s390x assembler pack update.

Configure

@@ -131,7 +131,7 @@ my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-
 my $sparcv8_asm=":sparcv8.o:des_enc-sparc.o fcrypt_b.o:::::::::::void";
 my $alpha_asm="alphacpuid.o:bn_asm.o alpha-mont.o::::::::::::void";
 my $mips3_asm=":bn-mips3.o::::::::::::void";
-my $s390x_asm=":bn-s390x.o::aes_cbc.o aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o:::::::void";
+my $s390x_asm="s390xcpuid.o:bn-s390x.o s390x-mont.o::aes-s390x.o:::sha1-s390x.o sha256-s390x.o sha512-s390x.o:::::::void";
 my $armv4_asm=":bn_asm.o armv4-mont.o::aes_cbc.o aes-armv4.o:::sha1-armv4-large.o sha256-armv4.o sha512-armv4.o:::::::void";
 my $ppc32_asm="ppccpuid.o:bn-ppc.o::aes_core.o aes_cbc.o aes-ppc.o:::sha1-ppc.o sha256-ppc.o::::::";
 my $ppc64_asm="ppccpuid.o:bn-ppc.o ppc-mont.o::aes_core.o aes_cbc.o aes-ppc.o:::sha1-ppc.o sha256-ppc.o sha512-ppc.o::::::";

TABLE

@@ -3542,10 +3542,10 @@ $thread_cflag = -D_REENTRANT
 $sys_id       = 
 $lflags       = -ldl
 $bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_INT DES_UNROLL
-$cpuid_obj    = 
+$cpuid_obj    = s390xcpuid.o
-$bn_obj       = bn-s390x.o
+$bn_obj       = bn-s390x.o s390x-mont.o
 $des_obj      = 
-$aes_obj      = aes_cbc.o aes-s390x.o
+$aes_obj      = aes-s390x.o
 $bf_obj       = 
 $md5_obj      = 
 $sha1_obj     = sha1-s390x.o sha256-s390x.o sha512-s390x.o

crypto/bn/asm/s390x-mont.pl

@@ -27,6 +27,11 @@
 # module performance by implementing dedicated squaring code-path and
 # possibly by unrolling loops...
 
+# January 2009.
+#
+# Reschedule to minimize/avoid Address Generation Interlock hazard,
+# make inner loops counter-based.
+
 $mn0="%r0";
 $num="%r1";
 
@@ -47,7 +52,7 @@ $nhi="%r10";
 $nlo="%r11";
 $AHI="%r12";
 $NHI="%r13";
-$fp="%r14";
+$count="%r14";
 $sp="%r15";
 
 $code.=<<___;
@@ -57,44 +62,46 @@ $code.=<<___;
 bn_mul_mont:
 	lgf	$num,164($sp)	# pull $num
 	sla	$num,3		# $num to enumerate bytes
-	la	$rp,0($num,$rp)	# pointers to point at the vectors' ends
-	la	$ap,0($num,$ap)
 	la	$bp,0($num,$bp)
-	la	$np,0($num,$np)
 
 	stmg	%r2,%r15,16($sp)
 
 	cghi	$num,16		#
 	lghi	%r2,0		#
 	blr	%r14		# if($num<16) return 0;
+	cghi	$num,128	#
+	bhr	%r14		# if($num>128) return 0;
 
-	lcgr	$num,$num	# -$num
+	lghi	$rp,-160-8	# leave room for carry bit
+	lcgr	$j,$num		# -$num
 	lgr	%r0,$sp
-	lgr	$fp,$sp
+	la	$rp,0($rp,$sp)
-	aghi	$fp,-160-8	# leave room for carry bit
+	la	$sp,0($j,$rp)	# alloca
-	la	$sp,0($num,$fp)	# alloca
+	stg	%r0,0($sp)	# back chain
-	stg	%r0,0($sp)
-	aghi	$fp,160-8	# $fp to point at tp[$num-1]
 
