From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pf1-x42c.google.com (mail-pf1-x42c.google.com [IPv6:2607:f8b0:4864:20::42c]) by sourceware.org (Postfix) with ESMTPS id 0F13F3858D1E for ; Tue, 25 Apr 2023 20:03:56 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 0F13F3858D1E Authentication-Results: sourceware.org; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: sourceware.org; spf=pass smtp.mailfrom=gmail.com Received: by mail-pf1-x42c.google.com with SMTP id d2e1a72fcca58-63b51fd2972so4965430b3a.3 for ; Tue, 25 Apr 2023 13:03:55 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20221208; t=1682453035; x=1685045035; h=content-transfer-encoding:in-reply-to:from:references:cc:to :content-language:subject:user-agent:mime-version:date:message-id :from:to:cc:subject:date:message-id:reply-to; bh=3SPUGm9n2i1quWUDHIFVNr7bXhahiV6WnirDohElus0=; b=jvt9gdG2VoZoRf0HShHdVyMseJMRvUuqxvEKp4VvKhGjjyr5GgCnbMyDQ1hVG8NXMv Afi+AtPOfLpJ3Zy35RypkTittzr/ElplCRy6YEDx9d+grWI8+yJUbWooP35GUFspxvCp PnY4SHJwh2XQyygEIsX5rN00LtTBqM/6/WZwZgeEX5rRA7H88jnvDwjrAS9HHeZBssLY ENOP9FlmK4We4MYAF81gur7ASJykY+TfDzugUibY9qL/V2YsQyiG5CPW7UMt7Ywyh/PS 69Xf/A3t+T/USJopUSUW9E7kj6dQ2fOBifY+hRYtaMznVaNMTSTmz5jhk2UuJUcCXD5Q 0B9g== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20221208; t=1682453035; x=1685045035; h=content-transfer-encoding:in-reply-to:from:references:cc:to :content-language:subject:user-agent:mime-version:date:message-id :x-gm-message-state:from:to:cc:subject:date:message-id:reply-to; bh=3SPUGm9n2i1quWUDHIFVNr7bXhahiV6WnirDohElus0=; b=RHe5RVj0BkIC+TxejtEAXqdZxMGtLVxeu6zN6xpH+2Xd1QFkr+pJUrNPFTAMj9D24+ w6f7E5CfTwCJ93ZMV2JcYz0xwxGzUkH1PB7SNvrP4DhZ3Y9zXFHAnR9xkPl96u/L4yLM lA+sSINxuW2ulChPfKpH2u/VwhFbzwBjR2b3vGkYdtUu54qfjwbRwRcu7sqe/KK3GkK5 2jNRgNdwgep0WzTKtpOg/emHgSsGq1/A7qQULebhmVwNZKwFx80jdUc1HTq4lZuP3XrK +C9HUOfKg3PCn5VruDIOIBVvprq/j1OYMWLp/VMJhBa9+Y8ucvzGvaS/UdwsSZwGtzFM MQ2Q== X-Gm-Message-State: AAQBX9coNWT9nocy9HzOvZROvrUTEeU/f2UHUGVPot7j3lzPYhrFIgQM fU/iNkGD6NgbInvXiuD7Uz4= X-Google-Smtp-Source: AKy350YR06eEH+jaQNoiiJzospWdtHP07g4YJjB4mbUTpvJ1ZvA0RIwKGaHOIN03a8GOq0MhlDfW0w== X-Received: by 2002:a17:902:ebc6:b0:1a5:22a6:4e6a with SMTP id p6-20020a170902ebc600b001a522a64e6amr19058472plg.51.1682453034699; Tue, 25 Apr 2023 13:03:54 -0700 (PDT) Received: from ?IPV6:2601:681:8600:13d0::99f? ([2601:681:8600:13d0::99f]) by smtp.gmail.com with ESMTPSA id a7-20020a1709027d8700b001a653a32173sm8595276plm.29.2023.04.25.13.03.53 (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Tue, 25 Apr 2023 13:03:54 -0700 (PDT) Message-ID: <3d4cda57-e0f0-9861-6969-f31f22ca9ef8@gmail.com> Date: Tue, 25 Apr 2023 14:03:52 -0600 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101 Thunderbird/102.9.1 Subject: Re: [Committed] riscv: relax splitter restrictions for creating pseudos Content-Language: en-US To: Vineet Gupta , gcc-patches@gcc.gnu.org Cc: kito.cheng@gmail.com, Palmer Dabbelt , Philipp Tomsich , Christoph Mullner , gnu-toolchain@rivosinc.co References: <20230418143635.980594-1-vineetg@rivosinc.com> <680c7bbe-5d6e-07cd-8468-247afc65e1dd@gmail.com> From: Jeff Law In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit X-Spam-Status: No, score=-1.9 required=5.0 tests=BAYES_00,BODY_8BITS,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FREEMAIL_FROM,NICE_REPLY_A,RCVD_IN_DNSWL_NONE,SPF_HELO_NONE,SPF_PASS,TXREP,T_SCC_BODY_TEXT_LINE autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: On 4/25/23 11:25, Vineet Gupta wrote: > > > On 4/18/23 11:36, Jeff Law via Gcc-patches wrote: >> >> >> On 4/18/23 08:36, Vineet Gupta wrote: >>> [partial addressing of PR/109279] >>> >>> RISCV splitters have restrictions to not create pesudos due to a combine >>> limitatation. And despite this being a split-during-combine limitation, >>> all split passes take the hit due to way define*_split are used in gcc. >>> >>> With the original combine issue being fixed 61bee6aed2 ("combine: Don't >>> record for UNDO_MODE pointers into regno_reg_rtx array [PR104985]") >>> the RV splitters can now be relaxed. >>> >>> This improves the codegen in general. e.g. >>> >>>     long long f(void) { return 0x0101010101010101ull; } >>> >>> Before >>> >>>     li    a0,0x01010000 >>>     addi    a0,0x0101 >>>     slli    a0,a0,16 >>>     addi    a0,a0,0x0101 >>>     slli    a0,a0,16 >>>     addi    a0,a0,0x0101 >>>     ret >>> >>> With patch >>> >>>     li    a5,0x01010000 >>>     addi    a5,a5,0x0101 >>>     mv    a0,a5 >>>     slli    a5,a5,32 >>>     add    a0,a5,a0 >>>     ret >>> >>> This is testsuite clean, no regression w/ patch. >>> >>>                 ========= Summary of gcc testsuite ========= >>>                              | # of unexpected case / # of unique >>> unexpected case >>>                              |          gcc |          g++ | gfortran | >>>   rv64imafdc/  lp64d/ medlow |    2 /     2 |    1 /     1 | 6 /     1 | >>>     rv64imac/   lp64/ medlow |    3 /     3 |    1 /     1 | 43 / 8 | >>>   rv32imafdc/ ilp32d/ medlow |    1 /     1 |    3 /     2 | 6 /     1 | >>>     rv32imac/  ilp32/ medlow |    1 /     1 |    3 /     2 | 43 / 8 | >>> >>> This came up as part of IRC chat on PR/109279 and was suggested by >>> Andrew Pinski. >>> >>> Signed-off-by: Vineet Gupta >>> --- >>>   gcc/config/riscv/riscv-protos.h |  4 +-- >>>   gcc/config/riscv/riscv.cc       | 46 +++++++++++++-------------------- >>>   gcc/config/riscv/riscv.md       |  8 +++--- >>>   3 files changed, 24 insertions(+), 34 deletions(-) >> This looks fine, except that you don't have a ChangeLog. > > Pushed with Changelog and some additional info about qemu icount > reductions with this patch > >     500.perlbench_r 0       1235310737733   1231742384460   0.29% >                     1       744489708820    743515759958 >                     2       714072106766    712875768625    0.17% >     502.gcc_r       0       197365353269    197178223030 >                     1       235614445254    235465240341 >                     2       226769189971    226604663947 >                     3       188315686133    188123584015 >                     4       289372107644    289187945424 >     503.bwaves_r    0       326291538768    326291539697 >                     1       515809487294    515809488863 >                     2       401647004144    401647005463 >                     3       488750661035    488750662484 >     505.mcf_r       0       681926695281    681925418147 >     507.cactuBSSN_r 0       3832240965352   3832226068734 >     508.namd_r      0       1919838790866   1919832527292 >     510.parest_r    0       3515999635520   3515878553435 >     511.povray_r    0       3073889223775   3074758622749 >     519.lbm_r       0       1194077464296   1194077464041 >     520.omnetpp_r   0       1014144252460   1011530791131   0.26% >     521.wrf_r       0       3966715533120   3966265425092 >     523.xalancbmk_r 0       1064914296949   1064506711802 >     525.x264_r      0       509290028335    509258131632 >                     1       2001424246635   2001677767181 >                     2       1914660798226   1914869407575 >     526.blender_r   0       1726083839515   1725974286174 >     527.cam4_r      0       2336526136415   2333656336419 >     531.deepsjeng_r 0       1689007489539   1686541299243   0.15% >     538.imagick_r   0       3247960667520   3247942048723 >     541.leela_r     0       2072315300365   2070248271250 >     544.nab_r       0       1527909091282   1527906483039 >     548.exchange2_r 0       2086120304280   2086314757502 >     549.fotonik3d_r 0       2261694058444   2261670330720 >     554.roms_r      0       2640547903140   2640512733483 >     557.xz_r        0       388736881767    386880875636    0.48% >                     1       959356981818    959993132842 >                     2       547643353034    546374038310    0.23% >     997.specrand_fr 0       512881578       512599641 >     999.specrand_ir 0       512881578       512599641 Just a note, there are some regressions in this table. For example xz input #2. So to be fair when making comparisons it's probably worth noting the regressions as well as improvements. Anyway, my results are in line with yours. Given the instruction counts and known IPC, I think we're taking about a 2.5% hit on deepsjeng and about a 1% hit on leela and x264#2 comparing our internal gcc-12 vs gcc-13 trees. Jeff