sha512-mips.pl revision 279264 
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# SHA2 block procedures for MIPS.
11
12# October 2010.
13#
14# SHA256 performance improvement on MIPS R5000 CPU is ~27% over gcc-
15# generated code in o32 build and ~55% in n32/64 build. SHA512 [which
16# for now can only be compiled for MIPS64 ISA] improvement is modest
17# ~17%, but it comes for free, because it's same instruction sequence.
18# Improvement coefficients are for aligned input.
19
20######################################################################
21# There is a number of MIPS ABI in use, O32 and N32/64 are most
22# widely used. Then there is a new contender: NUBI. It appears that if
23# one picks the latter, it's possible to arrange code in ABI neutral
24# manner. Therefore let's stick to NUBI register layout:
25#
26($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
27($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
28($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
29($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
30#
31# The return value is placed in $a0. Following coding rules facilitate
32# interoperability:
33#
34# - never ever touch $tp, "thread pointer", former $gp [o32 can be
35#   excluded from the rule, because it's specified volatile];
36# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
37#   old code];
38# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
39#
40# For reference here is register layout for N32/64 MIPS ABIs:
41#
42# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
43# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
44# ($t0,$t1,$t2,$t3,$t8,$t9)=map("\$$_",(12..15,24,25));
45# ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
46# ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
47#
48$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
49
50if ($flavour =~ /64|n32/i) {
51	$PTR_ADD="dadd";	# incidentally works even on n32
52	$PTR_SUB="dsub";	# incidentally works even on n32
53	$REG_S="sd";
54	$REG_L="ld";
55	$PTR_SLL="dsll";	# incidentally works even on n32
56	$SZREG=8;
57} else {
58	$PTR_ADD="add";
59	$PTR_SUB="sub";
60	$REG_S="sw";
61	$REG_L="lw";
62	$PTR_SLL="sll";
63	$SZREG=4;
64}
65$pf = ($flavour =~ /nubi/i) ? $t0 : $t2;
66#
67# <appro@openssl.org>
68#
69######################################################################
70
71$big_endian=(`echo MIPSEL | $ENV{CC} -E -`=~/MIPSEL/)?1:0 if ($ENV{CC});
72
73for (@ARGV) {	$output=$_ if (/^\w[\w\-]*\.\w+$/);	}
74open STDOUT,">$output";
75
76if (!defined($big_endian)) { $big_endian=(unpack('L',pack('N',1))==1); }
77
78if ($output =~ /512/) {
79	$label="512";
80	$SZ=8;
81	$LD="ld";		# load from memory
82	$ST="sd";		# store to memory
83	$SLL="dsll";		# shift left logical
