From patchwork Fri Dec 9 14:03:29 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Kyrill Tkachov X-Patchwork-Id: 704569 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]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 3tZv991wqjz9vDh for ; Sat, 10 Dec 2016 01:03:57 +1100 (AEDT) Authentication-Results: ozlabs.org; dkim=pass (1024-bit key; unprotected) header.d=gcc.gnu.org header.i=@gcc.gnu.org header.b="GFXMhrtG"; dkim-atps=neutral DomainKey-Signature: a=rsa-sha1; c=nofws; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender :message-id:date:from:mime-version:to:cc:subject:content-type; q=dns; s=default; b=wWKu9oU5KNrPgnXmANNzhgLSuW4ci7popFvt1mfyf9V NKK81SAtn7r16OfOAVKBXlxEgFMdXvQM4P8YfGCq2vCD8DltfFjRYewtqbSK1MP3 9LAEtMxPgZxoBw9u9N+UbJLqrahRNyjrnxZVlEWPXOvh6rWFDyPCWG2l2p5w5+K8 = DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=gcc.gnu.org; h=list-id :list-unsubscribe:list-archive:list-post:list-help:sender :message-id:date:from:mime-version:to:cc:subject:content-type; s=default; bh=6eiMbwsNu4lxJMg1oIZfKJEfO5Q=; b=GFXMhrtGHZC7Wgf9V QwL+ysZkm4xKX7JjkKZyab95WUDuArAMgPsfp6NHR8VYCAHZFtsMSvbPRd3zkYje 5LsXPucrULJ+Bic7nATNJ6VVx/KG1ZF799p4yVhf0kzidy/vyPzIyM9tWAamAL7w 8U0Uszirep9aMjzy9OIRFZZWO4= Received: (qmail 4699 invoked by alias); 9 Dec 2016 14:03:37 -0000 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 Received: (qmail 4669 invoked by uid 89); 9 Dec 2016 14:03:36 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-3.9 required=5.0 tests=BAYES_00, KAM_LAZY_DOMAIN_SECURITY, RP_MATCHES_RCVD autolearn=ham version=3.3.2 spammy=_Bool, _bool, 955 X-HELO: foss.arm.com Received: from foss.arm.com (HELO foss.arm.com) (217.140.101.70) by sourceware.org (qpsmtpd/0.93/v0.84-503-g423c35a) with ESMTP; Fri, 09 Dec 2016 14:03:33 +0000 Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.72.51.249]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id DBB9D1570; Fri, 9 Dec 2016 06:03:31 -0800 (PST) Received: from [10.2.207.77] (e100706-lin.cambridge.arm.com [10.2.207.77]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 3C8DF3F220; Fri, 9 Dec 2016 06:03:31 -0800 (PST) Message-ID: <584AB9B1.2070004@foss.arm.com> Date: Fri, 09 Dec 2016 14:03:29 +0000 From: Kyrill Tkachov User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.2.0 MIME-Version: 1.0 To: GCC Patches CC: Ramana Radhakrishnan , Richard Earnshaw Subject: [PATCH][ARM] PR target/78694: Avoid invalid RTL sharing in minipool code Hi all, In this ICE GCC reports invalid RTL sharing in the pattern: (insn 955 954 956 (unspec_volatile [ (const:SI (unspec:SI [ (symbol_ref:SI ("a") [flags 0xe8] ) (const_int 4 [0x4]) ] UNSPEC_TLS)) ] VUNSPEC_POOL_4) -1 (nil)) seb.c:116:1: error: shared rtx (const:SI (unspec:SI [ (symbol_ref:SI ("a") [flags 0xe8] ) (const_int 4 [0x4]) ] UNSPEC_TLS)) This is the consttable_4 pattern emitted by dump_minipool and the shared rtx that's causing the problem is the operand to that pattern that comes from the 'value' field of Mnode. I'm not very familiar with the exact restrictions on sharing RTL but I believe the right way to fix these is to do a copy_rtx of the value we want to use. So this patch does that when the pool entries are emitted. The ICE is quite fragile. See the PR for more details, but I've been able to reliably reproduce it on an arm-none-linux-gnueabihf target with -mcpu=arm7tdmi -mfloat-abi=soft -mfpu=vfp -mthumb and this patch fixes that there. Bootstrapped and tested on arm-none-linux-gnueabihf. Ok for trunk? Thanks, Kyrill 2016-12-09 Kyrylo Tkachov PR target/78694 * config/arm/arm.c (dump_minipool): Copy mp->value before emitting it in the minipool to avoid invalid RTL sharing. 