + blt zero, a7, L(loop_32) + + ld.d t4, a4, -32 + ld.d t5, a4, -24 + ld.d t6, a4, -16 + ld.d t7, a4, -8 + + st.d t4, a3, -32 + st.d t5, a3, -24 + st.d t6, a3, -16 + st.d t7, a3, -8 + + jr ra + +L(more_32bytes): + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + ld.d t4, a4, -32 + ld.d t5, a4, -24 + ld.d t6, a4, -16 + ld.d t7, a4, -8 + + st.d t0, a0, 0 + st.d t1, a0, 8 + st.d t2, a0, 16 + st.d t3, a0, 24 + + st.d t4, a3, -32 + st.d t5, a3, -24 + st.d t6, a3, -16 + st.d t7, a3, -8 + + jr ra + +L(less_16bytes): + srai.d a6, a2, 3 + beqz a6, L(less_8bytes) + + ld.d t0, a1, 0 + ld.d t1, a4, -8 + st.d t0, a0, 0 + st.d t1, a3, -8 + + jr ra + +L(less_8bytes): + srai.d a6, a2, 2 + beqz a6, L(less_4bytes) + + ld.w t0, a1, 0 + ld.w t1, a4, -4 + st.w t0, a0, 0 + st.w t1, a3, -4 + + jr ra + +L(less_4bytes): + srai.d a6, a2, 1 + beqz a6, L(less_2bytes) + + ld.h t0, a1, 0 + ld.h t1, a4, -2 + st.h t0, a0, 0 + st.h t1, a3, -2 + + jr ra + +L(less_2bytes): + beqz a2, L(less_1bytes) + + ld.b t0, a1, 0 + st.b t0, a0, 0 + jr ra + +L(less_1bytes): + jr ra + +L(long_bytes): + srli.d t8, a0, 3 + slli.d t8, t8, 3 + beq a0, t8, L(start) + ld.d t0, a1, 0 + + addi.d t8, t8, 0x8 + st.d t0, a0, 0 + sub.d a7, a0, t8 + sub.d a1, a1, a7 + +L(start): + addi.d a5, a3, -0x80 + blt a5, t8, L(align_end_proc) + +L(loop_128): + LD_64(a1, 0) + ST_64(t8, 0) + LD_64(a1, 64) + addi.d a1, a1, 0x80 + ST_64(t8, 64) + addi.d t8, t8, 0x80 + bge a5, t8, L(loop_128) + +L(align_end_proc): + sub.d a2, a3, t8 + pcaddi t1, 34 + andi t2, a2, 0x78 + sub.d t1, t1, t2 + jr t1 + + ld.d t0, a1, 112 + st.d t0, t8, 112 + ld.d t0, a1, 104 + st.d t0, t8, 104 + ld.d t0, a1, 96 + st.d t0, t8, 96 + ld.d t0, a1, 88 + st.d t0, t8, 88 + ld.d t0, a1, 80 + st.d t0, t8, 80 + ld.d t0, a1, 72 + st.d t0, t8, 72 + ld.d t0, a1, 64 + st.d t0, t8, 64 + ld.d t0, a1, 56 + st.d t0, t8, 56 + ld.d t0, a1, 48 + st.d t0, t8, 48 + ld.d t0, a1, 40 + st.d t0, t8, 40 + ld.d t0, a1, 32 + st.d t0, t8, 32 + ld.d t0, a1, 24 + st.d t0, t8, 24 + ld.d t0, a1, 16 + st.d t0, t8, 16 + ld.d t0, a1, 8 + st.d t0, t8, 8 + ld.d t0, a1, 0 + st.d t0, t8, 0 + ld.d t0, a4, -8 + st.d t0, a3, -8 + + jr ra +END(MEMCPY_NAME) + +libc_hidden_builtin_def (MEMCPY_NAME) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/memcpy.c b/sysdeps/loongarch/lp64/multiarch/memcpy.c new file mode 100644 index 0000000000..93b238ce27 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memcpy.c @@ -0,0 +1,37 @@ +/* Multiple versions of memcpy. + 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 + . */ + +/* Define multiple versions only for the definition in libc. */ +#if IS_IN (libc) +# define memcpy __redirect_memcpy +# include +# undef memcpy + +# define SYMBOL_NAME memcpy +# include "ifunc-lasx.h" + +libc_ifunc_redirected (__redirect_memcpy, memcpy, + IFUNC_SELECTOR ()); + +# ifdef SHARED +__hidden_ver1 (memcpy, __GI_memcpy, __redirect_memcpy) + __attribute__ ((visibility ("hidden"))) __attribute_copy__ (memcmp); +# endif + +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S b/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S new file mode 100644 index 0000000000..5354f38379 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memmove-aligned.S @@ -0,0 +1,20 @@ +/* Optimized memmove_aligned 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 + . */ + +/* memmove_aligned is part of memcpy_aligned, see memcpy-aligned.S. */ diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S b/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S new file mode 100644 index 0000000000..ff68e7a22b --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memmove-lasx.S @@ -0,0 +1,287 @@ +/* Optimized memmove 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 + +#ifndef MEMCPY_NAME +# define MEMCPY_NAME __memcpy_lasx +#endif + +#ifndef MEMMOVE_NAME +# define MEMMOVE_NAME __memmove_lasx +#endif + +LEAF(MEMCPY_NAME, 6) + li.d t0, 32 + add.d a3, a0, a2 + add.d a4, a1, a2 + bgeu t0, a2, L(less_32bytes) + + li.d t1, 64 + bltu t1, a2, L(copy_long) + xvld xr0, a1, 0 + xvld xr1, a4, -32 + + xvst xr0, a0, 0 + xvst xr1, a3, -32 + jr ra +L(less_32bytes): + srli.d t0, a2, 4 + + beqz t0, L(less_16bytes) + vld vr0, a1, 0 + vld vr1, a4, -16 + vst vr0, a0, 0 + + + vst vr1, a3, -16 + jr ra +L(less_16bytes): + srli.d t0, a2, 3 + beqz t0, L(less_8bytes) + + ld.d t0, a1, 0 + ld.d t1, a4, -8 + st.d t0, a0, 0 + st.d t1, a3, -8 + + jr ra +L(less_8bytes): + srli.d t0, a2, 2 + beqz t0, L(less_4bytes) + ld.w t0, a1, 0 + + ld.w t1, a4, -4 + st.w t0, a0, 0 + st.w t1, a3, -4 + jr ra + + +L(less_4bytes): + srli.d t0, a2, 1 + beqz t0, L(less_2bytes) + ld.h t0, a1, 0 + ld.h t1, a4, -2 + + st.h t0, a0, 0 + st.h t1, a3, -2 + jr ra +L(less_2bytes): + beqz a2, L(less_1bytes) + + ld.b t0, a1, 0 + st.b t0, a0, 0 +L(less_1bytes): + jr ra +END(MEMCPY_NAME) + +LEAF(MEMMOVE_NAME, 6) + + li.d t0, 32 + add.d a3, a0, a2 + add.d a4, a1, a2 + bgeu t0, a2, L(less_32bytes) + + li.d t1, 64 + bltu t1, a2, L(move_long) + xvld xr0, a1, 0 + xvld xr1, a4, -32 + + xvst xr0, a0, 0 + xvst xr1, a3, -32 + jr ra +L(move_long): + sub.d t2, a0, a1 + + bltu t2, a2, L(copy_back) +L(copy_long): + andi t2, a0, 0x1f + addi.d a2, a2, -1 + sub.d t2, t0, t2 + + + xvld xr8, a1, 0 + xvld xr9, a4, -32 + sub.d t3, a2, t2 + add.d a5, a0, t2 + + andi a2, t3, 0xff + add.d a1, a1, t2 + beq a2, t3, L(lt256) + sub.d a6, a4, a2 + + addi.d a6, a6, -1 +L(loop_256): + xvld xr0, a1, 0 + xvld xr1, a1, 32 + xvld xr2, a1, 64 + + xvld xr3, a1, 96 + xvld xr4, a1, 128 + xvld xr5, a1, 160 + xvld xr6, a1, 192 + + + xvld xr7, a1, 224 + addi.d a1, a1, 256 + xvst xr0, a5, 0 + xvst xr1, a5, 32 + + xvst xr2, a5, 64 + xvst xr3, a5, 96 + xvst xr4, a5, 128 + xvst xr5, a5, 160 + + xvst xr6, a5, 192 + xvst xr7, a5, 224 + addi.d a5, a5, 256 + bne a1, a6, L(loop_256) + +L(lt256): + srli.d t2, a2, 7 + beqz t2, L(lt128) + xvld xr0, a1, 0 + xvld xr1, a1, 32 + + + xvld xr2, a1, 