84	$SRL="dsrl";		# shift right logical
85	$ADDU="daddu";
86	@Sigma0=(28,34,39);
87	@Sigma1=(14,18,41);
88	@sigma0=( 7, 1, 8);	# right shift first
89	@sigma1=( 6,19,61);	# right shift first
90	$lastK=0x817;
91	$rounds=80;
92} else {
93	$label="256";
94	$SZ=4;
95	$LD="lw";		# load from memory
96	$ST="sw";		# store to memory
97	$SLL="sll";		# shift left logical
98	$SRL="srl";		# shift right logical
99	$ADDU="addu";
100	@Sigma0=( 2,13,22);
101	@Sigma1=( 6,11,25);
102	@sigma0=( 3, 7,18);	# right shift first
103	@sigma1=(10,17,19);	# right shift first
104	$lastK=0x8f2;
105	$rounds=64;
106}
107
108$MSB = $big_endian ? 0 : ($SZ-1);
109$LSB = ($SZ-1)&~$MSB;
110
111@V=($A,$B,$C,$D,$E,$F,$G,$H)=map("\$$_",(1,2,3,7,24,25,30,31));
112@X=map("\$$_",(8..23));
113
114$ctx=$a0;
115$inp=$a1;
116$len=$a2;	$Ktbl=$len;
117
118sub BODY_00_15 {
119my ($i,$a,$b,$c,$d,$e,$f,$g,$h)=@_;
120my ($T1,$tmp0,$tmp1,$tmp2)=(@X[4],@X[5],@X[6],@X[7]);
121
122$code.=<<___ if ($i<15);
123	${LD}l	@X[1],`($i+1)*$SZ+$MSB`($inp)
124	${LD}r	@X[1],`($i+1)*$SZ+$LSB`($inp)
125___
126$code.=<<___	if (!$big_endian && $i<16 && $SZ==4);
127	srl	$tmp0,@X[0],24		# byte swap($i)
128	srl	$tmp1,@X[0],8
129	andi	$tmp2,@X[0],0xFF00
130	sll	@X[0],@X[0],24
131	andi	$tmp1,0xFF00
132	sll	$tmp2,$tmp2,8
133	or	@X[0],$tmp0
134	or	$tmp1,$tmp2
135	or	@X[0],$tmp1
136___
137$code.=<<___	if (!$big_endian && $i<16 && $SZ==8);
138	ori	$tmp0,$zero,0xFF
139	dsll	$tmp2,$tmp0,32
140	or	$tmp0,$tmp2		# 0x000000FF000000FF
141	and	$tmp1,@X[0],$tmp0	# byte swap($i)
142	dsrl	$tmp2,@X[0],24
143	dsll	$tmp1,24
144	and	$tmp2,$tmp0
145	dsll	$tmp0,8			# 0x0000FF000000FF00
146	or	$tmp1,$tmp2
147	and	$tmp2,@X[0],$tmp0
148	dsrl	@X[0],8
149	dsll	$tmp2,8
150	and	@X[0],$tmp0
151	or	$tmp1,$tmp2
152	or	@X[0],$tmp1
153	dsrl	$tmp1,@X[0],32
154	dsll	@X[0],32
155	or	@X[0],$tmp1
156___
157$code.=<<___;
158	$ADDU	$T1,$X[0],$h			# $i
159	$SRL	$h,$e,@Sigma1[0]
160	xor	$tmp2,$f,$g
161	$SLL	$tmp1,$e,`$SZ*8-@Sigma1[2]`
162	and	$tmp2,$e
163	$SRL	$tmp0,$e,@Sigma1[1]
164	xor	$h,$tmp1
165	$SLL	$tmp1,$e,`$SZ*8-@Sigma1[1]`
166	xor	$h,$tmp0
167	$SRL	$tmp0,$e,@Sigma1[2]
168	xor	$h,$tmp1
169	$SLL	$tmp1,$e,`$SZ*8-@Sigma1[0]`
170	xor	$h,$tmp0
171	xor	$tmp2,$g			# Ch(e,f,g)
172	xor	$tmp0,$tmp1,$h			# Sigma1(e)
173
174	$SRL	$h,$a,@Sigma0[0]
175	$ADDU	$T1,$tmp2
176	$LD	$tmp2,`$i*$SZ`($Ktbl)		# K[$i]
177	$SLL	$tmp1,$a,`$SZ*8-@Sigma0[2]`
178	$ADDU	$T1,$tmp0
179	$SRL	$tmp0,$a,@Sigma0[1]
180	xor	$h,$tmp1
181	$SLL	$tmp1,$a,`$SZ*8-@Sigma0[1]`
182	xor	$h,$tmp0
183	$SRL	$tmp0,$a,@Sigma0[2]
184	xor	$h,$tmp1
185	$SLL	$tmp1,$a,`$SZ*8-@Sigma0[0]`
186	xor	$h,$tmp0
187	$ST	@X[0],`($i%16)*$SZ`($sp)	# offload to ring buffer
188	xor	$h,$tmp1			# Sigma0(a)
189
190	or	$tmp0,$a,$b
191	and	$tmp1,$a,$b
192	and	$tmp0,$c
193	or	$tmp1,$tmp0			# Maj(a,b,c)
