From patchwork Thu Jul 21 13:05:04 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Adhemerval Zanella Netto X-Patchwork-Id: 592135 Delivered-To: patch@linaro.org Received: by 2002:a05:7000:b811:0:0:0:0 with SMTP id fc17csp444327mab; Thu, 21 Jul 2022 06:05:41 -0700 (PDT) X-Google-Smtp-Source: AGRyM1tQM/NTMjAMilIia4SaGC14E9hE80ND+oQdJdm51qdMnOpwbPxdNz5zMZfEP473uv0DqnzJ X-Received: by 2002:a17:906:9bdd:b0:72b:3cab:eade with SMTP id de29-20020a1709069bdd00b0072b3cabeademr40196993ejc.58.1658408741044; Thu, 21 Jul 2022 06:05:41 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1658408741; cv=none; d=google.com; s=arc-20160816; b=AlUTckafx+oQQhyaV8Af8XThsbTIyF/SswAWEywFXUqQ8WxSRi88c/RjmXwRUBZ6+5 arz5LtbiSBKbBR6fd171as4HEVSNJaUMSqJshFhrWZSRLAno4N1Jfb2Ly56aQ8WzbUpc d9aLaA12lnhLvKe0i1p9CW31rXArKFvxT0W9JRld2WMv3cNTxqbRiRbMk73zcsSlxmqN 3nlzu1A+BVgma7Fkhx5uKHxfj95N1yIUaUUGmyQkmK3fDlWOnBCEtYSmltccHy5kCZgm aoyGlbyeZ/PCbNf2phz3y32uzd5kaUko7tcQ6iiYTpMrYeZHHhsSxNN65+G0k7nPk+qA GIvg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:reply-to:from:list-subscribe:list-help:list-post :list-archive:list-unsubscribe:list-id:precedence :content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:to:dmarc-filter:delivered-to:dkim-signature :dkim-filter; bh=BkfMBESRWI0BLNu+XM46uyo7yetnqgE/h2FbK8p9XPk=; b=W6mxSW4lNWAASkWQcqeFFL9E4RoXb+rcc6mw171awWFJG4WgzYzTeyCIrAa2CqSz/n dxR89aq+KQnFVCSBpYesREZgWiAEXyWa5bIS8jroc2A9TrQnxVqBjNoDuwXFlr8rwVr1 LyDC7Bz3MXyMSs6G0x+ADWpfQmbRh/usknLNl3Awsw6lKdGb5WhUVayopzcPhtLZZTCi AXyO6k46uwH6837zB01UeDa+i9+SeYdtoOUsc50q5exNKwlkPiP3oDd1woBFrvU8LU5C RkMrnNcufu30C11YoyKBh4QSgcz7ifhCvwXn4ksfKHYft8kLHqaMCBYmHK004Tb8HZIK vJTA== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@sourceware.org header.s=default header.b=JHWmXQgq; spf=pass (google.com: domain of libc-alpha-bounces+patch=linaro.org@sourceware.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) smtp.mailfrom="libc-alpha-bounces+patch=linaro.org@sourceware.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=sourceware.org Return-Path: Received: from sourceware.org (server2.sourceware.org. [2620:52:3:1:0:246e:9693:128c]) by mx.google.com with ESMTPS id gn21-20020a1709070d1500b007123c7ecf90si2750676ejc.476.2022.07.21.06.05.40 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 21 Jul 2022 06:05:41 -0700 (PDT) Received-SPF: pass (google.com: domain of libc-alpha-bounces+patch=linaro.org@sourceware.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) client-ip=2620:52:3:1:0:246e:9693:128c; Authentication-Results: mx.google.com; dkim=pass header.i=@sourceware.org header.s=default header.b=JHWmXQgq; spf=pass (google.com: domain of libc-alpha-bounces+patch=linaro.org@sourceware.org designates 2620:52:3:1:0:246e:9693:128c as permitted sender) smtp.mailfrom="libc-alpha-bounces+patch=linaro.org@sourceware.