^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1) #! /usr/bin/env perl
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) # SPDX-License-Identifier: GPL-2.0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4) # This code is taken from CRYPTOGAMs[1] and is included here using the option
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) # in the license to distribute the code under the GPL. Therefore this program
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) # is free software; you can redistribute it and/or modify it under the terms of
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) # the GNU General Public License version 2 as published by the Free Software
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) # Foundation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) # [1] https://www.openssl.org/~appro/cryptogams/
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) # Copyright (c) 2006-2017, CRYPTOGAMS by <appro@openssl.org>
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) # All rights reserved.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) # Redistribution and use in source and binary forms, with or without
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) # modification, are permitted provided that the following conditions
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17) # are met:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) # * Redistributions of source code must retain copyright notices,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20) # this list of conditions and the following disclaimer.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) # * Redistributions in binary form must reproduce the above
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23) # copyright notice, this list of conditions and the following
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) # disclaimer in the documentation and/or other materials
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) # provided with the distribution.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27) # * Neither the name of the CRYPTOGAMS nor the names of its
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) # copyright holder and contributors may be used to endorse or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29) # promote products derived from this software without specific
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) # prior written permission.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) # ALTERNATIVELY, provided that this notice is retained in full, this
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) # product may be distributed under the terms of the GNU General Public
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34) # License (GPL), in which case the provisions of the GPL apply INSTEAD OF
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) # those given above.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER AND CONTRIBUTORS
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38) # "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39) # LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40) # A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) # OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) # SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43) # LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) # DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45) # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46) # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) # project. The module is, however, dual licensed under OpenSSL and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52) # CRYPTOGAMS licenses depending on where you obtain it. For further
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53) # details see https://www.openssl.org/~appro/cryptogams/.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) # This module implements support for AES instructions as per PowerISA
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57) # specification version 2.07, first implemented by POWER8 processor.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) # The module is endian-agnostic in sense that it supports both big-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59) # and little-endian cases. Data alignment in parallelizable modes is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) # handled with VSX loads and stores, which implies MSR.VSX flag being
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61) # set. It should also be noted that ISA specification doesn't prohibit
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62) # alignment exceptions for these instructions on page boundaries.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63) # Initially alignment was handled in pure AltiVec/VMX way [when data
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) # is aligned programmatically, which in turn guarantees exception-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65) # free execution], but it turned to hamper performance when vcipher
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66) # instructions are interleaved. It's reckoned that eventual
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) # misalignment penalties at page boundaries are in average lower
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) # than additional overhead in pure AltiVec approach.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70) # May 2016
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72) # Add XTS subroutine, 9x on little- and 12x improvement on big-endian
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73) # systems were measured.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) ######################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) # Current large-block performance in cycles per byte processed with
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77) # 128-bit key (less is better).
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) # CBC en-/decrypt CTR XTS
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80) # POWER8[le] 3.96/0.72 0.74 1.1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81) # POWER8[be] 3.75/0.65 0.66 1.0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) $flavour = shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) if ($flavour =~ /64/) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) $SIZE_T =8;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) $LRSAVE =2*$SIZE_T;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) $STU ="stdu";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) $POP ="ld";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90) $PUSH ="std";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) $UCMP ="cmpld";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) $SHL ="sldi";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93) } elsif ($flavour =~ /32/) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) $SIZE_T =4;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) $LRSAVE =$SIZE_T;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) $STU ="stwu";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97) $POP ="lwz";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) $PUSH ="stw";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99) $UCMP ="cmplw";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) $SHL ="slwi";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) } else { die "nonsense $flavour"; }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) $LITTLE_ENDIAN = ($flavour=~/le$/) ? $SIZE_T : 0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) die "can't locate ppc-xlate.pl";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) $FRAME=8*$SIZE_T;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) $prefix="aes_p8";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) $sp="r1";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) $vrsave="r12";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) {{{ # Key setup procedures #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) my ($inp,$bits,$out,$ptr,$cnt,$rounds)=map("r$_",(3..8));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121) my ($zero,$in0,$in1,$key,$rcon,$mask,$tmp)=map("v$_",(0..6));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122) my ($stage,$outperm,$outmask,$outhead,$outtail)=map("v$_",(7..11));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) .machine "any"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) .text
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) .align 7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) rcon:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) .long 0x01000000, 0x01000000, 0x01000000, 0x01000000 ?rev
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) .long 0x1b000000, 0x1b000000, 0x1b000000, 0x1b000000 ?rev
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) .long 0x0d0e0f0c, 0x0d0e0f0c, 0x0d0e0f0c, 0x0d0e0f0c ?rev
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134) .long 0,0,0,0 ?asis
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) Lconsts:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) mflr r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) bcl 20,31,\$+4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) mflr $ptr #vvvvv "distance between . and rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) addi $ptr,$ptr,-0x48
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) mtlr r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) .byte 0,12,0x14,0,0,0,0,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) .asciz "AES for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146) .globl .${prefix}_set_encrypt_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) Lset_encrypt_key:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148) mflr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) $PUSH r11,$LRSAVE($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) li $ptr,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) ${UCMP}i $inp,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) beq- Lenc_key_abort # if ($inp==0) return -1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) ${UCMP}i $out,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155) beq- Lenc_key_abort # if ($out==0) return -1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) li $ptr,-2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) cmpwi $bits,128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) blt- Lenc_key_abort
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) cmpwi $bits,256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) bgt- Lenc_key_abort
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) andi. r0,$bits,0x3f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) bne- Lenc_key_abort
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) lis r0,0xfff0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) mfspr $vrsave,256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) bl Lconsts
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) mtlr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) neg r9,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) lvx $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) lvsr $key,0,r9 # borrow $key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) li r8,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) cmpwi $bits,192
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) lvx $in1,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) le?vspltisb $mask,0x0f # borrow $mask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) lvx $rcon,0,$ptr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) le?vxor $key,$key,$mask # adjust for byte swap
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) lvx $mask,r8,$ptr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) addi $ptr,$ptr,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) vperm $in0,$in0,$in1,$key # align [and byte swap in LE]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) li $cnt,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) vxor $zero,$zero,$zero
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) mtctr $cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) ?lvsr $outperm,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) vspltisb $outmask,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) lvx $outhead,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) ?vperm $outmask,$zero,$outmask,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) blt Loop128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194) addi $inp,$inp,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) beq L192
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) addi $inp,$inp,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) b L256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) Loop128:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201) vperm $key,$in0,$in0,$mask # rotate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) vadduwm $rcon,$rcon,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) bdnz Loop128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) lvx $rcon,0,$ptr # last two round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) vperm $key,$in0,$in0,$mask # rotate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) vadduwm $rcon,$rcon,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) vperm $key,$in0,$in0,$mask # rotate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) addi $inp,$out,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259) addi $out,$out,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) li $rounds,10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) b Ldone
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) L192:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) lvx $tmp,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) li $cnt,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273) vperm $in1,$in1,$tmp,$key # align [and byte swap in LE]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) vspltisb $key,8 # borrow $key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) mtctr $cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) vsububm $mask,$mask,$key # adjust the mask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278) Loop192:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) vperm $key,$in1,$in1,$mask # roate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) vsldoi $stage,$zero,$in1,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) vspltw $tmp,$in0,3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291) vxor $tmp,$tmp,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292) vsldoi $in1,$zero,$in1,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) vadduwm $rcon,$rcon,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) vxor $in1,$in1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) vxor $in1,$in1,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) vsldoi $stage,$stage,$in0,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) vperm $key,$in1,$in1,$mask # rotate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) vperm $outtail,$stage,$stage,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) vsldoi $stage,$in0,$in1,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311) vperm $outtail,$stage,$stage,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) vspltw $tmp,$in0,3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) vxor $tmp,$tmp,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322) vsldoi $in1,$zero,$in1,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) vadduwm $rcon,$rcon,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324) vxor $in1,$in1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) vxor $in1,$in1,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331) addi $inp,$out,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) bdnz Loop192
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335) li $rounds,12
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336) addi $out,$out,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) b Ldone
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340) L256:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) lvx $tmp,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) li $cnt,7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) li $rounds,14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) vperm $in1,$in1,$tmp,$key # align [and byte swap in LE]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) mtctr $cnt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352) Loop256:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) vperm $key,$in1,$in1,$mask # rotate-n-splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354) vsldoi $tmp,$zero,$in0,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) vperm $outtail,$in1,$in1,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) vcipherlast $key,$key,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) vxor $in0,$in0,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) vadduwm $rcon,$rcon,$rcon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) vxor $in0,$in0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369) vperm $outtail,$in0,$in0,$outperm # rotate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) vsel $stage,$outhead,$outtail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371) vmr $outhead,$outtail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372) stvx $stage,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) addi $inp,$out,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375) bdz Ldone
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377) vspltw $key,$in0,3 # just splat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378) vsldoi $tmp,$zero,$in1,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) vsbox $key,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381) vxor $in1,$in1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) vxor $in1,$in1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) vsldoi $tmp,$zero,$tmp,12 # >>32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385) vxor $in1,$in1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) vxor $in1,$in1,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388) b Loop256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) Ldone:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392) lvx $in1,0,$inp # redundant in aligned case
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393) vsel $in1,$outhead,$in1,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) stvx $in1,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) li $ptr,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) stw $rounds,0($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399) Lenc_key_abort:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400) mr r3,$ptr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) .byte 0,12,0x14,1,0,0,3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405) .size .${prefix}_set_encrypt_key,.-.${prefix}_set_encrypt_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) .globl .${prefix}_set_decrypt_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408) $STU $sp,-$FRAME($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409) mflr r10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410) $PUSH r10,$FRAME+$LRSAVE($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411) bl Lset_encrypt_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412) mtlr r10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) cmpwi r3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415) bne- Ldec_key_abort
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) slwi $cnt,$rounds,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418) subi $inp,$out,240 # first round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420) add $out,$inp,$cnt # last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) Ldeckey:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424) lwz r0, 0($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425) lwz r6, 4($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426) lwz r7, 8($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) lwz r8, 12($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429) lwz r9, 0($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) lwz r10,4($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) lwz r11,8($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432) lwz r12,12($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433) stw r0, 0($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434) stw r6, 4($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) stw r7, 8($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436) stw r8, 12($out)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437) subi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438) stw r9, -16($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) stw r10,-12($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440) stw r11,-8($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441) stw r12,-4($inp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442) bdnz Ldeckey
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444) xor r3,r3,r3 # return value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445) Ldec_key_abort:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) addi $sp,$sp,$FRAME
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) .byte 0,12,4,1,0x80,0,3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451) .size .${prefix}_set_decrypt_key,.-.${prefix}_set_decrypt_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) }}}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) {{{ # Single block en- and decrypt procedures #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456) sub gen_block () {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457) my $dir = shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) my $n = $dir eq "de" ? "n" : "";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) my ($inp,$out,$key,$rounds,$idx)=map("r$_",(3..7));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462) .globl .${prefix}_${dir}crypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) lwz $rounds,240($key)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464) lis r0,0xfc00
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465) mfspr $vrsave,256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) li $idx,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 468)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 469) lvx v0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 470) neg r11,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 471) lvx v1,$idx,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 472) lvsl v2,0,$inp # inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 473) le?vspltisb v4,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 474) ?lvsl v3,0,r11 # outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 475) le?vxor v2,v2,v4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 476) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 477) vperm v0,v0,v1,v2 # align [and byte swap in LE]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 478) lvx v1,0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 479) ?lvsl v5,0,$key # keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 480) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 481) lvx v2,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 482) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 483) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 484) ?vperm v1,v1,v2,v5 # align round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 485)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 486) vxor v0,v0,v1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 487) lvx v1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 488) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 489) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 490)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 491) Loop_${dir}c:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 492) ?vperm v2,v2,v1,v5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 493) v${n}cipher v0,v0,v2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 494) lvx v2,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 495) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 496) ?vperm v1,v1,v2,v5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 497) v${n}cipher v0,v0,v1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 498) lvx v1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 499) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 500) bdnz Loop_${dir}c
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 501)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 502) ?vperm v2,v2,v1,v5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 503) v${n}cipher v0,v0,v2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 504) lvx v2,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 505) ?vperm v1,v1,v2,v5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 506) v${n}cipherlast v0,v0,v1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 507)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 508) vspltisb v2,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 509) vxor v1,v1,v1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 510) li $idx,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 511) ?vperm v2,v1,v2,v3 # outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 512) le?vxor v3,v3,v4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 513) lvx v1,0,$out # outhead
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 514) vperm v0,v0,v0,v3 # rotate [and byte swap in LE]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 515) vsel v1,v1,v0,v2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 516) lvx v4,$idx,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 517) stvx v1,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 518) vsel v0,v0,v4,v2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 519) stvx v0,$idx,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 520)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 521) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 522) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 523) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 524) .byte 0,12,0x14,0,0,0,3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 525) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 526) .size .${prefix}_${dir}crypt,.-.${prefix}_${dir}crypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 527) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 528) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 529) &gen_block("en");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 530) &gen_block("de");
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 531) }}}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 532) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 533) {{{ # CBC en- and decrypt procedures #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 534) my ($inp,$out,$len,$key,$ivp,$enc,$rounds,$idx)=map("r$_",(3..10));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 535) my ($rndkey0,$rndkey1,$inout,$tmp)= map("v$_",(0..3));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 536) my ($ivec,$inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm)=
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 537) map("v$_",(4..10));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 538) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 539) .globl .${prefix}_cbc_encrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 540) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 541) bltlr-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 542)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 543) cmpwi $enc,0 # test direction
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 544) lis r0,0xffe0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 545) mfspr $vrsave,256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 546) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 547)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 548) li $idx,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 549) vxor $rndkey0,$rndkey0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 550) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 551)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 552) lvx $ivec,0,$ivp # load [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 553) lvsl $inpperm,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 554) lvx $inptail,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 555) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 556) vperm $ivec,$ivec,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 557)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 558) neg r11,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 559) ?lvsl $keyperm,0,$key # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 560) lwz $rounds,240($key)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 561)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 562) lvsr $inpperm,0,r11 # prepare for unaligned load
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 563) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 564) addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 565) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 566)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 567) ?lvsr $outperm,0,$out # prepare for unaligned store
