From patchwork Wed Jun 30 15:36:07 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Pedro Franco de Carvalho X-Patchwork-Id: 44064 X-Patchwork-Delegate: msc@linux.ibm.com 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 9642F393D022 for ; Wed, 30 Jun 2021 15:37:00 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 9642F393D022 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1625067420; bh=7PdzCxa/2VmbpWFVHVhheytUdkGLe4cnL6IO2KFsDhU=; h=To:Subject:Date:List-Id:List-Unsubscribe:List-Archive:List-Post: List-Help:List-Subscribe:From:Reply-To:From; b=O1JrGn9fpO8rKr4tlcu9rXVEaeQiH27HxrNR7muPJucpilbOQo1UlnX7t2u5+d5UD sC59sEdf7ffhJWTi17U3lHkae92IwXs8SpFaJmgT4bk9/PZ7qFDm7AJ9nvqOO72J2b mR0m+h2Zkbu0gIkvssXkhJhC3JYiydy8FL/Sfuzw= 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 E34D63857836 for ; Wed, 30 Jun 2021 15:36:32 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org E34D63857836 Received: from pps.filterd (m0187473.ppops.net [127.0.0.1]) by mx0a-001b2d01.pphosted.com (8.16.0.43/8.16.0.43) with SMTP id 15UFYLoU139840 for ; Wed, 30 Jun 2021 11:36:32 -0400 Received: from ppma04wdc.us.ibm.com (1a.90.2fa9.ip4.static.sl-reverse.com [169.47.144.26]) by mx0a-001b2d01.pphosted.com with ESMTP id 39gtqmhg3f-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT) for ; Wed, 30 Jun 2021 11:36:31 -0400 Received: from pps.filterd (ppma04wdc.us.ibm.com [127.0.0.1]) by ppma04wdc.us.ibm.com (8.16.1.2/8.16.1.2) with SMTP id 15UFTl87019686 for ; Wed, 30 Jun 2021 15:36:30 GMT Received: from b03cxnp08025.gho.boulder.ibm.com (b03cxnp08025.gho.boulder.ibm.com [9.17.130.17]) by ppma04wdc.us.ibm.com with ESMTP id 39duvd2756-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT) for ; Wed, 30 Jun 2021 15:36:30 +0000 Received: from b03ledav005.gho.boulder.ibm.com (b03ledav005.gho.boulder.ibm.com [9.17.130.236]) by b03cxnp08025.gho.boulder.ibm.com (8.14.9/8.14.9/NCO v10.0) with ESMTP id 15UFaTka47710608 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Wed, 30 Jun 2021 15:36:29 GMT Received: from b03ledav005.gho.boulder.ibm.com (unknown [127.0.0.1]) by IMSVA (Postfix) with ESMTP id 03ECFBE05A for ; Wed, 30 Jun 2021 15:36:29 +0000 (GMT) Received: from b03ledav005.gho.boulder.ibm.com (unknown [127.0.0.1]) by IMSVA (Postfix) with ESMTP id 3310CBE062 for ; Wed, 30 Jun 2021 15:36:28 +0000 (GMT) Received: from pedro.localdomain (unknown [9.171.84.154]) by b03ledav005.gho.boulder.ibm.com (Postfix) with SMTP for ; Wed, 30 Jun 2021 15:36:27 +0000 (GMT) Received: by pedro.localdomain (Postfix, from userid 1000) id C65B03C0F02; Wed, 30 Jun 2021 12:36:22 -0300 (-03) To: libc-alpha@sourceware.org Subject: [PATCH] powerpc: optimize strcpy/stpcpy for POWER9/10 Date: Wed, 30 Jun 2021 12:36:07 -0300 Message-Id: <20210630153607.63317-1-pedromfc@linux.ibm.com> X-Mailer: git-send-email 2.30.2 MIME-Version: 1.0 X-TM-AS-GCONF: 00 X-Proofpoint-GUID: s49KwPVS3ywVt8Ya2J_6KMh2EmBdRAzH X-Proofpoint-ORIG-GUID: s49KwPVS3ywVt8Ya2J_6KMh2EmBdRAzH X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:6.0.391, 18.0.790 definitions=2021-06-30_06:2021-06-30, 2021-06-30 signatures=0 X-Proofpoint-Spam-Details: rule=outbound_notspam policy=outbound score=0 lowpriorityscore=0 adultscore=0 priorityscore=1501 bulkscore=0 spamscore=0 suspectscore=0 phishscore=0 malwarescore=0 clxscore=1011 mlxscore=0 impostorscore=0 mlxlogscore=999 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2104190000 definitions=main-2106300092 X-Spam-Status: No, score=-10.5 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_EF, GIT_PATCH_0, KAM_NUMSUBJECT, 