aboutsummaryrefslogtreecommitdiff
path: root/openssl/crypto/rc4/asm/rc4-586.pl
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/rc4/asm/rc4-586.pl
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/rc4/asm/rc4-586.pl')
-rw-r--r--openssl/crypto/rc4/asm/rc4-586.pl230
1 files changed, 230 insertions, 0 deletions
diff --git a/openssl/crypto/rc4/asm/rc4-586.pl b/openssl/crypto/rc4/asm/rc4-586.pl
new file mode 100644
index 000000000..ef7eee766
--- /dev/null
+++ b/openssl/crypto/rc4/asm/rc4-586.pl
@@ -0,0 +1,230 @@
+#!/usr/local/bin/perl
+
+# At some point it became apparent that the original SSLeay RC4
+# assembler implementation performs suboptimaly on latest IA-32
+# microarchitectures. After re-tuning performance has changed as
+# following:
+#
+# Pentium +0%
+# Pentium III +17%
+# AMD +52%(*)
+# P4 +180%(**)
+#
+# (*) This number is actually a trade-off:-) It's possible to
+# achieve +72%, but at the cost of -48% off PIII performance.
+# In other words code performing further 13% faster on AMD
+# would perform almost 2 times slower on Intel PIII...
+# For reference! This code delivers ~80% of rc4-amd64.pl
+# performance on the same Opteron machine.
+# (**) This number requires compressed key schedule set up by
+# RC4_set_key and therefore doesn't apply to 0.9.7 [option for
+# compressed key schedule is implemented in 0.9.8 and later,
+# see commentary section in rc4_skey.c for further details].
+#
+# <appro@fy.chalmers.se>
+
+push(@INC,"perlasm","../../perlasm");
+require "x86asm.pl";
+
+&asm_init($ARGV[0],"rc4-586.pl");
+
+$x="eax";
+$y="ebx";
+$tx="ecx";
+$ty="edx";
+$in="esi";
+$out="edi";
+$d="ebp";
+
+&RC4("RC4");
+
+&asm_finish();
+
+sub RC4_loop
+ {
+ local($n,$p,$char)=@_;
+
+ &comment("Round $n");
+
+ if ($char)
+ {
+ if ($p >= 0)
+ {
+ &mov($ty, &swtmp(2));
+ &cmp($ty, $in);
+ &jbe(&label("finished"));
+ &inc($in);
+ }
+ else
+ {
+ &add($ty, 8);
+ &inc($in);
+ &cmp($ty, $in);
+ &jb(&label("finished"));
+ &mov(&swtmp(2), $ty);
+ }
+ }
+ # Moved out
+ # &mov( $tx, &DWP(0,$d,$x,4)) if $p < 0;
+
+ &add( &LB($y), &LB($tx));
+ &mov( $ty, &DWP(0,$d,$y,4));
+ # XXX
+ &mov( &DWP(0,$d,$x,4),$ty);
+ &add( $ty, $tx);
+ &mov( &DWP(0,$d,$y,4),$tx);
+ &and( $ty, 0xff);
+ &inc( &LB($x)); # NEXT ROUND
+ &mov( $tx, &DWP(0,$d,$x,4)) if $p < 1; # NEXT ROUND
+ &mov( $ty, &DWP(0,$d,$ty,4));
+
+ if (!$char)
+ {
+ #moved up into last round
+ if ($p >= 1)
+ {
+ &add( $out, 8)
+ }
+ &movb( &BP($n,"esp","",0), &LB($ty));
+ }
+ else
+ {
+ # Note in+=8 has occured
+ &movb( &HB($ty), &BP(-1,$in,"",0));
+ # XXX
+ &xorb(&LB($ty), &HB($ty));
+ # XXX
+ &movb(&BP($n,$out,"",0),&LB($ty));
+ }
+ }
+
+
+sub RC4
+ {
+ local($name)=@_;
+
+ &function_begin_B($name,"");
+
+ &mov($ty,&wparam(1)); # len
+ &cmp($ty,0);
+ &jne(&label("proceed"));
+ &ret();
+ &set_label("proceed");
+
+ &comment("");
+
+ &push("ebp");
+ &push("ebx");
+ &push("esi");
+ &xor( $x, $x); # avoid partial register stalls
+ &push("edi");
+ &xor( $y, $y); # avoid partial register stalls
+ &mov( $d, &wparam(0)); # key
+ &mov( $in, &wparam(2));
+
+ &movb( &LB($x), &BP(0,$d,"",1));
+ &movb( &LB($y), &BP(4,$d,"",1));
+
+ &mov( $out, &wparam(3));
+ &inc( &LB($x));
+
+ &stack_push(3); # 3 temp variables
+ &add( $d, 8);
+
+ # detect compressed schedule, see commentary section in rc4_skey.c...