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 568) vspltisb $outmask,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 569) lvx $outhead,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 570) ?vperm $outmask,$rndkey0,$outmask,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 571) le?vxor $outperm,$outperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 572)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 573) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 574) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 575) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 576) beq Lcbc_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 577)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 578) Lcbc_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 579) vmr $inout,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 580) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 581) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 582) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 583) subi $len,$len,16 # len-=16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 584)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 585) lvx $rndkey0,0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 586) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 587) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 588) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 589) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 590) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 591) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 592) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 593) vxor $inout,$inout,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 594)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 595) Loop_cbc_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 596) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 597) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 598) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 599) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 600) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 601) vcipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 602) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 603) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 604) bdnz Loop_cbc_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 605)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 606) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 607) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 608) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 609) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 610) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 611) vcipherlast $ivec,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 612) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 613)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 614) vperm $tmp,$ivec,$ivec,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 615) vsel $inout,$outhead,$tmp,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 616) vmr $outhead,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 617) stvx $inout,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 618) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 619) bge Lcbc_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 620)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 621) b Lcbc_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 622)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 623) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 624) Lcbc_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 625) ${UCMP}i $len,128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 626) bge _aesp8_cbc_decrypt8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 627) vmr $tmp,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 628) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 629) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 630) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 631) subi $len,$len,16 # len-=16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 632)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 633) lvx $rndkey0,0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 634) vperm $tmp,$tmp,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 635) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 636) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 637) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 638) vxor $inout,$tmp,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 639) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 640) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 641)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 642) Loop_cbc_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 643) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 644) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 645) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 646) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 647) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 648) vncipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 649) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 650) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 651) bdnz Loop_cbc_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 652)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 653) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 654) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 655) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 656) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 657) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 658) vncipherlast $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 659) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 660)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 661) vxor $inout,$inout,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 662) vmr $ivec,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 663) vperm $tmp,$inout,$inout,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 664) vsel $inout,$outhead,$tmp,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 665) vmr $outhead,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 666) stvx $inout,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 667) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 668) bge Lcbc_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 669)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 670) Lcbc_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 671) addi $out,$out,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 672) lvx $inout,0,$out # redundant in aligned case
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 673) vsel $inout,$outhead,$inout,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 674) stvx $inout,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 675)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 676) neg $enc,$ivp # write [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 677) li $idx,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 678) vxor $rndkey0,$rndkey0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 679) vspltisb $outmask,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 680) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 681) ?lvsl $outperm,0,$enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 682) ?vperm $outmask,$rndkey0,$outmask,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 683) le?vxor $outperm,$outperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 684) lvx $outhead,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 685) vperm $ivec,$ivec,$ivec,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 686) vsel $inout,$outhead,$ivec,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 687) lvx $inptail,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 688) stvx $inout,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 689) vsel $inout,$ivec,$inptail,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 690) stvx $inout,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 691)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 692) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 693) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 694) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 695) .byte 0,12,0x14,0,0,0,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 696) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 697) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 698) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 699) {{ # Optimized CBC decrypt procedure #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 700) my $key_="r11";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 701) my ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70)=map("r$_",(0,8,26..31));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 702) my ($in0, $in1, $in2, $in3, $in4, $in5, $in6, $in7 )=map("v$_",(0..3,10..13));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 703) my ($out0,$out1,$out2,$out3,$out4,$out5,$out6,$out7)=map("v$_",(14..21));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 704) my $rndkey0="v23"; # v24-v25 rotating buffer for first found keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 705) # v26-v31 last 6 round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 706) my ($tmp,$keyperm)=($in3,$in4); # aliases with "caller", redundant assignment
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 707)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 708) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 709) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 710) _aesp8_cbc_decrypt8x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 711) $STU $sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 712) li r10,`$FRAME+8*16+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 713) li r11,`$FRAME+8*16+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 714) stvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 715) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 716) stvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 717) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 718) stvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 719) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 720) stvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 721) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 722) stvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 723) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 724) stvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 725) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 726) stvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 727) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 728) stvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 729) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 730) stvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 731) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 732) stvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 733) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 734) stvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 735) stvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 736) li r0,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 737) stw $vrsave,`$FRAME+21*16-4`($sp) # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 738) li $x10,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 739) $PUSH r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 740) li $x20,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 741) $PUSH r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 742) li $x30,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 743) $PUSH r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 744) li $x40,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 745) $PUSH r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 746) li $x50,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 747) $PUSH r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 748) li $x60,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 749) $PUSH r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 750) li $x70,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 751) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 752)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 753) subi $rounds,$rounds,3 # -4 in total
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 754) subi $len,$len,128 # bias
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 755)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 756) lvx $rndkey0,$x00,$key # load key schedule
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 757) lvx v30,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 758) addi $key,$key,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 759) lvx v31,$x00,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 760) ?vperm $rndkey0,$rndkey0,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 761) addi $key_,$sp,$FRAME+15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 762) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 763)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 764) Load_cbc_dec_key:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 765) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 766) lvx v30,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 767) addi $key,$key,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 768) stvx v24,$x00,$key_ # off-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 769) ?vperm v25,v31,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 770) lvx v31,$x00,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 771) stvx v25,$x10,$key_ # off-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 772) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 773) bdnz Load_cbc_dec_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 774)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 775) lvx v26,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 776) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 777) lvx v27,$x20,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 778) stvx v24,$x00,$key_ # off-load round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 779) ?vperm v25,v31,v26,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 780) lvx v28,$x30,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 781) stvx v25,$x10,$key_ # off-load round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 782) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 783) ?vperm v26,v26,v27,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 784) lvx v29,$x40,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 785) ?vperm v27,v27,v28,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 786) lvx v30,$x50,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 787) ?vperm v28,v28,v29,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 788) lvx v31,$x60,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 789) ?vperm v29,v29,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 790) lvx $out0,$x70,$key # borrow $out0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 791) ?vperm v30,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 792) lvx v24,$x00,$key_ # pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 793) ?vperm v31,v31,$out0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 794) lvx v25,$x10,$key_ # pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 795)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 796) #lvx $inptail,0,$inp # "caller" already did this
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 797) #addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 798) subi $inp,$inp,15 # undo "caller"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 799)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 800) le?li $idx,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 801) lvx_u $in0,$x00,$inp # load first 8 "words"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 802) le?lvsl $inpperm,0,$idx
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 803) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 804) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 805) le?vxor $inpperm,$inpperm,$tmp # transform for lvx_u/stvx_u
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 806) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 807) le?vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 808) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 809) le?vperm $in1,$in1,$in1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 810) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 811) le?vperm $in2,$in2,$in2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 812) vxor $out0,$in0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 813) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 814) le?vperm $in3,$in3,$in3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 815) vxor $out1,$in1,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 816) lvx_u $in6,$x60,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 817) le?vperm $in4,$in4,$in4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 818) vxor $out2,$in2,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 819) lvx_u $in7,$x70,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 820) addi $inp,$inp,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 821) le?vperm $in5,$in5,$in5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 822) vxor $out3,$in3,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 823) le?vperm $in6,$in6,$in6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 824) vxor $out4,$in4,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 825) le?vperm $in7,$in7,$in7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 826) vxor $out5,$in5,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 827) vxor $out6,$in6,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 828) vxor $out7,$in7,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 829)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 830) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 831) b Loop_cbc_dec8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 832) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 833) Loop_cbc_dec8x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 834) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 835) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 836) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 837) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 838) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 839) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 840) vncipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 841) vncipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 842) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 843) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 844)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 845) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 846) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 847) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 848) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 849) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 850) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 851) vncipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 852) vncipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 853) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 854) bdnz Loop_cbc_dec8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 855)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 856) subic $len,$len,128 # $len-=128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 857) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 858) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 859) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 860) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 861) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 862) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 863) vncipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 864) vncipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 865)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 866) subfe. r0,r0,r0 # borrow?-1:0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 867) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 868) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 869) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 870) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 871) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 872) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 873) vncipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 874) vncipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 875)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 876) and r0,r0,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 877) vncipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 878) vncipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 879) vncipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 880) vncipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 881) vncipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 882) vncipher $out5,$out5,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 883) vncipher $out6,$out6,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 884) vncipher $out7,$out7,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 885)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 886) add $inp,$inp,r0 # $inp is adjusted in such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 887) # way that at exit from the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 888) # loop inX-in7 are loaded
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 889) # with last "words"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 890) vncipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 891) vncipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 892) vncipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 893) vncipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 894) vncipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 895) vncipher $out5,$out5,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 896) vncipher $out6,$out6,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 897) vncipher $out7,$out7,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 898)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 899) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 900) vncipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 901) vncipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 902) vncipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 903) vncipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 904) vncipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 905) vncipher $out5,$out5,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 906) vncipher $out6,$out6,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 907) vncipher $out7,$out7,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 908) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 909)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 910) vncipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 911) vncipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 912) vncipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 913) vncipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 914) vncipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 915) vncipher $out5,$out5,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 916) vncipher $out6,$out6,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 917) vncipher $out7,$out7,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 918) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 919)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 920) vncipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 921) vxor $ivec,$ivec,v31 # xor with last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 922) vncipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 923) vxor $in0,$in0,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 924) vncipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 925) vxor $in1,$in1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 926) vncipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 927) vxor $in2,$in2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 928) vncipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 929) vxor $in3,$in3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 930) vncipher $out5,$out5,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 931) vxor $in4,$in4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 932) vncipher $out6,$out6,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 933) vxor $in5,$in5,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 934) vncipher $out7,$out7,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 935) vxor $in6,$in6,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 936)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 937) vncipherlast $out0,$out0,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 938) vncipherlast $out1,$out1,$in0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 939) lvx_u $in0,$x00,$inp # load next input block
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 940) vncipherlast $out2,$out2,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 941) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 942) vncipherlast $out3,$out3,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 943) le?vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 944) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 945) vncipherlast $out4,$out4,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 946) le?vperm $in1,$in1,$in1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 947) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 948) vncipherlast $out5,$out5,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 949) le?vperm $in2,$in2,$in2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 950) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 951) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 952) le?vperm $in3,$in3,$in3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 953) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 954) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 955) le?vperm $in4,$in4,$in4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 956) lvx_u $in6,$x60,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 957) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 958) le?vperm $in5,$in5,$in5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 959) lvx_u $in7,$x70,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 960) addi $inp,$inp,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 961)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 962) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 963) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 964) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 965) le?vperm $in6,$in6,$in6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 966) vxor $out0,$in0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 967) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 968) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 969) le?vperm $in7,$in7,$in7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 970) vxor $out1,$in1,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 971) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 972) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 973) vxor $out2,$in2,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 974) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 975) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 976) vxor $out3,$in3,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 977) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 978) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 979) vxor $out4,$in4,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 980) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 981) stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 982) vxor $out5,$in5,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 983) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 984) stvx_u $out6,$x60,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 985) vxor $out6,$in6,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 986) stvx_u $out7,$x70,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 987) addi $out,$out,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 988) vxor $out7,$in7,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 989)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 990) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 991) beq Loop_cbc_dec8x # did $len-=128 borrow?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 992)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 993) addic. $len,$len,128
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 994) beq Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 995) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 996) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 997)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 998) Loop_cbc_dec8x_tail: # up to 7 "words" tail...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 999) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1000) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1001) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1002) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1003) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1004) vncipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1005) vncipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1006) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1007) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1008)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1009) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1010) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1011) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1012) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1013) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1014) vncipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1015) vncipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1016) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1017) bdnz Loop_cbc_dec8x_tail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1018)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1019) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1020) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1021) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1022) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1023) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1024) vncipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1025) vncipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1026)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1027) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1028) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1029) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1030) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1031) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1032) vncipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1033) vncipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1034)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1035) vncipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1036) vncipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1037) vncipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1038) vncipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1039) vncipher $out5,$out5,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1040) vncipher $out6,$out6,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1041) vncipher $out7,$out7,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1042)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1043) vncipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1044) vncipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1045) vncipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1046) vncipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1047) vncipher $out5,$out5,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1048) vncipher $out6,$out6,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1049) vncipher $out7,$out7,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1050)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1051) vncipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1052) vncipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1053) vncipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1054) vncipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1055) vncipher $out5,$out5,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1056) vncipher $out6,$out6,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1057) vncipher $out7,$out7,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1058)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1059) vncipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1060) vncipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1061) vncipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1062) vncipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1063) vncipher $out5,$out5,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1064) vncipher $out6,$out6,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1065) vncipher $out7,$out7,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1066)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1067) vncipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1068) vxor $ivec,$ivec,v31 # last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1069) vncipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1070) vxor $in1,$in1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1071) vncipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1072) vxor $in2,$in2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1073) vncipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1074) vxor $in3,$in3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1075) vncipher $out5,$out5,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1076) vxor $in4,$in4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1077) vncipher $out6,$out6,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1078) vxor $in5,$in5,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1079) vncipher $out7,$out7,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1080) vxor $in6,$in6,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1081)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1082) cmplwi $len,32 # switch($len)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1083) blt Lcbc_dec8x_one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1084) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1085) beq Lcbc_dec8x_two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1086) cmplwi $len,64
