1#!/usr/bin/perl -w
2#
3# MD5 optimized for AMD64.
4#
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.
8#
9
10use strict;
11
12my $code;
13
14# round1_step() does:
15#   dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
16#   %r10d = X[k_next]
17#   %r11d = z' (copy of z for the next step)
18# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
19sub round1_step
20{
21    my ($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);
24    $code .= <<EOF;
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,		$dst		/* dst += x */
34EOF
35}
36
37# round2_step() does:
38#   dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
39#   %r10d = X[k_next]
40#   %r11d = z' (copy of z for the next step)
41#   %r12d = z' (copy of z for the next step)
42# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
43sub round2_step
44{
45    my ($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);
49    $code .= <<EOF;
50	not	%r11d				/* not z */
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,		$dst		/* dst += x */
61EOF
62}
63
64# round3_step() does:
65#   dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
66#   %r10d = X[k_next]
67#   %r11d = y' (copy of y for the next step)
68# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
69sub round3_step
70{
71    my ($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);
74    $code .= <<EOF;
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,		$dst		/* dst += x */
83EOF
84}
85
86# round4_step() does:
87#   dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
88#   %r10d = X[k_next]
89#   %r11d = not z' (copy of not z for the next step)
90# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
91sub round4_step
92{
93    my ($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    $code .= " xor	%edx,		%r11d		/* (NEXT STEP) not z' = not %edx*/\n"
97    if ($pos == -1);
98    $code .= <<EOF;
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 */
108EOF
109}
110
111my $flavour = shift;
112my $output  = shift;
113if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
114
115my $win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
116
117$0 =~ m/(.*[\/\\])[^\/\\]+$/; my $dir=$1; my $xlate;
118( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
119( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
120die "can't locate x86_64-xlate.pl";
121
122no warnings qw(uninitialized);
123open STDOUT,"| $^X $xlate $flavour $output";
124
125$code .= <<EOF;
126.text
127.align 16
128
129.globl md5_block_asm_data_order
130.type md5_block_asm_data_order,\@function,3
131md5_block_asm_data_order:
132	push	%rbp
133	push	%rbx
134	push	%r12
135	push	%r14
136	push	%r15
137.Lprologue:
138
139	# rdi = arg #1 (ctx, MD5_CTX pointer)
140	# rsi = arg #2 (ptr, data pointer)
141	# rdx = arg #3 (nbr, number of 16-word blocks to process)
142	mov	%rdi,		%rbp	# rbp = ctx
143	shl	\$6,		%rdx	# rdx = nbr in bytes
144	lea	(%rsi,%rdx),	%rdi	# rdi = end
145	mov	0*4(%rbp),	%eax	# eax = ctx->A
146	mov	1*4(%rbp),	%ebx	# ebx = ctx->B
147	mov	2*4(%rbp),	%ecx	# ecx = ctx->C
148	mov	3*4(%rbp),	%edx	# edx = ctx->D
149	# end is 'rdi'
150	# ptr is 'rsi'
151	# A is 'eax'
152	# B is 'ebx'
153	# C is 'ecx'
154	# D is 'edx'
155
156	cmp	%rdi,		%rsi		# cmp end with ptr
157	je	.Lend				# jmp if ptr == end
158
159	# BEGIN of loop over 16-word blocks
160.Lloop:	# save old values of A, B, C, D
161	mov	%eax,		%r8d
162	mov	%ebx,		%r9d
163	mov	%ecx,		%r14d
164	mov	%edx,		%r15d
165EOF
166round1_step(-1,'%eax','%ebx','%ecx','%edx', '1','0xd76aa478', '7');
167round1_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xe8c7b756','12');
168round1_step( 0,'%ecx','%edx','%eax','%ebx', '3','0x242070db','17');
169round1_step( 0,'%ebx','%ecx','%edx','%eax', '4','0xc1bdceee','22');
170round1_step( 0,'%eax','%ebx','%ecx','%edx', '5','0xf57c0faf', '7');
