1160814Ssimon#!/usr/bin/env perl 2160814Ssimon# 3160814Ssimon# ==================================================================== 4160814Ssimon# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL 5194206Ssimon# project. The module is, however, dual licensed under OpenSSL and 6194206Ssimon# CRYPTOGAMS licenses depending on where you obtain it. For further 7194206Ssimon# details see http://www.openssl.org/~appro/cryptogams/. 8160814Ssimon# ==================================================================== 9160814Ssimon# 10160814Ssimon# SHA256/512_Transform for Itanium. 11160814Ssimon# 12160814Ssimon# sha512_block runs in 1003 cycles on Itanium 2, which is almost 50% 13160814Ssimon# faster than gcc and >60%(!) faster than code generated by HP-UX 14160814Ssimon# compiler (yes, HP-UX is generating slower code, because unlike gcc, 15160814Ssimon# it failed to deploy "shift right pair," 'shrp' instruction, which 16160814Ssimon# substitutes for 64-bit rotate). 17160814Ssimon# 18160814Ssimon# 924 cycles long sha256_block outperforms gcc by over factor of 2(!) 19160814Ssimon# and HP-UX compiler - by >40% (yes, gcc won sha512_block, but lost 20160814Ssimon# this one big time). Note that "formally" 924 is about 100 cycles 21160814Ssimon# too much. I mean it's 64 32-bit rounds vs. 80 virtually identical 22160814Ssimon# 64-bit ones and 1003*64/80 gives 802. Extra cycles, 2 per round, 23160814Ssimon# are spent on extra work to provide for 32-bit rotations. 32-bit 24160814Ssimon# rotations are still handled by 'shrp' instruction and for this 25160814Ssimon# reason lower 32 bits are deposited to upper half of 64-bit register 26160814Ssimon# prior 'shrp' issue. And in order to minimize the amount of such 27160814Ssimon# operations, X[16] values are *maintained* with copies of lower 28160814Ssimon# halves in upper halves, which is why you'll spot such instructions 29160814Ssimon# as custom 'mux2', "parallel 32-bit add," 'padd4' and "parallel 30160814Ssimon# 32-bit unsigned right shift," 'pshr4.u' instructions here. 31160814Ssimon# 32160814Ssimon# Rules of engagement. 33160814Ssimon# 34160814Ssimon# There is only one integer shifter meaning that if I have two rotate, 35160814Ssimon# deposit or extract instructions in adjacent bundles, they shall 36160814Ssimon# split [at run-time if they have to]. But note that variable and 37160814Ssimon# parallel shifts are performed by multi-media ALU and *are* pairable 38160814Ssimon# with rotates [and alike]. On the backside MMALU is rather slow: it 39160814Ssimon# takes 2 extra cycles before the result of integer operation is 40160814Ssimon# available *to* MMALU and 2(*) extra cycles before the result of MM 41160814Ssimon# operation is available "back" *to* integer ALU, not to mention that 42160814Ssimon# MMALU itself has 2 cycles latency. However! I explicitly scheduled 43160814Ssimon# these MM instructions to avoid MM stalls, so that all these extra 44160814Ssimon# latencies get "hidden" in instruction-level parallelism. 45160814Ssimon# 46160814Ssimon# (*) 2 cycles on Itanium 1 and 1 cycle on Itanium 2. But I schedule 47160814Ssimon# for 2 in order to provide for best *overall* performance, 48160814Ssimon# because on Itanium 1 stall on MM result is accompanied by 49160814Ssimon# pipeline flush, which takes 6 cycles:-( 50160814Ssimon# 51160814Ssimon# Resulting performance numbers for 900MHz Itanium 2 system: 52160814Ssimon# 53160814Ssimon# The 'numbers' are in 1000s of bytes per second processed. 54160814Ssimon# type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes 55160814Ssimon# sha1(*) 6210.14k 20376.30k 52447.83k 85870.05k 105478.12k 56160814Ssimon# sha256 7476.45k 20572.05k 41538.34k 56062.29k 62093.18k 57160814Ssimon# sha512 4996.56k 20026.28k 47597.20k 85278.79k 111501.31k 58160814Ssimon# 59160814Ssimon# (*) SHA1 numbers are for HP-UX compiler and are presented purely 60160814Ssimon# for reference purposes. I bet it can improved too... 61160814Ssimon# 62160814Ssimon# To generate code, pass the file name with either 256 or 512 in its 63160814Ssimon# name and compiler flags. 