From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wr1-x432.google.com (mail-wr1-x432.google.com [IPv6:2a00:1450:4864:20::432]) by sourceware.org (Postfix) with ESMTPS id DA973394D8BE for ; Thu, 7 Jan 2021 08:50:03 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org DA973394D8BE Received: by mail-wr1-x432.google.com with SMTP id r3so4825830wrt.2 for ; Thu, 07 Jan 2021 00:50:03 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc:content-transfer-encoding; bh=qcvLbSoMy6cV3tS2GNlPfc8I8aXIdl1r8BT0IjMr5vg=; b=HDedj8GYcf9M9HpAe9Qkvqpu+y5N4CaEQ7ECXuP3jOS2NDtQf6r7t5n7vTXIQ7pTLh 5f0/7cBKhUJi+er4zNk8dEAAgioMTYsQyoGZD/8+vp7o3D7GmVWNs+HAuUapuESI6H/b ecZFesaGmVZCroi0lhRhEsJpaXI/qXeg3nqKhc2/g3ERfTD8ua1RkVvBsG1jeZvFG0F1 KTs9u4Lw0rMoRmxQYWgFYf0A0XvFlesfdOWy3/xhmqljXYDvBUvrG2w//G4fK/dKijv3 jDvFNGrJlDVaCxtjyjtkrA8lGNCYbLdzvjd2YTNtIXgKRiH7YLx8cppSM0wd5tjSNABl lUMA== X-Gm-Message-State: AOAM533Pml65CbmUviwqYK3kW5x2hVcweXwP4BOJ1n1aSNNdssO3hIeu avY82xZ/9xiINuCYCC/uhOIMV5v1WcHvMRaXvEE= X-Google-Smtp-Source: ABdhPJzlXdtTxTNMVr3UJ2qvuXcMzptpBRer1lr4LV7rFiuYUJo4KcAlfSN26gYwmbMp0oE+S+Ls01ZepR4NAhB1lvU= X-Received: by 2002:adf:dd09:: with SMTP id a9mr7842934wrm.90.1610009402837; Thu, 07 Jan 2021 00:50:02 -0800 (PST) MIME-Version: 1.0 References: <20210106173303.27988-1-philipp.tomsich@vrull.eu> In-Reply-To: From: Kito Cheng Date: Thu, 7 Jan 2021 16:49:51 +0800 Message-ID: Subject: Re: [PATCH] RISC-V: Zihintpause: add __builtin_riscv_pause To: Philipp Tomsich Cc: Kito Cheng , Andrew Waterman , Jim Wilson , GCC Patches Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Spam-Status: No, score=-1.2 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP, URIBL_SBL, URIBL_SBL_A autolearn=no autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) 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: Thu, 07 Jan 2021 08:50:05 -0000 My point is tracking info and consistent behavior/scheme with other extensions, so personally I strongly prefer it should be guarded with -march. But maybe we could create an issue on riscv-c-api-doc[1] or riscv-toolchain-conventions[2] to get feedback from LLVM folks, since I think this behavior should align between LLVM and GCC. [1] https://github.com/riscv/riscv-c-api-doc [2] https://github.com/riscv/riscv-toolchain-conventions On Thu, Jan 7, 2021 at 2:53 PM Philipp Tomsich w= rote: > > Kito: > > We had originally considered to guard this with a -march, but decided aga= inst it > eventually: this instruction will be (among other cases) used in the cpu_= relax() of > the Linux kernel. For cases like that, we should consider this the basel= ine (i.e. > either there's no pause=E2=80=94in which case, the encoded fence will not= hurt=E2=80=94or the > Zihintpause extension)... but it all maps back to a single builtin-call. > > Note that the Zihintfence will be enabled for all (also older) targets, a= s the insn > is supported there as well (as a fence that doesn't do anything)... so gu= arding it > will not really change the behavior. > > That said, I'll get going on an v2 that will include the -march guard (an= d we can > still turn things back to how they are today). > > Thanks, > Philipp. > > On Thu, 7 Jan 2021 at 06:42, Kito Cheng wrote: >> >> Hi Andrew: >> >> It's safe to execute on old machine, but it is still a new extension not= included on baseline ISA, so I still prefer having -march to guard that, a= nd then we can track that in the ELF attribute to see what extensions and w= hich version are used in the executable / object files. >> >> >> On Thu, Jan 7, 2021 at 11:51 AM Andrew Waterman w= rote: >>> >>> I've got a contrary opinion: >>> >>> Since HINTs are guaranteed to execute as no-ops--e.g., this one is >>> just a FENCE instruction, which is already a mandatory part of the >>> base ISA--they don't _need_ to be called out as separate extensions in >>> the toolchain. >>> >>> Although there's nothing fundamentally wrong with Kito's suggestion, >>> it seems like an extra hoop to jump through without commensurate >>> benefit. I see no reason to restrict the use of __builtin_pause, >>> since all RISC-V implementations, including old ones, are required to >>> support it. And, of course, that's the reason we encoded it this way >>> :) >>> >>> >>> On Wed, Jan 6, 2021 at 7:35 PM Kito Cheng wrote: >>> > >>> > Hi Philipp: >>> > >>> > Could you add zihintpause to -march parser and guard that on the >>> > pattern and builtin like zifencei[1-2]? >>> > >>> > And could you sent a PR to >>> > https://github.com/riscv/riscv-c-api-doc/blob/master/riscv-c-api.md t= o >>> > mention __builtin_riscv_pause? >>> > >>> > Thanks! >>> > >>> > [1] march parser change: >>> > https://github.com/gcc-mirror/gcc/commit/b03be74bad08c382da47e048007a= 78fa3fb4ef49 >>> > [2] Default version for ext.: >>> > https://github.com/gcc-mirror/gcc/commit/4b81528241ca682025d92558ff6a= eec91dafdca8 >>> > >>> > >>> > > --- /dev/null >>> > > +++ b/gcc/testsuite/gcc.target/riscv/builtin_pause.c >>> > > @@ -0,0 +1,10 @@ >>> > > +/* { dg-do compile } */ >>> > > +/* { dg-options "-O2" } */ >>> > > + >>> > > +void test_pause() >>> > >>> > I would suggest you change the function name in the testcase, >>> > otherwise the scan-assembler test will always pass even if you didn't >>> > generate "pause" instruction. >>> > >>> > >>> > > +{ >>> > > + __builtin_riscv_pause (); >>> > > +} >>> > > + >>> > > +/* { dg-final { scan-assembler "pause" } } */ >>> > > + >>> > > -- >>> > > 2.18.4 >>> > >