1#! /usr/bin/env perl 2# Copyright 2004-2016 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# 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# Eternal question is what's wrong with compiler generated code? The 18# trick is that it's possible to reduce the number of shifts required 19# to perform rotations by maintaining copy of 32-bit value in upper 20# bits of 64-bit register. Just follow mux2 and shrp instructions... 21# Performance under big-endian OS such as HP-UX is 179MBps*1GHz, which 22# is >50% better than HP C and >2x better than gcc. 23 24# $output is the last argument if it looks like a file (it has an extension) 25$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef; 26 27$code=<<___; 28.ident \"sha1-ia64.s, version 1.3\" 29.ident \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\" 30.explicit 31 32___ 33 34 35if ($^O eq "hpux") { 36 $ADDP="addp4"; 37 for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); } 38} else { $ADDP="add"; } 39 40#$human=1; 41if ($human) { # useful for visual code auditing... 42 ($A,$B,$C,$D,$E) = ("A","B","C","D","E"); 43 ($h0,$h1,$h2,$h3,$h4) = ("h0","h1","h2","h3","h4"); 44 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) = 45 ( "K_00_19","K_20_39","K_40_59","K_60_79" ); 46 @X= ( "X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7", 47 "X8", "X9","X10","X11","X12","X13","X14","X15" ); 48} 49else { 50 ($A,$B,$C,$D,$E) = ("loc0","loc1","loc2","loc3","loc4"); 51 ($h0,$h1,$h2,$h3,$h4) = ("loc5","loc6","loc7","loc8","loc9"); 52 ($K_00_19, $K_20_39, $K_40_59, $K_60_79) = 53 ( "r14", "r15", "loc10", "loc11" ); 54 @X= ( "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", 55 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31" ); 56} 57 58sub BODY_00_15 { 59local *code=shift; 60my ($i,$a,$b,$c,$d,$e)=@_; 61my $j=$i+1; 62my $Xn=@X[$j%16]; 63 64$code.=<<___ if ($i==0); 65{ .mmi; ld1 $X[$i]=[inp],2 // MSB 66 ld1 tmp2=[tmp3],2 };; 67{ .mmi; ld1 tmp0=[inp],2 68 ld1 tmp4=[tmp3],2 // LSB 69 dep $X[$i]=$X[$i],tmp2,8,8 };; 70___ 71if ($i<15) { 72 $code.=<<___; 73{ .mmi; ld1 $Xn=[inp],2 // forward Xload 74 nop.m 0x0 75 dep tmp1=tmp0,tmp4,8,8 };; 76{ .mmi; ld1 tmp2=[tmp3],2 // forward Xload 77 and tmp4=$c,$b 78 dep $X[$i]=$X[$i],tmp1,16,16} //;; 79{ .mmi; add $e=$e,$K_00_19 // e+=K_00_19 80 andcm tmp1=$d,$b 81 dep.z tmp5=$a,5,27 };; // a<<5 82{ .mmi; add $e=$e,$X[$i] // e+=Xload 83 or tmp4=tmp4,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) 84 extr.u tmp1=$a,27,5 };; // a>>27 85{ .mmi; ld1 tmp0=[inp],2 // forward Xload 86 add $e=$e,tmp4 // e+=F_00_19(b,c,d) 87 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) 88{ .mmi; ld1 tmp4=[tmp3],2 // forward Xload 89 or tmp5=tmp1,tmp5 // ROTATE(a,5) 90 mux2 tmp6=$a,0x44 };; // see b in next iteration 91{ .mii; add $e=$e,tmp5 // e+=ROTATE(a,5) 92 dep $Xn=$Xn,tmp2,8,8 // forward Xload 93 mux2 $X[$i]=$X[$i],0x44 } //;; 94 95___ 96 } 97else { 98 $code.=<<___; 99{ .mii; and tmp3=$c,$b 100 dep tmp1=tmp0,tmp4,8,8;; 101 dep $X[$i]=$X[$i],tmp1,16,16} //;; 102{ .mmi; add $e=$e,$K_00_19 // e+=K_00_19 103 andcm tmp1=$d,$b 104 dep.z tmp5=$a,5,27 };; // a<<5 105{ .mmi; add $e=$e,$X[$i] // e+=Xupdate 106 or tmp4=tmp3,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) 107 extr.u tmp1=$a,27,5 } // a>>27 108{ .