64160814Ssimon 65160814Ssimon$output=shift; 66160814Ssimon 67160814Ssimonif ($output =~ /512.*\.[s|asm]/) { 68160814Ssimon $SZ=8; 69160814Ssimon $BITS=8*$SZ; 70160814Ssimon $LDW="ld8"; 71160814Ssimon $STW="st8"; 72160814Ssimon $ADD="add"; 73160814Ssimon $SHRU="shr.u"; 74160814Ssimon $TABLE="K512"; 75194206Ssimon $func="sha512_block_data_order"; 76160814Ssimon @Sigma0=(28,34,39); 77160814Ssimon @Sigma1=(14,18,41); 78160814Ssimon @sigma0=(1, 8, 7); 79160814Ssimon @sigma1=(19,61, 6); 80160814Ssimon $rounds=80; 81160814Ssimon} elsif ($output =~ /256.*\.[s|asm]/) { 82160814Ssimon $SZ=4; 83160814Ssimon $BITS=8*$SZ; 84160814Ssimon $LDW="ld4"; 85160814Ssimon $STW="st4"; 86160814Ssimon $ADD="padd4"; 87160814Ssimon $SHRU="pshr4.u"; 88160814Ssimon $TABLE="K256"; 89194206Ssimon $func="sha256_block_data_order"; 90160814Ssimon @Sigma0=( 2,13,22); 91160814Ssimon @Sigma1=( 6,11,25); 92160814Ssimon @sigma0=( 7,18, 3); 93160814Ssimon @sigma1=(17,19,10); 94160814Ssimon $rounds=64; 95160814Ssimon} else { die "nonsense $output"; } 96160814Ssimon 97160814Ssimonopen STDOUT,">$output" || die "can't open $output: $!"; 98160814Ssimon 99160814Ssimonif ($^O eq "hpux") { 100160814Ssimon $ADDP="addp4"; 101160814Ssimon for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); } 102160814Ssimon} else { $ADDP="add"; } 103160814Ssimonfor (@ARGV) { $big_endian=1 if (/\-DB_ENDIAN/); 104160814Ssimon $big_endian=0 if (/\-DL_ENDIAN/); } 105160814Ssimonif (!defined($big_endian)) 106160814Ssimon { $big_endian=(unpack('L',pack('N',1))==1); } 107160814Ssimon 108160814Ssimon$code=<<___; 109194206Ssimon.ident \"$output, version 1.1\" 110160814Ssimon.ident \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\" 111160814Ssimon.explicit 112160814Ssimon.text 113160814Ssimon 114194206Ssimonpfssave=r2; 115194206Ssimonlcsave=r3; 116160814Ssimonprsave=r14; 117160814SsimonK=r15; 118160814SsimonA=r16; B=r17; C=r18; D=r19; 119160814SsimonE=r20; F=r21; G=r22; H=r23; 120160814SsimonT1=r24; T2=r25; 121160814Ssimons0=r26; s1=r27; t0=r28; t1=r29; 122160814SsimonKtbl=r30; 123160814Ssimonctx=r31; // 1st arg 124160814Ssimoninput=r48; // 2nd arg 125160814Ssimonnum=r49; // 3rd arg 126160814Ssimonsgm0=r50; sgm1=r51; // small constants 127194206SsimonA_=r54; B_=r55; C_=r56; D_=r57; 128194206SsimonE_=r58; F_=r59; G_=r60; H_=r61; 129160814Ssimon 130160814Ssimon// void $func (SHA_CTX *ctx, const void *in,size_t num[,int host]) 131160814Ssimon.global $func# 132160814Ssimon.proc $func# 133160814Ssimon.align 32 134160814Ssimon$func: 135160814Ssimon .prologue 136194206Ssimon .save ar.pfs,pfssave 137194206Ssimon{ .mmi; alloc pfssave=ar.pfs,3,27,0,16 138160814Ssimon $ADDP ctx=0,r32 // 1st arg 139194206Ssimon .save ar.lc,lcsave 140194206Ssimon mov lcsave=ar.lc } 141160814Ssimon{ .mmi; $ADDP input=0,r33 // 2nd arg 142194206Ssimon mov num=r34 // 3rd arg 143194206Ssimon .save pr,prsave 144160814Ssimon mov prsave=pr };; 145160814Ssimon 146160814Ssimon .body 147160814Ssimon{ .mib; add r8=0*$SZ,ctx 148160814Ssimon add r9=1*$SZ,ctx 149194206Ssimon brp.loop.imp .L_first16,.L_first16_end-16 } 150160814Ssimon{ .mib; add r10=2*$SZ,ctx 151160814Ssimon add r11=3*$SZ,ctx 152194206Ssimon brp.loop.imp .L_rest,.L_rest_end-16 };; 153194206Ssimon 154160814Ssimon// load A-H 155194206Ssimon.Lpic_point: 156194206Ssimon{ .mmi; $LDW A_=[r8],4*$SZ 157194206Ssimon $LDW B_=[r9],4*$SZ 158194206Ssimon mov Ktbl=ip } 159194206Ssimon{ .mmi; $LDW C_=[r10],4*$SZ 160194206Ssimon $LDW D_=[r11],4*$SZ 161194206Ssimon mov sgm0=$sigma0[2] };; 162194206Ssimon{ .mmi; $LDW E_=[r8] 163194206Ssimon $LDW F_=[r9] 164194206Ssimon add Ktbl=($TABLE#-.Lpic_point),Ktbl } 165194206Ssimon{ .mmi; $LDW G_=[r10] 166194206Ssimon $LDW H_=[r11] 167194206Ssimon cmp.ne p0,p16=0,r0 };; // used in sha256_block 168194206Ssimon___ 169194206Ssimon$code.