From: Andy Polyakov <appro@openssl.org>
Date: Mon, 27 Jun 2011 10:00:31 +0000 (+0000)
Subject: s390x assembler pack: add s390x-gf2m.pl and harmonize AES_xts_[en|de]crypt.
X-Git-Tag: OpenSSL-fips-2_0-rc1~287
X-Git-Url: https://git.librecmc.org/?a=commitdiff_plain;h=0c237e42a46f0b0dd3de019f75f0c61e27809187;p=oweals%2Fopenssl.git

s390x assembler pack: add s390x-gf2m.pl and harmonize AES_xts_[en|de]crypt.
---

diff --git a/Configure b/Configure
index cdef7f8506..53dd946b76 100755
--- a/Configure
+++ b/Configure
@@ -134,7 +134,7 @@ my $sparcv8_asm=":sparcv8.o:des_enc-sparc.o fcrypt_b.o::::::::::::void";
 my $alpha_asm="alphacpuid.o:bn_asm.o alpha-mont.o:::::sha1-alpha.o:::::::ghash-alpha.o:void";
 my $mips32_asm=":bn-mips.o::aes_cbc.o aes-mips.o:::sha1-mips.o sha256-mips.o:::::::";
 my $mips64_asm=":bn-mips.o mips-mont.o::aes_cbc.o aes-mips.o:::sha1-mips.o sha256-mips.o sha512-mips.o:::::::";
-my $s390x_asm="s390xcap.o s390xcpuid.o:bn-s390x.o s390x-mont.o::aes_ctr.o aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o::rc4-s390x.o:::::ghash-s390x.o";
+my $s390x_asm="s390xcap.o s390xcpuid.o:bn-s390x.o s390x-mont.o s390x-gf2m.o::aes_ctr.o aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o::rc4-s390x.o:::::ghash-s390x.o";
 my $armv4_asm=":bn_asm.o armv4-mont.o armv4-gf2m.o::aes_cbc.o aes-armv4.o:::sha1-armv4-large.o sha256-armv4.o sha512-armv4.o:::::::ghash-armv4.o:void";
 my $parisc11_asm="pariscid.o:bn_asm.o parisc-mont.o::aes_core.o aes_cbc.o aes-parisc.o:::sha1-parisc.o sha256-parisc.o sha512-parisc.o::rc4-parisc.o:::::ghash-parisc.o:32";
 my $parisc20_asm="pariscid.o:pa-risc2W.o parisc-mont.o::aes_core.o aes_cbc.o aes-parisc.o:::sha1-parisc.o sha256-parisc.o sha512-parisc.o::rc4-parisc.o:::::ghash-parisc.o:64";
@@ -370,7 +370,7 @@ my %table=(
 # ldconfig and run-time linker to autodiscover. Unfortunately it
 # doesn't work just yet, because of couple of bugs in glibc
 # sysdep/s390/dl-procinfo.c affecting ldconfig and ld.so.1...
-"linux32-s390x",	"gcc:-m31 -Wa,-mzarch -DB_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL:s390xcap.o s390xcpuid.o:bn_asm.o s390x-mont.o::aes_ctr.o aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o::rc4-s390x.o:::::ghash-s390x.o:31:dlfcn:linux-shared:-fPIC:-m31:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR):::/highgprs",
+"linux32-s390x",	"gcc:-m31 -Wa,-mzarch -DB_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL:s390xcap.o s390xcpuid.o:bn_asm.o s390x-mont.o s390x-gf2m.o::aes_ctr.o aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o::rc4-s390x.o:::::ghash-s390x.o:31:dlfcn:linux-shared:-fPIC:-m31:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR):::/highgprs",
 #### SPARC Linux setups
 # Ray Miller <ray.miller@computing-services.oxford.ac.uk> has patiently
 # assisted with debugging of following two configs.
diff --git a/TABLE b/TABLE
index ccaee9b14f..2e405ec230 100644
--- a/TABLE
+++ b/TABLE
@@ -4105,7 +4105,7 @@ $sys_id       =
 $lflags       = -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL
 $cpuid_obj    = s390xcap.o s390xcpuid.o
-$bn_obj       = bn_asm.o s390x-mont.o
+$bn_obj       = bn_asm.o s390x-mont.o s390x-gf2m.o
 $des_obj      = 
 $aes_obj      = aes_ctr.o aes-s390x.o
 $bf_obj       = 
@@ -4137,7 +4137,7 @@ $sys_id       =
 $lflags       = -ldl
 $bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL
 $cpuid_obj    = s390xcap.o s390xcpuid.o
-$bn_obj       = bn-s390x.o s390x-mont.o
+$bn_obj       = bn-s390x.o s390x-mont.o s390x-gf2m.o
 $des_obj      = 
 $aes_obj      = aes_ctr.o aes-s390x.o
 $bf_obj       = 
diff --git a/crypto/aes/asm/aes-s390x.pl b/crypto/aes/asm/aes-s390x.pl
index d5d514dc2d..687bdd9499 100644
--- a/crypto/aes/asm/aes-s390x.pl
+++ b/crypto/aes/asm/aes-s390x.pl
@@ -78,9 +78,9 @@
 