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=sourceware.org Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 97C14385626D for ; Thu, 21 Jul 2022 13:05:39 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 97C14385626D DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=sourceware.org; s=default; t=1658408739; bh=BkfMBESRWI0BLNu+XM46uyo7yetnqgE/h2FbK8p9XPk=; 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=JHWmXQgqXL0ayCxN94apGPkUAydBAaPc0ivZmoynnFcaaTO19JUMdAw5tR+rADTcO GHJAcolVuq/yhbguTBRNcZCA9BQeiPEQXJ7+1fxwAm21esF+d7iQv9V+pJokfxCQOh NuZfiV8PzJ0q0bOR04iv+DVRh1fhs+vvDwv9Jt5k= X-Original-To: libc-alpha@sourceware.org Delivered-To: libc-alpha@sourceware.org Received: from mail-oi1-x22a.google.com (mail-oi1-x22a.google.com [IPv6:2607:f8b0:4864:20::22a]) by sourceware.org (Postfix) with ESMTPS id 512EE385AE63 for ; Thu, 21 Jul 2022 13:05:28 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 512EE385AE63 Received: by mail-oi1-x22a.google.com with SMTP id j70so1834313oih.10 for ; Thu, 21 Jul 2022 06:05:28 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=BkfMBESRWI0BLNu+XM46uyo7yetnqgE/h2FbK8p9XPk=; b=E9YbIYnui9UdZEl2GABPaWTcKunMv3ZVibrbaYK9GbrJxwjf89oR2PypmM8bSB3YRS yGe0lQ6AzHtPdSaQe66GqS4X5oqiZ0UlD9VtdJ1+nk/XpiJsZvtR6KOAnc9aAUwWiai4 xTQ1vWHSBtxaPO3ySNYxczph5kWZ+ngwxnmXDZQtjCS7RL0hlcVrUIVa6kzSSn3rbd45 MGutqEkms9uKzj+e1L5F4kPE9wf7T77TY2L9sE0DZAHbFeE4tLsLoq6n71UhmSR1kPxA GKuxHMAv7A/5xGl1L6M8FsLgK9DFP8h6UJK3oCnIsFeiMLzGAT2sWGKaaTopGKqXwngc +Akg== X-Gm-Message-State: AJIora+1WJYpDGuIWL1Yw3ufVXqM30gsfyEyFk4SYdikcPEYpP/zEMFI OwGN/ekExeaNt9Zydw1hBMcMaLK0cpZCDA== X-Received: by 2002:a05:6808:e87:b0:32e:4638:302c with SMTP id k7-20020a0568080e8700b0032e4638302cmr4387332oil.89.1658408726976; Thu, 21 Jul 2022 06:05:26 -0700 (PDT) Received: from mandiga.. ([2804:431:c7cb:8ded:c7ae:e809:e8fb:6cb9]) by smtp.gmail.com with ESMTPSA id a32-20020a056870a1a000b0010d997ffe7asm767553oaf.37.2022.07.21.06.05.25 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 21 Jul 2022 06:05:26 -0700 (PDT) To: libc-alpha@sourceware.org, Florian Weimer , Carlos O'Donell Subject: [PATCH v11 6/9] x86: Add AVX2 optimized chacha20 Date: Thu, 21 Jul 2022 10:05:04 -0300 Message-Id: <20220721130507.3017393-7-adhemerval.zanella@linaro.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20220721130507.3017393-1-adhemerval.zanella@linaro.org> References: <20220721130507.3017393-1-adhemerval.zanella@linaro.org> MIME-Version: 1.0 X-Spam-Status: No, score=-11.4 required=5.0 tests=BAYES_00, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_NUMSUBJECT, KAM_SHORT, RCVD_IN_DNSWL_NONE, SCC_5_SHORT_WORD_LINES, SPF_HELO_NONE, SPF_PASS, TXREP autolearn=ham autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) 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: Adhemerval Zanella via Libc-alpha From: Adhemerval Zanella Netto Reply-To: Adhemerval Zanella Errors-To: libc-alpha-bounces+patch=linaro.org@sourceware.org Sender: "Libc-alpha" From: Adhemerval Zanella Netto It adds vectorized ChaCha20 implementation based on libgcrypt cipher/chacha20-amd64-avx2.S. It is used only if AVX2 is supported and enabled by the architecture. As for generic implementation, the last step that XOR with the input is omited. The final state register clearing is also omitted. On a Ryzen 9 5900X it shows the following improvements (using formatted bench-arc4random data): SSE MB/s ----------------------------------------------- arc4random [single-thread] 