aboutsummaryrefslogtreecommitdiff
path: root/openssl/crypto/cast/asm
diff options
context:
space:
mode:
authormarha <marha@users.sourceforge.net>2009-06-28 22:07:26 +0000
committermarha <marha@users.sourceforge.net>2009-06-28 22:07:26 +0000
commit3562e78743202e43aec8727005182a2558117eca (patch)
tree8f9113a77d12470c5c851a2a8e4cb02e89df7d43 /openssl/crypto/cast/asm
downloadvcxsrv-3562e78743202e43aec8727005182a2558117eca.tar.gz
vcxsrv-3562e78743202e43aec8727005182a2558117eca.tar.bz2
vcxsrv-3562e78743202e43aec8727005182a2558117eca.zip
Checked in the following released items:
xkeyboard-config-1.4.tar.gz ttf-bitstream-vera-1.10.tar.gz font-alias-1.0.1.tar.gz font-sun-misc-1.0.0.tar.gz font-sun-misc-1.0.0.tar.gz font-sony-misc-1.0.0.tar.gz font-schumacher-misc-1.0.0.tar.gz font-mutt-misc-1.0.0.tar.gz font-misc-misc-1.0.0.tar.gz font-misc-meltho-1.0.0.tar.gz font-micro-misc-1.0.0.tar.gz font-jis-misc-1.0.0.tar.gz font-isas-misc-1.0.0.tar.gz font-dec-misc-1.0.0.tar.gz font-daewoo-misc-1.0.0.tar.gz font-cursor-misc-1.0.0.tar.gz font-arabic-misc-1.0.0.tar.gz font-winitzki-cyrillic-1.0.0.tar.gz font-misc-cyrillic-1.0.0.tar.gz font-cronyx-cyrillic-1.0.0.tar.gz font-screen-cyrillic-1.0.1.tar.gz font-xfree86-type1-1.0.1.tar.gz font-adobe-utopia-type1-1.0.1.tar.gz font-ibm-type1-1.0.0.tar.gz font-bitstream-type1-1.0.0.tar.gz font-bitstream-speedo-1.0.0.tar.gz font-bh-ttf-1.0.0.tar.gz font-bh-type1-1.0.0.tar.gz font-bitstream-100dpi-1.0.0.tar.gz font-bh-lucidatypewriter-100dpi-1.0.0.tar.gz font-bh-100dpi-1.0.0.tar.gz font-adobe-utopia-100dpi-1.0.1.tar.gz font-adobe-100dpi-1.0.0.tar.gz font-util-1.0.1.tar.gz font-bitstream-75dpi-1.0.0.tar.gz font-bh-lucidatypewriter-75dpi-1.0.0.tar.gz font-adobe-utopia-75dpi-1.0.1.tar.gz font-bh-75dpi-1.0.0.tar.gz bdftopcf-1.0.1.tar.gz font-adobe-75dpi-1.0.0.tar.gz mkfontscale-1.0.6.tar.gz openssl-0.9.8k.tar.gz bigreqsproto-1.0.2.tar.gz xtrans-1.2.2.tar.gz resourceproto-1.0.2.tar.gz inputproto-1.4.4.tar.gz compositeproto-0.4.tar.gz damageproto-1.1.0.tar.gz zlib-1.2.3.tar.gz xkbcomp-1.0.5.tar.gz freetype-2.3.9.tar.gz pthreads-w32-2-8-0-release.tar.gz pixman-0.12.0.tar.gz kbproto-1.0.3.tar.gz evieext-1.0.2.tar.gz fixesproto-4.0.tar.gz recordproto-1.13.2.tar.gz randrproto-1.2.2.tar.gz scrnsaverproto-1.1.0.tar.gz renderproto-0.9.3.tar.gz xcmiscproto-1.1.2.tar.gz fontsproto-2.0.2.tar.gz xextproto-7.0.3.tar.gz xproto-7.0.14.tar.gz libXdmcp-1.0.2.tar.gz libxkbfile-1.0.5.tar.gz libfontenc-1.0.4.tar.gz libXfont-1.3.4.tar.gz libX11-1.1.5.tar.gz libXau-1.0.4.tar.gz libxcb-1.1.tar.gz xorg-server-1.5.3.tar.gz
Diffstat (limited to 'openssl/crypto/cast/asm')
-rw-r--r--openssl/crypto/cast/asm/cast-586.pl176
-rw-r--r--openssl/crypto/cast/asm/readme7
2 files changed, 183 insertions, 0 deletions
diff --git a/openssl/crypto/cast/asm/cast-586.pl b/openssl/crypto/cast/asm/cast-586.pl
new file mode 100644
index 000000000..6be0bfe57
--- /dev/null
+++ b/openssl/crypto/cast/asm/cast-586.pl
@@ -0,0 +1,176 @@
+#!/usr/local/bin/perl
+
+# define for pentium pro friendly version
+$ppro=1;
+
+push(@INC,"perlasm","../../perlasm");
+require "x86asm.pl";
+require "cbc.pl";
+
+&asm_init($ARGV[0],"cast-586.pl",$ARGV[$#ARGV] eq "386");
+
+$CAST_ROUNDS=16;
+$L="edi";
+$R="esi";
+$K="ebp";
+$tmp1="ecx";
+$tmp2="ebx";
+$tmp3="eax";
+$tmp4="edx";
+$S1="CAST_S_table0";
+$S2="CAST_S_table1";
+$S3="CAST_S_table2";
+$S4="CAST_S_table3";
+
+@F1=("add","xor","sub");
+@F2=("xor","sub","add");
+@F3=("sub","add","xor");
+
+&CAST_encrypt("CAST_encrypt",1);
+&CAST_encrypt("CAST_decrypt",0);
+&cbc("CAST_cbc_encrypt","CAST_encrypt","CAST_decrypt",1,4,5,3,-1,-1);
+
+&asm_finish();
+
+sub CAST_encrypt {
+ local($name,$enc)=@_;
+
+ local($win_ex)=<<"EOF";
+EXTERN _CAST_S_table0:DWORD
+EXTERN _CAST_S_table1:DWORD
+EXTERN _CAST_S_table2:DWORD
+EXTERN _CAST_S_table3:DWORD
+EOF
+ &main::external_label(
+ "CAST_S_table0",
+ "CAST_S_table1",
+ "CAST_S_table2",
+ "CAST_S_table3",
+ );
+
+ &function_begin_B($name,$win_ex);
+
+ &comment("");
+
+ &push("ebp");
+ &push("ebx");
+ &mov($tmp2,&wparam(0));
+ &mov($K,&wparam(1));
+ &push("esi");
+ &push("edi");
+
