Kinson Chik | a8fa74c | 2011-07-29 11:33:41 -0700 | [diff] [blame] | 1 | #!/usr/bin/env perl |
| 2 | # |
| 3 | # ==================================================================== |
| 4 | # Written by Andy Polyakov <appro@fy.chalmers.se> 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 block transform for x86. September 2007. |
| 11 | # |
| 12 | # Performance in clock cycles per processed byte (less is better): |
| 13 | # |
| 14 | # Pentium PIII P4 AMD K8 Core2 |
| 15 | # gcc 46 36 41 27 26 |
| 16 | # icc 57 33 38 25 23 |
| 17 | # x86 asm 40 30 35 20 20 |
| 18 | # x86_64 asm(*) - - 21 15.8 16.5 |
| 19 | # |
| 20 | # (*) x86_64 assembler performance is presented for reference |
| 21 | # purposes. |
| 22 | # |
| 23 | # Performance improvement over compiler generated code varies from |
| 24 | # 10% to 40% [see above]. Not very impressive on some ยต-archs, but |
| 25 | # it's 5 times smaller and optimizies amount of writes. |
| 26 | |
| 27 | $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1; |
| 28 | push(@INC,"${dir}","${dir}../../perlasm"); |
| 29 | require "x86asm.pl"; |
| 30 | |
| 31 | &asm_init($ARGV[0],"sha512-586.pl",$ARGV[$#ARGV] eq "386"); |
| 32 | |
| 33 | $A="eax"; |
| 34 | $E="edx"; |
| 35 | $T="ebx"; |
| 36 | $Aoff=&DWP(0,"esp"); |
| 37 | $Boff=&DWP(4,"esp"); |
| 38 | $Coff=&DWP(8,"esp"); |
| 39 | $Doff=&DWP(12,"esp"); |
| 40 | $Eoff=&DWP(16,"esp"); |
| 41 | $Foff=&DWP(20,"esp"); |
| 42 | $Goff=&DWP(24,"esp"); |
| 43 | $Hoff=&DWP(28,"esp"); |
| 44 | $Xoff=&DWP(32,"esp"); |
| 45 | $K256="ebp"; |
| 46 | |
| 47 | sub BODY_00_15() { |
| 48 | my $in_16_63=shift; |
| 49 | |
| 50 | &mov ("ecx",$E); |
| 51 | &add ($T,&DWP(4*(8+15+16-9),"esp")) if ($in_16_63); # T += X[-7] |
| 52 | &ror ("ecx",6); |
| 53 | &mov ("edi",$E); |
| 54 | &ror ("edi",11); |
| 55 | &mov ("esi",$Foff); |
| 56 | &xor ("ecx","edi"); |
| 57 | &ror ("edi",25-11); |
| 58 | &mov (&DWP(4*(8+15),"esp"),$T) if ($in_16_63); # save X[0] |
| 59 | &xor ("ecx","edi"); # Sigma1(e) |
| 60 | &mov ("edi",$Goff); |
| 61 | &add ($T,"ecx"); # T += Sigma1(e) |
| 62 | &mov ($Eoff,$E); # modulo-scheduled |
| 63 | |
| 64 | &xor ("esi","edi"); |
| 65 | &mov ("ecx",$A); |
| 66 | &and ("esi",$E); |
| 67 | &mov ($E,$Doff); # e becomes d, which is e in next iteration |
| 68 | &xor ("esi","edi"); # Ch(e,f,g) |
| 69 | &mov ("edi",$A); |
| 70 | &add ($T,"esi"); # T += Ch(e,f,g) |
| 71 | |
| 72 | &ror ("ecx",2); |
| 73 | &add ($T,$Hoff); # T += h |
| 74 | &ror ("edi",13); |
| 75 | &mov ("esi",$Boff); |
| 76 | &xor ("ecx","edi"); |
| 77 | &ror ("edi",22-13); |
| 78 | &add ($E,$T); # d += T |
| 79 | &xor ("ecx","edi"); # Sigma0(a) |
| 80 | &mov ("edi",$Coff); |
| 81 | |
| 82 | &add ($T,"ecx"); # T += Sigma0(a) |
| 83 | &mov ($Aoff,$A); # modulo-scheduled |
| 84 | |
| 85 | &mov ("ecx",$A); |
| 86 | &sub ("esp",4); |
| 87 | &or ($A,"esi"); # a becomes h, which is a in next iteration |
| 88 | &and ("ecx","esi"); |
| 89 | &and ($A,"edi"); |
| 90 | &mov ("esi",&DWP(0,$K256)); |
| 91 | &or ($A,"ecx"); # h=Maj(a,b,c) |
| 92 | |
| 93 | &add ($K256,4); |