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1087) blt Lcbc_dec8x_three
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1088) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1089) beq Lcbc_dec8x_four
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1090) cmplwi $len,96
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1091) blt Lcbc_dec8x_five
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1092) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1093) beq Lcbc_dec8x_six
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1094)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1095) Lcbc_dec8x_seven:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1096) vncipherlast $out1,$out1,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1097) vncipherlast $out2,$out2,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1098) vncipherlast $out3,$out3,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1099) vncipherlast $out4,$out4,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1100) vncipherlast $out5,$out5,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1101) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1102) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1103) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1104)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1105) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1106) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1107) stvx_u $out1,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1108) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1109) stvx_u $out2,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1110) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1111) stvx_u $out3,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1112) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1113) stvx_u $out4,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1114) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1115) stvx_u $out5,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1116) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1117) stvx_u $out6,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1118) stvx_u $out7,$x60,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1119) addi $out,$out,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1120) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1121)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1122) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1123) Lcbc_dec8x_six:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1124) vncipherlast $out2,$out2,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1125) vncipherlast $out3,$out3,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1126) vncipherlast $out4,$out4,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1127) vncipherlast $out5,$out5,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1128) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1129) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1130) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1131)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1132) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1133) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1134) stvx_u $out2,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1135) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1136) stvx_u $out3,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1137) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1138) stvx_u $out4,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1139) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1140) stvx_u $out5,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1141) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1142) stvx_u $out6,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1143) stvx_u $out7,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1144) addi $out,$out,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1145) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1146)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1147) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1148) Lcbc_dec8x_five:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1149) vncipherlast $out3,$out3,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1150) vncipherlast $out4,$out4,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1151) vncipherlast $out5,$out5,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1152) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1153) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1154) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1155)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1156) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1157) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1158) stvx_u $out3,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1159) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1160) stvx_u $out4,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1161) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1162) stvx_u $out5,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1163) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1164) stvx_u $out6,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1165) stvx_u $out7,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1166) addi $out,$out,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1167) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1168)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1169) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1170) Lcbc_dec8x_four:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1171) vncipherlast $out4,$out4,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1172) vncipherlast $out5,$out5,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1173) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1174) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1175) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1176)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1177) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1178) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1179) stvx_u $out4,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1180) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1181) stvx_u $out5,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1182) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1183) stvx_u $out6,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1184) stvx_u $out7,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1185) addi $out,$out,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1186) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1187)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1188) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1189) Lcbc_dec8x_three:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1190) vncipherlast $out5,$out5,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1191) vncipherlast $out6,$out6,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1192) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1193) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1194)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1195) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1196) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1197) stvx_u $out5,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1198) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1199) stvx_u $out6,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1200) stvx_u $out7,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1201) addi $out,$out,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1202) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1203)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1204) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1205) Lcbc_dec8x_two:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1206) vncipherlast $out6,$out6,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1207) vncipherlast $out7,$out7,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1208) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1209)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1210) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1211) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1212) stvx_u $out6,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1213) stvx_u $out7,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1214) addi $out,$out,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1215) b Lcbc_dec8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1216)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1217) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1218) Lcbc_dec8x_one:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1219) vncipherlast $out7,$out7,$ivec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1220) vmr $ivec,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1221)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1222) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1223) stvx_u $out7,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1224) addi $out,$out,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1225)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1226) Lcbc_dec8x_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1227) le?vperm $ivec,$ivec,$ivec,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1228) stvx_u $ivec,0,$ivp # write [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1229)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1230) li r10,`$FRAME+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1231) li r11,`$FRAME+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1232) stvx $inpperm,r10,$sp # wipe copies of round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1233) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1234) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1235) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1236) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1237) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1238) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1239) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1240) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1241) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1242) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1243) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1244) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1245) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1246) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1247) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1248)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1249) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1250) lvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1251) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1252) lvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1253) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1254) lvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1255) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1256) lvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1257) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1258) lvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1259) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1260) lvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1261) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1262) lvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1263) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1264) lvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1265) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1266) lvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1267) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1268) lvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1269) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1270) lvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1271) lvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1272) $POP r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1273) $POP r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1274) $POP r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1275) $POP r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1276) $POP r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1277) $POP r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1278) addi $sp,$sp,`$FRAME+21*16+6*$SIZE_T`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1279) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1280) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1281) .byte 0,12,0x14,0,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1282) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1283) .size .${prefix}_cbc_encrypt,.-.${prefix}_cbc_encrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1284) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1285) }} }}}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1286)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1287) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1288) {{{ # CTR procedure[s] #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1289)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1290) ####################### WARNING: Here be dragons! #######################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1291) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1292) # This code is written as 'ctr32', based on a 32-bit counter used
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1293) # upstream. The kernel does *not* use a 32-bit counter. The kernel uses
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1294) # a 128-bit counter.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1295) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1296) # This leads to subtle changes from the upstream code: the counter
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1297) # is incremented with vaddu_q_m rather than vaddu_w_m. This occurs in
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1298) # both the bulk (8 blocks at a time) path, and in the individual block
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1299) # path. Be aware of this when doing updates.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1300) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1301) # See:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1302) # 1d4aa0b4c181 ("crypto: vmx - Fixing AES-CTR counter bug")
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1303) # 009b30ac7444 ("crypto: vmx - CTR: always increment IV as quadword")
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1304) # https://github.com/openssl/openssl/pull/8942
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1305) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1306) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1307) my ($inp,$out,$len,$key,$ivp,$x10,$rounds,$idx)=map("r$_",(3..10));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1308) my ($rndkey0,$rndkey1,$inout,$tmp)= map("v$_",(0..3));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1309) my ($ivec,$inptail,$inpperm,$outhead,$outperm,$outmask,$keyperm,$one)=
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1310) map("v$_",(4..11));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1311) my $dat=$tmp;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1312)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1313) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1314) .globl .${prefix}_ctr32_encrypt_blocks
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1315) ${UCMP}i $len,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1316) bltlr-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1317)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1318) lis r0,0xfff0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1319) mfspr $vrsave,256
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1320) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1321)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1322) li $idx,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1323) vxor $rndkey0,$rndkey0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1324) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1325)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1326) lvx $ivec,0,$ivp # load [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1327) lvsl $inpperm,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1328) lvx $inptail,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1329) vspltisb $one,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1330) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1331) vperm $ivec,$ivec,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1332) vsldoi $one,$rndkey0,$one,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1333)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1334) neg r11,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1335) ?lvsl $keyperm,0,$key # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1336) lwz $rounds,240($key)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1337)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1338) lvsr $inpperm,0,r11 # prepare for unaligned load
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1339) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1340) addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1341) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1342)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1343) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1344) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1345) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1346)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1347) ${UCMP}i $len,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1348) bge _aesp8_ctr32_encrypt8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1349)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1350) ?lvsr $outperm,0,$out # prepare for unaligned store
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1351) vspltisb $outmask,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1352) lvx $outhead,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1353) ?vperm $outmask,$rndkey0,$outmask,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1354) le?vxor $outperm,$outperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1355)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1356) lvx $rndkey0,0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1357) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1358) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1359) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1360) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1361) vxor $inout,$ivec,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1362) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1363) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1364) b Loop_ctr32_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1365)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1366) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1367) Loop_ctr32_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1368) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1369) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1370) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1371) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1372) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1373) vcipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1374) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1375) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1376) bdnz Loop_ctr32_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1377)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1378) vadduqm $ivec,$ivec,$one # Kernel change for 128-bit
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1379) vmr $dat,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1380) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1381) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1382) subic. $len,$len,1 # blocks--
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1383)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1384) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1385) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1386) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1387) vperm $dat,$dat,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1388) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1389) ?vperm $rndkey1,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1390) lvx $rndkey0,0,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1391) vxor $dat,$dat,$rndkey1 # last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1392) vcipherlast $inout,$inout,$dat
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1393)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1394) lvx $rndkey1,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1395) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1396) vperm $inout,$inout,$inout,$outperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1397) vsel $dat,$outhead,$inout,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1398) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1399) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1400) vmr $outhead,$inout
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1401) vxor $inout,$ivec,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1402) lvx $rndkey0,$idx,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1403) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1404) stvx $dat,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1405) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1406) bne Loop_ctr32_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1407)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1408) addi $out,$out,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1409) lvx $inout,0,$out # redundant in aligned case
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1410) vsel $inout,$outhead,$inout,$outmask
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1411) stvx $inout,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1412)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1413) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1414) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1415) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1416) .byte 0,12,0x14,0,0,0,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1417) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1418) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1419) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1420) {{ # Optimized CTR procedure #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1421) my $key_="r11";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1422) my ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70)=map("r$_",(0,8,26..31));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1423) my ($in0, $in1, $in2, $in3, $in4, $in5, $in6, $in7 )=map("v$_",(0..3,10,12..14));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1424) my ($out0,$out1,$out2,$out3,$out4,$out5,$out6,$out7)=map("v$_",(15..22));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1425) my $rndkey0="v23"; # v24-v25 rotating buffer for first found keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1426) # v26-v31 last 6 round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1427) my ($tmp,$keyperm)=($in3,$in4); # aliases with "caller", redundant assignment
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1428) my ($two,$three,$four)=($outhead,$outperm,$outmask);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1429)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1430) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1431) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1432) _aesp8_ctr32_encrypt8x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1433) $STU $sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1434) li r10,`$FRAME+8*16+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1435) li r11,`$FRAME+8*16+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1436) stvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1437) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1438) stvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1439) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1440) stvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1441) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1442) stvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1443) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1444) stvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1445) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1446) stvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1447) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1448) stvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1449) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1450) stvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1451) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1452) stvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1453) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1454) stvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1455) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1456) stvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1457) stvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1458) li r0,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1459) stw $vrsave,`$FRAME+21*16-4`($sp) # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1460) li $x10,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1461) $PUSH r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1462) li $x20,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1463) $PUSH r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1464) li $x30,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1465) $PUSH r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1466) li $x40,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1467) $PUSH r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1468) li $x50,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1469) $PUSH r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1470) li $x60,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1471) $PUSH r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1472) li $x70,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1473) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1474)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1475) subi $rounds,$rounds,3 # -4 in total
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1476)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1477) lvx $rndkey0,$x00,$key # load key schedule
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1478) lvx v30,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1479) addi $key,$key,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1480) lvx v31,$x00,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1481) ?vperm $rndkey0,$rndkey0,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1482) addi $key_,$sp,$FRAME+15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1483) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1484)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1485) Load_ctr32_enc_key:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1486) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1487) lvx v30,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1488) addi $key,$key,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1489) stvx v24,$x00,$key_ # off-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1490) ?vperm v25,v31,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1491) lvx v31,$x00,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1492) stvx v25,$x10,$key_ # off-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1493) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1494) bdnz Load_ctr32_enc_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1495)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1496) lvx v26,$x10,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1497) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1498) lvx v27,$x20,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1499) stvx v24,$x00,$key_ # off-load round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1500) ?vperm v25,v31,v26,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1501) lvx v28,$x30,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1502) stvx v25,$x10,$key_ # off-load round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1503) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1504) ?vperm v26,v26,v27,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1505) lvx v29,$x40,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1506) ?vperm v27,v27,v28,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1507) lvx v30,$x50,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1508) ?vperm v28,v28,v29,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1509) lvx v31,$x60,$key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1510) ?vperm v29,v29,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1511) lvx $out0,$x70,$key # borrow $out0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1512) ?vperm v30,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1513) lvx v24,$x00,$key_ # pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1514) ?vperm v31,v31,$out0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1515) lvx v25,$x10,$key_ # pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1516)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1517) vadduqm $two,$one,$one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1518) subi $inp,$inp,15 # undo "caller"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1519) $SHL $len,$len,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1520)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1521) vadduqm $out1,$ivec,$one # counter values ...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1522) vadduqm $out2,$ivec,$two # (do all ctr adds as 128-bit)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1523) vxor $out0,$ivec,$rndkey0 # ... xored with rndkey[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1524) le?li $idx,8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1525) vadduqm $out3,$out1,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1526) vxor $out1,$out1,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1527) le?lvsl $inpperm,0,$idx
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1528) vadduqm $out4,$out2,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1529) vxor $out2,$out2,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1530) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1531) vadduqm $out5,$out3,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1532) vxor $out3,$out3,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1533) le?vxor $inpperm,$inpperm,$tmp # transform for lvx_u/stvx_u
