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# SHA256/512 block procedures for s390x.
11
12# April 2007.
13#
14# sha256_block_data_order is reportedly >3 times faster than gcc 3.3
15# generated code (must be a bug in compiler, as improvement is
16# "pathologically" high, in particular in comparison to other SHA
17# modules). But the real twist is that it detects if hardware support
18# for SHA256 is available and in such case utilizes it. Then the
19# performance can reach >6.5x of assembler one for larger chunks.
20#
21# sha512_block_data_order is ~70% faster than gcc 3.3 generated code.
22
23# January 2009.
24#
25# Add support for hardware SHA512 and reschedule instructions to
26# favour dual-issue z10 pipeline. Hardware SHA256/512 is ~4.7x faster
27# than software.
28
29# November 2010.
30#
31# Adapt for -m31 build. If kernel supports what's called "highgprs"
32# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
33# instructions and achieve "64-bit" performance even in 31-bit legacy
34# application context. The feature is not specific to any particular
35# processor, as long as it's "z-CPU". Latter implies that the code
36# remains z/Architecture specific. On z900 SHA256 was measured to
37# perform 2.4x and SHA512 - 13x better than code generated by gcc 4.3.
38
39$flavour = shift;
40
41if ($flavour =~ /3[12]/) {
42	$SIZE_T=4;
43	$g="";
44} else {
45	$SIZE_T=8;
46	$g="g";
47}
48
49$t0="%r0";
50$t1="%r1";
51$ctx="%r2";	$t2="%r2";
52$inp="%r3";
53$len="%r4";	# used as index in inner loop
54
55$A="%r5";
56$B="%r6";
57$C="%r7";
58$D="%r8";
59$E="%r9";
60$F="%r10";
61$G="%r11";
62$H="%r12";	@V=($A,$B,$C,$D,$E,$F,$G,$H);
63$tbl="%r13";
64$T1="%r14";
65$sp="%r15";
66
67while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
68open STDOUT,">$output";
69
70if ($output =~ /512/) {
71	$label="512";
72	$SZ=8;
73	$LD="lg";	# load from memory
74	$ST="stg";	# store to memory
75	$ADD="alg";	# add with memory operand
76	$ROT="rllg";	# rotate left
77	$SHR="srlg";	# logical right shift [see even at the end]
78	@Sigma0=(25,30,36);
79	@Sigma1=(23,46,50);
80	@sigma0=(56,63, 7);
81	@sigma1=( 3,45, 6);
82	$rounds=80;
83	$kimdfunc=3;	# 0 means unknown/unsupported/unimplemented/disabled
84} else {
85	$label="256";
86	$SZ=4;
87	$LD="llgf";	# load from memory
88	$ST="st";	# store to memory
89	$ADD="al";	# add with memory operand
90	$ROT="rll";	# rotate left
91	$SHR="srl";	# logical right shift
92	@Sigma0=(10,19,30);
93	@Sigma1=( 7,21,26);
94	@sigma0=(14,25, 3);
95	@sigma1=(13,15,10);
96	$rounds=64;
97	$kimdfunc=2;	# magic function code for kimd instruction
98}
99$Func="sha${label}_block_data_order";
100$Table="K${label}";
101$stdframe=16*$SIZE_T+4*8;
102$frame=$stdframe+16*$SZ;
103
104sub BODY_00_15 {
105my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
106
107$code.=<<___ if ($i<16);
108	$LD	$T1,`$i*$SZ`($inp)	### $i
109___
110$code.=<<___;
111	$ROT	$t0,$e,$Sigma1[0]
112	$ROT	$t1,$e,$Sigma1[1]
113	 lgr	$t2,$f
114	xgr	$t0,$t1
115	$ROT	$t1,$t1,`$Sigma1[2]-$Sigma1[1]`
116	 xgr	$t2,$g
117	$ST	$T1,`$stdframe+$SZ*($i%16)`($sp)
118	xgr	$t0,$t1			# Sigma1(e)
119	algr	$T1,$h			# T1+=h
120	 ngr	$t2,$e
121	 lgr	$t1,$a
122	algr	$T1,$t0			# T1+=Sigma1(e)
123	$ROT	$h,$a,$Sigma0[0]
124	 xgr	$t2,$g			# Ch(e,f,g)
125	$ADD	$T1,`$i*$SZ`($len,$tbl)	# T1+=K[i]