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate 109 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate 110 nop.i 0 };; 111{ .mmi; add $e=$e,tmp4 // e+=F_00_19(b,c,d) 112 xor $Xn=$Xn,tmp3 // forward Xupdate 113 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) 114{ .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5) 115 mux2 tmp6=$a,0x44 };; // see b in next iteration 116{ .mii; add $e=$e,tmp1 // e+=ROTATE(a,5) 117 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) 118 mux2 $X[$i]=$X[$i],0x44 };; 119 120___ 121 } 122} 123 124sub BODY_16_19 { 125local *code=shift; 126my ($i,$a,$b,$c,$d,$e)=@_; 127my $j=$i+1; 128my $Xn=@X[$j%16]; 129 130$code.=<<___; 131{ .mib; add $e=$e,$K_00_19 // e+=K_00_19 132 dep.z tmp5=$a,5,27 } // a<<5 133{ .mib; andcm tmp1=$d,$b 134 and tmp0=$c,$b };; 135{ .mmi; add $e=$e,$X[$i%16] // e+=Xupdate 136 or tmp0=tmp0,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) 137 extr.u tmp1=$a,27,5 } // a>>27 138{ .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate 139 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate 140 nop.i 0 };; 141{ .mmi; add $e=$e,tmp0 // f+=F_00_19(b,c,d) 142 xor $Xn=$Xn,tmp3 // forward Xupdate 143 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) 144{ .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5) 145 mux2 tmp6=$a,0x44 };; // see b in next iteration 146{ .mii; add $e=$e,tmp1 // e+=ROTATE(a,5) 147 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) 148 nop.i 0 };; 149 150___ 151} 152 153sub BODY_20_39 { 154local *code=shift; 155my ($i,$a,$b,$c,$d,$e,$Konst)=@_; 156 $Konst = $K_20_39 if (!defined($Konst)); 157my $j=$i+1; 158my $Xn=@X[$j%16]; 159 160if ($i<79) { 161$code.=<<___; 162{ .mib; add $e=$e,$Konst // e+=K_XX_XX 163 dep.z tmp5=$a,5,27 } // a<<5 164{ .mib; xor tmp0=$c,$b 165 xor $Xn=$Xn,$X[($j+2)%16] };; // forward Xupdate 166{ .mib; add $e=$e,$X[$i%16] // e+=Xupdate 167 extr.u tmp1=$a,27,5 } // a>>27 168{ .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d 169 xor $Xn=$Xn,$X[($j+8)%16] };; // forward Xupdate 170{ .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d) 171 xor $Xn=$Xn,$X[($j+13)%16] // forward Xupdate 172 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) 173{ .mmi; or tmp1=tmp1,tmp5 // ROTATE(a,5) 174 mux2 tmp6=$a,0x44 };; // see b in next iteration 175{ .mii; add $e=$e,tmp1 // e+=ROTATE(a,5) 176 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) 177 nop.i 0 };; 178 179___ 180} 181else { 182$code.=<<___; 183{ .mib; add $e=$e,$Konst // e+=K_60_79 184 dep.z tmp5=$a,5,27 } // a<<5 185{ .mib; xor tmp0=$c,$b 186 add $h1=$h1,$a };; // wrap up 187{ .mib; add $e=$e,$X[$i%16] // e+=Xupdate 188 extr.u tmp1=$a,27,5 } // a>>27 189{ .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d 190 add $h3=$h3,$c };; // wrap up 191{ .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d) 192 or tmp1=tmp1,tmp5 // ROTATE(a,5) 193 shrp $b=tmp6,tmp6,2 };; // b=ROTATE(b,30) ;;? 