=<<___ if ($BITS==64); 170194206Ssimon{ .mii; and r8=7,input 171194206Ssimon and input=~7,input;; 172194206Ssimon cmp.eq p9,p0=1,r8 } 173194206Ssimon{ .mmi; cmp.eq p10,p0=2,r8 174194206Ssimon cmp.eq p11,p0=3,r8 175194206Ssimon cmp.eq p12,p0=4,r8 } 176194206Ssimon{ .mmi; cmp.eq p13,p0=5,r8 177194206Ssimon cmp.eq p14,p0=6,r8 178194206Ssimon cmp.eq p15,p0=7,r8 };; 179194206Ssimon___ 180194206Ssimon$code.=<<___; 181194206Ssimon.L_outer: 182194206Ssimon.rotr X[16] 183194206Ssimon{ .mmi; mov A=A_ 184194206Ssimon mov B=B_ 185194206Ssimon mov ar.lc=14 } 186194206Ssimon{ .mmi; mov C=C_ 187194206Ssimon mov D=D_ 188194206Ssimon mov E=E_ } 189194206Ssimon{ .mmi; mov F=F_ 190194206Ssimon mov G=G_ 191194206Ssimon mov ar.ec=2 } 192194206Ssimon{ .mmi; ld1 X[15]=[input],$SZ // eliminated in 64-bit 193194206Ssimon mov H=H_ 194160814Ssimon mov sgm1=$sigma1[2] };; 195160814Ssimon 196194206Ssimon___ 197194206Ssimon$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32); 198160814Ssimon.align 32 199160814Ssimon.L_first16: 200194206Ssimon{ .mmi; add r9=1-$SZ,input 201194206Ssimon add r10=2-$SZ,input 202194206Ssimon add r11=3-$SZ,input };; 203194206Ssimon{ .mmi; ld1 r9=[r9] 204194206Ssimon ld1 r10=[r10] 205160814Ssimon dep.z $t1=E,32,32 } 206194206Ssimon{ .mmi; $LDW K=[Ktbl],$SZ 207194206Ssimon ld1 r11=[r11] 208160814Ssimon zxt4 E=E };; 209194206Ssimon{ .mii; or $t1=$t1,E 210194206Ssimon dep X[15]=X[15],r9,8,8 211194206Ssimon dep r11=r10,r11,8,8 };; 212194206Ssimon{ .mmi; and T1=F,E 213194206Ssimon and T2=A,B 214194206Ssimon dep X[15]=X[15],r11,16,16 } 215160814Ssimon{ .mmi; andcm r8=G,E 216160814Ssimon and r9=A,C 217160814Ssimon mux2 $t0=A,0x44 };; // copy lower half to upper 218194206Ssimon{ .mmi; (p16) ld1 X[15-1]=[input],$SZ // prefetch 219194206Ssimon xor T1=T1,r8 // T1=((e & f) ^ (~e & g)) 220160814Ssimon _rotr r11=$t1,$Sigma1[0] } // ROTR(e,14) 221160814Ssimon{ .mib; and r10=B,C 222160814Ssimon xor T2=T2,r9 };; 223160814Ssimon___ 224160814Ssimon$t0="A", $t1="E", $code.=<<___ if ($BITS==64); 225194206Ssimon// in 64-bit mode I load whole X[16] at once and take care of alignment... 226194206Ssimon{ .mmi; add r8=1*$SZ,input 227194206Ssimon add r9=2*$SZ,input 228194206Ssimon add r10=3*$SZ,input };; 229194206Ssimon{ .mmb; $LDW X[15]=[input],4*$SZ 230194206Ssimon $LDW X[14]=[r8],4*$SZ 231194206Ssimon(p9) br.cond.dpnt.many .L1byte };; 232194206Ssimon{ .mmb; $LDW X[13]=[r9],4*$SZ 233194206Ssimon $LDW X[12]=[r10],4*$SZ 234194206Ssimon(p10) br.cond.dpnt.many .L2byte };; 235194206Ssimon{ .mmb; $LDW X[11]=[input],4*$SZ 236194206Ssimon $LDW X[10]=[r8],4*$SZ 237194206Ssimon(p11) br.cond.dpnt.many .L3byte };; 238194206Ssimon{ .mmb; $LDW X[ 9]=[r9],4*$SZ 239194206Ssimon $LDW X[ 8]=[r10],4*$SZ 240194206Ssimon(p12) br.cond.dpnt.many .L4byte };; 241194206Ssimon{ .mmb; $LDW X[ 7]=[input],4*$SZ 242194206Ssimon $LDW X[ 6]=[r8],4*$SZ 243194206Ssimon(p13) br.cond.dpnt.many .L5byte };; 244194206Ssimon{ .mmb; $LDW X[ 5]=[r9],4*$SZ 245194206Ssimon $LDW X[ 4]=[r10],4*$SZ 246194206Ssimon(p14) br.cond.dpnt.many .L6byte };; 247194206Ssimon{ .mmb; $LDW X[ 3]=[input],4*$SZ 248194206Ssimon $LDW X[ 2]=[r8],4*$SZ 249194206Ssimon(p15) br.cond.dpnt.many .L7byte };; 250194206Ssimon{ .mmb; $LDW X[ 1]=[r9],4*$SZ 251194206Ssimon $LDW X[ 0]=[r10],4*$SZ 252194206Ssimon br.many .L_first16 };; 253194206Ssimon.L1byte: 254194206Ssimon{ .mmi; $LDW X[13]=[r9],4*$SZ 255194206Ssimon $LDW X[12]=[r10],4*$SZ 256194206Ssimon shrp X[15]=X[15],X[14],56 };; 257194206Ssimon{ .mmi; $LDW X[11]=[input],4*$SZ 258194206Ssimon $LDW X[10]=[r8],4*$SZ 259194206Ssimon shrp X[14]=X[14],X[13],56 } 260194206Ssimon{ .mmi; $LDW X[ 9]=[r9],4*$SZ 261194206Ssimon $LDW X[ 8]=[r10],4*$SZ 262194206Ssimon shrp X[13]=X[13],X[12],56 };; 