 # February 2011.
 #
-# Add AES_xts_[en|de]crypt. This includes support for z196
-# km-xts-aes instructions, which deliver ~70% improvement at 8KB
-# block size over vanilla km-based code.
+# Add AES_xts_[en|de]crypt. This includes support for z196 km-xts-aes
+# instructions, which deliver ~70% improvement at 8KB block size over
+# vanilla km-based code, 37% - at most like 512-bytes block size.
 
 $flavour = shift;
 
@@ -1579,7 +1579,8 @@ ___
 
 ########################################################################
 # void AES_xts_encrypt(const char *inp,char *out,size_t len,
-#	const AES_KEY *key1, const AES_KEY *key2,u64 secno);
+#	const AES_KEY *key1, const AES_KEY *key2,
+#	const unsigned char iv[16]);
 #
 {
 my $inp="%r2";
@@ -1595,7 +1596,7 @@ $code.=<<___;
 .align	16
 _s390x_xts_km:
 ___
-$code.=<<___ if(0);
+$code.=<<___ if(1);
 	llgfr	$s0,%r0			# put aside the function code
 	lghi	$s1,0x7f
 	nr	$s1,%r0
@@ -1789,9 +1790,10 @@ $code.=<<___ if (!$softonly);
 	sllg	$len,$len,4		# $len&=~15
 	slgr	$out,$inp
 
-	lrvg	$s0,$stdframe($sp)	# load secno
-	lghi	$s1,0
+	# generate the tweak value
+	l${g}	$s3,$stdframe($sp)	# pointer to iv
 	la	$s2,$tweak($sp)
+	lmg	$s0,$s1,0($s3)
 	lghi	$s3,16
 	stmg	$s0,$s1,0($s2)
 	la	%r1,0($key2)		# $key2 is not needed anymore
@@ -1996,12 +1998,11 @@ $code.=<<___ if (!$softonly);
 	slgr	$out,$inp
 
 	# generate the tweak value
-	lrvg	$s0,$stdframe($sp)	# load secno
-	lghi	$s1,0
+	l${g}	$s3,$stdframe($sp)	# pointer to iv
 	la	$s2,$tweak($sp)
+	lmg	$s0,$s1,0($s3)
 	lghi	$s3,16
-	stg	$s0,0($s2)
-	stg	$s1,8($s2)
+	stmg	$s0,$s1,0($s2)
 	la	%r1,0($key2)		# $key2 is not needed past this point
 	.long	0xb92e00aa		# km $s2,$s2, generate the tweak
 	brc	1,.-4			# can this happen?
diff --git a/crypto/bn/Makefile b/crypto/bn/Makefile
index 74f1ac4b10..bbab19ec15 100644
--- a/crypto/bn/Makefile
+++ b/crypto/bn/Makefile
@@ -91,6 +91,8 @@ mips-mont.s:	asm/mips-mont.pl
 
 bn-s390x.o:	asm/s390x.S
 	$(CC) $(CFLAGS) -c -o $@ asm/s390x.S
+s390x-gf2m.s:	asm/s390x-gfm2.pl
+	$(PERL) asm/s390x-gfm2.pl $(PERLASM_SCHEME) $@
 
