From patchwork Wed Jun 15 08:45:15 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Revital Eres X-Patchwork-Id: 100589 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) by ozlabs.org (Postfix) with SMTP id 2F199B6F99 for ; Thu, 16 Jun 2011 16:00:28 +1000 (EST) Received: (qmail 14872 invoked by alias); 16 Jun 2011 06:00:25 -0000 Received: (qmail 14859 invoked by uid 22791); 16 Jun 2011 06:00:23 -0000 X-SWARE-Spam-Status: No, hits=-1.9 required=5.0 tests=AWL, BAYES_00, RCVD_IN_DNSWL_LOW, TW_DD X-Spam-Check-By: sourceware.org Received: from mail-ew0-f47.google.com (HELO mail-ew0-f47.google.com) (209.85.215.47) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Thu, 16 Jun 2011 06:00:06 +0000 Received: by ewy5 with SMTP id 5so487560ewy.20 for ; Wed, 15 Jun 2011 23:00:04 -0700 (PDT) MIME-Version: 1.0 Received: by 10.213.2.202 with SMTP id 10mr936573ebk.110.1308127515689; Wed, 15 Jun 2011 01:45:15 -0700 (PDT) Received: by 10.213.27.203 with HTTP; Wed, 15 Jun 2011 01:45:15 -0700 (PDT) In-Reply-To: References: Date: Wed, 15 Jun 2011 11:45:15 +0300 Message-ID: Subject: Re: [PATCH, SMS] Fix violation of memory dependence From: Revital Eres To: Ayal Zaks Cc: gcc-patches@gcc.gnu.org, Patch Tracking Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org Hello, >>>         * ddg.c (add_intra_loop_mem_dep): New function. > > You could check first thing if (from->cuid == to->cuid), for code clarity. Attached is the new version of the patch which addresses this. The patch was re-tested as follows: On ppc64-redhat-linux regtest as well as bootstrap with SMS flags enabling SMS also on loops with stage count 1. Regtested on SPU. On arm-linux-gnueabi bootstrap c language with SMS flags enabling SMS also on loops with stage count 1 and currently regression testing on c,c++. OK for mainline once regtest on arm-linux-gnueabi completes? Thanks, Revital gcc/ * ddg.c (add_intra_loop_mem_dep): New function. (build_intra_loop_deps): Call it. testsuite/ * gcc.dg/sms-9.c: New file. Index: ddg.c =================================================================== --- ddg.c (revision 174906) +++ ddg.c (working copy) @@ -390,6 +390,38 @@ insns_may_alias_p (rtx insn1, rtx insn2) &PATTERN (insn2)); } +/* Given two nodes, analyze their RTL insns and add intra-loop mem deps + to ddg G. */ +static void +add_intra_loop_mem_dep (ddg_ptr g, ddg_node_ptr from, ddg_node_ptr to) +{ + + if ((from->cuid == to->cuid) + || !insns_may_alias_p (from->insn, to->insn)) + /* Do not create edge if memory references have disjoint alias sets + or 'to' and 'from' are the same instruction. */ + return; + + if (mem_write_insn_p (from->insn)) + { + if (mem_read_insn_p (to->insn)) + create_ddg_dep_no_link (g, from, to, + DEBUG_INSN_P (to->insn) + ? ANTI_DEP : TRUE_DEP, MEM_DEP, 0); + else + create_ddg_dep_no_link (g, from, to, + DEBUG_INSN_P (to->insn) + ? ANTI_DEP : OUTPUT_DEP, MEM_DEP, 0); + } + else + { + if (mem_read_insn_p (to->insn)) + return; + else + create_ddg_dep_no_link (g, from, to, ANTI_DEP, MEM_DEP, 0); + } +} + /* Given two nodes, analyze their RTL insns and add inter-loop mem deps to ddg G. */ static void @@ -477,10 +509,22 @@ build_intra_loop_deps (ddg_ptr g) if (DEBUG_INSN_P (j_node->insn)) continue; if (mem_access_insn_p (j_node->insn)) - /* Don't bother calculating inter-loop dep if an intra-loop dep - already exists. */ + { + /* Don't bother calculating inter-loop dep if an intra-loop dep + already exists. */ if (! TEST_BIT (dest_node->successors, j)) add_inter_loop_mem_dep (g, dest_node, j_node); + /* If -fmodulo-sched-allow-regmoves + is set certain anti-dep edges are not created. + It might be that these anti-dep edges are on the + path from one memory instruction to another such that + removing these edges could cause a violation of the + memory dependencies. Thus we add intra edges between + every two memory instructions in this case. */ + if (flag_modulo_sched_allow_regmoves + && !TEST_BIT (dest_node->predecessors, j)) + add_intra_loop_mem_dep (g, j_node, dest_node); + } } } } Index: testsuite/gcc.dg/sms-9.c =================================================================== --- testsuite/gcc.dg/sms-9.c (revision 0) +++ testsuite/gcc.dg/sms-9.c (revision 0) @@ -0,0 +1,60 @@ +/* { dg-do run } */ +/* { dg-options "-O2 -fmodulo-sched -fno-auto-inc-dec -O2 -fmodulo-sched-allow-regmoves" } */ + +#include +#include + +struct df_ref_info +{ + unsigned int *begin; + unsigned int *count; +}; + +extern void *memset (void *s, int c, __SIZE_TYPE__ n); + + +__attribute__ ((noinline)) +int +df_reorganize_refs_by_reg_by_insn (struct df_ref_info *ref_info, + int num, unsigned int start) +{ + unsigned int m = num; + unsigned int offset = 77; + unsigned int r; + + for (r = start; r < m; r++) + { + ref_info->begin[r] = offset; + offset += ref_info->count[r]; + ref_info->count[r] = 0; + } + + return offset; +} + +int +main () +{ + struct df_ref_info temp; + int num = 100; + unsigned int start = 5; + int i, offset; + + temp.begin = malloc (100 * sizeof (unsigned int)); + temp.count = malloc (100 * sizeof (unsigned int)); + + memset (temp.begin, 0, sizeof (unsigned int) * num); + memset (temp.count, 0, sizeof (unsigned int) * num); + + for (i = 0; i < num; i++) + temp.count[i] = i + 1; + + offset = df_reorganize_refs_by_reg_by_insn (&temp, num, start); + + if (offset != 5112) + abort (); + + free (temp.begin); + free (temp.count); + return 0; +}