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1534) vadduqm $out6,$out4,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1535) vxor $out4,$out4,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1536) vadduqm $out7,$out5,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1537) vxor $out5,$out5,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1538) vadduqm $ivec,$out6,$two # next counter value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1539) vxor $out6,$out6,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1540) vxor $out7,$out7,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1541)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1542) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1543) b Loop_ctr32_enc8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1544) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1545) Loop_ctr32_enc8x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1546) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1547) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1548) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1549) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1550) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1551) vcipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1552) vcipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1553) vcipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1554) Loop_ctr32_enc8x_middle:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1555) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1556) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1557)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1558) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1559) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1560) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1561) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1562) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1563) vcipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1564) vcipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1565) vcipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1566) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1567) bdnz Loop_ctr32_enc8x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1568)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1569) subic r11,$len,256 # $len-256, borrow $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1570) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1571) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1572) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1573) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1574) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1575) vcipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1576) vcipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1577) vcipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1578)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1579) subfe r0,r0,r0 # borrow?-1:0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1580) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1581) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1582) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1583) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1584) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1585) vcipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1586) vcipher $out6,$out6,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1587) vcipher $out7,$out7,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1588)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1589) and r0,r0,r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1590) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1591) vcipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1592) vcipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1593) vcipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1594) vcipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1595) vcipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1596) vcipher $out5,$out5,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1597) vcipher $out6,$out6,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1598) vcipher $out7,$out7,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1599) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1600)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1601) subic $len,$len,129 # $len-=129
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1602) vcipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1603) addi $len,$len,1 # $len-=128 really
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1604) vcipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1605) vcipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1606) vcipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1607) vcipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1608) vcipher $out5,$out5,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1609) vcipher $out6,$out6,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1610) vcipher $out7,$out7,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1611) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1612)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1613) vcipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1614) lvx_u $in0,$x00,$inp # load input
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1615) vcipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1616) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1617) vcipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1618) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1619) vcipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1620) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1621) vcipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1622) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1623) vcipher $out5,$out5,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1624) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1625) vcipher $out6,$out6,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1626) lvx_u $in6,$x60,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1627) vcipher $out7,$out7,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1628) lvx_u $in7,$x70,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1629) addi $inp,$inp,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1630)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1631) vcipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1632) le?vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1633) vcipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1634) le?vperm $in1,$in1,$in1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1635) vcipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1636) le?vperm $in2,$in2,$in2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1637) vcipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1638) le?vperm $in3,$in3,$in3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1639) vcipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1640) le?vperm $in4,$in4,$in4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1641) vcipher $out5,$out5,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1642) le?vperm $in5,$in5,$in5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1643) vcipher $out6,$out6,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1644) le?vperm $in6,$in6,$in6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1645) vcipher $out7,$out7,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1646) le?vperm $in7,$in7,$in7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1647)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1648) add $inp,$inp,r0 # $inp is adjusted in such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1649) # way that at exit from the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1650) # loop inX-in7 are loaded
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1651) # with last "words"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1652) subfe. r0,r0,r0 # borrow?-1:0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1653) vcipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1654) vxor $in0,$in0,v31 # xor with last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1655) vcipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1656) vxor $in1,$in1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1657) vcipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1658) vxor $in2,$in2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1659) vcipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1660) vxor $in3,$in3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1661) vcipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1662) vxor $in4,$in4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1663) vcipher $out5,$out5,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1664) vxor $in5,$in5,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1665) vcipher $out6,$out6,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1666) vxor $in6,$in6,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1667) vcipher $out7,$out7,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1668) vxor $in7,$in7,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1669)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1670) bne Lctr32_enc8x_break # did $len-129 borrow?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1671)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1672) vcipherlast $in0,$out0,$in0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1673) vcipherlast $in1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1674) vadduqm $out1,$ivec,$one # counter values ...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1675) vcipherlast $in2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1676) vadduqm $out2,$ivec,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1677) vxor $out0,$ivec,$rndkey0 # ... xored with rndkey[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1678) vcipherlast $in3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1679) vadduqm $out3,$out1,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1680) vxor $out1,$out1,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1681) vcipherlast $in4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1682) vadduqm $out4,$out2,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1683) vxor $out2,$out2,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1684) vcipherlast $in5,$out5,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1685) vadduqm $out5,$out3,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1686) vxor $out3,$out3,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1687) vcipherlast $in6,$out6,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1688) vadduqm $out6,$out4,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1689) vxor $out4,$out4,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1690) vcipherlast $in7,$out7,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1691) vadduqm $out7,$out5,$two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1692) vxor $out5,$out5,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1693) le?vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1694) vadduqm $ivec,$out6,$two # next counter value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1695) vxor $out6,$out6,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1696) le?vperm $in1,$in1,$in1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1697) vxor $out7,$out7,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1698) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1699)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1700) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1701) stvx_u $in0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1702) le?vperm $in2,$in2,$in2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1703) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1704) stvx_u $in1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1705) le?vperm $in3,$in3,$in3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1706) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1707) stvx_u $in2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1708) le?vperm $in4,$in4,$in4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1709) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1710) stvx_u $in3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1711) le?vperm $in5,$in5,$in5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1712) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1713) stvx_u $in4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1714) le?vperm $in6,$in6,$in6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1715) vcipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1716) stvx_u $in5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1717) le?vperm $in7,$in7,$in7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1718) vcipher $out6,$out6,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1719) stvx_u $in6,$x60,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1720) vcipher $out7,$out7,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1721) stvx_u $in7,$x70,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1722) addi $out,$out,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1723)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1724) b Loop_ctr32_enc8x_middle
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1725)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1726) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1727) Lctr32_enc8x_break:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1728) cmpwi $len,-0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1729) blt Lctr32_enc8x_one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1730) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1731) beq Lctr32_enc8x_two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1732) cmpwi $len,-0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1733) blt Lctr32_enc8x_three
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1734) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1735) beq Lctr32_enc8x_four
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1736) cmpwi $len,-0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1737) blt Lctr32_enc8x_five
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1738) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1739) beq Lctr32_enc8x_six
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1740) cmpwi $len,0x00
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1741) blt Lctr32_enc8x_seven
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1742)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1743) Lctr32_enc8x_eight:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1744) vcipherlast $out0,$out0,$in0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1745) vcipherlast $out1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1746) vcipherlast $out2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1747) vcipherlast $out3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1748) vcipherlast $out4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1749) vcipherlast $out5,$out5,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1750) vcipherlast $out6,$out6,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1751) vcipherlast $out7,$out7,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1752)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1753) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1754) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1755) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1756) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1757) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1758) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1759) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1760) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1761) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1762) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1763) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1764) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1765) stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1766) le?vperm $out7,$out7,$out7,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1767) stvx_u $out6,$x60,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1768) stvx_u $out7,$x70,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1769) addi $out,$out,0x80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1770) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1771)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1772) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1773) Lctr32_enc8x_seven:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1774) vcipherlast $out0,$out0,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1775) vcipherlast $out1,$out1,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1776) vcipherlast $out2,$out2,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1777) vcipherlast $out3,$out3,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1778) vcipherlast $out4,$out4,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1779) vcipherlast $out5,$out5,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1780) vcipherlast $out6,$out6,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1781)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1782) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1783) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1784) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1785) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1786) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1787) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1788) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1789) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1790) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1791) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1792) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1793) le?vperm $out6,$out6,$out6,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1794) stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1795) stvx_u $out6,$x60,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1796) addi $out,$out,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1797) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1798)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1799) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1800) Lctr32_enc8x_six:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1801) vcipherlast $out0,$out0,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1802) vcipherlast $out1,$out1,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1803) vcipherlast $out2,$out2,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1804) vcipherlast $out3,$out3,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1805) vcipherlast $out4,$out4,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1806) vcipherlast $out5,$out5,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1807)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1808) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1809) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1810) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1811) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1812) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1813) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1814) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1815) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1816) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1817) le?vperm $out5,$out5,$out5,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1818) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1819) stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1820) addi $out,$out,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1821) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1822)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1823) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1824) Lctr32_enc8x_five:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1825) vcipherlast $out0,$out0,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1826) vcipherlast $out1,$out1,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1827) vcipherlast $out2,$out2,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1828) vcipherlast $out3,$out3,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1829) vcipherlast $out4,$out4,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1830)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1831) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1832) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1833) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1834) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1835) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1836) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1837) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1838) le?vperm $out4,$out4,$out4,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1839) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1840) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1841) addi $out,$out,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1842) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1843)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1844) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1845) Lctr32_enc8x_four:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1846) vcipherlast $out0,$out0,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1847) vcipherlast $out1,$out1,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1848) vcipherlast $out2,$out2,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1849) vcipherlast $out3,$out3,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1850)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1851) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1852) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1853) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1854) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1855) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1856) le?vperm $out3,$out3,$out3,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1857) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1858) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1859) addi $out,$out,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1860) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1861)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1862) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1863) Lctr32_enc8x_three:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1864) vcipherlast $out0,$out0,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1865) vcipherlast $out1,$out1,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1866) vcipherlast $out2,$out2,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1867)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1868) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1869) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1870) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1871) le?vperm $out2,$out2,$out2,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1872) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1873) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1874) addi $out,$out,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1875) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1876)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1877) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1878) Lctr32_enc8x_two:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1879) vcipherlast $out0,$out0,$in6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1880) vcipherlast $out1,$out1,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1881)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1882) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1883) le?vperm $out1,$out1,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1884) stvx_u $out0,$x00,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1885) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1886) addi $out,$out,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1887) b Lctr32_enc8x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1888)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1889) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1890) Lctr32_enc8x_one:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1891) vcipherlast $out0,$out0,$in7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1892)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1893) le?vperm $out0,$out0,$out0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1894) stvx_u $out0,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1895) addi $out,$out,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1896)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1897) Lctr32_enc8x_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1898) li r10,`$FRAME+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1899) li r11,`$FRAME+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1900) stvx $inpperm,r10,$sp # wipe copies of round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1901) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1902) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1903) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1904) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1905) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1906) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1907) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1908) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1909) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1910) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1911) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1912) stvx $inpperm,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1913) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1914) stvx $inpperm,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1915) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1916)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1917) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1918) lvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1919) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1920) lvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1921) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1922) lvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1923) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1924) lvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1925) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1926) lvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1927) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1928) lvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1929) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1930) lvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1931) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1932) lvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1933) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1934) lvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1935) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1936) lvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1937) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1938) lvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1939) lvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1940) $POP r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1941) $POP r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1942) $POP r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1943) $POP r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1944) $POP r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1945) $POP r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1946) addi $sp,$sp,`$FRAME+21*16+6*$SIZE_T`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1947) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1948) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1949) .byte 0,12,0x14,0,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1950) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1951) .size .${prefix}_ctr32_encrypt_blocks,.-.${prefix}_ctr32_encrypt_blocks
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1952) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1953) }} }}}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1954)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1955) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1956) {{{ # XTS procedures #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1957) # int aes_p8_xts_[en|de]crypt(const char *inp, char *out, size_t len, #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1958) # const AES_KEY *key1, const AES_KEY *key2, #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1959) # [const] unsigned char iv[16]); #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1960) # If $key2 is NULL, then a "tweak chaining" mode is engaged, in which #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1961) # input tweak value is assumed to be encrypted already, and last tweak #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1962) # value, one suitable for consecutive call on same chunk of data, is #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1963) # written back to original buffer. In addition, in "tweak chaining" #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1964) # mode only complete input blocks are processed. #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1965)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1966) my ($inp,$out,$len,$key1,$key2,$ivp,$rounds,$idx) = map("r$_",(3..10));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1967) my ($rndkey0,$rndkey1,$inout) = map("v$_",(0..2));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1968) my ($output,$inptail,$inpperm,$leperm,$keyperm) = map("v$_",(3..7));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1969) my ($tweak,$seven,$eighty7,$tmp,$tweak1) = map("v$_",(8..12));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1970) my $taillen = $key2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1971)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1972) ($inp,$idx) = ($idx,$inp); # reassign
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1973)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1974) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1975) .globl .${prefix}_xts_encrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1976) mr $inp,r3 # reassign
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1977) li r3,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1978) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1979) bltlr-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1980)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1981) lis r0,0xfff0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1982) mfspr r12,256 # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1983) li r11,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1984) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1985)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1986) vspltisb $seven,0x07 # 0x070707..07
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1987) le?lvsl $leperm,r11,r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1988) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1989) le?vxor $leperm,$leperm,$seven
