1160814Ssimon#!/usr/bin/perl -w 2160814Ssimon# 3160814Ssimon# MD5 optimized for AMD64. 4160814Ssimon# 5160814Ssimon# Author: Marc Bevand <bevand_m (at) epita.fr> 6160814Ssimon# Licence: I hereby disclaim the copyright on this code and place it 7160814Ssimon# in the public domain. 8160814Ssimon# 9160814Ssimon 10160814Ssimonuse strict; 11160814Ssimon 12160814Ssimonmy $code; 13160814Ssimon 14160814Ssimon# round1_step() does: 15160814Ssimon# dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s) 16160814Ssimon# %r10d = X[k_next] 17160814Ssimon# %r11d = z' (copy of z for the next step) 18160814Ssimon# Each round1_step() takes about 5.71 clocks (9 instructions, 1.58 IPC) 19160814Ssimonsub round1_step 20160814Ssimon{ 21160814Ssimon my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 22205128Ssimon $T_i = unpack("l",pack("l", hex($T_i))); # convert to 32-bit signed decimal 23160814Ssimon $code .= " mov 0*4(%rsi), %r10d /* (NEXT STEP) X[0] */\n" if ($pos == -1); 24160814Ssimon $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1); 25160814Ssimon $code .= <<EOF; 26160814Ssimon xor $y, %r11d /* y ^ ... */ 27160814Ssimon lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 28160814Ssimon and $x, %r11d /* x & ... */ 29160814Ssimon xor $z, %r11d /* z ^ ... */ 30160814Ssimon mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 31160814Ssimon add %r11d, $dst /* dst += ... */ 32160814Ssimon rol \$$s, $dst /* dst <<< s */ 33160814Ssimon mov $y, %r11d /* (NEXT STEP) z' = $y */ 34160814Ssimon add $x, $dst /* dst += x */ 35160814SsimonEOF 36160814Ssimon} 37160814Ssimon 38160814Ssimon# round2_step() does: 39160814Ssimon# dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s) 40160814Ssimon# %r10d = X[k_next] 41160814Ssimon# %r11d = y' (copy of y for the next step) 42160814Ssimon# Each round2_step() takes about 6.22 clocks (9 instructions, 1.45 IPC) 43160814Ssimonsub round2_step 44160814Ssimon{ 45160814Ssimon my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 46205128Ssimon $T_i = unpack("l",pack("l", hex($T_i))); # convert to 32-bit signed decimal 47160814Ssimon $code .= " mov 1*4(%rsi), %r10d /* (NEXT STEP) X[1] */\n" if ($pos == -1); 48160814Ssimon $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1); 49160814Ssimon $code .= <<EOF; 50160814Ssimon xor $x, %r11d /* x ^ ... */ 51160814Ssimon lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 52160814Ssimon and $z, %r11d /* z & ... */ 53160814Ssimon xor $y, %r11d /* y ^ ... */ 54160814Ssimon mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 55160814Ssimon add %r11d, $dst /* dst += ... */ 56160814Ssimon rol \$$s, $dst /* dst <<< s */ 57160814Ssimon mov $x, %r11d /* (NEXT STEP) y' = $x */ 58160814Ssimon add $x, $dst /* dst += x */ 59160814SsimonEOF 60160814Ssimon} 61160814Ssimon 62160814Ssimon# round3_step() does: 63160814Ssimon# dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s) 64160814Ssimon# %r10d = X[k_next] 65160814Ssimon# %r11d = y' (copy of y for the next step) 66160814Ssimon# Each round3_step() takes about 4.26 clocks (8 instructions, 1.88 IPC) 67160814Ssimonsub round3_step 68160814Ssimon{ 69160814Ssimon my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 70205128Ssimon $T_i = unpack("l",pack("l", hex($T_i))); # convert to 32-bit signed decimal 71160814Ssimon $code .= " mov 5*4(%rsi), %r10d /* (NEXT STEP) X[5] */\n" if ($pos == -1); 72160814Ssimon $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1); 73160814Ssimon $code .= <<EOF; 74160814Ssimon lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 75160814Ssimon mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 76160814Ssimon xor $z, %r11d /* z ^ ... */ 77160814Ssimon xor $x, %r11d /* x ^ ... */ 78160814Ssimon add %r11d, $dst /* dst += ... */ 79160814Ssimon rol \$$s, $dst /* dst <<< s */ 80160814Ssimon mov $x, %r11d /* (NEXT STEP) y' = $x */ 81160814Ssimon add $x, $dst /* dst += x */ 82160814SsimonEOF 83160814Ssimon} 84160814Ssimon 85160814Ssimon# round4_step() does: 86160814Ssimon# dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s) 87160814Ssimon# %r10d = X[k_next] 88160814Ssimon# %r11d = not z' (copy of not z for the next step) 89160814Ssimon# Each round4_step() takes about 5.27 clocks (9 instructions, 1.71 IPC) 90160814Ssimonsub round4_step 91160814Ssimon{ 92160814Ssimon my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_; 93205128Ssimon $T_i = unpack("l",pack("l", hex($T_i))); # convert to 32-bit signed decimal 94160814Ssimon $code .