263194206Ssimon{ .mmi; $LDW X[ 7]=[input],4*$SZ 264194206Ssimon $LDW X[ 6]=[r8],4*$SZ 265194206Ssimon shrp X[12]=X[12],X[11],56 } 266194206Ssimon{ .mmi; $LDW X[ 5]=[r9],4*$SZ 267194206Ssimon $LDW X[ 4]=[r10],4*$SZ 268194206Ssimon shrp X[11]=X[11],X[10],56 };; 269194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 270194206Ssimon $LDW X[ 2]=[r8],4*$SZ 271194206Ssimon shrp X[10]=X[10],X[ 9],56 } 272194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 273194206Ssimon $LDW X[ 0]=[r10],4*$SZ 274194206Ssimon shrp X[ 9]=X[ 9],X[ 8],56 };; 275194206Ssimon{ .mii; $LDW T1=[input] 276194206Ssimon shrp X[ 8]=X[ 8],X[ 7],56 277194206Ssimon shrp X[ 7]=X[ 7],X[ 6],56 } 278194206Ssimon{ .mii; shrp X[ 6]=X[ 6],X[ 5],56 279194206Ssimon shrp X[ 5]=X[ 5],X[ 4],56 };; 280194206Ssimon{ .mii; shrp X[ 4]=X[ 4],X[ 3],56 281194206Ssimon shrp X[ 3]=X[ 3],X[ 2],56 } 282194206Ssimon{ .mii; shrp X[ 2]=X[ 2],X[ 1],56 283194206Ssimon shrp X[ 1]=X[ 1],X[ 0],56 } 284194206Ssimon{ .mib; shrp X[ 0]=X[ 0],T1,56 285194206Ssimon br.many .L_first16 };; 286194206Ssimon.L2byte: 287194206Ssimon{ .mmi; $LDW X[11]=[input],4*$SZ 288194206Ssimon $LDW X[10]=[r8],4*$SZ 289194206Ssimon shrp X[15]=X[15],X[14],48 } 290194206Ssimon{ .mmi; $LDW X[ 9]=[r9],4*$SZ 291194206Ssimon $LDW X[ 8]=[r10],4*$SZ 292194206Ssimon shrp X[14]=X[14],X[13],48 };; 293194206Ssimon{ .mmi; $LDW X[ 7]=[input],4*$SZ 294194206Ssimon $LDW X[ 6]=[r8],4*$SZ 295194206Ssimon shrp X[13]=X[13],X[12],48 } 296194206Ssimon{ .mmi; $LDW X[ 5]=[r9],4*$SZ 297194206Ssimon $LDW X[ 4]=[r10],4*$SZ 298194206Ssimon shrp X[12]=X[12],X[11],48 };; 299194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 300194206Ssimon $LDW X[ 2]=[r8],4*$SZ 301194206Ssimon shrp X[11]=X[11],X[10],48 } 302194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 303194206Ssimon $LDW X[ 0]=[r10],4*$SZ 304194206Ssimon shrp X[10]=X[10],X[ 9],48 };; 305194206Ssimon{ .mii; $LDW T1=[input] 306194206Ssimon shrp X[ 9]=X[ 9],X[ 8],48 307194206Ssimon shrp X[ 8]=X[ 8],X[ 7],48 } 308194206Ssimon{ .mii; shrp X[ 7]=X[ 7],X[ 6],48 309194206Ssimon shrp X[ 6]=X[ 6],X[ 5],48 };; 310194206Ssimon{ .mii; shrp X[ 5]=X[ 5],X[ 4],48 311194206Ssimon shrp X[ 4]=X[ 4],X[ 3],48 } 312194206Ssimon{ .mii; shrp X[ 3]=X[ 3],X[ 2],48 313194206Ssimon shrp X[ 2]=X[ 2],X[ 1],48 } 314194206Ssimon{ .mii; shrp X[ 1]=X[ 1],X[ 0],48 315194206Ssimon shrp X[ 0]=X[ 0],T1,48 } 316194206Ssimon{ .mfb; br.many .L_first16 };; 317194206Ssimon.L3byte: 318194206Ssimon{ .mmi; $LDW X[ 9]=[r9],4*$SZ 319194206Ssimon $LDW X[ 8]=[r10],4*$SZ 320194206Ssimon shrp X[15]=X[15],X[14],40 };; 321194206Ssimon{ .mmi; $LDW X[ 7]=[input],4*$SZ 322194206Ssimon $LDW X[ 6]=[r8],4*$SZ 323194206Ssimon shrp X[14]=X[14],X[13],40 } 324194206Ssimon{ .mmi; $LDW X[ 5]=[r9],4*$SZ 325194206Ssimon $LDW X[ 4]=[r10],4*$SZ 326194206Ssimon shrp X[13]=X[13],X[12],40 };; 327194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 328194206Ssimon $LDW X[ 2]=[r8],4*$SZ 329194206Ssimon shrp X[12]=X[12],X[11],40 } 330194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 331194206Ssimon $LDW X[ 0]=[r10],4*$SZ 332194206Ssimon shrp X[11]=X[11],X[10],40 };; 333194206Ssimon{ .mii; $LDW T1=[input] 334194206Ssimon shrp X[10]=X[10],X[ 9],40 335194206Ssimon shrp X[ 9]=X[ 9],X[ 8],40 } 336194206Ssimon{ .mii; shrp X[ 8]=X[ 8],X[ 7],40 337194206Ssimon shrp X[ 7]=X[ 7],X[ 6],40 };; 338194206Ssimon{ .mii; shrp X[ 6]=X[ 6],X[ 5],40 339194206Ssimon shrp X[ 5]=X[ 5],X[ 4],40 } 340194206Ssimon{ .mii; shrp X[ 4]=X[ 4],X[ 3],40 341194206Ssimon shrp X[ 3]=X[ 3],X[ 2],40 } 342194206Ssimon{ .mii; shrp X[ 2]=X[ 2],X[ 1],40 343194206Ssimon shrp X[ 1]=X[ 1],X[ 0],40 } 344194206Ssimon{ .mib; shrp X[ 0]=X[ 0],T1,40 345194206Ssimon br.many .L_first16 };; 346194206Ssimon.L4byte: 347194206Ssimon{ .mmi; $LDW X[ 7]=[input],4*$SZ 348194206Ssimon $LDW X[ 6]=[r8],4*$SZ 349194206Ssimon shrp X[15]=X[15],X[14],32 } 350194206Ssimon{ .mmi; $LDW X[ 5]=[r9],4*$SZ 351194206Ssimon $LDW X[ 4]=[r10],4*$SZ 352194206Ssimon shrp X[14]=X[14],X[13],32 };; 353194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 354194206Ssimon $LDW X[ 2]=[r8],4*$SZ 355194206Ssimon shrp X[13]=X[13],X[12],32 } 356194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 357194206Ssimon $LDW X[ 0]=[r10],4*$SZ 358194206Ssimon shrp X[12]=X[12],X[11],32 };; 359194206Ssimon{ .mii; $LDW T1=[input] 360194206Ssimon shrp X[11]=X[11],X[10],32 361194206Ssimon shrp X[10]=X[10],X[ 9],32 } 362194206Ssimon{ .mii; shrp X[ 9]=X[ 9],X[ 8],32 363194206Ssimon shrp X[ 8]=X[ 8],X[ 7],32 };; 364194206Ssimon{ .mii; shrp X[ 7]=X[ 7],X[ 6],32 365194206Ssimon shrp X[ 6]=X[ 6],X[ 5],32 } 366194206Ssimon{ .mii; shrp X[ 5]=X[ 5],X[ 4],32 367194206Ssimon shrp X[ 4]=X[ 4],X[ 3],32 } 368194206Ssimon{ .mii; shrp X[ 3]=X[ 3],X[ 2],32 369194206Ssimon shrp X[ 2]=X[ 2],X[ 1],32 } 370194206Ssimon{ .mii; shrp X[ 1]=X[ 1],X[ 0],32 371194206Ssimon shrp X[ 0]=X[ 0],T1,32 } 372194206Ssimon{ .mfb; br.many .L_first16 };; 373194206Ssimon.L5byte: 374194206Ssimon{ .mmi; $LDW X[ 5]=[r9],4*$SZ 375194206Ssimon $LDW X[ 4]=[r10],4*$SZ 376194206Ssimon shrp X[15]=X[15],X[14],24 };; 377194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 378194206Ssimon $LDW X[ 2]=[r8],4*$SZ 379194206Ssimon shrp X[14]=X[14],X[13],24 } 380194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 381194206Ssimon $LDW X[ 0]=[r10],4*$SZ 382194206Ssimon shrp X[13]=X[13],X[12],24 };; 383194206Ssimon{ .mii; $LDW T1=[input] 384194206Ssimon shrp X[12]=X[12],X[11],24 385194206Ssimon shrp X[11]=X[11],X[10],24 } 386194206Ssimon{ .mii; shrp X[10]=X[10],X[ 9],24 387194206Ssimon shrp X[ 9]=X[ 9],X[ 8],24 };; 388194206Ssimon{ .mii; shrp X[ 8]=X[ 8],X[ 7],24 389194206Ssimon shrp X[ 7]=X[ 7],X[ 6],24 } 390194206Ssimon{ .mii; shrp X[ 6]=X[ 6],X[ 5],24 391194206Ssimon shrp X[ 5]=X[ 5],X[ 4],24 } 392194206Ssimon{ .mii; shrp X[ 4]=X[ 4],X[ 3],24 393194206Ssimon shrp X[ 3]=X[ 3],X[ 2],24 } 394194206Ssimon{ .mii; shrp X[ 2]=X[ 2],X[ 1],24 395194206Ssimon shrp X[ 1]=X[ 1],X[ 0],24 } 396194206Ssimon{ .mib; shrp X[ 0]=X[ 0],T1,24 397194206Ssimon br.many .L_first16 };; 398194206Ssimon.L6byte: 399194206Ssimon{ .mmi; $LDW X[ 3]=[input],4*$SZ 400194206Ssimon $LDW X[ 2]=[r8],4*$SZ 401194206Ssimon shrp X[15]=X[15],X[14],16 } 402194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 403194206Ssimon $LDW X[ 0]=[r10],4*$SZ 404194206Ssimon shrp X[14]=X[14],X[13],16 };; 405194206Ssimon{ .mii; $LDW T1=[input] 406194206Ssimon shrp X[13]=X[13],X[12],16 407194206Ssimon shrp X[12]=X[12],X[11],16 } 408194206Ssimon{ .mii; shrp X[11]=X[11],X[10],16 409194206Ssimon shrp X[10]=X[10],X[ 9],16 };; 410194206Ssimon{ .mii; shrp X[ 9]=X[ 9],X[ 8],16 411194206Ssimon shrp X[ 8]=X[ 8],X[ 7],16 } 412194206Ssimon{ .mii; shrp X[ 7]=X[ 7],X[ 6],16 413194206Ssimon shrp X[ 6]=X[ 6],X[ 5],16 } 414194206Ssimon{ .mii; shrp X[ 5]=X[ 5],X[ 4],16 415194206Ssimon shrp X[ 4]=X[ 4],X[ 3],16 } 416194206Ssimon{ .mii; shrp X[ 3]=X[ 3],X[ 2],16 417194206Ssimon shrp X[ 2]=X[ 2],X[ 1],16 } 418194206Ssimon{ .mii; shrp X[ 1]=X[ 1],X[ 0],16 419194206Ssimon shrp X[ 0]=X[ 0],T1,16 } 420194206Ssimon{ .mfb; br.many .L_first16 };; 421194206Ssimon.L7byte: 422194206Ssimon{ .mmi; $LDW X[ 1]=[r9],4*$SZ 423194206Ssimon $LDW X[ 0]=[r10],4*$SZ 424194206Ssimon shrp X[15]=X[15],X[14],8 };; 425194206Ssimon{ .mii; $LDW T1=[input] 426194206Ssimon shrp X[14]=X[14],X[13],8 427194206Ssimon shrp X[13]=X[13],X[12],8 } 428194206Ssimon{ .mii; shrp X[12]=X[12],X[11],8 429194206Ssimon shrp X[11]=X[11],X[10],8 };; 430194206Ssimon{ .mii; shrp X[10]=X[10],X[ 9],8 431194206Ssimon shrp X[ 9]=X[ 9],X[ 8],8 } 432194206Ssimon{ .mii; shrp X[ 8]=X[ 8],X[ 7],8 433194206Ssimon shrp X[ 7]=X[ 7],X[ 6],8 } 434194206Ssimon{ .mii; shrp X[ 6]=X[ 6],X[ 5],8 435194206Ssimon shrp X[ 5]=X[ 5],X[ 4],8 } 436194206Ssimon{ .mii; shrp X[ 