From: Andy Polyakov <appro@openssl.org>
Date: Sun, 31 Mar 2013 12:32:05 +0000 (+0200)
Subject: Add support for SPARC T4 DES opcode.
X-Git-Tag: master-post-reformat~1362
X-Git-Url: https://git.openssl.org/gitweb/?p=openssl.git;a=commitdiff_plain;h=c5d975a74313268a36b6a6103cd37221724137c2

Add support for SPARC T4 DES opcode.
---

diff --git a/Configure b/Configure
index 24c8dc2665..e5b8c3049b 100755
--- a/Configure
+++ b/Configure
@@ -130,7 +130,7 @@ my $x86_elf_asm="$x86_asm:elf";
 
 my $x86_64_asm="x86_64cpuid.o:x86_64-gcc.o x86_64-mont.o x86_64-mont5.o x86_64-gf2m.o modexp512-x86_64.o::aes-x86_64.o vpaes-x86_64.o bsaes-x86_64.o aesni-x86_64.o aesni-sha1-x86_64.o::md5-x86_64.o:sha1-x86_64.o sha256-x86_64.o sha512-x86_64.o::rc4-x86_64.o rc4-md5-x86_64.o:::wp-x86_64.o:cmll-x86_64.o cmll_misc.o:ghash-x86_64.o aesni-gcm-x86_64.o:e_padlock-x86_64.o";
 my $ia64_asm="ia64cpuid.o:bn-ia64.o ia64-mont.o::aes_core.o aes_cbc.o aes-ia64.o::md5-ia64.o:sha1-ia64.o sha256-ia64.o sha512-ia64.o::rc4-ia64.o rc4_skey.o:::::ghash-ia64.o::void";
-my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o:des_enc-sparc.o fcrypt_b.o:aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o::md5-sparcv9.o:sha1-sparcv9.o sha256-sparcv9.o sha512-sparcv9.o::::::camellia.o cmll_misc.o cmll_cbc.o cmllt4-sparcv9.o:ghash-sparcv9.o::void";
+my $sparcv9_asm="sparcv9cap.o sparccpuid.o:bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o:des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o:aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o::md5-sparcv9.o:sha1-sparcv9.o sha256-sparcv9.o sha512-sparcv9.o::::::camellia.o cmll_misc.o cmll_cbc.o cmllt4-sparcv9.o:ghash-sparcv9.o::void";
 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 $mips64_asm=":bn-mips.o mips-mont.o::aes_cbc.o aes-mips.o:::sha1-mips.o sha256-mips.o sha512-mips.o::::::::";
diff --git a/TABLE b/TABLE
index a3ca775a61..a3a58a4c0a 100644
--- a/TABLE
+++ b/TABLE
@@ -175,7 +175,7 @@ $lflags       =
 $bn_ops       = BN_LLONG RC2_CHAR RC4_CHUNK DES_INT DES_PTR DES_RISC2 BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -2683,7 +2683,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK_LL DES_PTR DES_RISC1 DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -2716,7 +2716,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -4432,7 +4432,7 @@ $lflags       = -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -4663,7 +4663,7 @@ $lflags       = -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_PTR DES_RISC1 DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -5521,7 +5521,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK_LL DES_PTR DES_RISC1 DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -5554,7 +5554,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -5653,7 +5653,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_PTR DES_RISC1 DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
@@ -5686,7 +5686,7 @@ $lflags       = -lsocket -lnsl -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_INT DES_PTR DES_RISC1 DES_UNROLL BF_PTR
 $cpuid_obj    = sparcv9cap.o sparccpuid.o
 $bn_obj       = bn-sparcv9.o sparcv9-mont.o sparcv9a-mont.o vis3-mont.o sparct4-mont.o sparcv9-gf2m.o
-$des_obj      = des_enc-sparc.o fcrypt_b.o
+$des_obj      = des_enc-sparc.o fcrypt_b.o dest4-sparcv9.o
 $aes_obj      = aes_core.o aes_cbc.o aes-sparcv9.o aest4-sparcv9.o
 $bf_obj       = 
 $md5_obj      = md5-sparcv9.o
diff --git a/crypto/des/Makefile b/crypto/des/Makefile
index 868f6087e8..1c061a6879 100644
--- a/crypto/des/Makefile
+++ b/crypto/des/Makefile
@@ -61,6 +61,8 @@ des: des.o cbc3_enc.o lib
 
 des_enc-sparc.S:	asm/des_enc.m4
 	m4 -B 8192 asm/des_enc.m4 > des_enc-sparc.S
+dest4-sparcv9.s:	asm/dest4-sparcv9.pl
+	$(PERL) asm/dest4-sparcv9.pl $(CFLAGS) > $@
 