-	la	$bp,0($num,$bp)	# restore $bp
+	sra	$num,3		# restore $num
+	la	$bp,0($j,$bp)	# restore $bp
+	ahi	$num,-1		# adjust $num for inner loop
 	lg	$n0,0($n0)	# pull n0
 
 	lg	$bi,0($bp)
-	lg	$alo,0($num,$ap)
+	lg	$alo,0($ap)
 	mlgr	$ahi,$bi	# ap[0]*bp[0]
 	lgr	$AHI,$ahi
 
 	lgr	$mn0,$alo	# "tp[0]"*n0
 	msgr	$mn0,$n0
 
-	lg	$nlo,0($num,$np)#
+	lg	$nlo,0($np)	#
 	mlgr	$nhi,$mn0	# np[0]*m1
 	algr	$nlo,$alo	# +="tp[0]"
 	lghi	$NHI,0
 	alcgr	$NHI,$nhi
 
-	lgr	$j,$num
+	la	$j,8(%r0)	# j=1
-	aghi	$j,8		# j=1
+	lr	$count,$num
+
+.align	16
 .L1st:
 	lg	$alo,0($j,$ap)
 	mlgr	$ahi,$bi	# ap[j]*bp[0]
@@ -110,43 +117,45 @@ bn_mul_mont:
 	algr	$nlo,$alo
 	alcgr	$NHI,$nhi
 
-	stg	$nlo,0($j,$fp)	# tp[j-1]=
+	stg	$nlo,160-8($j,$sp)	# tp[j-1]=
-	aghi	$j,8		# j++
+	la	$j,8($j)	# j++
-	jnz	.L1st
+	brct	$count,.L1st
 
 	algr	$NHI,$AHI
 	lghi	$AHI,0
 	alcgr	$AHI,$AHI	# upmost overflow bit
-	stg	$NHI,0($fp)
+	stg	$NHI,160-8($j,$sp)
-	stg	$AHI,8($fp)
+	stg	$AHI,160($j,$sp)
 	la	$bp,8($bp)	# bp++
 
 .Louter:
 	lg	$bi,0($bp)	# bp[i]
-	lg	$alo,0($num,$ap)
+	lg	$alo,0($ap)
 	mlgr	$ahi,$bi	# ap[0]*bp[i]
-	alg	$alo,8($num,$fp)# +=tp[0]
+	alg	$alo,160($sp)	# +=tp[0]
 	lghi	$AHI,0
 	alcgr	$AHI,$ahi
 
 	lgr	$mn0,$alo
 	msgr	$mn0,$n0	# tp[0]*n0
 
-	lg	$nlo,0($num,$np)# np[0]
+	lg	$nlo,0($np)	# np[0]
 	mlgr	$nhi,$mn0	# np[0]*m1
 	algr	$nlo,$alo	# +="tp[0]"
 	lghi	$NHI,0
 	alcgr	$NHI,$nhi
 
-	lgr	$j,$num
+	la	$j,8(%r0)	# j=1
-	aghi	$j,8		# j=1
+	lr	$count,$num
+
+.align	16
 .Linner:
 	lg	$alo,0($j,$ap)
 	mlgr	$ahi,$bi	# ap[j]*bp[i]
 	algr	$alo,$AHI
 	lghi	$AHI,0
 	alcgr	$ahi,$AHI
-	alg	$alo,8($j,$fp)	# +=tp[j]
+	alg	$alo,160($j,$sp)# +=tp[j]
 	alcgr	$AHI,$ahi
 
 	lg	$nlo,0($j,$np)
@@ -157,34 +166,29 @@ bn_mul_mont:
 	algr	$nlo,$alo	# +="tp[j]"
 	alcgr	$NHI,$nhi
 
-	stg	$nlo,0($j,$fp)	# tp[j-1]=
+	stg	$nlo,160-8($j,$sp)	# tp[j-1]=
-	aghi	$j,8		# j++
+	la	$j,8($j)	# j++
-	jnz	.Linner
+	brct	$count,.Linner
 
 	algr	$NHI,$AHI
 	lghi	$AHI,0
 	alcgr	$AHI,$AHI
-	alg	$NHI,8($fp)	# accumulate previous upmost overflow bit
+	alg	$NHI,160($j,$sp)# accumulate previous upmost overflow bit
 	lghi	$ahi,0
 	alcgr	$AHI,$ahi	# new upmost overflow bit
-	stg	$NHI,0($fp)
+	stg	$NHI,160-8($j,$sp)
-	stg	$AHI,8($fp)
+	stg	$AHI,160($j,$sp)
 
 	la	$bp,8($bp)	# bp++
-	clg	$bp,16+32($fp)	# compare to &bp[num]
+	clg	$bp,160+8+32($j,$sp)	# compare to &bp[num]
 	jne	.Louter
-___
 
