diff options
author | XiaokangQian <xiaokang.qian@arm.com> | 2021-06-09 06:35:46 +0000 |
---|---|---|
committer | Tomas Mraz <tomas@openssl.org> | 2022-11-11 10:02:44 +0100 |
commit | 34ca334e5de6837f2c6bc0b0b0df28bdd237e4d7 (patch) | |
tree | 7e1cc3c3cc26f34e53ac4c4a1f957bd5e892fb65 /crypto/modes/asm | |
parent | a2bdca6fe666c3a0a13e7f0a51626715608f8597 (diff) |
Optimize AES-GCM for uarchs with unroll and new instructions
Increase the block numbers to 8 for every iteration. Increase the hash
table capacity. Make use of EOR3 instruction to improve the performance.
This can improve performance 25-40% on out-of-order microarchitectures
with a large number of fast execution units, such as Neoverse V1. We also
see 20-30% performance improvements on other architectures such as the M1.
Assembly code reviewd by Tom Cosgrove (ARM).
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Paul Dale <pauli@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/15916)
(cherry picked from commit 954f45ba4c504570206ff5bed811e512cf92dc8e)
Diffstat (limited to 'crypto/modes/asm')
-rw-r--r-- | crypto/modes/asm/aes-gcm-armv8-unroll8_64.pl | 7369 | ||||
-rw-r--r-- | crypto/modes/asm/ghashv8-armx.pl | 105 |
2 files changed, 7466 insertions, 8 deletions
diff --git a/crypto/modes/asm/aes-gcm-armv8-unroll8_64.pl b/crypto/modes/asm/aes-gcm-armv8-unroll8_64.pl new file mode 100644 index 0000000000..1aaad663d7 --- /dev/null +++ b/crypto/modes/asm/aes-gcm-armv8-unroll8_64.pl @@ -0,0 +1,7369 @@ +#! /usr/bin/env perl +# Copyright 2020-2021 The OpenSSL Project Authors. All Rights Reserved. +# +# Licensed under the Apache License 2.0 (the "License"). You may not use +# this file except in compliance with the License. You can obtain a copy +# in the file LICENSE in the source distribution or at +# https://www.openssl.org/source/license.html + +# +#======================================================================== +# Written by Xiaokang Qian <xiaokang.qian@arm.com> for the OpenSSL project, +# derived from https://github.com/ARM-software/AArch64cryptolib, original +# author Samuel Lee <Samuel.Lee@arm.com>. The module is, however, dual +# licensed under OpenSSL and SPDX BSD-3-Clause licenses depending on where you +# obtain it. +#======================================================================== +# +# Approach - We want to reload constants as we have plenty of spare ASIMD slots around crypto units for loading +# Unroll x8 in main loop, main loop to act on 8 16B blocks per iteration, and then do modulo of the accumulated +# intermediate hashesfrom the 8 blocks. +# +# ____________________________________________________ +# | | +# | PRE | +# |____________________________________________________| +# | | | | +# | CTR block 8k+13| AES block 8k+8 | GHASH block 8k+0 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+14| AES block 8k+9 | GHASH block 8k+1 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+15| AES block 8k+10| GHASH block 8k+2 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+16| AES block 8k+11| GHASH block 8k+3 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+17| AES block 8k+12| GHASH block 8k+4 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+18| AES block 8k+13| GHASH block 8k+5 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+19| AES block 8k+14| GHASH block 8k+6 | +# |________________|________________|__________________| +# | | | | +# | CTR block 8k+20| AES block 8k+15| GHASH block 8k+7 | +# |________________|____(mostly)____|__________________| +# | | +# | MODULO | +# |____________________________________________________| +# +# PRE: +# Ensure previous generated intermediate hash is aligned and merged with result for GHASH 4k+0 +# EXT low_acc, low_acc, low_acc, #8 +# EOR res_curr (8k+0), res_curr (4k+0), low_acc +# +# CTR block: +# Increment and byte reverse counter in scalar registers and transfer to SIMD registers +# REV ctr32, rev_ctr32 +# ORR ctr64, constctr96_top32, ctr32, LSL #32 +# INS ctr_next.d[0], constctr96_bottom64 // Keeping this in scalar registers to free up space in SIMD RF +# INS ctr_next.d[1], ctr64X +# ADD rev_ctr32, #1 +# +# AES block: +# Do AES encryption/decryption on CTR block X and EOR it with input block X. Take 256 bytes key below for example. +# Doing small trick here of loading input in scalar registers, EORing with last key and then transferring +# Given we are very constrained in our ASIMD registers this is quite important +# +# Encrypt: +# LDR input_low, [ input_ptr ], #8 +# LDR input_high, [ input_ptr ], #8 +# EOR input_low, k14_low +# EOR input_high, k14_high +# INS res_curr.d[0], input_low +# INS