 des-586.s:	asm/des-586.pl ../perlasm/x86asm.pl ../perlasm/cbc.pl
 	$(PERL) asm/des-586.pl $(PERLASM_SCHEME) $(CFLAGS) > $@
diff --git a/crypto/des/asm/dest4-sparcv9.pl b/crypto/des/asm/dest4-sparcv9.pl
new file mode 100644
index 0000000000..5936a658e5
--- /dev/null
+++ b/crypto/des/asm/dest4-sparcv9.pl
@@ -0,0 +1,594 @@
+#!/usr/bin/env perl
+
+# ====================================================================
+# Written by David S. Miller <davem@devemloft.net> and Andy Polyakov
+# <appro@openssl.org>. The module is licensed under 2-clause BSD
+# license. March 2013. All rights reserved.
+# ====================================================================
+
+######################################################################
+# DES for SPARC T4.
+#
+# As with other hardware-assisted ciphers CBC encrypt results [for
+# aligned data] are virtually identical to critical path lengths:
+#
+#		DES		Triple-DES
+# CBC encrypt	4.14/4.15(*)	11.7/11.7
+# CBC decrypt	1.77/4.11(**)	6.42/7.47
+#
+#			 (*)	numbers after slash are for
+#				misaligned data;
+#			 (**)	this is result for largest
+#				block size, unlike all other
+#				cases smaller blocks results
+#				are better[?];
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../perlasm");
+require "sparcv9_modes.pl";
+
+&asm_init(@ARGV);
+
+$code.=<<___ if ($::abibits==64);
+.register       %g2,#scratch
+.register       %g3,#scratch
+___
+
+$code.=<<___;
+.text
+___
+
+{ my ($inp,$out)=("%o0","%o1");
+
+$code.=<<___;
+.align	32
+.globl	des_t4_key_expand
+.type	des_t4_key_expand,#function
+des_t4_key_expand:
+	andcc		$inp, 0x7, %g0
+	alignaddr	$inp, %g0, $inp
+	bz,pt		%icc, 1f
+	ldd		[$inp + 0x00], %f0
+	ldd		[$inp + 0x08], %f2
+	faligndata	%f0, %f2, %f0
+1:	des_kexpand	%f0, 0, %f0
+	des_kexpand	%f0, 1, %f2
+	std		%f0, [$out + 0x00]
+	des_kexpand	%f2, 3, %f6
+	std		%f2, [$out + 0x08]
+	des_kexpand	%f2, 2, %f4
+	des_kexpand	%f6, 3, %f10
+	std		%f6, [$out + 0x18]
+	des_kexpand	%f6, 2, %f8
+	std		%f4, [$out + 0x10]
+	des_kexpand	%f10, 3, %f14
+	std		%f10, [$out + 0x28]
+	des_kexpand	%f10, 2, %f12
+	std		%f8, [$out + 0x20]
+	des_kexpand	%f14, 1, %f16
+	std		%f14, [$out + 0x38]
+	des_kexpand	%f16, 3, %f20
+	std		%f12, [$out + 0x30]
+	des_kexpand	%f16, 2, %f18
+	std		%f16, [$out + 0x40]
+	des_kexpand	%f20, 3, %f24
+	std		%f20, [$out + 0x50]
+	des_kexpand	%f20, 2, %f22
+	std		%f18, [$out + 0x48]
+	des_kexpand	%f24, 3, %f28
+	std		%f24, [$out + 0x60]
+	des_kexpand	%f24, 2, %f26
+	std		%f22, [$out + 0x58]
+	des_kexpand	%f28, 1, %f30
+	std		%f28, [$out + 0x70]
+	std		%f26, [$out + 0x68]
+	retl
+	std		%f30, [$out + 0x78]
+.size	des_t4_key_expand,.-des_t4_key_expand
+___
+}
+{ my ($inp,$out,$len,$key,$ivec) = map("%o$_",(0..4));
+  my ($ileft,$iright,$omask) = map("%g$_",(1..3));
+
+$code.=<<___;
+.globl	des_t4_cbc_encrypt
+.align	32
+des_t4_cbc_encrypt:
+	ld		[$ivec + 0], %f0	! load ivec
+	ld		[$ivec + 4], %f1
+
+	and		$inp, 7, $ileft
+	andn		$inp, 7, $inp
+	sll		$ileft, 3, $ileft
+	mov		0xff, $omask
+	prefetch	[$inp], 20
+	prefetch	[$inp + 63], 20
+	sub		%g0, $ileft, $iright
+	and		$out, 7, %g4
+	alignaddrl	$out, %g0, $out
+	srl		$omask, %g4, $omask
+	srlx		$len, 3, $len
+	movrz		%g4, 0, $omask
+	prefetch	[$out], 22
+
+	ldd		[$key + 0x00], %f4	! load key schedule
+	ldd		[$key + 0x08], %f6
+	ldd		[$key + 0x10], %f8
+	ldd		[$key + 0x18], %f10
+	ldd		[$key + 0x20], %f12
+	ldd		[$key + 0x28], %f14
+	ldd		[$key + 0x30], %f16
+	ldd		[$key + 0x38], %f18
+	ldd		[$key + 0x40], %f20
+	ldd		[$key + 0x48], %f22
+	ldd		[$key + 0x50], %f24
+	ldd		[$key + 0x58], %f26
+	ldd		[$key + 0x60], %f28
+	ldd		[$key + 0x68], %f30
+	ldd		[$key + 0x70], %f32
+	ldd		[$key + 0x78], %f34
+
+.Ldes_cbc_enc_loop:
+	ldx		[$inp + 0], %g4
+	brz,pt		$ileft, 4f
+	nop
+
+	ldx		[$inp + 8], %g5
+	sllx		%g4, $ileft, %g4
+	srlx		%g5, $iright, %g5
+	or		%g5, %g4, %g4
+4:
+	movxtod		%g4, %f2
+	prefetch	[$inp + 8+63], 20
+	add		$inp, 8, $inp
+	fxor		%f2, %f0, %f0		! ^= ivec
+	prefetch	[$out + 63], 22
+
+	des_ip		%f0, %f0
+	des_round	%f4, %f6, %f0, %f0
+	des_round	%f8, %f10, %f0, %f0
+	des_round	%f12, %f14, %f0, %f0
+	des_round	%f16, %f18, %f0, %f0
+	des_round	%f20, %f22, %f0, %f0
+	des_round	%f24, %f26, %f0, %f0
+	des_round	%f28, %f30, %f0, %f0
+	des_round	%f32, %f34, %f0, %f0
+	des_iip		%f0, %f0
+
+	brnz,pn		$omask, 2f
+	sub		$len, 1, $len
+
+	std		%f0, [$out + 0]
+	brnz,pt		$len, .Ldes_cbc_enc_loop
+	add		$out, 8, $out
+
+	st		%f0, [$ivec + 0]	! write out ivec
+	retl
+	st		%f1, [$ivec + 4]
+
+.align	16
+2:	ldxa		[$inp]0x82, %g4		! avoid read-after-write hazard
+						! and ~4x deterioration
+						! in inp==out case
+	faligndata	%f0, %f0, %f2		! handle unaligned output
+
+	stda		%f8, [$out + $omask]0xc0	! partial store
+	add		$out, 8, $out
+	orn		%g0, $omask, $omask
+	stda		%f8, [$out + $omask]0xc0	! partial store
+
+	brnz,pt		$len, .Ldes_cbc_enc_loop+4
+	orn		%g0, $omask, $omask
+
+	st		%f0, [$ivec + 0]	! write out ivec
+	retl
+	st		%f1, [$ivec + 4]
+.type	des_t4_cbc_encrypt,#function
+.size	des_t4_cbc_encrypt,.-des_t4_cbc_encrypt
+
+.globl	des_t4_cbc_decrypt
+.align	32
+des_t4_cbc_decrypt:
+	ld		[$ivec + 0], %f2	! load ivec
+	ld		[$ivec + 4], %f3
+
+	and		$inp, 7, $ileft
+	andn		$inp, 7, $inp
+	sll		$ileft, 3, $ileft
