1#! /usr/bin/env perl 2# Copyright 2014-2020 The OpenSSL Project Authors. All Rights Reserved. 3# 4# Licensed under the OpenSSL license (the "License"). You may not use 5# this file except in compliance with the License. You can obtain a copy 6# in the file LICENSE in the source distribution or at 7# https://www.openssl.org/source/license.html 8 9 10# ==================================================================== 11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL 12# project. The module is, however, dual licensed under OpenSSL and 13# CRYPTOGAMS licenses depending on where you obtain it. For further 14# details see http://www.openssl.org/~appro/cryptogams/. 15# ==================================================================== 16 17# SHA256/512 for PowerISA v2.07. 18# 19# Accurate performance measurements are problematic, because it's 20# always virtualized setup with possibly throttled processor. 21# Relative comparison is therefore more informative. This module is 22# ~60% faster than integer-only sha512-ppc.pl. To anchor to something 23# else, SHA256 is 24% slower than sha1-ppc.pl and 2.5x slower than 24# hardware-assisted aes-128-cbc encrypt. SHA512 is 20% faster than 25# sha1-ppc.pl and 1.6x slower than aes-128-cbc. Another interesting 26# result is degree of computational resources' utilization. POWER8 is 27# "massively multi-threaded chip" and difference between single- and 28# maximum multi-process benchmark results tells that utilization is 29# whooping 94%. For sha512-ppc.pl we get [not unimpressive] 84% and 30# for sha1-ppc.pl - 73%. 100% means that multi-process result equals 31# to single-process one, given that all threads end up on the same 32# physical core. 33# 34###################################################################### 35# Believed-to-be-accurate results in cycles per processed byte [on 36# little-endian system]. Numbers in square brackets are for 64-bit 37# build of sha512-ppc.pl, presented for reference. 38# 39# POWER8 POWER9 40# SHA256 9.7 [15.8] 11.2 [12.5] 41# SHA512 6.1 [10.3] 7.0 [7.9] 42 43$flavour=shift; 44$output =shift; 45 46if ($flavour =~ /64/) { 47 $SIZE_T=8; 48 $LRSAVE=2*$SIZE_T; 49 $STU="stdu"; 50 $POP="ld"; 51 $PUSH="std"; 52} elsif ($flavour =~ /32/) { 53 $SIZE_T=4; 54 $LRSAVE=$SIZE_T; 55 $STU="stwu"; 56 $POP="lwz"; 57 $PUSH="stw"; 58} else { die "nonsense $flavour"; } 59 60$LENDIAN=($flavour=~/le/); 61 62$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; 63( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or 64( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or 65die "can't locate ppc-xlate.pl"; 66 67open STDOUT,"| $^X $xlate $flavour $output" || die "can't call $xlate: $!"; 68 69if ($output =~ /512/) { 70 $bits=512; 71 $SZ=8; 72 $sz="d"; 73 $rounds=80; 74} else { 75 $bits=256; 76 $SZ=4; 77 $sz="w"; 78 $rounds=64; 79} 80 81$func="sha${bits}_block_p8"; 82$LOCALS=8*$SIZE_T+8*16; 83$FRAME=$LOCALS+9*16+6*$SIZE_T; 84 85$sp ="r1"; 86$toc="r2"; 87$ctx="r3"; 88$inp="r4"; 89$num="r5"; 90$Tbl="r6"; 91$idx="r7"; 92$lrsave="r8"; 93$offload="r11"; 94$vrsave="r12"; 95@I = ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70) = (0,map("r$_",(10,26..31))); 96 97@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("v$_",(0..7)); 98@X=map("v$_",(8..19,24..27)); 99($Ki,$Func,$Sigma,$lemask)=map("v$_",(28..31)); 100 101sub ROUND { 102my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_; 103my $j=($i+1)%16; 104my $k=($i+2)%8; 105 106$code.=<<___ if ($i<15 && ($i%(16/$SZ))==(16/$SZ-1)); 107 lvx_u @X[$i+1],0,$inp ; load X[i] in advance 108 addi $inp,$inp,16 109___ 110$code.=<<___ if ($i<16 && ($i%(16/$SZ))); 111 vsldoi @X[$i],@X[$i-1],@X[$i-1],$SZ 112___ 113$code.=<<___ if ($LENDIAN && $i<16 && ($i%(16/$SZ))==0); 114 vperm @X[$i],@X[$i],@X[$i],$lemask 115___ 116$code.=<<___ if ($i>=15); 117 vshasigma${sz} $Sigma,@X[($j+1)%16],0,0 118 vaddu${sz}m @X[$j],@X[$j],$Sigma 119 vshasigma${sz} $Sigma,@X[($j+14)%16],0,15 120 vaddu${sz}m @X[$j],@X[$j],$Sigma 121 vaddu${sz}m @X[$j],@X[$j],@X[($j+9)%16] 122___ 123$code.=<<___; 124 vaddu${sz}m $h,$h,@X[$i%16] ; h+=X[i] 125 vsel $Func,$g,$f,$e ; Ch(e,f,g) 126 vaddu${sz}m $g,$g,$Ki ; future h+=K[i] 127 vaddu${sz}m $h,$h,$Func ; h+=Ch(e,f,g) 128 vshasigma${sz} $Sigma,$e,1,15 ; Sigma1(e) 129 vaddu${sz}m $h,$h,$Sigma ; h+=Sigma1(e) 130 vxor $Func,$a,$b 131 vsel $Func,$b,$c,$Func ; Maj(a,b,c) 132 vaddu${sz}m $d,$d,$h ; d+=h 133 vshasigma${sz} $Sigma,$a,1,0 ; Sigma0(a) 134 vaddu${sz}m $Sigma,$Sigma,$Func ; Sigma0(a)+Maj(a,b,c) 135 vaddu${sz}m $h,$h,$Sigma ; h+=Sigma0(a)+Maj(a,b,c) 136 lvx $Ki,@I[$k],$idx ; load next K[i] 137___ 138$code.=<<___ if ($k == 7); 139 addi $idx,$idx,0x80 140___ 141} 142 143$code=<<___; 144.machine "any" 145.text 146 147.globl $func 148.align 6 149$func: 150 $STU $sp,-$FRAME($sp) 151 mflr $lrsave 152 li r10,`$LOCALS+15` 153 li r11,`$LOCALS+31` 154 stvx v24,r10,$sp # ABI says so 155 addi r10,r10,32 156 mfspr $vrsave,256 157 stvx v25,r11,$sp 158 addi r11,r11,32 159 stvx v26,r10,$sp 160 addi r10,r10,32 161 stvx v27,r11,$sp 162 addi r11,r11,32 163 stvx v28,r10,$sp 164 addi r10,r10,32 165 stvx v29,r11,$sp 166 addi r11,r11,32 167 stvx v30,r10,$sp 168 stvx v31,r11,$sp 169 