xref: /openbsd/lib/libcrypto/sha/asm/sha1-mips.pl (revision 5af055cd) 
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 MIPS.
11
12# Performance improvement is 30% on unaligned input. The "secret" is
13# to deploy lwl/lwr pair to load unaligned input. One could have
14# vectorized Xupdate on MIPSIII/IV, but the goal was to code MIPS32-
15# compatible subroutine. There is room for minor optimization on
16# little-endian platforms...
17
18######################################################################
19# There is a number of MIPS ABI in use, O32 and N32/64 are most
20# widely used. Then there is a new contender: NUBI. It appears that if
21# one picks the latter, it's possible to arrange code in ABI neutral
22# manner. Therefore let's stick to NUBI register layout:
23#
24($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
25($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
26($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
27($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
28#
29# The return value is placed in $a0. Following coding rules facilitate
30# interoperability:
31#
32# - never ever touch $tp, "thread pointer", former $gp;
33# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
34#   old code];
35# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
36#
37# For reference here is register layout for N32/64 MIPS ABIs:
38#
39# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
40# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
41# ($t0,$t1,$t2,$t3,$t8,$t9)=map("\$$_",(12..15,24,25));
42# ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
43# ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
44#
45$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
46
47if ($flavour =~ /64|n32/i) {
48	$PTR_ADD="dadd";	# incidentally works even on n32
49	$PTR_SUB="dsub";	# incidentally works even on n32
50	$REG_S="sd";
51	$REG_L="ld";
52	$PTR_SLL="dsll";	# incidentally works even on n32
53	$SZREG=8;
54} else {
55	$PTR_ADD="add";
56	$PTR_SUB="sub";
57	$REG_S="sw";
58	$REG_L="lw";
59	$PTR_SLL="sll";
60	$SZREG=4;
61}
62#
63# <appro@openssl.org>
64#
65######################################################################
66
67$big_endian=(`echo MIPSEL | $ENV{CC} -E -P -`=~/MIPSEL/)?1:0;
68
69for (@ARGV) {	$output=$_ if (/^\w[\w\-]*\.\w+$/);   }
70open STDOUT,">$output";
71
72if (!defined($big_endian))
73            {   $big_endian=(unpack('L',pack('N',1))==1);   }
74
75# offsets of the Most and Least Significant Bytes
76$MSB=$big_endian?0:3;
77$LSB=3&~$MSB;
78
79@X=map("\$$_",(8..23));	# a4-a7,s0-s11
80
81$ctx=$a0;
82$inp=$a1;
83$num=$a2;
84$A="\$1";
85$B="\$2";
86$C="\$3";
87$D="\$7";
88$E="\$24";	@V=($A,$B,$C,$D,$E);
89$t0="\$25";
90$t1=$num;	# $num is offloaded to stack
91$t2="\$30";	# fp
92$K="\$31";	# ra
93
94sub BODY_00_14 {
95my ($i,$a,$b,$c,$d,$e)=@_;
96my $j=$i+1;
97$code.=<<___	if (!$big_endian);
98	srl	$t0,@X[$i],24	# byte swap($i)