+	mov		0xff, $omask
+	prefetch	[$inp], 20
+	prefetch	[$inp + 63], 20
+	sub		%g0, $ileft, $iright
+	and		$out, 7, %g4
+	alignaddrl	$out, %g0, $out
+	srl		$omask, %g4, $omask
+	srlx		$len, 3, $len
+	movrz		%g4, 0, $omask
+	prefetch	[$out], 22
+
+	ldd		[$key + 0x78], %f4	! load key schedule
+	ldd		[$key + 0x70], %f6
+	ldd		[$key + 0x68], %f8
+	ldd		[$key + 0x60], %f10
+	ldd		[$key + 0x58], %f12
+	ldd		[$key + 0x50], %f14
+	ldd		[$key + 0x48], %f16
+	ldd		[$key + 0x40], %f18
+	ldd		[$key + 0x38], %f20
+	ldd		[$key + 0x30], %f22
+	ldd		[$key + 0x28], %f24
+	ldd		[$key + 0x20], %f26
+	ldd		[$key + 0x18], %f28
+	ldd		[$key + 0x10], %f30
+	ldd		[$key + 0x08], %f32
+	ldd		[$key + 0x00], %f34
+
+.Ldes_cbc_dec_loop:
+	ldx		[$inp + 0], %g4
+	brz,pt		$ileft, 4f
+	nop
+
+	ldx		[$inp + 8], %g5
+	sllx		%g4, $ileft, %g4
+	srlx		%g5, $iright, %g5
+	or		%g5, %g4, %g4
+4:
+	movxtod		%g4, %f0
+	prefetch	[$inp + 8+63], 20
+	add		$inp, 8, $inp
+	prefetch	[$out + 63], 22
+
+	des_ip		%f0, %f0
+	des_round	%f4, %f6, %f0, %f0
+	des_round	%f8, %f10, %f0, %f0
+	des_round	%f12, %f14, %f0, %f0
+	des_round	%f16, %f18, %f0, %f0
+	des_round	%f20, %f22, %f0, %f0
+	des_round	%f24, %f26, %f0, %f0
+	des_round	%f28, %f30, %f0, %f0
+	des_round	%f32, %f34, %f0, %f0
+	des_iip		%f0, %f0
+
+	fxor		%f2, %f0, %f0		! ^= ivec
+	movxtod		%g4, %f2
+
+	brnz,pn		$omask, 2f
+	sub		$len, 1, $len
+
+	std		%f0, [$out + 0]
+	brnz,pt		$len, .Ldes_cbc_dec_loop
+	add		$out, 8, $out
+
+	st		%f2, [$ivec + 0]	! write out ivec
+	retl
+	st		%f3, [$ivec + 4]
+
+.align	16
+2:	ldxa		[$inp]0x82, %g4		! avoid read-after-write hazard
+						! and ~4x deterioration
+						! in inp==out case
+	faligndata	%f0, %f0, %f0		! handle unaligned output
+
+	stda		%f0, [$out + $omask]0xc0	! partial store
+	add		$out, 8, $out
+	orn		%g0, $omask, $omask
+	stda		%f0, [$out + $omask]0xc0	! partial store
+
+	brnz,pt		$len, .Ldes_cbc_dec_loop+4
+	orn		%g0, $omask, $omask
+
+	st		%f2, [$ivec + 0]	! write out ivec
+	retl
+	st		%f3, [$ivec + 4]
+.type	des_t4_cbc_decrypt,#function
+.size	des_t4_cbc_decrypt,.-des_t4_cbc_decrypt
+
+.globl	des_t4_ede3_cbc_encrypt
+.align	32
+des_t4_ede3_cbc_encrypt:
+	ld		[$ivec + 0], %f0	! load ivec
+	ld		[$ivec + 4], %f1
+
+	and		$inp, 7, $ileft
+	andn		$inp, 7, $inp
+	sll		$ileft, 3, $ileft
+	mov		0xff, $omask
+	prefetch	[$inp], 20
+	prefetch	[$inp + 63], 20
+	sub		%g0, $ileft, $iright
+	and		$out, 7, %g4
+	alignaddrl	$out, %g0, $out
+	srl		$omask, %g4, $omask
+	srlx		$len, 3, $len
+	movrz		%g4, 0, $omask
+	prefetch	[$out], 22
+
+	ldd		[$key + 0x00], %f4	! load key schedule
+	ldd		[$key + 0x08], %f6
+	ldd		[$key + 0x10], %f8
+	ldd		[$key + 0x18], %f10
+	ldd		[$key + 0x20], %f12
+	ldd		[$key + 0x28], %f14
+	ldd		[$key + 0x30], %f16
+	ldd		[$key + 0x38], %f18
+	ldd		[$key + 0x40], %f20
+	ldd		[$key + 0x48], %f22
+	ldd		[$key + 0x50], %f24
+	ldd		[$key + 0x58], %f26
+	ldd		[$key + 0x60], %f28
+	ldd		[$key + 0x68], %f30
+	ldd		[$key + 0x70], %f32
+	ldd		[$key + 0x78], %f34
+
+.Ldes_ede3_cbc_enc_loop:
+	ldx		[$inp + 0], %g4
+	brz,pt		$ileft, 4f
+	nop
+
+	ldx		[$inp + 8], %g5
+	sllx		%g4, $ileft, %g4
+	srlx		%g5, $iright, %g5
+	or		%g5, %g4, %g4
+4:
+	movxtod		%g4, %f2
+	prefetch	[$inp + 8+63], 20
+	add		$inp, 8, $inp
+	fxor		%f2, %f0, %f0		! ^= ivec
+	prefetch	[$out + 63], 22
+
+	des_ip		%f0, %f0
+	des_round	%f4, %f6, %f0, %f0
+	des_round	%f8, %f10, %f0, %f0
+	des_round	%f12, %f14, %f0, %f0
+	des_round	%f16, %f18, %f0, %f0
+	ldd		[$key + 0x100-0x08], %f36
+	ldd		[$key + 0x100-0x10], %f38
+	des_round	%f20, %f22, %f0, %f0
+	ldd		[$key + 0x100-0x18], %f40
+	ldd		[$key + 0x100-0x20], %f42
+	des_round	%f24, %f26, %f0, %f0
+	ldd		[$key + 0x100-0x28], %f44
+	ldd		[$key + 0x100-0x30], %f46
+	des_round	%f28, %f30, %f0, %f0
+	ldd		[$key + 0x100-0x38], %f48
+	ldd		[$key + 0x100-0x40], %f50
+	des_round	%f32, %f34, %f0, %f0
+	ldd		[$key + 0x100-0x48], %f52
+	ldd		[$key + 0x100-0x50], %f54
+	des_iip		%f0, %f0
+
+	ldd		[$key + 0x100-0x58], %f56
+	ldd		[$key + 0x100-0x60], %f58
+	des_ip		%f0, %f0
+	ldd		[$key + 0x100-0x68], %f60
+	ldd		[$key + 0x100-0x70], %f62
+	des_round	%f36, %f38, %f0, %f0
+	ldd		[$key + 0x100-0x78], %f36
+	ldd		[$key + 0x100-0x80], %f38
+	des_round	%f40, %f42, %f0, %f0
+	des_round	%f44, %f46, %f0, %f0
+	des_round	%f48, %f50, %f0, %f0
+	ldd		[$key + 0x100+0x00], %f40
+	ldd		[$key + 0x100+0x08], %f42
+	des_round	%f52, %f54, %f0, %f0
+	ldd		[$key + 0x100+0x10], %f44
+	ldd		[$key + 0x100+0x18], %f46
+	des_round	%f56, %f58, %f0, %f0
+	ldd		[$key + 0x100+0x20], %f48
+	ldd		[$key + 0x100+0x28], %f50
+	des_round	%f60, %f62, %f0, %f0
+	ldd		[$key + 0x100+0x30], %f52
+	ldd		[$key + 0x100+0x38], %f54
+	des_round	%f36, %f38, %f0, %f0
+	ldd		[$key + 0x100+0x40], %f56
+	ldd		[$key + 0x100+0x48], %f58
+	des_iip		%f0, %f0
+
+	ldd		[$key + 0x100+0x50], %f60
+	ldd		[$key + 0x100+0x58], %f62
+	des_ip		%f0, %f0
+	ldd		[$key + 0x100+0x60], %f36
+	ldd		[$key + 0x100+0x68], %f38
+	des_round	%f40, %f42, %f0, %f0
+	ldd		[$key + 0x100+0x70], %f40
+	ldd		[$key + 0x100+0x78], %f42
+	des_round	%f44, %f46, %f0, %f0
+	des_round	%f48, %f50, %f0, %f0
+	des_round	%f52, %f54, %f0, %f0
+	des_round	%f56, %f58, %f0, %f0
+	des_round	%f60, %f62, %f0, %f0
+	des_round	%f36, %f38, %f0, %f0
