From patchwork Wed May 12 14:27:17 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Matheus Castanho X-Patchwork-Id: 43394 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id C3D5D398B840; Wed, 12 May 2021 14:27:28 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org C3D5D398B840 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1620829648; bh=coJdJD4m/QfX2cu0bPEDS061wnm8/QMyKHETP9HQJ+4=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=sl/B4nsnYJp20YPqIIKDJSTGwKBL5n69W6ULN+Xp1NQPz9ABZg1xZxk/T6Qyy8P/i YZdVw3uOFdagWF1fw9f+KaOy6po96RJXm0WJP7AHlGgfm0U3sRy1ajUlSt1RTa1h2n mQJ8L43Bus6CjxnkHv6yFVR+O4J9TboNlJBpnjr8= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mx0a-001b2d01.pphosted.com (mx0a-001b2d01.pphosted.com [148.163.156.1]) by sourceware.org (Postfix) with ESMTPS id 571B6383583F for ; Wed, 12 May 2021 14:27:25 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org 571B6383583F Received: from pps.filterd (m0098399.ppops.net [127.0.0.1]) by mx0a-001b2d01.pphosted.com (8.16.0.43/8.16.0.43) with SMTP id 14CEKBth043029 for ; Wed, 12 May 2021 10:27:24 -0400 Received: from ppma02wdc.us.ibm.com (aa.5b.37a9.ip4.static.sl-reverse.com [169.55.91.170]) by mx0a-001b2d01.pphosted.com with ESMTP id 38gefnvx09-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT) for ; Wed, 12 May 2021 10:27:24 -0400 Received: from pps.filterd (ppma02wdc.us.ibm.com [127.0.0.1]) by ppma02wdc.us.ibm.com (8.16.0.43/8.16.0.43) with SMTP id 14CEGMYe008996 for ; Wed, 12 May 2021 14:27:23 GMT Received: from b03cxnp07028.gho.boulder.ibm.com (b03cxnp07028.gho.boulder.ibm.com [9.17.130.15]) by ppma02wdc.us.ibm.com with ESMTP id 38dj99nyb1-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT) for ; Wed, 12 May 2021 14:27:23 +0000 Received: from b03ledav004.gho.boulder.ibm.com (b03ledav004.gho.boulder.ibm.com [9.17.130.235]) by b03cxnp07028.gho.boulder.ibm.com (8.14.9/8.14.9/NCO v10.0) with ESMTP id 14CERMwt17367386 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Wed, 12 May 2021 14:27:22 GMT Received: from b03ledav004.gho.boulder.ibm.com (unknown [127.0.0.1]) by IMSVA (Postfix) with ESMTP id 0E4087805F for ; Wed, 12 May 2021 14:27:22 +0000 (GMT) Received: from b03ledav004.gho.boulder.ibm.com (unknown [127.0.0.1]) by IMSVA (Postfix) with ESMTP id 62A467806B for ; Wed, 12 May 2021 14:27:21 +0000 (GMT) Received: from localhost (unknown [9.160.142.149]) by b03ledav004.gho.boulder.ibm.com (Postfix) with ESMTP for ; Wed, 12 May 2021 14:27:21 +0000 (GMT) To: libc-alpha@sourceware.org Subject: [PATCH 2/2] powerpc: Add optimized rawmemchr for POWER10 Date: Wed, 12 May 2021 11:27:17 -0300 Message-Id: <20210512142717.225429-2-msc@linux.ibm.com> X-Mailer: git-send-email 2.31.1 In-Reply-To: <20210512142717.225429-1-msc@linux.ibm.com> References: <20210512142717.225429-1-msc@linux.ibm.com> X-TM-AS-GCONF: 00 X-Proofpoint-ORIG-GUID: VqIZ45b1yO5cnJXzhcvF5lQANU5uEKjR X-Proofpoint-GUID: VqIZ45b1yO5cnJXzhcvF5lQANU5uEKjR X-Proofpoint-UnRewURL: 0 URL was un-rewritten MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:6.0.391, 18.0.761 definitions=2021-05-12_07:2021-05-12, 2021-05-12 signatures=0 X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 spamscore=0 priorityscore=1501 phishscore=0 adultscore=0 lowpriorityscore=0 bulkscore=0 mlxlogscore=999 suspectscore=0 clxscore=1015 mlxscore=0 malwarescore=0 impostorscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2104190000 definitions=main-2105120094 X-Spam-Status: No, score=-12.2 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_EF, GIT_PATCH_0, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_LOW, RCVD_IN_MSPIKE_H2, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Matheus Castanho via Libc-alpha