3 # ====================================================================
4 # Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
5 # project. Rights for redistribution and usage in source and binary
6 # forms are granted according to the OpenSSL license.
7 # ====================================================================
9 # 2.22x RC4 tune-up:-) It should be noted though that my hand [as in
10 # "hand-coded assembler"] doesn't stand for the whole improvement
11 # coefficient. It turned out that eliminating RC4_CHAR from config
12 # line results in ~40% improvement (yes, even for C implementation).
13 # Presumably it has everything to do with AMD cache architecture and
14 # RAW or whatever penalties. Once again! The module *requires* config
15 # line *without* RC4_CHAR! As for coding "secret," I bet on partial
16 # register arithmetics. For example instead 'inc %r8; and $255,%r8'
17 # I simply 'inc %r8b'. Even though optimization manual discourages
18 # to operate on partial registers, it turned out to be the best bet.
19 # At least for AMD... How IA32E would perform remains to be seen...
23 $win64=1 if ($output =~ /win64.[s|asm]/);
25 open STDOUT,">$output" || die "can't open $output: $!";
27 if (defined($win64)) {
30 $inp="%rsi"; # r8, arg3 moves here
31 $out="%rdi"; # r9, arg4 moves here
46 if (defined($win64)) {
47 $ret =~ s/\[([\S]+)\+([\S]+)\]/[$2+$1]/g; # [%rN+%rM*4]->[%rM*4+%rN]
49 $ret =~ s/[\+\*]/,/g; # [%rN+%rM*4]->[%rN,%rM,4]
50 $ret =~ s/\[([^\]]+)\]/($1)/g; # [%rN]->(%rN)
55 $code=<<___ if (!defined($win64));
63 .byte 0xF3,0xC3 # repz ret, 2-byte ret
66 $code=<<___ if (defined($win64));
73 DB F3h,C3h ; repz ret, 2-byte ret
83 movl `&PTR("DWORD-8[$dat]")`,$XX#d
84 movl `&PTR("DWORD-4[$dat]")`,$YY#d
89 movq `&PTR("QWORD[$inp]")`,%rax
92 movl `&PTR("DWORD[$dat+$XX*4]")`,$TX#d
94 movl `&PTR("DWORD[$dat+$YY*4]")`,$TY#d
95 movl $TX#d,`&PTR("DWORD[$dat+$YY*4]")`
96 movl $TY#d,`&PTR("DWORD[$dat+$XX*4]")`
99 movl `&PTR("DWORD[$dat+$TX*4]")`,$TY#d
102 for ($i=1;$i<=6;$i++) {
104 movl `&PTR("DWORD[$dat+$XX*4]")`,$TX#d
106 movl `&PTR("DWORD[$dat+$YY*4]")`,$TY#d
107 movl $TX#d,`&PTR("DWORD[$dat+$YY*4]")`
108 movl $TY#d,`&PTR("DWORD[$dat+$XX*4]")`
110 movl `&PTR("DWORD[$dat+$TX*4]")`,$TY#d
117 movl `&PTR("DWORD[$dat+$XX*4]")`,$TX#d
119 movl `&PTR("DWORD[$dat+$YY*4]")`,$TY#d
120 movl $TX#d,`&PTR("DWORD[$dat+$YY*4]")`
121 movl $TY#d,`&PTR("DWORD[$dat+$XX*4]")`
125 movl `&PTR("DWORD[$dat+$TX*4]")`,$TY#d
130 mov %rax,`&PTR("QWORD-8[$out]")`
137 movl $XX#d,`&PTR("DWORD-8[$dat]")`
138 movl $YY#d,`&PTR("DWORD-4[$dat]")`
140 $code.=<<___ if (defined($win64));
144 DB F3h,C3h ; retz ret, 2-byte ret
146 $code.=<<___ if (!defined($win64));
147 .byte 0xF3,0xC3 # repz ret, 2-byte ret
152 movzb `&PTR("BYTE[$inp]")`,%rax
155 movl `&PTR("DWORD[$dat+$XX*4]")`,$TX#d
157 movl `&PTR("DWORD[$dat+$YY*4]")`,$TY#d
158 movl $TX#d,`&PTR("DWORD[$dat+$YY*4]")`
159 movl $TY#d,`&PTR("DWORD[$dat+$XX*4]")`
161 movl `&PTR("DWORD[$dat+$TX*4]")`,$TY#d
164 movb %al,`&PTR("BYTE[$out]")`
170 if (defined($win64)) {
173 $code.=".size RC4,.-RC4\n"
176 $code =~ s/#([bwd])/$1/gm;
177 $code =~ s/\`([^\`]*)\`/eval $1/gem;
179 if (defined($win64)) {
180 $code =~ s/\.align/ALIGN/gm;
181 $code =~ s/[\$%]//gm;
182 $code =~ s/\.L/\$L/gm;
183 $code =~ s/([\w]+)([\s]+)([\S]+),([\S]+)/$1$2$4,$3/gm;
184 $code =~ s/([QD]*WORD|BYTE)/$1 PTR /gm;
185 $code =~ s/(mov[z]*)[bwlq]/$1/gm;
187 $code =~ s/[QD]*WORD|BYTE//gm;