2
0

sha512p8-ppc.pl 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423
  1. #! /usr/bin/env perl
  2. # Copyright 2014-2018 The OpenSSL Project Authors. All Rights Reserved.
  3. #
  4. # Licensed under the Apache License 2.0 (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. # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
  10. # project. The module is, however, dual licensed under OpenSSL and
  11. # CRYPTOGAMS licenses depending on where you obtain it. For further
  12. # details see http://www.openssl.org/~appro/cryptogams/.
  13. # ====================================================================
  14. # SHA256/512 for PowerISA v2.07.
  15. #
  16. # Accurate performance measurements are problematic, because it's
  17. # always virtualized setup with possibly throttled processor.
  18. # Relative comparison is therefore more informative. This module is
  19. # ~60% faster than integer-only sha512-ppc.pl. To anchor to something
  20. # else, SHA256 is 24% slower than sha1-ppc.pl and 2.5x slower than
  21. # hardware-assisted aes-128-cbc encrypt. SHA512 is 20% faster than
  22. # sha1-ppc.pl and 1.6x slower than aes-128-cbc. Another interesting
  23. # result is degree of computational resources' utilization. POWER8 is
  24. # "massively multi-threaded chip" and difference between single- and
  25. # maximum multi-process benchmark results tells that utilization is
  26. # whooping 94%. For sha512-ppc.pl we get [not unimpressive] 84% and
  27. # for sha1-ppc.pl - 73%. 100% means that multi-process result equals
  28. # to single-process one, given that all threads end up on the same
  29. # physical core.
  30. #
  31. ######################################################################
  32. # Believed-to-be-accurate results in cycles per processed byte [on
  33. # little-endian system]. Numbers in square brackets are for 64-bit
  34. # build of sha512-ppc.pl, presented for reference.
  35. #
  36. # POWER8 POWER9
  37. # SHA256 9.7 [15.8] 11.2 [12.5]
  38. # SHA512 6.1 [10.3] 7.0 [7.9]
  39. # $output is the last argument if it looks like a file (it has an extension)
  40. # $flavour is the first argument if it doesn't look like a file
  41. $output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
  42. $flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
  43. if ($flavour =~ /64/) {
  44. $SIZE_T=8;
  45. $LRSAVE=2*$SIZE_T;
  46. $STU="stdu";
  47. $POP="ld";
  48. $PUSH="std";
  49. } elsif ($flavour =~ /32/) {
  50. $SIZE_T=4;
  51. $LRSAVE=$SIZE_T;
  52. $STU="stwu";
  53. $POP="lwz";
  54. $PUSH="stw";
  55. } else { die "nonsense $flavour"; }
  56. $LENDIAN=($flavour=~/le/);
  57. $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
  58. ( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
  59. ( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