2016-12-09 Kyrylo Tkachov PR target/78694 * gcc.c-torture/compile/pr78694.c: New test. commit 947b7660395c31e79605db30c4a1e306a4b050bf Author: Kyrylo Tkachov Date: Thu Dec 8 11:45:46 2016 +0000 [ARM] PR target/78694: Avoid invalid RTL sharing in minipool code diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c index 437da6f..6fcc23f 100644 --- a/gcc/config/arm/arm.c +++ b/gcc/config/arm/arm.c @@ -15959,35 +15959,37 @@ dump_minipool (rtx_insn *scan) fputc ('\n', dump_file); } + rtx val = copy_rtx (mp->value); + switch (GET_MODE_SIZE (mp->mode)) { #ifdef HAVE_consttable_1 case 1: - scan = emit_insn_after (gen_consttable_1 (mp->value), scan); + scan = emit_insn_after (gen_consttable_1 (val), scan); break; #endif #ifdef HAVE_consttable_2 case 2: - scan = emit_insn_after (gen_consttable_2 (mp->value), scan); + scan = emit_insn_after (gen_consttable_2 (val), scan); break; #endif #ifdef HAVE_consttable_4 case 4: - scan = emit_insn_after (gen_consttable_4 (mp->value), scan); + scan = emit_insn_after (gen_consttable_4 (val), scan); break; #endif #ifdef HAVE_consttable_8 case 8: - scan = emit_insn_after (gen_consttable_8 (mp->value), scan); + scan = emit_insn_after (gen_consttable_8 (val), scan); break; #endif #ifdef HAVE_consttable_16 case 16: - scan = emit_insn_after (gen_consttable_16 (mp->value), scan); + scan = emit_insn_after (gen_consttable_16 (val), scan); break; #endif diff --git a/gcc/testsuite/gcc.c-torture/compile/pr78694.c b/gcc/testsuite/gcc.c-torture/compile/pr78694.c new file mode 100644 index 0000000..bc31944 --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/compile/pr78694.c @@ -0,0 +1,118 @@ +/* PR target/78694. */ + +enum +{ + MEMMODEL_RELAXED, + MEMMODEL_ACQUIRE, + PRIORITY_INSERT_END +}; +enum +{ + PQ_CHILDREN, + PQ_TASKGROUP +}; +struct gomp_team_state +{ + struct gomp_team *team; +}; +enum gomp_task_kind +{ + GOMP_TASK_UNDEFERRED, + GOMP_TASK_WAITING +}; +struct gomp_taskwait +{ + _Bool in_taskwait; +}; +struct gomp_task +{ + struct gomp_task *parent; + int children_queue; + struct gomp_taskgroup *taskgroup; + int dependers; + struct gomp_taskwait taskwait; + enum gomp_task_kind kind; + _Bool in_tied_task; +} j, q, *n; +struct gomp_taskgroup +{ + _Bool in_taskgroup_wait; + int num_children; +} l; +struct gomp_team +{ + int task_queue; + int task_running_count; +}; +struct gomp_thread +{ + struct gomp_team_state ts; + struct gomp_task task; +} extern __thread a; + +int b, c, d, e, f, g, h, i, k, m, o, p, r; + +void priority_queue_next_task (struct gomp_task *, int, int); +int gomp_task_run_pre (struct gomp_task *, struct gomp_task, struct gomp_team); +void priority_queue_insert (int, struct gomp_task); +void priority_queue_insert2 (int, struct gomp_task, int, int, int); +void priority_queue_insert3 (int, struct gomp_task, int, int, int); +void gomp_sem_post (int); +void free (void *); + +_Bool s; +int +GOMP_taskgroup_end () +{ + struct gomp_thread *t = &a; + struct gomp_team u = *t->ts.team; + struct gomp_task *v = &t->task, *w; + if (__atomic_load_n (&l.num_children, MEMMODEL_ACQUIRE)) + while (1) + { + if (l.num_children) + priority_queue_next_task (v, u.task_queue, r); + else if (w) + free (w); + if (n->kind == GOMP_TASK_WAITING) + { + s = gomp_task_run_pre (n, q, u); + if (__builtin_expect (s, 0)) + { + if (w) + free (w); + goto finish_cancelled; + } + n = 0; + l.in_taskgroup_wait = 1; + } + if (w) + { + t->task = *n; + if (__builtin_expect (p, 0)) + if (o) + t->task = *v; + } + if (n) + { + struct gomp_task x = x; + for (; i; b++) + { + struct gomp_task y = j; + if (g) + continue; + priority_queue_insert (PQ_CHILDREN, x); + if (x.taskwait.in_taskwait) + priority_queue_insert2 (PQ_TASKGROUP, y, e, 0, d); + if (h) + gomp_sem_post (f); + priority_queue_insert3 (k, y, PRIORITY_INSERT_END, 0, d); + ++c; + } + } + finish_cancelled: + w = (struct gomp_task *) (n - u.task_running_count - v); + } + v->taskgroup = (struct gomp_taskgroup *) m; + return 1; +}