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 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
29 xor $z, %r11d /* z ^ ... */
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 $code .= " mov %edx, %r11d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
47 $code .= " mov %edx, %r12d /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
50 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
51 and $x, %r12d /* x & z */
52 and $y, %r11d /* y & (not z) */
53 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
54 or %r11d, %r12d /* (y & (not z)) | (x & z) */
55 mov $y, %r11d /* (NEXT STEP) z' = $y */
56 add %r12d, $dst /* dst += ... */
57 mov $y, %r12d /* (NEXT STEP) z' = $y */
58 rol \$$s, $dst /* dst <<< s */
59 add $x, $dst /* dst += x */
64 # dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
66 # %r11d = y' (copy of y for the next step)
67 # Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
70 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
71 $code .= " mov %ecx, %r11d /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
73 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
74 xor $z, %r11d /* z ^ ... */
75 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
76 xor $x, %r11d /* x ^ ... */
77 add %r11d, $dst /* dst += ... */
78 rol \$$s, $dst /* dst <<< s */
79 mov $x, %r11d /* (NEXT STEP) y' = $x */
80 add $x, $dst /* dst += x */
85 # dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
87 # %r11d = not z' (copy of not z for the next step)
88 # Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
91 my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
92 $code .= " mov \$0xffffffff, %r11d\n" if ($pos == -1);
93 $code .= " xor %edx, %r11d /* (NEXT STEP) not z' = not %edx*/\n"
96 lea $T_i($dst,%r10d),$dst /* Const + dst + ... */
97 or $x, %r11d /* x | ... */
98 mov $k_next*4(%rsi),%r10d /* (NEXT STEP) X[$k_next] */
99 xor $y, %r11d /* y ^ ... */
100 add %r11d, $dst /* dst += ... */
101 mov \$0xffffffff, %r11d
102 rol \$$s, $dst /* dst <<< s */
103 xor $y, %r11d /* (NEXT STEP) not z' = not $y */
104 add $x, $dst /* dst += x */
110 if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
112 my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
114 $0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
115 ( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
116 ( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
117 die "can't locate x86_64-xlate.pl";
119 no warnings qw(uninitialized);
120 open OUT,"| \"$^X\" $xlate $flavour $output";
127 .globl md5_block_asm_data_order
128 .type md5_block_asm_data_order,\@function,3
129 md5_block_asm_data_order:
137 # rdi = arg #1 (ctx, MD5_CTX pointer)
138 # rsi = arg #2 (ptr, data pointer)
139 # rdx = arg #3 (nbr, number of 16-word blocks to process)
140 mov %rdi, %rbp # rbp = ctx
141 shl \$6, %rdx # rdx = nbr in bytes
142 lea (%rsi,%rdx), %rdi # rdi = end
143 mov 0*4(%rbp), %eax # eax = ctx->A
144 mov 1*4(%rbp), %ebx # ebx = ctx->B
145 mov 2*4(%rbp), %ecx # ecx = ctx->C
146 mov 3*4(%rbp), %edx # edx = ctx->D
154 cmp %rdi, %rsi # cmp end with ptr
155 je .Lend # jmp if ptr == end
157 # BEGIN of loop over 16-word blocks
158 .Lloop: # save old values of A, B, C, D
164 round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
165 round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
166 round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
167 round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
168 round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
169 round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
170 round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
171 round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
172 round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
173 round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
174 round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
175 round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
176 round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
177 round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
178 round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
179 round1_step( 1,'%ebx','%ecx','%edx','%eax', '1','0x49b40821','22');
181 round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
182 round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
183 round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
184 round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
185 round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
186 round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
187 round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
188 round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
189 round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
190 round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
191 round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
192 round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
193 round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
194 round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
195 round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
196 round2_step( 1,'%ebx','%ecx','%edx','%eax', '5','0x8d2a4c8a','20');
198 round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
199 round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
200 round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
201 round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
202 round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
203 round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
204 round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
205 round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
206 round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
207 round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
208 round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
209 round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
210 round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
211 round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
212 round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
213 round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
215 round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
216 round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
217 round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
218 round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
219 round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
220 round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
221 round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
222 round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
223 round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
224 round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
225 round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
226 round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
227 round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
228 round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
229 round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
230 round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
232 # add old values of A, B, C, D
239 add \$64, %rsi # ptr += 64
240 cmp %rdi, %rsi # cmp end with ptr
241 jb .Lloop # jmp if ptr < end
242 # END of loop over 16-word blocks
245 mov %eax, 0*4(%rbp) # ctx->A = A
246 mov %ebx, 1*4(%rbp) # ctx->B = B
247 mov %ecx, 2*4(%rbp) # ctx->C = C
248 mov %edx, 3*4(%rbp) # ctx->D = D
258 .size md5_block_asm_data_order,.-md5_block_asm_data_order
261 # EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
262 # CONTEXT *context,DISPATCHER_CONTEXT *disp)
270 .extern __imp_RtlVirtualUnwind
271 .type se_handler,\@abi-omnipotent
285 mov 120($context),%rax # pull context->Rax
286 mov 248($context),%rbx # pull context->Rip
288 lea .Lprologue(%rip),%r10
289 cmp %r10,%rbx # context->Rip<.Lprologue
292 mov 152($context),%rax # pull context->Rsp
294 lea .Lepilogue(%rip),%r10
295 cmp %r10,%rbx # context->Rip>=.Lepilogue
305 mov %rbx,144($context) # restore context->Rbx
306 mov %rbp,160($context) # restore context->Rbp
307 mov %r12,216($context) # restore context->R12
308 mov %r14,232($context) # restore context->R14
309 mov %r15,240($context) # restore context->R15
314 mov %rax,152($context) # restore context->Rsp
315 mov %rsi,168($context) # restore context->Rsi
316 mov %rdi,176($context) # restore context->Rdi
318 mov 40($disp),%rdi # disp->ContextRecord
319 mov $context,%rsi # context
320 mov \$154,%ecx # sizeof(CONTEXT)
321 .long 0xa548f3fc # cld; rep movsq
324 xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
325 mov 8(%rsi),%rdx # arg2, disp->ImageBase
326 mov 0(%rsi),%r8 # arg3, disp->ControlPc
327 mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
328 mov 40(%rsi),%r10 # disp->ContextRecord
329 lea 56(%rsi),%r11 # &disp->HandlerData
330 lea 24(%rsi),%r12 # &disp->EstablisherFrame
331 mov %r10,32(%rsp) # arg5
332 mov %r11,40(%rsp) # arg6
333 mov %r12,48(%rsp) # arg7
334 mov %rcx,56(%rsp) # arg8, (NULL)
335 call *__imp_RtlVirtualUnwind(%rip)
337 mov \$1,%eax # ExceptionContinueSearch
349 .size se_handler,.-se_handler
353 .rva .LSEH_begin_md5_block_asm_data_order
354 .rva .LSEH_end_md5_block_asm_data_order
355 .rva .LSEH_info_md5_block_asm_data_order
359 .LSEH_info_md5_block_asm_data_order: