From patchwork Tue Nov 7 23:34:38 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Khem Raj X-Patchwork-Id: 118209 Delivered-To: patch@linaro.org Received: by 10.140.22.164 with SMTP id 33csp4592505qgn; Tue, 7 Nov 2017 15:34:51 -0800 (PST) X-Google-Smtp-Source: ABhQp+SMVTQsKpEKPHYKbaRk1XsVRY9V0rZXwZZN+OtaTrmXbe3rXPNCpB4eyMKZYxA13NZRZEuj X-Received: by 10.159.207.129 with SMTP id z1mr369873plo.174.1510097690938; Tue, 07 Nov 2017 15:34:50 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1510097690; cv=none; d=google.com; s=arc-20160816; b=jgQAhncRsDMUyNMqzk4DSt5xKtNqNDsiY8iNTxp44SB7h9pIXXO/KcknnkR6Xd9GKd rlU0AXdLJRzrRdBSoj7D302l7u2nscwh3aeEjEtA4wjGY0z3My4I6f26xZZJTkE3A9cC OoJIftex6k9tBYEK+2ass5BWtIU9CeKTu+RS2/zD5zC9NMDqZkYktONea/xjLNZlqoEG hRY8nwHFAybTatdjnFIbsUSdHTMyY2+vJeXgKtJEVRDJrIHms8954xuvx91B2FdvAdFR +bjqUx2lEjGM3zXS36XwOZpXka2iNFeHg2CLETIMq0vU94EaWCNc/A7lpRG+BuGYTlLZ MFpA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=errors-to:sender:content-transfer-encoding:mime-version :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:message-id:date:to:from:dkim-signature :delivered-to:arc-authentication-results; bh=Zyyklq4ZLgUzJF46hIDHhlAulqIH40558smoq6q7Gzw=; b=y+ryJDOoqcLApWKvYHASvsk0NMNwlv/n/yblMgX0PjOJUqL/ivEil5pFKt9BbfEc8l DTNZLjoxwzVrjumkB7WsfIqEdlAkzr9hP0DWvVKj7HvHjbVOdaM/ojKPbdhAGg0RN84h bS2aFb6s2iElNouki0ThfXXG/z0AVGKjRN5tBfE5Dkt1zRfoobQwtmDGFON4jeAkkczD jlns8geU6sEKCKV2owG8Clca/GWhAr+n3gMvCGIhJlMt5dDf9pAYsUShY2YDLpYb4Tkp tzT9oxEmmrqy7E6kHUvjFLFO+A93jINhH+RyOWsrpVRPneWgLGjM7gIYa6aal8W14Wtv 4iqA== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.s=20161025 header.b=i6Duv+hv; spf=pass (google.com: best guess record for domain of openembedded-core-bounces@lists.openembedded.org designates 140.211.169.62 as permitted sender) smtp.mailfrom=openembedded-core-bounces@lists.openembedded.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Return-Path: Received: from mail.openembedded.org (mail.openembedded.org. [140.211.169.62]) by mx.google.com with ESMTP id i12si2339655pgn.69.2017.11.07.15.34.50; Tue, 07 Nov 2017 15:34:50 -0800 (PST) Received-SPF: pass (google.com: best guess record for domain of openembedded-core-bounces@lists.openembedded.org designates 140.211.169.62 as permitted sender) client-ip=140.211.169.62; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com header.s=20161025 header.b=i6Duv+hv; spf=pass (google.com: best guess record for domain of openembedded-core-bounces@lists.openembedded.org designates 140.211.169.62 as permitted sender) smtp.mailfrom=openembedded-core-bounces@lists.openembedded.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Received: from review.yoctoproject.org (localhost [127.0.0.1]) by mail.openembedded.org (Postfix) with ESMTP id 380847848F; Tue, 7 Nov 2017 23:34:48 +0000 (UTC) X-Original-To: openembedded-core@lists.openembedded.org Delivered-To: openembedded-core@lists.openembedded.org Received: from mail-pl0-f44.google.com (mail-pl0-f44.google.com [209.85.160.44]) by mail.openembedded.org (Postfix) with ESMTP id 7B70E7822A for ; Tue, 7 Nov 2017 23:34:46 +0000 (UTC) Received: by mail-pl0-f44.google.com with SMTP id y61so61455plh.3 for ; Tue, 07 Nov 2017 15:34:48 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id; bh=LTt12y/ccPNJkcu10FCAcpVKsj2rhK3BVvLHvRgY83o=; b=i6Duv+hvczN75sQXE6UsmxaV8p0oRUCXiBQ/Z50JCPPE6XO/3/D/q9WNKIUxVl8Awz H6rEg9cHZWPcFQNOlk2n397LhwrBpwJYD5/TCPIPeosv2q7sa3zKzxC6GE939bWYoXmm uUfkNgADIHV7qARBjMwuR2LR5pyYzxFZfljxDhXx7p75GBpCtV/xUVEJcp39K/rPzGls