res_curr.d[1], input_high +# AESE ctr_curr, k0; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k1; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k2; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k3; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k4; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k5; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k6; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k7; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k8; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k9; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k10; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k11; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k12; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k13 +# EOR res_curr, res_curr, ctr_curr +# ST1 { res_curr.16b }, [ output_ptr ], #16 +# +# Decrypt: +# AESE ctr_curr, k0; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k1; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k2; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k3; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k4; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k5; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k6; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k7; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k8; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k9; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k10; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k11; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k12; AESMC ctr_curr, ctr_curr +# AESE ctr_curr, k13 +# LDR res_curr, [ input_ptr ], #16 +# EOR res_curr, res_curr, ctr_curr +# MOV output_low, res_curr.d[0] +# MOV output_high, res_curr.d[1] +# EOR output_low, k14_low +# EOR output_high, k14_high +# STP output_low, output_high, [ output_ptr ], #16 + +# GHASH block X: +# Do 128b karatsuba polynomial multiplication on block +# We only have 64b->128b polynomial multipliers, naively that means we need to do 4 64b multiplies to generate a 128b +# +# multiplication: +# Pmull(A,B) == (Pmull(Ah,Bh)<<128 | Pmull(Al,Bl)) ^ (Pmull(Ah,Bl) ^ Pmull(Al,Bh))<<64 +# +# The idea behind Karatsuba multiplication is that we can do just 3 64b multiplies: +# Pmull(A,B) == (Pmull(Ah,Bh)<<128 | Pmull(Al,Bl)) ^ (Pmull(Ah^Al,Bh^Bl) ^ Pmull(Ah,Bh) ^ Pmull(Al,Bl))<<64 +# +# There is some complication here because the bit order of GHASH's PMULL is reversed compared to elsewhere, so we are +# multiplying with "twisted" powers of H +# +# Note: We can PMULL directly into the acc_x in first GHASH of the loop +# Note: For scheduling big cores we want to split the processing to happen over two loop iterations - otherwise the critical +# path latency dominates the performance +# +# This has a knock on effect on register pressure, so we have to be a bit more clever with our temporary registers +# than indicated here +# REV64 res_curr, res_curr +# INS t_m.d[0], res_curr.d[1] +# EOR t_m.8B, t_m.8B, res_curr.8B +# PMULL2 t_h, res_curr, HX +# PMULL t_l, res_curr, HX +# PMULL t_m, t_m, HX_k +# EOR acc_h, acc_h, t_h +# EOR acc_l, acc_l, t_l +# EOR acc_m, acc_m, t_m +# +# MODULO: take the partial accumulators (~representing sum of 256b multiplication results), from GHASH and do modulo reduction on them +# There is some complication here because the bit order of GHASH's PMULL is reversed compared to elsewhere, so we are doing modulo +# with a reversed constant +# EOR3 acc_m, acc_m, acc_l, acc_h // Finish off karatsuba processing +# PMULL t_mod, acc_h, mod_constant +# EXT acc_h, acc_h, acc_h, #8 +# EOR3 acc_m, acc_m, t_mod, acc_h +# PMULL acc_h, acc_m, mod_constant +# EXT acc_m, acc_m, acc_m, #8 +# EOR3 acc_l, acc_l, acc_m, acc_h + +$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef; +$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef; + +$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; +( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or +( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate ) or +die "can't locate arm-xlate.pl"; + +die "only for 64 bit" if $flavour !~ /64/; + +open OUT,"| \"$^X\" $xlate $flavour $output"; +*STDOUT=*OUT; + +$code=<<___; +#include "arm_arch.h" + +#if __ARM_MAX_ARCH__>=8 +___ +$code.