4]=X[ 4],X[ 3],8 437194206Ssimon shrp X[ 3]=X[ 3],X[ 2],8 } 438194206Ssimon{ .mii; shrp X[ 2]=X[ 2],X[ 1],8 439194206Ssimon shrp X[ 1]=X[ 1],X[ 0],8 } 440194206Ssimon{ .mib; shrp X[ 0]=X[ 0],T1,8 441194206Ssimon br.many .L_first16 };; 442194206Ssimon 443194206Ssimon.align 32 444194206Ssimon.L_first16: 445194206Ssimon{ .mmi; $LDW K=[Ktbl],$SZ 446160814Ssimon and T1=F,E 447160814Ssimon and T2=A,B } 448194206Ssimon{ .mmi; //$LDW X[15]=[input],$SZ // X[i]=*input++ 449160814Ssimon andcm r8=G,E 450160814Ssimon and r9=A,C };; 451160814Ssimon{ .mmi; xor T1=T1,r8 //T1=((e & f) ^ (~e & g)) 452160814Ssimon and r10=B,C 453160814Ssimon _rotr r11=$t1,$Sigma1[0] } // ROTR(e,14) 454160814Ssimon{ .mmi; xor T2=T2,r9 455160814Ssimon mux1 X[15]=X[15],\@rev };; // eliminated in big-endian 456160814Ssimon___ 457160814Ssimon$code.=<<___; 458160814Ssimon{ .mib; add T1=T1,H // T1=Ch(e,f,g)+h 459160814Ssimon _rotr r8=$t1,$Sigma1[1] } // ROTR(e,18) 460160814Ssimon{ .mib; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c)) 461160814Ssimon mov H=G };; 462160814Ssimon{ .mib; xor r11=r8,r11 463160814Ssimon _rotr r9=$t1,$Sigma1[2] } // ROTR(e,41) 464160814Ssimon{ .mib; mov G=F 465160814Ssimon mov F=E };; 466160814Ssimon{ .mib; xor r9=r9,r11 // r9=Sigma1(e) 467160814Ssimon _rotr r10=$t0,$Sigma0[0] } // ROTR(a,28) 468160814Ssimon{ .mib; add T1=T1,K // T1=Ch(e,f,g)+h+K512[i] 469160814Ssimon mov E=D };; 470160814Ssimon{ .mib; add T1=T1,r9 // T1+=Sigma1(e) 471160814Ssimon _rotr r11=$t0,$Sigma0[1] } // ROTR(a,34) 472160814Ssimon{ .mib; mov D=C 473160814Ssimon mov C=B };; 474160814Ssimon{ .mib; add T1=T1,X[15] // T1+=X[i] 475160814Ssimon _rotr r8=$t0,$Sigma0[2] } // ROTR(a,39) 476160814Ssimon{ .mib; xor r10=r10,r11 477160814Ssimon mux2 X[15]=X[15],0x44 };; // eliminated in 64-bit 478160814Ssimon{ .mmi; xor r10=r8,r10 // r10=Sigma0(a) 479160814Ssimon mov B=A 480160814Ssimon add A=T1,T2 };; 481160814Ssimon{ .mib; add E=E,T1 482160814Ssimon add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a) 483160814Ssimon br.ctop.sptk .L_first16 };; 484194206Ssimon.L_first16_end: 485160814Ssimon 486194206Ssimon{ .mii; mov ar.lc=$rounds-17 487194206Ssimon mov ar.ec=1 };; 488194206Ssimon 489160814Ssimon.align 32 490160814Ssimon.L_rest: 491160814Ssimon.rotr X[16] 492160814Ssimon{ .mib; $LDW K=[Ktbl],$SZ 493160814Ssimon _rotr r8=X[15-1],$sigma0[0] } // ROTR(s0,1) 494160814Ssimon{ .mib; $ADD X[15]=X[15],X[15-9] // X[i&0xF]+=X[(i+9)&0xF] 495160814Ssimon $SHRU s0=X[15-1],sgm0 };; // s0=X[(i+1)&0xF]>>7 496160814Ssimon{ .mib; and T1=F,E 497160814Ssimon _rotr r9=X[15-1],$sigma0[1] } // ROTR(s0,8) 498160814Ssimon{ .mib; andcm r10=G,E 499160814Ssimon $SHRU s1=X[15-14],sgm1 };; // s1=X[(i+14)&0xF]>>6 500160814Ssimon{ .mmi; xor T1=T1,r10 // T1=((e & f) ^ (~e & g)) 501160814Ssimon xor r9=r8,r9 502160814Ssimon _rotr r10=X[15-14],$sigma1[0] };;// ROTR(s1,19) 503160814Ssimon{ .mib; and T2=A,B 504160814Ssimon _rotr r11=X[15-14],$sigma1[1] }// ROTR(s1,61) 505160814Ssimon{ .mib; and r8=A,C };; 506160814Ssimon___ 507160814Ssimon$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32); 508160814Ssimon// I adhere to mmi; in order to hold Itanium 1 back and avoid 6 cycle 509160814Ssimon// pipeline flush in last bundle. Note that even on Itanium2 the 510160814Ssimon// latter stalls for one clock cycle... 511160814Ssimon{ .mmi; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF]) 512160814Ssimon dep.z $t1=E,32,32 } 513160814Ssimon{ .mmi; xor r10=r11,r10 514160814Ssimon zxt4 E=E };; 515160814Ssimon{ .mmi; or $t1=$t1,E 516160814Ssimon xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF]) 517160814Ssimon mux2 $t0=A,0x44 };; // copy lower half to upper 518160814Ssimon{ .mmi; xor T2=T2,r8 519160814Ssimon _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14) 520160814Ssimon{ .mmi; and r10=B,C 521160814Ssimon add T1=T1,H // T1=Ch(e,f,g)+h 522160814Ssimon $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF]) 523160814Ssimon___ 524160814Ssimon$t0="A", $t1="E", $code.