-undef $bi;
+	lg	$rp,160+8+16($j,$sp)	# reincarnate rp
-$count=$bp; undef $bp;
+	la	$ap,160($sp)
+	ahi	$num,1		# restore $num, incidentally clears "borrow"
 
-$code.=<<___;
+	la	$j,0(%r0)
-	lg	$rp,16+16($fp)	# reincarnate rp
+	lr	$count,$num
-	la	$ap,8($fp)
-	lgr	$j,$num
-
-	lcgr	$count,$num
-	sra	$count,3	# incidentally clears "borrow"
 .Lsub:	lg	$alo,0($j,$ap)
 	slbg	$alo,0($j,$np)
 	stg	$alo,0($j,$rp)
@@ -198,15 +202,17 @@ $code.=<<___;
 	xgr	$np,$AHI
 	ngr	$np,$rp
 	ogr	$ap,$np		# ap=borrow?tp:rp
-	lgr	$j,$num
 
+	la	$j,0(%r0)
+	lgr	$count,$num
 .Lcopy:	lg	$alo,0($j,$ap)	# copy or in-place refresh
-	stg	$j,8($j,$fp)	# zap tp
+	stg	$j,160($j,$sp)	# zap tp
 	stg	$alo,0($j,$rp)
-	aghi	$j,8
+	la	$j,8($j)
-	jnz	.Lcopy
+	brct	$count,.Lcopy
 
-	lmg	%r6,%r15,16+48($fp)
+	la	%r1,160+8+48($j,$sp)
+	lmg	%r6,%r15,0(%r1)
 	lghi	%r2,1		# signal "processed"
 	br	%r14
 .size	bn_mul_mont,.-bn_mul_mont

crypto/s390xcpuid.S

@@ -0,0 +1,83 @@
+.text
+
+.globl	OPENSSL_s390x_facilities
+.type	OPENSSL_s390x_facilities,@function
+.align	16
+OPENSSL_s390x_facilities:
+	lghi	%r0,0
+	.long	0xb2b0f010	# stfle	16(%r15)
+	lg	%r2,16(%r15)
+	br	%r14
+.size	OPENSSL_s390x_facilities,.-OPENSSL_s390x_facilities
+
+.globl	OPENSSL_rdtsc
+.type	OPENSSL_rdtsc,@function
+.align	16
+OPENSSL_rdtsc:
+	stck	16(%r15)
+	lg	%r2,16(%r15)
+	br	%r14
+.size	OPENSSL_rdtsc,.-OPENSSL_rdtsc
+
+.globl	OPENSSL_atomic_add
+.type	OPENSSL_atomic_add,@function
+.align	16
+OPENSSL_atomic_add:
+	l	%r1,0(%r2)
+.Lspin:	lr	%r0,%r1
+	ar	%r0,%r3
+	cs	%r1,%r0,0(%r2)
+	brc	4,.Lspin
+	lgfr	%r2,%r0		# OpenSSL expects the new value
+	br	%r14
+.size	OPENSSL_atomic_add,.-OPENSSL_atomic_add
+
+.globl	OPENSSL_wipe_cpu
+.type	OPENSSL_wipe_cpu,@function
+.align	16
+OPENSSL_wipe_cpu:
+	xgr	%r0,%r0
+	xgr	%r1,%r1
+	lgr	%r2,%r15
+	xgr	%r3,%r3
+	xgr	%r4,%r4
+	lzdr	%f0
+	lzdr	%f1
+	lzdr	%f2
+	lzdr	%f3
+	lzdr	%f4
+	lzdr	%f5
+	lzdr	%f6
+	lzdr	%f7
+	br	%r14
+.size	OPENSSL_wipe_cpu,.-OPENSSL_wipe_cpu
+
+.globl	OPENSSL_cleanse
+.type	OPENSSL_cleanse,@function
+.align	16
+OPENSSL_cleanse:
+	lghi	%r4,15
+	lghi	%r0,0
+	clgr	%r3,%r4
+	jh	.Lot
+.Little:
+	stc	%r0,0(%r2)
+	la	%r2,1(%r2)
+	brctg	%r3,.Little
+	br	%r14
+.align	4
+.Lot:	tmll	%r2,7
+	jz	.Laligned
+	stc	%r0,0(%r2)
+	la	%r2,1(%r2)
+	brctg	%r3,.Lot
+.Laligned:
+	srlg	%r4,%r3,3
+.Loop:	stg	%r0,0(%r2)
+	la	%r2,8(%r2)
+	brctg	%r4,.Loop
+	lghi	%r4,7
+	ngr	%r3,%r4
+	jnz	.Little
+	br	%r14
+.size	OPENSSL_cleanse,.-OPENSSL_cleanse