=".arch armv8.2-a+crypto\n.arch_extension sha3\n.text\n"; + +$input_ptr="x0"; #argument block +$bit_length="x1"; +$output_ptr="x2"; +$current_tag="x3"; +$counter="x16"; +$constant_temp="x15"; +$modulo_constant="x10"; +$cc="x8"; +{ +my ($end_input_ptr,$main_end_input_ptr,$temp0_x,$temp1_x)=map("x$_",(4..7)); +my ($temp2_x,$temp3_x)=map("x$_",(13..14)); +my ($ctr0b,$ctr1b,$ctr2b,$ctr3b,$ctr4b,$ctr5b,$ctr6b,$ctr7b,$res0b,$res1b,$res2b,$res3b,$res4b,$res5b,$res6b,$res7b)=map("v$_.16b",(0..15)); +my ($ctr0,$ctr1,$ctr2,$ctr3,$ctr4,$ctr5,$ctr6,$ctr7,$res0,$res1,$res2,$res3,$res4,$res5,$res6,$res7)=map("v$_",(0..15)); +my ($ctr0d,$ctr1d,$ctr2d,$ctr3d,$ctr4d,$ctr5d,$ctr6d,$ctr7d)=map("d$_",(0..7)); +my ($ctr0q,$ctr1q,$ctr2q,$ctr3q,$ctr4q,$ctr5q,$ctr6q,$ctr7q)=map("q$_",(0..7)); +my ($res0q,$res1q,$res2q,$res3q,$res4q,$res5q,$res6q,$res7q)=map("q$_",(8..15)); + +my ($ctr_t0,$ctr_t1,$ctr_t2,$ctr_t3,$ctr_t4,$ctr_t5,$ctr_t6,$ctr_t7)=map("v$_",(8..15)); +my ($ctr_t0b,$ctr_t1b,$ctr_t2b,$ctr_t3b,$ctr_t4b,$ctr_t5b,$ctr_t6b,$ctr_t7b)=map("v$_.16b",(8..15)); +my ($ctr_t0q,$ctr_t1q,$ctr_t2q,$ctr_t3q,$ctr_t4q,$ctr_t5q,$ctr_t6q,$ctr_t7q)=map("q$_",(8..15)); + +my ($acc_hb,$acc_mb,$acc_lb)=map("v$_.16b",(17..19)); +my ($acc_h,$acc_m,$acc_l)=map("v$_",(17..19)); + +my ($h1,$h12k,$h2,$h3,$h34k,$h4)=map("v$_",(20..25)); +my ($h5,$h56k,$h6,$h7,$h78k,$h8)=map("v$_",(20..25)); +my ($h1q,$h12kq,$h2q,$h3q,$h34kq,$h4q)=map("q$_",(20..25)); +my ($h5q,$h56kq,$h6q,$h7q,$h78kq,$h8q)=map("q$_",(20..25)); + +my $t0="v16"; +my $t0d="d16"; + +my $t1="v29"; +my $t2=$res1; +my $t3=$t1; + +my $t4=$res0; +my $t5=$res2; +my $t6=$t0; + +my $t7=$res3; +my $t8=$res4; +my $t9=$res5; + +my $t10=$res6; +my $t11="v21"; +my $t12=$t1; + +my $rtmp_ctr="v30"; +my $rtmp_ctrq="q30"; +my $rctr_inc="v31"; +my $rctr_incd="d31"; + +my $mod_constantd=$t0d; +my $mod_constant=$t0; + +my ($rk0,$rk1,$rk2)=map("v$_.16b",(26..28)); +my ($rk3,$rk4,$rk5)=map("v$_.16b",(26..28)); +my ($rk6,$rk7,$rk8)=map("v$_.16b",(26..28)); +my ($rk9,$rk10,$rk11)=map("v$_.16b",(26..28)); +my ($rk12,$rk13,$rk14)=map("v$_.16b",(26..28)); +my ($rk0q,$rk1q,$rk2q)=map("q$_",(26..28)); +my ($rk3q,$rk4q,$rk5q)=map("q$_",(26..28)); +my ($rk6q,$rk7q,$rk8q)=map("q$_",(26..28)); +my ($rk9q,$rk10q,$rk11q)=map("q$_",(26..28)); +my ($rk12q,$rk13q,$rk14q)=map("q$_",(26..28)); +my $rk2q1="v28.1q"; +my $rk3q1="v26.1q"; +my $rk4v="v27"; + + +######################################################################################### +# size_t unroll8_eor3_aes_gcm_enc_128_kernel(const unsigned char *in, +# size_t len, +# unsigned char *out, +# const void *key, +# unsigned char ivec[16], +# u64 *Xi); +# +$code.=<<___; +.global unroll8_eor3_aes_gcm_enc_128_kernel +.type unroll8_eor3_aes_gcm_enc_128_kernel,%function +.align 4 +unroll8_eor3_aes_gcm_enc_128_kernel: + AARCH64_VALID_CALL_TARGET + cbz x1, .L128_enc_ret + stp d8, d9, [sp, #-80]! + mov $counter, x4 + mov $cc, x5 + stp d10, d11, [sp, #16] + stp d12, d13, [sp, #32] + stp d14, d15, [sp, #48] + mov x5, #0xc200000000000000 + stp x5, xzr, [sp, #64] + add $modulo_constant, sp, #64 + + mov $constant_temp, #0x100000000 @ set up counter increment + movi $rctr_inc.16b, #0x0 + mov $rctr_inc.d[1], $constant_temp + lsr $main_end_input_ptr, $bit_length, #3 @ byte_len + ld1 { $ctr0b}, [$counter] @ CTR block 0 + + sub $main_end_input_ptr, $main_end_input_ptr, #1 @ byte_len - 1 + + and $main_end_input_ptr, $main_end_input_ptr, #0xffffffffffffff80 @ number of bytes to be processed in main loop (at least 1 byte must be handled by tail) + + rev32 $rtmp_ctr.16b, $ctr0.16b @ set up reversed counter + + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 0 + + rev32 $ctr1.16b, $rtmp_ctr.16b @ CTR block 1 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 1 + + rev32 $ctr2.16b, $rtmp_ctr.16b @ CTR block 2 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 2 + + rev32 $ctr3.16b, $rtmp_ctr.16b @ CTR block 3 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 3 + + rev32 $ctr4.16b, $rtmp_ctr.16b @ CTR block 4 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 4 + + rev32 $ctr5.16b, $rtmp_ctr.16b @ CTR block 5 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 5 + ldp $rk0q, $rk1q, [$cc, #0] @ load rk0, rk1 + + rev32 $ctr6.16b, $rtmp_ctr.16b @ CTR block 6 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 6 + + rev32 $ctr7.16b, $rtmp_ctr.16b @ CTR block 7 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 7 + + aese $ctr4b, $rk0 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 0 + aese $ctr6b, $rk0 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 0 + aese $ctr3b, $rk0 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 0 + + aese $ctr0b, $rk0 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 0 + aese $ctr1b, $rk0 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 0 + aese $ctr2b, $rk0 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 0 + + aese $ctr7b, $rk0 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 0 + aese $ctr5b, $rk0 