+	des_round	%f40, %f42, %f0, %f0
+	des_iip		%f0, %f0
+
+	brnz,pn		$omask, 2f
+	sub		$len, 1, $len
+
+	std		%f0, [$out + 0]
+	brnz,pt		$len, .Ldes_ede3_cbc_enc_loop
+	add		$out, 8, $out
+
+	st		%f0, [$ivec + 0]	! write out ivec
+	retl
+	st		%f1, [$ivec + 4]
+
+.align	16
+2:	ldxa		[$inp]0x82, %g4		! avoid read-after-write hazard
+						! and ~2x deterioration
+						! in inp==out case
+	faligndata	%f0, %f0, %f2		! handle unaligned output
+
+	stda		%f2, [$out + $omask]0xc0	! partial store
+	add		$out, 8, $out
+	orn		%g0, $omask, $omask
+	stda		%f2, [$out + $omask]0xc0	! partial store
+
+	brnz,pt		$len, .Ldes_ede3_cbc_enc_loop+4
+	orn		%g0, $omask, $omask
+
+	st		%f0, [$ivec + 0]	! write out ivec
+	retl
+	st		%f1, [$ivec + 4]
+.type	des_t4_ede3_cbc_encrypt,#function
+.size	des_t4_ede3_cbc_encrypt,.-des_t4_ede3_cbc_encrypt
+
+.globl	des_t4_ede3_cbc_decrypt
+.align	32
+des_t4_ede3_cbc_decrypt:
+	ld		[$ivec + 0], %f2	! load ivec
+	ld		[$ivec + 4], %f3
+
+	and		$inp, 7, $ileft
+	andn		$inp, 7, $inp
+	sll		$ileft, 3, $ileft
+	mov		0xff, $omask
+	prefetch	[$inp], 20
+	prefetch	[$inp + 63], 20
+	sub		%g0, $ileft, $iright
+	and		$out, 7, %g4
+	alignaddrl	$out, %g0, $out
+	srl		$omask, %g4, $omask
+	srlx		$len, 3, $len
+	movrz		%g4, 0, $omask
+	prefetch	[$out], 22
+
+	ldd		[$key + 0x100+0x78], %f4	! load key schedule
+	ldd		[$key + 0x100+0x70], %f6
+	ldd		[$key + 0x100+0x68], %f8
+	ldd		[$key + 0x100+0x60], %f10
+	ldd		[$key + 0x100+0x58], %f12
+	ldd		[$key + 0x100+0x50], %f14
+	ldd		[$key + 0x100+0x48], %f16
+	ldd		[$key + 0x100+0x40], %f18
+	ldd		[$key + 0x100+0x38], %f20
+	ldd		[$key + 0x100+0x30], %f22
+	ldd		[$key + 0x100+0x28], %f24
+	ldd		[$key + 0x100+0x20], %f26
+	ldd		[$key + 0x100+0x18], %f28
+	ldd		[$key + 0x100+0x10], %f30
+	ldd		[$key + 0x100+0x08], %f32
+	ldd		[$key + 0x100+0x00], %f34
+
+.Ldes_ede3_cbc_dec_loop:
+	ldx		[$inp + 0], %g4
+	brz,pt		$ileft, 4f
+	nop
+
+	ldx		[$inp + 8], %g5
+	sllx		%g4, $ileft, %g4
+	srlx		%g5, $iright, %g5
+	or		%g5, %g4, %g4
+4:
+	movxtod		%g4, %f0
+	prefetch	[$inp + 8+63], 20
+	add		$inp, 8, $inp
+	prefetch	[$out + 63], 22
+
+	des_ip		%f0, %f0
+	des_round	%f4, %f6, %f0, %f0
+	des_round	%f8, %f10, %f0, %f0
+	des_round	%f12, %f14, %f0, %f0
+	des_round	%f16, %f18, %f0, %f0
+	ldd		[$key + 0x80+0x00], %f36
+	ldd		[$key + 0x80+0x08], %f38
+	des_round	%f20, %f22, %f0, %f0
+	ldd		[$key + 0x80+0x10], %f40
+	ldd		[$key + 0x80+0x18], %f42
+	des_round	%f24, %f26, %f0, %f0
+	ldd		[$key + 0x80+0x20], %f44
+	ldd		[$key + 0x80+0x28], %f46
+	des_round	%f28, %f30, %f0, %f0
+	ldd		[$key + 0x80+0x30], %f48
+	ldd		[$key + 0x80+0x38], %f50
+	des_round	%f32, %f34, %f0, %f0
+	ldd		[$key + 0x80+0x40], %f52
+	ldd		[$key + 0x80+0x48], %f54
+	des_iip		%f0, %f0
+
+	ldd		[$key + 0x80+0x50], %f56
+	ldd		[$key + 0x80+0x58], %f58
+	des_ip		%f0, %f0
+	ldd		[$key + 0x80+0x60], %f60
+	ldd		[$key + 0x80+0x68], %f62
+	des_round	%f36, %f38, %f0, %f0
+	ldd		[$key + 0x80+0x70], %f36
+	ldd		[$key + 0x80+0x78], %f38
+	des_round	%f40, %f42, %f0, %f0
+	des_round	%f44, %f46, %f0, %f0
+	des_round	%f48, %f50, %f0, %f0
+	ldd		[$key + 0x80-0x08], %f40
+	ldd		[$key + 0x80-0x10], %f42
+	des_round	%f52, %f54, %f0, %f0
+	ldd		[$key + 0x80-0x18], %f44
+	ldd		[$key + 0x80-0x20], %f46
+	des_round	%f56, %f58, %f0, %f0
+	ldd		[$key + 0x80-0x28], %f48
+	ldd		[$key + 0x80-0x30], %f50
+	des_round	%f60, %f62, %f0, %f0
+	ldd		[$key + 0x80-0x38], %f52
+	ldd		[$key + 0x80-0x40], %f54
+	des_round	%f36, %f38, %f0, %f0
+	ldd		[$key + 0x80-0x48], %f56
+	ldd		[$key + 0x80-0x50], %f58
+	des_iip		%f0, %f0
+
+	ldd		[$key + 0x80-0x58], %f60
+	ldd		[$key + 0x80-0x60], %f62
+	des_ip		%f0, %f0
+	ldd		[$key + 0x80-0x68], %f36
+	ldd		[$key + 0x80-0x70], %f38
+	des_round	%f40, %f42, %f0, %f0
+	ldd		[$key + 0x80-0x78], %f40
+	ldd		[$key + 0x80-0x80], %f42
+	des_round	%f44, %f46, %f0, %f0
+	des_round	%f48, %f50, %f0, %f0
+	des_round	%f52, %f54, %f0, %f0
+	des_round	%f56, %f58, %f0, %f0
+	des_round	%f60, %f62, %f0, %f0
+	des_round	%f36, %f38, %f0, %f0
+	des_round	%f40, %f42, %f0, %f0
+	des_iip		%f0, %f0
+
+	fxor		%f2, %f0, %f0		! ^= ivec
+	movxtod		%g4, %f2
+
+	brnz,pn		$omask, 2f
+	sub		$len, 1, $len
+
+	std		%f0, [$out + 0]
+	brnz,pt		$len, .Ldes_ede3_cbc_dec_loop
+	add		$out, 8, $out
+
+	st		%f2, [$ivec + 0]	! write out ivec
+	retl
+	st		%f3, [$ivec + 4]
+
+.align	16
+2:	ldxa		[$inp]0x82, %g4		! avoid read-after-write hazard
+						! and ~3x deterioration
+						! in inp==out case
+	faligndata	%f0, %f0, %f0		! handle unaligned output
+
+	stda		%f0, [$out + $omask]0xc0	! partial store
+	add		$out, 8, $out
+	orn		%g0, $omask, $omask
+	stda		%f0, [$out + $omask]0xc0	! partial store
+
+	brnz,pt		$len, .Ldes_ede3_cbc_dec_loop+4
+	orn		%g0, $omask, $omask
+
+	st		%f2, [$ivec + 0]	! write out ivec
+	retl
+	st		%f3, [$ivec + 4]
+.type	des_t4_ede3_cbc_decrypt,#function
+.size	des_t4_ede3_cbc_decrypt,.-des_t4_ede3_cbc_decrypt
+___
+}
+$code.=<<___;
+.asciz  "DES for SPARC T4, David S. Miller, Andy Polyakov"
+.align  4
+___
+
+&emit_assembler();
+
+close STDOUT;
diff --git a/crypto/evp/e_des.c b/crypto/evp/e_des.c
index ca009f2c52..bf4366590f 100644
--- a/crypto/evp/e_des.c
+++ b/crypto/evp/e_des.c
@@ -65,6 +65,30 @@
 #include <openssl/des.h>
 #include <openssl/rand.h>
 