= " mov 0*4(%rsi), %r10d /* (NEXT STEP) X[0] */\n" if ($pos == -1); 95160814Ssimon $code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1); 96160814Ssimon $code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n" 97160814Ssimon if ($pos == -1); 98160814Ssimon $code .= <<EOF; 99160814Ssimon lea $T_i($dst,%r10d),$dst /* Const + dst + ... */ 100160814Ssimon or $x, %r11d /* x | ... */ 101160814Ssimon xor $y, %r11d /* y ^ ... */ 102160814Ssimon add %r11d, $dst /* dst += ... */ 103160814Ssimon mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */ 104160814Ssimon mov \$0xffffffff, %r11d 105160814Ssimon rol \$$s, $dst /* dst <<< s */ 106160814Ssimon xor $y, %r11d /* (NEXT STEP) not z' = not $y */ 107160814Ssimon add $x, $dst /* dst += x */ 108160814SsimonEOF 109160814Ssimon} 110160814Ssimon 111160814Ssimonmy $output = shift; 112160814Ssimonopen STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output"; 113160814Ssimon 114160814Ssimon$code .= <<EOF; 115160814Ssimon.text 116160814Ssimon.align 16 117160814Ssimon 118194206Ssimon.globl md5_block_asm_data_order 119194206Ssimon.type md5_block_asm_data_order,\@function,3 120194206Ssimonmd5_block_asm_data_order: 121160814Ssimon push %rbp 122160814Ssimon push %rbx 123160814Ssimon push %r14 124160814Ssimon push %r15 125160814Ssimon 126160814Ssimon # rdi = arg #1 (ctx, MD5_CTX pointer) 127160814Ssimon # rsi = arg #2 (ptr, data pointer) 128160814Ssimon # rdx = arg #3 (nbr, number of 16-word blocks to process) 129160814Ssimon mov %rdi, %rbp # rbp = ctx 130160814Ssimon shl \$6, %rdx # rdx = nbr in bytes 131160814Ssimon lea (%rsi,%rdx), %rdi # rdi = end 132160814Ssimon mov 0*4(%rbp), %eax # eax = ctx->A 133160814Ssimon mov 1*4(%rbp), %ebx # ebx = ctx->B 134160814Ssimon mov 2*4(%rbp), %ecx # ecx = ctx->C 135160814Ssimon mov 3*4(%rbp), %edx # edx = ctx->D 136160814Ssimon # end is 'rdi' 137160814Ssimon # ptr is 'rsi' 138160814Ssimon # A is 'eax' 139160814Ssimon # B is 'ebx' 140160814Ssimon # C is 'ecx' 141160814Ssimon # D is 'edx' 142160814Ssimon 143160814Ssimon cmp %rdi, %rsi # cmp end with ptr 144160814Ssimon je .Lend # jmp if ptr == end 145160814Ssimon 146160814Ssimon # BEGIN of loop over 16-word blocks 147160814Ssimon.Lloop: # save old values of A, B, C, D 148160814Ssimon mov %eax, %r8d 149160814Ssimon mov %ebx, %r9d 150160814Ssimon mov %ecx, %r14d 151160814Ssimon mov %edx, %r15d 152160814SsimonEOF 153160814Ssimonround1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7'); 154160814Ssimonround1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12'); 155160814Ssimonround1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17'); 156160814Ssimonround1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22'); 157160814Ssimonround1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7'); 158160814Ssimonround1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12'); 159160814Ssimonround1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17'); 160160814Ssimonround1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22'); 161160814Ssimonround1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7'); 162160814Ssimonround1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12'); 163160814Ssimonround1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17'); 164160814Ssimonround1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22'); 165160814Ssimonround1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7'); 166160814Ssimonround1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12'); 167160814Ssimonround1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17'); 168160814Ssimonround1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22'); 169160814Ssimon 170160814Ssimonround2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5'); 171160814Ssimonround2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9'); 172160814Ssimonround2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14'); 173160814Ssimonround2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20'); 174160814Ssimonround2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5'); 175160814Ssimonround2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9'); 176160814Ssimonround2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14'); 