194	$ADDU	$T1,$tmp2			# +=K[$i]
195	$ADDU	$h,$tmp1
196
197	$ADDU	$d,$T1
198	$ADDU	$h,$T1
199___
200$code.=<<___ if ($i>=13);
201	$LD	@X[3],`(($i+3)%16)*$SZ`($sp)	# prefetch from ring buffer
202___
203}
204
205sub BODY_16_XX {
206my $i=@_[0];
207my ($tmp0,$tmp1,$tmp2,$tmp3)=(@X[4],@X[5],@X[6],@X[7]);
208
209$code.=<<___;
210	$SRL	$tmp2,@X[1],@sigma0[0]		# Xupdate($i)
211	$ADDU	@X[0],@X[9]			# +=X[i+9]
212	$SLL	$tmp1,@X[1],`$SZ*8-@sigma0[2]`
213	$SRL	$tmp0,@X[1],@sigma0[1]
214	xor	$tmp2,$tmp1
215	$SLL	$tmp1,`@sigma0[2]-@sigma0[1]`
216	xor	$tmp2,$tmp0
217	$SRL	$tmp0,@X[1],@sigma0[2]
218	xor	$tmp2,$tmp1
219
220	$SRL	$tmp3,@X[14],@sigma1[0]
221	xor	$tmp2,$tmp0			# sigma0(X[i+1])
222	$SLL	$tmp1,@X[14],`$SZ*8-@sigma1[2]`
223	$ADDU	@X[0],$tmp2
224	$SRL	$tmp0,@X[14],@sigma1[1]
225	xor	$tmp3,$tmp1
226	$SLL	$tmp1,`@sigma1[2]-@sigma1[1]`
227	xor	$tmp3,$tmp0
228	$SRL	$tmp0,@X[14],@sigma1[2]
229	xor	$tmp3,$tmp1
230
231	xor	$tmp3,$tmp0			# sigma1(X[i+14])
232	$ADDU	@X[0],$tmp3
233___
234	&BODY_00_15(@_);
235}
236
237$FRAMESIZE=16*$SZ+16*$SZREG;
238$SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc0fff008 : 0xc0ff0000;
239
240$code.=<<___;
241#ifdef OPENSSL_FIPSCANISTER
242# include <openssl/fipssyms.h>
243#endif
244
245.text
246.set	noat
247#if !defined(__vxworks) || defined(__pic__)
248.option	pic2
249#endif
250
251.align	5
252.globl	sha${label}_block_data_order
253.ent	sha${label}_block_data_order
254sha${label}_block_data_order:
255	.frame	$sp,$FRAMESIZE,$ra
256	.mask	$SAVED_REGS_MASK,-$SZREG
257	.set	noreorder
258___
259$code.=<<___ if ($flavour =~ /o32/i);	# o32 PIC-ification
260	.cpload	$pf
261___
262$code.=<<___;
263	$PTR_SUB $sp,$FRAMESIZE
264	$REG_S	$ra,$FRAMESIZE-1*$SZREG($sp)
265	$REG_S	$fp,$FRAMESIZE-2*$SZREG($sp)
266	$REG_S	$s11,$FRAMESIZE-3*$SZREG($sp)
267	$REG_S	$s10,$FRAMESIZE-4*$SZREG($sp)
268	$REG_S	$s9,$FRAMESIZE-5*$SZREG($sp)
269	$REG_S	$s8,$FRAMESIZE-6*$SZREG($sp)
270	$REG_S	$s7,$FRAMESIZE-7*$SZREG($sp)
271	$REG_S	$s6,$FRAMESIZE-8*$SZREG($sp)
272	$REG_S	$s5,$FRAMESIZE-9*$SZREG($sp)
273	$REG_S	$s4,$FRAMESIZE-10*$SZREG($sp)
274___
275$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
276	$REG_S	$s3,$FRAMESIZE-11*$SZREG($sp)
277	$REG_S	$s2,$FRAMESIZE-12*$SZREG($sp)
278	$REG_S	$s1,$FRAMESIZE-13*$SZREG($sp)
279	$REG_S	$s0,$FRAMESIZE-14*$SZREG($sp)
280	$REG_S	$gp,$FRAMESIZE-15*$SZREG($sp)
281___
282$code.=<<___;
283	$PTR_SLL @X[15],$len,`log(16*$SZ)/log(2)`
284___
285$code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
286	.cplocal	$Ktbl
287	.cpsetup	$pf,$zero,sha${label}_block_data_order
288___
289$code.=<<___;
290	.set	reorder
291	la	$Ktbl,K${label}		# PIC-ified 'load address'
292
293	$LD	$A,0*$SZ($ctx)		# load context
294	$LD	$B,1*$SZ($ctx)
295	$LD	$C,2*$SZ($ctx)
296	$LD	$D,3*$SZ($ctx)
297	$LD	$E,4*$SZ($ctx)