704.25 arc4random_buf(16) [single-thread] 1018.17 arc4random_buf(32) [single-thread] 1315.27 arc4random_buf(48) [single-thread] 1449.36 arc4random_buf(64) [single-thread] 1511.16 arc4random_buf(80) [single-thread] 1539.48 arc4random_buf(96) [single-thread] 1571.06 arc4random_buf(112) [single-thread] 1596.16 arc4random_buf(128) [single-thread] 1613.48 ----------------------------------------------- AVX2 MB/s ----------------------------------------------- arc4random [single-thread] 922.61 arc4random_buf(16) [single-thread] 1478.70 arc4random_buf(32) [single-thread] 2241.80 arc4random_buf(48) [single-thread] 2681.28 arc4random_buf(64) [single-thread] 2913.43 arc4random_buf(80) [single-thread] 3009.73 arc4random_buf(96) [single-thread] 3141.16 arc4random_buf(112) [single-thread] 3254.46 arc4random_buf(128) [single-thread] 3305.02 ----------------------------------------------- Checked on x86_64-linux-gnu. --- LICENSES | 5 +- sysdeps/x86_64/Makefile | 1 + sysdeps/x86_64/chacha20-amd64-avx2.S | 328 +++++++++++++++++++++++++++ sysdeps/x86_64/chacha20-amd64-sse2.S | 5 + sysdeps/x86_64/chacha20_arch.h | 27 ++- 5 files changed, 359 insertions(+), 7 deletions(-) create mode 100644 sysdeps/x86_64/chacha20-amd64-avx2.S diff --git a/LICENSES b/LICENSES index f0117ef79f..80168d0b1b 100644 --- a/LICENSES +++ b/LICENSES @@ -390,8 +390,9 @@ Copyright 2001 by Stephen L. Moshier License along with this library; if not, see . */ -sysdeps/aarch64/chacha20-aarch64.S and sysdeps/x86_64/chacha20-amd64-sse2.S -imports code from libgcrypt, with the following notices: +sysdeps/aarch64/chacha20-aarch64.S, sysdeps/x86_64/chacha20-amd64-sse2.S, +and sysdeps/x86_64/chacha20-amd64-avx2.S imports code from libgcrypt, +with the following notices: Copyright (C) 2017-2019 Jussi Kivilinna diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile index 3acd9755a5..1178475d75 100644 --- a/sysdeps/x86_64/Makefile +++ b/sysdeps/x86_64/Makefile @@ -8,6 +8,7 @@ endif ifeq ($(subdir),stdlib) sysdep_routines += \ chacha20-amd64-sse2 \ + chacha20-amd64-avx2 \ # sysdep_routines endif diff --git a/sysdeps/x86_64/chacha20-amd64-avx2.S b/sysdeps/x86_64/chacha20-amd64-avx2.S new file mode 100644 index 0000000000..aefd1cdbd0 --- /dev/null +++ b/sysdeps/x86_64/chacha20-amd64-avx2.S @@ -0,0 +1,328 @@ +/* Optimized AVX2 implementation of ChaCha20 cipher. + Copyright (C) 2022 Free Software Foundation, Inc. + + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +/* chacha20-amd64-avx2.S - AVX2 implementation of ChaCha20 cipher + + Copyright (C) 2017-2019 Jussi Kivilinna + + This file is part of Libgcrypt. + + Libgcrypt is free software; you can redistribute it and/or modify + it under the terms of the GNU Lesser General Public License as + published by the Free Software Foundation; either version 2.1 of + the License, or (at your option) any later version. + + Libgcrypt is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with this program; if not, see . +*/ + +/* Based on D. J. Bernstein reference implementation at + http://cr.yp.to/chacha.html: + + chacha-regs.c version 20080118 + D. J. Bernstein + Public domain. */ + +#include + +#ifdef