crypto/sha/asm/sha1-s390x.pl

@@ -15,14 +15,20 @@
 # twist is that SHA1 hardware support is detected and utilized. In
 # which case performance can reach further >4.5x for larger chunks.
 
+# January 2009.
+#
+# Optimize Xupdate for amount of memory references and reschedule
+# instructions to favour dual-issue z10 pipeline. On z10 hardware is
+# "only" ~2.3x faster than software.
+
 $kimdfunc=1;	# magic function code for kimd instruction
 
 $output=shift;
 open STDOUT,">$output";
 
-$t0="%r0";
+$K_00_39="%r0"; $K=$K_00_39;
-$t1="%r1";
+$K_40_79="%r1";
-$ctx="%r2";
+$ctx="%r2";	$prefetch="%r2";
 $inp="%r3";
 $len="%r4";
 
@@ -31,119 +37,107 @@ $B="%r6";
 $C="%r7";
 $D="%r8";
 $E="%r9";	@V=($A,$B,$C,$D,$E);
-$K_00_19="%r10";
+$t0="%r10";
-$K_20_39="%r11";
+$t1="%r11";
-$K_40_59="%r12";
+@X=("%r12","%r13","%r14");
-$K_60_79="%r13";
-$Xi="%r14";
 $sp="%r15";
 
 $frame=160+16*4;
 
-sub BODY_00_15 {
-my ($i,$a,$b,$c,$d,$e)=@_;
-my $xi=($i&1)?$Xi:$t1;
-
-$code.=<<___ if ($i<16 && !($i&1));
-	lg	$Xi,`$i*4`($inp)
-___
-$code.=<<___;
-	alr	$e,$K_00_19	### $i
-	rll	$t0,$a,5
-	alr	$e,$t0
-	lr	$t0,$d
-	xr	$t0,$c
-	nr	$t0,$b
-	xr	$t0,$d
-	alr	$e,$t0
-	rll	$b,$b,30
-___
-$code.=<<___ if ($i<16 && !($i&1));
-	srlg	$xi,$Xi,32
-	stg	$Xi,`160+$i*4`($sp)
-___
-$code.=<<___;
-	alr	$e,$xi
-___
-}
-
 sub Xupdate {
 my $i=shift;
 
+$code.=<<___ if ($i==15);
+	lg	$prefetch,160($sp)	### Xupdate(16) warm-up
+	lr	$X[0],$X[2]
+___
 return if ($i&1);	# Xupdate is vectorized and executed every 2nd cycle
-$code.=<<___;
+$code.=<<___ if ($i<16);
-	lg	$Xi,`160+4*($i%16)`($sp)	### Xupdate($i)
+	lg	$X[0],`$i*4`($inp)	### Xload($i)
-	xg	$Xi,`160+4*(($i+2)%16)`($sp)
+	rllg	$X[1],$X[0],32
-	xg	$Xi,`160+4*(($i+8)%16)`($sp)
 ___
-if ((($i+13)%16)==15) {
+$code.=<<___ if ($i>=16);
-$code.=<<___;
+	xgr	$X[0],$prefetch		### Xupdate($i)
-	llgf	$t0,`160+4*15`($sp)
+	lg	$prefetch,`160+4*(($i+2)%16)`($sp)
-	x	$Xi,`160+0`($sp)
+	xg	$X[0],`160+4*(($i+8)%16)`($sp)
-	sllg	$t0,$t0,32
+	xgr	$X[0],$prefetch
-	xgr	$Xi,$t0
+	rll	$X[0],$X[0],1
+	rllg	$X[1],$X[0],32
+	rll	$X[1],$X[1],1
+	rllg	$X[0],$X[1],32
+	lr	$X[2],$X[1]		# feedback
 ___
-} else {
+$code.=<<___ if ($i<=70);
-$code.=<<___;
+	stg	$X[0],`160+4*($i%16)`($sp)
-	xg	$Xi,`160+4*(($i+13)%16)`($sp)
-___
-}
-$code.=<<___;
-	rll	$Xi,$Xi,1
-	rllg	$t1,$Xi,32
-	rll	$t1,$t1,1
-	rllg	$Xi,$t1,32
-	stg	$Xi,`160+4*($i%16)`($sp)
 ___
+unshift(@X,pop(@X));
 }
 
