1#!/usr/bin/env perl 2# 3# ==================================================================== 4# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL 5# project. The module is, however, dual licensed under OpenSSL and 6# CRYPTOGAMS licenses depending on where you obtain it. For further 7# details see http://www.openssl.org/~appro/cryptogams/. 8# ==================================================================== 9# 10# SHA256/512_Transform for Itanium. 11# 12# sha512_block runs in 1003 cycles on Itanium 2, which is almost 50% 13# faster than gcc and >60%(!) faster than code generated by HP-UX 14# compiler (yes, HP-UX is generating slower code, because unlike gcc, 15# it failed to deploy "shift right pair," 'shrp' instruction, which 16# substitutes for 64-bit rotate). 17# 18# 924 cycles long sha256_block outperforms gcc by over factor of 2(!) 19# and HP-UX compiler - by >40% (yes, gcc won sha512_block, but lost 20# this one big time). Note that "formally" 924 is about 100 cycles 21# too much. I mean it's 64 32-bit rounds vs. 80 virtually identical 22# 64-bit ones and 1003*64/80 gives 802. Extra cycles, 2 per round, 23# are spent on extra work to provide for 32-bit rotations. 32-bit 24# rotations are still handled by 'shrp' instruction and for this 25# reason lower 32 bits are deposited to upper half of 64-bit register 26# prior 'shrp' issue. And in order to minimize the amount of such 27# operations, X[16] values are *maintained* with copies of lower 28# halves in upper halves, which is why you'll spot such instructions 29# as custom 'mux2', "parallel 32-bit add," 'padd4' and "parallel 30# 32-bit unsigned right shift," 'pshr4.u' instructions here. 31# 32# Rules of engagement. 33# 34# There is only one integer shifter meaning that if I have two rotate, 35# deposit or extract instructions in adjacent bundles, they shall 36# split [at run-time if they have to]. But note that variable and 37# parallel shifts are performed by multi-media ALU and *are* pairable 38# with rotates [and alike]. On the backside MMALU is rather slow: it 39# takes 2 extra cycles before the result of integer operation is 40# available *to* MMALU and 2(*) extra cycles before the result of MM 41# operation is available "back" *to* integer ALU, not to mention that 42# MMALU itself has 2 cycles latency. However! I explicitly scheduled 43# these MM instructions to avoid MM stalls, so that all these extra 44# latencies get "hidden" in instruction-level parallelism. 45# 46# (*) 2 cycles on Itanium 1 and 1 cycle on Itanium 2. But I schedule 47# for 2 in order to provide for best *overall* performance, 48# because on Itanium 1 stall on MM result is accompanied by 49# pipeline flush, which takes 6 cycles:-( 50# 51# June 2012 52# 53# Improve performance by 15-20%. Note about "rules of engagement" 54# above. Contemporary cores are equipped with additional shifter, 55# so that they should perform even better than below, presumably 56# by ~10%. 