PIC +# define rRIP (%rip) +#else +# define rRIP +#endif + +/* register macros */ +#define INPUT %rdi +#define DST %rsi +#define SRC %rdx +#define NBLKS %rcx +#define ROUND %eax + +/* stack structure */ +#define STACK_VEC_X12 (32) +#define STACK_VEC_X13 (32 + STACK_VEC_X12) +#define STACK_TMP (32 + STACK_VEC_X13) +#define STACK_TMP1 (32 + STACK_TMP) + +#define STACK_MAX (32 + STACK_TMP1) + +/* vector registers */ +#define X0 %ymm0 +#define X1 %ymm1 +#define X2 %ymm2 +#define X3 %ymm3 +#define X4 %ymm4 +#define X5 %ymm5 +#define X6 %ymm6 +#define X7 %ymm7 +#define X8 %ymm8 +#define X9 %ymm9 +#define X10 %ymm10 +#define X11 %ymm11 +#define X12 %ymm12 +#define X13 %ymm13 +#define X14 %ymm14 +#define X15 %ymm15 + +#define X0h %xmm0 +#define X1h %xmm1 +#define X2h %xmm2 +#define X3h %xmm3 +#define X4h %xmm4 +#define X5h %xmm5 +#define X6h %xmm6 +#define X7h %xmm7 +#define X8h %xmm8 +#define X9h %xmm9 +#define X10h %xmm10 +#define X11h %xmm11 +#define X12h %xmm12 +#define X13h %xmm13 +#define X14h %xmm14 +#define X15h %xmm15 + +/********************************************************************** + helper macros + **********************************************************************/ + +/* 4x4 32-bit integer matrix transpose */ +#define transpose_4x4(x0,x1,x2,x3,t1,t2) \ + vpunpckhdq x1, x0, t2; \ + vpunpckldq x1, x0, x0; \ + \ + vpunpckldq x3, x2, t1; \ + vpunpckhdq x3, x2, x2; \ + \ + vpunpckhqdq t1, x0, x1; \ + vpunpcklqdq t1, x0, x0; \ + \ + vpunpckhqdq x2, t2, x3; \ + vpunpcklqdq x2, t2, x2; + +/* 2x2 128-bit matrix transpose */ +#define transpose_16byte_2x2(x0,x1,t1) \ + vmovdqa x0, t1; \ + vperm2i128 $0x20, x1, x0, x0; \ + vperm2i128 $0x31, x1, t1, x1; + +/********************************************************************** + 8-way chacha20 + **********************************************************************/ + +#define ROTATE2(v1,v2,c,tmp) \ + vpsrld $(32 - (c)), v1, tmp; \ + vpslld $(c), v1, v1; \ + vpaddb tmp, v1, v1; \ + vpsrld $(32 - (c)), v2, tmp; \ + vpslld $(c), v2, v2; \ + vpaddb tmp, v2, v2; + +#define ROTATE_SHUF_2(v1,v2,shuf) \ + vpshufb shuf, v1, v1; \ + vpshufb shuf, v2, v2; + +#define XOR(ds,s) \ + vpxor s, ds, ds; + +#define PLUS(ds,s) \ + vpaddd s, ds, ds; + +#define QUARTERROUND2(a1,b1,c1,d1,a2,b2,c2,d2,ign,tmp1,\ + interleave_op1,interleave_op2,\ + interleave_op3,interleave_op4) \ + vbroadcasti128 .Lshuf_rol16 rRIP, tmp1; \ + interleave_op1; \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE_SHUF_2(d1, d2, tmp1); \ + interleave_op2; \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 12, tmp1); \ + vbroadcasti128 .Lshuf_rol8 rRIP, tmp1; \ + interleave_op3; \ + PLUS(a1,b1); PLUS(a2,b2); XOR(d1,a1); XOR(d2,a2); \ + ROTATE_SHUF_2(d1, d2, tmp1); \ + interleave_op4; \ + PLUS(c1,d1); PLUS(c2,d2); XOR(b1,c1); XOR(b2,c2); \ + ROTATE2(b1, b2, 7, tmp1); + + .section .text.avx2, "ax", @progbits + .align 32 +chacha20_data: +L(shuf_rol16): + .byte 2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13 +L(shuf_rol8): + .byte 3,0,1,2,7,4,5,6,11,8,9,10,15,12,13,14 +L(inc_counter): + .byte 0,1,2,3,4,5,6,7 +L(unsigned_cmp): + .long 0x80000000 + + .hidden __chacha20_avx2_blocks8 +ENTRY (__chacha20_avx2_blocks8) + /* input: + * %rdi: input + * %rsi: dst + * %rdx: src + * %rcx: nblks (multiple of 8) + */ + vzeroupper; + + pushq %rbp; + cfi_adjust_cfa_offset(8); + cfi_rel_offset(rbp, 0) + movq %rsp, %rbp; + cfi_def_cfa_register(rbp); + + subq $STACK_MAX, %rsp; + andq $~31, %rsp; + +L(loop8): + mov $20, ROUND; + + /* Construct counter vectors X12 and X13 */ + vpmovzxbd L(inc_counter) rRIP, X0; + vpbroadcastd L(unsigned_cmp) rRIP, X2; + vpbroadcastd (12 * 4)(INPUT), X12; + vpbroadcastd (13 * 4)(INPUT), X13; + vpaddd X0, X12, X12; + vpxor X2, X0, X0; + vpxor X2, X12, X1; + vpcmpgtd X1, X0, X0; + vpsubd X0, X13, X13; + vmovdqa X12, (STACK_VEC_X12)(%rsp); + vmovdqa X13, (STACK_VEC_X13)(%rsp); + + /* Load vectors */ + vpbroadcastd (0 * 4)(INPUT), X0; + vpbroadcastd (1 * 4)(INPUT), X1; + vpbroadcastd (2 * 4)(INPUT), X2; + vpbroadcastd (3 * 4)(INPUT), X3; + vpbroadcastd (4 * 4)(INPUT), X4; + vpbroadcastd (5 * 4)(INPUT), X5; + vpbroadcastd (6 * 4)(INPUT), X6; + vpbroadcastd (7 * 4)(INPUT), X7; + vpbroadcastd (8 * 4)(INPUT), X8; + vpbroadcastd (9 * 4)(INPUT), X9; + vpbroadcastd (10 * 4)(INPUT), X10; + vpbroadcastd (11 * 4)(INPUT), X11; + vpbroadcastd (14 * 4)(INPUT), X14; + vpbroadcastd (15 * 4)(INPUT), X15; + vmovdqa X15, (STACK_TMP)(%rsp); + +L(round2): + QUARTERROUND2(X0, X4, X8, X12, X1, X5, X9, X13, tmp:=,X15,,,,) + vmovdqa (STACK_TMP)(%rsp), X15; + vmovdqa X8, (STACK_TMP)(%rsp); + QUARTERROUND2(X2, X6, X10, X14, X3, X7, X11, X15, tmp:=,X8,,,,) + QUARTERROUND2(X0, X5, X10, X15, X1, X6, X11, X12, tmp:=,X8,,,,) + vmovdqa (STACK_TMP)(%rsp), X8; + vmovdqa X15, (STACK_TMP)(%rsp); + QUARTERROUND2(X2, X7, X8, X13, X3, X4, X9, X14, tmp:=,X15,,,,) + sub $2, ROUND; + jnz L(round2); + + vmovdqa X8, (STACK_TMP1)(%rsp); + + /* tmp := X15 */ + vpbroadcastd (0 * 4)(INPUT), X15; + PLUS(X0, X15); + vpbroadcastd (1 * 4)(INPUT), X15; + PLUS(X1, X15); + vpbroadcastd (2 * 4)(INPUT), X15; + PLUS(X2, X15); + vpbroadcastd (3 * 4)(INPUT), X15; + PLUS(X3, X15); + vpbroadcastd (4 * 4)(INPUT), X15; + PLUS(X4, X15); + vpbroadcastd (5 * 4)(INPUT), X15; + PLUS(X5, X15); + vpbroadcastd (6 * 4)(INPUT), X15; + PLUS(X6, X15); + vpbroadcastd (7 * 4)(INPUT), X15; + PLUS(X7, X15); + transpose_4x4(X0, X1, X2, X3, X8, X15); + transpose_4x4(X4, X5, X6, X7, X8, X15); + vmovdqa (STACK_TMP1)(%rsp), X8; + transpose_16byte_2x2(X0, X4, X15); + transpose_16byte_2x2(X1, X5, X15); + transpose_16byte_2x2(X2, X6, X15); + transpose_16byte_2x2(X3, X7, X15); + vmovdqa (STACK_TMP)(%rsp), X15; + vmovdqu X0, (64 * 0 + 16 * 0)(DST) + vmovdqu X1, (64 * 1 + 16 * 0)(DST) + vpbroadcastd (8 * 4)(INPUT), X0; + PLUS(X8, X0); + vpbroadcastd (9 * 4)(INPUT), X0; + PLUS(X9, X0); + vpbroadcastd (10 * 4)(INPUT), X0; + PLUS(X10, X0); + vpbroadcastd (11 * 4)(INPUT), X0; + PLUS(X11, X0); + vmovdqa (STACK_VEC_X12)(%rsp), X0; + PLUS(X12, X0); + vmovdqa (STACK_VEC_X13)(%rsp), X0; + PLUS(X13, X0); + vpbroadcastd (14 * 4)(INPUT), X0; + PLUS(X14, X0); + vpbroadcastd (15 * 4)(INPUT), X0; + PLUS(X15, X0); + vmovdqu X2, (64 * 2 + 16 * 0)(DST) + vmovdqu X3, (64 * 3 + 16 * 0)(DST) + + /* Update counter */ + addq $8, (12 * 