64 + xvld xr3, a1, 96 + addi.d a1, a1, 128 + addi.d a2, a2, -128 + + xvst xr0, a5, 0 + xvst xr1, a5, 32 + xvst xr2, a5, 64 + xvst xr3, a5, 96 + + addi.d a5, a5, 128 +L(lt128): + bltu a2, t1, L(lt64) + xvld xr0, a1, 0 + xvld xr1, a1, 32 + + addi.d a1, a1, 64 + addi.d a2, a2, -64 + xvst xr0, a5, 0 + xvst xr1, a5, 32 + + + addi.d a5, a5, 64 +L(lt64): + bltu a2, t0, L(lt32) + xvld xr0, a1, 0 + xvst xr0, a5, 0 + +L(lt32): + xvst xr8, a0, 0 + xvst xr9, a3, -32 + jr ra + nop + +L(copy_back): + addi.d a3, a3, -1 + addi.d a2, a2, -2 + andi t2, a3, 0x1f + xvld xr8, a1, 0 + + xvld xr9, a4, -32 + sub.d t3, a2, t2 + sub.d a5, a3, t2 + sub.d a4, a4, t2 + + + andi a2, t3, 0xff + beq a2, t3, L(back_lt256) + add.d a6, a1, a2 + addi.d a6, a6, 2 + +L(back_loop_256): + xvld xr0, a4, -33 + xvld xr1, a4, -65 + xvld xr2, a4, -97 + xvld xr3, a4, -129 + + xvld xr4, a4, -161 + xvld xr5, a4, -193 + xvld xr6, a4, -225 + xvld xr7, a4, -257 + + addi.d a4, a4, -256 + xvst xr0, a5, -32 + xvst xr1, a5, -64 + xvst xr2, a5, -96 + + + xvst xr3, a5, -128 + xvst xr4, a5, -160 + xvst xr5, a5, -192 + xvst xr6, a5, -224 + + xvst xr7, a5, -256 + addi.d a5, a5, -256 + bne a4, a6, L(back_loop_256) +L(back_lt256): + srli.d t2, a2, 7 + + beqz t2, L(back_lt128) + xvld xr0, a4, -33 + xvld xr1, a4, -65 + xvld xr2, a4, -97 + + xvld xr3, a4, -129 + addi.d a2, a2, -128 + addi.d a4, a4, -128 + xvst xr0, a5, -32 + + + xvst xr1, a5, -64 + xvst xr2, a5, -96 + xvst xr3, a5, -128 + addi.d a5, a5, -128 + +L(back_lt128): + blt a2, t1, L(back_lt64) + xvld xr0, a4, -33 + xvld xr1, a4, -65 + addi.d a2, a2, -64 + + addi.d a4, a4, -64 + xvst xr0, a5, -32 + xvst xr1, a5, -64 + addi.d a5, a5, -64 + +L(back_lt64): + bltu a2, t0, L(back_lt32) + xvld xr0, a4, -33 + xvst xr0, a5, -32 +L(back_lt32): + xvst xr8, a0, 0 + + + xvst xr9, a3, -31 + jr ra +END(MEMMOVE_NAME) + +libc_hidden_builtin_def (MEMCPY_NAME) +libc_hidden_builtin_def (MEMMOVE_NAME) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S b/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S new file mode 100644 index 0000000000..9e1502a79b --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memmove-lsx.S @@ -0,0 +1,534 @@ +/* Optimized memmove 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 MEMCPY_NAME __memcpy_lsx +# define MEMMOVE_NAME __memmove_lsx + +LEAF(MEMCPY_NAME, 6) + li.d t6, 16 + add.d a3, a0, a2 + add.d a4, a1, a2 + bgeu t6, a2, L(less_16bytes) + + li.d t8, 64 + li.d t7, 32 + bltu t8, a2, L(copy_long) + bltu t7, a2, L(more_32bytes) + + vld vr0, a1, 0 + vld vr1, a4, -16 + vst vr0, a0, 0 + vst vr1, a3, -16 + + jr ra +L(more_32bytes): + vld vr0, a1, 0 + vld vr1, a1, 16 + vld vr2, a4, -32 + + + vld vr3, a4, -16 + vst vr0, a0, 0 + vst vr1, a0, 16 + vst vr2, a3, -32 + + vst vr3, a3, -16 + jr ra +L(less_16bytes): + srli.d t0, a2, 3 + beqz t0, L(less_8bytes) + + vldrepl.d vr0, a1, 0 + vldrepl.d vr1, a4, -8 + vstelm.d vr0, a0, 0, 0 + vstelm.d vr1, a3, -8, 0 + + jr ra +L(less_8bytes): + srli.d t0, a2, 2 + beqz t0, L(less_4bytes) + vldrepl.w vr0, a1, 0 + + + vldrepl.w vr1, a4, -4 + vstelm.w vr0, a0, 0, 0 + vstelm.w vr1, a3, -4, 0 + jr ra + +L(less_4bytes): + srli.d t0, a2, 1 + beqz t0, L(less_2bytes) + vldrepl.h vr0, a1, 0 + vldrepl.h vr1, a4, -2 + + vstelm.h vr0, a0, 0, 0 + vstelm.h vr1, a3, -2, 0 + jr ra +L(less_2bytes): + beqz a2, L(less_1bytes) + + ld.b t0, a1, 0 + st.b t0, a0, 0 +L(less_1bytes): + jr ra + nop +END(MEMCPY_NAME) + +LEAF(MEMMOVE_NAME, 6) + li.d t6, 16 + add.d a3, a0, a2 + add.d a4, a1, a2 + bgeu t6, a2, L(less_16bytes) + + li.d t8, 64 + li.d t7, 32 + bltu t8, a2, L(move_long) + bltu t7, a2, L(more_32bytes) + + vld vr0, a1, 0 + vld vr1, a4, -16 + vst vr0, a0, 0 + vst vr1, a3, -16 + + jr ra + nop +L(move_long): + sub.d t0, a0, a1 + bltu t0, a2, L(copy_back) + + +L(copy_long): + vld vr2, a1, 0 + andi t0, a0, 0xf + sub.d t0, t6, t0 + add.d a1, a1, t0 + + sub.d a2, a2, t0 + andi t1, a1, 0xf + bnez t1, L(unaligned) + vld vr0, a1, 0 + + addi.d a2, a2, -16 + vst vr2, a0, 0 + andi t2, a2, 0x7f + add.d a5, a0, t0 + + beq a2, t2, L(al_less_128) + sub.d t3, a2, t2 + move a2, t2 + add.d a6, a1, t3 + + +L(al_loop): + vld vr1, a1, 16 + vld vr2, a1, 32 + vld vr3, a1, 48 + vld vr4, a1, 64 + + vld vr5, a1, 80 + vld vr6, a1, 96 + vld vr7, a1, 112 + vst vr0, a5, 0 + + vld vr0, a1, 128 + addi.d a1, a1, 128 + vst vr1, a5, 16 + vst vr2, a5, 32 + + vst vr3, a5, 48 + vst vr4, a5, 64 + vst vr5, a5, 80 + vst vr6, a5, 96 + + + vst vr7, a5, 112 + addi.d a5, a5, 128 + bne a1, a6, L(al_loop) +L(al_less_128): + blt a2, t8, L(al_less_64) + + vld vr1, a1, 16 + vld vr2, a1, 32 + vld vr3, a1, 48 + addi.d a2, a2, -64 + + vst vr0, a5, 0 + vld vr0, a1, 64 + addi.d a1, a1, 64 + vst vr1, a5, 16 + + vst vr2, a5, 32 + vst vr3, a5, 48 + addi.d a5, a5, 64 +L(al_less_64): + blt a2, t7, L(al_less_32) + + + vld vr1, a1, 16 + addi.d a2, a2, -32 + vst vr0, a5, 0 + vld vr0, a1, 32 + + addi.d a1, a1, 32 + vst vr1, a5, 16 + addi.d a5, a5, 32 +L(al_less_32): + blt a2, t6, L(al_less_16) + + vst vr0, a5, 0 + vld vr0, a1, 16 + addi.d a5, a5, 16 +L(al_less_16): + vld vr1, a4, -16 + + vst vr0, a5, 0 + vst vr1, a3, -16 + jr ra + nop + + +L(magic_num): + .dword 0x0706050403020100 + .dword 0x0f0e0d0c0b0a0908 +L(unaligned): + pcaddi t2, -4 + bstrins.d a1, zero, 3, 0 + vld vr8, t2, 0 + vld vr0, a1, 0 + + vld vr1, a1, 16 + addi.d a2, a2, -16 + vst vr2, a0, 0 + add.d a5, a0, t0 + + vreplgr2vr.b vr9, t1 + andi t2, a2, 0x7f + vadd.b vr9, vr9, vr8 + addi.d a1, a1, 32 + + + beq t2, a2, L(un_less_128) + sub.d t3, a2, t2 + move a2, t2 + add.d a6, a1, t3 + +L(un_loop): + vld vr2, a1, 0 + vld vr3, a1, 16 + vld vr4, a1, 32 + vld vr5, a1, 48 + + vld vr6, a1, 64 + vld vr7, a1, 80 + vshuf.b vr8, vr1, vr0, vr9 + vld vr0, a1, 96 + + vst vr8, a5, 0 + vshuf.b vr8, vr2, vr1, vr9 + vld vr1, a1, 112 + vst vr8, a5, 16 + + + addi.d a1, a1, 128 + vshuf.b vr2, vr3, vr2, vr9 + vshuf.b vr3, vr4, vr3, vr9 + vst