57# 58###################################################################### 59# Current performance in cycles per processed byte for Itanium 2 60# pre-9000 series [little-endian] system: 61# 62# SHA1(*) 5.7 63# SHA256 12.6 64# SHA512 6.7 65# 66# (*) SHA1 result is presented purely for reference purposes. 67# 68# To generate code, pass the file name with either 256 or 512 in its 69# name and compiler flags. 70 71$output=shift; 72 73if ($output =~ /512.*\.[s|asm]/) { 74 $SZ=8; 75 $BITS=8*$SZ; 76 $LDW="ld8"; 77 $STW="st8"; 78 $ADD="add"; 79 $SHRU="shr.u"; 80 $TABLE="K512"; 81 $func="sha512_block_data_order"; 82 @Sigma0=(28,34,39); 83 @Sigma1=(14,18,41); 84 @sigma0=(1, 8, 7); 85 @sigma1=(19,61, 6); 86 $rounds=80; 87} elsif ($output =~ /256.*\.[s|asm]/) { 88 $SZ=4; 89 $BITS=8*$SZ; 90 $LDW="ld4"; 91 $STW="st4"; 92 $ADD="padd4"; 93 $SHRU="pshr4.u"; 94 $TABLE="K256"; 95 $func="sha256_block_data_order"; 96 @Sigma0=( 2,13,22); 97 @Sigma1=( 6,11,25); 98 @sigma0=( 7,18, 3); 99 @sigma1=(17,19,10); 100 $rounds=64; 101} else { die "nonsense $output"; } 102 103open STDOUT,">$output" || die "can't open $output: $!"; 104 105if ($^O eq "hpux") { 106 $ADDP="addp4"; 107 for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); } 108} else { $ADDP="add"; } 109for (@ARGV) { $big_endian=1 if (/\-DB_ENDIAN/); 110 $big_endian=0 if (/\-DL_ENDIAN/); } 111if (!defined($big_endian)) 112 { $big_endian=(unpack('L',pack('N',1))==1); } 113 114$code=<<___; 115.ident \"$output, version 2.0\" 116.ident \"IA-64 ISA artwork by Andy Polyakov <appro\@openssl.org>\" 117.explicit 118.text 119 120pfssave=r2; 121lcsave=r3; 122prsave=r14; 123K=r15; 124A_=r16; B_=r17; C_=r18; D_=r19; 125E_=r20; F_=r21; G_=r22; H_=r23; 126T1=r24; T2=r25; 127s0=r26; s1=r27; t0=r28; t1=r29; 128Ktbl=r30; 129ctx=r31; // 1st arg 130input=r56; // 2nd arg 131num=r57; // 3rd arg 132sgm0=r58; sgm1=r59; // small constants 133 134// void $func (SHA_CTX *ctx, const void *in,size_t num[,int host]) 135.global $func# 136.proc $func# 137.align 32 138.skip 16 139$func: 140 .prologue 141 .save ar.pfs,pfssave 142{ .mmi; alloc pfssave=ar.pfs,3,25,0,24 143 $ADDP ctx=0,r32 // 1st arg 144 .save ar.lc,lcsave 145 mov lcsave=ar.lc } 146{ .mmi; $ADDP input=0,r33 // 2nd arg 147 mov num=r34 // 3rd arg 148 .save pr,prsave 149 mov prsave=pr };; 150 151 .body 152{ .mib; add r8=0*$SZ,ctx 153 add r9=1*$SZ,ctx } 154{ .mib; add r10=2*$SZ,ctx 155 add r11=3*$SZ,ctx };; 156 157// load A-H 158.Lpic_point: 159{ .mmi; $LDW A_=[r8],4*$SZ 160 $LDW B_=[r9],4*$SZ 161 mov Ktbl=ip } 162{ .mmi; $LDW C_=[r10],4*$SZ 163 $LDW D_=[r11],4*$SZ 164 mov sgm0=$sigma0[2] };; 165{ .mmi; $LDW E_=[r8] 166 $LDW F_=[r9] 167 add Ktbl=($TABLE#-.Lpic_point),Ktbl } 168{ .mmi; $LDW G_=[r10] 169 $LDW H_=[r11] 170 cmp.ne p0,p16=0,r0 };; 171___ 172$code.