  60. die "can't locate ppc-xlate.pl";
  61. open STDOUT,"| $^X $xlate $flavour \"$output\""
  62. or die "can't call $xlate: $!";
  63. if ($output =~ /512/) {
  64. $bits=512;
  65. $SZ=8;
  66. $sz="d";
  67. $rounds=80;
  68. } else {
  69. $bits=256;
  70. $SZ=4;
  71. $sz="w";
  72. $rounds=64;
  73. }
  74. $func="sha${bits}_block_p8";
  75. $LOCALS=8*$SIZE_T+8*16;
  76. $FRAME=$LOCALS+9*16+6*$SIZE_T;
  77. $sp ="r1";
  78. $toc="r2";
  79. $ctx="r3";
  80. $inp="r4";
  81. $num="r5";
  82. $Tbl="r6";
  83. $idx="r7";
  84. $lrsave="r8";
  85. $offload="r11";
  86. $vrsave="r12";
  87. @I = ($x00,$x10,$x20,$x30,$x40,$x50,$x60,$x70) = (0,map("r$_",(10,26..31)));
  88. @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("v$_",(0..7));
  89. @X=map("v$_",(8..19,24..27));
  90. ($Ki,$Func,$Sigma,$lemask)=map("v$_",(28..31));
  91. sub ROUND {
  92. my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
  93. my $j=($i+1)%16;
  94. my $k=($i+2)%8;
  95. $code.=<<___ if ($i<15 && ($i%(16/$SZ))==(16/$SZ-1));
  96. lvx_u @X[$i+1],0,$inp ; load X[i] in advance
  97. addi $inp,$inp,16
  98. ___
  99. $code.=<<___ if ($i<16 && ($i%(16/$SZ)));
  100. vsldoi @X[$i],@X[$i-1],@X[$i-1],$SZ
  101. ___
  102. $code.=<<___ if ($LENDIAN && $i<16 && ($i%(16/$SZ))==0);
  103. vperm @X[$i],@X[$i],@X[$i],$lemask
  104. ___
  105. $code.=<<___ if ($i>=15);
  106. vshasigma${sz} $Sigma,@X[($j+1)%16],0,0
  107. vaddu${sz}m @X[$j],@X[$j],$Sigma
  108. vshasigma${sz} $Sigma,@X[($j+14)%16],0,15
  109. vaddu${sz}m @X[$j],@X[$j],$Sigma
  110. vaddu${sz}m @X[$j],@X[$j],@X[($j+9)%16]
  111. ___
  112. $code.=<<___;
  113. vaddu${sz}m $h,$h,@X[$i%16] ; h+=X[i]
  114. vsel $Func,$g,$f,$e ; Ch(e,f,g)
  115. vaddu${sz}m $g,$g,$Ki ; future h+=K[i]
  116. vaddu${sz}m $h,$h,$Func ; h+=Ch(e,f,g)
  117. vshasigma${sz} $Sigma,$e,1,15 ; Sigma1(e)
  118. vaddu${sz}m $h,$h,$Sigma ; h+=Sigma1(e)
  119. vxor $Func,$a,$b
  120. vsel $Func,$b,$c,$Func ; Maj(a,b,c)
  121. vaddu${sz}m $d,$d,$h ; d+=h
  122. vshasigma${sz} $Sigma,$a,1,0 ; Sigma0(a)
  123. vaddu${sz}m $Sigma,$Sigma,$Func ; Sigma0(a)+Maj(a,b,c)
  124. vaddu${sz}m $h,$h,$Sigma ; h+=Sigma0(a)+Maj(a,b,c)
  125. lvx $Ki,@I[$k],$idx ; load next K[i]
  126. ___
  127. $code.=<<___ if ($k == 7);
  128. addi $idx,$idx,0x80
  129. ___
  130. }
  131. $code=<<___;
  132. .machine "any"
  133. .text
  134. .globl $func
  135. .align 6
  136. $func:
  137. $STU $sp,-$FRAME($sp)
  138. mflr $lrsave
  139. li r10,`$LOCALS+15`
  140. li r11,`$LOCALS+31`
  141. stvx v24,r10,$sp # ABI says so
  142. addi r10,r10,32
  143. mfspr $vrsave,256
  144. stvx v25,r11,$sp
  145. addi r11,r11,32
  146. stvx v26,r10,$sp
  147. addi r10,r10,32
  148. stvx v27,r11,$sp
  149. addi r11,r11,32
  150. stvx v28,r10,$sp
  151. addi r10,r10,32
  152. stvx v29,r11,$sp
  153. addi r11,r11,32
