1#!/usr/bin/env perl 2# SPDX-License-Identifier: GPL-2.0 3 4# This code is taken from the OpenSSL project but the author (Andy Polyakov) 5# has relicensed it under the GPLv2. Therefore this program is free software; 6# you can redistribute it and/or modify it under the terms of the GNU General 7# Public License version 2 as published by the Free Software Foundation. 8# 9# The original headers, including the original license headers, are 10# included below for completeness. 11 12# ==================================================================== 13# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL 14# project. The module is, however, dual licensed under OpenSSL and 15# CRYPTOGAMS licenses depending on where you obtain it. For further 16# details see https://www.openssl.org/~appro/cryptogams/. 17# ==================================================================== 18# 19# GHASH for PowerISA v2.07. 20# 21# July 2014 22# 23# Accurate performance measurements are problematic, because it's 24# always virtualized setup with possibly throttled processor. 25# Relative comparison is therefore more informative. This initial 26# version is ~2.1x slower than hardware-assisted AES-128-CTR, ~12x 27# faster than "4-bit" integer-only compiler-generated 64-bit code. 28# "Initial version" means that there is room for futher improvement. 29 30$flavour=shift; 31$output =shift; 32 33if ($flavour =~ /64/) { 34 $SIZE_T=8; 35 $LRSAVE=2*$SIZE_T; 36 $STU="stdu"; 37 $POP="ld"; 38 $PUSH="std"; 39} elsif ($flavour =~ /32/) { 40 $SIZE_T=4; 41 $LRSAVE=$SIZE_T; 42 $STU="stwu"; 43 $POP="lwz"; 44 $PUSH="stw"; 45} else { die "nonsense $flavour"; } 46 47$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 48( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or 49( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or 50die "can't locate ppc-xlate.pl"; 51 52open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!"; 53 54my ($Xip,$Htbl,$inp,$len)=map("r$_",(3..6)); # argument block 55 56my ($Xl,$Xm,$Xh,$IN)=map("v$_",(0..3)); 57my ($zero,$t0,$t1,$t2,$xC2,$H,$Hh,$Hl,$lemask)=map("v$_",(4..12)); 58my $vrsave="r12"; 59 60$code=<<___; 61.machine "any" 62 63.text 64 65.globl .gcm_init_p8 66 lis r0,0xfff0 67 li r8,0x10 68 mfspr $vrsave,256 69 li r9,0x20 70 mtspr 256,r0 71 li r10,0x30 72 lvx_u $H,0,r4 # load H 73 le?xor r7,r7,r7 74 le?addi r7,r7,0x8 # need a vperm start with 08 75 le?lvsr 5,0,r7 76 le?vspltisb 6,0x0f 77 le?vxor 5,5,6 # set a b-endian mask 78 le?vperm $H,$H,$H,5 79 80 vspltisb $xC2,-16 # 0xf0 81 vspltisb $t0,1 # one 82 vaddubm $xC2,$xC2,$xC2 # 0xe0 83 vxor $zero,$zero,$zero 84 vor $xC2,$xC2,$t0 # 0xe1 85 vsldoi $xC2,$xC2,$zero,15 # 0xe1... 86 vsldoi $t1,$zero,$t0,1 # ...1 87 vaddubm $xC2,$xC2,$xC2 # 0xc2... 88 vspltisb $t2,7 89 vor $xC2,$xC2,$t1 # 0xc2....01 90 vspltb $t1,$H,0 # most significant byte 91 vsl $H,$H,$t0 # H<<=1 92 vsrab $t1,$t1,$t2 # broadcast carry bit 93 vand $t1,$t1,$xC2 94 vxor $H,$H,$t1 # twisted H 95 96 vsldoi $H,$H,$H,8 # twist even more ... 