=<<___ if ($BITS==64); 173{ .mii; and r8=7,input 174 and input=~7,input;; 175 cmp.eq p9,p0=1,r8 } 176{ .mmi; cmp.eq p10,p0=2,r8 177 cmp.eq p11,p0=3,r8 178 cmp.eq p12,p0=4,r8 } 179{ .mmi; cmp.eq p13,p0=5,r8 180 cmp.eq p14,p0=6,r8 181 cmp.eq p15,p0=7,r8 };; 182___ 183$code.=<<___; 184.L_outer: 185.rotr R[8],X[16] 186A=R[0]; B=R[1]; C=R[2]; D=R[3]; E=R[4]; F=R[5]; G=R[6]; H=R[7] 187{ .mmi; ld1 X[15]=[input],$SZ // eliminated in sha512 188 mov A=A_ 189 mov ar.lc=14 } 190{ .mmi; mov B=B_ 191 mov C=C_ 192 mov D=D_ } 193{ .mmi; mov E=E_ 194 mov F=F_ 195 mov ar.ec=2 };; 196{ .mmi; mov G=G_ 197 mov H=H_ 198 mov sgm1=$sigma1[2] } 199{ .mib; mov r8=0 200 add r9=1-$SZ,input 201 brp.loop.imp .L_first16,.L_first16_end-16 };; 202___ 203$t0="A", $t1="E", $code.=<<___ if ($BITS==64); 204// in sha512 case I load whole X[16] at once and take care of alignment... 205{ .mmi; add r8=1*$SZ,input 206 add r9=2*$SZ,input 207 add r10=3*$SZ,input };; 208{ .mmb; $LDW X[15]=[input],4*$SZ 209 $LDW X[14]=[r8],4*$SZ 210(p9) br.cond.dpnt.many .L1byte };; 211{ .mmb; $LDW X[13]=[r9],4*$SZ 212 $LDW X[12]=[r10],4*$SZ 213(p10) br.cond.dpnt.many .L2byte };; 214{ .mmb; $LDW X[11]=[input],4*$SZ 215 $LDW X[10]=[r8],4*$SZ 216(p11) br.cond.dpnt.many .L3byte };; 217{ .mmb; $LDW X[ 9]=[r9],4*$SZ 218 $LDW X[ 8]=[r10],4*$SZ 219(p12) br.cond.dpnt.many .L4byte };; 220{ .mmb; $LDW X[ 7]=[input],4*$SZ 221 $LDW X[ 6]=[r8],4*$SZ 222(p13) br.cond.dpnt.many .L5byte };; 223{ .mmb; $LDW X[ 5]=[r9],4*$SZ 224 $LDW X[ 4]=[r10],4*$SZ 225(p14) br.cond.dpnt.many .L6byte };; 226{ .mmb; $LDW X[ 3]=[input],4*$SZ 227 $LDW X[ 2]=[r8],4*$SZ 228(p15) br.cond.dpnt.many .L7byte };; 229{ .mmb; $LDW X[ 1]=[r9],4*$SZ 230 $LDW X[ 0]=[r10],4*$SZ } 231{ .mib; mov r8=0 232 mux1 X[15]=X[15],\@rev // eliminated on big-endian 233 br.many .L_first16 };; 234.L1byte: 235{ .mmi; $LDW X[13]=[r9],4*$SZ 236 $LDW X[12]=[r10],4*$SZ 237 shrp X[15]=X[15],X[14],56 };; 238{ .mmi; $LDW X[11]=[input],4*$SZ 239 $LDW X[10]=[r8],4*$SZ 240 shrp X[14]=X[14],X[13],56 } 241{ .mmi; $LDW X[ 9]=[r9],4*$SZ 242 $LDW X[ 8]=[r10],4*$SZ 243 shrp X[13]=X[13],X[12],56 };; 244{ .mmi; $LDW X[ 7]=[input],4*$SZ 245 $LDW X[ 6]=[r8],4*$SZ 246 shrp X[12]=X[12],X[11],56 } 247{ .mmi; $LDW X[ 5]=[r9],4*$SZ 248 $LDW X[ 4]=[r10],4*$SZ 249 shrp X[11]=X[11],X[10],56 };; 250{ .mmi; $LDW X[ 3]=[input],4*$SZ 251 $LDW X[ 2]=[r8],4*$SZ 252 shrp X[10]=X[10],X[ 9],56 } 253{ .mmi; $LDW X[ 1]=[r9],4*$SZ 254 $LDW X[ 0]=[r10],4*$SZ 255 shrp X[ 9]=X[ 9],X[ 8],56 };; 256{ .mii; $LDW T1=[input] 257 shrp X[ 8]=X[ 8],X[ 7],56 258 shrp X[ 7]=X[ 7],X[ 6],56 } 259{ .mii; shrp X[ 6]=X[ 6],X[ 5],56 260 shrp X[ 5]=X[ 5],X[ 4],56 };; 261{ .mii; shrp X[ 4]=X[ 4],X[ 3],56 262 shrp X[ 3]=X[ 3],X[ 2],56 } 263{ .mii; shrp X[ 2]=X[ 2],X[ 1],56 264 shrp X[ 1]=X[ 1],X[ 0],56 } 265{ .mib; shrp X[ 0]=X[ 0],T1,56 } 266{ .mib; mov r8=0 267 mux1 X[15]=X[15],\@rev // eliminated on big-endian 268 br.many .L_first16 };; 269.L2byte: 270{ .mmi; $LDW X[11]=[input],4*$SZ 