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1990)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1991) li $idx,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1992) lvx $tweak,0,$ivp # load [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1993) lvsl $inpperm,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1994) lvx $inptail,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1995) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1996) vperm $tweak,$tweak,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1997)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1998) neg r11,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1999) lvsr $inpperm,0,r11 # prepare for unaligned load
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2000) lvx $inout,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2001) addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2002) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2003)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2004) ${UCMP}i $key2,0 # key2==NULL?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2005) beq Lxts_enc_no_key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2006)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2007) ?lvsl $keyperm,0,$key2 # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2008) lwz $rounds,240($key2)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2009) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2010) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2011) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2012)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2013) lvx $rndkey0,0,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2014) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2015) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2016) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2017) vxor $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2018) lvx $rndkey0,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2019) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2020) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2021)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2022) Ltweak_xts_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2023) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2024) vcipher $tweak,$tweak,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2025) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2026) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2027) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2028) vcipher $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2029) lvx $rndkey0,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2030) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2031) bdnz Ltweak_xts_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2032)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2033) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2034) vcipher $tweak,$tweak,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2035) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2036) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2037) vcipherlast $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2038)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2039) li $ivp,0 # don't chain the tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2040) b Lxts_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2041)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2042) Lxts_enc_no_key2:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2043) li $idx,-16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2044) and $len,$len,$idx # in "tweak chaining"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2045) # mode only complete
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2046) # blocks are processed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2047) Lxts_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2048) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2049) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2050)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2051) ?lvsl $keyperm,0,$key1 # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2052) lwz $rounds,240($key1)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2053) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2054) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2055) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2056)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2057) vslb $eighty7,$seven,$seven # 0x808080..80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2058) vor $eighty7,$eighty7,$seven # 0x878787..87
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2059) vspltisb $tmp,1 # 0x010101..01
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2060) vsldoi $eighty7,$eighty7,$tmp,15 # 0x870101..01
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2061)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2062) ${UCMP}i $len,96
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2063) bge _aesp8_xts_encrypt6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2064)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2065) andi. $taillen,$len,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2066) subic r0,$len,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2067) subi $taillen,$taillen,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2068) subfe r0,r0,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2069) and r0,r0,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2070) add $inp,$inp,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2071)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2072) lvx $rndkey0,0,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2073) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2074) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2075) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2076) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2077) vxor $inout,$inout,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2078) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2079) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2080) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2081) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2082) b Loop_xts_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2083)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2084) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2085) Loop_xts_enc:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2086) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2087) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2088) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2089) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2090) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2091) vcipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2092) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2093) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2094) bdnz Loop_xts_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2095)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2096) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2097) vcipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2098) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2099) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2100) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2101) vxor $rndkey0,$rndkey0,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2102) vcipherlast $output,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2103)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2104) le?vperm $tmp,$output,$output,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2105) be?nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2106) le?stvx_u $tmp,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2107) be?stvx_u $output,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2108) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2109)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2110) subic. $len,$len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2111) beq Lxts_enc_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2112)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2113) vmr $inout,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2114) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2115) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2116) lvx $rndkey0,0,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2117) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2118) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2119)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2120) subic r0,$len,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2121) subfe r0,r0,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2122) and r0,r0,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2123) add $inp,$inp,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2124)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2125) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2126) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2127) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2128) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2129) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2130)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2131) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2132) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2133) vxor $inout,$inout,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2134) vxor $output,$output,$rndkey0 # just in case $len<16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2135) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2136) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2137) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2138)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2139) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2140) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2141) bge Loop_xts_enc
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2142)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2143) vxor $output,$output,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2144) lvsr $inpperm,0,$len # $inpperm is no longer needed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2145) vxor $inptail,$inptail,$inptail # $inptail is no longer needed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2146) vspltisb $tmp,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2147) vperm $inptail,$inptail,$tmp,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2148) vsel $inout,$inout,$output,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2149)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2150) subi r11,$out,17
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2151) subi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2152) mtctr $len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2153) li $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2154) Loop_xts_enc_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2155) lbzu r0,1(r11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2156) stb r0,16(r11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2157) bdnz Loop_xts_enc_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2158)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2159) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2160) b Loop_xts_enc # one more time...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2161)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2162) Lxts_enc_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2163) ${UCMP}i $ivp,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2164) beq Lxts_enc_ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2165)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2166) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2167) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2168) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2169) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2170) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2171)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2172) le?vperm $tweak,$tweak,$tweak,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2173) stvx_u $tweak,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2174)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2175) Lxts_enc_ret:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2176) mtspr 256,r12 # restore vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2177) li r3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2178) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2179) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2180) .byte 0,12,0x04,0,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2181) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2182) .size .${prefix}_xts_encrypt,.-.${prefix}_xts_encrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2183)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2184) .globl .${prefix}_xts_decrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2185) mr $inp,r3 # reassign
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2186) li r3,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2187) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2188) bltlr-
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2189)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2190) lis r0,0xfff8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2191) mfspr r12,256 # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2192) li r11,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2193) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2194)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2195) andi. r0,$len,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2196) neg r0,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2197) andi. r0,r0,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2198) sub $len,$len,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2199)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2200) vspltisb $seven,0x07 # 0x070707..07
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2201) le?lvsl $leperm,r11,r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2202) le?vspltisb $tmp,0x0f
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2203) le?vxor $leperm,$leperm,$seven
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2204)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2205) li $idx,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2206) lvx $tweak,0,$ivp # load [unaligned] iv
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2207) lvsl $inpperm,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2208) lvx $inptail,$idx,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2209) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2210) vperm $tweak,$tweak,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2211)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2212) neg r11,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2213) lvsr $inpperm,0,r11 # prepare for unaligned load
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2214) lvx $inout,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2215) addi $inp,$inp,15 # 15 is not typo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2216) le?vxor $inpperm,$inpperm,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2217)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2218) ${UCMP}i $key2,0 # key2==NULL?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2219) beq Lxts_dec_no_key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2220)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2221) ?lvsl $keyperm,0,$key2 # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2222) lwz $rounds,240($key2)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2223) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2224) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2225) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2226)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2227) lvx $rndkey0,0,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2228) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2229) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2230) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2231) vxor $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2232) lvx $rndkey0,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2233) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2234) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2235)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2236) Ltweak_xts_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2237) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2238) vcipher $tweak,$tweak,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2239) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2240) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2241) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2242) vcipher $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2243) lvx $rndkey0,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2244) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2245) bdnz Ltweak_xts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2246)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2247) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2248) vcipher $tweak,$tweak,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2249) lvx $rndkey1,$idx,$key2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2250) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2251) vcipherlast $tweak,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2252)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2253) li $ivp,0 # don't chain the tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2254) b Lxts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2255)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2256) Lxts_dec_no_key2:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2257) neg $idx,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2258) andi. $idx,$idx,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2259) add $len,$len,$idx # in "tweak chaining"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2260) # mode only complete
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2261) # blocks are processed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2262) Lxts_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2263) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2264) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2265)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2266) ?lvsl $keyperm,0,$key1 # prepare for unaligned key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2267) lwz $rounds,240($key1)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2268) srwi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2269) subi $rounds,$rounds,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2270) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2271)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2272) vslb $eighty7,$seven,$seven # 0x808080..80
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2273) vor $eighty7,$eighty7,$seven # 0x878787..87
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2274) vspltisb $tmp,1 # 0x010101..01
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2275) vsldoi $eighty7,$eighty7,$tmp,15 # 0x870101..01
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2276)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2277) ${UCMP}i $len,96
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2278) bge _aesp8_xts_decrypt6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2279)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2280) lvx $rndkey0,0,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2281) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2282) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2283) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2284) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2285) vxor $inout,$inout,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2286) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2287) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2288) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2289) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2290)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2291) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2292) blt Ltail_xts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2293) be?b Loop_xts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2294)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2295) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2296) Loop_xts_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2297) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2298) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2299) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2300) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2301) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2302) vncipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2303) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2304) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2305) bdnz Loop_xts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2306)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2307) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2308) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2309) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2310) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2311) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2312) vxor $rndkey0,$rndkey0,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2313) vncipherlast $output,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2314)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2315) le?vperm $tmp,$output,$output,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2316) be?nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2317) le?stvx_u $tmp,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2318) be?stvx_u $output,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2319) addi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2320)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2321) subic. $len,$len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2322) beq Lxts_dec_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2323)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2324) vmr $inout,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2325) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2326) addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2327) lvx $rndkey0,0,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2328) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2329) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2330)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2331) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2332) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2333) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2334) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2335) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2336)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2337) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2338) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2339) vxor $inout,$inout,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2340) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2341) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2342) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2343)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2344) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2345) ${UCMP}i $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2346) bge Loop_xts_dec
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2347)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2348) Ltail_xts_dec:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2349) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2350) vaddubm $tweak1,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2351) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2352) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2353) vxor $tweak1,$tweak1,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2354)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2355) subi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2356) add $inp,$inp,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2357)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2358) vxor $inout,$inout,$tweak # :-(
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2359) vxor $inout,$inout,$tweak1 # :-)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2360)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2361) Loop_xts_dec_short:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2362) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2363) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2364) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2365) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2366) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2367) vncipher $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2368) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2369) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2370) bdnz Loop_xts_dec_short
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2371)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2372) ?vperm $rndkey1,$rndkey1,$rndkey0,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2373) vncipher $inout,$inout,$rndkey1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2374) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2375) li $idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2376) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2377) vxor $rndkey0,$rndkey0,$tweak1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2378) vncipherlast $output,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2379)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2380) le?vperm $tmp,$output,$output,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2381) be?nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2382) le?stvx_u $tmp,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2383) be?stvx_u $output,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2384)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2385) vmr $inout,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2386) lvx $inptail,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2387) #addi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2388) lvx $rndkey0,0,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2389) lvx $rndkey1,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2390) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2391) vperm $inout,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2392) ?vperm $rndkey0,$rndkey0,$rndkey1,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2393)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2394) lvsr $inpperm,0,$len # $inpperm is no longer needed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2395) vxor $inptail,$inptail,$inptail # $inptail is no longer needed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2396) vspltisb $tmp,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2397) vperm $inptail,$inptail,$tmp,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2398) vsel $inout,$inout,$output,$inptail
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2399)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2400) vxor $rndkey0,$rndkey0,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2401) vxor $inout,$inout,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2402) lvx $rndkey0,$idx,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2403) addi $idx,$idx,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2404)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2405) subi r11,$out,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2406) mtctr $len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2407) li $len,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2408) Loop_xts_dec_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2409) lbzu r0,1(r11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2410) stb r0,16(r11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2411) bdnz Loop_xts_dec_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2412)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2413) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2414) b Loop_xts_dec # one more time...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2415)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2416) Lxts_dec_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2417) ${UCMP}i $ivp,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2418) beq Lxts_dec_ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2419)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2420) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2421) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2422) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2423) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2424) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2425)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2426) le?vperm $tweak,$tweak,$tweak,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2427) stvx_u $tweak,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2428)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2429) Lxts_dec_ret:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2430) mtspr 256,r12 # restore vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2431) li r3,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2432) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2433) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2434) .byte 0,12,0x04,0,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2435) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2436) .size .${prefix}_xts_decrypt,.-.${prefix}_xts_decrypt
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2437) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2438) #########################################################################
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2439) {{ # Optimized XTS procedures #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2440) my $key_=$key2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2441) my ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70)=map("r$_",(0,3,26..31));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2442) $x00=0 if ($flavour =~ /osx/);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2443) my ($in0, $in1, $in2, $in3, $in4, $in5 )=map("v$_",(0..5));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2444) my ($out0, $out1, $out2, $out3, $out4, $out5)=map("v$_",(7,12..16));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2445) my ($twk0, $twk1, $twk2, $twk3, $twk4, $twk5)=map("v$_",(17..22));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2446) my $rndkey0="v23"; # v24-v25 rotating buffer for first found keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2447) # v26-v31 last 6 round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2448) my ($keyperm)=($out0); # aliases with "caller", redundant assignment