From: Matheus Castanho Reply-To: Matheus Castanho Errors-To: libc-alpha-bounces@sourceware.org Sender: "Libc-alpha" Reuse code for optimized strlen to implement a faster version of rawmemchr. This takes advantage of the same benefits provided by the strlen implementation, but needs some extra steps. The functionality and performance of __strlen_power10 is unchanged after this commit. rawmemchr returns a pointer to the char found, while strlen returns only the length, so we have to take that into account when preparing the return value. To quickly check 64B, the loop on __strlen_power10 merges the whole block into 16B by using unsigned minimum vector operations (vminub) and checks if there are any \0 on the resulting vector. The same code is used by rawmemchr if the char c is 0. However, this approach does not work when c != 0. We first need to subtract each byte by c, so that the value we are looking for is converted to a 0, then taking the minimum and checking for nulls works again. The new code branches after it has compared ~256 bytes and chooses which of the two strategies above will be used in the main loop, based on the char c. This extra branch adds some overhead (~5%) for length ~256, but is quickly amortized by the faster loop for larger sizes. Compared to __rawmemchr_power9, this version is ~20% faster for length < 256. Because of the optimized main loop, the improvement becomes ~35% for c != 0 and ~50% for c = 0 for strings longer than 256. Reviewed-by: Lucas A. M. Magalhaes --- .../powerpc/powerpc64/le/power10/rawmemchr.S | 22 +++ sysdeps/powerpc/powerpc64/le/power10/strlen.S | 156 +++++++++++++++--- sysdeps/powerpc/powerpc64/multiarch/Makefile | 4 +- .../powerpc64/multiarch/ifunc-impl-list.c | 4 + .../powerpc64/multiarch/rawmemchr-power10.S | 21 +++ .../powerpc/powerpc64/multiarch/rawmemchr.c | 4 + 6 files changed, 184 insertions(+), 27 deletions(-) create mode 100644 sysdeps/powerpc/powerpc64/le/power10/rawmemchr.S create mode 100644 sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power10.S diff --git a/sysdeps/powerpc/powerpc64/le/power10/rawmemchr.S b/sysdeps/powerpc/powerpc64/le/power10/rawmemchr.S new file mode 100644 index 0000000000..5351c2634f --- /dev/null +++ b/sysdeps/powerpc/powerpc64/le/power10/rawmemchr.S @@ -0,0 +1,22 @@ +/* Optimized rawmemchr implementation for POWER10 LE. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include + +#define USE_AS_RAWMEMCHR 1 +#include diff --git a/sysdeps/powerpc/powerpc64/le/power10/strlen.S b/sysdeps/powerpc/powerpc64/le/power10/strlen.S index ca7e9eb3d8..18222ebeb8 100644 --- a/sysdeps/powerpc/powerpc64/le/power10/strlen.S +++ b/sysdeps/powerpc/powerpc64/le/power10/strlen.S @@ -18,10 +18,50 @@ #include -#ifndef STRLEN -# define STRLEN __strlen -# define DEFINE_STRLEN_HIDDEN_DEF 1 -#endif +/* To reuse the code for rawmemchr, we have some extra steps compared to the + strlen implementation: + - Sum the initial value of r3 with the position at which the char was + found, to guarantee we return a pointer and not the length. + - In the main loop, subtract each byte by the char we are looking for, + so we can keep using vminub to quickly check 64B at once. */ +#ifdef USE_AS_RAWMEMCHR +# ifndef RAWMEMCHR +# define FUNCNAME __rawmemchr +# else +# define FUNCNAME RAWMEMCHR +# endif +# define MCOUNT_NARGS 2 +# define VREG_ZERO v20 +# define OFF_START_LOOP 256 +# define RAWMEMCHR_SUBTRACT_VECTORS \ + vsububm v4,v4,v18; \ + vsububm v5,v5,v18; \ + vsububm v6,v6,v18; \ + vsububm v7,v7,v18; +# define TAIL(vreg,increment) \ + vctzlsbb r4,vreg; \ + addi r4,r4,increment; \ + add r3,r5,r4; \ + blr + +#else /* strlen */ + +# ifndef STRLEN +# define FUNCNAME __strlen +# define