=<<___ if ($BITS==64); 525160814Ssimon{ .mib; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF]) 526160814Ssimon _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14) 527160814Ssimon{ .mib; xor r10=r11,r10 528160814Ssimon xor T2=T2,r8 };; 529160814Ssimon{ .mib; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF]) 530160814Ssimon add T1=T1,H } 531160814Ssimon{ .mib; and r10=B,C 532160814Ssimon $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF]) 533160814Ssimon___ 534160814Ssimon$code.=<<___; 535160814Ssimon{ .mmi; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c)) 536160814Ssimon mov H=G 537160814Ssimon _rotr r8=$t1,$Sigma1[1] };; // ROTR(e,18) 538160814Ssimon{ .mmi; xor r11=r8,r9 539160814Ssimon $ADD X[15]=X[15],s1 // X[i&0xF]+=sigma1(X[(i+14)&0xF]) 540160814Ssimon _rotr r9=$t1,$Sigma1[2] } // ROTR(e,41) 541160814Ssimon{ .mmi; mov G=F 542160814Ssimon mov F=E };; 543160814Ssimon{ .mib; xor r9=r9,r11 // r9=Sigma1(e) 544160814Ssimon _rotr r10=$t0,$Sigma0[0] } // ROTR(a,28) 545160814Ssimon{ .mib; add T1=T1,K // T1=Ch(e,f,g)+h+K512[i] 546160814Ssimon mov E=D };; 547160814Ssimon{ .mib; add T1=T1,r9 // T1+=Sigma1(e) 548160814Ssimon _rotr r11=$t0,$Sigma0[1] } // ROTR(a,34) 549160814Ssimon{ .mib; mov D=C 550160814Ssimon mov C=B };; 551160814Ssimon{ .mmi; add T1=T1,X[15] // T1+=X[i] 552160814Ssimon xor r10=r10,r11 553160814Ssimon _rotr r8=$t0,$Sigma0[2] };; // ROTR(a,39) 554160814Ssimon{ .mmi; xor r10=r8,r10 // r10=Sigma0(a) 555160814Ssimon mov B=A 556160814Ssimon add A=T1,T2 };; 557160814Ssimon{ .mib; add E=E,T1 558160814Ssimon add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a) 559160814Ssimon br.ctop.sptk .L_rest };; 560194206Ssimon.L_rest_end: 561160814Ssimon 562194206Ssimon{ .mmi; add A_=A_,A 563194206Ssimon add B_=B_,B 564194206Ssimon add C_=C_,C } 565194206Ssimon{ .mmi; add D_=D_,D 566194206Ssimon add E_=E_,E 567194206Ssimon cmp.ltu p16,p0=1,num };; 568194206Ssimon{ .mmi; add F_=F_,F 569194206Ssimon add G_=G_,G 570194206Ssimon add H_=H_,H } 571194206Ssimon{ .mmb; add Ktbl=-$SZ*$rounds,Ktbl 572194206Ssimon(p16) add num=-1,num 573194206Ssimon(p16) br.dptk.many .L_outer };; 574194206Ssimon 575160814Ssimon{ .mib; add r8=0*$SZ,ctx 576160814Ssimon add r9=1*$SZ,ctx } 577160814Ssimon{ .mib; add r10=2*$SZ,ctx 578160814Ssimon add r11=3*$SZ,ctx };; 579194206Ssimon{ .mmi; $STW [r8]=A_,4*$SZ 580194206Ssimon $STW [r9]=B_,4*$SZ 581194206Ssimon mov ar.lc=lcsave } 582194206Ssimon{ .mmi; $STW [r10]=C_,4*$SZ 583194206Ssimon $STW [r11]=D_,4*$SZ 584194206Ssimon mov pr=prsave,0x1ffff };; 585194206Ssimon{ .mmb; $STW [r8]=E_ 586194206Ssimon $STW [r9]=F_ } 587194206Ssimon{ .mmb; $STW [r10]=G_ 588194206Ssimon $STW [r11]=H_ 589160814Ssimon br.ret.sptk.many b0 };; 590160814Ssimon.endp $func# 591160814Ssimon___ 592160814Ssimon 593160814Ssimon$code =~ s/\`([^\`]*)\`/eval $1/gem; 594160814Ssimon$code =~ s/_rotr(\s+)([^=]+)=([^,]+),([0-9]+)/shrp$1$2=$3,$3,$4/gm; 595160814Ssimonif ($BITS==64) { 596160814Ssimon $code =~ s/mux2(\s+)\S+/nop.i$1 0x0/gm; 597194206Ssimon $code =~ s/mux1(\s+)\S+/nop.i$1 0x0/gm if ($big_endian); 598194206Ssimon $code =~ s/(shrp\s+X\[[^=]+)=([^,]+),([^,]+),([1-9]+)/$1=$3,$2,64-$4/gm 599194206Ssimon if (!