  154. stvx v30,r10,$sp
  155. stvx v31,r11,$sp
  156. li r11,-4096+255 # 0xfffff0ff
  157. stw $vrsave,`$FRAME-6*$SIZE_T-4`($sp) # save vrsave
  158. li $x10,0x10
  159. $PUSH r26,`$FRAME-6*$SIZE_T`($sp)
  160. li $x20,0x20
  161. $PUSH r27,`$FRAME-5*$SIZE_T`($sp)
  162. li $x30,0x30
  163. $PUSH r28,`$FRAME-4*$SIZE_T`($sp)
  164. li $x40,0x40
  165. $PUSH r29,`$FRAME-3*$SIZE_T`($sp)
  166. li $x50,0x50
  167. $PUSH r30,`$FRAME-2*$SIZE_T`($sp)
  168. li $x60,0x60
  169. $PUSH r31,`$FRAME-1*$SIZE_T`($sp)
  170. li $x70,0x70
  171. $PUSH $lrsave,`$FRAME+$LRSAVE`($sp)
  172. mtspr 256,r11
  173. bl LPICmeup
  174. addi $offload,$sp,`8*$SIZE_T+15`
  175. ___
  176. $code.=<<___ if ($LENDIAN);
  177. li $idx,8
  178. lvsl $lemask,0,$idx
  179. vspltisb $Ki,0x0f
  180. vxor $lemask,$lemask,$Ki
  181. ___
  182. $code.=<<___ if ($SZ==4);
  183. lvx_4w $A,$x00,$ctx
  184. lvx_4w $E,$x10,$ctx
  185. vsldoi $B,$A,$A,4 # unpack
  186. vsldoi $C,$A,$A,8
  187. vsldoi $D,$A,$A,12
  188. vsldoi $F,$E,$E,4
  189. vsldoi $G,$E,$E,8
  190. vsldoi $H,$E,$E,12
  191. ___
  192. $code.=<<___ if ($SZ==8);
  193. lvx_u $A,$x00,$ctx
  194. lvx_u $C,$x10,$ctx
  195. lvx_u $E,$x20,$ctx
  196. vsldoi $B,$A,$A,8 # unpack
  197. lvx_u $G,$x30,$ctx
  198. vsldoi $D,$C,$C,8
  199. vsldoi $F,$E,$E,8
  200. vsldoi $H,$G,$G,8
  201. ___
  202. $code.=<<___;
  203. li r0,`($rounds-16)/16` # inner loop counter
  204. b Loop
  205. .align 5
  206. Loop:
  207. lvx $Ki,$x00,$Tbl
  208. lvx_u @X[0],0,$inp
  209. addi $inp,$inp,16
  210. mr $idx,$Tbl # copy $Tbl
  211. stvx $A,$x00,$offload # offload $A-$H
  212. stvx $B,$x10,$offload
  213. stvx $C,$x20,$offload
  214. stvx $D,$x30,$offload
  215. stvx $E,$x40,$offload
  216. stvx $F,$x50,$offload
  217. stvx $G,$x60,$offload
  218. stvx $H,$x70,$offload
  219. vaddu${sz}m $H,$H,$Ki # h+K[i]
  220. lvx $Ki,$x10,$Tbl
  221. ___
  222. for ($i=0;$i<16;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); }
  223. $code.=<<___;
  224. mtctr r0
  225. b L16_xx
  226. .align 5
  227. L16_xx:
  228. ___
  229. for (;$i<32;$i++) { &ROUND($i,@V); unshift(@V,pop(@V)); }
  230. $code.=<<___;
  231. bdnz L16_xx
  232. lvx @X[2],$x00,$offload
  233. subic. $num,$num,1
  234. lvx @X[3],$x10,$offload
  235. vaddu${sz}m $A,$A,@X[2]
  236. lvx @X[4],$x20,$offload
  237. vaddu${sz}m $B,$B,@X[3]
  238. lvx @X[5],$x30,$offload
  239. vaddu${sz}m $C,$C,@X[4]
  240. lvx @X[6],$x40,$offload
  241. vaddu${sz}m $D,$D,@X[5]
  242. lvx @X[7],$x50,$offload
  243. vaddu${sz}m $E,$E,@X[6]
  244. lvx @X[8],$x60,$offload
  245. vaddu${sz}m $F,$F,@X[7]
  246. lvx @X[9],$x70,$offload
  247. vaddu${sz}m $G,$G,@X[8]
  248. vaddu${sz}m $H,$H,@X[9]
  249. bne Loop
  250. ___
  251. $code.=<<___ if ($SZ==4);
  252. lvx @X[0],$x20,$idx
  253. vperm $A,$A,$B,$Ki # pack the answer
  254. lvx @X[1],$x30,$idx
  255. vperm $E,$E,$F,$Ki