 x86_64-gcc.o:	asm/x86_64-gcc.c
 	$(CC) $(CFLAGS) -c -o $@ asm/x86_64-gcc.c
diff --git a/crypto/bn/asm/s390x-gf2m.pl b/crypto/bn/asm/s390x-gf2m.pl
new file mode 100644
index 0000000000..eb389b323a
--- /dev/null
+++ b/crypto/bn/asm/s390x-gf2m.pl
@@ -0,0 +1,220 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# May 2011
+#
+# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
+# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
+# the time being... gcc 4.3 appeared to generate poor code, therefore
+# the effort. The module delivers 55%-90% improvement on haviest ECDSA
+# verify and ECDH benchmarks for 163- and 571-bit keys on z990, and
+# 25%-30% - on z196(*). This is for 64-bit build. In 32-bit "highgprs"
+# case improvement is even higher, for example on z990 it was measured
+# 80%-150%. ECDSA sign is modest 9%-12% faster. Keep in mind that
+# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
+# all CPU time is burnt in it...
+#
+# (*)	Though no improvement could be measured if compared to code
+#	generated by gcc 4.1. Keep in mind that z196 is out-of-order
+#	execution core and is better at executing poor code.
+
+$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,">$output";
+
+$stdframe=16*$SIZE_T+4*8;
+
+$rp="%r2";
+$a1="%r3";
+$a0="%r4";
+$b1="%r5";
+$b0="%r6";
+
+$ra="%r14";
+$sp="%r15";
+
+@T=("%r0","%r1");
+@i=("%r12","%r13");
+
+($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(6..11));
+($lo,$hi,$b)=map("%r$_",(3..5)); $a=$lo; $mask=$a8;
+
+$code.=<<___;
+.text
+
+.type	_mul_1x1,\@function
+.align	16
+_mul_1x1:
+	lgr	$a1,$a
+	sllg	$a2,$a,1
+	sllg	$a4,$a,2
+	sllg	$a8,$a,3
+
+	srag	$lo,$a1,63			# broadcast 63rd bit
+	nihh	$a1,0x1fff
+	srag	@i[0],$a2,63			# broadcast 62nd bit
+	nihh	$a2,0x3fff
+	srag	@i[1],$a4,63			# broadcast 61st bit
+	nihh	$a4,0x7fff
+	ngr	$lo,$b
+	ngr	@i[0],$b
+	ngr	@i[1],$b
+
+	lghi	@T[0],0
+	lgr	$a12,$a1
+	stg	@T[0],`$stdframe+0*8`($sp)	# tab[0]=0
+	xgr	$a12,$a2
+	stg	$a1,`$stdframe+1*8`($sp)	# tab[1]=a1
+	 lgr	$a48,$a4
+	stg	$a2,`$stdframe+2*8`($sp)	# tab[2]=a2
+	 xgr	$a48,$a8
+	stg	$a12,`$stdframe+3*8`($sp)	# tab[3]=a1^a2
+	 xgr	$a1,$a4
+
+	stg	$a4,`$stdframe+4*8`($sp)	# tab[4]=a4
+	xgr	$a2,$a4
+	stg	$a1,`$stdframe+5*8`($sp)	# tab[5]=a1^a4
+	xgr	$a12,$a4
+	stg	$a2,`$stdframe+6*8`($sp)	# tab[6]=a2^a4
+	 xgr	$a1,$a48
+	stg	$a12,`$stdframe+7*8`($sp)	# tab[7]=a1^a2^a4
+	 xgr	$a2,$a48
+
+	stg	$a8,`$stdframe+8*8`($sp)	# tab[8]=a8
+	xgr	$a12,$a48
+	stg	$a1,`$stdframe+9*8`($sp)	# tab[9]=a1^a8
+	 xgr	$a1,$a4
+	stg	$a2,`$stdframe+10*8`($sp)	# tab[10]=a2^a8
