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# sha1_block procedure for x86_64.
11#
12# It was brought to my attention that on EM64T compiler-generated code
13# was far behind 32-bit assembler implementation. This is unlike on
14# Opteron where compiler-generated code was only 15% behind 32-bit
15# assembler, which originally made it hard to motivate the effort.
16# There was suggestion to mechanically translate 32-bit code, but I
17# dismissed it, reasoning that x86_64 offers enough register bank
18# capacity to fully utilize SHA-1 parallelism. Therefore this fresh
19# implementation:-) However! While 64-bit code does performs better
20# on Opteron, I failed to beat 32-bit assembler on EM64T core. Well,
21# x86_64 does offer larger *addressable* bank, but out-of-order core
22# reaches for even more registers through dynamic aliasing, and EM64T
23# core must have managed to run-time optimize even 32-bit code just as
24# good as 64-bit one. Performance improvement is summarized in the
25# following table:
26#
27#		gcc 3.4		32-bit asm	cycles/byte
28# Opteron	+45%		+20%		6.8
29# Xeon P4	+65%		+0%		9.9
30# Core2		+60%		+10%		7.0
31
32#
33# OpenSolaris OS modifications
34#
35# Sun elects to use this software under the BSD license.
36#
37# This source originates from OpenSSL file sha1-x86_64.pl at
38# ftp://ftp.openssl.org/snapshot/openssl-0.9.8-stable-SNAP-20080131.tar.gz
39# (presumably for future OpenSSL release 0.9.8h), with these changes:
40#
41# 1. Added perl "use strict" and declared variables.
42#
43# 2. Added OpenSolaris ENTRY_NP/SET_SIZE macros from
44# /usr/include/sys/asm_linkage.h, .ident keywords, and lint(1B) guards.
45#
46# 3. Removed x86_64-xlate.pl script (not needed for as(1) or gas(1) assemblers).
47#
48
49use strict;
50my ($code, $ctx, $inp, $num, $xi, $t0, $t1, $i, @V, $A, $B, $C, $D, $E, $T);
51my $output = shift;
52open STDOUT,">$output";
53
54
55#
56# void sha1_block_data_order(SHA1_CTX *ctx, const void *inpp, size_t blocks);
57#
58
59# Arguments:
60$ctx="%rdi";	# 1st arg
61$inp="%rsi";	# 2nd arg
62$num="%rdx";	# 3rd arg
63
64# reassign arguments in order to produce more compact code
65$ctx="%r8";
66$inp="%r9";
67$num="%r10";
68
69# Temporaries:
70$xi="%eax";
71$t0="%ebx";
72$t1="%ecx";
73# State information from SHA-1 context:
74$A="%edx";
75$B="%esi";
76$C="%edi";
77$D="%ebp";
78$E="%r11d";
79# Temporary:
80$T="%r12d";
81
82@V=($A,$B,$C,$D,$E,$T);
83
84sub PROLOGUE {
85my $func=shift;
86$code.=<<___;
87ENTRY_NP($func)
88	push	%rbx
89	push	%rbp
90	push	%r12
91	mov	%rsp,%rax
92	mov	%rdi,$ctx	# reassigned argument
93	sub	\$`8+16*4`,%rsp
94	mov	%rsi,$inp	# reassigned argument
95	and	\$-64,%rsp
96	mov	%rdx,$num	# reassigned argument
97	mov	%rax,`16*4`(%rsp)
98
99	mov	0($ctx),$A
100	mov	4($ctx),$B
101	mov	8($ctx),$C
102	mov	12($ctx),$D
103	mov	16($ctx),$E
104___
105}
106
107sub EPILOGUE {
108my $func=shift;
109$code.=<<___;
110	mov	`16*4`(%rsp),%rsp
111	pop	%r12
112	pop	%rbp
113	pop	%rbx
114	ret
115SET_SIZE($func)
116___
117}
118
119sub BODY_00_19 {
120my ($i,$a,$b,$c,$d,$e,$f,$host)=@_;
121my $j=$i+1;
122$code.=<<___ if ($i==0);
123	mov	`4*$i`($inp),$xi
124	`"bswap	$xi"	if(!defined($host))`
125	mov	$xi,`4*$i`(%rsp)
126___
127$code.=<<___ if ($i<15);
128	lea	0x5a827999($xi,$e),$f