+ # in 0.9.7 context ~50 bytes below RC4_CHAR label remain redundant,
+ # as compressed key schedule is set up in 0.9.8 and later.
+ &cmp(&DWP(256,$d),-1);
+ &je(&label("RC4_CHAR"));
+
+ &lea( $ty, &DWP(-8,$ty,$in));
+
+ # check for 0 length input
+
+ &mov( &swtmp(2), $ty); # this is now address to exit at
+ &mov( $tx, &DWP(0,$d,$x,4));
+
+ &cmp( $ty, $in);
+ &jb( &label("end")); # less than 8 bytes
+
+ &set_label("start");
+
+ # filling DELAY SLOT
+ &add( $in, 8);
+
+ &RC4_loop(0,-1,0);
+ &RC4_loop(1,0,0);
+ &RC4_loop(2,0,0);
+ &RC4_loop(3,0,0);
+ &RC4_loop(4,0,0);
+ &RC4_loop(5,0,0);
+ &RC4_loop(6,0,0);
+ &RC4_loop(7,1,0);
+
+ &comment("apply the cipher text");
+ # xor the cipher data with input
+
+ #&add( $out, 8); #moved up into last round
+
+ &mov( $tx, &swtmp(0));
+ &mov( $ty, &DWP(-8,$in,"",0));
+ &xor( $tx, $ty);
+ &mov( $ty, &DWP(-4,$in,"",0));
+ &mov( &DWP(-8,$out,"",0), $tx);
+ &mov( $tx, &swtmp(1));
+ &xor( $tx, $ty);
+ &mov( $ty, &swtmp(2)); # load end ptr;
+ &mov( &DWP(-4,$out,"",0), $tx);
+ &mov( $tx, &DWP(0,$d,$x,4));
+ &cmp($in, $ty);
+ &jbe(&label("start"));
+
+ &set_label("end");
+
+ # There is quite a bit of extra crap in RC4_loop() for this
+ # first round
+ &RC4_loop(0,-1,1);
+ &RC4_loop(1,0,1);
+ &RC4_loop(2,0,1);
+ &RC4_loop(3,0,1);
+ &RC4_loop(4,0,1);
+ &RC4_loop(5,0,1);
+ &RC4_loop(6,1,1);
+
+ &jmp(&label("finished"));
+
+ &align(16);
+ # this is essentially Intel P4 specific codepath, see rc4_skey.c,
+ # and is engaged in 0.9.8 and later context...
+ &set_label("RC4_CHAR");
+
+ &lea ($ty,&DWP(0,$in,$ty));
+ &mov (&swtmp(2),$ty);
+ &movz ($tx,&BP(0,$d,$x));
+
+ # strangely enough unrolled loop performs over 20% slower...
+ &set_label("RC4_CHAR_loop");
+ &add (&LB($y),&LB($tx));
+ &movz ($ty,&BP(0,$d,$y));
+ &movb (&BP(0,$d,$y),&LB($tx));
+ &movb (&BP(0,$d,$x),&LB($ty));
+ &add (&LB($ty),&LB($tx));
+ &movz ($ty,&BP(0,$d,$ty));
+ &add (&LB($x),1);
+ &xorb (&LB($ty),&BP(0,$in));
+ &lea ($in,&DWP(1,$in));
+ &movz ($tx,&BP(0,$d,$x));
+ &cmp ($in,&swtmp(2));
+ &movb (&BP(0,$out),&LB($ty));
+ &lea ($out,&DWP(1,$out));
+ &jb (&label("RC4_CHAR_loop"));
+
+ &set_label("finished");
+ &dec( $x);
+ &stack_pop(3);
+ &movb( &BP(-4,$d,"",0),&LB($y));
+ &movb( &BP(-8,$d,"",0),&LB($x));
+
+ &function_end($name);
+ }
+