177160814Ssimonround2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20'); 178160814Ssimonround2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5'); 179160814Ssimonround2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9'); 180160814Ssimonround2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14'); 181160814Ssimonround2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20'); 182160814Ssimonround2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5'); 183160814Ssimonround2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9'); 184160814Ssimonround2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14'); 185160814Ssimonround2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20'); 186160814Ssimon 187160814Ssimonround3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4'); 188160814Ssimonround3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11'); 189160814Ssimonround3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16'); 190160814Ssimonround3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23'); 191160814Ssimonround3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4'); 192160814Ssimonround3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11'); 193160814Ssimonround3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16'); 194160814Ssimonround3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23'); 195160814Ssimonround3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4'); 196160814Ssimonround3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11'); 197160814Ssimonround3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16'); 198160814Ssimonround3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23'); 199160814Ssimonround3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4'); 200160814Ssimonround3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11'); 201160814Ssimonround3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16'); 202160814Ssimonround3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23'); 203160814Ssimon 204160814Ssimonround4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6'); 205160814Ssimonround4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10'); 206160814Ssimonround4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15'); 207160814Ssimonround4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21'); 208160814Ssimonround4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6'); 209160814Ssimonround4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10'); 210160814Ssimonround4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15'); 211160814Ssimonround4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21'); 212160814Ssimonround4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6'); 213160814Ssimonround4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10'); 214160814Ssimonround4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15'); 215160814Ssimonround4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21'); 216160814Ssimonround4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6'); 217160814Ssimonround4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10'); 218160814Ssimonround4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15'); 219160814Ssimonround4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21'); 220160814Ssimon$code .= <<EOF; 221160814Ssimon # add old values of A, B, C, D 222160814Ssimon add %r8d, %eax 223160814Ssimon add %r9d, %ebx 224160814Ssimon add %r14d, %ecx 225160814Ssimon add %r15d, %edx 226160814Ssimon 227160814Ssimon # loop control 228160814Ssimon add \$64, %rsi # ptr += 64 229160814Ssimon cmp %rdi, %rsi # cmp end with ptr 230160814Ssimon jb .Lloop # jmp if ptr < end 231160814Ssimon # END of loop over 16-word blocks 232160814Ssimon 233160814Ssimon.Lend: 234160814Ssimon mov %eax, 0*4(%rbp) # ctx->A = A 235160814Ssimon mov %ebx, 1*4(%rbp) # ctx->B = B 236160814Ssimon mov %ecx, 2*4(%rbp) # ctx->C = C 237160814Ssimon mov %edx, 3*4(%rbp) # ctx->D = D 238160814Ssimon 239160814Ssimon pop %r15 240160814Ssimon pop %r14 241160814Ssimon pop %rbx 242160814Ssimon pop %rbp 243160814Ssimon ret 244194206Ssimon.size md5_block_asm_data_order,.-md5_block_asm_data_order 245160814SsimonEOF 246160814Ssimon 247160814Ssimonprint $code; 248160814Ssimon 249160814Ssimonclose STDOUT; 250