• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1#! /usr/bin/env perl
2# Copyright 2007-2020 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the Apache License 2.0 (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9
10# ====================================================================
11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12# project. The module is, however, dual licensed under OpenSSL and
13# CRYPTOGAMS licenses depending on where you obtain it. For further
14# details see http://www.openssl.org/~appro/cryptogams/.
15# ====================================================================
16
17# SHA256/512 block procedures for s390x.
18
19# April 2007.
20#
21# sha256_block_data_order is reportedly >3 times faster than gcc 3.3
22# generated code (must be a bug in compiler, as improvement is
23# "pathologically" high, in particular in comparison to other SHA
24# modules). But the real twist is that it detects if hardware support
25# for SHA256 is available and in such case utilizes it. Then the
26# performance can reach >6.5x of assembler one for larger chunks.
27#
28# sha512_block_data_order is ~70% faster than gcc 3.3 generated code.
29
30# January 2009.
31#
32# Add support for hardware SHA512 and reschedule instructions to
33# favour dual-issue z10 pipeline. Hardware SHA256/512 is ~4.7x faster
34# than software.
35
36# November 2010.
37#
38# Adapt for -m31 build. If kernel supports what's called "highgprs"
39# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
40# instructions and achieve "64-bit" performance even in 31-bit legacy
41# application context. The feature is not specific to any particular
42# processor, as long as it's "z-CPU". Latter implies that the code
43# remains z/Architecture specific. On z990 SHA256 was measured to
44# perform 2.4x and SHA512 - 13x better than code generated by gcc 4.3.
45
46# $output is the last argument if it looks like a file (it has an extension)
47# $flavour is the first argument if it doesn't look like a file
48$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
49$flavour = $#ARGV >= 0 && $ARGV[0] !~ m|\.| ? shift : undef;
50
51if ($flavour =~ /3[12]/) {
52	$SIZE_T=4;
53	$g="";
54} else {
55	$SIZE_T=8;
56	$g="g";
57}
58
59$t0="%r0";
60$t1="%r1";
61$ctx="%r2";	$t2="%r2";
62$inp="%r3";
63$len="%r4";	# used as index in inner loop
64
65$A="%r5";
66$B="%r6";
67$C="%r7";
68$D="%r8";
69$E="%r9";
70$F="%r10";
71$G="%r11";
72$H="%r12";	@V=($A,$B,$C,$D,$E,$F,$G,$H);
73$tbl="%r13";
74$T1="%r14";
75$sp="%r15";
76
77open STDOUT,">$output";
78
79if ($output =~ /512/) {
80	$label="512";
81	$SZ=8;
82	$LD="lg";	# load from memory
83	$ST="stg";	# store to memory
84	$ADD="alg";	# add with memory operand
85	$ROT="rllg";	# rotate left
86	$SHR="srlg";	# logical right shift [see even at the end]
87	@Sigma0=(25,30,36);
88	@Sigma1=(23,46,50);
89	@sigma0=(56,63, 7);
90	@sigma1=( 3,45, 6);
91	$rounds=80;
92	$kimdfunc=3;	# 0 means unknown/unsupported/unimplemented/disabled
93} else {
94	$label="256";
95	$SZ=4;
96	$LD="llgf";	# load from memory
97	$ST="st";	# store to memory
98	$ADD="al";	# add with memory operand
99	$ROT="rll";	# rotate left
100	$SHR="srl";	# logical right shift
101	@Sigma0=(10,19,30);
102	@Sigma1=( 7,21,26);
103	@sigma0=(14,25, 3);
104	@sigma1=(13,15,10);
105	$rounds=64;
106	$kimdfunc=2;	# magic function code for kimd instruction
107}
108$Func="sha${label}_block_data_order";
109$Table="K${label}";
110$stdframe=16*$SIZE_T+4*8;
111$frame=$stdframe+16*$SZ;
112
113sub BODY_00_15 {