vr2, a5, 32 + + vshuf.b vr4, vr5, vr4, vr9 + vst vr3, a5, 48 + vshuf.b vr5, vr6, vr5, vr9 + vst vr4, a5, 64 + + vshuf.b vr6, vr7, vr6, vr9 + vst vr5, a5, 80 + vshuf.b vr7, vr0, vr7, vr9 + vst vr6, a5, 96 + + vst vr7, a5, 112 + addi.d a5, a5, 128 + bne a1, a6, L(un_loop) +L(un_less_128): + blt a2, t8, L(un_less_64) + + + vld vr2, a1, 0 + vld vr3, a1, 16 + vshuf.b vr4, vr1, vr0, vr9 + vld vr0, a1, 32 + + vst vr4, a5, 0 + addi.d a2, a2, -64 + vshuf.b vr4, vr2, vr1, vr9 + vld vr1, a1, 48 + + addi.d a1, a1, 64 + vst vr4, a5, 16 + vshuf.b vr2, vr3, vr2, vr9 + vshuf.b vr3, vr0, vr3, vr9 + + vst vr2, a5, 32 + vst vr3, a5, 48 + addi.d a5, a5, 64 +L(un_less_64): + blt a2, t7, L(un_less_32) + + + vshuf.b vr3, vr1, vr0, vr9 + vld vr0, a1, 0 + vst vr3, a5, 0 + addi.d a2, a2, -32 + + vshuf.b vr3, vr0, vr1, vr9 + vld vr1, a1, 16 + addi.d a1, a1, 32 + vst vr3, a5, 16 + + addi.d a5, a5, 32 +L(un_less_32): + blt a2, t6, L(un_less_16) + vshuf.b vr2, vr1, vr0, vr9 + vor.v vr0, vr1, vr1 + + vld vr1, a1, 0 + vst vr2, a5, 0 + addi.d a5, a5, 16 +L(un_less_16): + vld vr2, a4, -16 + + + vshuf.b vr0, vr1, vr0, vr9 + vst vr0, a5, 0 + vst vr2, a3, -16 + jr ra + +L(copy_back): + addi.d t0, a3, -1 + vld vr2, a4, -16 + andi t0, t0, 0xf + addi.d t0, t0, 1 + + sub.d a4, a4, t0 + sub.d a2, a2, t0 + andi t1, a4, 0xf + bnez t1, L(back_unaligned) + + vld vr0, a4, -16 + addi.d a2, a2, -16 + vst vr2, a3, -16 + andi t2, a2, 0x7f + + + sub.d a3, a3, t0 + beq t2, a2, L(back_al_less_128) + sub.d t3, a2, t2 + move a2, t2 + + sub.d a6, a4, t3 +L(back_al_loop): + vld vr1, a4, -32 + vld vr2, a4, -48 + vld vr3, a4, -64 + + vld vr4, a4, -80 + vld vr5, a4, -96 + vld vr6, a4, -112 + vld vr7, a4, -128 + + vst vr0, a3, -16 + vld vr0, a4, -144 + addi.d a4, a4, -128 + vst vr1, a3, -32 + + + vst vr2, a3, -48 + vst vr3, a3, -64 + vst vr4, a3, -80 + vst vr5, a3, -96 + + vst vr6, a3, -112 + vst vr7, a3, -128 + addi.d a3, a3, -128 + bne a4, a6, L(back_al_loop) + +L(back_al_less_128): + blt a2, t8, L(back_al_less_64) + vld vr1, a4, -32 + vld vr2, a4, -48 + vld vr3, a4, -64 + + addi.d a2, a2, -64 + vst vr0, a3, -16 + vld vr0, a4, -80 + addi.d a4, a4, -64 + + + vst vr1, a3, -32 + vst vr2, a3, -48 + vst vr3, a3, -64 + addi.d a3, a3, -64 + +L(back_al_less_64): + blt a2, t7, L(back_al_less_32) + vld vr1, a4, -32 + addi.d a2, a2, -32 + vst vr0, a3, -16 + + vld vr0, a4, -48 + vst vr1, a3, -32 + addi.d a3, a3, -32 + addi.d a4, a4, -32 + +L(back_al_less_32): + blt a2, t6, L(back_al_less_16) + vst vr0, a3, -16 + vld vr0, a4, -32 + addi.d a3, a3, -16 + + +L(back_al_less_16): + vld vr1, a1, 0 + vst vr0, a3, -16 + vst vr1, a0, 0 + jr ra + +L(magic_num_2): + .dword 0x0706050403020100 + .dword 0x0f0e0d0c0b0a0908 +L(back_unaligned): + pcaddi t2, -4 + bstrins.d a4, zero, 3, 0 + vld vr8, t2, 0 + vld vr0, a4, 0 + + vld vr1, a4, -16 + addi.d a2, a2, -16 + vst vr2, a3, -16 + sub.d a3, a3, t0 + + + vreplgr2vr.b vr9, t1 + andi t2, a2, 0x7f + vadd.b vr9, vr9, vr8 + addi.d a4, a4, -16 + + beq t2, a2, L(back_un_less_128) + sub.d t3, a2, t2 + move a2, t2 + sub.d a6, a4, t3 + +L(back_un_loop): + vld vr2, a4, -16 + vld vr3, a4, -32 + vld vr4, a4, -48 + + vld vr5, a4, -64 + vld vr6, a4, -80 + vld vr7, a4, -96 + vshuf.b vr8, vr0, vr1, vr9 + + + vld vr0, a4, -112 + vst vr8, a3, -16 + vshuf.b vr8, vr1, vr2, vr9 + vld vr1, a4, -128 + + vst vr8, a3, -32 + addi.d a4, a4, -128 + vshuf.b vr2, vr2, vr3, vr9 + vshuf.b vr3, vr3, vr4, vr9 + + vst vr2, a3, -48 + vshuf.b vr4, vr4, vr5, vr9 + vst vr3, a3, -64 + vshuf.b vr5, vr5, vr6, vr9 + + vst vr4, a3, -80 + vshuf.b vr6, vr6, vr7, vr9 + vst vr5, a3, -96 + vshuf.b vr7, vr7, vr0, vr9 + + + vst vr6, a3, -112 + vst vr7, a3, -128 + addi.d a3, a3, -128 + bne a4, a6, L(back_un_loop) + +L(back_un_less_128): + blt a2, t8, L(back_un_less_64) + vld vr2, a4, -16 + vld vr3, a4, -32 + vshuf.b vr4, vr0, vr1, vr9 + + vld vr0, a4, -48 + vst vr4, a3, -16 + addi.d a2, a2, -64 + vshuf.b vr4, vr1, vr2, vr9 + + vld vr1, a4, -64 + addi.d a4, a4, -64 + vst vr4, a3, -32 + vshuf.b vr2, vr2, vr3, vr9 + + + vshuf.b vr3, vr3, vr0, vr9 + vst vr2, a3, -48 + vst vr3, a3, -64 + addi.d a3, a3, -64 + +L(back_un_less_64): + blt a2, t7, L(back_un_less_32) + vshuf.b vr3, vr0, vr1, vr9 + vld vr0, a4, -16 + vst vr3, a3, -16 + + addi.d a2, a2, -32 + vshuf.b vr3, vr1, vr0, vr9 + vld vr1, a4, -32 + addi.d a4, a4, -32 + + vst vr3, a3, -32 + addi.d a3, a3, -32 +L(back_un_less_32): + blt a2, t6, L(back_un_less_16) + vshuf.b vr2, vr0, vr1, vr9 + + + vor.v vr0, vr1, vr1 + vld vr1, a4, -16 + vst vr2, a3, -16 + addi.d a3, a3, -16 + +L(back_un_less_16): + vld vr2, a1, 0 + vshuf.b vr0, vr0, vr1, vr9 + vst vr0, a3, -16 + vst vr2, a0, 0 + + jr ra +END(MEMMOVE_NAME) + +libc_hidden_builtin_def (MEMCPY_NAME) +libc_hidden_builtin_def (MEMMOVE_NAME) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S b/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S new file mode 100644 index 0000000000..90a64b6bb9 --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memmove-unaligned.S @@ -0,0 +1,380 @@ +/* Optimized memmove_unaligned 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 MEMMOVE_NAME __memmove_unaligned + +# define LD_64(reg, n) \ + ld.d t0, reg, n; \ + ld.d t1, reg, n + 8; \ + ld.d t2, reg, n + 16; \ + ld.d t3, reg, n + 24; \ + ld.d t4, reg, n + 32; \ + ld.d t5, reg, n + 40; \ + ld.d t6, reg, n + 48; \ + ld.d t7, reg, n + 56; + +# define ST_64(reg, n) \ + st.d t0, reg, n; \ + st.d t1, reg, n + 8; \ + st.d t2, reg, n + 16; \ + st.d t3, reg, n + 24; \ + st.d t4, reg, n + 32; \ + st.d t5, reg, n + 40; \ + st.d t6, reg, n + 48; \ + st.d t7, reg, n + 56; + +LEAF(MEMMOVE_NAME, 3) + add.d a4, a1, a2 + add.d a3, a0, a2 + beq a1, a0, L(less_1bytes) + move t8, a0 + + srai.d a6, a2, 4 + beqz a6, L(less_16bytes) + srai.d a6, a2, 6 + bnez a6, L(more_64bytes) + srai.d a6, a2, 5 + beqz a6, L(less_32bytes) + + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a1, 