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2449) my $taillen=$x70;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2450)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2451) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2452) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2453) _aesp8_xts_encrypt6x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2454) $STU $sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2455) mflr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2456) li r7,`$FRAME+8*16+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2457) li r3,`$FRAME+8*16+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2458) $PUSH r11,`$FRAME+21*16+6*$SIZE_T+$LRSAVE`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2459) stvx v20,r7,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2460) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2461) stvx v21,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2462) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2463) stvx v22,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2464) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2465) stvx v23,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2466) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2467) stvx v24,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2468) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2469) stvx v25,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2470) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2471) stvx v26,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2472) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2473) stvx v27,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2474) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2475) stvx v28,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2476) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2477) stvx v29,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2478) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2479) stvx v30,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2480) stvx v31,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2481) li r0,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2482) stw $vrsave,`$FRAME+21*16-4`($sp) # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2483) li $x10,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2484) $PUSH r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2485) li $x20,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2486) $PUSH r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2487) li $x30,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2488) $PUSH r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2489) li $x40,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2490) $PUSH r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2491) li $x50,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2492) $PUSH r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2493) li $x60,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2494) $PUSH r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2495) li $x70,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2496) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2497)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2498) subi $rounds,$rounds,3 # -4 in total
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2499)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2500) lvx $rndkey0,$x00,$key1 # load key schedule
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2501) lvx v30,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2502) addi $key1,$key1,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2503) lvx v31,$x00,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2504) ?vperm $rndkey0,$rndkey0,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2505) addi $key_,$sp,$FRAME+15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2506) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2507)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2508) Load_xts_enc_key:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2509) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2510) lvx v30,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2511) addi $key1,$key1,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2512) stvx v24,$x00,$key_ # off-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2513) ?vperm v25,v31,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2514) lvx v31,$x00,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2515) stvx v25,$x10,$key_ # off-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2516) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2517) bdnz Load_xts_enc_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2518)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2519) lvx v26,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2520) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2521) lvx v27,$x20,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2522) stvx v24,$x00,$key_ # off-load round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2523) ?vperm v25,v31,v26,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2524) lvx v28,$x30,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2525) stvx v25,$x10,$key_ # off-load round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2526) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2527) ?vperm v26,v26,v27,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2528) lvx v29,$x40,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2529) ?vperm v27,v27,v28,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2530) lvx v30,$x50,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2531) ?vperm v28,v28,v29,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2532) lvx v31,$x60,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2533) ?vperm v29,v29,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2534) lvx $twk5,$x70,$key1 # borrow $twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2535) ?vperm v30,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2536) lvx v24,$x00,$key_ # pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2537) ?vperm v31,v31,$twk5,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2538) lvx v25,$x10,$key_ # pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2539)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2540) vperm $in0,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2541) subi $inp,$inp,31 # undo "caller"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2542) vxor $twk0,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2543) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2544) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2545) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2546) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2547) vxor $out0,$in0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2548) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2549)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2550) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2551) vxor $twk1,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2552) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2553) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2554) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2555) le?vperm $in1,$in1,$in1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2556) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2557) vxor $out1,$in1,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2558) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2559)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2560) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2561) andi. $taillen,$len,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2562) vxor $twk2,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2563) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2564) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2565) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2566) le?vperm $in2,$in2,$in2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2567) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2568) vxor $out2,$in2,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2569) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2570)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2571) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2572) sub $len,$len,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2573) vxor $twk3,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2574) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2575) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2576) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2577) le?vperm $in3,$in3,$in3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2578) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2579) vxor $out3,$in3,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2580) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2581)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2582) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2583) subi $len,$len,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2584) vxor $twk4,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2585) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2586) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2587) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2588) le?vperm $in4,$in4,$in4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2589) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2590) vxor $out4,$in4,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2591) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2592)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2593) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2594) addi $inp,$inp,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2595) vxor $twk5,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2596) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2597) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2598) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2599) le?vperm $in5,$in5,$in5,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2600) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2601) vxor $out5,$in5,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2602) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2603)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2604) vxor v31,v31,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2605) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2606) b Loop_xts_enc6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2607)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2608) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2609) Loop_xts_enc6x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2610) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2611) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2612) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2613) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2614) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2615) vcipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2616) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2617) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2618)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2619) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2620) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2621) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2622) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2623) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2624) vcipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2625) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2626) bdnz Loop_xts_enc6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2627)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2628) subic $len,$len,96 # $len-=96
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2629) vxor $in0,$twk0,v31 # xor with last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2630) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2631) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2632) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2633) vxor $twk0,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2634) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2635) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2636) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2637) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2638) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2639) vcipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2640)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2641) subfe. r0,r0,r0 # borrow?-1:0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2642) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2643) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2644) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2645) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2646) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2647) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2648) vxor $in1,$twk1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2649) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2650) vxor $twk1,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2651) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2652) vcipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2653)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2654) and r0,r0,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2655) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2656) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2657) vcipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2658) vcipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2659) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2660) vcipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2661) vcipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2662) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2663) vcipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2664) vcipher $out5,$out5,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2665)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2666) add $inp,$inp,r0 # $inp is adjusted in such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2667) # way that at exit from the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2668) # loop inX-in5 are loaded
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2669) # with last "words"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2670) vxor $in2,$twk2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2671) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2672) vxor $twk2,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2673) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2674) vcipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2675) vcipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2676) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2677) vcipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2678) vcipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2679) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2680) vcipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2681) vcipher $out5,$out5,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2682)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2683) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2684) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2685) vcipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2686) vcipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2687) vxor $in3,$twk3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2688) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2689) vxor $twk3,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2690) vcipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2691) vcipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2692) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2693) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2694) vcipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2695) vcipher $out5,$out5,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2696) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2697) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2698)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2699) vcipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2700) vcipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2701) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2702) vcipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2703) vcipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2704) vxor $in4,$twk4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2705) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2706) vxor $twk4,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2707) vcipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2708) vcipher $out5,$out5,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2709) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2710) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2711) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2712)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2713) vcipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2714) vcipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2715) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2716) vcipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2717) vcipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2718) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2719) vcipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2720) vcipher $out5,$out5,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2721) vxor $in5,$twk5,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2722) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2723) vxor $twk5,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2724)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2725) vcipherlast $out0,$out0,$in0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2726) lvx_u $in0,$x00,$inp # load next input block
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2727) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2728) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2729) vcipherlast $out1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2730) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2731) vcipherlast $out2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2732) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2733) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2734) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2735) vcipherlast $out3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2736) le?vperm $in1,$in1,$in1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2737) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2738) vcipherlast $out4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2739) le?vperm $in2,$in2,$in2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2740) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2741) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2742) vcipherlast $tmp,$out5,$in5 # last block might be needed
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2743) # in stealing mode
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2744) le?vperm $in3,$in3,$in3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2745) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2746) addi $inp,$inp,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2747) le?vperm $in4,$in4,$in4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2748) le?vperm $in5,$in5,$in5,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2749)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2750) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2751) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2752) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2753) vxor $out0,$in0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2754) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2755) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2756) vxor $out1,$in1,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2757) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2758) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2759) vxor $out2,$in2,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2760) le?vperm $out4,$out4,$out4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2761) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2762) vxor $out3,$in3,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2763) le?vperm $out5,$tmp,$tmp,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2764) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2765) vxor $out4,$in4,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2766) le?stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2767) be?stvx_u $tmp, $x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2768) vxor $out5,$in5,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2769) addi $out,$out,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2770)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2771) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2772) beq Loop_xts_enc6x # did $len-=96 borrow?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2773)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2774) addic. $len,$len,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2775) beq Lxts_enc6x_zero
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2776) cmpwi $len,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2777) blt Lxts_enc6x_one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2778) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2779) beq Lxts_enc6x_two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2780) cmpwi $len,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2781) blt Lxts_enc6x_three
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2782) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2783) beq Lxts_enc6x_four
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2784)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2785) Lxts_enc6x_five:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2786) vxor $out0,$in1,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2787) vxor $out1,$in2,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2788) vxor $out2,$in3,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2789) vxor $out3,$in4,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2790) vxor $out4,$in5,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2791)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2792) bl _aesp8_xts_enc5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2793)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2794) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2795) vmr $twk0,$twk5 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2796) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2797) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2798) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2799) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2800) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2801) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2802) vxor $tmp,$out4,$twk5 # last block prep for stealing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2803) le?vperm $out4,$out4,$out4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2804) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2805) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2806) addi $out,$out,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2807) bne Lxts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2808) b Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2809)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2810) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2811) Lxts_enc6x_four:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2812) vxor $out0,$in2,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2813) vxor $out1,$in3,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2814) vxor $out2,$in4,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2815) vxor $out3,$in5,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2816) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2817)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2818) bl _aesp8_xts_enc5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2819)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2820) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2821) vmr $twk0,$twk4 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2822) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2823) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2824) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2825) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2826) vxor $tmp,$out3,$twk4 # last block prep for stealing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2827) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2828) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2829) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2830) addi $out,$out,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2831) bne Lxts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2832) b Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2833)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2834) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2835) Lxts_enc6x_three:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2836) vxor $out0,$in3,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2837) vxor $out1,$in4,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2838) vxor $out2,$in5,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2839) vxor $out3,$out3,$out3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2840) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2841)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2842) bl _aesp8_xts_enc5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2843)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2844) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2845) vmr $twk0,$twk3 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2846) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2847) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2848) vxor $tmp,$out2,$twk3 # last block prep for stealing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2849) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2850) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2851) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2852) addi $out,$out,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2853) bne Lxts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2854) b Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2855)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2856) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2857) Lxts_enc6x_two:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2858) vxor $out0,$in4,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2859) vxor $out1,$in5,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2860) vxor $out2,$out2,$out2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2861) vxor $out3,$out3,$out3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2862) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2863)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2864) bl _aesp8_xts_enc5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2865)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2866) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2867) vmr $twk0,$twk2 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2868) vxor $tmp,$out1,$twk2 # last block prep for stealing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2869) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2870) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2871) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2872) addi $out,$out,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2873) bne Lxts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2874) b Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2875)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2876) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2877) Lxts_enc6x_one:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2878) vxor $out0,$in5,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2879) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2880) Loop_xts_enc1x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2881) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2882) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2883) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2884)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2885) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2886) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2887) bdnz Loop_xts_enc1x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2888)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2889) add $inp,$inp,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2890) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2891) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2892)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2893) subi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2894) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2895)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2896) lvsr $inpperm,0,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2897) vcipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2898)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2899) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2900) vcipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2901)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2902) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2903) vcipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2904) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2905)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2906) vcipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2907) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2908) vxor $twk0,$twk0,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2909)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2910) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2911) vcipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2912)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2913) vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2914) vcipherlast $out0,$out0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2915)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2916) vmr $twk0,$twk1 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2917) vxor $tmp,$out0,$twk1 # last block prep for stealing
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2918) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2919) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2920) addi $out,$out,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2921) bne Lxts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2922) b Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2923)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2924) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2925) Lxts_enc6x_zero:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2926) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2927) beq Lxts_enc6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2928)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2929) add $inp,$inp,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2930) subi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2931) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2932) lvsr $inpperm,0,$taillen # $in5 is no more
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2933) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2934) vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2935) vxor $tmp,$tmp,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2936) Lxts_enc6x_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2937) vxor $in0,$in0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2938) vxor $out0,$out0,$out0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2939) vspltisb $out1,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2940) vperm $out0,$out0,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2941) vsel $out0,$in0,$tmp,$out0 # $tmp is last block, remember?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2942)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2943) subi r30,$out,17