271 $LDW X[10]=[r8],4*$SZ 272 shrp X[15]=X[15],X[14],48 } 273{ .mmi; $LDW X[ 9]=[r9],4*$SZ 274 $LDW X[ 8]=[r10],4*$SZ 275 shrp X[14]=X[14],X[13],48 };; 276{ .mmi; $LDW X[ 7]=[input],4*$SZ 277 $LDW X[ 6]=[r8],4*$SZ 278 shrp X[13]=X[13],X[12],48 } 279{ .mmi; $LDW X[ 5]=[r9],4*$SZ 280 $LDW X[ 4]=[r10],4*$SZ 281 shrp X[12]=X[12],X[11],48 };; 282{ .mmi; $LDW X[ 3]=[input],4*$SZ 283 $LDW X[ 2]=[r8],4*$SZ 284 shrp X[11]=X[11],X[10],48 } 285{ .mmi; $LDW X[ 1]=[r9],4*$SZ 286 $LDW X[ 0]=[r10],4*$SZ 287 shrp X[10]=X[10],X[ 9],48 };; 288{ .mii; $LDW T1=[input] 289 shrp X[ 9]=X[ 9],X[ 8],48 290 shrp X[ 8]=X[ 8],X[ 7],48 } 291{ .mii; shrp X[ 7]=X[ 7],X[ 6],48 292 shrp X[ 6]=X[ 6],X[ 5],48 };; 293{ .mii; shrp X[ 5]=X[ 5],X[ 4],48 294 shrp X[ 4]=X[ 4],X[ 3],48 } 295{ .mii; shrp X[ 3]=X[ 3],X[ 2],48 296 shrp X[ 2]=X[ 2],X[ 1],48 } 297{ .mii; shrp X[ 1]=X[ 1],X[ 0],48 298 shrp X[ 0]=X[ 0],T1,48 } 299{ .mib; mov r8=0 300 mux1 X[15]=X[15],\@rev // eliminated on big-endian 301 br.many .L_first16 };; 302.L3byte: 303{ .mmi; $LDW X[ 9]=[r9],4*$SZ 304 $LDW X[ 8]=[r10],4*$SZ 305 shrp X[15]=X[15],X[14],40 };; 306{ .mmi; $LDW X[ 7]=[input],4*$SZ 307 $LDW X[ 6]=[r8],4*$SZ 308 shrp X[14]=X[14],X[13],40 } 309{ .mmi; $LDW X[ 5]=[r9],4*$SZ 310 $LDW X[ 4]=[r10],4*$SZ 311 shrp X[13]=X[13],X[12],40 };; 312{ .mmi; $LDW X[ 3]=[input],4*$SZ 313 $LDW X[ 2]=[r8],4*$SZ 314 shrp X[12]=X[12],X[11],40 } 315{ .mmi; $LDW X[ 1]=[r9],4*$SZ 316 $LDW X[ 0]=[r10],4*$SZ 317 shrp X[11]=X[11],X[10],40 };; 318{ .mii; $LDW T1=[input] 319 shrp X[10]=X[10],X[ 9],40 320 shrp X[ 9]=X[ 9],X[ 8],40 } 321{ .mii; shrp X[ 8]=X[ 8],X[ 7],40 322 shrp X[ 7]=X[ 7],X[ 6],40 };; 323{ .mii; shrp X[ 6]=X[ 6],X[ 5],40 324 shrp X[ 5]=X[ 5],X[ 4],40 } 325{ .mii; shrp X[ 4]=X[ 4],X[ 3],40 326 shrp X[ 3]=X[ 3],X[ 2],40 } 327{ .mii; shrp X[ 2]=X[ 2],X[ 1],40 328 shrp X[ 1]=X[ 1],X[ 0],40 } 329{ .mib; shrp X[ 0]=X[ 0],T1,40 } 330{ .mib; mov r8=0 331 mux1 X[15]=X[15],\@rev // eliminated on big-endian 332 br.many .L_first16 };; 333.L4byte: 334{ .mmi; $LDW X[ 7]=[input],4*$SZ 335 $LDW X[ 6]=[r8],4*$SZ 336 shrp X[15]=X[15],X[14],32 } 337{ .mmi; $LDW X[ 5]=[r9],4*$SZ 338 $LDW X[ 4]=[r10],4*$SZ 339 shrp X[14]=X[14],X[13],32 };; 340{ .mmi; $LDW X[ 3]=[input],4*$SZ 341 $LDW X[ 2]=[r8],4*$SZ 342 shrp X[13]=X[13],X[12],32 } 343{ .mmi; $LDW X[ 1]=[r9],4*$SZ 344 $LDW X[ 0]=[r10],4*$SZ 345 shrp X[12]=X[12],X[11],32 };; 346{ .mii; $LDW T1=[input] 347 shrp X[11]=X[11],X[10],32 348 shrp X[10]=X[10],X[ 9],32 } 349{ .mii; shrp X[ 9]=X[ 9],X[ 8],32 350 shrp X[ 8]=X[ 8],X[ 7],32 };; 351{ .mii; shrp X[ 7]=X[ 7],X[ 6],32 352 shrp X[ 6]=X[ 6],X[ 5],32 } 353{ .mii; shrp X[ 5]=X[ 5],X[ 4],32 354 shrp X[ 4]=X[ 4],X[ 3],32 } 355{ .mii; shrp X[ 3]=X[ 3],X[ 2],32 356 shrp X[ 2]=X[ 2],X[ 1],32 } 357{ .mii; shrp X[ 1]=X[ 1],X[ 0],32 358 shrp X[ 0]=X[ 0],T1,32 } 359{ .mib; mov r8=0 360 mux1 X[15]=X[15],\@rev // eliminated on big-endian 361 