aboutsummaryrefslogtreecommitdiff
path: root/openssl/crypto/rc4/asm/rc4-s390x.pl
diff options
context:
space:
mode:
Diffstat (limited to 'openssl/crypto/rc4/asm/rc4-s390x.pl')
-rw-r--r--openssl/crypto/rc4/asm/rc4-s390x.pl39
1 files changed, 34 insertions, 5 deletions
diff --git a/openssl/crypto/rc4/asm/rc4-s390x.pl b/openssl/crypto/rc4/asm/rc4-s390x.pl
index 96681fa05..1aa754820 100644
--- a/openssl/crypto/rc4/asm/rc4-s390x.pl
+++ b/openssl/crypto/rc4/asm/rc4-s390x.pl
@@ -13,6 +13,29 @@
# "cluster" Address Generation Interlocks, so that one pipeline stall
# resolves several dependencies.
+# November 2010.
+#
+# Adapt for -m31 build. If kernel supports what's called "highgprs"
+# feature on Linux [see /proc/cpuinfo], it's possible to use 64-bit
+# instructions and achieve "64-bit" performance even in 31-bit legacy
+# application context. The feature is not specific to any particular
+# processor, as long as it's "z-CPU". Latter implies that the code
+# remains z/Architecture specific. On z990 it was measured to perform
+# 50% better than code generated by gcc 4.3.
+
+$flavour = shift;
+
+if ($flavour =~ /3[12]/) {
+ $SIZE_T=4;
+ $g="";
+} else {
+ $SIZE_T=8;
+ $g="g";
+}
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
$rp="%r14";
$sp="%r15";
$code=<<___;
@@ -39,7 +62,12 @@ $code.=<<___;
.type RC4,\@function
.align 64
RC4:
- stmg %r6,%r11,48($sp)
+ stm${g} %r6,%r11,6*$SIZE_T($sp)
+___
+$code.=<<___ if ($flavour =~ /3[12]/);
+ llgfr $len,$len
+___
+$code.=<<___;
llgc $XX[0],0($key)
llgc $YY,1($key)
la $XX[0],1($XX[0])
@@ -90,7 +118,7 @@ $code.=<<___;
xgr $acc,$TX[1]
stg $acc,0($out)
la $out,8($out)
- brct $cnt,.Loop8
+ brctg $cnt,.Loop8
.Lshort:
lghi $acc,7
@@ -122,7 +150,7 @@ $code.=<<___;
ahi $XX[0],-1
stc $XX[0],0($key)
stc $YY,1($key)
- lmg %r6,%r11,48($sp)
+ lm${g} %r6,%r11,6*$SIZE_T($sp)
br $rp
.size RC4,.-RC4
.string "RC4 for s390x, CRYPTOGAMS by <appro\@openssl.org>"
@@ -147,7 +175,7 @@ $code.=<<___;
.type RC4_set_key,\@function
.align 64
RC4_set_key:
- stmg %r6,%r8,48($sp)
+ stm${g} %r6,%r8,6*$SIZE_T($sp)
lhi $cnt,256
la $idx,0(%r0)
sth $idx,0($key)
@@ -180,7 +208,7 @@ RC4_set_key:
la $iinp,0(%r0)
j .L2ndloop
.Ldone:
- lmg %r6,%r8,48($sp)
+ lm${g} %r6,%r8,6*$SIZE_T($sp)
br $rp
.size RC4_set_key,.-RC4_set_key
@@ -203,3 +231,4 @@ RC4_options:
___
print $code;
+close STDOUT; # force flush