114my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
115
116$code.=<<___ if ($i<16);
117	$LD	$T1,`$i*$SZ`($inp)	### $i
118___
119$code.=<<___;
120	$ROT	$t0,$e,$Sigma1[0]
121	$ROT	$t1,$e,$Sigma1[1]
122	 lgr	$t2,$f
123	xgr	$t0,$t1
124	$ROT	$t1,$t1,`$Sigma1[2]-$Sigma1[1]`
125	 xgr	$t2,$g
126	$ST	$T1,`$stdframe+$SZ*($i%16)`($sp)
127	xgr	$t0,$t1			# Sigma1(e)
128	algr	$T1,$h			# T1+=h
129	 ngr	$t2,$e
130	 lgr	$t1,$a
131	algr	$T1,$t0			# T1+=Sigma1(e)
132	$ROT	$h,$a,$Sigma0[0]
133	 xgr	$t2,$g			# Ch(e,f,g)
134	$ADD	$T1,`$i*$SZ`($len,$tbl)	# T1+=K[i]
135	$ROT	$t0,$a,$Sigma0[1]
136	algr	$T1,$t2			# T1+=Ch(e,f,g)
137	 ogr	$t1,$b
138	xgr	$h,$t0
139	 lgr	$t2,$a
140	 ngr	$t1,$c
141	$ROT	$t0,$t0,`$Sigma0[2]-$Sigma0[1]`
142	xgr	$h,$t0			# h=Sigma0(a)
143	 ngr	$t2,$b
144	algr	$h,$T1			# h+=T1
145	 ogr	$t2,$t1			# Maj(a,b,c)
146	algr	$d,$T1			# d+=T1
147	algr	$h,$t2			# h+=Maj(a,b,c)
148___
149}
150
151sub BODY_16_XX {
152my ($i,$a,$b,$c,$d,$e,$f,$g,$h) = @_;
153
154$code.=<<___;
155	$LD	$T1,`$stdframe+$SZ*(($i+1)%16)`($sp)	### $i
156	$LD	$t1,`$stdframe+$SZ*(($i+14)%16)`($sp)
157	$ROT	$t0,$T1,$sigma0[0]
158	$SHR	$T1,$sigma0[2]
159	$ROT	$t2,$t0,`$sigma0[1]-$sigma0[0]`
160	xgr	$T1,$t0
161	$ROT	$t0,$t1,$sigma1[0]
162	xgr	$T1,$t2					# sigma0(X[i+1])
163	$SHR	$t1,$sigma1[2]
164	$ADD	$T1,`$stdframe+$SZ*($i%16)`($sp)	# +=X[i]
165	xgr	$t1,$t0
166	$ROT	$t0,$t0,`$sigma1[1]-$sigma1[0]`
167	$ADD	$T1,`$stdframe+$SZ*(($i+9)%16)`($sp)	# +=X[i+9]
168	xgr	$t1,$t0				# sigma1(X[i+14])
169	algr	$T1,$t1				# +=sigma1(X[i+14])
170___
171	&BODY_00_15(@_);
172}
173
174$code.=<<___;
175#include "s390x_arch.h"
176
177.text
178.align	64
179.type	$Table,\@object
180$Table:
181___
182$code.=<<___ if ($SZ==4);
183	.long	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
184	.long	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
185	.long	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
186	.long	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
187	.long	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
188	.long	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
189	.long	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
190	.long	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
191	.long	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
192	.long	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
193	.long	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
194	.long	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
195	.long	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
196	.long	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
197	.long	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
198	.long	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
199___
200$code.=<<___ if ($SZ==8);
201	.quad	0x428a2f98d728ae22,0x7137449123ef65cd
202	.quad	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
203	.quad	0x3956c25bf348b538,0x59f111f1b605d019
204	.quad	0x923f82a4af194f9b,0xab1c5ed5da6d8118
205	.quad	0xd807aa98a3030242,0x12835b0145706fbe
206	.quad	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
207	.quad	0x72be5d74f27b896f,0x80deb1fe3b1696b1
208	.quad	0x9bdc06a725c71235,0xc19bf174cf692694
209	.quad	0xe49b69c19ef14ad2,0xefbe4786384f25e3
210	.quad	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
211	.quad	0x2de92c6f592b0275,0x4a7484aa6ea6e483
212	.quad	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
213	.quad	0x983e5152ee66dfab,0xa831c66d2db43210
214	.quad	0xb00327c898fb213f,0xbf597fc7beef0ee4