br.many .L_first16 };; 362.L5byte: 363{ .mmi; $LDW X[ 5]=[r9],4*$SZ 364 $LDW X[ 4]=[r10],4*$SZ 365 shrp X[15]=X[15],X[14],24 };; 366{ .mmi; $LDW X[ 3]=[input],4*$SZ 367 $LDW X[ 2]=[r8],4*$SZ 368 shrp X[14]=X[14],X[13],24 } 369{ .mmi; $LDW X[ 1]=[r9],4*$SZ 370 $LDW X[ 0]=[r10],4*$SZ 371 shrp X[13]=X[13],X[12],24 };; 372{ .mii; $LDW T1=[input] 373 shrp X[12]=X[12],X[11],24 374 shrp X[11]=X[11],X[10],24 } 375{ .mii; shrp X[10]=X[10],X[ 9],24 376 shrp X[ 9]=X[ 9],X[ 8],24 };; 377{ .mii; shrp X[ 8]=X[ 8],X[ 7],24 378 shrp X[ 7]=X[ 7],X[ 6],24 } 379{ .mii; shrp X[ 6]=X[ 6],X[ 5],24 380 shrp X[ 5]=X[ 5],X[ 4],24 } 381{ .mii; shrp X[ 4]=X[ 4],X[ 3],24 382 shrp X[ 3]=X[ 3],X[ 2],24 } 383{ .mii; shrp X[ 2]=X[ 2],X[ 1],24 384 shrp X[ 1]=X[ 1],X[ 0],24 } 385{ .mib; shrp X[ 0]=X[ 0],T1,24 } 386{ .mib; mov r8=0 387 mux1 X[15]=X[15],\@rev // eliminated on big-endian 388 br.many .L_first16 };; 389.L6byte: 390{ .mmi; $LDW X[ 3]=[input],4*$SZ 391 $LDW X[ 2]=[r8],4*$SZ 392 shrp X[15]=X[15],X[14],16 } 393{ .mmi; $LDW X[ 1]=[r9],4*$SZ 394 $LDW X[ 0]=[r10],4*$SZ 395 shrp X[14]=X[14],X[13],16 };; 396{ .mii; $LDW T1=[input] 397 shrp X[13]=X[13],X[12],16 398 shrp X[12]=X[12],X[11],16 } 399{ .mii; shrp X[11]=X[11],X[10],16 400 shrp X[10]=X[10],X[ 9],16 };; 401{ .mii; shrp X[ 9]=X[ 9],X[ 8],16 402 shrp X[ 8]=X[ 8],X[ 7],16 } 403{ .mii; shrp X[ 7]=X[ 7],X[ 6],16 404 shrp X[ 6]=X[ 6],X[ 5],16 } 405{ .mii; shrp X[ 5]=X[ 5],X[ 4],16 406 shrp X[ 4]=X[ 4],X[ 3],16 } 407{ .mii; shrp X[ 3]=X[ 3],X[ 2],16 408 shrp X[ 2]=X[ 2],X[ 1],16 } 409{ .mii; shrp X[ 1]=X[ 1],X[ 0],16 410 shrp X[ 0]=X[ 0],T1,16 } 411{ .mib; mov r8=0 412 mux1 X[15]=X[15],\@rev // eliminated on big-endian 413 br.many .L_first16 };; 414.L7byte: 415{ .mmi; $LDW X[ 1]=[r9],4*$SZ 416 $LDW X[ 0]=[r10],4*$SZ 417 shrp X[15]=X[15],X[14],8 };; 418{ .mii; $LDW T1=[input] 419 shrp X[14]=X[14],X[13],8 420 shrp X[13]=X[13],X[12],8 } 421{ .mii; shrp X[12]=X[12],X[11],8 422 shrp X[11]=X[11],X[10],8 };; 423{ .mii; shrp X[10]=X[10],X[ 9],8 424 shrp X[ 9]=X[ 9],X[ 8],8 } 425{ .mii; shrp X[ 8]=X[ 8],X[ 7],8 426 shrp X[ 7]=X[ 7],X[ 6],8 } 427{ .mii; shrp X[ 6]=X[ 6],X[ 5],8 428 shrp X[ 5]=X[ 5],X[ 4],8 } 429{ .mii; shrp X[ 4]=X[ 4],X[ 3],8 430 shrp X[ 3]=X[ 3],X[ 2],8 } 431{ .mii; shrp X[ 2]=X[ 2],X[ 1],8 432 shrp X[ 1]=X[ 1],X[ 0],8 } 433{ .mib; shrp X[ 0]=X[ 0],T1,8 } 434{ .mib; mov r8=0 435 mux1 X[15]=X[15],\@rev };; // eliminated on big-endian 436 437.align 32 438.L_first16: 439{ .mmi; $LDW K=[Ktbl],$SZ 440 add A=A,r8 // H+=Sigma(0) from the past 441 _rotr r10=$t1,$Sigma1[0] } // ROTR(e,14) 442{ .mmi; and T1=F,E 443 andcm r8=G,E 444 (p16) mux1 X[14]=X[14],\@rev };; // eliminated on big-endian 445{ .mmi; and T2=A,B 446 and r9=A,C 447 _rotr r11=$t1,$Sigma1[1] } // ROTR(e,41) 448{ .mmi; xor T1=T1,r8 // T1=((e & f) ^ (~e & g)) 449 and r8=B,C };; 450___ 451$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32); 452.align 32 