-sub BODY_16_19 {
+sub BODY_00_19 {
-	&Xupdate(@_[0]);
+my ($i,$a,$b,$c,$d,$e)=@_;
-	&BODY_00_15(@_);
+my $xi=$X[1];
+
+	&Xupdate($i);
+$code.=<<___;
+	alr	$e,$K		### $i
+	rll	$t1,$a,5
+	lr	$t0,$d
+	xr	$t0,$c
+	alr	$e,$t1
+	nr	$t0,$b
+	alr	$e,$xi
+	xr	$t0,$d
+	rll	$b,$b,30
+	alr	$e,$t0
+___
 }
 
 sub BODY_20_39 {
 my ($i,$a,$b,$c,$d,$e)=@_;
-my $xi=($i&1)?$Xi:$t1;
+my $xi=$X[1];
-my $K_XX_XX=($i<40)?$K_20_39:$K_60_79;
 
 	&Xupdate($i);
 $code.=<<___;
-	alr	$e,$K_XX_XX	### $i
+	alr	$e,$K		### $i
-	rll	$t0,$a,5
+	rll	$t1,$a,5
-	alr	$e,$t0
 	lr	$t0,$b
+	alr	$e,$t1
 	xr	$t0,$c
-	xr	$t0,$d
-	alr	$e,$t0
-	rll	$b,$b,30
 	alr	$e,$xi
+	xr	$t0,$d
+	rll	$b,$b,30
+	alr	$e,$t0
 ___
 }
 
 sub BODY_40_59 {
 my ($i,$a,$b,$c,$d,$e)=@_;
-my $xi=($i&1)?$Xi:$t1;
+my $xi=$X[1];
 
 	&Xupdate($i);
 $code.=<<___;
-	alr	$e,$K_40_59	### $i
+	alr	$e,$K		### $i
-	rll	$t0,$a,5
+	rll	$t1,$a,5
-	alr	$e,$t0
 	lr	$t0,$b
+	alr	$e,$t1
 	or	$t0,$c
-	nr	$t0,$d
-	alr	$e,$xi
 	lr	$t1,$b
+	nr	$t0,$d
 	nr	$t1,$c
+	alr	$e,$xi
 	or	$t0,$t1
-	alr	$e,$t0
 	rll	$b,$b,30
+	alr	$e,$t0
 ___
 }
 
 $code.=<<___;
 .text
+.align	64
+.type	Ktable,\@object
+Ktable: .long	0x5a827999,0x6ed9eba1,0x8f1bbcdc,0xca62c1d6
+	.skip	48	#.long	0,0,0,0,0,0,0,0,0,0,0,0
+.size	Ktable,.-Ktable
 .globl	sha1_block_data_order
 .type	sha1_block_data_order,\@function
 sha1_block_data_order:
@@ -165,37 +159,43 @@ $code.=<<___ if ($kimdfunc);
 .Lsoftware:
 ___
 $code.=<<___;
+	lghi	%r1,-$frame
+	stg	$ctx,16($sp)
 	stmg	%r6,%r15,48($sp)
 	lgr	%r0,$sp
-	aghi	$sp,-$frame
+	la	$sp,0(%r1,$sp)
 	stg	%r0,0($sp)
 
-	sllg	$len,$len,6
+	larl	$t0,Ktable
-	la	$len,0($inp,$len)
-
 	llgf	$A,0($ctx)
 	llgf	$B,4($ctx)
 	llgf	$C,8($ctx)
 	llgf	$D,12($ctx)
 	llgf	$E,16($ctx)
 