dpUtPV3bPTww1IjNNSHAH+skTsKpdmcarfqHG0kL1WERqKNSSfy+i040Bl+Ve/mSdwLZ CORGx81BBE+nUgftiCWYcdJ6F7RaleFJUglCcrqw0cRJnPvKwJEYqmJOHYg5OWDgJzqr Mksg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id; bh=LTt12y/ccPNJkcu10FCAcpVKsj2rhK3BVvLHvRgY83o=; b=NYe+GktGVf2nGMak/SnEwsybGtLwegEHNn1jWOaFO9zfpJn/CX+65iLEjFUOOMRPvQ ontrY2InunhS55a0iIJcSSqXCdrAhPSVMaSpsG2dbsKgPp94H0gjDfZMDu2JQQOkEoLb hWeIYhaS5R07552uLzdG5xnLBI4NnrX4vejvBj5CQPpQjEqtMjs0ezUrW3WDjHqGMw68 5Zb9tKn/ROGuegsrqH97UOy0VfEtWD0rkf7YsE5yc3CjirMatYeCpnVcf8nOLO2Tm3pg aA7W29N7r4BimH6JnQnhuLQSoh48kYtLni7BUuuym43VH+GXoz4W/WlI3iVkgkE6k1Ts xxGQ== X-Gm-Message-State: AJaThX4d36ITsk3055wVlhV1ru1PFPurdCcUWWht0UMLr67mWi0F7UIE 6/nd6oJ+K0CldEhcWJR2A4iMzA== X-Received: by 10.159.252.7 with SMTP id n7mr332824pls.355.1510097687198; Tue, 07 Nov 2017 15:34:47 -0800 (PST) Received: from localhost.localdomain ([2601:646:8882:b8c::64bd]) by smtp.gmail.com with ESMTPSA id s87sm5322995pfi.23.2017.11.07.15.34.46 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Tue, 07 Nov 2017 15:34:46 -0800 (PST) From: Khem Raj To: openembedded-core@lists.openembedded.org Date: Tue, 7 Nov 2017 15:34:38 -0800 Message-Id: <20171107233438.33636-1-raj.khem@gmail.com> X-Mailer: git-send-email 2.15.0 Subject: [OE-core] [PATCH] gcc7/gcc6: Fix unaligned STRD issue on ARM X-BeenThere: openembedded-core@lists.openembedded.org X-Mailman-Version: 2.1.12 Precedence: list List-Id: Patches and discussions about the oe-core layer List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , MIME-Version: 1.0 Sender: openembedded-core-bounces@lists.openembedded.org Errors-To: openembedded-core-bounces@lists.openembedded.org Backport https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82445 Fixes [YOCTO 12297] Signed-off-by: Khem Raj --- meta/recipes-devtools/gcc/gcc-6.4.inc | 1 + ...5-suppress-32-bit-aligned-ldrd-strd-peeph.patch | 194 +++++++++++++++++++++ meta/recipes-devtools/gcc/gcc-7.2.inc | 1 + ...5-suppress-32-bit-aligned-ldrd-strd-peeph.patch | 194 +++++++++++++++++++++ 4 files changed, 390 insertions(+) create mode 100644 meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch create mode 100644 meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch -- 2.15.0 -- _______________________________________________ Openembedded-core mailing list Openembedded-core@lists.openembedded.org http://lists.openembedded.org/mailman/listinfo/openembedded-core diff --git a/meta/recipes-devtools/gcc/gcc-6.4.inc b/meta/recipes-devtools/gcc/gcc-6.4.inc index a42b7d83ca..37e996afb7 100644 --- a/meta/recipes-devtools/gcc/gcc-6.4.inc +++ b/meta/recipes-devtools/gcc/gcc-6.4.inc @@ -80,6 +80,7 @@ SRC_URI = "\ " BACKPORTS = "\ file://CVE-2016-6131.patch \ + file://0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch \ " SRC_URI[md5sum] = "11ba51a0cfb8471927f387c8895fe232" SRC_URI[sha256sum] = "850bf21eafdfe5cd5f6827148184c08c4a0852a37ccf36ce69855334d2c914d4" diff --git a/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch b/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch new file mode 100644 index 0000000000..0214ab83d9 --- /dev/null +++ b/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch @@ -0,0 +1,194 @@ +From ad5bf450aef2ffee6d57ed193fabc5f72f8eaa65 Mon Sep 17 00:00:00 2001 +From: rearnsha +Date: Thu, 19 Oct 2017 13:16:42 +0000 +Subject: [PATCH] [ARM] PR 82445 - suppress 32-bit aligned ldrd/strd peepholing + with -mno-unaligned-access + +Peephole patterns exist in the arm backend to spot load/store +operations to adjacent memory operations in order to convert them