453.L_first16: 454{ .mmi; add A=A,r8 // H+=Sigma(0) from the past 455 add r10=2-$SZ,input 456 add r11=3-$SZ,input };; 457{ .mmi; ld1 r9=[r9] 458 ld1 r10=[r10] 459 dep.z $t1=E,32,32 } 460{ .mmi; ld1 r11=[r11] 461 $LDW K=[Ktbl],$SZ 462 zxt4 E=E };; 463{ .mii; or $t1=$t1,E 464 dep X[15]=X[15],r9,8,8 465 mux2 $t0=A,0x44 };; // copy lower half to upper 466{ .mmi; and T1=F,E 467 andcm r8=G,E 468 dep r11=r10,r11,8,8 };; 469{ .mmi; and T2=A,B 470 and r9=A,C 471 dep X[15]=X[15],r11,16,16 };; 472{ .mmi; (p16) ld1 X[15-1]=[input],$SZ // prefetch 473 xor T1=T1,r8 // T1=((e & f) ^ (~e & g)) 474 _rotr r10=$t1,$Sigma1[0] } // ROTR(e,14) 475{ .mmi; and r8=B,C 476 _rotr r11=$t1,$Sigma1[1] };; // ROTR(e,18) 477___ 478$code.=<<___; 479{ .mmi; add T1=T1,H // T1=Ch(e,f,g)+h 480 xor r10=r10,r11 481 _rotr r11=$t1,$Sigma1[2] } // ROTR(e,41) 482{ .mmi; xor T2=T2,r9 483 add K=K,X[15] };; 484{ .mmi; add T1=T1,K // T1+=K[i]+X[i] 485 xor T2=T2,r8 // T2=((a & b) ^ (a & c) ^ (b & c)) 486 _rotr r8=$t0,$Sigma0[0] } // ROTR(a,28) 487{ .mmi; xor r11=r11,r10 // Sigma1(e) 488 _rotr r9=$t0,$Sigma0[1] };; // ROTR(a,34) 489{ .mmi; add T1=T1,r11 // T+=Sigma1(e) 490 xor r8=r8,r9 491 _rotr r9=$t0,$Sigma0[2] };; // ROTR(a,39) 492{ .mmi; xor r8=r8,r9 // Sigma0(a) 493 add D=D,T1 494 mux2 H=X[15],0x44 } // mov H=X[15] in sha512 495{ .mib; (p16) add r9=1-$SZ,input // not used in sha512 496 add X[15]=T1,T2 // H=T1+Maj(a,b,c) 497 br.ctop.sptk .L_first16 };; 498.L_first16_end: 499 500{ .mib; mov ar.lc=$rounds-17 501 brp.loop.imp .L_rest,.L_rest_end-16 } 502{ .mib; mov ar.ec=1 503 br.many .L_rest };; 504 505.align 32 506.L_rest: 507{ .mmi; $LDW K=[Ktbl],$SZ 508 add A=A,r8 // H+=Sigma0(a) from the past 509 _rotr r8=X[15-1],$sigma0[0] } // ROTR(s0,1) 510{ .mmi; add X[15]=X[15],X[15-9] // X[i&0xF]+=X[(i+9)&0xF] 511 $SHRU s0=X[15-1],sgm0 };; // s0=X[(i+1)&0xF]>>7 512{ .mib; and T1=F,E 513 _rotr r9=X[15-1],$sigma0[1] } // ROTR(s0,8) 514{ .mib; andcm r10=G,E 515 $SHRU s1=X[15-14],sgm1 };; // s1=X[(i+14)&0xF]>>6 516// Pair of mmi; splits on Itanium 1 and prevents pipeline flush 517// upon $SHRU output usage 518{ .mmi; xor T1=T1,r10 // T1=((e & f) ^ (~e & g)) 519 xor r9=r8,r9 520 _rotr r10=X[15-14],$sigma1[0] }// ROTR(s1,19) 521{ .mmi; and T2=A,B 522 and r8=A,C 523 _rotr r11=X[15-14],$sigma1[1] };;// ROTR(s1,61) 524___ 525$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32); 526{ .mib; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF]) 527 dep.z $t1=E,32,32 } 528{ .mib; xor r10=r11,r10 529 zxt4 E=E };; 530{ .mii; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF]) 531 shrp r9=E,$t1,32+$Sigma1[0] // ROTR(e,14) 532 mux2 $t0=A,0x44 };; // copy lower half to upper 533// Pair of mmi; splits on Itanium 1 and prevents pipeline flush 534// upon mux2 output usage 535{ .mmi; xor T2=T2,r8 536 shrp r8=E,$t1,32+$Sigma1[1]} // ROTR(e,18) 537{ .mmi; and r10=B,C 538 add T1=T1,H // T1=Ch(e,f,g)+h 539 or $t1=$t1,E };; 540___ 541$t0="A", $t1="E", $code.