194{ .mmi; add $e=$e,tmp1 // e+=ROTATE(a,5) 195 add tmp3=1,inp // used in unaligned codepath 196 add $h4=$h4,$d };; // wrap up 197 198___ 199} 200} 201 202sub BODY_40_59 { 203local *code=shift; 204my ($i,$a,$b,$c,$d,$e)=@_; 205my $j=$i+1; 206my $Xn=@X[$j%16]; 207 208$code.=<<___; 209{ .mib; add $e=$e,$K_40_59 // e+=K_40_59 210 dep.z tmp5=$a,5,27 } // a<<5 211{ .mib; and tmp1=$c,$d 212 xor tmp0=$c,$d };; 213{ .mmi; add $e=$e,$X[$i%16] // e+=Xupdate 214 add tmp5=tmp5,tmp1 // a<<5+(c&d) 215 extr.u tmp1=$a,27,5 } // a>>27 216{ .mmi; and tmp0=tmp0,$b 217 xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate 218 xor tmp3=$X[($j+8)%16],$X[($j+13)%16] };; // forward Xupdate 219{ .mmi; add $e=$e,tmp0 // e+=b&(c^d) 220 add tmp5=tmp5,tmp1 // ROTATE(a,5)+(c&d) 221 shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) 222{ .mmi; xor $Xn=$Xn,tmp3 223 mux2 tmp6=$a,0x44 };; // see b in next iteration 224{ .mii; add $e=$e,tmp5 // e+=ROTATE(a,5)+(c&d) 225 shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) 226 nop.i 0x0 };; 227 228___ 229} 230sub BODY_60_79 { &BODY_20_39(@_,$K_60_79); } 231 232$code.=<<___; 233.text 234 235tmp0=r8; 236tmp1=r9; 237tmp2=r10; 238tmp3=r11; 239ctx=r32; // in0 240inp=r33; // in1 241 242// void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num); 243.global sha1_block_data_order# 244.proc sha1_block_data_order# 245.align 32 246sha1_block_data_order: 247 .prologue 248{ .mmi; alloc tmp1=ar.pfs,3,14,0,0 249 $ADDP tmp0=4,ctx 250 .save ar.lc,r3 251 mov r3=ar.lc } 252{ .mmi; $ADDP ctx=0,ctx 253 $ADDP inp=0,inp 254 mov r2=pr };; 255tmp4=in2; 256tmp5=loc12; 257tmp6=loc13; 258 .body 259{ .mlx; ld4 $h0=[ctx],8 260 movl $K_00_19=0x5a827999 } 261{ .mlx; ld4 $h1=[tmp0],8 262 movl $K_20_39=0x6ed9eba1 };; 263{ .mlx; ld4 $h2=[ctx],8 264 movl $K_40_59=0x8f1bbcdc } 265{ .mlx; ld4 $h3=[tmp0] 266 movl $K_60_79=0xca62c1d6 };; 267{ .mmi; ld4 $h4=[ctx],-16 268 add in2=-1,in2 // adjust num for ar.lc 269 mov ar.ec=1 };; 270{ .mmi; nop.m 0 271 add tmp3=1,inp 272 mov ar.lc=in2 };; // brp.loop.imp: too far 273 274.Ldtop: 275{ .mmi; mov $A=$h0 276 mov $B=$h1 277 mux2 tmp6=$h1,0x44 } 278{ .mmi; mov $C=$h2 279 mov $D=$h3 280 mov $E=$h4 };; 281 282___ 283 284{ my $i; 285 my @V=($A,$B,$C,$D,$E); 286 287 for($i=0;$i<16;$i++) { &BODY_00_15(\$code,$i,@V); unshift(@V,pop(@V)); } 288 for(;$i<20;$i++) { &BODY_16_19(\$code,$i,@V); unshift(@V,pop(@V)); } 289 for(;$i<40;$i++) { &BODY_20_39(\$code,$i,@V); unshift(@V,pop(@V)); } 290 for(;$i<60;$i++) { &BODY_40_59(\$code,$i,@V); unshift(@V,pop(@V)); } 291 for(;$i<80;$i++) { &BODY_60_79(\$code,$i,@V); unshift(@V,pop(@V)); } 292 293 (($V[0] eq $A) and ($V[4] eq $E)) or die; # double-check 294} 295 296$code.=<<___; 297{ .mmb; add $h0=$h0,$A 298 add $h2=$h2,$C 299 br.ctop.dptk.many .Ldtop };; 300.Ldend: 301{ .mmi; add tmp0=4,ctx 302 mov ar.lc=r3 };; 303{ .mmi; st4 [ctx]=$h0,8 304 st4 [tmp0]=$h1,8 };; 305{ .mmi; st4 [ctx]=$h2,8 306 st4 [tmp0]=$h3 };; 307{ .mib; st4 [ctx]=$h4,-16 308 mov pr=r2,0x1ffff 309 br.ret.sptk.many b0 };; 310.endp sha1_block_data_order# 311stringz "SHA1 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>" 312___ 313 314open STDOUT,">$output" if $output; 315print $code; 316