129	mov	$c,$t0
130	mov	`4*$j`($inp),$xi
131	mov	$a,$e
132	xor	$d,$t0
133	`"bswap	$xi"	if(!defined($host))`
134	rol	\$5,$e
135	and	$b,$t0
136	mov	$xi,`4*$j`(%rsp)
137	add	$e,$f
138	xor	$d,$t0
139	rol	\$30,$b
140	add	$t0,$f
141___
142$code.=<<___ if ($i>=15);
143	lea	0x5a827999($xi,$e),$f
144	mov	`4*($j%16)`(%rsp),$xi
145	mov	$c,$t0
146	mov	$a,$e
147	xor	`4*(($j+2)%16)`(%rsp),$xi
148	xor	$d,$t0
149	rol	\$5,$e
150	xor	`4*(($j+8)%16)`(%rsp),$xi
151	and	$b,$t0
152	add	$e,$f
153	xor	`4*(($j+13)%16)`(%rsp),$xi
154	xor	$d,$t0
155	rol	\$30,$b
156	add	$t0,$f
157	rol	\$1,$xi
158	mov	$xi,`4*($j%16)`(%rsp)
159___
160}
161
162sub BODY_20_39 {
163my ($i,$a,$b,$c,$d,$e,$f)=@_;
164my $j=$i+1;
165my $K=($i<40)?"0x6ed9eba1":"-0x359d3e2a";
166$code.=<<___ if ($i<79);
167	lea	$K($xi,$e),$f
168	mov	`4*($j%16)`(%rsp),$xi
169	mov	$c,$t0
170	mov	$a,$e
171	xor	`4*(($j+2)%16)`(%rsp),$xi
172	xor	$b,$t0
173	rol	\$5,$e
174	xor	`4*(($j+8)%16)`(%rsp),$xi
175	xor	$d,$t0
176	add	$e,$f
177	xor	`4*(($j+13)%16)`(%rsp),$xi
178	rol	\$30,$b
179	add	$t0,$f
180	rol	\$1,$xi
181___
182$code.=<<___ if ($i<76);
183	mov	$xi,`4*($j%16)`(%rsp)
184___
185$code.=<<___ if ($i==79);
186	lea	$K($xi,$e),$f
187	mov	$c,$t0
188	mov	$a,$e
189	xor	$b,$t0
190	rol	\$5,$e
191	xor	$d,$t0
192	add	$e,$f
193	rol	\$30,$b
194	add	$t0,$f
195___
196}
197
198sub BODY_40_59 {
199my ($i,$a,$b,$c,$d,$e,$f)=@_;
200my $j=$i+1;
201$code.=<<___;
202	lea	-0x70e44324($xi,$e),$f
203	mov	`4*($j%16)`(%rsp),$xi
204	mov	$b,$t0
205	mov	$b,$t1
206	xor	`4*(($j+2)%16)`(%rsp),$xi
207	mov	$a,$e
208	and	$c,$t0
209	xor	`4*(($j+8)%16)`(%rsp),$xi
210	or	$c,$t1
211	rol	\$5,$e
212	xor	`4*(($j+13)%16)`(%rsp),$xi
213	and	$d,$t1
214	add	$e,$f
215	rol	\$1,$xi
216	or	$t1,$t0
217	rol	\$30,$b
218	mov	$xi,`4*($j%16)`(%rsp)
219	add	$t0,$f
220___
221}
222
223
224#
225# Execution begins here
226#
227
228$code=<<___;
229#if defined(lint) || defined(__lint)
230#include <sys/stdint.h>
231#include <sys/sha1.h>
232
233/* ARGSUSED */
234void
235sha1_block_data_order(SHA1_CTX *ctx, const void *inpp, size_t blocks)
236{
237}
238
239#else
240#include <sys/asm_linkage.h>
241___
242
243
244&PROLOGUE("sha1_block_data_order");
245$code.=".align	4\n.Lloop:\n";
246for($i=0;$i<20;$i++)	{ &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
247for(;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
248for(;$i<60;$i++)	{ &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
249for(;$i<80;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
250$code.=<<___;
251	/ Update and save state information in SHA-1 context
252	add	0($ctx),$E
253	add	4($ctx),$T
254	add	8($ctx),$A
255	add	12($ctx),$B
256	add	16($ctx),$C
257	mov	$E,0($ctx)
258	mov	$T,4($ctx)
259	mov	$A,8($ctx)
260	mov	$B,12($ctx)
261	mov	$C,16($ctx)
262
263	xchg	$E,$A	# mov	$E,$A
264	xchg	$T,$B	# mov	$T,$B
265	xchg	$E,$C	# mov	$A,$C
266	xchg	$T,$D	# mov	$B,$D
267			# mov	$C,$E
268	lea	`16*4`($inp),$inp
269	sub	\$1,$num
270	jnz	.Lloop
271___
272&EPILOGUE("sha1_block_data_order");
273$code.=<<___;
274.asciz	"SHA1 block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
275
276#endif /* lint || __lint */
277___
278
279####################################################################
280
281$code =~ s/\`([^\`]*)\`/eval $1/gem;
282print $code;
283close STDOUT;
284