171round1_step( 0,'%edx','%eax','%ebx','%ecx', '6','0x4787c62a','12');
172round1_step( 0,'%ecx','%edx','%eax','%ebx', '7','0xa8304613','17');
173round1_step( 0,'%ebx','%ecx','%edx','%eax', '8','0xfd469501','22');
174round1_step( 0,'%eax','%ebx','%ecx','%edx', '9','0x698098d8', '7');
175round1_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8b44f7af','12');
176round1_step( 0,'%ecx','%edx','%eax','%ebx','11','0xffff5bb1','17');
177round1_step( 0,'%ebx','%ecx','%edx','%eax','12','0x895cd7be','22');
178round1_step( 0,'%eax','%ebx','%ecx','%edx','13','0x6b901122', '7');
179round1_step( 0,'%edx','%eax','%ebx','%ecx','14','0xfd987193','12');
180round1_step( 0,'%ecx','%edx','%eax','%ebx','15','0xa679438e','17');
181round1_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x49b40821','22');
182
183round2_step(-1,'%eax','%ebx','%ecx','%edx', '6','0xf61e2562', '5');
184round2_step( 0,'%edx','%eax','%ebx','%ecx','11','0xc040b340', '9');
185round2_step( 0,'%ecx','%edx','%eax','%ebx', '0','0x265e5a51','14');
186round2_step( 0,'%ebx','%ecx','%edx','%eax', '5','0xe9b6c7aa','20');
187round2_step( 0,'%eax','%ebx','%ecx','%edx','10','0xd62f105d', '5');
188round2_step( 0,'%edx','%eax','%ebx','%ecx','15', '0x2441453', '9');
189round2_step( 0,'%ecx','%edx','%eax','%ebx', '4','0xd8a1e681','14');
190round2_step( 0,'%ebx','%ecx','%edx','%eax', '9','0xe7d3fbc8','20');
191round2_step( 0,'%eax','%ebx','%ecx','%edx','14','0x21e1cde6', '5');
192round2_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xc33707d6', '9');
193round2_step( 0,'%ecx','%edx','%eax','%ebx', '8','0xf4d50d87','14');
194round2_step( 0,'%ebx','%ecx','%edx','%eax','13','0x455a14ed','20');
195round2_step( 0,'%eax','%ebx','%ecx','%edx', '2','0xa9e3e905', '5');
196round2_step( 0,'%edx','%eax','%ebx','%ecx', '7','0xfcefa3f8', '9');
197round2_step( 0,'%ecx','%edx','%eax','%ebx','12','0x676f02d9','14');
198round2_step( 1,'%ebx','%ecx','%edx','%eax', '0','0x8d2a4c8a','20');
199
200round3_step(-1,'%eax','%ebx','%ecx','%edx', '8','0xfffa3942', '4');
201round3_step( 0,'%edx','%eax','%ebx','%ecx','11','0x8771f681','11');
202round3_step( 0,'%ecx','%edx','%eax','%ebx','14','0x6d9d6122','16');
203round3_step( 0,'%ebx','%ecx','%edx','%eax', '1','0xfde5380c','23');
204round3_step( 0,'%eax','%ebx','%ecx','%edx', '4','0xa4beea44', '4');
205round3_step( 0,'%edx','%eax','%ebx','%ecx', '7','0x4bdecfa9','11');
206round3_step( 0,'%ecx','%edx','%eax','%ebx','10','0xf6bb4b60','16');
207round3_step( 0,'%ebx','%ecx','%edx','%eax','13','0xbebfbc70','23');
208round3_step( 0,'%eax','%ebx','%ecx','%edx', '0','0x289b7ec6', '4');
209round3_step( 0,'%edx','%eax','%ebx','%ecx', '3','0xeaa127fa','11');
210round3_step( 0,'%ecx','%edx','%eax','%ebx', '6','0xd4ef3085','16');
211round3_step( 0,'%ebx','%ecx','%edx','%eax', '9', '0x4881d05','23');
212round3_step( 0,'%eax','%ebx','%ecx','%edx','12','0xd9d4d039', '4');
213round3_step( 0,'%edx','%eax','%ebx','%ecx','15','0xe6db99e5','11');
214round3_step( 0,'%ecx','%edx','%eax','%ebx', '2','0x1fa27cf8','16');
215round3_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xc4ac5665','23');
216
217round4_step(-1,'%eax','%ebx','%ecx','%edx', '7','0xf4292244', '6');
218round4_step( 0,'%edx','%eax','%ebx','%ecx','14','0x432aff97','10');
219round4_step( 0,'%ecx','%edx','%eax','%ebx', '5','0xab9423a7','15');
220round4_step( 0,'%ebx','%ecx','%edx','%eax','12','0xfc93a039','21');
221round4_step( 0,'%eax','%ebx','%ecx','%edx', '3','0x655b59c3', '6');
222round4_step( 0,'%edx','%eax','%ebx','%ecx','10','0x8f0ccc92','10');
223round4_step( 0,'%ecx','%edx','%eax','%ebx', '1','0xffeff47d','15');
224round4_step( 0,'%ebx','%ecx','%edx','%eax', '8','0x85845dd1','21');
225round4_step( 0,'%eax','%ebx','%ecx','%edx','15','0x6fa87e4f', '6');
226round4_step( 0,'%edx','%eax','%ebx','%ecx', '6','0xfe2ce6e0','10');