215	.quad	0xc6e00bf33da88fc2,0xd5a79147930aa725
216	.quad	0x06ca6351e003826f,0x142929670a0e6e70
217	.quad	0x27b70a8546d22ffc,0x2e1b21385c26c926
218	.quad	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
219	.quad	0x650a73548baf63de,0x766a0abb3c77b2a8
220	.quad	0x81c2c92e47edaee6,0x92722c851482353b
221	.quad	0xa2bfe8a14cf10364,0xa81a664bbc423001
222	.quad	0xc24b8b70d0f89791,0xc76c51a30654be30
223	.quad	0xd192e819d6ef5218,0xd69906245565a910
224	.quad	0xf40e35855771202a,0x106aa07032bbd1b8
225	.quad	0x19a4c116b8d2d0c8,0x1e376c085141ab53
226	.quad	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
227	.quad	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
228	.quad	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
229	.quad	0x748f82ee5defb2fc,0x78a5636f43172f60
230	.quad	0x84c87814a1f0ab72,0x8cc702081a6439ec
231	.quad	0x90befffa23631e28,0xa4506cebde82bde9
232	.quad	0xbef9a3f7b2c67915,0xc67178f2e372532b
233	.quad	0xca273eceea26619c,0xd186b8c721c0c207
234	.quad	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
235	.quad	0x06f067aa72176fba,0x0a637dc5a2c898a6
236	.quad	0x113f9804bef90dae,0x1b710b35131c471b
237	.quad	0x28db77f523047d84,0x32caab7b40c72493
238	.quad	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
239	.quad	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
240	.quad	0x5fcb6fab3ad6faec,0x6c44198c4a475817
241___
242$code.=<<___;
243.size	$Table,.-$Table
244.globl	$Func
245.type	$Func,\@function
246$Func:
247	sllg	$len,$len,`log(16*$SZ)/log(2)`
248___
249$code.=<<___ if ($kimdfunc);
250	larl	%r1,OPENSSL_s390xcap_P
251	lg	%r0,S390X_KIMD(%r1)	# check kimd capabilities
252	tmhh	%r0,`0x8000>>$kimdfunc`
253	jz	.Lsoftware
254	lghi	%r0,$kimdfunc
255	lgr	%r1,$ctx
256	lgr	%r2,$inp
257	lgr	%r3,$len
258	.long	0xb93e0002	# kimd %r0,%r2
259	brc	1,.-4		# pay attention to "partial completion"
260	br	%r14
261.align	16
262.Lsoftware:
263___
264$code.=<<___;
265	lghi	%r1,-$frame
266	la	$len,0($len,$inp)
267	stm${g}	$ctx,%r15,`2*$SIZE_T`($sp)
268	lgr	%r0,$sp
269	la	$sp,0(%r1,$sp)
270	st${g}	%r0,0($sp)
271
272	larl	$tbl,$Table
273	$LD	$A,`0*$SZ`($ctx)
274	$LD	$B,`1*$SZ`($ctx)
275	$LD	$C,`2*$SZ`($ctx)
276	$LD	$D,`3*$SZ`($ctx)
277	$LD	$E,`4*$SZ`($ctx)
278	$LD	$F,`5*$SZ`($ctx)
279	$LD	$G,`6*$SZ`($ctx)
280	$LD	$H,`7*$SZ`($ctx)
281
282.Lloop:
283	lghi	$len,0
284___
285for ($i=0;$i<16;$i++)	{ &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
286$code.=".Lrounds_16_xx:\n";
287for (;$i<32;$i++)	{ &BODY_16_XX($i,@V); unshift(@V,pop(@V)); }
288$code.=<<___;
289	aghi	$len,`16*$SZ`
290	lghi	$t0,`($rounds-16)*$SZ`
291	clgr	$len,$t0
292	jne	.Lrounds_16_xx
293
294	l${g}	$ctx,`$frame+2*$SIZE_T`($sp)
295	la	$inp,`16*$SZ`($inp)
296	$ADD	$A,`0*$SZ`($ctx)
297	$ADD	$B,`1*$SZ`($ctx)
298	$ADD	$C,`2*$SZ`($ctx)
299	$ADD	$D,`3*$SZ`($ctx)
300	$ADD	$E,`4*$SZ`($ctx)
301	$ADD	$F,`5*$SZ`($ctx)
302	$ADD	$G,`6*$SZ`($ctx)
303	$ADD	$H,`7*$SZ`($ctx)
304	$ST	$A,`0*$SZ`($ctx)
305	$ST	$B,`1*$SZ`($ctx)
306	$ST	$C,`2*$SZ`($ctx)
307	$ST	$D,`3*$SZ`($ctx)
308	$ST	$E,`4*$SZ`($ctx)
309	$ST	$F,`5*$SZ`($ctx)
310	$ST	$G,`6*$SZ`($ctx)
311	$ST	$H,`7*$SZ`($ctx)
312	cl${g}	$inp,`$frame+4*$SIZE_T`($sp)
313	jne	.Lloop
314
315	lm${g}	%r6,%r15,`$frame+6*$SIZE_T`($sp)
316	br	%r14
317.size	$Func,.-$Func
318.string	"SHA${label} block transform for s390x, CRYPTOGAMS by <appro\@openssl.org>"
319___
320
321$code =~ s/\`([^\`]*)\`/eval $1/gem;
322# unlike 32-bit shift 64-bit one takes three arguments
323$code =~ s/(srlg\s+)(%r[0-9]+),/$1$2,$2,/gm;
324
325print $code;
326close STDOUT or die "error closing STDOUT: $!";
327