-	llilh	$K_00_19,0x5a82
+	lg	$K_00_39,0($t0)
-	oill	$K_00_19,0x7999
+	lg	$K_40_79,8($t0)
-	llilh	$K_20_39,0x6ed9
+
-	oill	$K_20_39,0xeba1
-	llilh	$K_40_59,0x8f1b
-	oill	$K_40_59,0xbcdc
-	llilh	$K_60_79,0xca62
-	oill	$K_60_79,0xc1d6
 .Lloop:
+	rllg	$K_00_39,$K_00_39,32
+___
+for ($i=0;$i<20;$i++)	{ &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+	rllg	$K_00_39,$K_00_39,32
 ___
-for ($i=0;$i<16;$i++)	{ &BODY_00_15($i,@V); unshift(@V,pop(@V)); }
-for (;$i<20;$i++)	{ &BODY_16_19($i,@V); unshift(@V,pop(@V)); }
 for (;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;	$K=$K_40_79;
+	rllg	$K_40_79,$K_40_79,32
+___
 for (;$i<60;$i++)	{ &BODY_40_59($i,@V); unshift(@V,pop(@V)); }
+$code.=<<___;
+	rllg	$K_40_79,$K_40_79,32
+___
 for (;$i<80;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
 $code.=<<___;
 
+	lg	$ctx,`$frame+16`($sp)
+	la	$inp,64($inp)
 	al	$A,0($ctx)
 	al	$B,4($ctx)
 	al	$C,8($ctx)
@@ -206,9 +206,7 @@ $code.=<<___;
 	st	$C,8($ctx)
 	st	$D,12($ctx)
 	st	$E,16($ctx)
-	la	$inp,64($inp)
+	brct	$len,.Lloop
-	clgr	$inp,$len
-	jne	.Lloop
 
 	lmg	%r6,%r15,`$frame+48`($sp)
 	br	%r14

crypto/sha/asm/sha512-s390x.pl

@@ -20,9 +20,15 @@
 #
 # sha512_block_data_order is ~70% faster than gcc 3.3 generated code.
 
+# January 2009.
+#
+# Add support for hardware SHA512 and reschedule instructions to
+# favour dual-issue z10 pipeline. Hardware SHA256/512 is ~4.7x faster
+# than software.
+
 $t0="%r0";
 $t1="%r1";
-$ctx="%r2";
+$ctx="%r2";	$t2="%r2";
 $inp="%r3";
 $len="%r4";	# used as index in inner loop
 
@@ -54,7 +60,7 @@ if ($output =~ /512/) {
 	@sigma0=(56,63, 7);
 	@sigma1=( 3,45, 6);
 	$rounds=80;
-	$kimdfunc=0;	# 0 means unknown/unsupported/unimplemented
+	$kimdfunc=3;	# 0 means unknown/unsupported/unimplemented/disabled
 } else {
 	$label="256";
 	$SZ=4;
@@ -83,32 +89,32 @@ ___
 $code.=<<___;
 	$ROT	$t0,$e,$Sigma1[0]
 	$ROT	$t1,$e,$Sigma1[1]
+	 lgr	$t2,$f
 	xgr	$t0,$t1
 	$ROT	$t1,$t1,`$Sigma1[2]-$Sigma1[1]`
-	xgr	$t0,$t1			# Sigma1(e)
+	 xgr	$t2,$g
 	$ST	$T1,`160+$SZ*($i%16)`($sp)
+	xgr	$t0,$t1			# Sigma1(e)
+	la	$T1,0($T1,$h)		# T1+=h
+	 ngr	$t2,$e
+	 lgr	$t1,$a
 	algr	$T1,$t0			# T1+=Sigma1(e)
-	algr	$T1,$h			# T1+=h
-	$ADD	$T1,`$i*$SZ`($len,$tbl)	# T1+=K[i]
-	lgr	$t0,$f
-	xgr	$t0,$g
-	ngr	$t0,$e
-	xgr	$t0,$g			# Ch(e,f,g)
-	algr	$T1,$t0			# T1+=Ch(e,f,g)
 	$ROT	$h,$a,$Sigma0[0]
+	 xgr	$t2,$g			# Ch(e,f,g)
+	$ADD	$T1,`$i*$SZ`($len,$tbl)	# T1+=K[i]
 	$ROT	$t0,$a,$Sigma0[1]
+	algr	$T1,$t2			# T1+=Ch(e,f,g)
+	 ogr	$t1,$b
 	xgr	$h,$t0
+	 lgr	$t2,$a
+	 ngr	$t1,$c
 	$ROT	$t0,$t0,`$Sigma0[2]-$Sigma0[1]`
 	xgr	$h,$t0			# h=Sigma0(a)
-	lgr	$t0,$a
+	 ngr	$t2,$b
-	ogr	$t0,$b
-	ngr	$t0,$c
-	lgr	$t1,$a
-	ngr	$t1,$b
-	ogr	$t0,$t1			# Maj(a,b,c)
-	algr	$h,$t0			# h+=Maj(a,b,c)
-	algr	$d,$T1			# d+=T1
 	algr	$h,$T1			# h+=T1
+	 ogr	$t2,$t1			# Maj(a,b,c)
+	la	$d,0($d,$T1)		# d+=T1
+	algr	$h,$t2			# h+=Maj(a,b,c)
 ___
 }
 