4)(INPUT); + + transpose_4x4(X8, X9, X10, X11, X0, X1); + transpose_4x4(X12, X13, X14, X15, X0, X1); + vmovdqu X4, (64 * 4 + 16 * 0)(DST) + vmovdqu X5, (64 * 5 + 16 * 0)(DST) + transpose_16byte_2x2(X8, X12, X0); + transpose_16byte_2x2(X9, X13, X0); + transpose_16byte_2x2(X10, X14, X0); + transpose_16byte_2x2(X11, X15, X0); + vmovdqu X6, (64 * 6 + 16 * 0)(DST) + vmovdqu X7, (64 * 7 + 16 * 0)(DST) + vmovdqu X8, (64 * 0 + 16 * 2)(DST) + vmovdqu X9, (64 * 1 + 16 * 2)(DST) + vmovdqu X10, (64 * 2 + 16 * 2)(DST) + vmovdqu X11, (64 * 3 + 16 * 2)(DST) + vmovdqu X12, (64 * 4 + 16 * 2)(DST) + vmovdqu X13, (64 * 5 + 16 * 2)(DST) + vmovdqu X14, (64 * 6 + 16 * 2)(DST) + vmovdqu X15, (64 * 7 + 16 * 2)(DST) + + sub $8, NBLKS; + lea (8 * 64)(DST), DST; + lea (8 * 64)(SRC), SRC; + jnz L(loop8); + + vzeroupper; + + /* eax zeroed by round loop. */ + leave; + cfi_adjust_cfa_offset(-8) + cfi_def_cfa_register(%rsp); + ret; + int3; +END(__chacha20_avx2_blocks8) diff --git a/sysdeps/x86_64/chacha20-amd64-sse2.S b/sysdeps/x86_64/chacha20-amd64-sse2.S index 1178f06306..351a1109c6 100644 --- a/sysdeps/x86_64/chacha20-amd64-sse2.S +++ b/sysdeps/x86_64/chacha20-amd64-sse2.S @@ -44,6 +44,9 @@ Public domain. */ #include +#include + +#if MINIMUM_X86_ISA_LEVEL <= 2 #ifdef PIC # define rRIP (%rip) @@ -304,3 +307,5 @@ L(round2_4): cfi_def_cfa_register(%rsp); ret_spec_stop; END (__chacha20_sse2_blocks4) + +#endif /* if MINIMUM_X86_ISA_LEVEL <= 2 */ diff --git a/sysdeps/x86_64/chacha20_arch.h b/sysdeps/x86_64/chacha20_arch.h index 60dee08045..6f3784e392 100644 --- a/sysdeps/x86_64/chacha20_arch.h +++ b/sysdeps/x86_64/chacha20_arch.h @@ -16,6 +16,7 @@ License along with the GNU C Library; if not, see . */ +#include #include #include #include @@ -23,16 +24,32 @@ unsigned int __chacha20_sse2_blocks4 (uint32_t *state, uint8_t *dst, const uint8_t *src, size_t nblks) attribute_hidden; +unsigned int __chacha20_avx2_blocks8 (uint32_t *state, uint8_t *dst, + const uint8_t *src, size_t nblks) + attribute_hidden; static inline void chacha20_crypt (uint32_t *state, uint8_t *dst, const uint8_t *src, size_t bytes) { - _Static_assert (CHACHA20_BUFSIZE % 4 == 0, - "CHACHA20_BUFSIZE not multiple of 4"); - _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 4, - "CHACHA20_BUFSIZE <= CHACHA20_BLOCK_SIZE * 4"); + _Static_assert (CHACHA20_BUFSIZE % 4 == 0 && CHACHA20_BUFSIZE % 8 == 0, + "CHACHA20_BUFSIZE not multiple of 4 or 8"); + _Static_assert (CHACHA20_BUFSIZE >= CHACHA20_BLOCK_SIZE * 8, + "CHACHA20_BUFSIZE < CHACHA20_BLOCK_SIZE * 8"); - __chacha20_sse2_blocks4 (state, dst, src, +#if MINIMUM_X86_ISA_LEVEL > 2 + __chacha20_avx2_blocks8 (state, dst, src, CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); +#else + const struct cpu_features* cpu_features = __get_cpu_features (); + + /* AVX2 version uses vzeroupper, so disable it if RTM is enabled. */ + if (X86_ISA_CPU_FEATURE_USABLE_P (cpu_features, AVX2) + && X86_ISA_CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER, !)) + __chacha20_avx2_blocks8 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); + else + __chacha20_sse2_blocks4 (state, dst, src, + CHACHA20_BUFSIZE / CHACHA20_BLOCK_SIZE); +#endif }