99	srl	$t1,@X[$i],8
100	andi	$t2,@X[$i],0xFF00
101	sll	@X[$i],@X[$i],24
102	andi	$t1,0xFF00
103	sll	$t2,$t2,8
104	or	@X[$i],$t0
105	or	$t1,$t2
106	or	@X[$i],$t1
107___
108$code.=<<___;
109	 lwl	@X[$j],$j*4+$MSB($inp)
110	sll	$t0,$a,5	# $i
111	addu	$e,$K
112	 lwr	@X[$j],$j*4+$LSB($inp)
113	srl	$t1,$a,27
114	addu	$e,$t0
115	xor	$t0,$c,$d
116	addu	$e,$t1
117	sll	$t2,$b,30
118	and	$t0,$b
119	srl	$b,$b,2
120	xor	$t0,$d
121	addu	$e,@X[$i]
122	or	$b,$t2
123	addu	$e,$t0
124___
125}
126
127sub BODY_15_19 {
128my ($i,$a,$b,$c,$d,$e)=@_;
129my $j=$i+1;
130
131$code.=<<___	if (!$big_endian && $i==15);
132	srl	$t0,@X[$i],24	# byte swap($i)
133	srl	$t1,@X[$i],8
134	andi	$t2,@X[$i],0xFF00
135	sll	@X[$i],@X[$i],24
136	andi	$t1,0xFF00
137	sll	$t2,$t2,8
138	or	@X[$i],$t0
139	or	@X[$i],$t1
140	or	@X[$i],$t2
141___
142$code.=<<___;
143	 xor	@X[$j%16],@X[($j+2)%16]
144	sll	$t0,$a,5	# $i
145	addu	$e,$K
146	srl	$t1,$a,27
147	addu	$e,$t0
148	 xor	@X[$j%16],@X[($j+8)%16]
149	xor	$t0,$c,$d
150	addu	$e,$t1
151	 xor	@X[$j%16],@X[($j+13)%16]
152	sll	$t2,$b,30
153	and	$t0,$b
154	 srl	$t1,@X[$j%16],31
155	 addu	@X[$j%16],@X[$j%16]
156	srl	$b,$b,2
157	xor	$t0,$d
158	 or	@X[$j%16],$t1
159	addu	$e,@X[$i%16]
160	or	$b,$t2
161	addu	$e,$t0
162___
163}
164
165sub BODY_20_39 {
166my ($i,$a,$b,$c,$d,$e)=@_;
167my $j=$i+1;
168$code.=<<___ if ($i<79);
169	 xor	@X[$j%16],@X[($j+2)%16]
170	sll	$t0,$a,5	# $i
171	addu	$e,$K
172	srl	$t1,$a,27
173	addu	$e,$t0
174	 xor	@X[$j%16],@X[($j+8)%16]
175	xor	$t0,$c,$d
176	addu	$e,$t1
177	 xor	@X[$j%16],@X[($j+13)%16]
178	sll	$t2,$b,30
179	xor	$t0,$b
180	 srl	$t1,@X[$j%16],31
181	 addu	@X[$j%16],@X[$j%16]
182	srl	$b,$b,2
183	addu	$e,@X[$i%16]
184	 or	@X[$j%16],$t1
185	or	$b,$t2
186	addu	$e,$t0
187___
188$code.=<<___ if ($i==79);
189	 lw	@X[0],0($ctx)
190	sll	$t0,$a,5	# $i
191	addu	$e,$K
192	 lw	@X[1],4($ctx)
193	srl	$t1,$a,27
194	addu	$e,$t0
195	 lw	@X[2],8($ctx)
196	xor	$t0,$c,$d
197	addu	$e,$t1
198	 lw	@X[3],12($ctx)
199	sll	$t2,$b,30
200	xor	$t0,$b
201	 lw	@X[4],16($ctx)
202	srl	$b,$b,2
203	addu	$e,@X[$i%16]
204	or	$b,$t2
205	addu	$e,$t0
206___
207}
208
209sub BODY_40_59 {
210my ($i,$a,$b,$c,$d,$e)=@_;
211my $j=$i+1;
212$code.=<<___ if ($i<79);
213	 xor	@X[$j%16],@X[($j+2)%16]
214	sll	$t0,$a,5	# $i
215	addu	$e,$K
216	srl	$t1,$a,27
217	addu	$e,$t0
218	 xor	@X[$j%16],@X[($j+8)%16]
219	and	$t0,$c,$d
220	addu	$e,$t1
221	 xor	@X[$j%16],@X[($j+13)%16]
222	sll	$t2,$b,30
223	addu	$e,$t0
224	 srl	$t1,@X[$j%16],31
225	xor	$t0,$c,$d
226	 addu	@X[$j%16],@X[$j%16]
227	and	$t0,$b
228	srl	$b,$b,2
229	 or	@X[$j%16],$t1
230	addu	$e,@X[$i%16]
231	or	$b,$t2
232	addu	$e,$t0
233___
234}
235
236$FRAMESIZE=16;	# large enough to accommodate NUBI saved registers
237$SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc0fff008 : 0xc0ff0000;
238
239$code=<<___;
240.text
241
242.set	noat
243.set	noreorder
244.align	5
245.globl	sha1_block_data_order
246.ent	sha1_block_data_order
247sha1_block_data_order:
248	.frame	$sp,$FRAMESIZE*$SZREG,$ra
249	.mask	$SAVED_REGS_MASK,-$SZREG
250	.set	noreorder
251	$PTR_SUB $sp,$FRAMESIZE*$SZREG
252	$REG_S	$ra,($FRAMESIZE-1)*$SZREG($sp)
253	$REG_S	$fp,($FRAMESIZE-2)*$SZREG($sp)
254	$REG_S	$s11,($FRAMESIZE-3)*$SZREG($sp)
255	$REG_S	$s10,($FRAMESIZE-4)*$SZREG($sp)
256	$REG_S	$s9,($FRAMESIZE-5)*$SZREG($sp)
257	$REG_S	$s8,($FRAMESIZE-6)*$SZREG($sp)
258	$REG_S	$s7,($FRAMESIZE-7)*$SZREG($sp)
259	$REG_S	$s6,($FRAMESIZE-8)*$SZREG($sp)
260	$REG_S	$s5,($FRAMESIZE-9)*$SZREG($sp)
261	$REG_S	$s4,($FRAMESIZE-10)*$SZREG($sp)
262___
263$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
264	$REG_S	$s3,($FRAMESIZE-11)*$SZREG($sp)
265	$REG_S	$s2,($FRAMESIZE-12)*$SZREG($sp)
266	$REG_S	$s1,($FRAMESIZE-13)*$SZREG($sp)
267	$REG_S	$s0,($FRAMESIZE-14)*$SZREG($sp)
268	$REG_S	$gp,($FRAMESIZE-15)*$SZREG($sp)
269___
270$code.=<<___;
271	$PTR_SLL $num,6
272	$PTR_ADD $num,$inp
273	$REG_S	$num,0($sp)
274	lw	$A,0($ctx)
275	lw	$B,4($ctx)
276	lw	$C,8($ctx)
277	lw	$D,12($ctx)
278	b	.Loop
279	lw	$E,16($ctx)
280.align	4
281.Loop:
282	.set	reorder
283	lwl	@X[0],$MSB($inp)
284	lui	$K,0x5a82
285	lwr	@X[0],$LSB($inp)
286	ori	$K,0x7999	# K_00_19
287___
288for ($i=0;$i<15;$i++)	{ &BODY_00_14($i,@V); unshift(@V,pop(@V)); }
289for (;$i<20;$i++)	{ &BODY_15_19($i,@V); unshift(@V,pop(@V)); }
290$code.=<<___;
291	lui	$K,0x6ed9
292	ori	$K,0xeba1	# K_20_39
293___
294for (;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
295$code.=<<___;
296	lui	$K,0x8f1b
297	ori	$K,0xbcdc	# K_40_59
298___
299for (;$i<60;$i++)	{ &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
300$code.=<<___;
301	lui	$K,0xca62
302	ori	$K,0xc1d6	# K_60_79
303___
304for (;$i<80;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
305$code.=<<___;
306	$PTR_ADD $inp,64
307	$REG_L	$num,0($sp)
308
309	addu	$A,$X[0]
310	addu	$B,$X[1]
311	sw	$A,0($ctx)
312	addu	$C,$X[2]
313	addu	$D,$X[3]
314	sw	$B,4($ctx)
315	addu	$E,$X[4]
316	sw	$C,8($ctx)
317	sw	$D,12($ctx)
318	sw	$E,16($ctx)
319	.set	noreorder
320	bne	$inp,$num,.Loop
321	nop
322
323	.set	noreorder
324	$REG_L	$ra,($FRAMESIZE-1)*$SZREG($sp)
325	$REG_L	$fp,($FRAMESIZE-2)*$SZREG($sp)
326	$REG_L	$s11,($FRAMESIZE-3)*$SZREG($sp)
327	$REG_L	$s10,($FRAMESIZE-4)*$SZREG($sp)
328	$REG_L	$s9,($FRAMESIZE-5)*$SZREG($sp)
329	$REG_L	$s8,($FRAMESIZE-6)*$SZREG($sp)
330	$REG_L	$s7,($FRAMESIZE-7)*$SZREG($sp)
331	$REG_L	$s6,($FRAMESIZE-8)*$SZREG($sp)
332	$REG_L	$s5,($FRAMESIZE-9)*$SZREG($sp)
333	$REG_L	$s4,($FRAMESIZE-10)*$SZREG($sp)
334___
335$code.=<<___ if ($flavour =~ /nubi/i);
336	$REG_L	$s3,($FRAMESIZE-11)*$SZREG($sp)
337	$REG_L	$s2,($FRAMESIZE-12)*$SZREG($sp)
338	$REG_L	$s1,($FRAMESIZE-13)*$SZREG($sp)
339	$REG_L	$s0,($FRAMESIZE-14)*$SZREG($sp)
340	$REG_L	$gp,($FRAMESIZE-15)*$SZREG($sp)
341___
342$code.=<<___;
343	jr	$ra
344	$PTR_ADD $sp,$FRAMESIZE*$SZREG
345.end	sha1_block_data_order
346.rdata
347.asciiz	"SHA1 for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
348___
349print $code;
350close STDOUT;
351