+typedef struct
+	{
+	union { double align; DES_key_schedule ks; } ks;
+	union {
+		void (*cbc)(const void *,void *,size_t,const void *,void *);
+	} stream;
+	} EVP_DES_KEY;
+
+#if defined(AES_ASM) && (defined(__sparc) || defined(__sparc__))
+/* ---------^^^ this is not a typo, just a way to detect that
+ * assembler support was in general requested... */
+#include "sparc_arch.h"
+
+extern unsigned int OPENSSL_sparcv9cap_P[];
+
+#define SPARC_DES_CAPABLE	(OPENSSL_sparcv9cap_P[1] & CFR_DES)
+
+void	des_t4_key_expand(const void *key, DES_key_schedule *ks);
+void	des_t4_cbc_encrypt(const void *inp,void *out,size_t len,
+				DES_key_schedule *ks,unsigned char iv[8]);
+void	des_t4_cbc_decrypt(const void *inp,void *out,size_t len,
+				DES_key_schedule *ks,unsigned char iv[8]);
+#endif
+
 static int des_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 			const unsigned char *iv, int enc);
 static int des_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr);
@@ -99,6 +123,13 @@ static int des_ofb_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 static int des_cbc_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 			  const unsigned char *in, size_t inl)
 {
+	EVP_DES_KEY *dat = (EVP_DES_KEY *)ctx->cipher_data;
+
+	if (dat->stream.cbc)
+		{
+		(*dat->stream.cbc)(in,out,inl,&dat->ks.ks,ctx->iv);
+		return 1;
+		}
 	while(inl>=EVP_MAXCHUNK)
 		{
 		DES_ncbc_encrypt(in, out, (long)EVP_MAXCHUNK, ctx->cipher_data,
@@ -176,18 +207,18 @@ static int des_cfb8_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
     return 1;
     }
 