DEFINE_STRLEN_HIDDEN_DEF 1 +# else +# define FUNCNAME STRLEN +# endif +# define MCOUNT_NARGS 1 +# define VREG_ZERO v18 +# define OFF_START_LOOP 192 +# define TAIL(vreg,increment) \ + vctzlsbb r4,vreg; \ + subf r3,r3,r5; \ + addi r4,r4,increment; \ + add r3,r3,r4; \ + blr +#endif /* USE_AS_RAWMEMCHR */ /* TODO: Replace macros by the actual instructions when minimum binutils becomes >= 2.35. This is used to keep compatibility with older versions. */ @@ -50,19 +90,13 @@ li r6,offset; \ LXVP(v4+32,offset,addr); \ LXVP(v6+32,offset+32,addr); \ + RAWMEMCHR_SUBTRACT_VECTORS; \ vminub v14,v4,v5; \ vminub v15,v6,v7; \ vminub v16,v14,v15; \ - vcmpequb. v0,v16,v18; \ + vcmpequb. v0,v16,VREG_ZERO; \ bne cr6,L(label) -#define TAIL(vreg,increment) \ - vctzlsbb r4,vreg; \ - subf r3,r3,r5; \ - addi r4,r4,increment; \ - add r3,r3,r4; \ - blr - /* Implements the function int [r3] strlen (const void *s [r3]) @@ -72,11 +106,21 @@ .machine power9 -ENTRY_TOCLESS (STRLEN, 4) - CALL_MCOUNT 1 +ENTRY_TOCLESS (FUNCNAME, 4) + CALL_MCOUNT MCOUNT_NARGS + +#ifdef USE_AS_RAWMEMCHR + xori r5,r4,0xff - vspltisb v18,0 + mtvsrd v18+32,r4 /* matching char in v18 */ + mtvsrd v19+32,r5 /* non matching char in v19 */ + + vspltb v18,v18,7 /* replicate */ + vspltb v19,v19,7 /* replicate */ +#else vspltisb v19,-1 +#endif + vspltisb VREG_ZERO,0 /* Next 16B-aligned address. Prepare address for L(aligned). */ addi r5,r3,16 @@ -90,16 +134,25 @@ ENTRY_TOCLESS (STRLEN, 4) vcmpequb. v6,v0,v18 beq cr6,L(aligned) +#ifdef USE_AS_RAWMEMCHR + vctzlsbb r6,v6 + add r3,r3,r6 +#else vctzlsbb r3,v6 +#endif blr - /* Test next 176B, 16B at a time. The main loop is optimized for longer - strings, so checking the first bytes in 16B chunks benefits a lot - small strings. */ + /* Test up to OFF_START_LOOP-16 bytes in 16B chunks. The main loop is + optimized for longer strings, so checking the first bytes in 16B + chunks benefits a lot small strings. */ .p2align 5 L(aligned): +#ifdef USE_AS_RAWMEMCHR + cmpdi cr5,r4,0 /* Check if c == 0. This will be useful to + choose how we will perform the main loop. */ +#endif /* Prepare address for the loop. */ - addi r4,r3,192 + addi r4,r3,OFF_START_LOOP clrrdi r4,r4,6 CHECK16(v0,0,r5,tail1) @@ -113,15 +166,43 @@ L(aligned): CHECK16(v8,128,r5,tail9) CHECK16(v9,144,r5,tail10) CHECK16(v10,160,r5,tail11) +#ifdef USE_AS_RAWMEMCHR + CHECK16(v0,176,r5,tail12) + CHECK16(v1,192,r5,tail13) + CHECK16(v2,208,r5,tail14) + CHECK16(v3,224,r5,tail15) +#endif addi r5,r4,128 +#ifdef USE_AS_RAWMEMCHR + /* If c == 0, use the same loop as strlen, without the vsububm. */ + beq cr5,L(loop) + + /* This is very similar to the block after L(loop), the difference is + that here RAWMEMCHR_SUBTRACT_VECTORS is not empty, and we subtract + each byte loaded by the char we are looking for, this way we can keep + using vminub to merge the results and checking for nulls. */ + .p2align 5 +L(rawmemchr_loop): + CHECK64(0,r4,pre_tail_64b) + CHECK64(64,r4,pre_tail_64b) + addi r4,r4,256 + + CHECK64(0,r5,tail_64b) + CHECK64(64,r5,tail_64b) + addi r5,r5,256 + + b L(rawmemchr_loop) +#endif /* Switch to a more aggressive approach checking 64B each time. Use 2 pointers 128B apart and unroll the loop once to make the pointer updates and usages separated enough to avoid stalls waiting for address calculation. */ .p2align 5 L(loop): +#undef RAWMEMCHR_SUBTRACT_VECTORS +#define RAWMEMCHR_SUBTRACT_VECTORS /* nothing */ CHECK64(0,r4,pre_tail_64b) CHECK64(64,r4,pre_tail_64b) addi r4,r4,256 @@ -140,10 +221,10 @@ L(tail_64b): block and mark it in its corresponding VR. lxvp vx,0(ry) puts the low 16B bytes into vx+1, and the high into vx, so the order here is v5, v4, v7, v6. */ - vcmpequb v1,v5,v18 - vcmpequb