97 vsldoi $xC2,$zero,$xC2,8 # 0xc2.0 98 vsldoi $Hl,$zero,$H,8 # ... and split 99 vsldoi $Hh,$H,$zero,8 100 101 stvx_u $xC2,0,r3 # save pre-computed table 102 stvx_u $Hl,r8,r3 103 stvx_u $H, r9,r3 104 stvx_u $Hh,r10,r3 105 106 mtspr 256,$vrsave 107 blr 108 .long 0 109 .byte 0,12,0x14,0,0,0,2,0 110 .long 0 111.size .gcm_init_p8,.-.gcm_init_p8 112 113.globl .gcm_gmult_p8 114 lis r0,0xfff8 115 li r8,0x10 116 mfspr $vrsave,256 117 li r9,0x20 118 mtspr 256,r0 119 li r10,0x30 120 lvx_u $IN,0,$Xip # load Xi 121 122 lvx_u $Hl,r8,$Htbl # load pre-computed table 123 le?lvsl $lemask,r0,r0 124 lvx_u $H, r9,$Htbl 125 le?vspltisb $t0,0x07 126 lvx_u $Hh,r10,$Htbl 127 le?vxor $lemask,$lemask,$t0 128 lvx_u $xC2,0,$Htbl 129 le?vperm $IN,$IN,$IN,$lemask 130 vxor $zero,$zero,$zero 131 132 vpmsumd $Xl,$IN,$Hl # H.lo·Xi.lo 133 vpmsumd $Xm,$IN,$H # H.hi·Xi.lo+H.lo·Xi.hi 134 vpmsumd $Xh,$IN,$Hh # H.hi·Xi.hi 135 136 vpmsumd $t2,$Xl,$xC2 # 1st phase 137 138 vsldoi $t0,$Xm,$zero,8 139 vsldoi $t1,$zero,$Xm,8 140 vxor $Xl,$Xl,$t0 141 vxor $Xh,$Xh,$t1 142 143 vsldoi $Xl,$Xl,$Xl,8 144 vxor $Xl,$Xl,$t2 145 146 vsldoi $t1,$Xl,$Xl,8 # 2nd phase 147 vpmsumd $Xl,$Xl,$xC2 148 vxor $t1,$t1,$Xh 149 vxor $Xl,$Xl,$t1 150 151 le?vperm $Xl,$Xl,$Xl,$lemask 152 stvx_u $Xl,0,$Xip # write out Xi 153 154 mtspr 256,$vrsave 155 blr 156 .long 0 157 .byte 0,12,0x14,0,0,0,2,0 158 .long 0 159.size .gcm_gmult_p8,.-.gcm_gmult_p8 160 161.globl .gcm_ghash_p8 162 lis r0,0xfff8 163 li r8,0x10 164 mfspr $vrsave,256 165 li r9,0x20 166 mtspr 256,r0 167 li r10,0x30 168 lvx_u $Xl,0,$Xip # load Xi 169 170 lvx_u $Hl,r8,$Htbl # load pre-computed table 171 le?lvsl $lemask,r0,r0 172 lvx_u $H, r9,$Htbl 173 le?vspltisb $t0,0x07 174 lvx_u $Hh,r10,$Htbl 175 le?vxor $lemask,$lemask,$t0 176 lvx_u $xC2,0,$Htbl 177 le?vperm $Xl,$Xl,$Xl,$lemask 178 vxor $zero,$zero,$zero 179 180 lvx_u $IN,0,$inp 181 addi $inp,$inp,16 182 subi $len,$len,16 183 le?vperm $IN,$IN,$IN,$lemask 184 vxor $IN,$IN,$Xl 185 b Loop 186 187.align 5 188Loop: 189 subic $len,$len,16 190 vpmsumd $Xl,$IN,$Hl # H.lo·Xi.lo 191 subfe. r0,r0,r0 # borrow?-1:0 192 vpmsumd $Xm,$IN,$H # H.hi·Xi.lo+H.lo·Xi.hi 193 and r0,r0,$len 194 vpmsumd $Xh,$IN,$Hh # H.hi·Xi.hi 195 add $inp,$inp,r0 196 197 vpmsumd $t2,$Xl,$xC2 # 1st phase 198 199 vsldoi $t0,$Xm,$zero,8 200 vsldoi $t1,$zero,$Xm,8 201 vxor $Xl,$Xl,$t0 202 vxor $Xh,$Xh,$t1 203 204 vsldoi $Xl,$Xl,$Xl,8 205 vxor $Xl,$Xl,$t2 206 lvx_u $IN,0,$inp 207 addi $inp,$inp,16 208 209 vsldoi $t1,$Xl,$Xl,8 # 2nd phase 210 vpmsumd $Xl,$Xl,$xC2 211 le?vperm $IN,$IN,$IN,$lemask 212 vxor $t1,$t1,$Xh 213 vxor $IN,$IN,$t1 214 vxor $IN,$IN,$Xl 215 beq Loop # did $len-=16 borrow? 216 217 vxor $Xl,$Xl,$t1 218 le?vperm $Xl,$Xl,$Xl,$lemask 219 stvx_u $Xl,0,$Xip # write out Xi 220 221 mtspr 256,$vrsave 222 blr 223 .long 0 224 .byte 0,12,0x14,0,0,0,4,0 225 .long 0 226.size .gcm_ghash_p8,.-.gcm_ghash_p8 227 228.asciz "GHASH for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>" 229.align 2 230___ 231 232foreach (split("\n",$code)) { 233 if ($flavour =~ /le$/o) { # little-endian 234 s/le\?//o or 235 s/be\?/#be#/o; 236 } else { 237 s/le\?/#le#/o or 238 s/be\?//o; 239 } 240 print $_,"\n"; 241} 242 243close STDOUT; # enforce flush 244