+ &comment("Load the 2 words");
+ &mov($L,&DWP(0,$tmp2,"",0));
+ &mov($R,&DWP(4,$tmp2,"",0));
+
+ &comment('Get short key flag');
+ &mov($tmp3,&DWP(128,$K,"",0));
+ if($enc) {
+ &push($tmp3);
+ } else {
+ &or($tmp3,$tmp3);
+ &jnz(&label('cast_dec_skip'));
+ }
+
+ &xor($tmp3, $tmp3);
+
+ # encrypting part
+
+ if ($enc) {
+ &E_CAST( 0,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 1,$S,$R,$L,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 2,$S,$L,$R,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 3,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 4,$S,$L,$R,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 5,$S,$R,$L,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 6,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 7,$S,$R,$L,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 8,$S,$L,$R,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 9,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(10,$S,$L,$R,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(11,$S,$R,$L,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &comment('test short key flag');
+ &pop($tmp4);
+ &or($tmp4,$tmp4);
+ &jnz(&label('cast_enc_done'));
+ &E_CAST(12,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(13,$S,$R,$L,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(14,$S,$L,$R,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(15,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ } else {
+ &E_CAST(15,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(14,$S,$R,$L,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(13,$S,$L,$R,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(12,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &set_label('cast_dec_skip');
+ &E_CAST(11,$S,$L,$R,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST(10,$S,$R,$L,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 9,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 8,$S,$R,$L,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 7,$S,$L,$R,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 6,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 5,$S,$L,$R,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 4,$S,$R,$L,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 3,$S,$L,$R,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 2,$S,$R,$L,$K,@F3,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 1,$S,$L,$R,$K,@F2,$tmp1,$tmp2,$tmp3,$tmp4);
+ &E_CAST( 0,$S,$R,$L,$K,@F1,$tmp1,$tmp2,$tmp3,$tmp4);
+ }
+
+ &set_label('cast_enc_done') if $enc;
+# Why the nop? - Ben 17/1/99
+ &nop();
+ &mov($tmp3,&wparam(0));
+ &mov(&DWP(4,$tmp3,"",0),$L);
+ &mov(&DWP(0,$tmp3,"",0),$R);
+ &function_end($name);
+}
+
+sub E_CAST {
+ local($i,$S,$L,$R,$K,$OP1,$OP2,$OP3,$tmp1,$tmp2,$tmp3,$tmp4)=@_;
+ # Ri needs to have 16 pre added.
+
+ &comment("round $i");
+ &mov( $tmp4, &DWP($i*8,$K,"",1));
+
+ &mov( $tmp1, &DWP($i*8+4,$K,"",1));
+ &$OP1( $tmp4, $R);
+
+ &rotl( $tmp4, &LB($tmp1));
+
+ if ($ppro) {
+ &mov( $tmp2, $tmp4); # B
+ &xor( $tmp1, $tmp1);
+
+ &movb( &LB($tmp1), &HB($tmp4)); # A
+ &and( $tmp2, 0xff);
+
+ &shr( $tmp4, 16); #
+ &xor( $tmp3, $tmp3);
+ } else {
+ &mov( $tmp2, $tmp4); # B
+ &movb( &LB($tmp1), &HB($tmp4)); # A # BAD BAD BAD
+
+ &shr( $tmp4, 16); #
+ &and( $tmp2, 0xff);
+ }
+
+ &movb( &LB($tmp3), &HB($tmp4)); # C # BAD BAD BAD
+ &and( $tmp4, 0xff); # D
+
+ &mov( $tmp1, &DWP($S1,"",$tmp1,4));
+ &mov( $tmp2, &DWP($S2,"",$tmp2,4));
+
+ &$OP2( $tmp1, $tmp2);
+ &mov( $tmp2, &DWP($S3,"",$tmp3,4));
+
+ &$OP3( $tmp1, $tmp2);
+ &mov( $tmp2, &DWP($S4,"",$tmp4,4));
+
+ &$OP1( $tmp1, $tmp2);
+ # XXX
+
+ &xor( $L, $tmp1);
+ # XXX
+}
+
diff --git a/openssl/crypto/cast/asm/readme b/openssl/crypto/cast/asm/readme
new file mode 100644
index 000000000..fbcd76289
--- /dev/null
+++ b/openssl/crypto/cast/asm/readme
@@ -0,0 +1,7 @@
+There is a ppro flag in cast-586 which turns on/off
+generation of pentium pro/II friendly code
+
+This flag makes the inner loop one cycle longer, but generates
+code that runs %30 faster on the pentium pro/II, while only %7 slower
+on the pentium. By default, this flag is on.
+