298	$LD	$F,5*$SZ($ctx)
299	$LD	$G,6*$SZ($ctx)
300	$LD	$H,7*$SZ($ctx)
301
302	$PTR_ADD @X[15],$inp		# pointer to the end of input
303	$REG_S	@X[15],16*$SZ($sp)
304	b	.Loop
305
306.align	5
307.Loop:
308	${LD}l	@X[0],$MSB($inp)
309	${LD}r	@X[0],$LSB($inp)
310___
311for ($i=0;$i<16;$i++)
312{ &BODY_00_15($i,@V); unshift(@V,pop(@V)); push(@X,shift(@X)); }
313$code.=<<___;
314	b	.L16_xx
315.align	4
316.L16_xx:
317___
318for (;$i<32;$i++)
319{ &BODY_16_XX($i,@V); unshift(@V,pop(@V)); push(@X,shift(@X)); }
320$code.=<<___;
321	and	@X[6],0xfff
322	li	@X[7],$lastK
323	.set	noreorder
324	bne	@X[6],@X[7],.L16_xx
325	$PTR_ADD $Ktbl,16*$SZ		# Ktbl+=16
326
327	$REG_L	@X[15],16*$SZ($sp)	# restore pointer to the end of input
328	$LD	@X[0],0*$SZ($ctx)
329	$LD	@X[1],1*$SZ($ctx)
330	$LD	@X[2],2*$SZ($ctx)
331	$PTR_ADD $inp,16*$SZ
332	$LD	@X[3],3*$SZ($ctx)
333	$ADDU	$A,@X[0]
334	$LD	@X[4],4*$SZ($ctx)
335	$ADDU	$B,@X[1]
336	$LD	@X[5],5*$SZ($ctx)
337	$ADDU	$C,@X[2]
338	$LD	@X[6],6*$SZ($ctx)
339	$ADDU	$D,@X[3]
340	$LD	@X[7],7*$SZ($ctx)
341	$ADDU	$E,@X[4]
342	$ST	$A,0*$SZ($ctx)
343	$ADDU	$F,@X[5]
344	$ST	$B,1*$SZ($ctx)
345	$ADDU	$G,@X[6]
346	$ST	$C,2*$SZ($ctx)
347	$ADDU	$H,@X[7]
348	$ST	$D,3*$SZ($ctx)
349	$ST	$E,4*$SZ($ctx)
350	$ST	$F,5*$SZ($ctx)
351	$ST	$G,6*$SZ($ctx)
352	$ST	$H,7*$SZ($ctx)
353
354	bne	$inp,@X[15],.Loop
355	$PTR_SUB $Ktbl,`($rounds-16)*$SZ`	# rewind $Ktbl
356
357	$REG_L	$ra,$FRAMESIZE-1*$SZREG($sp)
358	$REG_L	$fp,$FRAMESIZE-2*$SZREG($sp)
359	$REG_L	$s11,$FRAMESIZE-3*$SZREG($sp)
360	$REG_L	$s10,$FRAMESIZE-4*$SZREG($sp)
361	$REG_L	$s9,$FRAMESIZE-5*$SZREG($sp)
362	$REG_L	$s8,$FRAMESIZE-6*$SZREG($sp)
363	$REG_L	$s7,$FRAMESIZE-7*$SZREG($sp)
364	$REG_L	$s6,$FRAMESIZE-8*$SZREG($sp)
365	$REG_L	$s5,$FRAMESIZE-9*$SZREG($sp)
366	$REG_L	$s4,$FRAMESIZE-10*$SZREG($sp)
367___
368$code.=<<___ if ($flavour =~ /nubi/i);
369	$REG_L	$s3,$FRAMESIZE-11*$SZREG($sp)
370	$REG_L	$s2,$FRAMESIZE-12*$SZREG($sp)
371	$REG_L	$s1,$FRAMESIZE-13*$SZREG($sp)
372	$REG_L	$s0,$FRAMESIZE-14*$SZREG($sp)
373	$REG_L	$gp,$FRAMESIZE-15*$SZREG($sp)
374___
375$code.=<<___;
376	jr	$ra
377	$PTR_ADD $sp,$FRAMESIZE
378.end	sha${label}_block_data_order
379
380.rdata
381.align	5
382K${label}:
383___
384if ($SZ==4) {
385$code.=<<___;
386	.word	0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5
387	.word	0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5
388	.word	0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3
389	.word	0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174
390	.word	0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc
391	.word	0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da
392	.word	0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7
393	.word	0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967
