3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> 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 # ====================================================================
12 # The module implements "4-bit" GCM GHASH function and underlying
13 # single multiplication operation in GF(2^128). "4-bit" means that it
14 # uses 256 bytes per-key table [+128 bytes shared table]. Performance
15 # was measured to be ~18 cycles per processed byte on z10, which is
16 # almost 40% better than gcc-generated code. It should be noted that
17 # 18 cycles is worse result than expected: loop is scheduled for 12
18 # and the result should be close to 12. In the lack of instruction-
19 # level profiling data it's impossible to tell why...
21 while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
22 open STDOUT,">$output";
24 $softonly=1; # disable hardware support for now
29 $Xi="%r2"; # argument block
34 $rem0="%r6"; # variables
53 $code.=<<___ if(!$softonly);
54 larl %r1,OPENSSL_s390xcap_P
56 tmhl %r0,0x4000 # check for message-security-assist
60 .long 0xb93e0004 # kimd %r0,%r4
62 tmhh %r1,0x4000 # check for function 65
64 stg %r0,16($sp) # arrange 16 bytes of zero input
66 lghi %r0,65 # function 65
67 la %r1,0($Xi) # H lies right after Xi in gcm128_context
70 .long 0xb93e0004 # kimd %r0,$inp
71 brc 1,.-4 # pay attention to "partial completion"
82 larl $rem_4bit,rem_4bit
86 .type gcm_gmult_4bit,\@function
87 .size gcm_gmult_4bit,(.-gcm_gmult_4bit)
93 $code.=<<___ if(!$softonly);
94 larl %r1,OPENSSL_s390xcap_P
96 tmhl %r0,0x4000 # check for message-security-assist
100 .long 0xb93e0004 # kimd %r0,%r4
102 tmhh %r1,0x4000 # check for function 65
104 lghi %r0,65 # function 65
105 la %r1,0($Xi) # H lies right after Xi in gcm128_context
106 .long 0xb93e0004 # kimd %r0,$inp
107 brc 1,.-4 # pay attention to "partial completion"
113 stmg %r6,%r14,48($sp)
118 larl $rem_4bit,rem_4bit
120 lg $Zlo,8+1($Xi) # Xi
124 xg $Zhi,0($inp) # Xi ^= inp
133 srlg $xi,$Zlo,8 # extract second byte
139 lg $Zlo,8($nlo,$Htbl)
140 lg $Zhi,0($nlo,$Htbl)
151 xg $Zlo,8($nhi,$Htbl)
152 xg $Zhi,0($nhi,$Htbl)
162 xg $Zlo,8($nlo,$Htbl)
165 xg $Zhi,0($nlo,$Htbl)
167 xg $Zhi,0($rem0,$rem_4bit)
177 xg $Zlo,8($nhi,$Htbl)
178 xg $Zhi,0($nhi,$Htbl)
180 xg $Zhi,0($rem1,$rem_4bit)
184 brct $cnt,.Lghash_inner
189 xg $Zlo,8($nlo,$Htbl)
190 xg $Zhi,0($nlo,$Htbl)
192 xg $Zhi,0($rem0,$rem_4bit)
199 xg $Zlo,8($nhi,$Htbl)
200 xg $Zhi,0($nhi,$Htbl)
202 xg $Zhi,0($rem1,$rem_4bit)
204 lg $tmp,0($xi,$rem_4bit)
206 sllg $tmp,$tmp,4 # correct last rem_4bit[rem]
214 .type gcm_ghash_4bit,\@function
215 .size gcm_ghash_4bit,(.-gcm_ghash_4bit)
219 .long `0x0000<<12`,0,`0x1C20<<12`,0,`0x3840<<12`,0,`0x2460<<12`,0
220 .long `0x7080<<12`,0,`0x6CA0<<12`,0,`0x48C0<<12`,0,`0x54E0<<12`,0
221 .long `0xE100<<12`,0,`0xFD20<<12`,0,`0xD940<<12`,0,`0xC560<<12`,0
222 .long `0x9180<<12`,0,`0x8DA0<<12`,0,`0xA9C0<<12`,0,`0xB5E0<<12`,0
223 .type rem_4bit,\@object
224 .size rem_4bit,(.-rem_4bit)
225 .string "GHASH for s390x, CRYPTOGAMS by <appro\@openssl.org>"
228 $code =~ s/\`([^\`]*)\`/eval $1/gem;