=<<___ if ($BITS==64); 542{ .mib; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF]) 543 _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14) 544{ .mib; xor r10=r11,r10 545 xor T2=T2,r8 };; 546{ .mib; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF]) 547 _rotr r8=$t1,$Sigma1[1] } // ROTR(e,18) 548{ .mib; and r10=B,C 549 add T1=T1,H };; // T1+=H 550___ 551$code.=<<___; 552{ .mib; xor r9=r9,r8 553 _rotr r8=$t1,$Sigma1[2] } // ROTR(e,41) 554{ .mib; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c)) 555 add X[15]=X[15],s0 };; // X[i]+=sigma0(X[i+1]) 556{ .mmi; xor r9=r9,r8 // Sigma1(e) 557 add X[15]=X[15],s1 // X[i]+=sigma0(X[i+14]) 558 _rotr r8=$t0,$Sigma0[0] };; // ROTR(a,28) 559{ .mmi; add K=K,X[15] 560 add T1=T1,r9 // T1+=Sigma1(e) 561 _rotr r9=$t0,$Sigma0[1] };; // ROTR(a,34) 562{ .mmi; add T1=T1,K // T1+=K[i]+X[i] 563 xor r8=r8,r9 564 _rotr r9=$t0,$Sigma0[2] };; // ROTR(a,39) 565{ .mib; add D=D,T1 566 mux2 H=X[15],0x44 } // mov H=X[15] in sha512 567{ .mib; xor r8=r8,r9 // Sigma0(a) 568 add X[15]=T1,T2 // H=T1+Maj(a,b,c) 569 br.ctop.sptk .L_rest };; 570.L_rest_end: 571 572{ .mmi; add A=A,r8 };; // H+=Sigma0(a) from the past 573{ .mmi; add A_=A_,A 574 add B_=B_,B 575 add C_=C_,C } 576{ .mmi; add D_=D_,D 577 add E_=E_,E 578 cmp.ltu p16,p0=1,num };; 579{ .mmi; add F_=F_,F 580 add G_=G_,G 581 add H_=H_,H } 582{ .mmb; add Ktbl=-$SZ*$rounds,Ktbl 583(p16) add num=-1,num 584(p16) br.dptk.many .L_outer };; 585 586{ .mib; add r8=0*$SZ,ctx 587 add r9=1*$SZ,ctx } 588{ .mib; add r10=2*$SZ,ctx 589 add r11=3*$SZ,ctx };; 590{ .mmi; $STW [r8]=A_,4*$SZ 591 $STW [r9]=B_,4*$SZ 592 mov ar.lc=lcsave } 593{ .mmi; $STW [r10]=C_,4*$SZ 594 $STW [r11]=D_,4*$SZ 595 mov pr=prsave,0x1ffff };; 596{ .mmb; $STW [r8]=E_ 597 $STW [r9]=F_ } 598{ .mmb; $STW [r10]=G_ 599 $STW [r11]=H_ 600 br.ret.sptk.many b0 };; 601.endp $func# 602___ 603 604foreach(split($/,$code)) { 605 s/\`([^\`]*)\`/eval $1/gem; 606 s/_rotr(\s+)([^=]+)=([^,]+),([0-9]+)/shrp$1$2=$3,$3,$4/gm; 607 if ($BITS==64) { 608 s/mux2(\s+)([^=]+)=([^,]+),\S+/mov$1 $2=$3/gm; 609 s/mux1(\s+)\S+/nop.i$1 0x0/gm if ($big_endian); 610 s/(shrp\s+X\[[^=]+)=([^,]+),([^,]+),([1-9]+)/$1=$3,$2,64-$4/gm 611 if (!$big_endian); 612 s/ld1(\s+)X\[\S+/nop.m$1 0x0/gm; 613 } 614 615 print $_,"\n"; 616} 617 618print<<___ if ($BITS==32); 619.align 64 620.type K256#,\@object 621K256: data4 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5 622 data4 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5 623 data4 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3 624 data4 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174 625 data4 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc 626 data4 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da 627 data4 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7 628 data4 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967 