  256. vperm $A,$A,$C,@X[0]
  257. vperm $E,$E,$G,@X[0]
  258. vperm $A,$A,$D,@X[1]
  259. vperm $E,$E,$H,@X[1]
  260. stvx_4w $A,$x00,$ctx
  261. stvx_4w $E,$x10,$ctx
  262. ___
  263. $code.=<<___ if ($SZ==8);
  264. vperm $A,$A,$B,$Ki # pack the answer
  265. vperm $C,$C,$D,$Ki
  266. vperm $E,$E,$F,$Ki
  267. vperm $G,$G,$H,$Ki
  268. stvx_u $A,$x00,$ctx
  269. stvx_u $C,$x10,$ctx
  270. stvx_u $E,$x20,$ctx
  271. stvx_u $G,$x30,$ctx
  272. ___
  273. $code.=<<___;
  274. addi $offload,$sp,`$LOCALS+15`
  275. mtlr $lrsave
  276. mtspr 256,$vrsave
  277. lvx v24,$x00,$offload # ABI says so
  278. lvx v25,$x10,$offload
  279. lvx v26,$x20,$offload
  280. lvx v27,$x30,$offload
  281. lvx v28,$x40,$offload
  282. lvx v29,$x50,$offload
  283. lvx v30,$x60,$offload
  284. lvx v31,$x70,$offload
  285. $POP r26,`$FRAME-6*$SIZE_T`($sp)
  286. $POP r27,`$FRAME-5*$SIZE_T`($sp)
  287. $POP r28,`$FRAME-4*$SIZE_T`($sp)
  288. $POP r29,`$FRAME-3*$SIZE_T`($sp)
  289. $POP r30,`$FRAME-2*$SIZE_T`($sp)
  290. $POP r31,`$FRAME-1*$SIZE_T`($sp)
  291. addi $sp,$sp,$FRAME
  292. blr
  293. .long 0
  294. .byte 0,12,4,1,0x80,6,3,0
  295. .long 0
  296. .size $func,.-$func
  297. ___
  298. # Ugly hack here, because PPC assembler syntax seem to vary too
  299. # much from platforms to platform...
  300. $code.=<<___;
  301. .align 6
  302. LPICmeup:
  303. mflr r0
  304. bcl 20,31,\$+4
  305. mflr $Tbl ; vvvvvv "distance" between . and 1st data entry
  306. addi $Tbl,$Tbl,`64-8`
  307. mtlr r0
  308. blr
  309. .long 0
  310. .byte 0,12,0x14,0,0,0,0,0
  311. .space `64-9*4`
  312. ___
  313. if ($SZ==8) {
  314. local *table = sub {
  315. foreach(@_) { $code.=".quad $_,$_\n"; }
  316. };
  317. table(
  318. "0x428a2f98d728ae22","0x7137449123ef65cd",
  319. "0xb5c0fbcfec4d3b2f","0xe9b5dba58189dbbc",
  320. "0x3956c25bf348b538","0x59f111f1b605d019",
  321. "0x923f82a4af194f9b","0xab1c5ed5da6d8118",
  322. "0xd807aa98a3030242","0x12835b0145706fbe",
  323. "0x243185be4ee4b28c","0x550c7dc3d5ffb4e2",
  324. "0x72be5d74f27b896f","0x80deb1fe3b1696b1",
  325. "0x9bdc06a725c71235","0xc19bf174cf692694",
  326. "0xe49b69c19ef14ad2","0xefbe4786384f25e3",
  327. "0x0fc19dc68b8cd5b5","0x240ca1cc77ac9c65",
  328. "0x2de92c6f592b0275","0x4a7484aa6ea6e483",
  329. "0x5cb0a9dcbd41fbd4","0x76f988da831153b5",
  330. "0x983e5152ee66dfab","0xa831c66d2db43210",
  331. "0xb00327c898fb213f","0xbf597fc7beef0ee4",
  332. "0xc6e00bf33da88fc2","0xd5a79147930aa725",
  333. "0x06ca6351e003826f","0x142929670a0e6e70",
  334. "0x27b70a8546d22ffc","0x2e1b21385c26c926",
  335. "0x4d2c6dfc5ac42aed","0x53380d139d95b3df",
  336. "0x650a73548baf63de","0x766a0abb3c77b2a8",
  337. "0x81c2c92e47edaee6","0x92722c851482353b",
  338. "0xa2bfe8a14cf10364","0xa81a664bbc423001",
  339. "0xc24b8b70d0f89791","0xc76c51a30654be30",
  340. "0xd192e819d6ef5218","0xd69906245565a910",