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 0 + ldp $rk2q, $rk3q, [$cc, #32] @ load rk2, rk3 + + aese $ctr3b, $rk1 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 1 + + aese $ctr7b, $rk1 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 1 + aese $ctr5b, $rk1 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 1 + aese $ctr4b, $rk1 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 1 + + aese $ctr2b, $rk1 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 1 + aese $ctr6b, $rk1 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 1 + aese $ctr0b, $rk1 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 1 + + aese $ctr5b, $rk2 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 2 + aese $ctr1b, $rk1 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 1 + aese $ctr0b, $rk2 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 2 + + aese $ctr2b, $rk2 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 2 + aese $ctr3b, $rk2 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 2 + aese $ctr7b, $rk2 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 2 + + aese $ctr1b, $rk2 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 2 + aese $ctr6b, $rk2 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 2 + aese $ctr4b, $rk2 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 2 + + aese $ctr2b, $rk3 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 3 + + ldp $rk4q, $rk5q, [$cc, #64] @ load rk4, rk5 + aese $ctr5b, $rk3 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 3 + aese $ctr0b, $rk3 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 3 + + aese $ctr4b, $rk3 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 3 + aese $ctr3b, $rk3 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 3 + aese $ctr6b, $rk3 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 3 + + aese $ctr7b, $rk3 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 3 + + aese $ctr6b, $rk4 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 4 + aese $ctr1b, $rk3 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 3 + aese $ctr5b, $rk4 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 4 + + aese $ctr7b, $rk4 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 4 + aese $ctr4b, $rk4 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 4 + aese $ctr0b, $rk4 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 4 + + aese $ctr1b, $rk4 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 4 + aese $ctr2b, $rk4 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 4 + aese $ctr3b, $rk4 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 4 + + aese $ctr7b, $rk5 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 5 + aese $ctr0b, $rk5 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 5 + ldp $rk6q, $rk7q, [$cc, #96] @ load rk6, rk7 + + aese $ctr1b, $rk5 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 5 + aese $ctr3b, $rk5 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 5 + aese $ctr2b, $rk5 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 5 + + aese $ctr4b, $rk5 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 5 + aese $ctr5b, $rk5 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 5 + aese $ctr6b, $rk5 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 5 + + aese $ctr4b, $rk6 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 6 + aese $ctr3b, $rk6 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 6 + aese $ctr2b, $rk6 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 6 + + aese $ctr7b, $rk6 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 6 + aese $ctr6b, $rk6 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 6 + aese $ctr5b, $rk6 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 6 + + aese $ctr0b, $rk6 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 6 + aese $ctr1b, $rk6 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 6 + ldp $rk8q, $rk9q, [$cc, #128] @ load rk8, rk9 + + aese $ctr5b, $rk7 \n aesmc $ctr5b, $ctr5b @ AES block 5 - round 7 + + ld1 { $acc_lb}, [$current_tag] + ext $acc_lb, $acc_lb, $acc_lb, #8 + rev64 $acc_lb, $acc_lb + + aese $ctr7b, $rk7 \n aesmc $ctr7b, $ctr7b @ AES block 7 - round 7 + + aese $ctr4b, $rk7 \n aesmc $ctr4b, $ctr4b @ AES block 4 - round 7 + aese $ctr3b, $rk7 \n aesmc $ctr3b, $ctr3b @ AES block 3 - round 7 + aese $ctr6b, $rk7 \n aesmc $ctr6b, $ctr6b @ AES block 6 - round 7 + + aese $ctr1b, $rk7 \n aesmc $ctr1b, $ctr1b @ AES block 1 - round 7 + aese $ctr2b, $rk7 \n aesmc $ctr2b, $ctr2b @ AES block 2 - round 7 + aese $ctr0b, $rk7 \n aesmc $ctr0b, $ctr0b @ AES block 0 - round 7 + + aese $ctr3b, $rk8 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 