$big_endian); 600194206Ssimon $code =~ s/ld1(\s+)X\[\S+/nop.m$1 0x0/gm; 601160814Ssimon} 602160814Ssimon 603160814Ssimonprint $code; 604160814Ssimon 605160814Ssimonprint<<___ if ($BITS==32); 606160814Ssimon.align 64 607160814Ssimon.type K256#,\@object 608160814SsimonK256: data4 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5 609160814Ssimon data4 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5 610160814Ssimon data4 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3 611160814Ssimon data4 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174 612160814Ssimon data4 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc 613160814Ssimon data4 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da 614160814Ssimon data4 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7 615160814Ssimon data4 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967 616160814Ssimon data4 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13 617160814Ssimon data4 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85 618160814Ssimon data4 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3 619160814Ssimon data4 0xd192e819,0xd6990624,0xf40e3585,0x106aa070 620160814Ssimon data4 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5 621160814Ssimon data4 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3 622160814Ssimon data4 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208 623160814Ssimon data4 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2 624160814Ssimon.size K256#,$SZ*$rounds 625194206Ssimonstringz "SHA256 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>" 626160814Ssimon___ 627160814Ssimonprint<<___ if ($BITS==64); 628160814Ssimon.align 64 629160814Ssimon.type K512#,\@object 630160814SsimonK512: data8 0x428a2f98d728ae22,0x7137449123ef65cd 631160814Ssimon data8 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc 632160814Ssimon data8 0x3956c25bf348b538,0x59f111f1b605d019 633160814Ssimon data8 0x923f82a4af194f9b,0xab1c5ed5da6d8118 634160814Ssimon data8 0xd807aa98a3030242,0x12835b0145706fbe 635160814Ssimon data8 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2 636160814Ssimon data8 0x72be5d74f27b896f,0x80deb1fe3b1696b1 637160814Ssimon data8 0x9bdc06a725c71235,0xc19bf174cf692694 638160814Ssimon data8 0xe49b69c19ef14ad2,0xefbe4786384f25e3 639160814Ssimon data8 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65 640160814Ssimon data8 0x2de92c6f592b0275,0x4a7484aa6ea6e483 641160814Ssimon data8 0x5cb0a9dcbd41fbd4,0x76f988da831153b5 642160814Ssimon data8 0x983e5152ee66dfab,0xa831c66d2db43210 643160814Ssimon data8 0xb00327c898fb213f,0xbf597fc7beef0ee4 644160814Ssimon data8 0xc6e00bf33da88fc2,0xd5a79147930aa725 645160814Ssimon data8 0x06ca6351e003826f,0x142929670a0e6e70 646160814Ssimon data8 0x27b70a8546d22ffc,0x2e1b21385c26c926 647160814Ssimon data8 0x4d2c6dfc5ac42aed,0x53380d139d95b3df 648160814Ssimon data8 0x650a73548baf63de,0x766a0abb3c77b2a8 649160814Ssimon data8 0x81c2c92e47edaee6,0x92722c851482353b 650160814Ssimon data8 0xa2bfe8a14cf10364,0xa81a664bbc423001 651160814Ssimon data8 0xc24b8b70d0f89791,0xc76c51a30654be30 652160814Ssimon data8 0xd192e819d6ef5218,0xd69906245565a910 653160814Ssimon data8 0xf40e35855771202a,0x106aa07032bbd1b8 654160814Ssimon data8 0x19a4c116b8d2d0c8,0x1e376c085141ab53 655160814Ssimon data8 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8 656160814Ssimon data8 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb 657160814Ssimon data8 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3 658160814Ssimon data8 0x748f82ee5defb2fc,0x78a5636f43172f60 659160814Ssimon data8 0x84c87814a1f0ab72,0x8cc702081a6439ec 660160814Ssimon data8 0x90befffa23631e28,0xa4506cebde82bde9 661160814Ssimon data8 0xbef9a3f7b2c67915,0xc67178f2e372532b 662160814Ssimon data8 0xca273eceea26619c,0xd186b8c721c0c207 663160814Ssimon data8 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178 664160814Ssimon data8 0x06f067aa72176fba,0x0a637dc5a2c898a6 665160814Ssimon data8 0x113f9804bef90dae,0x1b710b35131c471b 666160814Ssimon data8 0x28db77f523047d84,0x32caab7b40c72493 667160814Ssimon data8 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c 668160814Ssimon data8 0x4cc5d4becb3e42b6,0x597f299cfc657e2a 669160814Ssimon data8 0x5fcb6fab3ad6faec,0x6c44198c4a475817 670160814Ssimon.size K512#,$SZ*$rounds 671194206Ssimonstringz "SHA512 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>" 672160814Ssimon___ 673