-BLOCK_CIPHER_defs(des, DES_key_schedule, NID_des, 8, 8, 8, 64,
+BLOCK_CIPHER_defs(des, EVP_DES_KEY, NID_des, 8, 8, 8, 64,
 			EVP_CIPH_RAND_KEY, des_init_key, NULL,
 			EVP_CIPHER_set_asn1_iv,
 			EVP_CIPHER_get_asn1_iv,
 			des_ctrl)
 
-BLOCK_CIPHER_def_cfb(des,DES_key_schedule,NID_des,8,8,1,
+BLOCK_CIPHER_def_cfb(des,EVP_DES_KEY,NID_des,8,8,1,
 		     EVP_CIPH_RAND_KEY, des_init_key,NULL,
 		     EVP_CIPHER_set_asn1_iv,
 		     EVP_CIPHER_get_asn1_iv,des_ctrl)
 
-BLOCK_CIPHER_def_cfb(des,DES_key_schedule,NID_des,8,8,8,
+BLOCK_CIPHER_def_cfb(des,EVP_DES_KEY,NID_des,8,8,8,
 		     EVP_CIPH_RAND_KEY,des_init_key,NULL,
 		     EVP_CIPHER_set_asn1_iv,
 		     EVP_CIPHER_get_asn1_iv,des_ctrl)
@@ -196,8 +227,25 @@ static int des_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 			const unsigned char *iv, int enc)
 	{
 	DES_cblock *deskey = (DES_cblock *)key;
+	EVP_DES_KEY *dat = (EVP_DES_KEY *)ctx->cipher_data;
+
+	dat->stream.cbc = NULL;
+#if defined(SPARC_DES_CAPABLE)
+	if (SPARC_DES_CAPABLE)
+		{
+		int mode = ctx->cipher->flags & EVP_CIPH_MODE;
+
+		if (mode == EVP_CIPH_CBC_MODE)
+			{
+			des_t4_key_expand(key,&dat->ks.ks);
+			dat->stream.cbc = enc ? des_t4_cbc_encrypt :
+						des_t4_cbc_decrypt;
+			return 1;
+			}
+		}
+#endif
 #ifdef EVP_CHECK_DES_KEY
-	if(DES_set_key_checked(deskey,ctx->cipher_data) != 0)
+	if(DES_set_key_checked(deskey,dat->ks.ks) != 0)
 		return 0;
 #else
 	DES_set_key_unchecked(deskey,ctx->cipher_data);
diff --git a/crypto/evp/e_des3.c b/crypto/evp/e_des3.c
index 2f016f0431..6b0bd2b367 100644
--- a/crypto/evp/e_des3.c
+++ b/crypto/evp/e_des3.c
@@ -67,6 +67,33 @@
 #include <openssl/des.h>
 #include <openssl/rand.h>
 