v2,v4,v18 - vcmpequb v3,v7,v18 - vcmpequb v4,v6,v18 + vcmpequb v1,v5,VREG_ZERO + vcmpequb v2,v4,VREG_ZERO + vcmpequb v3,v7,VREG_ZERO + vcmpequb v4,v6,VREG_ZERO /* Take into account the other 64B blocks we had already checked. */ add r5,r5,r6 @@ -165,7 +246,9 @@ L(tail_64b): or r10,r8,r7 cnttzd r0,r10 /* Count trailing zeros before the match. */ +#ifndef USE_AS_RAWMEMCHR subf r5,r3,r5 +#endif add r3,r5,r0 /* Compute final length. */ blr @@ -213,9 +296,32 @@ L(tail10): L(tail11): TAIL(v10,160) -END (STRLEN) +#ifdef USE_AS_RAWMEMCHR + .p2align 5 +L(tail12): + TAIL(v0,176) + + .p2align 5 +L(tail13): + TAIL(v1,192) + + .p2align 5 +L(tail14): + TAIL(v2,208) + + .p2align 5 +L(tail15): + TAIL(v3,224) +#endif + +END (FUNCNAME) -#ifdef DEFINE_STRLEN_HIDDEN_DEF +#ifdef USE_AS_RAWMEMCHR +weak_alias (__rawmemchr,rawmemchr) +libc_hidden_builtin_def (__rawmemchr) +#else +# ifdef DEFINE_STRLEN_HIDDEN_DEF weak_alias (__strlen, strlen) libc_hidden_builtin_def (strlen) +# endif #endif diff --git a/sysdeps/powerpc/powerpc64/multiarch/Makefile b/sysdeps/powerpc/powerpc64/multiarch/Makefile index ea50b61674..e6e013db17 100644 --- a/sysdeps/powerpc/powerpc64/multiarch/Makefile +++ b/sysdeps/powerpc/powerpc64/multiarch/Makefile @@ -33,9 +33,9 @@ sysdep_routines += memcpy-power8-cached memcpy-power7 memcpy-a2 memcpy-power6 \ ifneq (,$(filter %le,$(config-machine))) sysdep_routines += memcpy-power10 memmove-power10 memset-power10 \ + rawmemchr-power9 rawmemchr-power10 \ strcmp-power9 strncmp-power9 strcpy-power9 stpcpy-power9 \ - rawmemchr-power9 strlen-power9 strncpy-power9 stpncpy-power9 \ - strlen-power10 + strlen-power9 strncpy-power9 stpncpy-power9 strlen-power10 endif CFLAGS-strncase-power7.c += -mcpu=power7 -funroll-loops CFLAGS-strncase_l-power7.c += -mcpu=power7 -funroll-loops diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c index b123c6a3d3..a92b67448e 100644 --- a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c +++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c @@ -257,6 +257,10 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array, /* Support sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c. */ IFUNC_IMPL (i, name, rawmemchr, #ifdef __LITTLE_ENDIAN__ + IFUNC_IMPL_ADD (array, i, rawmemchr, + (hwcap2 & PPC_FEATURE2_ARCH_3_1) + && (hwcap & PPC_FEATURE_HAS_VSX), + __rawmemchr_power10) IFUNC_IMPL_ADD (array, i, rawmemchr, hwcap2 & PPC_FEATURE2_ARCH_3_00, __rawmemchr_power9) diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power10.S b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power10.S new file mode 100644 index 0000000000..bf1ed7e194 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr-power10.S @@ -0,0 +1,21 @@ +/* Optimized rawmemchr implementation for PowerPC64/POWER10. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#define RAWMEMCHR __rawmemchr_power10 + +#include diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c index d906a4ea50..c0ffea2b93 100644 --- a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c @@ -26,6 +26,7 @@ extern __typeof (__rawmemchr) __rawmemchr_ppc attribute_hidden; extern __typeof (__rawmemchr) __rawmemchr_power7 attribute_hidden; # ifdef __LITTLE_ENDIAN__ extern __typeof (__rawmemchr) __rawmemchr_power9 attribute_hidden; +extern __typeof (__rawmemchr) __rawmemchr_power10 attribute_hidden; # endif # undef __rawmemchr @@ -34,6 +35,9 @@ extern __typeof (__rawmemchr) __rawmemchr_power9 attribute_hidden; ifunc symbol properly. */ libc_ifunc_redirected (__redirect___rawmemchr, __rawmemchr, # ifdef __LITTLE_ENDIAN__ + (hwcap2 & PPC_FEATURE2_ARCH_3_1) + && (hwcap & PPC_FEATURE_HAS_VSX) + ? __rawmemchr_power10 : (hwcap2 & PPC_FEATURE2_ARCH_3_00) ? __rawmemchr_power9 : # endif