+	 xgr	$a2,$a4
+	stg	$a12,`$stdframe+11*8`($sp)	# tab[11]=a1^a2^a8
+
+	xgr	$a12,$a4
+	stg	$a48,`$stdframe+12*8`($sp)	# tab[12]=a4^a8
+	 srlg	$hi,$lo,1
+	stg	$a1,`$stdframe+13*8`($sp)	# tab[13]=a1^a4^a8
+	 sllg	$lo,$lo,63
+	stg	$a2,`$stdframe+14*8`($sp)	# tab[14]=a2^a4^a8
+	 srlg	@T[0],@i[0],2
+	stg	$a12,`$stdframe+15*8`($sp)	# tab[15]=a1^a2^a4^a8
+
+	lghi	$mask,`0xf<<3`
+	sllg	$a1,@i[0],62
+	 sllg	@i[0],$b,3
+	srlg	@T[1],@i[1],3
+	 ngr	@i[0],$mask
+	sllg	$a2,@i[1],61
+	 srlg	@i[1],$b,4-3
+	xgr	$hi,@T[0]
+	 ngr	@i[1],$mask
+	xgr	$lo,$a1
+	xgr	$hi,@T[1]
+	xgr	$lo,$a2
+
+	xg	$lo,$stdframe(@i[0],$sp)
+	srlg	@i[0],$b,8-3
+	ngr	@i[0],$mask
+___
+for($n=1;$n<14;$n++) {
+$code.=<<___;
+	lg	@T[1],$stdframe(@i[1],$sp)
+	srlg	@i[1],$b,`($n+2)*4`-3
+	sllg	@T[0],@T[1],`$n*4`
+	ngr	@i[1],$mask
+	srlg	@T[1],@T[1],`64-$n*4`
+	xgr	$lo,@T[0]
+	xgr	$hi,@T[1]
+___
+	push(@i,shift(@i)); push(@T,shift(@T));
+}
+$code.=<<___;
+	lg	@T[1],$stdframe(@i[1],$sp)
+	sllg	@T[0],@T[1],`$n*4`
+	srlg	@T[1],@T[1],`64-$n*4`
+	xgr	$lo,@T[0]
+	xgr	$hi,@T[1]
+
+	lg	@T[0],$stdframe(@i[0],$sp)
+	sllg	@T[1],@T[0],`($n+1)*4`
+	srlg	@T[0],@T[0],`64-($n+1)*4`
+	xgr	$lo,@T[1]
+	xgr	$hi,@T[0]
+
+	br	$ra
+.size	_mul_1x1,.-_mul_1x1
+
+.globl	bn_GF2m_mul_2x2
+.type	bn_GF2m_mul_2x2,\@function
+.align	16
+bn_GF2m_mul_2x2:
+	stm${g}	%r3,%r15,3*$SIZE_T($sp)
+
+	lghi	%r1,-$stdframe-128
+	la	%r0,0($sp)
+	la	$sp,0(%r1,$sp)			# alloca
+	st${g}	%r0,0($sp)			# back chain
+___
+if ($SIZE_T==8) {
+my @r=map("%r$_",(6..9));
+$code.=<<___;
+	bras	$ra,_mul_1x1			# a1·b1
+	stmg	$lo,$hi,16($rp)
+
+	lg	$a,`$stdframe+128+4*$SIZE_T`($sp)
+	lg	$b,`$stdframe+128+6*$SIZE_T`($sp)
+	bras	$ra,_mul_1x1			# a0·b0
+	stmg	$lo,$hi,0($rp)
+
+	lg	$a,`$stdframe+128+3*$SIZE_T`($sp)
+	lg	$b,`$stdframe+128+5*$SIZE_T`($sp)
+	xg	$a,`$stdframe+128+4*$SIZE_T`($sp)
+	xg	$b,`$stdframe+128+6*$SIZE_T`($sp)
+	bras	$ra,_mul_1x1			# (a0+a1)·(b0+b1)
+	lmg	@r[0],@r[3],0($rp)
+
+	xgr	$lo,$hi
+	xgr	$hi,@r[1]
+	xgr	$lo,@r[0]
+	xgr	$hi,@r[2]
+	xgr	$lo,@r[3]	
+	xgr	$hi,@r[3]
+	xgr	$lo,$hi
+	stg	$hi,16($rp)
+	stg	$lo,8($rp)
+___
+} else {
+$code.=<<___;
+	sllg	%r3,%r3,32
+	sllg	%r5,%r5,32
+	or	%r3,%r4
+	or	%r5,%r6
+	bras	$ra,_mul_1x1
+	rllg	$lo,$lo,32
+	rllg	$hi,$hi,32
+	stmg	$lo,$hi,0($rp)
+___
+}
+$code.=<<___;
+	lm${g}	%r6,%r15,`$stdframe+128+6*$SIZE_T`($sp)
+	br	$ra
+.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
+.string	"GF(2^m) Multiplication for s390x, CRYPTOGAMS by <appro\@openssl.org>"
+___
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+print $code;
+close STDOUT;