+typedef struct
+	{
+	union { double align; DES_key_schedule ks[3]; } ks;
+	union {
+		void (*cbc)(const void *,void *,size_t,const void *,void *);
+	} stream;
+	} DES_EDE_KEY;
+#define ks1 ks.ks[0]
+#define ks2 ks.ks[1]
+#define ks3 ks.ks[2]
+
+#if defined(AES_ASM) && (defined(__sparc) || defined(__sparc__))
+/* ---------^^^ this is not a typo, just a way to detect that
+ * assembler support was in general requested... */
+#include "sparc_arch.h"
+
+extern unsigned int OPENSSL_sparcv9cap_P[];
+
+#define SPARC_DES_CAPABLE	(OPENSSL_sparcv9cap_P[1] & CFR_DES)
+
+void	des_t4_key_expand(const void *key, DES_key_schedule *ks);
+void	des_t4_ede3_cbc_encrypt(const void *inp,void *out,size_t len,
+				DES_key_schedule *ks,unsigned char iv[8]);
+void	des_t4_ede3_cbc_decrypt(const void *inp,void *out,size_t len,
+				DES_key_schedule *ks,unsigned char iv[8]);
+#endif
+
 static int des_ede_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 			    const unsigned char *iv,int enc);
 
@@ -75,13 +102,6 @@ static int des_ede3_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 
 static int des3_ctrl(EVP_CIPHER_CTX *c, int type, int arg, void *ptr);
 
-typedef struct
-    {
-    DES_key_schedule ks1;/* key schedule */
-    DES_key_schedule ks2;/* key schedule (for ede) */
-    DES_key_schedule ks3;/* key schedule (for ede3) */
-    } DES_EDE_KEY;
-
 #define data(ctx) ((DES_EDE_KEY *)(ctx)->cipher_data)
 