126	$ROT	$t0,$a,$Sigma0[1]
127	algr	$T1,$t2			# T1+=Ch(e,f,g)
128	 ogr	$t1,$b
129	xgr	$h,$t0
130	 lgr	$t2,$a
131	 ngr	$t1,$c
132	$ROT	$t0,$t0,`$Sigma0[2]-$Sigma0[1]`
133	xgr	$h,$t0			# h=Sigma0(a)
134	 ngr	$t2,$b
135	algr	$h,$T1			# h+=T1
136	 ogr	$t2,$t1			# Maj(a,b,c)
137	algr	$d,$T1			# d+=T1
138	algr	$h,$t2			# h+=Maj(a,b,c)
139___
140}
141
142sub BODY_16_XX {
143my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
144
145$code.=<<___;
146	$LD	$T1,`$stdframe+$SZ*(($i+1)%16)`($sp)	### $i
147	$LD	$t1,`$stdframe+$SZ*(($i+14)%16)`($sp)
148	$ROT	$t0,$T1,$sigma0[0]
149	$SHR	$T1,$sigma0[2]
150	$ROT	$t2,$t0,`$sigma0[1]-$sigma0[0]`
151	xgr	$T1,$t0
152	$ROT	$t0,$t1,$sigma1[0]
153	xgr	$T1,$t2					# sigma0(X[i+1])
154	$SHR	$t1,$sigma1[2]
155	$ADD	$T1,`$stdframe+$SZ*($i%16)`($sp)	# +=X[i]
156	xgr	$t1,$t0
157	$ROT	$t0,$t0,`$sigma1[1]-$sigma1[0]`
158	$ADD	$T1,`$stdframe+$SZ*(($i+9)%16)`($sp)	# +=X[i+9]
159	xgr	$t1,$t0				# sigma1(X[i+14])
160	algr	$T1,$t1				# +=sigma1(X[i+14])
161___
162	&BODY_00_15(@_);
163}
164
165$code.=<<___;
166.text
167.align	64
168.type	$Table,\@object
169$Table:
170___
171$code.=<<___ if ($SZ==4);
172	.long	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
173	.long	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
174	.long	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
175	.long	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
176	.long	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
177	.long	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
178	.long	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
179	.long	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
180	.long	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
181	.long	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
182	.long	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
183	.long	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
184	.long	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
185	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
186	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
187	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
188___
189$code.=<<___ if ($SZ==8);
190	.quad	0x428a2f98d728ae22,0x7137449123ef65cd
191	.quad	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
192	.quad	0x3956c25bf348b538,0x59f111f1b605d019
193	.quad	0x923f82a4af194f9b,0xab1c5ed5da6d8118
194	.quad	0xd807aa98a3030242,0x12835b0145706fbe
195	.quad	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
196	.quad	0x72be5d74f27b896f,0x80deb1fe3b1696b1
197	.quad	0x9bdc06a725c71235,0xc19bf174cf692694
198	.quad	0xe49b69c19ef14ad2,0xefbe4786384f25e3
199	.quad	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
200	.quad	0x2de92c6f592b0275,0x4a7484aa6ea6e483
201	.quad	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
202	.quad	0x983e5152ee66dfab,0xa831c66d2db43210
203	.quad	0xb00327c898fb213f,0xbf597fc7beef0ee4
204	.quad	0xc6e00bf33da88fc2,0xd5a79147930aa725
205	.quad	0x06ca6351e003826f,0x142929670a0e6e70
206	.quad	0x27b70a8546d22ffc,0x2e1b21385c26c926
207	.quad	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
208	.quad	0x650a73548baf63de,0x766a0abb3c77b2a8
209	.quad	0x81c2c92e47edaee6,0x92722c851482353b
210	.quad	0xa2bfe8a14cf10364,0xa81a664bbc423001
211	.quad	0xc24b8b70d0f89791,0xc76c51a30654be30