@@ -120,15 +126,15 @@ $code.=<<___;
 	$LD	$t1,`160+$SZ*(($i+14)%16)`($sp)
 	$ROT	$t0,$T1,$sigma0[0]
 	$SHR	$T1,$sigma0[2]
+	$ROT	$t2,$t0,`$sigma0[1]-$sigma0[0]`
 	xgr	$T1,$t0
-	$ROT	$t0,$t0,`$sigma0[1]-$sigma0[0]`
-	xgr	$T1,$t0				# sigma0(X[i+1])
 	$ROT	$t0,$t1,$sigma1[0]
-	$ADD	$T1,`160+$SZ*($i%16)`($sp)	# +=X[i]
+	xgr	$T1,$t2				# sigma0(X[i+1])
 	$SHR	$t1,$sigma1[2]
+	$ADD	$T1,`160+$SZ*($i%16)`($sp)	# +=X[i]
 	xgr	$t1,$t0
-	$ADD	$T1,`160+$SZ*(($i+9)%16)`($sp)	# +=X[i+9]
 	$ROT	$t0,$t0,`$sigma1[1]-$sigma1[0]`
+	$ADD	$T1,`160+$SZ*(($i+9)%16)`($sp)	# +=X[i+9]
 	xgr	$t1,$t0				# sigma1(X[i+14])
 	algr	$T1,$t1				# +=sigma1(X[i+14])
 ___
@@ -225,15 +231,14 @@ $code.=<<___ if ($kimdfunc);
 ___
 $code.=<<___;
 	sllg	$len,$len,`log(16*$SZ)/log(2)`
-	la	$len,0($inp,$len)
+	lghi	%r1,-$frame
-	stmg	$len,%r15,32($sp)
+	agr	$len,$inp
+	stmg	$ctx,%r15,16($sp)
 	lgr	%r0,$sp
-	aghi	$sp,-$frame
+	la	$sp,0(%r1,$sp)
 	stg	%r0,0($sp)
 
-	bras	$tbl,.Lpic
+	larl	$tbl,$Table
-.Lpic:	aghi	$tbl,$Table-.Lpic
-
 	$LD	$A,`0*$SZ`($ctx)
 	$LD	$B,`1*$SZ`($ctx)
 	$LD	$C,`2*$SZ`($ctx)
@@ -255,6 +260,8 @@ $code.=<<___;
 	clgr	$len,$t0
 	jne	.Lrounds_16_xx
 
+	lg	$ctx,`$frame+16`($sp)
+	la	$inp,`16*$SZ`($inp)
 	$ADD	$A,`0*$SZ`($ctx)
 	$ADD	$B,`1*$SZ`($ctx)
 	$ADD	$C,`2*$SZ`($ctx)
@@ -271,7 +278,6 @@ $code.=<<___;
 	$ST	$F,`5*$SZ`($ctx)
 	$ST	$G,`6*$SZ`($ctx)
 	$ST	$H,`7*$SZ`($ctx)
-	la	$inp,`16*$SZ`($inp)
 	clg	$inp,`$frame+32`($sp)
 	jne	.Lloop