8 + aese $ctr6b, $rk8 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 8 + aese $ctr2b, $rk8 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 8 + + aese $ctr7b, $rk8 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 8 + aese $ctr0b, $rk8 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 8 + ldr $rk10q, [$cc, #160] @ load rk10 + + aese $ctr3b, $rk9 @ AES block 8k+11 - round 9 + aese $ctr4b, $rk8 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 8 + aese $ctr2b, $rk9 @ AES block 8k+10 - round 9 + + aese $ctr5b, $rk8 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 8 + aese $ctr1b, $rk8 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 8 + aese $ctr6b, $rk9 @ AES block 8k+14 - round 9 + + aese $ctr4b, $rk9 @ AES block 8k+12 - round 9 + add $main_end_input_ptr, $main_end_input_ptr, $input_ptr + aese $ctr0b, $rk9 @ AES block 8k+8 - round 9 + + aese $ctr7b, $rk9 @ AES block 8k+15 - round 9 + aese $ctr5b, $rk9 @ AES block 8k+13 - round 9 + aese $ctr1b, $rk9 @ AES block 8k+9 - round 9 + + add $end_input_ptr, $input_ptr, $bit_length, lsr #3 @ end_input_ptr + cmp $input_ptr, $main_end_input_ptr @ check if we have <= 8 blocks + b.ge .L128_enc_tail @ handle tail + + ldp $ctr_t0q, $ctr_t1q, [$input_ptr], #32 @ AES block 0, 1 - load plaintext + + ldp $ctr_t2q, $ctr_t3q, [$input_ptr], #32 @ AES block 2, 3 - load plaintext + + ldp $ctr_t4q, $ctr_t5q, [$input_ptr], #32 @ AES block 4, 5 - load plaintext + + ldp $ctr_t6q, $ctr_t7q, [$input_ptr], #32 @ AES block 6, 7 - load plaintext + cmp $input_ptr, $main_end_input_ptr @ check if we have <= 8 blocks + + eor3 $res0b, $ctr_t0b, $ctr0b, $rk10 @ AES block 0 - result + rev32 $ctr0.16b, $rtmp_ctr.16b @ CTR block 8 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8 + + eor3 $res1b, $ctr_t1b, $ctr1b, $rk10 @ AES block 1 - result + stp $res0q, $res1q, [$output_ptr], #32 @ AES block 0, 1 - store result + + rev32 $ctr1.16b, $rtmp_ctr.16b @ CTR block 9 + eor3 $res5b, $ctr_t5b, $ctr5b, $rk10 @ AES block 5 - result + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 9 + + eor3 $res2b, $ctr_t2b, $ctr2b, $rk10 @ AES block 2 - result + eor3 $res6b, $ctr_t6b, $ctr6b, $rk10 @ AES block 6 - result + eor3 $res4b, $ctr_t4b, $ctr4b, $rk10 @ AES block 4 - result + + rev32 $ctr2.16b, $rtmp_ctr.16b @ CTR block 10 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 10 + + eor3 $res3b, $ctr_t3b, $ctr3b, $rk10 @ AES block 3 - result + eor3 $res7b, $ctr_t7b, $ctr7b,$rk10 @ AES block 7 - result + stp $res2q, $res3q, [$output_ptr], #32 @ AES block 2, 3 - store result + + rev32 $ctr3.16b, $rtmp_ctr.16b @ CTR block 11 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 11 + stp $res4q, $res5q, [$output_ptr], #32 @ AES block 4, 5 - store result + + stp $res6q, $res7q, [$output_ptr], #32 @ AES block 6, 7 - store result + + rev32 $ctr4.16b, $rtmp_ctr.16b @ CTR block 12 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 12 + b.ge .L128_enc_prepretail @ do prepretail + +.L128_enc_main_loop: @ main loop start + rev32 $ctr5.16b, $rtmp_ctr.16b @ CTR block 8k+13 + ldr $h5q, [$current_tag, #128] @ load h5l | h5h + ext $h5.16b, $h5.16b, $h5.16b, #8 + ldr $h6q, [$current_tag, #160] @ load h6l | h6h + ext $h6.16b, $h6.16b, $h6.16b, #8 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+13 + + rev64 $res1b, $res1b @ GHASH block 8k+1 + rev64 $res0b, $res0b @ GHASH block 8k + ldr $h7q, [$current_tag, #176] @ load h7l | h7h + ext $h7.16b, $h7.16b, $h7.16b, #8 + ldr $h8q, [$current_tag, #208] @ load h8l | h8h + ext $h8.16b, $h8.16b, $h8.16b, #8 + + rev32 $ctr6.16b, $rtmp_ctr.16b @ CTR block 8k+14 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+14 + ext $acc_lb, $acc_lb, $acc_lb, #8 @ PRE 0 + + ldr $h56kq, [$current_tag, #144] @ load h6k | h5k + ldr $h78kq, [$current_tag, #192] @ load h8k | h7k + rev64 $res5b, $res5b @ GHASH block 8k+5 (t0, t1, t2 and t3 free) + rev64 $res3b, $res3b @ GHASH block 8k+3 + + ldp $rk0q, $rk1q, [$cc, #0] @ load rk0, rk1 + eor $res0b, $res0b, $acc_lb @ PRE 1 + rev32 $ctr7.16b, $rtmp_ctr.16b @ CTR block 8k+15 + + rev64 $res7b, $res7b @ GHASH block 8k+7 (t0, t1, t2 and t3 free) + + pmull2 $t0.1q, $res1.2d, $h7.2d @ GHASH block 8k+1 - high + rev64 $res2b, $res2b @ GHASH block 8k+2 + pmull2 $acc_h.1q, $res0.2d, $h8.2d @ GHASH block 8k - high + + pmull $h7.1q, $res1.1d, $h7.1d @ GHASH block 8k+1 - low + trn1 $acc_m.2d, $res1.2d, $res0.2d @ GHASH block 8k, 8k+1 - mid + pmull $acc_l.1q, $res0.1d, $h8.1d @ GHASH block 8k - low + + trn2 $res0.2d, $res1.2d, $res0.2d @ GHASH block 8k, 8k+1 - mid + pmull2 $t1.1q, $res2.2d, $h6.2d @ GHASH block 8k+2 - high + pmull2 $t2.1q, $res3.2d, $h5.2d @ GHASH block 