From e822c756b66024d49ab936bf77b745206660fcd2 Mon Sep 17 00:00:00 2001
From: Andy Polyakov <appro@openssl.org>
Date: Mon, 29 Nov 2010 20:52:43 +0000
Subject: s390x assembler pack: adapt for -m31 build, see commentary in
 Configure for more details.

---
 crypto/rc4/asm/rc4-s390x.pl | 35 ++++++++++++++++++++++++++++++-----
 1 file changed, 30 insertions(+), 5 deletions(-)

(limited to 'crypto/rc4')

diff --git a/crypto/rc4/asm/rc4-s390x.pl b/crypto/rc4/asm/rc4-s390x.pl
index f26c515e78..1aa754820c 100644
--- a/crypto/rc4/asm/rc4-s390x.pl
+++ b/crypto/rc4/asm/rc4-s390x.pl
@@ -13,6 +13,26 @@
 # "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";
 
@@ -42,7 +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])
@@ -93,7 +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
@@ -125,7 +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>"
@@ -150,7 +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)
@@ -183,7 +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
 
-- 
cgit v1.2.3