From patchwork Mon May 3 08:44:38 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Noah Goldstein X-Patchwork-Id: 43207 Return-Path: X-Original-To: patchwork@sourceware.org Delivered-To: patchwork@sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 0BC71393F873; Mon, 3 May 2021 08:46:11 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 0BC71393F873 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1620031571; bh=BJJCM5pTuFIFebpk8V/bvzn8mYdMTXg7C1e2Wmwz+dE=; h=To:Subject:Date:In-Reply-To:References:List-Id:List-Unsubscribe: List-Archive:List-Post:List-Help:List-Subscribe:From:Reply-To: From; b=IJnonyZJRyHlCegTZsQplGnGWNn+VgkmnSixxv4emEk+oP9kRd50gLv150hJtF9Oo igWBHSQf5xfxozHpoBcwYeTR6Pm+SIzghEExBm/u+XQqN56jPz6JALh0L5Hc5MMFyq MgtiJLy5tQkFAALb6meK44Fd9pZOvR8GneOWFCFo= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-qv1-xf2d.google.com (mail-qv1-xf2d.google.com [IPv6:2607:f8b0:4864:20::f2d]) by sourceware.org (Postfix) with ESMTPS id D9112385DC33 for ; Mon, 3 May 2021 08:46:04 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org D9112385DC33 Received: by mail-qv1-xf2d.google.com with SMTP id i8so2209307qvv.0 for ; Mon, 03 May 2021 01:46:04 -0700 (PDT) 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:in-reply-to :references:mime-version:content-transfer-encoding; bh=BJJCM5pTuFIFebpk8V/bvzn8mYdMTXg7C1e2Wmwz+dE=; b=PdilHfWgDq2VZr/86irNIZrR+5eJiGSR4JoXREc5E0hhMlHEEtBD1DgnGZZNQ2prRM mEiSAKeyCRwArojufMxbi2UGCWU7JhiWBdMSIdsZL4LWfbNhdMDtzCVPgcaThhpBJheC 9kRjYfNfmBdwK6sQ5+M7pNdBguxACGpGRLXbVIT4Jrqp8aIjQ3NCBKbu1frJLdyl+sOf 3e03aRHbubmeMJpCLXTK8C/FFnkuQvBFHC+3KIZjbKGMQRUPO/O8gJTC5fqUysGZaEJL Abq5L3BdlrdNfCouwbUytP7UUVhIOzu5q4VnNA4HncYLNyOF/VXYLJyFtSYpmI7DelEh 9kSA== X-Gm-Message-State: AOAM533tbkUApITYE4jDTz7svhwsKFBCv6a28bP7O3kWRTgiArYQQ1Rd /3Tiu/0kqtoBEyp2HZaG/VyMFwmO9ok= X-Google-Smtp-Source: ABdhPJy9UNh8UELnrQ9ZhRMi3GPipn/+vLAr1k4xREXS63UCPsLIuMMsok5rEa6+S4OzpNFAcvVG8w== X-Received: by 2002:a05:6214:19ed:: with SMTP id q13mr9045666qvc.58.1620031562800; Mon, 03 May 2021 01:46:02 -0700 (PDT) Received: from localhost.localdomain (pool-71-245-178-39.pitbpa.fios.verizon.net. [71.245.178.39]) by smtp.googlemail.com with ESMTPSA id a3sm7954970qto.2.2021.05.03.01.46.01 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 03 May 2021 01:46:02 -0700 (PDT) To: libc-alpha@sourceware.org Subject: [PATCH v1 3/3] x86: Optimize memchr-evex.S Date: Mon, 3 May 2021 04:44:38 -0400 Message-Id: <20210503084435.160548-3-goldstein.w.n@gmail.com> X-Mailer: git-send-email 2.29.2 In-Reply-To: <20210503084435.160548-1-goldstein.w.n@gmail.com> References: <20210503084435.160548-1-goldstein.w.n@gmail.com> MIME-Version: 1.0 X-Spam-Status: No, score=-12.6 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, FREEMAIL_FROM, GIT_PATCH_0, RCVD_IN_DNSWL_NONE, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org X-BeenThere: libc-alpha@sourceware.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Libc-alpha mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Noah Goldstein via Libc-alpha From: Noah Goldstein Reply-To: Noah Goldstein Errors-To: libc-alpha-bounces@sourceware.org Sender: "Libc-alpha" No bug. This commit optimizes memchr-evex.S. The optimizations include replacing some branches with cmovcc, avoiding some branches entirely in the less_4x_vec case, making the page cross logic less strict, saving some ALU in the alignment process, and most importantly increasing ILP in the 4x loop. test-memchr, test-rawmemchr, and test-wmemchr are all passing. Signed-off-by: Noah Goldstein --- Tests where run on the following CPUs: Tigerlake: https://ark.intel.com/content/www/us/en/ark/products/208921/intel-core-i7-1165g7-processor-12m-cache-up-to-4-70-ghz-with-ipu.html Icelake: https://ark.intel.com/content/www/us/en/ark/products/196597/intel-core-i7-1065g7-processor-8m-cache-up-to-3-90-ghz.html Skylake: https://ark.intel.com/content/www/us/en/ark/products/149091/intel-core-i7-8565u-processor-8m-cache-up-to-4-60-ghz.html All times are the geometric mean of N=20. The unit of time is seconds. "Cur" refers to the current implementation "New" refers to this patches implementation Note: The numbers for size = [1, 32] are highly dependent on function alignment. That being said the new implementation which uses cmovcc instead of a branch (mostly for the reason of high variance with different alignments) for the [1, 32] case is far more consistent and performs about as well (and should only be a bigger improvement in cases where the sizes / position are not 100% predictable). For memchr-evex the numbers are a near universal improvement. The case where the current implement as better is for size = 0 and for size = [1, 32] with pos < size the two implementations are about the same. For size = [1, 32] with pos > size, for medium range sizes, and large size, however, the new implementation is faster. Results For Tigerlake memchr-evex size , algn , Pos , Cur T , New T , Win , Dif 2048 , 0 , , 32 5.58 , 5.22 , New , 0.36 256 , 1 , , 64 5.22 , 4.93 , New , 0.29 2048 , 0 , , 64 5.22 , 4.89 , New , 0.33 256 , 2 , , 64 5.14 , 4.81 , New , 0.33 2048 , 0 , , 128 6.3 , 5.67 , New , 0.63 256 , 3 , , 64 5.22 , 4.9 , New , 0.32 2048 , 0 , , 256 11.07 , 10.92 , New , 0.15 256 , 4 , , 64 5.16 , 4.86 , New , 0.3 2048 , 0 , , 512 15.66 , 14.81 , New , 0.85 256 , 5 , , 64 5.15 , 4.84 , New , 0.31 2048 , 0 , , 1024 25.7 , 23.02 , New , 2.68 256 , 6 , , 64 5.12 , 4.89 , New , 0.23 2048 , 0 , , 2048 42.34 , 37.71 , New , 4.63 256 , 7 , , 64 5.03 , 4.62 , New , 0.41 192 , 1 , , 32 4.96 , 4.28 , New , 0.68 256 , 1 , , 32 4.95 , 4.28 , New , 0.67 512 , 1 , , 32 4.94 , 4.29 , New , 0.65 192 , 2 , , 64 5.1 , 4.8 , New , 0.3 512 , 2 , , 64 5.12 , 4.72 , New , 0.4 192 , 3 , , 96 5.54 , 5.12 , New , 0.42 256 , 3 , , 96 5.52 , 5.15 , New , 0.37 512 , 3 , , 96 5.51 , 5.16 , New , 0.35 192 , 4 , , 128 6.1 , 5.53 , New , 0.57 256 , 4 , , 128 6.09 , 5.49 , New , 0.6 512 , 4 , , 128 6.08 , 5.48 , New , 0.6 192 , 5 , , 160 7.42 , 6.71 , New , 0.71 256 , 5 , , 160 6.86 , 6.71 , New , 0.15 512 , 5 , , 160 9.28 , 8.68 , New , 0.6 192 , 6 , , 192 7.94 , 7.47 , New , 0.47 256 , 6 , , 192 7.62 , 7.17 , New , 0.45 512 , 6 , , 192 9.2 , 9.16 , New , 0.04 192 , 7 , , 224 8.02 , 7.43 , New , 0.59 256 , 7 , , 224 8.34 , 7.85 , New , 0.49 512 , 7 , , 224 9.89 , 9.16 , New , 0.73 2 , 0 , , 1 3.0 , 3.0 , Eq , 0.0 2 , 1 , , 1 3.0 , 3.0 , Eq , 0.0 0 , 0 , , 1 3.01 , 3.6 , Cur , 0.59 0 , 1 , , 1 3.01 , 3.6 , Cur , 0.59 3 , 0 , , 2 3.0 , 3.0 , Eq , 0.0 3 , 2 , , 2 3.0 , 3.0 , Eq , 0.0 1 , 0 , , 2 3.6 , 3.0 , New , 0.6 1 , 2 , , 2 3.6 , 3.0 , New , 0.6 4 , 0 , , 3 3.01 , 3.01 , Eq , 0.0 4 , 3 , , 3 3.01 , 3.01 , Eq , 0.0 2 , 0 , , 3 3.62 , 3.02 , New , 0.6 2 , 3 , , 3 3.62 , 3.03 , New , 0.59 5 , 0 , , 4 3.02 , 3.03 , Cur , 0.01 5 , 4 , , 4 3.02 , 3.02 , Eq , 0.0 3 , 0 , , 4 3.63 , 3.02 , New , 0.61 3 , 4 , , 4 3.63 , 3.04 , New , 0.59 6 , 0 , , 5 3.05 , 3.04 , New , 0.01 6 , 5 , , 5 3.02 , 3.02 , Eq , 0.0 4 , 0 , , 5 3.63 , 3.02 , New , 0.61 4 , 5 , , 5 3.64 , 3.03 , New , 0.61 7 , 0 , , 6 3.03 , 3.03 , Eq , 0.0 7 , 6 , , 6 3.02 , 3.02 , Eq , 0.0 5 , 0 , , 6 3.64 , 3.01 , New , 0.63 5 , 6 , , 6 3.64 , 3.03 , New , 0.61 8 , 0 , , 7 3.03 , 3.04 , Cur , 0.01 8 , 7 , , 7 3.04 , 3.04 , Eq , 0.0 6 , 0 , , 7 3.67 , 3.04 , New , 0.63 6 , 7 , , 7 3.65 , 3.05 , New , 0.6 9 , 0 , , 8 3.05 , 3.05 , Eq , 0.0 7 , 0 , , 8 3.67 , 3.05 , New , 0.62 10 , 0 , , 9 3.06 , 3.06 , Eq , 0.0 10 , 1 , , 9 3.06 , 3.06 , Eq , 0.0 8 , 0 , , 9 3.67 , 3.06 , New , 0.61 8 , 1 , , 9 3.67 , 3.06 , New , 0.61 11 , 0 , , 10 3.06 , 3.06 , Eq , 0.0 11 , 2 , , 10 3.07 , 3.06 , New , 0.01 9 , 0 , , 10 3.67 , 3.05 , New , 0.62 9 , 2 , , 10 3.67 , 3.06 , New , 0.61 12 , 0 , , 11 3.06 , 3.06 , Eq , 0.0 12 , 3 , , 11 3.06 , 3.06 , Eq , 0.0 10 , 0 , , 11 3.67 , 3.06 , New , 0.61 10 , 3 , , 11 3.67 , 3.06 , New , 0.61 13 , 0 , , 12 3.06 , 3.07 , Cur , 0.01 13 , 4 , , 12 3.06 , 3.07 , Cur , 0.01 11 , 0 , , 12 3.67 , 3.11 , New , 0.56 11 , 4 , , 12 3.68 , 3.12 , New , 0.56 14 , 0 , , 13 3.07 , 3.1 , Cur , 0.03 14 , 5 , , 13 3.06 , 3.07 , Cur , 0.01 12 , 0 , , 13 3.67 , 3.07 , New , 0.6 12 , 5 , , 13 3.67 , 3.08 , New , 0.59 15 , 0 , , 14 3.06 , 3.06 , Eq , 0.0 15 , 6 , , 14 3.07 , 3.06 , New , 0.01 13 , 0 , , 14 3.67 , 3.06 , New , 0.61 13 , 6 , , 14 3.68 , 3.06 , New , 0.62 16 , 0 , , 15 3.06 , 3.06 , Eq , 0.0 16 , 7 , , 15 3.06 , 3.05 , New , 0.01 14 , 0 , , 15 3.68 , 3.06 , New , 0.62 14 , 7 , , 15 3.67 , 3.06 , New , 0.61 17 , 0 , , 16 3.07 , 3.06 , New , 0.01 15 , 0 , , 16 3.68 , 3.06 , New , 0.62 18 , 0 , , 17 3.06 , 3.06 , Eq , 0.0 18 , 1 , , 17 3.06 , 3.06 , Eq , 0.0 16 , 0 , , 17 3.67 , 3.06 , New , 0.61 16 , 1 , , 17 3.67 , 3.05 , New , 0.62 19 , 0 , , 18 3.07 , 3.06 , New , 0.01 19 , 2 , , 18 3.06 , 3.06 , Eq , 0.0 17 , 0 , , 18 3.68 , 3.08 , New , 0.6 17 , 2 , , 18 3.68 , 3.06 , New , 0.62 20 , 0 , , 19 3.06 , 3.06 , Eq , 0.0 20 , 3 , , 19 3.06 , 3.06 , Eq , 0.0 18 , 0 , , 19 3.68 , 3.06 , New , 0.62 18 , 3 , , 19 3.68 , 3.06 , New , 0.62 21 , 0 , , 20 3.06 , 3.06 , Eq , 0.0 21 , 4 , , 20 3.06 , 3.06 , Eq , 0.0 19 , 0 , , 20 3.67 , 3.06 , New , 0.61 19 , 4 , , 20 3.67 , 3.06 , New , 0.61 22 , 0 , , 21 3.06 , 3.06 , Eq , 0.0 22 , 5 , , 21 3.06 , 3.06 , Eq , 0.0 20 , 0 , , 21 3.67 , 3.05 , New , 0.62 20 , 5 , , 21 3.68 , 3.06 , New , 0.62 23 , 0 , , 22 3.07 , 3.06 , New , 0.01 23 , 6 , , 22 3.06 , 3.06 , Eq , 0.0 21 , 0 , , 22 3.68 , 3.07 , New , 0.61 21 , 6 , , 22 3.67 , 3.06 , New , 0.61 24 , 0 , , 23 3.19 , 3.06 , New , 0.13 24 , 7 , , 23 3.08 , 3.06 , New , 0.02 22 , 0 , , 23 3.69 , 3.06 , New , 0.63 22 , 7 , , 23 3.68 , 3.06 , New , 0.62 25 , 0 , , 24 3.07 , 3.06 , New , 0.01 23 , 0 , , 24 3.68 , 3.06 , New , 0.62 26 , 0 , , 25 3.06 , 3.05 , New , 0.01 26 , 1 , , 25 3.07 , 3.06 , New , 0.01 24 , 0 , , 25 3.67 , 3.05 , New , 0.62 24 , 1 , , 25 3.68 , 3.06 , New , 0.62 27 , 0 , , 26 3.12 , 3.06 , New , 0.06 27 , 2 , , 26 3.08 , 3.06 , New , 0.02 25 , 0 , , 26 3.69 , 3.06 , New , 0.63 25 , 2 , , 26 3.67 , 3.06 , New , 0.61 28 , 0 , , 27 3.06 , 3.06 , Eq , 0.0 28 , 3 , , 27 3.06 , 3.06 , Eq , 0.0 26 , 0 , , 27 3.67 , 3.06 , New , 0.61 26 , 3 , , 27 3.67 , 3.06 , New , 0.61 29 , 0 , , 28 3.06 , 3.06 , Eq , 0.0 29 , 4 , , 28 3.06 , 3.06 , Eq , 0.0 27 , 0 , , 28 3.68 , 3.05 , New , 0.63 27 , 4 , , 28 3.67 , 3.06 , New , 0.61 30 , 0 , , 29 3.06 , 3.06 , Eq , 0.0 30 , 5 , , 29 3.06 , 3.06 , Eq , 0.0 28 , 0 , , 29 3.67 , 3.06 , New , 0.61 28 , 5 , , 29 3.68 , 3.06 , New , 0.62 31 , 0 , , 30 3.06 , 3.06 , Eq , 0.0 31 , 6 , , 30 3.06 , 3.06 , Eq , 0.0 29 , 0 , , 30 3.68 , 3.06 , New , 0.62 29 , 6 , , 30 3.7 , 3.06 , New , 0.64 32 , 0 , , 31 3.17 , 3.06 , New , 0.11 32 , 7 , , 31 3.12 , 3.06 , New , 0.06 30 , 0 , , 31 3.68 , 3.06 , New , 0.62 30 , 7 , , 31 3.68 , 3.06 , New , 0.62 Results For Icelake memchr-evex size , algn , Pos , Cur T , New T , Win , Dif 2048 , 0 , , 32 4.94 , 4.26 , New , 0.68 256 , 1 , , 64 4.5 , 4.13 , New , 0.37 2048 , 0 , , 64 4.19 , 3.9 , New , 0.29 256 , 2 , , 64 4.19 , 3.87 , New , 0.32 2048 , 0 , , 128 4.96 , 4.53 , New , 0.43 256 , 3 , , 64 4.07 , 3.86 , New , 0.21 2048 , 0 , , 256 8.77 , 8.61 , New , 0.16 256 , 4 , , 64 4.08 , 3.87 , New , 0.21 2048 , 0 , , 512 12.22 , 11.67 , New , 0.55 256 , 5 , , 64 4.12 , 3.83 , New , 0.29 2048 , 0 , , 1024 20.06 , 18.09 , New , 1.97 256 , 6 , , 64 4.2 , 3.95 , New , 0.25 2048 , 0 , , 2048 33.83 , 30.62 , New , 3.21 256 , 7 , , 64 4.3 , 4.04 , New , 0.26 192 , 1 , , 32 4.2 , 3.71 , New , 0.49 256 , 1 , , 32 4.24 , 3.76 , New , 0.48 512 , 1 , , 32 4.29 , 3.74 , New , 0.55 192 , 2 , , 64 4.42 , 4.0 , New , 0.42 512 , 2 , , 64 4.17 , 3.83 , New , 0.34 192 , 3 , , 96 4.44 , 4.26 , New , 0.18 256 , 3 , , 96 4.45 , 4.14 , New , 0.31 512 , 3 , , 96 4.42 , 4.15 , New , 0.27 192 , 4 , , 128 4.93 , 4.45 , New , 0.48 256 , 4 , , 128 4.93 , 4.47 , New , 0.46 512 , 4 , , 128 4.95 , 4.47 , New , 0.48 192 , 5 , , 160 5.95 , 5.44 , New , 0.51 256 , 5 , , 160 5.59 , 5.47 , New , 0.12 512 , 5 , , 160 7.59 , 7.34 , New , 0.25 192 , 6 , , 192 6.53 , 6.08 , New , 0.45 256 , 6 , , 192 6.2 , 5.88 , New , 0.32 512 , 6 , , 192 7.53 , 7.62 , Cur , 0.09 192 , 7 , , 224 6.62 , 6.12 , New , 0.5 256 , 7 , , 224 6.79 , 6.51 , New , 0.28 512 , 7 , , 224 8.12 , 7.61 , New , 0.51 2 , 0 , , 1 2.5 , 2.54 , Cur , 0.04 2 , 1 , , 1 2.56 , 2.55 , New , 0.01 0 , 0 , , 1 2.57 , 3.12 , Cur , 0.55 0 , 1 , , 1 2.59 , 3.14 , Cur , 0.55 3 , 0 , , 2 2.62 , 2.63 , Cur , 0.01 3 , 2 , , 2 2.66 , 2.67 , Cur , 0.01 1 , 0 , , 2 3.24 , 2.72 , New , 0.52 1 , 2 , , 2 3.28 , 2.75 , New , 0.53 4 , 0 , , 3 2.78 , 2.8 , Cur , 0.02 4 , 3 , , 3 2.8 , 2.82 , Cur , 0.02 2 , 0 , , 3 3.38 , 2.86 , New , 0.52 2 , 3 , , 3 3.41 , 2.89 , New , 0.52 5 , 0 , , 4 2.88 , 2.91 , Cur , 0.03 5 , 4 , , 4 2.88 , 2.92 , Cur , 0.04 3 , 0 , , 4 3.48 , 2.93 , New , 0.55 3 , 4 , , 4 3.47 , 2.93 , New , 0.54 6 , 0 , , 5 2.95 , 2.94 , New , 0.01 6 , 5 , , 5 2.91 , 2.92 , Cur , 0.01 4 , 0 , , 5 3.47 , 2.9 , New , 0.57 4 , 5 , , 5 3.43 , 2.91 , New , 0.52 7 , 0 , , 6 2.87 , 2.9 , Cur , 0.03 7 , 6 , , 6 2.87 , 2.89 , Cur , 0.02 5 , 0 , , 6 3.44 , 2.88 , New , 0.56 5 , 6 , , 6 3.41 , 2.87 , New , 0.54 8 , 0 , , 7 2.86 , 2.87 , Cur , 0.01 8 , 7 , , 7 2.86 , 2.87 , Cur , 0.01 6 , 0 , , 7 3.43 , 2.87 , New , 0.56 6 , 7 , , 7 3.44 , 2.87 , New , 0.57 9 , 0 , , 8 2.86 , 2.88 , Cur , 0.02 7 , 0 , , 8 3.41 , 2.89 , New , 0.52 10 , 0 , , 9 2.83 , 2.87 , Cur , 0.04 10 , 1 , , 9 2.82 , 2.87 , Cur , 0.05 8 , 0 , , 9 3.4 , 2.89 , New , 0.51 8 , 1 , , 9 3.41 , 2.87 , New , 0.54 11 , 0 , , 10 2.83 , 2.88 , Cur , 0.05 11 , 2 , , 10 2.84 , 2.88 , Cur , 0.04 9 , 0 , , 10 3.41 , 2.87 , New , 0.54 9 , 2 , , 10 3.41 , 2.88 , New , 0.53 12 , 0 , , 11 2.83 , 2.89 , Cur , 0.06 12 , 3 , , 11 2.85 , 2.87 , Cur , 0.02 10 , 0 , , 11 3.41 , 2.87 , New , 0.54 10 , 3 , , 11 3.42 , 2.88 , New , 0.54 13 , 0 , , 12 2.86 , 2.87 , Cur , 0.01 13 , 4 , , 12 2.84 , 2.88 , Cur , 0.04 11 , 0 , , 12 3.43 , 2.87 , New , 0.56 11 , 4 , , 12 3.49 , 2.87 , New , 0.62 14 , 0 , , 13 2.85 , 2.86 , Cur , 0.01 14 , 5 , , 13 2.85 , 2.86 , Cur , 0.01 12 , 0 , , 13 3.41 , 2.86 , New , 0.55 12 , 5 , , 13 3.44 , 2.85 , New , 0.59 15 , 0 , , 14 2.83 , 2.87 , Cur , 0.04 15 , 6 , , 14 2.82 , 2.86 , Cur , 0.04 13 , 0 , , 14 3.41 , 2.86 , New , 0.55 13 , 6 , , 14 3.4 , 2.86 , New , 0.54 16 , 0 , , 15 2.84 , 2.86 , Cur , 0.02 16 , 7 , , 15 2.83 , 2.85 , Cur , 0.02 14 , 0 , , 15 3.41 , 2.85 , New , 0.56 14 , 7 , , 15 3.39 , 2.87 , New , 0.52 17 , 0 , , 16 2.83 , 2.87 , Cur , 0.04 15 , 0 , , 16 3.4 , 2.85 , New , 0.55 18 , 0 , , 17 2.83 , 2.86 , Cur , 0.03 18 , 1 , , 17 2.85 , 2.84 , New , 0.01 16 , 0 , , 17 3.41 , 2.85 , New , 0.56 16 , 1 , , 17 3.4 , 2.86 , New , 0.54 19 , 0 , , 18 2.8 , 2.84 , Cur , 0.04 19 , 2 , , 18 2.82 , 2.83 , Cur , 0.01 17 , 0 , , 18 3.39 , 2.86 , New , 0.53 17 , 2 , , 18 3.39 , 2.84 , New , 0.55 20 , 0 , , 19 2.85 , 2.87 , Cur , 0.02 20 , 3 , , 19 2.88 , 2.87 , New , 0.01 18 , 0 , , 19 3.38 , 2.85 , New , 0.53 18 , 3 , , 19 3.4 , 2.85 , New , 0.55 21 , 0 , , 20 2.83 , 2.85 , Cur , 0.02 21 , 4 , , 20 2.88 , 2.85 , New , 0.03 19 , 0 , , 20 3.39 , 2.84 , New , 0.55 19 , 4 , , 20 3.39 , 2.96 , New , 0.43 22 , 0 , , 21 2.84 , 2.9 , Cur , 0.06 22 , 5 , , 21 2.81 , 2.84 , Cur , 0.03 20 , 0 , , 21 3.41 , 2.81 , New , 0.6 20 , 5 , , 21 3.38 , 2.83 , New , 0.55 23 , 0 , , 22 2.8 , 2.82 , Cur , 0.02 23 , 6 , , 22 2.81 , 2.83 , Cur , 0.02 21 , 0 , , 22 3.35 , 2.81 , New , 0.54 21 , 6 , , 22 3.34 , 2.81 , New , 0.53 24 , 0 , , 23 2.77 , 2.84 , Cur , 0.07 24 , 7 , , 23 2.78 , 2.8 , Cur , 0.02 22 , 0 , , 23 3.34 , 2.79 , New , 0.55 22 , 7 , , 23 3.32 , 2.79 , New , 0.53 25 , 0 , , 24 2.77 , 2.8 , Cur , 0.03 23 , 0 , , 24 3.29 , 2.79 , New , 0.5 26 , 0 , , 25 2.73 , 2.78 , Cur , 0.05 26 , 1 , , 25 2.75 , 2.79 , Cur , 0.04 24 , 0 , , 25 3.27 , 2.79 , New , 0.48 24 , 1 , , 25 3.27 , 2.77 , New , 0.5 27 , 0 , , 26 2.72 , 2.78 , Cur , 0.06 27 , 2 , , 26 2.75 , 2.76 , Cur , 0.01 25 , 0 , , 26 3.29 , 2.73 , New , 0.56 25 , 2 , , 26 3.3 , 2.76 , New , 0.54 28 , 0 , , 27 2.75 , 2.79 , Cur , 0.04 28 , 3 , , 27 2.77 , 2.77 , Eq , 0.0 26 , 0 , , 27 3.28 , 2.78 , New , 0.5 26 , 3 , , 27 3.29 , 2.78 , New , 0.51 29 , 0 , , 28 2.74 , 2.76 , Cur , 0.02 29 , 4 , , 28 2.74 , 2.77 , Cur , 0.03 27 , 0 , , 28 3.3 , 2.76 , New , 0.54 27 , 4 , , 28 3.3 , 2.74 , New , 0.56 30 , 0 , , 29 2.72 , 2.76 , Cur , 0.04 30 , 5 , , 29 2.74 , 2.75 , Cur , 0.01 28 , 0 , , 29 3.25 , 2.73 , New , 0.52 28 , 5 , , 29 3.3 , 2.73 , New , 0.57 31 , 0 , , 30 2.73 , 2.77 , Cur , 0.04 31 , 6 , , 30 2.74 , 2.76 , Cur , 0.02 29 , 0 , , 30 3.25 , 2.73 , New , 0.52 29 , 6 , , 30 3.26 , 2.74 , New , 0.52 32 , 0 , , 31 2.73 , 2.74 , Cur , 0.01 32 , 7 , , 31 2.73 , 2.75 , Cur , 0.02 30 , 0 , , 31 3.24 , 2.72 , New , 0.52 30 , 7 , , 31 3.24 , 2.72 , New , 0.52 For memchr-avx2 the improvements are more modest though again near universal. The improvement is most significant for medium sizes and small sizes with pos > size. For small sizes with pos < size and large sizes the two implementations perform roughly the same for large sizes. Results For Tigerlake memchr-avx2 size , algn , Pos , Cur T , New T , Win , Dif 2048 , 0 , , 32 6.15 , 6.27 , Cur , 0.12 256 , 1 , , 64 6.21 , 6.03 , New , 0.18 2048 , 0 , , 64 6.07 , 5.95 , New , 0.12 256 , 2 , , 64 6.01 , 5.8 , New , 0.21 2048 , 0 , , 128 7.05 , 6.55 , New , 0.5 256 , 3 , , 64 6.14 , 5.83 , New , 0.31 2048 , 0 , , 256 11.78 , 11.78 , Eq , 0.0 256 , 4 , , 64 6.1 , 5.85 , New , 0.25 2048 , 0 , , 512 16.32 , 15.96 , New , 0.36 256 , 5 , , 64 6.1 , 5.77 , New , 0.33 2048 , 0 , , 1024 25.38 , 25.18 , New , 0.2 256 , 6 , , 64 6.08 , 5.88 , New , 0.2 2048 , 0 , , 2048 38.56 , 38.32 , New , 0.24 256 , 7 , , 64 5.93 , 5.68 , New , 0.25 192 , 1 , , 32 5.49 , 5.3 , New , 0.19 256 , 1 , , 32 5.5 , 5.28 , New , 0.22 512 , 1 , , 32 5.48 , 5.32 , New , 0.16 192 , 2 , , 64 6.1 , 5.73 , New , 0.37 512 , 2 , , 64 5.88 , 5.72 , New , 0.16 192 , 3 , , 96 6.31 , 5.93 , New , 0.38 256 , 3 , , 96 6.32 , 5.93 , New , 0.39 512 , 3 , , 96 6.2 , 5.94 , New , 0.26 192 , 4 , , 128 6.65 , 6.4 , New , 0.25 256 , 4 , , 128 6.6 , 6.37 , New , 0.23 512 , 4 , , 128 6.74 , 6.33 , New , 0.41 192 , 5 , , 160 7.78 , 7.4 , New , 0.38 256 , 5 , , 160 7.18 , 7.4 , Cur , 0.22 512 , 5 , , 160 9.81 , 9.44 , New , 0.37 192 , 6 , , 192 9.12 , 7.77 , New , 1.35 256 , 6 , , 192 7.97 , 7.66 , New , 0.31 512 , 6 , , 192 10.14 , 9.95 , New , 0.19 192 , 7 , , 224 8.96 , 7.78 , New , 1.18 256 , 7 , , 224 8.52 , 8.23 , New , 0.29 512 , 7 , , 224 10.33 , 9.98 , New , 0.35 2 , 0 , , 1 3.61 , 3.6 , New , 0.01 2 , 1 , , 1 3.6 , 3.6 , Eq , 0.0 0 , 0 , , 1 3.02 , 3.0 , New , 0.02 0 , 1 , , 1 3.0 , 3.0 , Eq , 0.0 3 , 0 , , 2 3.6 , 3.6 , Eq , 0.0 3 , 2 , , 2 3.61 , 3.6 , New , 0.01 1 , 0 , , 2 4.82 , 3.6 , New , 1.22 1 , 2 , , 2 4.81 , 3.6 , New , 1.21 4 , 0 , , 3 3.61 , 3.61 , Eq , 0.0 4 , 3 , , 3 3.62 , 3.61 , New , 0.01 2 , 0 , , 3 4.82 , 3.62 , New , 1.2 2 , 3 , , 3 4.83 , 3.63 , New , 1.2 5 , 0 , , 4 3.63 , 3.64 , Cur , 0.01 5 , 4 , , 4 3.63 , 3.62 , New , 0.01 3 , 0 , , 4 4.84 , 3.62 , New , 1.22 3 , 4 , , 4 4.84 , 3.64 , New , 1.2 