16 + ld.d t3, a1, 24 + + ld.d t4, a4, -32 + ld.d t5, a4, -24 + ld.d t6, a4, -16 + ld.d t7, a4, -8 + + st.d t0, a0, 0 + st.d t1, a0, 8 + st.d t2, a0, 16 + st.d t3, a0, 24 + + st.d t4, a3, -32 + st.d t5, a3, -24 + st.d t6, a3, -16 + st.d t7, a3, -8 + + jr ra + +L(less_32bytes): + ld.d t0, a1, 0 + ld.d t1, a1, 8 + ld.d t2, a4, -16 + ld.d t3, a4, -8 + + st.d t0, a0, 0 + st.d t1, a0, 8 + st.d t2, a3, -16 + st.d t3, a3, -8 + + jr ra + +L(less_16bytes): + srai.d a6, a2, 3 + beqz a6, L(less_8bytes) + + ld.d t0, a1, 0 + ld.d t1, a4, -8 + st.d t0, a0, 0 + st.d t1, a3, -8 + + jr ra + +L(less_8bytes): + srai.d a6, a2, 2 + beqz a6, L(less_4bytes) + + ld.w t0, a1, 0 + ld.w t1, a4, -4 + st.w t0, a0, 0 + st.w t1, a3, -4 + + jr ra + +L(less_4bytes): + srai.d a6, a2, 1 + beqz a6, L(less_2bytes) + + ld.h t0, a1, 0 + ld.h t1, a4, -2 + st.h t0, a0, 0 + st.h t1, a3, -2 + + jr ra + +L(less_2bytes): + beqz a2, L(less_1bytes) + + ld.b t0, a1, 0 + st.b t0, a0, 0 + + jr ra + +L(less_1bytes): + jr ra + +L(more_64bytes): + sub.d a7, a0, a1 + bltu a7, a2, L(copy_backward) + +L(copy_forward): + srli.d a0, a0, 3 + slli.d a0, a0, 3 + beq a0, t8, L(all_align) + addi.d a0, a0, 0x8 + sub.d a7, t8, a0 + sub.d a1, a1, a7 + add.d a2, a7, a2 + +L(start_unalign_proc): + pcaddi t1, 18 + slli.d a6, a7, 3 + add.d t1, t1, a6 + jr t1 + + ld.b t0, a1, -7 + st.b t0, a0, -7 + ld.b t0, a1, -6 + st.b t0, a0, -6 + ld.b t0, a1, -5 + st.b t0, a0, -5 + ld.b t0, a1, -4 + st.b t0, a0, -4 + ld.b t0, a1, -3 + st.b t0, a0, -3 + ld.b t0, a1, -2 + st.b t0, a0, -2 + ld.b t0, a1, -1 + st.b t0, a0, -1 +L(start_over): + + addi.d a2, a2, -0x80 + blt a2, zero, L(end_unalign_proc) + +L(loop_less): + LD_64(a1, 0) + ST_64(a0, 0) + LD_64(a1, 64) + ST_64(a0, 64) + + addi.d a0, a0, 0x80 + addi.d a1, a1, 0x80 + addi.d a2, a2, -0x80 + bge a2, zero, L(loop_less) + +L(end_unalign_proc): + addi.d a2, a2, 0x80 + + pcaddi t1, 36 + andi t2, a2, 0x78 + add.d a1, a1, t2 + add.d a0, a0, t2 + sub.d t1, t1, t2 + jr t1 + + ld.d t0, a1, -120 + st.d t0, a0, -120 + ld.d t0, a1, -112 + st.d t0, a0, -112 + ld.d t0, a1, -104 + st.d t0, a0, -104 + ld.d t0, a1, -96 + st.d t0, a0, -96 + ld.d t0, a1, -88 + st.d t0, a0, -88 + ld.d t0, a1, -80 + st.d t0, a0, -80 + ld.d t0, a1, -72 + st.d t0, a0, -72 + ld.d t0, a1, -64 + st.d t0, a0, -64 + ld.d t0, a1, -56 + st.d t0, a0, -56 + ld.d t0, a1, -48 + st.d t0, a0, -48 + ld.d t0, a1, -40 + st.d t0, a0, -40 + ld.d t0, a1, -32 + st.d t0, a0, -32 + ld.d t0, a1, -24 + st.d t0, a0, -24 + ld.d t0, a1, -16 + st.d t0, a0, -16 + ld.d t0, a1, -8 + st.d t0, a0, -8 + + andi a2, a2, 0x7 + pcaddi t1, 18 + slli.d a2, a2, 3 + sub.d t1, t1, a2 + jr t1 + + ld.b t0, a4, -7 + st.b t0, a3, -7 + ld.b t0, a4, -6 + st.b t0, a3, -6 + ld.b t0, a4, -5 + st.b t0, a3, -5 + ld.b t0, a4, -4 + st.b t0, a3, -4 + ld.b t0, a4, -3 + st.b t0, a3, -3 + ld.b t0, a4, -2 + st.b t0, a3, -2 + ld.b t0, a4, -1 + st.b t0, a3, -1 +L(end): + move a0, t8 + jr ra + +L(all_align): + addi.d