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2944) subi $out,$out,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2945) mtctr $taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2946) Loop_xts_enc6x_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2947) lbzu r0,1(r30)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2948) stb r0,16(r30)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2949) bdnz Loop_xts_enc6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2950)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2951) li $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2952) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2953) b Loop_xts_enc1x # one more time...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2954)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2955) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2956) Lxts_enc6x_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2957) ${UCMP}i $ivp,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2958) beq Lxts_enc6x_ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2959)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2960) vxor $tweak,$twk0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2961) le?vperm $tweak,$tweak,$tweak,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2962) stvx_u $tweak,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2963)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2964) Lxts_enc6x_ret:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2965) mtlr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2966) li r10,`$FRAME+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2967) li r11,`$FRAME+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2968) stvx $seven,r10,$sp # wipe copies of round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2969) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2970) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2971) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2972) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2973) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2974) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2975) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2976) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2977) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2978) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2979) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2980) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2981) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2982) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2983) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2984)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2985) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2986) lvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2987) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2988) lvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2989) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2990) lvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2991) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2992) lvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2993) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2994) lvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2995) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2996) lvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2997) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2998) lvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2999) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3000) lvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3001) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3002) lvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3003) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3004) lvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3005) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3006) lvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3007) lvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3008) $POP r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3009) $POP r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3010) $POP r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3011) $POP r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3012) $POP r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3013) $POP r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3014) addi $sp,$sp,`$FRAME+21*16+6*$SIZE_T`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3015) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3016) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3017) .byte 0,12,0x04,1,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3018) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3019)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3020) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3021) _aesp8_xts_enc5x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3022) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3023) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3024) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3025) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3026) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3027) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3028) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3029)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3030) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3031) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3032) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3033) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3034) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3035) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3036) bdnz _aesp8_xts_enc5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3037)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3038) add $inp,$inp,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3039) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3040) vcipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3041) vcipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3042) vcipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3043) vcipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3044) vcipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3045)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3046) subi $inp,$inp,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3047) vcipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3048) vcipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3049) vcipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3050) vcipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3051) vcipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3052) vxor $twk0,$twk0,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3053)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3054) vcipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3055) lvsr $inpperm,r0,$taillen # $in5 is no more
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3056) vcipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3057) vcipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3058) vcipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3059) vcipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3060) vxor $in1,$twk1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3061)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3062) vcipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3063) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3064) vcipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3065) vcipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3066) vcipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3067) vcipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3068) vxor $in2,$twk2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3069)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3070) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3071) vcipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3072) vcipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3073) vcipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3074) vcipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3075) vcipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3076) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3077) vxor $in3,$twk3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3078)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3079) vcipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3080) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3081) vcipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3082) vcipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3083) vcipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3084) vcipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3085) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3086) vxor $in4,$twk4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3087)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3088) vcipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3089) vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3090) vcipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3091) vcipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3092) vcipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3093) vcipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3094)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3095) vcipherlast $out0,$out0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3096) vcipherlast $out1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3097) vcipherlast $out2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3098) vcipherlast $out3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3099) vcipherlast $out4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3100) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3101) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3102) .byte 0,12,0x14,0,0,0,0,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3103)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3104) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3105) _aesp8_xts_decrypt6x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3106) $STU $sp,-`($FRAME+21*16+6*$SIZE_T)`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3107) mflr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3108) li r7,`$FRAME+8*16+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3109) li r3,`$FRAME+8*16+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3110) $PUSH r11,`$FRAME+21*16+6*$SIZE_T+$LRSAVE`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3111) stvx v20,r7,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3112) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3113) stvx v21,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3114) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3115) stvx v22,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3116) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3117) stvx v23,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3118) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3119) stvx v24,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3120) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3121) stvx v25,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3122) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3123) stvx v26,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3124) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3125) stvx v27,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3126) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3127) stvx v28,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3128) addi r7,r7,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3129) stvx v29,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3130) addi r3,r3,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3131) stvx v30,r7,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3132) stvx v31,r3,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3133) li r0,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3134) stw $vrsave,`$FRAME+21*16-4`($sp) # save vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3135) li $x10,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3136) $PUSH r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3137) li $x20,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3138) $PUSH r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3139) li $x30,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3140) $PUSH r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3141) li $x40,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3142) $PUSH r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3143) li $x50,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3144) $PUSH r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3145) li $x60,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3146) $PUSH r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3147) li $x70,0x70
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3148) mtspr 256,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3149)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3150) subi $rounds,$rounds,3 # -4 in total
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3151)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3152) lvx $rndkey0,$x00,$key1 # load key schedule
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3153) lvx v30,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3154) addi $key1,$key1,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3155) lvx v31,$x00,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3156) ?vperm $rndkey0,$rndkey0,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3157) addi $key_,$sp,$FRAME+15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3158) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3159)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3160) Load_xts_dec_key:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3161) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3162) lvx v30,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3163) addi $key1,$key1,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3164) stvx v24,$x00,$key_ # off-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3165) ?vperm v25,v31,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3166) lvx v31,$x00,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3167) stvx v25,$x10,$key_ # off-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3168) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3169) bdnz Load_xts_dec_key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3170)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3171) lvx v26,$x10,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3172) ?vperm v24,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3173) lvx v27,$x20,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3174) stvx v24,$x00,$key_ # off-load round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3175) ?vperm v25,v31,v26,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3176) lvx v28,$x30,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3177) stvx v25,$x10,$key_ # off-load round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3178) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3179) ?vperm v26,v26,v27,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3180) lvx v29,$x40,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3181) ?vperm v27,v27,v28,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3182) lvx v30,$x50,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3183) ?vperm v28,v28,v29,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3184) lvx v31,$x60,$key1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3185) ?vperm v29,v29,v30,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3186) lvx $twk5,$x70,$key1 # borrow $twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3187) ?vperm v30,v30,v31,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3188) lvx v24,$x00,$key_ # pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3189) ?vperm v31,v31,$twk5,$keyperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3190) lvx v25,$x10,$key_ # pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3191)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3192) vperm $in0,$inout,$inptail,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3193) subi $inp,$inp,31 # undo "caller"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3194) vxor $twk0,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3195) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3196) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3197) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3198) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3199) vxor $out0,$in0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3200) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3201)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3202) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3203) vxor $twk1,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3204) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3205) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3206) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3207) le?vperm $in1,$in1,$in1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3208) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3209) vxor $out1,$in1,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3210) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3211)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3212) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3213) andi. $taillen,$len,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3214) vxor $twk2,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3215) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3216) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3217) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3218) le?vperm $in2,$in2,$in2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3219) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3220) vxor $out2,$in2,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3221) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3222)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3223) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3224) sub $len,$len,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3225) vxor $twk3,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3226) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3227) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3228) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3229) le?vperm $in3,$in3,$in3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3230) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3231) vxor $out3,$in3,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3232) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3233)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3234) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3235) subi $len,$len,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3236) vxor $twk4,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3237) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3238) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3239) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3240) le?vperm $in4,$in4,$in4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3241) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3242) vxor $out4,$in4,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3243) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3244)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3245) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3246) addi $inp,$inp,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3247) vxor $twk5,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3248) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3249) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3250) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3251) le?vperm $in5,$in5,$in5,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3252) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3253) vxor $out5,$in5,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3254) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3255)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3256) vxor v31,v31,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3257) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3258) b Loop_xts_dec6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3259)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3260) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3261) Loop_xts_dec6x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3262) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3263) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3264) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3265) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3266) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3267) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3268) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3269) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3270)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3271) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3272) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3273) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3274) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3275) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3276) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3277) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3278) bdnz Loop_xts_dec6x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3279)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3280) subic $len,$len,96 # $len-=96
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3281) vxor $in0,$twk0,v31 # xor with last round key
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3282) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3283) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3284) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3285) vxor $twk0,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3286) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3287) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3288) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3289) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3290) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3291) vncipher $out5,$out5,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3292)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3293) subfe. r0,r0,r0 # borrow?-1:0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3294) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3295) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3296) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3297) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3298) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3299) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3300) vxor $in1,$twk1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3301) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3302) vxor $twk1,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3303) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3304) vncipher $out5,$out5,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3305)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3306) and r0,r0,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3307) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3308) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3309) vncipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3310) vncipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3311) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3312) vncipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3313) vncipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3314) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3315) vncipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3316) vncipher $out5,$out5,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3317)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3318) add $inp,$inp,r0 # $inp is adjusted in such
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3319) # way that at exit from the
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3320) # loop inX-in5 are loaded
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3321) # with last "words"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3322) vxor $in2,$twk2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3323) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3324) vxor $twk2,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3325) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3326) vncipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3327) vncipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3328) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3329) vncipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3330) vncipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3331) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3332) vncipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3333) vncipher $out5,$out5,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3334)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3335) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3336) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3337) vncipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3338) vncipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3339) vxor $in3,$twk3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3340) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3341) vxor $twk3,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3342) vncipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3343) vncipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3344) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3345) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3346) vncipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3347) vncipher $out5,$out5,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3348) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3349) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3350)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3351) vncipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3352) vncipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3353) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3354) vncipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3355) vncipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3356) vxor $in4,$twk4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3357) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3358) vxor $twk4,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3359) vncipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3360) vncipher $out5,$out5,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3361) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3362) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3363) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3364)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3365) vncipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3366) vncipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3367) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3368) vncipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3369) vncipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3370) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3371) vncipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3372) vncipher $out5,$out5,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3373) vxor $in5,$twk5,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3374) vsrab $tmp,$tweak,$seven # next tweak value
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3375) vxor $twk5,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3376)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3377) vncipherlast $out0,$out0,$in0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3378) lvx_u $in0,$x00,$inp # load next input block
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3379) vaddubm $tweak,$tweak,$tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3380) vsldoi $tmp,$tmp,$tmp,15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3381) vncipherlast $out1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3382) lvx_u $in1,$x10,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3383) vncipherlast $out2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3384) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3385) lvx_u $in2,$x20,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3386) vand $tmp,$tmp,$eighty7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3387) vncipherlast $out3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3388) le?vperm $in1,$in1,$in1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3389) lvx_u $in3,$x30,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3390) vncipherlast $out4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3391) le?vperm $in2,$in2,$in2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3392) lvx_u $in4,$x40,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3393) vxor $tweak,$tweak,$tmp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3394) vncipherlast $out5,$out5,$in5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3395) le?vperm $in3,$in3,$in3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3396) lvx_u $in5,$x50,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3397) addi $inp,$inp,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3398) le?vperm $in4,$in4,$in4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3399) le?vperm $in5,$in5,$in5,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3400)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3401) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3402) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3403) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3404) vxor $out0,$in0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3405) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3406) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3407) vxor $out1,$in1,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3408) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3409) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3410) vxor $out2,$in2,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3411) le?vperm $out4,$out4,$out4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3412) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3413) vxor $out3,$in3,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3414) le?vperm $out5,$out5,$out5,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3415) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3416) vxor $out4,$in4,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3417) stvx_u $out5,$x50,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3418) vxor $out5,$in5,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3419) addi $out,$out,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3420)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3421) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3422) beq Loop_xts_dec6x # did $len-=96 borrow?