212	.quad	0xd192e819d6ef5218,0xd69906245565a910
213	.quad	0xf40e35855771202a,0x106aa07032bbd1b8
214	.quad	0x19a4c116b8d2d0c8,0x1e376c085141ab53
215	.quad	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
216	.quad	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
217	.quad	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
218	.quad	0x748f82ee5defb2fc,0x78a5636f43172f60
219	.quad	0x84c87814a1f0ab72,0x8cc702081a6439ec
220	.quad	0x90befffa23631e28,0xa4506cebde82bde9
221	.quad	0xbef9a3f7b2c67915,0xc67178f2e372532b
222	.quad	0xca273eceea26619c,0xd186b8c721c0c207
223	.quad	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
224	.quad	0x06f067aa72176fba,0x0a637dc5a2c898a6
225	.quad	0x113f9804bef90dae,0x1b710b35131c471b
226	.quad	0x28db77f523047d84,0x32caab7b40c72493
227	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
228	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
229	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
230___
231$code.=<<___;
232.size	$Table,.-$Table
233.globl	$Func
234.type	$Func,\@function
235$Func:
236	sllg	$len,$len,`log(16*$SZ)/log(2)`
237___
238$code.=<<___ if ($kimdfunc);
239	larl	%r1,OPENSSL_s390xcap_P
240	lg	%r0,0(%r1)
241	tmhl	%r0,0x4000	# check for message-security assist
242	jz	.Lsoftware
243	lghi	%r0,0
244	la	%r1,`2*$SIZE_T`($sp)
245	.long	0xb93e0002	# kimd %r0,%r2
246	lg	%r0,`2*$SIZE_T`($sp)
247	tmhh	%r0,`0x8000>>$kimdfunc`
248	jz	.Lsoftware
249	lghi	%r0,$kimdfunc
250	lgr	%r1,$ctx
251	lgr	%r2,$inp
252	lgr	%r3,$len
253	.long	0xb93e0002	# kimd %r0,%r2
254	brc	1,.-4		# pay attention to "partial completion"
255	br	%r14
256.align	16
257.Lsoftware:
258___
259$code.=<<___;
260	lghi	%r1,-$frame
261	la	$len,0($len,$inp)
262	stm${g}	$ctx,%r15,`2*$SIZE_T`($sp)
263	lgr	%r0,$sp
264	la	$sp,0(%r1,$sp)
265	st${g}	%r0,0($sp)
266
267	larl	$tbl,$Table
268	$LD	$A,`0*$SZ`($ctx)
269	$LD	$B,`1*$SZ`($ctx)
270	$LD	$C,`2*$SZ`($ctx)
271	$LD	$D,`3*$SZ`($ctx)
272	$LD	$E,`4*$SZ`($ctx)
273	$LD	$F,`5*$SZ`($ctx)
274	$LD	$G,`6*$SZ`($ctx)
275	$LD	$H,`7*$SZ`($ctx)
276
277.Lloop:
278	lghi	$len,0
279___
280for ($i=0;$i<16;$i++)	{ &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
281$code.=".Lrounds_16_xx:\n";
282for (;$i<32;$i++)	{ &BODY_16_XX($i,@V); unshift(@V,pop(@V)); }
283$code.=<<___;
284	aghi	$len,`16*$SZ`
285	lghi	$t0,`($rounds-16)*$SZ`
286	clgr	$len,$t0
287	jne	.Lrounds_16_xx
288
289	l${g}	$ctx,`$frame+2*$SIZE_T`($sp)
290	la	$inp,`16*$SZ`($inp)
291	$ADD	$A,`0*$SZ`($ctx)
292	$ADD	$B,`1*$SZ`($ctx)
293	$ADD	$C,`2*$SZ`($ctx)
294	$ADD	$D,`3*$SZ`($ctx)
295	$ADD	$E,`4*$SZ`($ctx)
296	$ADD	$F,`5*$SZ`($ctx)
297	$ADD	$G,`6*$SZ`($ctx)
298	$ADD	$H,`7*$SZ`($ctx)
299	$ST	$A,`0*$SZ`($ctx)
300	$ST	$B,`1*$SZ`($ctx)
301	$ST	$C,`2*$SZ`($ctx)
302	$ST	$D,`3*$SZ`($ctx)
303	$ST	$E,`4*$SZ`($ctx)
304	$ST	$F,`5*$SZ`($ctx)
305	$ST	$G,`6*$SZ`($ctx)
306	$ST	$H,`7*$SZ`($ctx)
307	cl${g}	$inp,`$frame+4*$SIZE_T`($sp)
308	jne	.Lloop
309
310	lm${g}	%r6,%r15,`$frame+6*$SIZE_T`($sp)
311	br	%r14
312.size	$Func,.-$Func
313.string	"SHA${label} block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
314.comm	OPENSSL_s390xcap_P,16,8
315___
316
317$code =~ s/\`([^\`]*)\`/eval $1/gem;
318# unlike 32-bit shift 64-bit one takes three arguments
319$code =~ s/(srlg\s+)(%r[0-9]+),/$1$2,$2,/gm;
320
321print $code;
322close STDOUT;
323