a1, a1, 0x8 + addi.d a0, a0, 0x8 + ld.d t0, a1, -8 + st.d t0, a0, -8 + addi.d a2, a2, -8 + b L(start_over) + +L(all_align_back): + addi.d a4, a4, -0x8 + addi.d a3, a3, -0x8 + ld.d t0, a4, 0 + st.d t0, a3, 0 + addi.d a2, a2, -8 + b L(start_over_back) + +L(copy_backward): + move a5, a3 + srli.d a3, a3, 3 + slli.d a3, a3, 3 + beq a3, a5, L(all_align_back) + sub.d a7, a3, a5 + add.d a4, a4, a7 + add.d a2, a7, a2 + + pcaddi t1, 18 + slli.d a6, a7, 3 + add.d t1, t1, a6 + jr t1 + + ld.b t0, a4, 6 + st.b t0, a3, 6 + ld.b t0, a4, 5 + st.b t0, a3, 5 + ld.b t0, a4, 4 + st.b t0, a3, 4 + ld.b t0, a4, 3 + st.b t0, a3, 3 + ld.b t0, a4, 2 + st.b t0, a3, 2 + ld.b t0, a4, 1 + st.b t0, a3, 1 + ld.b t0, a4, 0 + st.b t0, a3, 0 +L(start_over_back): + addi.d a2, a2, -0x80 + blt a2, zero, L(end_unalign_proc_back) + +L(loop_less_back): + LD_64(a4, -64) + ST_64(a3, -64) + LD_64(a4, -128) + ST_64(a3, -128) + + addi.d a4, a4, -0x80 + addi.d a3, a3, -0x80 + addi.d a2, a2, -0x80 + bge a2, zero, L(loop_less_back) + +L(end_unalign_proc_back): + addi.d a2, a2, 0x80 + + pcaddi t1, 36 + andi t2, a2, 0x78 + sub.d a4, a4, t2 + sub.d a3, a3, t2 + sub.d t1, t1, t2 + jr t1 + + ld.d t0, a4, 112 + st.d t0, a3, 112 + ld.d t0, a4, 104 + st.d t0, a3, 104 + ld.d t0, a4, 96 + st.d t0, a3, 96 + ld.d t0, a4, 88 + st.d t0, a3, 88 + ld.d t0, a4, 80 + st.d t0, a3, 80 + ld.d t0, a4, 72 + st.d t0, a3, 72 + ld.d t0, a4, 64 + st.d t0, a3, 64 + ld.d t0, a4, 56 + st.d t0, a3, 56 + ld.d t0, a4, 48 + st.d t0, a3, 48 + ld.d t0, a4, 40 + st.d t0, a3, 40 + ld.d t0, a4, 32 + st.d t0, a3, 32 + ld.d t0, a4, 24 + st.d t0, a3, 24 + ld.d t0, a4, 16 + st.d t0, a3, 16 + ld.d t0, a4, 8 + st.d t0, a3, 8 + ld.d t0, a4, 0 + st.d t0, a3, 0 + + andi a2, a2, 0x7 + pcaddi t1, 18 + slli.d a2, a2, 3 + sub.d t1, t1, a2 + jr t1 + + ld.b t0, a1, 6 + st.b t0, a0, 6 + ld.b t0, a1, 5 + st.b t0, a0, 5 + ld.b t0, a1, 4 + st.b t0, a0, 4 + ld.b t0, a1, 3 + st.b t0, a0, 3 + ld.b t0, a1, 2 + st.b t0, a0, 2 + ld.b t0, a1, 1 + st.b t0, a0, 1 + ld.b t0, a1, 0 + st.b t0, a0, 0 + + move a0, t8 + jr ra +END(MEMMOVE_NAME) + +libc_hidden_builtin_def (MEMMOVE_NAME) +#endif diff --git a/sysdeps/loongarch/lp64/multiarch/memmove.c b/sysdeps/loongarch/lp64/multiarch/memmove.c new file mode 100644 index 0000000000..7e3ca4c47b --- /dev/null +++ b/sysdeps/loongarch/lp64/multiarch/memmove.c @@ -0,0 +1,38 @@ +/* Multiple versions of memmove. + 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 + . */ + +/* Define multiple versions only for the definition in libc. */ +#if IS_IN (libc) +# define memmove __redirect_memmove +# include +# undef memmove + +# define SYMBOL_NAME memmove +# include "ifunc-lasx.h" + +libc_ifunc_redirected (__redirect_memmove, __libc_memmove, + IFUNC_SELECTOR ()); +strong_alias (__libc_memmove, memmove); + +# ifdef SHARED +__hidden_ver1 (__libc_memmove, __GI_memmove, __redirect_memmove) + __attribute__ ((visibility ("hidden"))); +# endif + +#endif