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3423)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3424) addic. $len,$len,0x60
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3425) beq Lxts_dec6x_zero
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3426) cmpwi $len,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3427) blt Lxts_dec6x_one
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3428) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3429) beq Lxts_dec6x_two
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3430) cmpwi $len,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3431) blt Lxts_dec6x_three
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3432) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3433) beq Lxts_dec6x_four
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3434)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3435) Lxts_dec6x_five:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3436) vxor $out0,$in1,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3437) vxor $out1,$in2,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3438) vxor $out2,$in3,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3439) vxor $out3,$in4,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3440) vxor $out4,$in5,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3441)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3442) bl _aesp8_xts_dec5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3443)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3444) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3445) vmr $twk0,$twk5 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3446) vxor $twk1,$tweak,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3447) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3448) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3449) vxor $out0,$in0,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3450) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3451) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3452) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3453) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3454) le?vperm $out4,$out4,$out4,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3455) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3456) stvx_u $out4,$x40,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3457) addi $out,$out,0x50
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3458) bne Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3459) b Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3460)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3461) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3462) Lxts_dec6x_four:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3463) vxor $out0,$in2,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3464) vxor $out1,$in3,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3465) vxor $out2,$in4,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3466) vxor $out3,$in5,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3467) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3468)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3469) bl _aesp8_xts_dec5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3470)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3471) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3472) vmr $twk0,$twk4 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3473) vmr $twk1,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3474) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3475) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3476) vxor $out0,$in0,$twk5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3477) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3478) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3479) le?vperm $out3,$out3,$out3,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3480) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3481) stvx_u $out3,$x30,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3482) addi $out,$out,0x40
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3483) bne Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3484) b Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3485)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3486) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3487) Lxts_dec6x_three:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3488) vxor $out0,$in3,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3489) vxor $out1,$in4,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3490) vxor $out2,$in5,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3491) vxor $out3,$out3,$out3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3492) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3493)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3494) bl _aesp8_xts_dec5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3495)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3496) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3497) vmr $twk0,$twk3 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3498) vmr $twk1,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3499) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3500) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3501) vxor $out0,$in0,$twk4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3502) le?vperm $out2,$out2,$out2,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3503) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3504) stvx_u $out2,$x20,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3505) addi $out,$out,0x30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3506) bne Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3507) b Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3508)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3509) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3510) Lxts_dec6x_two:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3511) vxor $out0,$in4,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3512) vxor $out1,$in5,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3513) vxor $out2,$out2,$out2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3514) vxor $out3,$out3,$out3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3515) vxor $out4,$out4,$out4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3516)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3517) bl _aesp8_xts_dec5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3518)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3519) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3520) vmr $twk0,$twk2 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3521) vmr $twk1,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3522) le?vperm $out1,$out1,$out1,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3523) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3524) vxor $out0,$in0,$twk3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3525) stvx_u $out1,$x10,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3526) addi $out,$out,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3527) bne Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3528) b Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3529)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3530) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3531) Lxts_dec6x_one:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3532) vxor $out0,$in5,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3533) nop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3534) Loop_xts_dec1x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3535) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3536) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3537) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3538)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3539) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3540) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3541) bdnz Loop_xts_dec1x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3542)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3543) subi r0,$taillen,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3544) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3545)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3546) andi. r0,r0,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3547) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3548) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3549)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3550) sub $inp,$inp,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3551) vncipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3552)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3553) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3554) vncipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3555)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3556) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3557) vncipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3558) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3559)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3560) vncipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3561) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3562) vxor $twk0,$twk0,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3563)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3564) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3565) vncipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3566)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3567) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3568) vncipherlast $out0,$out0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3569)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3570) vmr $twk0,$twk1 # unused tweak
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3571) vmr $twk1,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3572) le?vperm $out0,$out0,$out0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3573) stvx_u $out0,$x00,$out # store output
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3574) addi $out,$out,0x10
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3575) vxor $out0,$in0,$twk2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3576) bne Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3577) b Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3578)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3579) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3580) Lxts_dec6x_zero:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3581) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3582) beq Lxts_dec6x_done
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3583)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3584) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3585) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3586) vxor $out0,$in0,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3587) Lxts_dec6x_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3588) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3589) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3590) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3591)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3592) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3593) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3594) bdnz Lxts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3595)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3596) add $inp,$inp,$taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3597) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3598)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3599) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3600) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3601)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3602) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3603) vncipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3604)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3605) lvsr $inpperm,0,$taillen # $in5 is no more
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3606) vncipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3607)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3608) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3609) vncipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3610) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3611)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3612) vncipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3613) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3614) vxor $twk1,$twk1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3615)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3616) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3617) vncipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3618)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3619) vperm $in0,$in0,$in0,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3620) vncipherlast $tmp,$out0,$twk1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3621)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3622) le?vperm $out0,$tmp,$tmp,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3623) le?stvx_u $out0,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3624) be?stvx_u $tmp,0,$out
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3625)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3626) vxor $out0,$out0,$out0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3627) vspltisb $out1,-1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3628) vperm $out0,$out0,$out1,$inpperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3629) vsel $out0,$in0,$tmp,$out0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3630) vxor $out0,$out0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3631)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3632) subi r30,$out,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3633) mtctr $taillen
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3634) Loop_xts_dec6x_steal:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3635) lbzu r0,1(r30)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3636) stb r0,16(r30)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3637) bdnz Loop_xts_dec6x_steal
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3638)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3639) li $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3640) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3641) b Loop_xts_dec1x # one more time...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3642)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3643) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3644) Lxts_dec6x_done:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3645) ${UCMP}i $ivp,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3646) beq Lxts_dec6x_ret
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3647)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3648) vxor $tweak,$twk0,$rndkey0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3649) le?vperm $tweak,$tweak,$tweak,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3650) stvx_u $tweak,0,$ivp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3651)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3652) Lxts_dec6x_ret:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3653) mtlr r11
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3654) li r10,`$FRAME+15`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3655) li r11,`$FRAME+31`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3656) stvx $seven,r10,$sp # wipe copies of round keys
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3657) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3658) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3659) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3660) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3661) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3662) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3663) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3664) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3665) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3666) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3667) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3668) stvx $seven,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3669) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3670) stvx $seven,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3671) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3672)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3673) mtspr 256,$vrsave
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3674) lvx v20,r10,$sp # ABI says so
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3675) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3676) lvx v21,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3677) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3678) lvx v22,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3679) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3680) lvx v23,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3681) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3682) lvx v24,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3683) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3684) lvx v25,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3685) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3686) lvx v26,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3687) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3688) lvx v27,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3689) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3690) lvx v28,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3691) addi r10,r10,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3692) lvx v29,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3693) addi r11,r11,32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3694) lvx v30,r10,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3695) lvx v31,r11,$sp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3696) $POP r26,`$FRAME+21*16+0*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3697) $POP r27,`$FRAME+21*16+1*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3698) $POP r28,`$FRAME+21*16+2*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3699) $POP r29,`$FRAME+21*16+3*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3700) $POP r30,`$FRAME+21*16+4*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3701) $POP r31,`$FRAME+21*16+5*$SIZE_T`($sp)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3702) addi $sp,$sp,`$FRAME+21*16+6*$SIZE_T`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3703) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3704) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3705) .byte 0,12,0x04,1,0x80,6,6,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3706) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3707)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3708) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3709) _aesp8_xts_dec5x:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3710) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3711) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3712) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3713) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3714) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3715) lvx v24,$x20,$key_ # round[3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3716) addi $key_,$key_,0x20
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3717)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3718) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3719) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3720) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3721) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3722) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3723) lvx v25,$x10,$key_ # round[4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3724) bdnz _aesp8_xts_dec5x
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3725)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3726) subi r0,$taillen,1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3727) vncipher $out0,$out0,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3728) vncipher $out1,$out1,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3729) vncipher $out2,$out2,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3730) vncipher $out3,$out3,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3731) vncipher $out4,$out4,v24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3732)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3733) andi. r0,r0,16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3734) cmpwi $taillen,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3735) vncipher $out0,$out0,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3736) vncipher $out1,$out1,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3737) vncipher $out2,$out2,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3738) vncipher $out3,$out3,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3739) vncipher $out4,$out4,v25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3740) vxor $twk0,$twk0,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3741)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3742) sub $inp,$inp,r0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3743) vncipher $out0,$out0,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3744) vncipher $out1,$out1,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3745) vncipher $out2,$out2,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3746) vncipher $out3,$out3,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3747) vncipher $out4,$out4,v26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3748) vxor $in1,$twk1,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3749)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3750) vncipher $out0,$out0,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3751) lvx_u $in0,0,$inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3752) vncipher $out1,$out1,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3753) vncipher $out2,$out2,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3754) vncipher $out3,$out3,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3755) vncipher $out4,$out4,v27
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3756) vxor $in2,$twk2,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3757)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3758) addi $key_,$sp,$FRAME+15 # rewind $key_
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3759) vncipher $out0,$out0,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3760) vncipher $out1,$out1,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3761) vncipher $out2,$out2,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3762) vncipher $out3,$out3,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3763) vncipher $out4,$out4,v28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3764) lvx v24,$x00,$key_ # re-pre-load round[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3765) vxor $in3,$twk3,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3766)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3767) vncipher $out0,$out0,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3768) le?vperm $in0,$in0,$in0,$leperm
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3769) vncipher $out1,$out1,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3770) vncipher $out2,$out2,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3771) vncipher $out3,$out3,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3772) vncipher $out4,$out4,v29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3773) lvx v25,$x10,$key_ # re-pre-load round[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3774) vxor $in4,$twk4,v31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3775)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3776) vncipher $out0,$out0,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3777) vncipher $out1,$out1,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3778) vncipher $out2,$out2,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3779) vncipher $out3,$out3,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3780) vncipher $out4,$out4,v30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3781)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3782) vncipherlast $out0,$out0,$twk0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3783) vncipherlast $out1,$out1,$in1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3784) vncipherlast $out2,$out2,$in2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3785) vncipherlast $out3,$out3,$in3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3786) vncipherlast $out4,$out4,$in4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3787) mtctr $rounds
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3788) blr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3789) .long 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3790) .byte 0,12,0x14,0,0,0,0,0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3791) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3792) }} }}}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3793)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3794) my $consts=1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3795) foreach(split("\n",$code)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3796) s/\`([^\`]*)\`/eval($1)/geo;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3797)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3798) # constants table endian-specific conversion
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3799) if ($consts && m/\.(long|byte)\s+(.+)\s+(\?[a-z]*)$/o) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3800) my $conv=$3;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3801) my @bytes=();
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3802)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3803) # convert to endian-agnostic format
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3804) if ($1 eq "long") {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3805) foreach (split(/,\s*/,$2)) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3806) my $l = /^0/?oct:int;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3807) push @bytes,($l>>24)&0xff,($l>>16)&0xff,($l>>8)&0xff,$l&0xff;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3808) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3809) } else {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3810) @bytes = map(/^0/?oct:int,split(/,\s*/,$2));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3811) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3812)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3813) # little-endian conversion
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3814) if ($flavour =~ /le$/o) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3815) SWITCH: for($conv) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3816) /\?inv/ && do { @bytes=map($_^0xf,@bytes); last; };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3817) /\?rev/ && do { @bytes=reverse(@bytes); last; };
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3818) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3819) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3820)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3821) #emit
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3822) print ".byte\t",join(',',map (sprintf("0x%02x",$_),@bytes)),"\n";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3823) next;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3824) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3825) $consts=0 if (m/Lconsts:/o); # end of table
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3826)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3827) # instructions prefixed with '?' are endian-specific and need
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3828) # to be adjusted accordingly...
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3829) if ($flavour =~ /le$/o) { # little-endian
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3830) s/le\?//o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3831) s/be\?/#be#/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3832) s/\?lvsr/lvsl/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3833) s/\?lvsl/lvsr/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3834) s/\?(vperm\s+v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+,\s*)(v[0-9]+)/$1$3$2$4/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3835) s/\?(vsldoi\s+v[0-9]+,\s*)(v[0-9]+,)\s*(v[0-9]+,\s*)([0-9]+)/$1$3$2 16-$4/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3836) s/\?(vspltw\s+v[0-9]+,\s*)(v[0-9]+,)\s*([0-9])/$1$2 3-$3/o;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3837) } else { # big-endian
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3838) s/le\?/#le#/o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3839) s/be\?//o or
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3840) s/\?([a-z]+)/$1/o;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3841) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3842)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3843) print $_,"\n";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3844) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3845)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3846) close STDOUT;
Orange Pi5 kernel
Deprecated Linux kernel 5.10.110 for OrangePi 5/5B/5+ boards
3 Commits
0 Branches
0 Tags