into +ldrd/strd instructions. However, when we have strict alignment +enforced, then we can only do this if the accesses are known to be +64-bit aligned; this is unlikely to be the case for most loads. The +patch adds some alignment checking to the code that validates the +addresses for use in the peephole patterns. This should also fix +incorrect generation of ldrd/strd with unaligned accesses that could +previously have occurred on ARMv5e where all such operations must be +64-bit aligned. + +I've added some new tests as well. In doing so I discovered that the +ldrd/strd peephole tests could never fail since they would match the +source file name in the scanned assembly as well as any instructions +of the intended type. I've fixed those by tightening the scan results +slightly. + +gcc: + +* config/arm/arm.c (align_ok_ldrd_strd): New function. +(mem_ok_for_ldrd_strd): New parameter align. Extract the alignment of the +mem into it. +(gen_operands_ldrd_strd): Validate the alignment of the accesses. + +testsuite: + +* gcc.target/arm/peep-ldrd-1.c: Tighten test scan pattern. +* gcc.target/arm/peep-strd-1.c: Likewise. +* gcc.target/arm/peep-ldrd-2.c: New test. +* gcc.target/arm/peep-strd-2.c: New test. + + + +git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-6-branch@253892 138bc75d-0d04-0410-961f-82ee72b054a4 +--- +Upstream-Status: Backport +Signed-off-by: Khem Raj + + gcc/ChangeLog | 8 +++++++ + gcc/config/arm/arm.c | 27 ++++++++++++++++++---- + gcc/testsuite/ChangeLog | 8 +++++++ + gcc/testsuite/gcc.target/arm/peep-ldrd-1.c | 2 +- + .../arm/{peep-ldrd-1.c => peep-ldrd-2.c} | 4 ++-- + gcc/testsuite/gcc.target/arm/peep-strd-1.c | 2 +- + .../arm/{peep-strd-1.c => peep-strd-2.c} | 4 ++-- + 7 files changed, 44 insertions(+), 11 deletions(-) + copy gcc/testsuite/gcc.target/arm/{peep-ldrd-1.c => peep-ldrd-2.c} (63%) + copy gcc/testsuite/gcc.target/arm/{peep-strd-1.c => peep-strd-2.c} (58%) + +diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c +index 9c0813d598d..e3da9f77fb6 100644 +--- a/gcc/config/arm/arm.c ++++ b/gcc/config/arm/arm.c +@@ -15926,12 +15926,23 @@ operands_ok_ldrd_strd (rtx rt, rtx rt2, rtx rn, HOST_WIDE_INT offset, + return true; + } + ++/* Return true if a 64-bit access with alignment ALIGN and with a ++ constant offset OFFSET from the base pointer is permitted on this ++ architecture. */ ++static bool ++align_ok_ldrd_strd (HOST_WIDE_INT align, HOST_WIDE_INT offset) ++{ ++ return (unaligned_access ++ ? (align >= BITS_PER_WORD && (offset & 3) == 0) ++ : (align >= 2 * BITS_PER_WORD && (offset & 7) == 0)); ++} ++ + /* Helper for gen_operands_ldrd_strd. Returns true iff the memory + operand MEM's address contains an immediate offset from the base +- register and has no side effects, in which case it sets BASE and +- OFFSET accordingly. */ ++ register and has no side effects, in which case it sets BASE, ++ OFFSET and ALIGN accordingly. */ + static bool +-mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset) ++mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset, HOST_WIDE_INT *align) + { + rtx addr; + +@@ -15950,6 +15961,7 @@ mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset) + gcc_assert (MEM_P (mem)); + + *offset = const0_rtx; ++ *align = MEM_ALIGN (mem); + + addr = XEXP (mem, 0); + +@@ -15990,7 +16002,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + bool const_store, bool commute) + { + int nops = 2; +- HOST_WIDE_INT offsets[2], offset; ++ HOST_WIDE_INT offsets[2], offset, align[2]; + rtx base = NULL_RTX; + rtx