629 data4 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13 630 data4 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85 631 data4 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3 632 data4 0xd192e819,0xd6990624,0xf40e3585,0x106aa070 633 data4 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5 634 data4 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3 635 data4 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208 636 data4 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2 637.size K256#,$SZ*$rounds 638stringz "SHA256 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>" 639___ 640print<<___ if ($BITS==64); 641.align 64 642.type K512#,\@object 643K512: data8 0x428a2f98d728ae22,0x7137449123ef65cd 644 data8 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc 645 data8 0x3956c25bf348b538,0x59f111f1b605d019 646 data8 0x923f82a4af194f9b,0xab1c5ed5da6d8118 647 data8 0xd807aa98a3030242,0x12835b0145706fbe 648 data8 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2 649 data8 0x72be5d74f27b896f,0x80deb1fe3b1696b1 650 data8 0x9bdc06a725c71235,0xc19bf174cf692694 651 data8 0xe49b69c19ef14ad2,0xefbe4786384f25e3 652 data8 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65 653 data8 0x2de92c6f592b0275,0x4a7484aa6ea6e483 654 data8 0x5cb0a9dcbd41fbd4,0x76f988da831153b5 655 data8 0x983e5152ee66dfab,0xa831c66d2db43210 656 data8 0xb00327c898fb213f,0xbf597fc7beef0ee4 657 data8 0xc6e00bf33da88fc2,0xd5a79147930aa725 658 data8 0x06ca6351e003826f,0x142929670a0e6e70 659 data8 0x27b70a8546d22ffc,0x2e1b21385c26c926 660 data8 0x4d2c6dfc5ac42aed,0x53380d139d95b3df 661 data8 0x650a73548baf63de,0x766a0abb3c77b2a8 662 data8 0x81c2c92e47edaee6,0x92722c851482353b 663 data8 0xa2bfe8a14cf10364,0xa81a664bbc423001 664 data8 0xc24b8b70d0f89791,0xc76c51a30654be30 665 data8 0xd192e819d6ef5218,0xd69906245565a910 666 data8 0xf40e35855771202a,0x106aa07032bbd1b8 667 data8 0x19a4c116b8d2d0c8,0x1e376c085141ab53 668 data8 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8 669 data8 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb 670 data8 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3 671 data8 0x748f82ee5defb2fc,0x78a5636f43172f60 672 data8 0x84c87814a1f0ab72,0x8cc702081a6439ec 673 data8 0x90befffa23631e28,0xa4506cebde82bde9 674 data8 0xbef9a3f7b2c67915,0xc67178f2e372532b 675 data8 0xca273eceea26619c,0xd186b8c721c0c207 676 data8 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178 677 data8 0x06f067aa72176fba,0x0a637dc5a2c898a6 678 data8 0x113f9804bef90dae,0x1b710b35131c471b 679 data8 0x28db77f523047d84,0x32caab7b40c72493 680 data8 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c 681 data8 0x4cc5d4becb3e42b6,0x597f299cfc657e2a 682 data8 0x5fcb6fab3ad6faec,0x6c44198c4a475817 683.size K512#,$SZ*$rounds 684stringz "SHA512 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>" 685___ 686