| 94 | &add ($A,$T); # h += T |
| 95 | &mov ($T,&DWP(4*(8+15+16-1),"esp")) if ($in_16_63); # preload T |
| 96 | &add ($E,"esi"); # d += K256[i] |
| 97 | &add ($A,"esi"); # h += K256[i] |
| 98 | } |
| 99 | |
| 100 | &function_begin("sha256_block_data_order"); |
| 101 | &mov ("esi",wparam(0)); # ctx |
| 102 | &mov ("edi",wparam(1)); # inp |
| 103 | &mov ("eax",wparam(2)); # num |
| 104 | &mov ("ebx","esp"); # saved sp |
| 105 | |
| 106 | &call (&label("pic_point")); # make it PIC! |
| 107 | &set_label("pic_point"); |
| 108 | &blindpop($K256); |
| 109 | &lea ($K256,&DWP(&label("K256")."-".&label("pic_point"),$K256)); |
| 110 | |
| 111 | &sub ("esp",16); |
| 112 | &and ("esp",-64); |
| 113 | |
| 114 | &shl ("eax",6); |
| 115 | &add ("eax","edi"); |
| 116 | &mov (&DWP(0,"esp"),"esi"); # ctx |
| 117 | &mov (&DWP(4,"esp"),"edi"); # inp |
| 118 | &mov (&DWP(8,"esp"),"eax"); # inp+num*128 |
| 119 | &mov (&DWP(12,"esp"),"ebx"); # saved sp |
| 120 | |
| 121 | &set_label("loop",16); |
| 122 | # copy input block to stack reversing byte and dword order |
| 123 | for($i=0;$i<4;$i++) { |
| 124 | &mov ("eax",&DWP($i*16+0,"edi")); |
| 125 | &mov ("ebx",&DWP($i*16+4,"edi")); |
| 126 | &mov ("ecx",&DWP($i*16+8,"edi")); |
| 127 | &mov ("edx",&DWP($i*16+12,"edi")); |
| 128 | &bswap ("eax"); |
| 129 | &bswap ("ebx"); |
| 130 | &bswap ("ecx"); |
| 131 | &bswap ("edx"); |
| 132 | &push ("eax"); |
| 133 | &push ("ebx"); |
| 134 | &push ("ecx"); |
| 135 | &push ("edx"); |
| 136 | } |
| 137 | &add ("edi",64); |
| 138 | &sub ("esp",4*8); # place for A,B,C,D,E,F,G,H |
| 139 | &mov (&DWP(4*(8+16)+4,"esp"),"edi"); |
| 140 | |
| 141 | # copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack |
| 142 | &mov ($A,&DWP(0,"esi")); |
| 143 | &mov ("ebx",&DWP(4,"esi")); |
| 144 | &mov ("ecx",&DWP(8,"esi")); |
| 145 | &mov ("edi",&DWP(12,"esi")); |
| 146 | # &mov ($Aoff,$A); |
| 147 | &mov ($Boff,"ebx"); |
| 148 | &mov ($Coff,"ecx"); |
| 149 | &mov ($Doff,"edi"); |
| 150 | &mov ($E,&DWP(16,"esi")); |
| 151 | &mov ("ebx",&DWP(20,"esi")); |
| 152 | &mov ("ecx",&DWP(24,"esi")); |
| 153 | &mov ("edi",&DWP(28,"esi")); |
| 154 | # &mov ($Eoff,$E); |
| 155 | &mov ($Foff,"ebx"); |
| 156 | &mov ($Goff,"ecx"); |
| 157 | &mov ($Hoff,"edi"); |
| 158 | |
| 159 | &set_label("00_15",16); |
| 160 | &mov ($T,&DWP(4*(8+15),"esp")); |
| 161 | |
| 162 | &BODY_00_15(); |
| 163 | |
| 164 | &cmp ("esi",0xc19bf174); |
| 165 | &jne (&label("00_15")); |
| 166 | |
| 167 | &mov ($T,&DWP(4*(8+15+16-1),"esp")); # preloaded in BODY_00_15(1) |
| 168 | &set_label("16_63",16); |
| 169 | &mov ("esi",$T); |
| 170 | &mov ("ecx",&DWP(4*(8+15+16-14),"esp")); |
| 171 | &shr ($T,3); |
| 172 | &ror ("esi",7); |
| 173 | &xor ($T,"esi"); |
| 174 | &ror ("esi",18-7); |
| 175 | &mov ("edi","ecx"); |
| 176 | &xor ($T,"esi"); # T = sigma0(X[-15]) |
| 177 | |
| 178 | &shr ("ecx",10); |
| 179 | &mov ("esi",&DWP(4*(8+15+16),"esp")); |
| 180 | &ror ("edi",17); |
| 181 | &xor ("ecx","edi"); |
| 182 | &ror ("edi",19-17); |
| 183 | &add ($T,"esi"); # T += X[-16] |
| 184 | &xor ("edi","ecx") # sigma1(X[-2]) |