8k+3 - high + + eor $acc_lb, $acc_lb, $h7.16b @ GHASH block 8k+1 - low + ldr $h3q, [$current_tag, #80] @ load h3l | h3h + ext $h3.16b, $h3.16b, $h3.16b, #8 + ldr $h4q, [$current_tag, #112] @ load h3l | h3h + ext $h4.16b, $h4.16b, $h4.16b, #8 + aese $ctr5b, $rk0 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 0 + + aese $ctr1b, $rk0 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 0 + aese $ctr4b, $rk0 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 0 + eor $acc_hb, $acc_hb, $t0.16b @ GHASH block 8k+1 - high + + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+15 + aese $ctr2b, $rk0 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 0 + eor $res0.16b, $res0.16b, $acc_m.16b @ GHASH block 8k, 8k+1 - mid + + aese $ctr6b, $rk0 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 0 + aese $ctr1b, $rk1 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 1 + aese $ctr0b, $rk0 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 0 + + aese $ctr2b, $rk1 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 1 + aese $ctr3b, $rk0 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 0 + pmull $h6.1q, $res2.1d, $h6.1d @ GHASH block 8k+2 - low + + aese $ctr5b, $rk1 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 1 + aese $ctr7b, $rk0 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 0 + aese $ctr0b, $rk1 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 1 + + eor3 $acc_hb, $acc_hb, $t1.16b,$t2.16b @ GHASH block 8k+2, 8k+3 - high + trn1 $t3.2d, $res3.2d, $res2.2d @ GHASH block 8k+2, 8k+3 - mid + trn2 $res2.2d, $res3.2d, $res2.2d @ GHASH block 8k+2, 8k+3 - mid + + ldp $rk2q, $rk3q, [$cc, #32] @ load rk2, rk3 + aese $ctr4b, $rk1 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 1 + aese $ctr3b, $rk1 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 1 + + pmull $h5.1q, $res3.1d, $h5.1d @ GHASH block 8k+3 - low + aese $ctr7b, $rk1 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 1 + aese $ctr6b, $rk1 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 1 + + pmull2 $acc_m.1q, $res0.2d, $h78k.2d @ GHASH block 8k - mid + eor $res2.16b, $res2.16b, $t3.16b @ GHASH block 8k+2, 8k+3 - mid + pmull $h78k.1q, $res0.1d, $h78k.1d @ GHASH block 8k+1 - mid + + rev64 $res6b, $res6b @ GHASH block 8k+6 (t0, t1, and t2 free) + eor3 $acc_lb, $acc_lb, $h6.16b, $h5.16b @ GHASH block 8k+2, 8k+3 - low + + pmull2 $t3.1q, $res2.2d, $h56k.2d @ GHASH block 8k+2 - mid + eor $acc_mb, $acc_mb, $h78k.16b @ GHASH block 8k+1 - mid + pmull $h56k.1q, $res2.1d, $h56k.1d @ GHASH block 8k+3 - mid + + aese $ctr5b, $rk2 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 2 + aese $ctr4b, $rk2 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 2 + aese $ctr2b, $rk2 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 2 + + aese $ctr1b, $rk2 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 2 + eor3 $acc_mb, $acc_mb, $h56k.16b, $t3.16b @ GHASH block 8k+2, 8k+3 - mid + aese $ctr6b, $rk2 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 2 + + aese $ctr0b, $rk2 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 2 + aese $ctr3b, $rk2 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 2 + aese $ctr7b, $rk2 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 2 + + aese $ctr6b, $rk3 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 3 + ldr $h12kq, [$current_tag, #48] @ load h2k | h1k + ldr $h34kq, [$current_tag, #96] @ load h4k | h3k + rev64 $res4b, $res4b @ GHASH block 8k+4 (t0, t1, and t2 free) + + ldp $rk4q, $rk5q, [$cc, #64] @ load rk4, rk5 + aese $ctr2b, $rk3 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 3 + aese $ctr1b, $rk3 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 3 + + ldr $h1q, [$current_tag, #32] @ load h1l | h1h + ext $h1.16b, $h1.16b, $h1.16b, #8 + ldr $h2q, [$current_tag, #64] @ load h1l | h1h + ext $h2.16b, $h2.16b, $h2.16b, #8 + pmull2 $t4.1q, $res4.2d, $h4.2d @ GHASH block 8k+4 - high + pmull $h4.1q, $res4.1d, $h4.1d @ GHASH block 8k+4 - low + + trn1 $t6.2d, $res5.2d, $res4.2d @ GHASH block 8k+4, 8k+5 - mid + trn2 $res4.2d, $res5.2d, $res4.2d @ GHASH block 8k+4, 8k+5 - mid + + aese $ctr0b, $rk3 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 3 + aese $ctr3b, $rk3 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 3 + + aese $ctr7b, $rk3 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 3 + aese $ctr4b, $rk3 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 3 + + aese $ctr5b, $rk3 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 3 + aese $ctr0b, $rk4 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 4 + + aese $ctr7b, $rk4 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 