cur_base, cur_offset, tmp; + int i, gap; +@@ -16002,7 +16014,8 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + registers, and the corresponding memory offsets. */ + for (i = 0; i < nops; i++) + { +- if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset)) ++ if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset, ++ &align[i])) + return false; + + if (i == 0) +@@ -16114,6 +16127,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + /* Swap the instructions such that lower memory is accessed first. */ + std::swap (operands[0], operands[1]); + std::swap (operands[2], operands[3]); ++ std::swap (align[0], align[1]); + if (const_store) + std::swap (operands[4], operands[5]); + } +@@ -16127,6 +16141,9 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + if (gap != 4) + return false; + ++ if (!align_ok_ldrd_strd (align[0], offset)) ++ return false; ++ + /* Make sure we generate legal instructions. */ + if (operands_ok_ldrd_strd (operands[0], operands[1], base, offset, + false, load)) +diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +index eb2b86ee7b6..d49eff6b87e 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q) + *p = a; + return a; + } +-/* { dg-final { scan-assembler "ldrd" } } */ ++/* { dg-final { scan-assembler "ldrd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +similarity index 63% +copy from gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +copy to gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +index eb2b86ee7b6..6822c2b1454 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +@@ -1,6 +1,6 @@ + /* { dg-do compile } */ + /* { dg-require-effective-target arm_prefer_ldrd_strd } */ +-/* { dg-options "-O2" } */ ++/* { dg-options "-O2 -mno-unaligned-access" } */ + int foo(int a, int b, int* p, int *q) + { + a = p[2] + p[3]; +@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q) + *p = a; + return a; + } +-/* { dg-final { scan-assembler "ldrd" } } */ ++/* { dg-final { scan-assembler-not "ldrd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-1.c +index bd330769599..fe1beac7229 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-strd-1.c +@@ -6,4 +6,4 @@ void foo(int a, int b, int* p) + p[2] = a; + p[3] = b; + } +-/* { dg-final { scan-assembler "strd" } } */ ++/* { dg-final { scan-assembler "strd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-2.c +similarity index 58% +copy from gcc/testsuite/gcc.target/arm/peep-strd-1.c +copy to gcc/testsuite/gcc.target/arm/peep-strd-2.c +index bd330769599..bfc5ebe9eec 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-strd-2.c +@@ -1,9 +1,9 @@ + /* { dg-do compile } */ + /* { dg-require-effective-target arm_prefer_ldrd_strd } */ +-/* { dg-options "-O2" } */ ++/* { dg-options "-O2 -mno-unaligned-access" } */ + void foo(int a, int b, int* p) + { + p[2] = a; + p[3] = b; + } +-/* { dg-final { scan-assembler "strd" } } */ ++/* { dg-final { scan-assembler-not "strd\\t" } } */ +-- +2.15.0 + diff --git a/meta/recipes-devtools/gcc/gcc-7.2.inc b/meta/recipes-devtools/gcc/gcc-7.2.inc index 04edad77cb..02b8e03482 100644 --- a/meta/recipes-devtools/gcc/gcc-7.2.inc +++ b/meta/recipes-devtools/gcc/gcc-7.2.inc @@ -79,6 +79,7 @@ SRC_URI = "\ ${BACKPORTS} \ " BACKPORTS = "\ + file://0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch \ " SRC_URI[md5sum] = "ff370482573133a7fcdd96cd2f552292" SRC_URI[sha256sum] = "1cf7adf8ff4b5aa49041c8734bbcf1ad18cc4c94d0029aae0f4e48841088479a" diff --git a/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch b/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch new file mode 100644 index 0000000000..021250700d --- /dev/null +++ b/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch @@ -0,0 +1,194 @@ +From 8c18b422211878ba02503462cb22a2cc25a0a325 Mon Sep 17 00:00:00 2001 +From: rearnsha +Date: Thu, 19 Oct 2017 13:14:55 +0000 +Subject: [PATCH] [ARM] PR 82445 - suppress 32-bit aligned ldrd/strd peepholing + with -mno-unaligned-access + +Peephole patterns exist in the arm backend to spot load/store +operations to adjacent memory operations in order to convert them into +ldrd/strd instructions. However, when we have strict alignment +enforced, then we can only do this if the accesses are known to be +64-bit aligned; this is unlikely to be the case for most loads. The +patch adds some alignment checking to the code that validates the +addresses for use in the peephole patterns. This should also fix +incorrect generation of ldrd/strd with unaligned accesses that could +previously have occurred on ARMv5e where all such operations must be +64-bit aligned. + +I've added some new tests as well. In doing so I discovered that the +ldrd/strd peephole tests could never fail since they would match the +source file name in the scanned assembly as well as any instructions +of the intended type. I've fixed those by tightening the scan results +slightly. + +gcc: + +* config/arm/arm.c (align_ok_ldrd_strd): New function. +(mem_ok_for_ldrd_strd): New parameter align. Extract the alignment of the +mem into it. +(gen_operands_ldrd_strd): Validate the alignment of the accesses. + +testsuite: + +* gcc.target/arm/peep-ldrd-1.c: Tighten test scan pattern. +* gcc.target/arm/peep-strd-1.c: Likewise. +* gcc.target/arm/peep-ldrd-2.c: New test. +* gcc.target/arm/peep-strd-2.c: New test. + + + +git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-7-branch@253891 138bc75d-0d04-0410-961f-82ee72b054a4 +--- +Upstream-Status: Backport +Signed-off-by: Khem Raj + + gcc/ChangeLog | 8 +++++++ + gcc/config/arm/arm.c | 27 ++++++++++++++++++---- + gcc/testsuite/ChangeLog | 8 +++++++ + gcc/testsuite/gcc.target/arm/peep-ldrd-1.c | 2 +- + .../arm/{peep-ldrd-1.c => peep-ldrd-2.c} | 4 ++-- + gcc/testsuite/gcc.target/arm/peep-strd-1.c | 2 +- + .../arm/{peep-strd-1.c => peep-strd-2.c} | 4 ++-- + 7 files changed, 44 insertions(+), 11 deletions(-) + copy gcc/testsuite/gcc.target/arm/{peep-ldrd-1.c => peep-ldrd-2.c} (63%) + copy gcc/testsuite/gcc.target/arm/{peep-strd-1.c => peep-strd-2.c} (58%) + +diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c +index 1ded0d2a17d..989957f048e 100644 +--- a/gcc/config/arm/arm.c ++++ b/gcc/config/arm/arm.c +@@ -15199,12 +15199,23 @@ operands_ok_ldrd_strd (rtx rt, rtx rt2, rtx rn, HOST_WIDE_INT offset, + return true; + } + ++/* Return true if a 64-bit access with alignment ALIGN and with a ++ constant offset OFFSET from the base pointer is permitted on this ++ architecture. */ ++static bool ++align_ok_ldrd_strd (HOST_WIDE_INT align, HOST_WIDE_INT offset) ++{ ++ return (unaligned_access ++ ? (align >= BITS_PER_WORD && (offset & 3) == 0) ++ : (align >= 2 * BITS_PER_WORD && (offset & 7) == 0)); ++} ++ + /* Helper for gen_operands_ldrd_strd. Returns true iff the memory + operand MEM's address contains an immediate offset from the base +- register and has no side effects, in which case it sets BASE and +- OFFSET accordingly. */ ++ register and has no side effects, in which case it sets BASE, ++ OFFSET and ALIGN accordingly. */ + static bool +-mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset) ++mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset, HOST_WIDE_INT *align) + { + rtx addr; + +@@ -15223,6 +15234,7 @@ mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset) + gcc_assert (MEM_P (mem)); + + *offset = const0_rtx; ++ *align = MEM_ALIGN (mem); + + addr = XEXP (mem, 0); + +@@ -15263,7 +15275,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + bool const_store, bool commute) + { + int nops = 2; +- HOST_WIDE_INT offsets[2], offset; ++ HOST_WIDE_INT offsets[2], offset, align[2]; + rtx base = NULL_RTX; + rtx cur_base, cur_offset, tmp; + int i, gap; +@@ -15275,7 +15287,8 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + registers, and the corresponding memory offsets. */ + for (i = 0; i < nops; i++) + { +- if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset)) ++ if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset, ++ &align[i])) + return false; + + if (i == 0) +@@ -15389,6 +15402,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + /* Swap the instructions such that lower memory is accessed first. */ + std::swap (operands[0], operands[1]); + std::swap (operands[2], operands[3]); ++ std::swap (align[0], align[1]); + if (const_store) + std::swap (operands[4], operands[5]); + } +@@ -15402,6 +15416,9 @@ gen_operands_ldrd_strd (rtx *operands, bool load, + if (gap != 4) + return false; + ++ if (!align_ok_ldrd_strd (align[0], offset)) ++ return false; ++ + /* Make sure we generate legal instructions. */ + if (operands_ok_ldrd_strd (operands[0], operands[1], base, offset, + false, load)) +diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +index eb2b86ee7b6..d49eff6b87e 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q) + *p = a; + return a; + } +-/* { dg-final { scan-assembler "ldrd" } } */ ++/* { dg-final { scan-assembler "ldrd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +similarity index 63% +copy from gcc/testsuite/gcc.target/arm/peep-ldrd-1.c +copy to gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +index eb2b86ee7b6..6822c2b1454 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c +@@ -1,6 +1,6 @@ + /* { dg-do compile } */ + /* { dg-require-effective-target arm_prefer_ldrd_strd } */ +-/* { dg-options "-O2" } */ ++/* { dg-options "-O2 -mno-unaligned-access" } */ + int foo(int a, int b, int* p, int *q) + { + a = p[2] + p[3]; +@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q) + *p = a; + return a; + } +-/* { dg-final { scan-assembler "ldrd" } } */ ++/* { dg-final { scan-assembler-not "ldrd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-1.c +index bd330769599..fe1beac7229 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-strd-1.c +@@ -6,4 +6,4 @@ void foo(int a, int b, int* p) + p[2] = a; + p[3] = b; + } +-/* { dg-final { scan-assembler "strd" } } */ ++/* { dg-final { scan-assembler "strd\\t" } } */ +diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-2.c +similarity index 58% +copy from gcc/testsuite/gcc.target/arm/peep-strd-1.c +copy to gcc/testsuite/gcc.target/arm/peep-strd-2.c +index bd330769599..bfc5ebe9eec 100644 +--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c ++++ b/gcc/testsuite/gcc.target/arm/peep-strd-2.c +@@ -1,9 +1,9 @@ + /* { dg-do compile } */ + /* { dg-require-effective-target arm_prefer_ldrd_strd } */ +-/* { dg-options "-O2" } */ ++/* { dg-options "-O2 -mno-unaligned-access" } */ + void foo(int a, int b, int* p) + { + p[2] = a; + p[3] = b; + } +-/* { dg-final { scan-assembler "strd" } } */ ++/* { dg-final { scan-assembler-not "strd\\t" } } */ +-- +2.15.0 +