227round4_step( 0,'%ecx','%edx','%eax','%ebx','13','0xa3014314','15');
228round4_step( 0,'%ebx','%ecx','%edx','%eax', '4','0x4e0811a1','21');
229round4_step( 0,'%eax','%ebx','%ecx','%edx','11','0xf7537e82', '6');
230round4_step( 0,'%edx','%eax','%ebx','%ecx', '2','0xbd3af235','10');
231round4_step( 0,'%ecx','%edx','%eax','%ebx', '9','0x2ad7d2bb','15');
232round4_step( 1,'%ebx','%ecx','%edx','%eax', '0','0xeb86d391','21');
233$code .= <<EOF;
234	# add old values of A, B, C, D
235	add	%r8d,	%eax
236	add	%r9d,	%ebx
237	add	%r14d,	%ecx
238	add	%r15d,	%edx
239
240	# loop control
241	add	\$64,		%rsi		# ptr += 64
242	cmp	%rdi,		%rsi		# cmp end with ptr
243	jb	.Lloop				# jmp if ptr < end
244	# END of loop over 16-word blocks
245
246.Lend:
247	mov	%eax,		0*4(%rbp)	# ctx->A = A
248	mov	%ebx,		1*4(%rbp)	# ctx->B = B
249	mov	%ecx,		2*4(%rbp)	# ctx->C = C
250	mov	%edx,		3*4(%rbp)	# ctx->D = D
251
252	mov	(%rsp),%r15
253	mov	8(%rsp),%r14
254	mov	16(%rsp),%r12
255	mov	24(%rsp),%rbx
256	mov	32(%rsp),%rbp
257	add	\$40,%rsp
258.Lepilogue:
259	ret
260.size md5_block_asm_data_order,.-md5_block_asm_data_order
261EOF
262
263# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
264#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
265if ($win64) {
266my $rec="%rcx";
267my $frame="%rdx";
268my $context="%r8";
269my $disp="%r9";
270
271$code.=<<___;
272.extern	__imp_RtlVirtualUnwind
273.type	se_handler,\@abi-omnipotent
274.align	16
275se_handler:
276	push	%rsi
277	push	%rdi
278	push	%rbx
279	push	%rbp
280	push	%r12
281	push	%r13
282	push	%r14
283	push	%r15
284	pushfq
285	sub	\$64,%rsp
286
287	mov	120($context),%rax	# pull context->Rax
288	mov	248($context),%rbx	# pull context->Rip
289
290	lea	.Lprologue(%rip),%r10
291	cmp	%r10,%rbx		# context->Rip<.Lprologue
292	jb	.Lin_prologue
293
294	mov	152($context),%rax	# pull context->Rsp
295
296	lea	.Lepilogue(%rip),%r10
297	cmp	%r10,%rbx		# context->Rip>=.Lepilogue
298	jae	.Lin_prologue
299
300	lea	40(%rax),%rax
301
302	mov	-8(%rax),%rbp
303	mov	-16(%rax),%rbx
304	mov	-24(%rax),%r12
305	mov	-32(%rax),%r14
306	mov	-40(%rax),%r15
307	mov	%rbx,144($context)	# restore context->Rbx
308	mov	%rbp,160($context)	# restore context->Rbp
309	mov	%r12,216($context)	# restore context->R12
310	mov	%r14,232($context)	# restore context->R14
311	mov	%r15,240($context)	# restore context->R15
312
313.Lin_prologue:
314	mov	8(%rax),%rdi
315	mov	16(%rax),%rsi
316	mov	%rax,152($context)	# restore context->Rsp
317	mov	%rsi,168($context)	# restore context->Rsi
318	mov	%rdi,176($context)	# restore context->Rdi
319
320	mov	40($disp),%rdi		# disp->ContextRecord
321	mov	$context,%rsi		# context
322	mov	\$154,%ecx		# sizeof(CONTEXT)
323	.long	0xa548f3fc		# cld; rep movsq
324
325	mov	$disp,%rsi
326	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
327	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
328	mov	0(%rsi),%r8		# arg3, disp->ControlPc
329	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
330	mov	40(%rsi),%r10		# disp->ContextRecord
331	lea	56(%rsi),%r11		# &disp->HandlerData
332	lea	24(%rsi),%r12		# &disp->EstablisherFrame
333	mov	%r10,32(%rsp)		# arg5
334	mov	%r11,40(%rsp)		# arg6
335	mov	%r12,48(%rsp)		# arg7
336	mov	%rcx,56(%rsp)		# arg8, (NULL)
337	call	*__imp_RtlVirtualUnwind(%rip)
338
339	mov	\$1,%eax		# ExceptionContinueSearch
340	add	\$64,%rsp
341	popfq
342	pop	%r15
343	pop	%r14
344	pop	%r13
345	pop	%r12
346	pop	%rbp
347	pop	%rbx
348	pop	%rdi
349	pop	%rsi
350	ret
351.size	se_handler,.-se_handler
352
353.section	.pdata
354.align	4
355	.rva	.LSEH_begin_md5_block_asm_data_order
356	.rva	.LSEH_end_md5_block_asm_data_order
357	.rva	.LSEH_info_md5_block_asm_data_order
358
359.section	.xdata
360.align	8
361.LSEH_info_md5_block_asm_data_order:
362	.byte	9,0,0,0
363	.rva	se_handler
364___
365}
366
367print $code;
368
369close STDOUT;
370