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: Pedro Franco de Carvalho via Libc-alpha From: Pedro Franco de Carvalho Reply-To: Pedro Franco de Carvalho Errors-To: libc-alpha-bounces+patchwork=sourceware.org@sourceware.org Sender: "Libc-alpha" This patch modifies the current POWER9 implementation of strcpy and stpcpy to optimize it for POWER9/10. Since no new POWER10 instructions are used, the original POWER9 strcpy is modified instead of creating a new implementation for POWER10. This implementation is based on both the original POWER9 implementation of strcpy and the preamble of the new POWER10 implementation of strlen. The changes also affect stpcpy, which uses the same implementation with some additional code before returning. On POWER9, averaging improvements across the benchmark inputs (length/source alignement/destination alignment), for an expermient that ran the benchmark five times, bench-strcpy showed an improvement of 5.23%, and bench-stpcpy showed an improvement of 6.59%. On POWER10, bench-strcpy showed 13.16%, and bench-stpcpy showed 13.59%. The changes are: 1. Removed the null string optimization. Although this results in a few extra cycles for the null string, in combination with the second change, this resulted in improvements for for other cases. 2. Adapted the preamble from strlen for POWER10. This is the part of the function that handles up to the first 16 bytes of the string. 3. Increased number of unrolled iterations in the main loop to 6. Reviewed-by: Matheus Castanho --- sysdeps/powerpc/powerpc64/le/power9/strcpy.S | 160 +++++++++++-------- 1 file changed, 89 insertions(+), 71 deletions(-) diff --git a/sysdeps/powerpc/powerpc64/le/power9/strcpy.S b/sysdeps/powerpc/powerpc64/le/power9/strcpy.S index 76cfcae200..b9e5afdcbe 100644 --- a/sysdeps/powerpc/powerpc64/le/power9/strcpy.S +++ b/sysdeps/powerpc/powerpc64/le/power9/strcpy.S @@ -45,91 +45,78 @@ The implementation can load bytes past a null terminator, but only up to the next 16B boundary, so it never crosses a page. */ +/* Load quadword at addr+offset to vreg, check for null bytes, + and branch to label if any are found. */ +#define CHECK16(vreg,offset,addr,label) \ + lxv vreg+32,offset(addr); \ + vcmpequb. v6,vreg,v18; \ + bne cr6,L(label); + .machine power9 ENTRY_TOCLESS (FUNC_NAME, 4) CALL_MCOUNT 2 - /* NULL string optimisation */ - lbz r0,0(r4) - stb r0,0(r3) - cmpwi r0,0 - beqlr - - addi r4,r4,1 - addi r11,r3,1 - vspltisb v18,0 /* Zeroes in v18 */ + vspltisb v19,-1 /* 0xFF bytes in v19 */ - neg r5,r4 - rldicl r9,r5,0,60 /* How many bytes to get source 16B aligned? */ + /* Next 16B-aligned address. Prepare address for L(loop). */ + addi r5,r4,16 + clrrdi r5,r5,4 + subf r8,r4,r5 + add r11,r3,r8 - /* Get source 16B aligned */ + /* Align data and fill bytes not loaded with non matching char. */ lvx v0,0,r4 lvsr v1,0,r4 - vperm v0,v18,v0,v1 - - vcmpequb v6,v0,v18 /* 0xff if byte is NULL, 0x00 otherwise */ - vctzlsbb r7,v6 /* Number of trailing zeroes */ - addi r8,r7,1 /* Add null terminator */ + vperm v0,v19,v0,v1 - /* r8 = bytes including null - r9 = bytes to get source 16B aligned - if r8 > r9 - no null, copy r9 bytes - else - there is a null, copy r8 bytes and return. */ - cmpd r8,r9 - bgt L(no_null) + vcmpequb. v6,v0,v18 /* 0xff if byte is NULL, 0x00 otherwise */ + beq cr6,L(no_null) - sldi r10,r8,56 /* stxvl wants size in top 8 bits */ - stxvl 32+v0,r11,r10 /* Partial store */ + /* There's a null byte. */ + vctzlsbb r8,v6 /* Number of trailing zeroes */ + addi r9,r8,1 /* Add null byte. */ + sldi