6 , 0 , , 5 3.66 , 3.64 , New , 0.02 6 , 5 , , 5 3.65 , 3.62 , New , 0.03 4 , 0 , , 5 4.83 , 3.63 , New , 1.2 4 , 5 , , 5 4.85 , 3.64 , New , 1.21 7 , 0 , , 6 3.76 , 3.79 , Cur , 0.03 7 , 6 , , 6 3.76 , 3.72 , New , 0.04 5 , 0 , , 6 4.84 , 3.62 , New , 1.22 5 , 6 , , 6 4.85 , 3.64 , New , 1.21 8 , 0 , , 7 3.64 , 3.65 , Cur , 0.01 8 , 7 , , 7 3.65 , 3.65 , Eq , 0.0 6 , 0 , , 7 4.88 , 3.64 , New , 1.24 6 , 7 , , 7 4.87 , 3.65 , New , 1.22 9 , 0 , , 8 3.66 , 3.66 , Eq , 0.0 7 , 0 , , 8 4.89 , 3.66 , New , 1.23 10 , 0 , , 9 3.67 , 3.67 , Eq , 0.0 10 , 1 , , 9 3.67 , 3.67 , Eq , 0.0 8 , 0 , , 9 4.9 , 3.67 , New , 1.23 8 , 1 , , 9 4.9 , 3.67 , New , 1.23 11 , 0 , , 10 3.68 , 3.67 , New , 0.01 11 , 2 , , 10 3.69 , 3.67 , New , 0.02 9 , 0 , , 10 4.9 , 3.67 , New , 1.23 9 , 2 , , 10 4.9 , 3.67 , New , 1.23 12 , 0 , , 11 3.71 , 3.68 , New , 0.03 12 , 3 , , 11 3.71 , 3.67 , New , 0.04 10 , 0 , , 11 4.9 , 3.67 , New , 1.23 10 , 3 , , 11 4.9 , 3.67 , New , 1.23 13 , 0 , , 12 4.24 , 4.23 , New , 0.01 13 , 4 , , 12 4.23 , 4.23 , Eq , 0.0 11 , 0 , , 12 4.9 , 3.7 , New , 1.2 11 , 4 , , 12 4.9 , 3.73 , New , 1.17 14 , 0 , , 13 3.99 , 4.01 , Cur , 0.02 14 , 5 , , 13 3.98 , 3.98 , Eq , 0.0 12 , 0 , , 13 4.9 , 3.69 , New , 1.21 12 , 5 , , 13 4.9 , 3.69 , New , 1.21 15 , 0 , , 14 3.99 , 3.97 , New , 0.02 15 , 6 , , 14 4.0 , 4.0 , Eq , 0.0 13 , 0 , , 14 4.9 , 3.67 , New , 1.23 13 , 6 , , 14 4.9 , 3.67 , New , 1.23 16 , 0 , , 15 3.99 , 4.02 , Cur , 0.03 16 , 7 , , 15 4.01 , 3.96 , New , 0.05 14 , 0 , , 15 4.93 , 3.67 , New , 1.26 14 , 7 , , 15 4.92 , 3.67 , New , 1.25 17 , 0 , , 16 4.04 , 3.99 , New , 0.05 15 , 0 , , 16 5.42 , 4.22 , New , 1.2 18 , 0 , , 17 4.01 , 3.97 , New , 0.04 18 , 1 , , 17 3.99 , 3.98 , New , 0.01 16 , 0 , , 17 5.22 , 3.98 , New , 1.24 16 , 1 , , 17 5.19 , 3.98 , New , 1.21 19 , 0 , , 18 4.0 , 3.99 , New , 0.01 19 , 2 , , 18 4.03 , 3.97 , New , 0.06 17 , 0 , , 18 5.18 , 3.99 , New , 1.19 17 , 2 , , 18 5.18 , 3.98 , New , 1.2 20 , 0 , , 19 4.02 , 3.98 , New , 0.04 20 , 3 , , 19 4.0 , 3.98 , New , 0.02 18 , 0 , , 19 5.19 , 3.97 , New , 1.22 18 , 3 , , 19 5.21 , 3.98 , New , 1.23 21 , 0 , , 20 3.98 , 4.0 , Cur , 0.02 21 , 4 , , 20 4.0 , 4.0 , Eq , 0.0 19 , 0 , , 20 5.19 , 3.99 , New , 1.2 19 , 4 , , 20 5.17 , 3.99 , New , 1.18 22 , 0 , , 21 4.03 , 3.98 , New , 0.05 22 , 5 , , 21 4.01 , 3.95 , New , 0.06 20 , 0 , , 21 5.19 , 4.0 , New , 1.19 20 , 5 , , 21 5.21 , 3.99 , New , 1.22 23 , 0 , , 22 4.06 , 3.97 , New , 0.09 23 , 6 , , 22 4.02 , 3.98 , New , 0.04 21 , 0 , , 22 5.2 , 4.02 , New , 1.18 21 , 6 , , 22 5.22 , 4.0 , New , 1.22 24 , 0 , , 23 4.15 , 3.98 , New , 0.17 24 , 7 , , 23 4.0 , 4.01 , Cur , 0.01 22 , 0 , , 23 5.28 , 4.0 , New , 1.28 22 , 7 , , 23 5.22 , 3.99 , New , 1.23 25 , 0 , , 24 4.1 , 4.04 , New , 0.06 23 , 0 , , 24 5.23 , 4.04 , New , 1.19 26 , 0 , , 25 4.1 , 4.06 , New , 0.04 26 , 1 , , 25 4.07 , 3.99 , New , 0.08 24 , 0 , , 25 5.26 , 4.02 , New , 1.24 24 , 1 , , 25 5.21 , 4.0 , New , 1.21 27 , 0 , , 26 4.17 , 4.03 , New , 0.14 27 , 2 , , 26 4.09 , 4.03 , New , 0.06 25 , 0 , , 26 5.29 , 4.1 , New , 1.19 25 , 2 , , 26 5.25 , 4.0 , New , 1.25 28 , 0 , , 27 4.06 , 4.1 , Cur , 0.04 28 , 3 , , 27 4.09 , 4.04 , New , 0.05 26 , 0 , , 27 5.26 , 4.04 , New , 1.22 26 , 3 , , 27 5.28 , 4.01 , New , 1.27 29 , 0 , , 28 4.07 , 4.02 , New , 0.05 29 , 4 , , 28 4.07 , 4.05 , New , 0.02 27 , 0 , , 28 5.25 , 4.02 , New , 1.23 27 , 4 , , 28 5.25 , 4.03 , New , 1.22 30 , 0 , , 29 4.14 , 4.06 , New , 0.08 30 , 5 , , 29 4.08 , 4.04 , New , 0.04 28 , 0 , , 29 5.26 , 4.07 , New , 1.19 28 , 5 , , 29 5.28 , 4.04 , New , 1.24 31 , 0 , , 30 4.09 , 4.08 , New , 0.01 31 , 6 , , 30 4.1 , 4.08 , New , 0.02 29 , 0 , , 30 5.28 , 4.05 , New , 1.23 29 , 6 , , 30 5.24 , 4.07 , New , 1.17 32 , 0 , , 31 4.1 , 4.13 , Cur , 0.03 32 , 7 , , 31 4.16 , 4.09 , New , 0.07 30 , 0 , , 31 5.31 , 4.09 , New , 1.22 30 , 7 , , 31 5.28 , 4.08 , New , 1.2 Results For Icelake memchr-avx2 size , algn , Pos , Cur T , New T , Win , Dif 2048 , 0 , , 32 5.74 , 5.08 , New , 0.66 256 , 1 , , 64 5.16 , 4.93 , New , 0.23 2048 , 0 , , 64 4.86 , 4.69 , New , 0.17 256 , 2 , , 64 4.78 , 4.7 , New , 0.08 2048 , 0 , , 128 5.64 , 5.0 , New , 0.64 256 , 3 , , 64 4.64 , 4.59 , New , 0.05 2048 , 0 , , 256 9.07 , 9.17 , Cur , 0.1 256 , 4 , , 64 4.7 , 4.6 , New , 0.1 2048 , 0 , , 512 12.56 , 12.33 , New , 0.23 256 , 5 , , 64 4.72 , 4.61 , New , 0.11 2048 , 0 , , 1024 19.36 , 19.49 , Cur , 0.13 256 , 6 , , 64 4.82 , 4.69 , New , 0.13 2048 , 0 , , 2048 29.99 , 30.53 , Cur , 0.54 256 , 7 , , 64 4.9 , 4.85 , New , 0.05 192 , 1 , , 32 4.89 , 4.45 , New , 0.44 256 , 1 , , 32 4.93 , 4.44 , New , 0.49 512 , 1 , , 32 4.97 , 4.45 , New , 0.52 192 , 2 , , 64 5.04 , 4.65 , New , 0.39 512 , 2 , , 64 4.75 , 4.66 , New , 0.09 192 , 3 , , 96 5.14 , 4.66 , New , 0.48 256 , 3 , , 96 5.12 , 4.66 , New , 0.46 512 , 3 , , 96 5.13 , 4.62 , New , 0.51 192 , 4 , , 128 5.65 , 4.95 , New , 0.7 256 , 4 , , 128 5.63 , 4.95 , New , 0.68 512 , 4 , , 128 5.68 , 4.96 , New , 0.72 192 , 5 , , 160 6.1 , 5.84 , New , 0.26 256 , 5 , , 160 5.58 , 5.84 , Cur , 0.26 512 , 5 , , 160 7.95 , 7.74 , New , 0.21 192 , 6 , , 192 7.07 , 6.23 , New , 0.84 256 , 6 , , 192 6.34 , 6.09 , New , 0.25 512 , 6 , , 192 8.17 , 8.13 , New , 0.04 192 , 7 , , 224 7.06 , 6.23 , New , 0.83 256 , 7 , , 224 6.76 , 6.65 , New , 0.11 512 , 7 , , 224 8.29 , 8.08 , New , 0.21 2 , 0 , , 1 3.0 , 3.04 , Cur , 0.04 2 , 1 , , 1 3.06 , 3.07 , Cur , 0.01 0 , 0 , , 1 2.57 , 2.59 , Cur , 0.02 0 , 1 , , 1 2.6 , 2.61 , Cur , 0.01 3 , 0 , , 2 3.15 , 3.17 , Cur , 0.02 3 , 2 , , 2 3.19 , 3.21 , Cur , 0.02 1 , 0 , , 2 4.32 , 3.25 , New , 1.07 1 , 2 , , 2 4.36 , 3.31 , New , 1.05 4 , 0 , , 3 3.5 , 3.52 , Cur , 0.02 4 , 3 , , 3 3.52 , 3.54 , Cur , 0.02 2 , 0 , , 3 4.51 , 3.43 , New , 1.08 2 , 3 , , 3 4.56 , 3.47 , New , 1.09 5 , 0 , , 4 3.61 , 3.65 , Cur , 0.04 5 , 4 , , 4 3.63 , 3.67 , Cur , 0.04 3 , 0 , , 4 4.64 , 3.51 , New , 1.13 3 , 4 , , 4 4.7 , 3.51 , New , 1.19 6 , 0 , , 5 3.66 , 3.68 , Cur , 0.02 6 , 5 , , 5 3.69 , 3.65 , New , 0.04 4 , 0 , , 5 4.7 , 3.49 , New , 1.21 4 , 5 , , 5 4.58 , 3.48 , New , 1.1 7 , 0 , , 6 3.6 , 3.65 , Cur , 0.05 7 , 6 , , 6 3.59 , 3.64 , Cur , 0.05 5 , 0 , , 6 4.74 , 3.65 , New , 1.09 5 , 6 , , 6 4.73 , 3.64 , New , 1.09 8 , 0 , , 7 3.6 , 3.61 , Cur , 0.01 8 , 7 , , 7 3.6 , 3.61 , Cur , 0.01 6 , 0 , , 7 4.73 , 3.6 , New , 1.13 6 , 7 , , 