394	.word	0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13
395	.word	0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85
396	.word	0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3
397	.word	0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070
398	.word	0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5
399	.word	0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3
400	.word	0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208
401	.word	0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2
402___
403} else {
404$code.=<<___;
405	.dword	0x428a2f98d728ae22, 0x7137449123ef65cd
406	.dword	0xb5c0fbcfec4d3b2f, 0xe9b5dba58189dbbc
407	.dword	0x3956c25bf348b538, 0x59f111f1b605d019
408	.dword	0x923f82a4af194f9b, 0xab1c5ed5da6d8118
409	.dword	0xd807aa98a3030242, 0x12835b0145706fbe
410	.dword	0x243185be4ee4b28c, 0x550c7dc3d5ffb4e2
411	.dword	0x72be5d74f27b896f, 0x80deb1fe3b1696b1
412	.dword	0x9bdc06a725c71235, 0xc19bf174cf692694
413	.dword	0xe49b69c19ef14ad2, 0xefbe4786384f25e3
414	.dword	0x0fc19dc68b8cd5b5, 0x240ca1cc77ac9c65
415	.dword	0x2de92c6f592b0275, 0x4a7484aa6ea6e483
416	.dword	0x5cb0a9dcbd41fbd4, 0x76f988da831153b5
417	.dword	0x983e5152ee66dfab, 0xa831c66d2db43210
418	.dword	0xb00327c898fb213f, 0xbf597fc7beef0ee4
419	.dword	0xc6e00bf33da88fc2, 0xd5a79147930aa725
420	.dword	0x06ca6351e003826f, 0x142929670a0e6e70
421	.dword	0x27b70a8546d22ffc, 0x2e1b21385c26c926
422	.dword	0x4d2c6dfc5ac42aed, 0x53380d139d95b3df
423	.dword	0x650a73548baf63de, 0x766a0abb3c77b2a8
424	.dword	0x81c2c92e47edaee6, 0x92722c851482353b
425	.dword	0xa2bfe8a14cf10364, 0xa81a664bbc423001
426	.dword	0xc24b8b70d0f89791, 0xc76c51a30654be30
427	.dword	0xd192e819d6ef5218, 0xd69906245565a910
428	.dword	0xf40e35855771202a, 0x106aa07032bbd1b8
429	.dword	0x19a4c116b8d2d0c8, 0x1e376c085141ab53
430	.dword	0x2748774cdf8eeb99, 0x34b0bcb5e19b48a8
431	.dword	0x391c0cb3c5c95a63, 0x4ed8aa4ae3418acb
432	.dword	0x5b9cca4f7763e373, 0x682e6ff3d6b2b8a3
433	.dword	0x748f82ee5defb2fc, 0x78a5636f43172f60
434	.dword	0x84c87814a1f0ab72, 0x8cc702081a6439ec
435	.dword	0x90befffa23631e28, 0xa4506cebde82bde9
436	.dword	0xbef9a3f7b2c67915, 0xc67178f2e372532b
437	.dword	0xca273eceea26619c, 0xd186b8c721c0c207
438	.dword	0xeada7dd6cde0eb1e, 0xf57d4f7fee6ed178
439	.dword	0x06f067aa72176fba, 0x0a637dc5a2c898a6
440	.dword	0x113f9804bef90dae, 0x1b710b35131c471b
441	.dword	0x28db77f523047d84, 0x32caab7b40c72493
442	.dword	0x3c9ebe0a15c9bebc, 0x431d67c49c100d4c
443	.dword	0x4cc5d4becb3e42b6, 0x597f299cfc657e2a
444	.dword	0x5fcb6fab3ad6faec, 0x6c44198c4a475817
445___
446}
447$code.=<<___;
448.asciiz	"SHA${label} for MIPS, CRYPTOGAMS by <appro\@openssl.org>"
449.align	5
450
451___
452
453$code =~ s/\`([^\`]*)\`/eval $1/gem;
454print $code;
455close STDOUT;
456