  341. "0xf40e35855771202a","0x106aa07032bbd1b8",
  342. "0x19a4c116b8d2d0c8","0x1e376c085141ab53",
  343. "0x2748774cdf8eeb99","0x34b0bcb5e19b48a8",
  344. "0x391c0cb3c5c95a63","0x4ed8aa4ae3418acb",
  345. "0x5b9cca4f7763e373","0x682e6ff3d6b2b8a3",
  346. "0x748f82ee5defb2fc","0x78a5636f43172f60",
  347. "0x84c87814a1f0ab72","0x8cc702081a6439ec",
  348. "0x90befffa23631e28","0xa4506cebde82bde9",
  349. "0xbef9a3f7b2c67915","0xc67178f2e372532b",
  350. "0xca273eceea26619c","0xd186b8c721c0c207",
  351. "0xeada7dd6cde0eb1e","0xf57d4f7fee6ed178",
  352. "0x06f067aa72176fba","0x0a637dc5a2c898a6",
  353. "0x113f9804bef90dae","0x1b710b35131c471b",
  354. "0x28db77f523047d84","0x32caab7b40c72493",
  355. "0x3c9ebe0a15c9bebc","0x431d67c49c100d4c",
  356. "0x4cc5d4becb3e42b6","0x597f299cfc657e2a",
  357. "0x5fcb6fab3ad6faec","0x6c44198c4a475817","0");
  358. $code.=<<___ if (!$LENDIAN);
  359. .quad 0x0001020304050607,0x1011121314151617
  360. ___
  361. $code.=<<___ if ($LENDIAN); # quad-swapped
  362. .quad 0x1011121314151617,0x0001020304050607
  363. ___
  364. } else {
  365. local *table = sub {
  366. foreach(@_) { $code.=".long $_,$_,$_,$_\n"; }
  367. };
  368. table(
  369. "0x428a2f98","0x71374491","0xb5c0fbcf","0xe9b5dba5",
  370. "0x3956c25b","0x59f111f1","0x923f82a4","0xab1c5ed5",
  371. "0xd807aa98","0x12835b01","0x243185be","0x550c7dc3",
  372. "0x72be5d74","0x80deb1fe","0x9bdc06a7","0xc19bf174",
  373. "0xe49b69c1","0xefbe4786","0x0fc19dc6","0x240ca1cc",
  374. "0x2de92c6f","0x4a7484aa","0x5cb0a9dc","0x76f988da",
  375. "0x983e5152","0xa831c66d","0xb00327c8","0xbf597fc7",
  376. "0xc6e00bf3","0xd5a79147","0x06ca6351","0x14292967",
  377. "0x27b70a85","0x2e1b2138","0x4d2c6dfc","0x53380d13",
  378. "0x650a7354","0x766a0abb","0x81c2c92e","0x92722c85",
  379. "0xa2bfe8a1","0xa81a664b","0xc24b8b70","0xc76c51a3",
  380. "0xd192e819","0xd6990624","0xf40e3585","0x106aa070",
  381. "0x19a4c116","0x1e376c08","0x2748774c","0x34b0bcb5",
  382. "0x391c0cb3","0x4ed8aa4a","0x5b9cca4f","0x682e6ff3",
  383. "0x748f82ee","0x78a5636f","0x84c87814","0x8cc70208",
  384. "0x90befffa","0xa4506ceb","0xbef9a3f7","0xc67178f2","0");
  385. $code.=<<___ if (!$LENDIAN);
  386. .long 0x00010203,0x10111213,0x10111213,0x10111213
  387. .long 0x00010203,0x04050607,0x10111213,0x10111213
  388. .long 0x00010203,0x04050607,0x08090a0b,0x10111213
  389. ___
  390. $code.=<<___ if ($LENDIAN); # word-swapped
  391. .long 0x10111213,0x10111213,0x10111213,0x00010203
  392. .long 0x10111213,0x10111213,0x04050607,0x00010203
  393. .long 0x10111213,0x08090a0b,0x04050607,0x00010203
  394. ___
  395. }
  396. $code.=<<___;
  397. .asciz "SHA${bits} for PowerISA 2.07, CRYPTOGAMS by <appro\@openssl.org>"
  398. .align 2
  399. ___
  400. $code =~ s/\`([^\`]*)\`/eval $1/gem;
  401. print $code;
  402. close STDOUT;