7 4.73 , 3.62 , New , 1.11 9 , 0 , , 8 3.59 , 3.62 , Cur , 0.03 7 , 0 , , 8 4.72 , 3.64 , New , 1.08 10 , 0 , , 9 3.57 , 3.62 , Cur , 0.05 10 , 1 , , 9 3.56 , 3.61 , Cur , 0.05 8 , 0 , , 9 4.69 , 3.63 , New , 1.06 8 , 1 , , 9 4.71 , 3.61 , New , 1.1 11 , 0 , , 10 3.58 , 3.62 , Cur , 0.04 11 , 2 , , 10 3.59 , 3.63 , Cur , 0.04 9 , 0 , , 10 4.72 , 3.61 , New , 1.11 9 , 2 , , 10 4.7 , 3.61 , New , 1.09 12 , 0 , , 11 3.58 , 3.63 , Cur , 0.05 12 , 3 , , 11 3.58 , 3.62 , Cur , 0.04 10 , 0 , , 11 4.7 , 3.6 , New , 1.1 10 , 3 , , 11 4.73 , 3.64 , New , 1.09 13 , 0 , , 12 3.6 , 3.6 , Eq , 0.0 13 , 4 , , 12 3.57 , 3.62 , Cur , 0.05 11 , 0 , , 12 4.73 , 3.62 , New , 1.11 11 , 4 , , 12 4.79 , 3.61 , New , 1.18 14 , 0 , , 13 3.61 , 3.62 , Cur , 0.01 14 , 5 , , 13 3.59 , 3.59 , Eq , 0.0 12 , 0 , , 13 4.7 , 3.61 , New , 1.09 12 , 5 , , 13 4.75 , 3.58 , New , 1.17 15 , 0 , , 14 3.58 , 3.62 , Cur , 0.04 15 , 6 , , 14 3.59 , 3.62 , Cur , 0.03 13 , 0 , , 14 4.68 , 3.6 , New , 1.08 13 , 6 , , 14 4.68 , 3.63 , New , 1.05 16 , 0 , , 15 3.57 , 3.6 , Cur , 0.03 16 , 7 , , 15 3.55 , 3.59 , Cur , 0.04 14 , 0 , , 15 4.69 , 3.61 , New , 1.08 14 , 7 , , 15 4.69 , 3.61 , New , 1.08 17 , 0 , , 16 3.56 , 3.61 , Cur , 0.05 15 , 0 , , 16 4.71 , 3.58 , New , 1.13 18 , 0 , , 17 3.57 , 3.65 , Cur , 0.08 18 , 1 , , 17 3.58 , 3.59 , Cur , 0.01 16 , 0 , , 17 4.7 , 3.58 , New , 1.12 16 , 1 , , 17 4.68 , 3.59 , New , 1.09 19 , 0 , , 18 3.51 , 3.58 , Cur , 0.07 19 , 2 , , 18 3.55 , 3.58 , Cur , 0.03 17 , 0 , , 18 4.69 , 3.61 , New , 1.08 17 , 2 , , 18 4.68 , 3.61 , New , 1.07 20 , 0 , , 19 3.57 , 3.6 , Cur , 0.03 20 , 3 , , 19 3.59 , 3.59 , Eq , 0.0 18 , 0 , , 19 4.68 , 3.59 , New , 1.09 18 , 3 , , 19 4.67 , 3.57 , New , 1.1 21 , 0 , , 20 3.61 , 3.58 , New , 0.03 21 , 4 , , 20 3.62 , 3.6 , New , 0.02 19 , 0 , , 20 4.74 , 3.57 , New , 1.17 19 , 4 , , 20 4.69 , 3.7 , New , 0.99 22 , 0 , , 21 3.57 , 3.64 , Cur , 0.07 22 , 5 , , 21 3.55 , 3.6 , Cur , 0.05 20 , 0 , , 21 4.72 , 3.55 , New , 1.17 20 , 5 , , 21 4.66 , 3.55 , New , 1.11 23 , 0 , , 22 3.56 , 3.56 , Eq , 0.0 23 , 6 , , 22 3.54 , 3.56 , Cur , 0.02 21 , 0 , , 22 4.65 , 3.53 , New , 1.12 21 , 6 , , 22 4.62 , 3.56 , New , 1.06 24 , 0 , , 23 3.5 , 3.54 , Cur , 0.04 24 , 7 , , 23 3.52 , 3.53 , Cur , 0.01 22 , 0 , , 23 4.61 , 3.51 , New , 1.1 22 , 7 , , 23 4.6 , 3.51 , New , 1.09 25 , 0 , , 24 3.5 , 3.53 , Cur , 0.03 23 , 0 , , 24 4.54 , 3.5 , New , 1.04 26 , 0 , , 25 3.47 , 3.49 , Cur , 0.02 26 , 1 , , 25 3.46 , 3.51 , Cur , 0.05 24 , 0 , , 25 4.53 , 3.51 , New , 1.02 24 , 1 , , 25 4.51 , 3.51 , New , 1.0 27 , 0 , , 26 3.44 , 3.51 , Cur , 0.07 27 , 2 , , 26 3.51 , 3.52 , Cur , 0.01 25 , 0 , , 26 4.56 , 3.46 , New , 1.1 25 , 2 , , 26 4.55 , 3.47 , New , 1.08 28 , 0 , , 27 3.47 , 3.5 , Cur , 0.03 28 , 3 , , 27 3.48 , 3.47 , New , 0.01 26 , 0 , , 27 4.52 , 3.44 , New , 1.08 26 , 3 , , 27 4.55 , 3.46 , New , 1.09 29 , 0 , , 28 3.45 , 3.49 , Cur , 0.04 29 , 4 , , 28 3.5 , 3.5 , Eq , 0.0 27 , 0 , , 28 4.56 , 3.49 , New , 1.07 27 , 4 , , 28 4.5 , 3.49 , New , 1.01 30 , 0 , , 29 3.44 , 3.48 , Cur , 0.04 30 , 5 , , 29 3.46 , 3.47 , Cur , 0.01 28 , 0 , , 29 4.49 , 3.43 , New , 1.06 28 , 5 , , 29 4.57 , 3.45 , New , 1.12 31 , 0 , , 30 3.48 , 3.48 , Eq , 0.0 31 , 6 , , 30 3.46 , 3.49 , Cur , 0.03 29 , 0 , , 30 4.49 , 3.44 , New , 1.05 29 , 6 , , 30 4.53 , 3.44 , New , 1.09 32 , 0 , , 31 3.44 , 3.45 , Cur , 0.01 32 , 7 , , 31 3.46 , 3.51 , Cur , 0.05 30 , 0 , , 31 4.48 , 3.42 , New , 1.06 30 , 7 , , 31 4.48 , 3.44 , New , 1.04 Results For Skylake memchr-avx2 size , algn , Pos , Cur T , New T , Win , Dif 2048 , 0 , , 32 6.61 , 5.4 , New , 1.21 256 , 1 , , 64 6.52 , 5.68 , New , 0.84 2048 , 0 , , 64 6.03 , 5.47 , New , 0.56 256 , 2 , , 64 6.07 , 5.42 , New , 0.65 2048 , 0 , , 128 7.01 , 5.83 , New , 1.18 256 , 3 , , 64 6.24 , 5.68 , New , 0.56 2048 , 0 , , 256 11.03 , 9.86 , New , 1.17 256 , 4 , , 64 6.17 , 5.49 , New , 0.68 2048 , 0 , , 512 14.11 , 13.41 , New , 0.7 256 , 5 , , 64 6.03 , 5.45 , New , 0.58 2048 , 0 , , 1024 19.82 , 19.92 , Cur , 0.1 256 , 6 , , 64 6.14 , 5.7 , New , 0.44 2048 , 0 , , 2048 30.9 , 30.59 , New , 0.31 256 , 7 , , 64 6.05 , 5.64 , New , 0.41 192 , 1 , , 32 5.6 , 4.89 , New , 0.71 256 , 1 , , 32 5.59 , 5.07 , New , 0.52 512 , 1 , , 32 5.58 , 4.93 , New , 0.65 192 , 2 , , 64 6.14 , 5.46 , New , 0.68 512 , 2 , , 64 5.95 , 5.38 , New , 0.57 192 , 3 , , 96 6.6 , 5.74 , New , 0.86 256 , 3 , , 96 6.48 , 5.37 , New , 1.11 512 , 3 , , 96 6.56 , 5.44 , New , 1.12 192 , 4 , , 128 7.04 , 6.02 , New , 1.02 256 , 4 , , 128 6.96 , 5.89 , New , 1.07 512 , 4 , , 128 6.97 , 5.99 , New , 0.98 192 , 5 , , 160 8.49 , 7.07 , New , 1.42 256 , 5 , , 160 8.1 , 6.96 , New , 1.14 512 , 5 , , 160 10.48 , 9.14 , New , 1.34 192 , 6 , , 192 8.46 , 8.52 , Cur , 0.06 256 , 6 , , 192 8.53 , 7.58 , New , 0.95 512 , 6 , , 192 10.88 , 9.06 , New , 1.82 192 , 7 , , 224 8.59 , 8.35 , New , 0.24 256 , 7 , , 224 8.86 , 7.91 , New , 0.95 512 , 7 , , 224 10.89 , 8.98 , New , 1.91 2 , 0 , , 1 4.28 , 3.62 , New , 0.66 2 , 1 , , 1 4.32 , 3.75 , New , 0.57 0 , 0 , , 1 3.76 , 3.24 , New , 0.52 0 , 1 , , 1 3.7 , 3.19 , New , 0.51 3 , 0 , , 2 4.16 , 3.67 , New , 0.49 3 , 2 , , 2 4.21 , 3.68 , New , 0.53 1 , 0 , , 2 4.25 , 3.74 , New , 0.51 1 , 2 , , 2 4.4 , 3.82 , New , 0.58 4 , 0 , , 3 4.43 , 3.88 , New , 0.55 4 , 3 , , 3 4.34 , 3.8 , New , 0.54 2 , 0 , , 3 4.33 , 3.79 , New , 0.54 2 , 3 , , 3 4.37 , 3.84 , New , 0.53 5 , 0 , , 4 4.45 , 3.87 , New , 0.58 5 , 4 , , 4 4.41 , 3.84 , New , 0.57 3 , 0 , , 4 4.34 , 3.83 , New , 0.51 3 , 4 , , 4 4.35 , 3.82 , New , 0.53 6 , 0 , , 5 4.41 , 3.88 , New , 0.53 6 , 5 , , 5 4.41 , 3.88 , New , 0.53 4 , 0 , , 5 4.35 , 3.84 , New , 0.51 4 , 5 , , 5 4.37 , 3.85 , New , 0.52 7 , 0 , , 6 4.4 , 3.84 , New , 0.56 7 , 6 , , 6 4.39 , 3.83 , New , 0.56 5 , 0 , , 6 4.37 , 3.85 , New , 0.52 5 , 6 , , 6 4.4 , 3.86 , New , 0.54 8 , 0 , , 7 4.39 , 3.88 , New , 0.51 8 , 7 , , 7 4.4 , 3.83 , New , 0.57 6 , 0 , , 7 4.39 , 3.85 , New , 0.54 6 , 7 , , 7 4.38 , 3.87 , New , 0.51 9 , 0 , , 8 4.47 , 3.96 , New , 0.51 7 , 0 , , 8 4.37 , 3.85 , New , 0.52 10 , 0 , , 9 4.61 , 4.08 , New , 0.53 10 , 1 , , 9 4.61 , 4.09 , New , 0.52 8 , 0 , , 9 4.37 , 3.85 , New , 0.52 8 , 1 , , 9 4.37 , 3.85 , New , 0.52 11 , 0 , , 10 4.68 , 4.06 , New , 0.62 11 , 2 , , 10 4.56 , 4.1 , New , 0.46 9 , 0 , , 10 4.36 , 3.83 , New , 0.53 9 , 2 , , 10 4.37 , 3.83 , New , 0.54 12 , 0 , , 11 4.62 , 4.05 , New , 0.57 12 , 3 , , 11 4.63 , 4.06 , New , 0.57 10 , 0 , , 11 4.38 , 3.86 , New , 0.52 10 , 3 , , 11 4.41 , 3.86 , New , 0.55 13 , 0 , , 12 4.57 , 4.08 , New , 0.49 13 , 4 , , 12 4.59 , 4.12 , New , 0.47 11 , 0 , , 12 4.45 , 4.0 , New , 0.45 11 , 4 , , 12 4.51 , 4.04 , New , 0.47 14 , 0 , , 13 4.64 , 4.16 , New , 0.48 14 , 5 , , 13 4.67 , 4.1 , New , 0.57 12 , 0 , , 13 4.58 , 4.08 , New , 0.5 12 , 5 , , 13 4.6 , 4.1 , New , 0.5 15 , 0 , , 14 4.61 , 4.05 , New , 0.56 15 , 6 , , 14 4.59 , 4.06 , New , 0.53 13 , 0 , , 14 4.57 , 4.06 , New , 0.51 13 , 6 , , 14 4.57 , 4.05 , New , 0.52 16 , 0 , , 15 4.62 , 4.05 , New , 0.57 16 , 7 , , 15 4.63 , 4.06 , New , 0.57 14 , 0 , , 15 4.61 , 4.06 , New , 0.55 14 , 7 , , 15 4.59 , 4.05 , New , 0.54 17 , 0 , , 16 4.58 , 4.08 , New , 0.5 15 , 0 , , 16 4.64 , 4.06 , New , 0.58 18 , 0 , , 17 4.56 , 4.17 , New , 0.39 18 , 1 , , 17 4.59 , 4.09 , New , 0.5 16 , 0 , , 17 4.59 , 4.07 , New , 0.52 16 , 1 , , 17 4.58 , 4.04 , New , 0.54 19 , 0 , , 18 4.61 , 4.05 , New , 0.56 19 , 2 , , 18 4.6 , 4.08 , New , 0.52 17 , 0 , , 18 4.64 , 4.11 , New , 0.53 17 , 2 , , 18 4.56 , 4.13 , New , 0.43 20 , 0 , , 19 4.77 , 4.3 , New , 0.47 20 , 3 , , 19 4.6 , 4.14 , New , 0.46 18 , 0 , , 19 4.72 , 4.02 , New , 0.7 18 , 3 , , 19 4.53 , 4.01 , New , 0.52 21 , 0 , , 20 4.66 , 4.26 , New , 0.4 21 , 4 , , 20 4.74 , 4.07 , New , 0.67 19 , 0 , , 20 4.62 , 4.12 , New , 0.5 19 , 4 , , 20 4.57 , 4.04 , New , 0.53 22 , 0 , , 21 4.61 , 4.13 , New , 0.48 22 , 5 , , 21 4.64 , 4.08 , New , 0.56 20 , 0 , , 21 4.49 , 4.01 , New , 0.48 20 , 5 , , 21 4.58 , 4.06 , New , 0.52 23 , 0 , , 22 4.62 , 4.13 , New , 0.49 23 , 6 , , 22 4.72 , 4.27 , New , 0.45 21 , 0 , , 22 4.65 , 3.97 , New , 0.68 21 , 6 , , 22 4.5 , 4.02 , New , 0.48 24 , 0 , , 23 4.78 , 4.07 , New , 0.71 24 , 7 , , 23 4.67 , 4.23 , New , 0.44 22 , 0 , , 23 4.49 , 3.99 , New , 0.5 22 , 7 , , 23 4.56 , 4.03 , New , 0.53 25 , 0 , , 24 4.6 , 4.15 , New , 0.45 23 , 0 , , 24 4.57 , 4.06 , New , 0.51 26 , 0 , , 25 4.54 , 4.14 , New , 0.4 26 , 1 , , 25 4.72 , 4.1 , New , 0.62 24 , 0 , , 25 4.52 , 4.13 , New , 0.39 24 , 1 , , 25 4.55 , 4.0 , New , 0.55 27 , 0 , , 26 4.51 , 4.06 , New , 0.45 27 , 2 , , 26 4.53 , 4.16 , New , 0.37 25 , 0 , , 26 4.59 , 4.09 , New , 0.5 25 , 2 , , 26 4.55 , 4.01 , New , 0.54 28 , 0 , , 27 4.59 , 3.99 , New , 0.6 28 , 3 , , 27 4.57 , 3.95 , New , 0.62 26 , 0 , , 27 4.55 , 4.15 , New , 0.4 26 , 3 , , 27 4.57 , 3.99 , New , 0.58 29 , 0 , , 28 4.41 , 4.03 , New , 0.38 29 , 4 , , 28 4.59 , 4.02 , New , 0.57 27 , 0 , , 28 4.63 , 4.08 , New , 0.55 27 , 4 , , 28 4.44 , 4.02 , New , 0.42 30 , 0 , , 29 4.53 , 3.93 , New , 0.6 30 , 5 , , 29 4.55 , 3.88 , New , 0.67 28 , 0 , , 29 4.49 , 3.9 , New , 0.59 28 , 5 , , 29 4.44 , 3.94 , New , 0.5 31 , 0 , , 30 4.41 , 3.85 , New , 0.56 31 , 6 , , 30 4.48 , 3.86 , New , 0.62 29 , 0 , , 30 4.55 , 3.94 , New , 0.61 29 , 6 , , 30 4.32 , 3.95 , New , 0.37 32 , 0 , , 31 4.36 , 3.91 , New , 0.45 32 , 7 , , 31 4.37 , 3.89 , New , 0.48 30 , 0 , , 31 4.65 , 3.9 , New , 0.75 30 , 7 , , 31 4.42 , 3.93 , New , 0.49 sysdeps/x86_64/multiarch/memchr-evex.S | 580 +++++++++++++++---------- 1 file changed, 349 insertions(+), 231 deletions(-) diff --git a/sysdeps/x86_64/multiarch/memchr-evex.S b/sysdeps/x86_64/multiarch/memchr-evex.S index 6dd5d67b90..65c16ef8a4 100644 --- a/sysdeps/x86_64/multiarch/memchr-evex.S +++ b/sysdeps/x86_64/multiarch/memchr-evex.S @@ -26,14 +26,28 @@ # ifdef USE_AS_WMEMCHR # define VPBROADCAST vpbroadcastd -# define VPCMP vpcmpd -# define SHIFT_REG r8d +# define VPMINU vpminud +# define VPCMP vpcmpd +# define VPCMPEQ vpcmpeqd +# define CHAR_SIZE 4 # else # define VPBROADCAST vpbroadcastb -# define VPCMP vpcmpb -# define SHIFT_REG ecx +# define VPMINU vpminub +# define VPCMP vpcmpb +# define VPCMPEQ vpcmpeqb +# define CHAR_SIZE 1 # endif +# ifdef USE_AS_RAWMEMCHR +# define RAW_PTR_REG rcx +# define ALGN_PTR_REG rdi +# else +# define RAW_PTR_REG rdi +# define ALGN_PTR_REG rcx +# endif + +#define XZERO xmm23 +#define YZERO ymm23 # define XMMMATCH xmm16 # define YMMMATCH ymm16 # define YMM1 ymm17 @@ -44,18 +58,16 @@ # define YMM6 ymm22 # define VEC_SIZE 32 +# define CHAR_PER_VEC (VEC_SIZE / CHAR_SIZE) +# define PAGE_SIZE 4096 .section .text.evex,"ax",@progbits -ENTRY (MEMCHR) +ENTRY(MEMCHR) # ifndef USE_AS_RAWMEMCHR /* Check for zero length. */ test %RDX_LP, %RDX_LP jz L(zero) -# endif - movl %edi, %ecx -# ifdef USE_AS_WMEMCHR - shl $2, %RDX_LP -# else + # ifdef __ILP32__ /* Clear the upper 32 bits. */ movl %edx, %edx @@ -63,319 +75,425 @@ ENTRY (MEMCHR) # endif /* Broadcast CHAR to YMMMATCH. */ VPBROADCAST %esi, %YMMMATCH - /* Check if we may cross page boundary with one vector load. */ - andl $(2 * VEC_SIZE - 1), %ecx - cmpl $VEC_SIZE, %ecx - ja L(cros_page_boundary) + /* Check if we may cross page boundary with one + vector load. */ + movl %edi, %eax + andl $(PAGE_SIZE - 1), %eax + cmpl $(PAGE_SIZE - VEC_SIZE), %eax + ja L(cross_page_boundary) /* Check the first VEC_SIZE bytes. */ - VPCMP $0, (%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - testl %eax, %eax - + VPCMP $0, (%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax # ifndef USE_AS_RAWMEMCHR - jnz L(first_vec_x0_check) - /* Adjust length and check the end of data. */ - subq $VEC_SIZE, %rdx - jbe L(zero) + /* If length < CHAR_PER_VEC handle special. */ + cmpq $CHAR_PER_VEC, %rdx + jbe L(first_vec_x0) +# endif + testl %eax, %eax + jz L(aligned_more) + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (%rdi, %rax, CHAR_SIZE), %rax # else - jnz L(first_vec_x0) + addq %rdi, %rax # endif - - /* Align data for aligned loads in the loop. */ - addq $VEC_SIZE, %rdi - andl $(VEC_SIZE - 1), %ecx - andq $-VEC_SIZE, %rdi + ret # ifndef USE_AS_RAWMEMCHR - /* Adjust length. */ - addq %rcx, %rdx - - subq $(VEC_SIZE * 4), %rdx - jbe L(last_4x_vec_or_less) -# endif - jmp L(more_4x_vec) +L(zero): + xorl %eax, %eax + ret + .p2align 5 +L(first_vec_x0): + /* Check if first match was before length. */ + tzcntl %eax, %eax + xorl %ecx, %ecx + cmpl %eax, %edx + leaq (%rdi, %rax, CHAR_SIZE), %rax + cmovle %rcx, %rax + ret +# else + /* NB: first_vec_x0 is 17 bytes which will leave + cross_page_boundary (which is relatively cold) close + enough to ideal alignment. So only realign + L(cross_page_boundary) if rawmemchr. */ .p2align 4 -L(cros_page_boundary): - andl $(VEC_SIZE - 1), %ecx +# endif +L(cross_page_boundary): + /* Save pointer before aligning as its original + value is necessary for computer return address if byte is + found or adjusting length if it is not and this is + memchr. */ + movq %rdi, %rcx + /* Align data to VEC_SIZE. ALGN_PTR_REG is rcx + for memchr and rdi for rawmemchr. */ + andq $-VEC_SIZE, %ALGN_PTR_REG + VPCMP $0, (%ALGN_PTR_REG), %YMMMATCH, %k0 + kmovd %k0, %r8d # ifdef USE_AS_WMEMCHR - /* NB: Divide shift count by 4 since each bit in K1 represent 4 - bytes. */ - movl %ecx, %SHIFT_REG - sarl $2, %SHIFT_REG + /* NB: Divide shift count by 4 since each bit in + K0 represent 4 bytes. */ + sarl $2, %eax +# endif +# ifndef USE_AS_RAWMEMCHR + movl $(PAGE_SIZE / CHAR_SIZE), %esi + subl %eax, %esi # endif - andq $-VEC_SIZE, %rdi - VPCMP $0, (%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - /* Remove the leading bytes. */ - sarxl %SHIFT_REG, %eax, %eax - testl %eax, %eax - jz L(aligned_more) - tzcntl %eax, %eax # ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - sall $2, %eax + andl $(CHAR_PER_VEC - 1), %eax # endif + /* Remove the leading bytes. */ + sarxl %eax, %r8d, %eax # ifndef USE_AS_RAWMEMCHR /* Check the end of data. */ - cmpq %rax, %rdx - jbe L(zero) + cmpq %rsi, %rdx + jbe L(first_vec_x0) +# endif + testl %eax, %eax + jz L(cross_page_continue) + tzcntl %eax, %eax +# ifdef USE_AS_WMEMCHR + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (%RAW_PTR_REG, %rax, CHAR_SIZE), %rax +# else + addq %RAW_PTR_REG, %rax # endif - addq %rdi, %rax - addq %rcx, %rax ret .p2align 4 -L(aligned_more): -# ifndef USE_AS_RAWMEMCHR - /* Calculate "rdx + rcx - VEC_SIZE" with "rdx - (VEC_SIZE - rcx)" - instead of "(rdx + rcx) - VEC_SIZE" to void possible addition - overflow. */ - negq %rcx - addq $VEC_SIZE, %rcx +L(first_vec_x1): + tzcntl %eax, %eax + leaq VEC_SIZE(%rdi, %rax, CHAR_SIZE), %rax + ret - /* Check the end of data. */ - subq %rcx, %rdx - jbe L(zero) -# endif + .p2align 4 +L(first_vec_x2): + tzcntl %eax, %eax + leaq (VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax + ret - addq $VEC_SIZE, %rdi + .p2align 4 +L(first_vec_x3): + tzcntl %eax, %eax + leaq (VEC_SIZE * 3)(%rdi, %rax, CHAR_SIZE), %rax + ret + + .p2align 4 +L(first_vec_x4): + tzcntl %eax, %eax + leaq (VEC_SIZE * 4)(%rdi, %rax, CHAR_SIZE), %rax + ret + + .p2align 5 +L(aligned_more): + /* Check the first 4 * VEC_SIZE. Only one + VEC_SIZE at a time since data is only aligned to + VEC_SIZE. */ # ifndef USE_AS_RAWMEMCHR - subq $(VEC_SIZE * 4), %rdx + /* Align data to VEC_SIZE. */ +L(cross_page_continue): + xorl %ecx, %ecx + subl %edi, %ecx + andq $-VEC_SIZE, %rdi + /* esi is for adjusting length to see if near the + end. */ + leal (VEC_SIZE * 5)(%rdi, %rcx), %esi +# ifdef USE_AS_WMEMCHR + /* NB: Divide bytes by 4 to get the wchar_t + count. */ + sarl $2, %esi +# endif +# else + andq $-VEC_SIZE, %rdi +L(cross_page_continue): +# endif + /* Load first VEC regardless. */ + VPCMP $0, (VEC_SIZE)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax +# ifndef USE_AS_RAWMEMCHR + /* Adjust length. If near end handle specially. + */ + subq %rsi, %rdx jbe L(last_4x_vec_or_less) # endif - -L(more_4x_vec): - /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time - since data is only aligned to VEC_SIZE. */ - VPCMP $0, (%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - testl %eax, %eax - jnz L(first_vec_x0) - - VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax testl %eax, %eax jnz L(first_vec_x1) - VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax testl %eax, %eax jnz L(first_vec_x2) - VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax testl %eax, %eax jnz L(first_vec_x3) - addq $(VEC_SIZE * 4), %rdi + VPCMP $0, (VEC_SIZE * 4)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(first_vec_x4) + # ifndef USE_AS_RAWMEMCHR - subq $(VEC_SIZE * 4), %rdx - jbe L(last_4x_vec_or_less) -# endif + /* Check if at last CHAR_PER_VEC * 4 length. */ + subq $(CHAR_PER_VEC * 4), %rdx + jbe L(last_4x_vec_or_less_cmpeq) + addq $VEC_SIZE, %rdi - /* Align data to 4 * VEC_SIZE. */ - movq %rdi, %rcx - andl $(4 * VEC_SIZE - 1), %ecx + /* Align data to VEC_SIZE * 4 for the loop and + readjust length. */ +# ifdef USE_AS_WMEMCHR + movl %edi, %ecx andq $-(4 * VEC_SIZE), %rdi - -# ifndef USE_AS_RAWMEMCHR - /* Adjust length. */ + andl $(VEC_SIZE * 4 - 1), %ecx + /* NB: Divide bytes by 4 to get the wchar_t + count. */ + sarl $2, %ecx addq %rcx, %rdx +# else + addq %rdi, %rdx + andq $-(4 * VEC_SIZE), %rdi + subq %rdi, %rdx +# endif +# else + addq $VEC_SIZE, %rdi + andq $-(4 * VEC_SIZE), %rdi # endif + vpxorq %XZERO, %XZERO, %XZERO + + /* Compare 4 * VEC at a time forward. */ .p2align 4 L(loop_4x_vec): - /* Compare 4 * VEC at a time forward. */ - VPCMP $0, (%rdi), %YMMMATCH, %k1 - VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k2 - kord %k1, %k2, %k5 - VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k3 - VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k4 - - kord %k3, %k4, %k6 - kortestd %k5, %k6 - jnz L(4x_vec_end) - - addq $(VEC_SIZE * 4), %rdi - + /* It would be possible to save some instructions + using 4x VPCMP but bottleneck on port 5 makes it not woth + it. */ + VPCMP $4, (VEC_SIZE * 4)(%rdi), %YMMMATCH, %k1 + /* xor will set bytes match esi to zero. */ + vpxorq (VEC_SIZE * 5)(%rdi), %YMMMATCH, %YMM2 + vpxorq (VEC_SIZE * 6)(%rdi), %YMMMATCH, %YMM3 + VPCMP $0, (VEC_SIZE * 7)(%rdi), %YMMMATCH, %k3 + /* Reduce VEC2 / VEC3 with min and VEC1 with zero + mask. */ + VPMINU %YMM2, %YMM3, %YMM3 {%k1} {z} + VPCMP $0, %YMM3, %YZERO, %k2 # ifdef USE_AS_RAWMEMCHR - jmp L(loop_4x_vec) + subq $-(VEC_SIZE * 4), %rdi + kortestd %k2, %k3 + jz L(loop_4x_vec) # else - subq $(VEC_SIZE * 4), %rdx - ja L(loop_4x_vec) + kortestd %k2, %k3 + jnz L(loop_4x_vec_end) -L(last_4x_vec_or_less): - /* Less than 4 * VEC and aligned to VEC_SIZE. */ - addl $(VEC_SIZE * 2), %edx - jle L(last_2x_vec) + subq $-(VEC_SIZE * 4), %rdi - VPCMP $0, (%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - testl %eax, %eax - jnz L(first_vec_x0) + subq $(CHAR_PER_VEC * 4), %rdx + ja L(loop_4x_vec) - VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax + /* Fall through into less than 4 remaining + vectors of length case. */ + VPCMP $0, (VEC_SIZE * 4)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + addq $(VEC_SIZE * 3), %rdi + .p2align 4 +L(last_4x_vec_or_less): + /* Check if first VEC contained match. */ testl %eax, %eax - jnz L(first_vec_x1) + jnz L(first_vec_x1_check) - VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - testl %eax, %eax + /* If remaining length > CHAR_PER_VEC * 2. */ + addl $(CHAR_PER_VEC * 2), %edx + jg L(last_4x_vec) - jnz L(first_vec_x2_check) - subl $VEC_SIZE, %edx - jle L(zero) +L(last_2x_vec): + /* If remaining length < CHAR_PER_VEC. */ + addl $CHAR_PER_VEC, %edx + jle L(zero_end) + + /* Check VEC2 and compare any match with + remaining length. */ + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + tzcntl %eax, %eax + cmpl %eax, %edx + jbe L(set_zero_end) + leaq (VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax +L(zero_end): + ret - VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax - testl %eax, %eax - jnz L(first_vec_x3_check) + .p2align 4 +L(first_vec_x1_check): + tzcntl %eax, %eax + /* Adjust length. */ + subl $-(CHAR_PER_VEC * 4), %edx + /* Check if match within remaining length. */ + cmpl %eax, %edx + jbe L(set_zero_end) + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq VEC_SIZE(%rdi, %rax, CHAR_SIZE), %rax + ret +L(set_zero_end): xorl %eax, %eax ret .p2align 4 -L(last_2x_vec): - addl $(VEC_SIZE * 2), %edx - VPCMP $0, (%rdi), %YMMMATCH, %k1 +L(loop_4x_vec_end): +# endif + /* rawmemchr will fall through into this if match + was found in loop. */ + + /* k1 has not of matches with VEC1. */ kmovd %k1, %eax - testl %eax, %eax +# ifdef USE_AS_WMEMCHR + subl $((1 << CHAR_PER_VEC) - 1), %eax +# else + incl %eax +# endif + jnz L(last_vec_x1_return) - jnz L(first_vec_x0_check) - subl $VEC_SIZE, %edx - jle L(zero) + VPCMP $0, %YMM2, %YZERO, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(last_vec_x2_return) - VPCMP $0, VEC_SIZE(%rdi), %YMMMATCH, %k1 - kmovd %k1, %eax + kmovd %k2, %eax testl %eax, %eax - jnz L(first_vec_x1_check) - xorl %eax, %eax - ret + jnz L(last_vec_x3_return) - .p2align 4 -L(first_vec_x0_check): + kmovd %k3, %eax tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - sall $2, %eax +# ifdef USE_AS_RAWMEMCHR + leaq (VEC_SIZE * 3)(%rdi, %rax, CHAR_SIZE), %rax +# else + leaq (VEC_SIZE * 7)(%rdi, %rax, CHAR_SIZE), %rax # endif - /* Check the end of data. */ - cmpq %rax, %rdx - jbe L(zero) - addq %rdi, %rax ret .p2align 4 -L(first_vec_x1_check): +L(last_vec_x1_return): tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - sall $2, %eax -# endif - /* Check the end of data. */ - cmpq %rax, %rdx - jbe L(zero) - addq $VEC_SIZE, %rax +# ifdef USE_AS_RAWMEMCHR +# ifdef USE_AS_WMEMCHR + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (%rdi, %rax, CHAR_SIZE), %rax +# else addq %rdi, %rax - ret - - .p2align 4 -L(first_vec_x2_check): - tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - sall $2, %eax +# endif +# else + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (VEC_SIZE * 4)(%rdi, %rax, CHAR_SIZE), %rax # endif - /* Check the end of data. */ - cmpq %rax, %rdx - jbe L(zero) - addq $(VEC_SIZE * 2), %rax - addq %rdi, %rax ret .p2align 4 -L(first_vec_x3_check): +L(last_vec_x2_return): tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - sall $2, %eax +# ifdef USE_AS_RAWMEMCHR + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq VEC_SIZE(%rdi, %rax, CHAR_SIZE), %rax +# else + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (VEC_SIZE * 5)(%rdi, %rax, CHAR_SIZE), %rax # endif - /* Check the end of data. */ - cmpq %rax, %rdx - jbe L(zero) - addq $(VEC_SIZE * 3), %rax - addq %rdi, %rax ret .p2align 4 -L(zero): - xorl %eax, %eax - ret -# endif - - .p2align 4 -L(first_vec_x0): +L(last_vec_x3_return): tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - leaq (%rdi, %rax, 4), %rax +# ifdef USE_AS_RAWMEMCHR + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax # else - addq %rdi, %rax + /* NB: Multiply bytes by CHAR_SIZE to get the + wchar_t count. */ + leaq (VEC_SIZE * 6)(%rdi, %rax, CHAR_SIZE), %rax # endif ret + +# ifndef USE_AS_RAWMEMCHR +L(last_4x_vec_or_less_cmpeq): + VPCMP $0, (VEC_SIZE * 5)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + subq $-(VEC_SIZE * 4), %rdi + /* Check first VEC regardless. */ + testl %eax, %eax + jnz L(first_vec_x1_check) + + /* If remaining length <= CHAR_PER_VEC * 2. */ + addl $(CHAR_PER_VEC * 2), %edx + jle L(last_2x_vec) + .p2align 4 -L(first_vec_x1): +L(last_4x_vec): + VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + testl %eax, %eax + jnz L(last_vec_x2) + + + VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + /* Create mask for possible matches within + remaining length. */ +# ifdef USE_AS_WMEMCHR + movl $((1 << (CHAR_PER_VEC * 2)) - 1), %ecx + bzhil %edx, %ecx, %ecx +# else + movq $-1, %rcx + bzhiq %rdx, %rcx, %rcx +# endif + /* Test matches in data against length match. */ + andl %ecx, %eax + jnz L(last_vec_x3) + + /* if remaining length <= CHAR_PER_VEC * 3 (Note + this is after remaining length was found to be > + CHAR_PER_VEC * 2. */ + subl $CHAR_PER_VEC, %edx + jbe L(zero_end2) + + + VPCMP $0, (VEC_SIZE * 4)(%rdi), %YMMMATCH, %k0 + kmovd %k0, %eax + /* Shift remaining length mask for last VEC. */ +# ifdef USE_AS_WMEMCHR + shrl $CHAR_PER_VEC, %ecx +# else + shrq $CHAR_PER_VEC, %rcx +# endif + andl %ecx, %eax + jz L(zero_end2) tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - leaq VEC_SIZE(%rdi, %rax, 4), %rax -# else - addq $VEC_SIZE, %rax - addq %rdi, %rax -# endif + leaq (VEC_SIZE * 4)(%rdi, %rax, CHAR_SIZE), %rax +L(zero_end2): ret - .p2align 4 -L(first_vec_x2): +L(last_vec_x2): tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - leaq (VEC_SIZE * 2)(%rdi, %rax, 4), %rax -# else - addq $(VEC_SIZE * 2), %rax - addq %rdi, %rax -# endif + leaq (VEC_SIZE * 2)(%rdi, %rax, CHAR_SIZE), %rax ret .p2align 4 -L(4x_vec_end): - kmovd %k1, %eax - testl %eax, %eax - jnz L(first_vec_x0) - kmovd %k2, %eax - testl %eax, %eax - jnz L(first_vec_x1) - kmovd %k3, %eax - testl %eax, %eax - jnz L(first_vec_x2) - kmovd %k4, %eax - testl %eax, %eax -L(first_vec_x3): +L(last_vec_x3): tzcntl %eax, %eax -# ifdef USE_AS_WMEMCHR - /* NB: Multiply wchar_t count by 4 to get the number of bytes. */ - leaq (VEC_SIZE * 3)(%rdi, %rax, 4), %rax -# else - addq $(VEC_SIZE * 3), %rax - addq %rdi, %rax -# endif + leaq (VEC_SIZE * 3)(%rdi, %rax, CHAR_SIZE), %rax ret +# endif -END (MEMCHR) +END(MEMCHR) #endif