li r11,-4096+255 # 0xfffff0ff 170 stw $vrsave,`$FRAME-6*$SIZE_T-4`($sp) # save vrsave 171 li $x10,0x10 172 $PUSH r26,`$FRAME-6*$SIZE_T`($sp) 173 li $x20,0x20 174 $PUSH r27,`$FRAME-5*$SIZE_T`($sp) 175 li $x30,0x30 176 $PUSH r28,`$FRAME-4*$SIZE_T`($sp) 177 li $x40,0x40 178 $PUSH r29,`$FRAME-3*$SIZE_T`($sp) 179 li $x50,0x50 180 $PUSH r30,`$FRAME-2*$SIZE_T`($sp) 181 li $x60,0x60 182 $PUSH r31,`$FRAME-1*$SIZE_T`($sp) 183 li $x70,0x70 184 $PUSH $lrsave,`$FRAME+$LRSAVE`($sp) 185 mtspr 256,r11 186 187 bl LPICmeup 188 addi $offload,$sp,`8*$SIZE_T+15` 189___ 190$code.=<<___ if ($LENDIAN); 191 li $idx,8 192 lvsl $lemask,0,$idx 193 vspltisb $Ki,0x0f 194 vxor $lemask,$lemask,$Ki 195___ 196$code.=<<___ if ($SZ==4); 197 lvx_4w $A,$x00,$ctx 198 lvx_4w $E,$x10,$ctx 199 vsldoi $B,$A,$A,4 # unpack 200 vsldoi $C,$A,$A,8 201 vsldoi $D,$A,$A,12 202 vsldoi $F,$E,$E,4 203 vsldoi $G,$E,$E,8 204 vsldoi $H,$E,$E,12 205___ 206$code.=<<___ if ($SZ==8); 207 lvx_u $A,$x00,$ctx 208 lvx_u $C,$x10,$ctx 209 lvx_u $E,$x20,$ctx 210 vsldoi $B,$A,$A,8 # unpack 211 lvx_u $G,$x30,$ctx 212 vsldoi $D,$C,$C,8 213 vsldoi $F,$E,$E,8 214 vsldoi $H,$G,$G,8 215___ 216$code.=<<___; 217 li r0,`($rounds-16)/16` # inner loop counter 218 b Loop 219.align 5 220Loop: 221 lvx $Ki,$x00,$Tbl 222 lvx_u @X[0],0,$inp 223 addi $inp,$inp,16 224 mr $idx,$Tbl # copy $Tbl 225 stvx $A,$x00,$offload # offload $A-$H 226 stvx $B,$x10,$offload 227 stvx $C,$x20,$offload 228 stvx $D,$x30,$offload 229 stvx $E,$x40,$offload 230 stvx $F,$x50,$offload 231 stvx $G,$x60,$offload 232 stvx $H,$x70,$offload 233 vaddu${sz}m $H,$H,$Ki # h+K[i] 234 lvx $Ki,$x10,$Tbl 235___ 236for ($i=0;$i<16;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); } 237$code.=<<___; 238 mtctr r0 239 b L16_xx 240.align 5 241L16_xx: 242___ 243for (;$i<32;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); } 244$code.=<<___; 245 bdnz L16_xx 246 247 lvx @X[2],$x00,$offload 248 subic. $num,$num,1 249 lvx @X[3],$x10,$offload 250 vaddu${sz}m $A,$A,@X[2] 251 lvx @X[4],$x20,$offload 252 vaddu${sz}m $B,$B,@X[3] 253 lvx @X[5],$x30,$offload 254 vaddu${sz}m $C,$C,@X[4] 255 lvx @X[6],$x40,$offload 256 vaddu${sz}m $D,$D,@X[5] 257 lvx @X[7],$x50,$offload 258 vaddu${sz}m $E,$E,@X[6] 259 lvx @X[8],$x60,$offload 260 vaddu${sz}m $F,$F,@X[7] 261 lvx @X[9],$x70,$offload 262 vaddu${sz}m $G,$G,@X[8] 263 vaddu${sz}m $H,$H,@X[9] 264 bne Loop 265___ 266$code.=<<___ if ($SZ==4); 267 lvx @X[0],$x20,$idx 268 vperm $A,$A,$B,$Ki # pack the answer 269 lvx @X[1],$x30,$idx 270 vperm $E,$E,$F,$Ki 271 vperm $A,$A,$C,@X[0] 272 vperm $E,$E,$G,@X[0] 273 vperm $A,$A,$D,@X[1] 274 vperm $E,$E,$H,@X[1] 275 stvx_4w $A,$x00,$ctx 276 stvx_4w $E,$x10,$ctx 277___ 278$code.