4 + aese $ctr3b, $rk4 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 4 + aese $ctr4b, $rk4 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 4 + + aese $ctr5b, $rk4 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 4 + aese $ctr6b, $rk4 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 4 + aese $ctr1b, $rk4 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 4 + + pmull2 $t5.1q, $res5.2d, $h3.2d @ GHASH block 8k+5 - high + eor $res4.16b, $res4.16b, $t6.16b @ GHASH block 8k+4, 8k+5 - mid + pmull $h3.1q, $res5.1d, $h3.1d @ GHASH block 8k+5 - low + + aese $ctr2b, $rk4 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 4 + ldp $rk6q, $rk7q, [$cc, #96] @ load rk6, rk7 + trn1 $t9.2d, $res7.2d, $res6.2d @ GHASH block 8k+6, 8k+7 - mid + + pmull2 $t6.1q, $res4.2d, $h34k.2d @ GHASH block 8k+4 - mid + pmull $h34k.1q, $res4.1d, $h34k.1d @ GHASH block 8k+5 - mid + pmull2 $t7.1q, $res6.2d, $h2.2d @ GHASH block 8k+6 - high + + pmull2 $t8.1q, $res7.2d, $h1.2d @ GHASH block 8k+7 - high + aese $ctr2b, $rk5 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 5 + aese $ctr5b, $rk5 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 5 + + pmull $h2.1q, $res6.1d, $h2.1d @ GHASH block 8k+6 - low + eor3 $acc_hb, $acc_hb, $t4.16b, $t5.16b @ GHASH block 8k+4, 8k+5 - high + trn2 $res6.2d, $res7.2d, $res6.2d @ GHASH block 8k+6, 8k+7 - mid + + eor3 $acc_lb, $acc_lb, $h4.16b, $h3.16b @ GHASH block 8k+4, 8k+5 - low + aese $ctr6b, $rk5 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 5 + + eor $res6.16b, $res6.16b, $t9.16b @ GHASH block 8k+6, 8k+7 - mid + aese $ctr7b, $rk5 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 5 + aese $ctr1b, $rk5 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 5 + + aese $ctr3b, $rk5 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 5 + aese $ctr4b, $rk5 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 5 + aese $ctr0b, $rk5 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 5 + + eor3 $acc_mb, $acc_mb, $h34k.16b, $t6.16b @ GHASH block 8k+4, 8k+5 - mid + ldr $mod_constantd, [$modulo_constant] @ MODULO - load modulo constant + pmull $h1.1q, $res7.1d, $h1.1d @ GHASH block 8k+7 - low + + aese $ctr7b, $rk6 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 6 + aese $ctr5b, $rk6 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 6 + + pmull2 $t9.1q, $res6.2d, $h12k.2d @ GHASH block 8k+6 - mid + aese $ctr1b, $rk6 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 6 + aese $ctr2b, $rk6 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 6 + + pmull $h12k.1q, $res6.1d, $h12k.1d @ GHASH block 8k+7 - mid + eor3 $acc_lb, $acc_lb, $h2.16b, $h1.16b @ GHASH block 8k+6, 8k+7 - low + ldp $ctr_t0q, $ctr_t1q, [$input_ptr], #32 @ AES block 8k+8, 8k+9 - load plaintext + + aese $ctr3b, $rk6 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 6 + rev32 $h1.16b, $rtmp_ctr.16b @ CTR block 8k+16 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+16 + + aese $ctr4b, $rk6 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 6 + aese $ctr0b, $rk6 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 6 + aese $ctr6b, $rk6 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 6 + + eor3 $acc_mb, $acc_mb, $h12k.16b, $t9.16b @ GHASH block 8k+6, 8k+7 - mid + ldp $rk8q, $rk9q, [$cc, #128] @ load rk8, rk9 + eor3 $acc_hb, $acc_hb, $t7.16b, $t8.16b @ GHASH block 8k+6, 8k+7 - high + + aese $ctr2b, $rk7 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 7 + aese $ctr7b, $rk7 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 7 + ldp $ctr_t2q, $ctr_t3q, [$input_ptr], #32 @ AES block 8k+10, 8k+11 - load plaintext + + aese $ctr5b, $rk7 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 7 + aese $ctr6b, $rk7 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 7 + aese $ctr1b, $rk7 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 7 + + pmull $t11.1q, $acc_h.1d, $mod_constant.1d @ MODULO - top 64b align with mid + aese $ctr0b, $rk7 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 7 + aese $ctr4b, $rk7 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 7 + + rev32 $h2.16b, $rtmp_ctr.16b @ CTR block 8k+17 + aese $ctr3b, $rk7 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 7 + + aese $ctr5b, $rk8 \n aesmc $ctr5b, $ctr5b @ AES block 8k+13 - round 8 + ldp $ctr_t4q, $ctr_t5q, [$input_ptr], #32 @ AES block 8k+12, 8k+13 - load plaintext + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+17 + + aese $ctr2b, $rk8 \n aesmc $ctr2b, $ctr2b @ AES block 8k+10 - round 8 + aese $ctr1b, $rk8 \n aesmc $ctr1b, $ctr1b @ AES block 8k+9 - round 8 + aese $ctr7b, $rk8 \n aesmc $ctr7b, $ctr7b @ AES block 8k+15 - round 8 + + aese $ctr4b, $rk8 \n aesmc $ctr4b, $ctr4b @ AES block 8k+12 - round 8 + eor3 $acc_mb, $acc_mb, $acc_hb, $acc_lb @ MODULO - karatsuba tidy up + ldr $rk10q, [$cc, #160] @ load rk10 + + ext $t12.16b, $acc_hb, $acc_hb, #8 @ MODULO - other top alignment + rev32 $h3.16b, $rtmp_ctr.16b @ CTR block 8k+18 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+18 + aese $ctr3b, $rk8 \n aesmc $ctr3b, $ctr3b @ AES block 8k+11 - round 8 + + aese $ctr0b, $rk8 \n aesmc $ctr0b, $ctr0b @ AES block 8k+8 - round 8 + eor3 $acc_mb, $acc_mb, $t12.16b, $t11.16b @ MODULO - fold into mid + aese $ctr6b, $rk8 \n aesmc $ctr6b, $ctr6b @ AES block 8k+14 - round 8 + + aese $ctr2b, $rk9 @ AES block 8k+10 - round 9 + aese $ctr4b, $rk9 @ AES block 8k+12 - round 9 + aese $ctr1b, $rk9 @ AES block 8k+9 - round 9 + + ldp $ctr_t6q, $ctr_t7q, [$input_ptr], #32 @ AES block 8k+14, 8k+15 - load plaintext + rev32 $h4.16b, $rtmp_ctr.16b @ CTR block 8k+19 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+19 + + cmp $input_ptr, $main_end_input_ptr @ LOOP CONTROL + eor3 $res4b, $ctr_t4b, $ctr4b, $rk10 @ AES block 4 - result + aese $ctr7b, $rk9 @ AES block 8k+15 - round 9 + + aese $ctr6b, $rk9 @ AES block 8k+14 - round 9 + aese $ctr3b, $rk9 @ AES block 8k+11 - round 9 + + eor3 $res2b, $ctr_t2b, $ctr2b, $rk10 @ AES block 8k+10 - result + + mov $ctr2.16b, $h3.16b @ CTR block 8k+18 + aese $ctr0b, $rk9 @ AES block 8k+8 - round 9 + + rev32 $ctr4.16b, $rtmp_ctr.16b @ CTR block 8k+20 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+20 + + eor3 $res7b, $ctr_t7b, $ctr7b, $rk10 @ AES block 7 - result + aese $ctr5b, $rk9 @ AES block 8k+13 - round 9 + pmull $acc_h.1q, $acc_m.1d, $mod_constant.1d @ MODULO - mid 64b align with low + + eor3 $res1b, $ctr_t1b, $ctr1b, $rk10 @ AES block 8k+9 - result + eor3 $res3b, $ctr_t3b, $ctr3b, $rk10 @ AES block 8k+11 - result + mov $ctr3.16b, $h4.16b @ CTR block 8k+19 + + ext $t11.16b, $acc_mb, $acc_mb, #8 @ MODULO - other mid alignment + eor3 $res5b, $ctr_t5b, $ctr5b, $rk10 @ AES block 5 - result + mov $ctr1.16b, $h2.16b @ CTR block 8k+17 + + eor3 $res0b, $ctr_t0b, $ctr0b, $rk10 @ AES block 8k+8 - result + mov $ctr0.16b, $h1.16b @ CTR block 8k+16 + stp $res0q, $res1q, [$output_ptr], #32 @ AES block 8k+8, 8k+9 - store result + + stp $res2q, $res3q, [$output_ptr], #32 @ AES block 8k+10, 8k+11 - store result + eor3 $res6b, $ctr_t6b, $ctr6b, $rk10 @ AES block 6 - result + + stp $res4q, $res5q, [$output_ptr], #32 @ AES block 8k+12, 8k+13 - store result + eor3 $acc_lb, $acc_lb, $acc_hb, $t11.16b @ MODULO - fold into low + + stp $res6q, $res7q, [$output_ptr], #32 @ AES block 8k+14, 8k+15 - store result + b.lt .L128_enc_main_loop + +.L128_enc_prepretail: @ PREPRETAIL + rev32 $ctr5.16b, $rtmp_ctr.16b @ CTR block 8k+13 + ldr $h7q, [$current_tag, #176] @ load h7l | h7h + ext $h7.16b, $h7.16b, $h7.16b, #8 + ldr $h8q, [$current_tag, #208] @ load h8l | h8h + ext $h8.16b, $h8.16b, $h8.16b, #8 + ext $acc_lb, $acc_lb, $acc_lb, #8 @ PRE 0 + + ldr $h5q, [$current_tag, #128] @ load h5l | h5h + ext $h5.16b, $h5.16b, $h5.16b, #8 + ldr $h6q, [$current_tag, #160] @ load h6l | h6h + ext $h6.16b, $h6.16b, $h6.16b, #8 + rev64 $res0b, $res0b @ GHASH block 8k + rev64 $res1b, $res1b @ GHASH block 8k+1 + + ldr $h56kq, [$current_tag, #144] @ load h6k | h5k + ldr $h78kq, [$current_tag, #192] @ load h6k | h5k + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+13 + rev64 $res3b, $res3b @ GHASH block 8k+3 + + rev64 $res2b, $res2b @ GHASH block 8k+2 + eor $res0b, $res0b, $acc_lb @ PRE 1 + + rev32 $ctr6.16b, $rtmp_ctr.16b @ CTR block 8k+14 + + pmull2 $t0.1q, $res1.2d, $h7.2d @ GHASH block 8k+1 - high + pmull $acc_l.1q, $res0.1d, $h8.1d @ GHASH block 8k - low + pmull2 $acc_h.1q, $res0.2d, $h8.2d @ GHASH block 8k - high + + rev64 $res5b, $res5b @ GHASH block 8k+5 (t0, t1, t2 and t3 free) + trn1 $acc_m.2d, $res1.2d, $res0.2d @ GHASH block 8k, 8k+1 - mid + + pmull $h7.1q, $res1.1d, $h7.1d @ GHASH block 8k+1 - low + eor $acc_hb, $acc_hb, $t0.16b @ GHASH block 8k+1 - high + trn2 $res0.2d, $res1.2d, $res0.2d @ GHASH block 8k, 8k+1 - mid + + eor $acc_lb, $acc_lb, $h7.16b @ GHASH block 8k+1 - low + eor $res0.16b, $res0.16b, $acc_m.16b @ GHASH block 8k, 8k+1 - mid + + ldp $rk0q, $rk1q, [$cc, #0] @ load rk0, rk1 + add $rtmp_ctr.4s, $rtmp_ctr.4s, $rctr_inc.4s @ CTR block 8k+14 + + pmull2 $acc_m.1q, $res0.2d, $h78k.2d @ GHASH block 8k - mid + pmull $h78k.1q, $res0.1d, $h78k.1d @ GHASH block 8k+1 - mid + + rev64 $res4b, $res4b @ GHASH block 8k+4 (t0, t1, and t2 free) + rev64 $res7b, $res7b @ GHASH block 8k+7 (t0, t1, t2 and t3 free) + + eor $acc_mb, $acc_mb, $h78k.16b @ GHASH block 8k+1 - mid + |