r10,r9,56 /* stxvl wants size in top 8 bits. */ + stxvl 32+v0,r3,r10 /* Partial store */ #ifdef USE_AS_STPCPY /* stpcpy returns the dest address plus the size not counting the final '\0'. */ - add r3,r11,r7 + add r3,r3,r8 #endif blr L(no_null): - sldi r10,r9,56 /* stxvl wants size in top 8 bits */ - stxvl 32+v0,r11,r10 /* Partial store */ - - add r4,r4,r9 - add r11,r11,r9 + sldi r10,r8,56 /* stxvl wants size in top 8 bits */ + stxvl 32+v0,r3,r10 /* Partial store */ + .p2align 4 L(loop): - lxv 32+v0,0(r4) - vcmpequb. v6,v0,v18 /* Any zero bytes? */ - bne cr6,L(tail1) - - lxv 32+v1,16(r4) - vcmpequb. v6,v1,v18 /* Any zero bytes? */ - bne cr6,L(tail2) - - lxv 32+v2,32(r4) - vcmpequb. v6,v2,v18 /* Any zero bytes? */ - bne cr6,L(tail3) - - lxv 32+v3,48(r4) - vcmpequb. v6,v3,v18 /* Any zero bytes? */ - bne cr6,L(tail4) + CHECK16(v0,0,r5,tail1) + CHECK16(v1,16,r5,tail2) + CHECK16(v2,32,r5,tail3) + CHECK16(v3,48,r5,tail4) + CHECK16(v4,64,r5,tail5) + CHECK16(v5,80,r5,tail6) stxv 32+v0,0(r11) stxv 32+v1,16(r11) stxv 32+v2,32(r11) stxv 32+v3,48(r11) + stxv 32+v4,64(r11) + stxv 32+v5,80(r11) - addi r4,r4,64 - addi r11,r11,64 + addi r5,r5,96 + addi r11,r11,96 b L(loop) + .p2align 4 L(tail1): - vctzlsbb r8,v6 - addi r9,r8,1 + vctzlsbb r8,v6 /* Number of trailing zeroes */ + addi r9,r8,1 /* Add null terminator */ sldi r9,r9,56 /* stxvl wants size in top 8 bits */ - stxvl 32+v0,r11,r9 + stxvl 32+v0,r11,r9 /* Partial store */ #ifdef USE_AS_STPCPY /* stpcpy returns the dest address plus the size not counting the final '\0'. */ @@ -137,50 +124,81 @@ L(tail1): #endif blr + .p2align 4 L(tail2): stxv 32+v0,0(r11) - vctzlsbb r8,v6 /* Number of trailing zeroes */ - addi r9,r8,1 /* Add null terminator */ - sldi r10,r9,56 /* stxvl wants size in top 8 bits */ + vctzlsbb r8,v6 + addi r9,r8,1 + sldi r9,r9,56 addi r11,r11,16 - stxvl 32+v1,r11,r10 /* Partial store */ + stxvl 32+v1,r11,r9 #ifdef USE_AS_STPCPY - /* stpcpy returns the dest address plus the size not counting the - final '\0'. */ add r3,r11,r8 #endif blr + .p2align 4 L(tail3): stxv 32+v0,0(r11) stxv 32+v1,16(r11) - vctzlsbb r8,v6 /* Number of trailing zeroes */ - addi r9,r8,1 /* Add null terminator */ - sldi r10,r9,56 /* stxvl wants size in top 8 bits */ + vctzlsbb r8,v6 + addi r9,r8,1 + sldi r9,r9,56 addi r11,r11,32 - stxvl 32+v2,r11,r10 /* Partial store */ + stxvl 32+v2,r11,r9 #ifdef USE_AS_STPCPY - /* stpcpy returns the dest address plus the size not counting the - final '\0'. */ add r3,r11,r8 #endif blr + .p2align 4 L(tail4): stxv 32+v0,0(r11) stxv 32+v1,16(r11) stxv 32+v2,32(r11) - vctzlsbb r8,v6 /* Number of trailing zeroes */ - addi r9,r8,1 /* Add null terminator */ - sldi r10,r9,56 /* stxvl wants size in top 8 bits */ + vctzlsbb r8,v6 + addi r9,r8,1 + sldi r9,r9,56 addi r11,r11,48 - stxvl 32+v3,r11,r10 /* Partial store */ + stxvl 32+v3,r11,r9 #ifdef USE_AS_STPCPY - /* stpcpy returns the dest address plus the size not counting the - final '\0'. */ add r3,r11,r8 #endif blr + + .p2align 4 +L(tail5): + stxv 32+v0,0(r11) + stxv 32+v1,16(r11) + stxv 32+v2,32(r11) + stxv 32+v3,48(r11) + vctzlsbb r8,v6 + addi r9,r8,1 + sldi r9,r9,56 + addi r11,r11,64 + stxvl 32+v4,r11,r9 +#ifdef USE_AS_STPCPY + add r3,r11,r8 +#endif + blr + + .p2align 4 +L(tail6): + stxv 32+v0,0(r11) + stxv 32+v1,16(r11) + stxv 32+v2,32(r11) + stxv 32+v3,48(r11) + stxv 32+v4,64(r11) + vctzlsbb r8,v6 + addi r9,r8,1 + sldi r9,r9,56 + addi r11,r11,80 + stxvl 32+v5,r11,r9 +#ifdef USE_AS_STPCPY + add r3,r11,r8 +#endif + blr + END (FUNC_NAME) #ifndef USE_AS_STPCPY libc_hidden_builtin_def (strcpy)