From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from rock.gnat.com (rock.gnat.com [205.232.38.15]) by sourceware.org (Postfix) with ESMTPS id EE7423858D28; Wed, 8 Dec 2021 05:37:56 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org EE7423858D28 Received: from localhost (localhost.localdomain [127.0.0.1]) by filtered-rock.gnat.com (Postfix) with ESMTP id ACD611169B3; Wed, 8 Dec 2021 00:37:56 -0500 (EST) X-Virus-Scanned: Debian amavisd-new at gnat.com Received: from rock.gnat.com ([127.0.0.1]) by localhost (rock.gnat.com [127.0.0.1]) (amavisd-new, port 10024) with LMTP id 0eLUHv3s5Dhg; Wed, 8 Dec 2021 00:37:56 -0500 (EST) Received: from free.home (tron.gnat.com [IPv6:2620:20:4000:0:46a8:42ff:fe0e:e294]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by rock.gnat.com (Postfix) with ESMTPS id 47660116990; Wed, 8 Dec 2021 00:37:56 -0500 (EST) Received: from livre (livre.home [172.31.160.2]) by free.home (8.15.2/8.15.2) with ESMTPS id 1B85blLp916965 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=NOT); Wed, 8 Dec 2021 02:37:47 -0300 From: Alexandre Oliva To: gcc-patches@gcc.gnu.org Cc: zsojka@seznam.cz, wilson@gcc.gnu.org, vmakarov@redhat.com Subject: [PR103302] skip multi-word pre-move clobber during lra Organization: Free thinker, does not speak for AdaCore Errors-To: aoliva@lxoliva.fsfla.org Date: Wed, 08 Dec 2021 02:37:47 -0300 Message-ID: User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/25.2 (gnu/linux) MIME-Version: 1.0 Content-Type: text/plain X-Scanned-By: MIMEDefang 2.84 X-Spam-Status: No, score=-12.2 required=5.0 tests=BAYES_00, GIT_PATCH_0, KAM_DMARC_STATUS, KAM_SHORT, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.4 X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 08 Dec 2021 05:37:58 -0000 If we emit clobbers before multi-word moves during lra, we get confused if a copy ends up with input or output replaced with each other: the clobber then kills the previous set, and it gets deleted. This patch avoids emitting such clobbers when lra_in_progress. Regstrapped on x86_64-linux-gnu. Verified that, applied on a riscv64 compiler that failed the test, the asm statements are no longer dropped in the reload dumps. Running a x86_64-x-riscv64 regression testing now. Ok to install? for gcc/ChangeLog PR target/103302 expr.c (emit_move_multi_word): Skip clobber during lra. for gcc/testsuite/ChangeLog PR target/103302 * gcc.target/riscv/pr103302.c: New. --- gcc/expr.c | 2 + gcc/testsuite/gcc.target/riscv/pr103302.c | 47 +++++++++++++++++++++++++++++ 2 files changed, 48 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/gcc.target/riscv/pr103302.c diff --git a/gcc/expr.c b/gcc/expr.c index b281525750978..0365625e7b835 100644 --- a/gcc/expr.c +++ b/gcc/expr.c @@ -3929,7 +3929,7 @@ emit_move_multi_word (machine_mode mode, rtx x, rtx y) hard regs shouldn't appear here except as return values. We never want to emit such a clobber after reload. */ if (x != y - && ! (reload_in_progress || reload_completed) + && ! (lra_in_progress || reload_in_progress || reload_completed) && need_clobber != 0) emit_clobber (x); diff --git a/gcc/testsuite/gcc.target/riscv/pr103302.c b/gcc/testsuite/gcc.target/riscv/pr103302.c new file mode 100644 index 0000000000000..822c408741645 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/pr103302.c @@ -0,0 +1,47 @@ +/* { dg-do run } */ +/* { dg-options "-Og -fharden-compares -fno-tree-dce -fno-tree-fre " } */ + +typedef unsigned char u8; +typedef unsigned char __attribute__((__vector_size__ (32))) v256u8; +typedef unsigned short __attribute__((__vector_size__ (32))) v256u16; +typedef unsigned short __attribute__((__vector_size__ (64))) v512u16; +typedef unsigned int u32; +typedef unsigned int __attribute__((__vector_size__ (4))) v512u32; +typedef unsigned long long __attribute__((__vector_size__ (32))) v256u64; +typedef unsigned long long __attribute__((__vector_size__ (64))) v512u64; +typedef unsigned __int128 __attribute__((__vector_size__ (32))) v256u128; +typedef unsigned __int128 __attribute__((__vector_size__ (64))) v512u128; + +v512u16 g; + +void +foo0 (u8 u8_0, v256u16 v256u16_0, v512u16 v512u16_0, u32 u32_0, v512u32, + v256u64 v256u64_0, v512u64 v512u64_0, v256u128 v256u128_0, + v512u128 v512u128_0) +{ + u32_0 <= (v512u128) (v512u128_0 != u8_0); + v512u64 v512u64_1 = + __builtin_shufflevector (v256u64_0, v512u64_0, 7, 8, 0, 9, 5, 0, 3, 1); + g = v512u16_0; + (v256u8) v256u16_0 + (v256u8) v256u128_0; +} + +int +main (void) +{ + foo0 (40, (v256u16) + { + }, (v512u16) + { + }, 0, (v512u32) + { + }, (v256u64) + { + }, (v512u64) + { + }, (v256u128) + { + }, (v512u128) + { + }); +} -- Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/ Free Software Activist GNU Toolchain Engineer Disinformation flourishes because many people care deeply about injustice but very few check the facts. Ask me about