| 185 | |
| 186 | &add ($T,"edi"); # T += sigma1(X[-2]) |
| 187 | # &add ($T,&DWP(4*(8+15+16-9),"esp")); # T += X[-7], moved to BODY_00_15(1) |
| 188 | # &mov (&DWP(4*(8+15),"esp"),$T); # save X[0] |
| 189 | |
| 190 | &BODY_00_15(1); |
| 191 | |
| 192 | &cmp ("esi",0xc67178f2); |
| 193 | &jne (&label("16_63")); |
| 194 | |
| 195 | &mov ("esi",&DWP(4*(8+16+64)+0,"esp"));#ctx |
| 196 | # &mov ($A,$Aoff); |
| 197 | &mov ("ebx",$Boff); |
| 198 | &mov ("ecx",$Coff); |
| 199 | &mov ("edi",$Doff); |
| 200 | &add ($A,&DWP(0,"esi")); |
| 201 | &add ("ebx",&DWP(4,"esi")); |
| 202 | &add ("ecx",&DWP(8,"esi")); |
| 203 | &add ("edi",&DWP(12,"esi")); |
| 204 | &mov (&DWP(0,"esi"),$A); |
| 205 | &mov (&DWP(4,"esi"),"ebx"); |
| 206 | &mov (&DWP(8,"esi"),"ecx"); |
| 207 | &mov (&DWP(12,"esi"),"edi"); |
| 208 | # &mov ($E,$Eoff); |
| 209 | &mov ("eax",$Foff); |
| 210 | &mov ("ebx",$Goff); |
| 211 | &mov ("ecx",$Hoff); |
| 212 | &mov ("edi",&DWP(4*(8+16+64)+4,"esp"));#inp |
| 213 | &add ($E,&DWP(16,"esi")); |
| 214 | &add ("eax",&DWP(20,"esi")); |
| 215 | &add ("ebx",&DWP(24,"esi")); |
| 216 | &add ("ecx",&DWP(28,"esi")); |
| 217 | &mov (&DWP(16,"esi"),$E); |
| 218 | &mov (&DWP(20,"esi"),"eax"); |
| 219 | &mov (&DWP(24,"esi"),"ebx"); |
| 220 | &mov (&DWP(28,"esi"),"ecx"); |
| 221 | |
| 222 | &add ("esp",4*(8+16+64)); # destroy frame |
| 223 | &sub ($K256,4*64); # rewind K |
| 224 | |
| 225 | &cmp ("edi",&DWP(8,"esp")); # are we done yet? |
| 226 | &jb (&label("loop")); |
| 227 | |
| 228 | &mov ("esp",&DWP(12,"esp")); # restore sp |
| 229 | &function_end_A(); |
| 230 | |
| 231 | &set_label("K256",64); # Yes! I keep it in the code segment! |
| 232 | &data_word(0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5); |
| 233 | &data_word(0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5); |
| 234 | &data_word(0xd807aa98,0x12835b01,0x243185be,0x550c7dc3); |
| 235 | &data_word(0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174); |
| 236 | &data_word(0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc); |
| 237 | &data_word(0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da); |
| 238 | &data_word(0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7); |
| 239 | &data_word(0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967); |
| 240 | &data_word(0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13); |
| 241 | &data_word(0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85); |
| 242 | &data_word(0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3); |
| 243 | &data_word(0xd192e819,0xd6990624,0xf40e3585,0x106aa070); |
| 244 | &data_word(0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5); |
| 245 | &data_word(0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3); |
| 246 | &data_word(0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208); |
| 247 | &data_word(0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2); |
| 248 | &function_end_B("sha256_block_data_order"); |
| 249 | &asciz("SHA256 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>"); |
| 250 | |
| 251 | &asm_finish(); |