=<<___ if ($SZ==8); 279 vperm $A,$A,$B,$Ki # pack the answer 280 vperm $C,$C,$D,$Ki 281 vperm $E,$E,$F,$Ki 282 vperm $G,$G,$H,$Ki 283 stvx_u $A,$x00,$ctx 284 stvx_u $C,$x10,$ctx 285 stvx_u $E,$x20,$ctx 286 stvx_u $G,$x30,$ctx 287___ 288$code.=<<___; 289 addi $offload,$sp,`$LOCALS+15` 290 mtlr $lrsave 291 mtspr 256,$vrsave 292 lvx v24,$x00,$offload # ABI says so 293 lvx v25,$x10,$offload 294 lvx v26,$x20,$offload 295 lvx v27,$x30,$offload 296 lvx v28,$x40,$offload 297 lvx v29,$x50,$offload 298 lvx v30,$x60,$offload 299 lvx v31,$x70,$offload 300 $POP r26,`$FRAME-6*$SIZE_T`($sp) 301 $POP r27,`$FRAME-5*$SIZE_T`($sp) 302 $POP r28,`$FRAME-4*$SIZE_T`($sp) 303 $POP r29,`$FRAME-3*$SIZE_T`($sp) 304 $POP r30,`$FRAME-2*$SIZE_T`($sp) 305 $POP r31,`$FRAME-1*$SIZE_T`($sp) 306 addi $sp,$sp,$FRAME 307 blr 308 .long 0 309 .byte 0,12,4,1,0x80,6,3,0 310 .long 0 311.size $func,.-$func 312___ 313 314# Ugly hack here, because PPC assembler syntax seem to vary too 315# much from platforms to platform... 316$code.=<<___; 317.align 6 318LPICmeup: 319 mflr r0 320 bcl 20,31,\$+4 321 mflr $Tbl ; vvvvvv "distance" between . and 1st data entry 322 addi $Tbl,$Tbl,`64-8` 323 mtlr r0 324 blr 325 .long 0 326 .byte 0,12,0x14,0,0,0,0,0 327 .space `64-9*4` 328___ 329 330if ($SZ==8) { 331 local *table = sub { 332 foreach(@_) { $code.=".quad $_,$_\n"; } 333 }; 334 table( 335 "0x428a2f98d728ae22","0x7137449123ef65cd", 336 "0xb5c0fbcfec4d3b2f","0xe9b5dba58189dbbc", 337 "0x3956c25bf348b538","0x59f111f1b605d019", 338 "0x923f82a4af194f9b","0xab1c5ed5da6d8118", 339 "0xd807aa98a3030242","0x12835b0145706fbe", 340 "0x243185be4ee4b28c","0x550c7dc3d5ffb4e2", 341 "0x72be5d74f27b896f","0x80deb1fe3b1696b1", 342 "0x9bdc06a725c71235","0xc19bf174cf692694", 343 "0xe49b69c19ef14ad2","0xefbe4786384f25e3", 344 "0x0fc19dc68b8cd5b5","0x240ca1cc77ac9c65", 345 "0x2de92c6f592b0275","0x4a7484aa6ea6e483", 346 "0x5cb0a9dcbd41fbd4","0x76f988da831153b5", 347 "0x983e5152ee66dfab","0xa831c66d2db43210", 348 "0xb00327c898fb213f","0xbf597fc7beef0ee4", 349 "0xc6e00bf33da88fc2","0xd5a79147930aa725", 350 "0x06ca6351e003826f","0x142929670a0e6e70", 351 "0x27b70a8546d22ffc","0x2e1b21385c26c926", 352 "0x4d2c6dfc5ac42aed","0x53380d139d95b3df", 353 "0x650a73548baf63de","0x766a0abb3c77b2a8", 354 "0x81c2c92e47edaee6","0x92722c851482353b", 355 "0xa2bfe8a14cf10364","0xa81a664bbc423001", 356 "0xc24b8b70d0f89791","0xc76c51a30654be30", 357 "0xd192e819d6ef5218","0xd69906245565a910", 358 "0xf40e35855771202a","0x106aa07032bbd1b8", 359 "0x19a4c116b8d2d0c8","0x1e376c085141ab53", 360 "0x2748774cdf8eeb99","0x34b0bcb5e19b48a8", 361 "0x391c0cb3c5c95a63","0x4ed8aa4ae3418acb", 362 "0x5b9cca4f7763e373","0x682e6ff3d6b2b8a3", 363 "0x748f82ee5defb2fc","0x78a5636f43172f60", 364 "0x84c87814a1f0ab72","0x8cc702081a6439ec", 365 "0x90befffa23631e28","0xa4506cebde82bde9", 366 "0xbef9a3f7b2c67915","0xc67178f2e372532b", 367 "0xca273eceea26619c","0xd186b8c721c0c207", 368 "0xeada7dd6cde0eb1e","0xf57d4f7fee6ed178", 369 "0x06f067aa72176fba","0x0a637dc5a2c898a6", 370 "0x113f9804bef90dae","0x1b710b35131c471b", 371 "0x28db77f523047d84","0x32caab7b40c72493", 372 "0x3c9ebe0a15c9bebc","0x431d67c49c100d4c", 373 "0x4cc5d4becb3e42b6","0x597f299cfc657e2a", 374 "0x5fcb6fab3ad6faec","0x6c44198c4a475817","0"); 375$code.=<<___ if (!$LENDIAN); 376.quad 0x0001020304050607,0x1011121314151617 377___ 378$code.=<<___ if ($LENDIAN); # quad-swapped 379.quad 0x1011121314151617,0x0001020304050607 380___ 381} else { 382 local *table = sub { 383 foreach(@_) { $code.=".long $_,$_,$_,$_\n"; } 384 }; 385 table( 386 "0x428a2f98","0x71374491","0xb5c0fbcf","0xe9b5dba5", 387 "0x3956c25b","0x59f111f1","0x923f82a4","0xab1c5ed5", 388 "0xd807aa98","0x12835b01","0x243185be","0x550c7dc3", 389 "0x72be5d74","0x80deb1fe","0x9bdc06a7","0xc19bf174", 390 "0xe49b69c1","0xefbe4786","0x0fc19dc6","0x240ca1cc", 391 "0x2de92c6f","0x4a7484aa","0x5cb0a9dc","0x76f988da", 392 "0x983e5152","0xa831c66d","0xb00327c8","0xbf597fc7", 393 "0xc6e00bf3","0xd5a79147","0x06ca6351","0x14292967", 394 "0x27b70a85","0x2e1b2138","0x4d2c6dfc","0x53380d13", 395 "0x650a7354","0x766a0abb","0x81c2c92e","0x92722c85", 396 "0xa2bfe8a1","0xa81a664b","0xc24b8b70","0xc76c51a3", 397 "0xd192e819","0xd6990624","0xf40e3585","0x106aa070", 398 "0x19a4c116","0x1e376c08","0x2748774c","0x34b0bcb5", 399 "0x391c0cb3","0x4ed8aa4a","0x5b9cca4f","0x682e6ff3", 400 "0x748f82ee","0x78a5636f","0x84c87814","0x8cc70208", 401 "0x90befffa","0xa4506ceb","0xbef9a3f7","0xc67178f2","0"); 402$code.=<<___ if (!$LENDIAN); 403.long 0x00010203,0x10111213,0x10111213,0x10111213 404.long 0x00010203,0x04050607,0x10111213,0x10111213 405.long 0x00010203,0x04050607,0x08090a0b,0x10111213 406___ 407$code.=<<___ if ($LENDIAN); # word-swapped 408.long 0x10111213,0x10111213,0x10111213,0x00010203 409.long 0x10111213,0x10111213,0x04050607,0x00010203 410.long 0x10111213,0x08090a0b,0x04050607,0x00010203 411___ 412} 413$code.=<<___; 414.asciz "SHA${bits} for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>" 415.align 2 416___ 417 418$code =~ s/\`([^\`]*)\`/eval $1/gem; 419print $code; 420close STDOUT or die "error closing STDOUT: $!"; 421