3 # MD5 optimized for AMD64.
5 # Author: Marc Bevand <bevand_m (at) epita.fr>
6 # Licence: I hereby disclaim the copyright on this code and place it
7 # in the public domain.
15 # dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
17 # %r11d = z' (copy of z for the next step)
18 # Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
21 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
22 $code .= " mov 0*4(%rsi), %r10d /* (NEXT STEP) X[0] */\n" if ($pos == -1);
23 $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
25 xor $y, %r11d /* y ^ ... */
26 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
27 and $x, %r11d /* x & ... */
28 xor $z, %r11d /* z ^ ... */
29 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
30 add %r11d, $dst /* dst += ... */
31 rol \$$s, $dst /* dst <<< s */
32 mov $y, %r11d /* (NEXT STEP) z' = $y */
33 add $x, $dst /* dst += x */
38 # dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
40 # %r11d = z' (copy of z for the next step)
41 # %r12d = z' (copy of z for the next step)
42 # Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
45 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
46 $code .= " mov 1*4(%rsi), %r10d /* (NEXT STEP) X[1] */\n" if ($pos == -1);
47 $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
48 $code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
51 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
52 and $x, %r12d /* x & z */
53 and $y, %r11d /* y & (not z) */
54 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
55 or %r11d, %r12d /* (y & (not z)) | (x & z) */
56 mov $y, %r11d /* (NEXT STEP) z' = $y */
57 add %r12d, $dst /* dst += ... */
58 mov $y, %r12d /* (NEXT STEP) z' = $y */
59 rol \$$s, $dst /* dst <<< s */
60 add $x, $dst /* dst += x */
65 # dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
67 # %r11d = y' (copy of y for the next step)
68 # Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
71 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
72 $code .= " mov 5*4(%rsi), %r10d /* (NEXT STEP) X[5] */\n" if ($pos == -1);
73 $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
75 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
76 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
77 xor $z, %r11d /* z ^ ... */
78 xor $x, %r11d /* x ^ ... */
79 add %r11d, $dst /* dst += ... */
80 rol \$$s, $dst /* dst <<< s */
81 mov $x, %r11d /* (NEXT STEP) y' = $x */
82 add $x, $dst /* dst += x */
87 # dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
89 # %r11d = not z' (copy of not z for the next step)
90 # Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
93 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
94 $code .= " mov 0*4(%rsi), %r10d /* (NEXT STEP) X[0] */\n" if ($pos == -1);
95 $code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1);
96 $code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n"
99 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
100 or $x, %r11d /* x | ... */
101 xor $y, %r11d /* y ^ ... */
102 add %r11d, $dst /* dst += ... */
103 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
104 mov \$0xffffffff, %r11d
105 rol \$$s, $dst /* dst <<< s */
106 xor $y, %r11d /* (NEXT STEP) not z' = not $y */
107 add $x, $dst /* dst += x */
112 open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
118 .globl md5_block_asm_host_order
119 .type md5_block_asm_host_order,\@function,3
120 md5_block_asm_host_order:
127 # rdi = arg #1 (ctx, MD5_CTX pointer)
128 # rsi = arg #2 (ptr, data pointer)
129 # rdx = arg #3 (nbr, number of 16-word blocks to process)
130 mov %rdi, %rbp # rbp = ctx
131 shl \$6, %rdx # rdx = nbr in bytes
132 lea (%rsi,%rdx), %rdi # rdi = end
133 mov 0*4(%rbp), %eax # eax = ctx->A
134 mov 1*4(%rbp), %ebx # ebx = ctx->B
135 mov 2*4(%rbp), %ecx # ecx = ctx->C
136 mov 3*4(%rbp), %edx # edx = ctx->D
144 cmp %rdi, %rsi # cmp end with ptr
145 je .Lend # jmp if ptr == end
147 # BEGIN of loop over 16-word blocks
148 .Lloop: # save old values of A, B, C, D
154 round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
155 round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
156 round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
157 round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
158 round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
159 round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
160 round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
161 round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
162 round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
163 round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
164 round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
165 round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
166 round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
167 round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
168 round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
169 round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');
171 round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
172 round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
173 round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
174 round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
175 round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
176 round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
177 round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
178 round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
179 round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
180 round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
181 round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
182 round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
183 round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
184 round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
185 round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
186 round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');
188 round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
189 round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
190 round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
191 round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
192 round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
193 round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
194 round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
195 round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
196 round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
197 round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
198 round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
199 round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
200 round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
201 round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
202 round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
203 round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
205 round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
206 round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
207 round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
208 round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
209 round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
210 round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
211 round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
212 round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
213 round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
214 round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
215 round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
216 round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
217 round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
218 round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
219 round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
220 round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
222 # add old values of A, B, C, D
229 add \$64, %rsi # ptr += 64
230 cmp %rdi, %rsi # cmp end with ptr
231 jb .Lloop # jmp if ptr < end
232 # END of loop over 16-word blocks
235 mov %eax, 0*4(%rbp) # ctx->A = A
236 mov %ebx, 1*4(%rbp) # ctx->B = B
237 mov %ecx, 2*4(%rbp) # ctx->C = C
238 mov %edx, 3*4(%rbp) # ctx->D = D
246 .size md5_block_asm_host_order,.-md5_block_asm_host_order