 /* Because of various casts and different args can't use IMPLEMENT_BLOCK_CIPHER */
@@ -121,6 +141,8 @@ static int des_ede_ofb_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 static int des_ede_cbc_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 			      const unsigned char *in, size_t inl)
 {
+	DES_EDE_KEY *dat = data(ctx);
+
 #ifdef KSSL_DEBUG
 	{
         int i;
@@ -132,10 +154,16 @@ static int des_ede_cbc_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 	printf("\n");
 	}
 #endif    /* KSSL_DEBUG */
+	if (dat->stream.cbc)
+		{
+		(*dat->stream.cbc)(in,out,inl,&dat->ks,ctx->iv);
+		return 1;
+		}
+
 	if (inl>=EVP_MAXCHUNK)
 		{
 		DES_ede3_cbc_encrypt(in, out, (long)EVP_MAXCHUNK,
-			     &data(ctx)->ks1, &data(ctx)->ks2, &data(ctx)->ks3,
+			     &dat->ks1, &dat->ks2, &dat->ks3,
 			     (DES_cblock *)ctx->iv, ctx->encrypt);
 		inl-=EVP_MAXCHUNK;
 		in +=EVP_MAXCHUNK;
@@ -143,7 +171,7 @@ static int des_ede_cbc_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 		}
 	if (inl)
 		DES_ede3_cbc_encrypt(in, out, (long)inl,
-			     &data(ctx)->ks1, &data(ctx)->ks2, &data(ctx)->ks3,
+			     &dat->ks1, &dat->ks2, &dat->ks3,
                              (DES_cblock *)ctx->iv, ctx->encrypt);
 	return 1;
 }
@@ -236,16 +264,35 @@ static int des_ede_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 			    const unsigned char *iv, int enc)
 	{
 	DES_cblock *deskey = (DES_cblock *)key;
+	DES_EDE_KEY *dat = data(ctx);
+
+	dat->stream.cbc = NULL;
+#if defined(SPARC_DES_CAPABLE)
+	if (SPARC_DES_CAPABLE)
+		{
+		int mode = ctx->cipher->flags & EVP_CIPH_MODE;
+
+		if (mode == EVP_CIPH_CBC_MODE)
+			{
+			des_t4_key_expand(&deskey[0],&dat->ks1);
+			des_t4_key_expand(&deskey[1],&dat->ks2);
+			memcpy(&dat->ks3,&dat->ks1,sizeof(dat->ks1));
+			dat->stream.cbc = enc ? des_t4_ede3_cbc_encrypt :
+						des_t4_ede3_cbc_decrypt;
+			return 1;
+			}
+		}
+#endif
 #ifdef EVP_CHECK_DES_KEY
-	if (DES_set_key_checked(&deskey[0],&data(ctx)->ks1)
-		!! DES_set_key_checked(&deskey[1],&data(ctx)->ks2))
+	if (DES_set_key_checked(&deskey[0],&dat->ks1)
+		!! DES_set_key_checked(&deskey[1],&dat->ks2))
 		return 0;
 #else
-	DES_set_key_unchecked(&deskey[0],&data(ctx)->ks1);
-	DES_set_key_unchecked(&deskey[1],&data(ctx)->ks2);
+	DES_set_key_unchecked(&deskey[0],&dat->ks1);
+	DES_set_key_unchecked(&deskey[1],&dat->ks2);
 #endif
-	memcpy(&data(ctx)->ks3,&data(ctx)->ks1,
-	       sizeof(data(ctx)->ks1));
+	memcpy(&dat->ks3,&dat->ks1,
+	       sizeof(dat->ks1));
 	return 1;
 	}
 
@@ -253,6 +300,8 @@ static int des_ede3_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 			     const unsigned char *iv, int enc)
 	{
 	DES_cblock *deskey = (DES_cblock *)key;
+	DES_EDE_KEY *dat = data(ctx);
+
 #ifdef KSSL_DEBUG
 	{
         int i;
@@ -264,15 +313,32 @@ static int des_ede3_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
 	}
 #endif	/* KSSL_DEBUG */
 
+	dat->stream.cbc = NULL;
+#if defined(SPARC_DES_CAPABLE)
+	if (SPARC_DES_CAPABLE)
+		{
+		int mode = ctx->cipher->flags & EVP_CIPH_MODE;
+
+		if (mode == EVP_CIPH_CBC_MODE)
+			{
+			des_t4_key_expand(&deskey[0],&dat->ks1);
+			des_t4_key_expand(&deskey[1],&dat->ks2);
+			des_t4_key_expand(&deskey[2],&dat->ks3);
+			dat->stream.cbc = enc ? des_t4_ede3_cbc_encrypt :
+						des_t4_ede3_cbc_decrypt;
+			return 1;
+			}
+		}
+#endif
 #ifdef EVP_CHECK_DES_KEY
-	if (DES_set_key_checked(&deskey[0],&data(ctx)->ks1)
-		|| DES_set_key_checked(&deskey[1],&data(ctx)->ks2)
-		|| DES_set_key_checked(&deskey[2],&data(ctx)->ks3))
+	if (DES_set_key_checked(&deskey[0],&dat->ks1)
+		|| DES_set_key_checked(&deskey[1],&dat->ks2)
+		|| DES_set_key_checked(&deskey[2],&dat->ks3))
 		return 0;
 #else
-	DES_set_key_unchecked(&deskey[0],&data(ctx)->ks1);
-	DES_set_key_unchecked(&deskey[1],&data(ctx)->ks2);
-	DES_set_key_unchecked(&deskey[2],&data(ctx)->ks3);
+	DES_set_key_unchecked(&deskey[0],&dat->ks1);
+	DES_set_key_unchecked(&deskey[1],&dat->ks2);
+	DES_set_key_unchecked(&deskey[2],&dat->ks3);
 #endif
 	return 1;
 	}
diff --git a/crypto/perlasm/sparcv9_modes.pl b/crypto/perlasm/sparcv9_modes.pl
index 363da32d59..a13ffcec2d 100644
--- a/crypto/perlasm/sparcv9_modes.pl
+++ b/crypto/perlasm/sparcv9_modes.pl
@@ -1582,6 +1582,63 @@ my %movxopf = (	"movdtox"	=> 0x110,
     }
 }
 
+sub undes {
+my ($mnemonic)=shift;
+my @args=@_;
+my ($ref,$opf);
+my %desopf = (	"des_round"	=> 0b1001,
+		"des_ip"	=> 0b100110100,
+		"des_iip"	=> 0b100110101,
+		"des_kexpand"	=> 0b100110110	);
+
+    $ref = "$mnemonic\t".join(",",@_);
+
+    if (defined($opf=$desopf{$mnemonic})) {	# 4-arg
+	if ($mnemonic eq "des_round") {
+	    foreach (@args[0..3]) {
+		return $ref if (!/%f([0-9]{1,2})/);
+		$_=$1;
+		if ($1>=32) {
+		    return $ref if ($1&1);
+		    # re-encode for upper double register addressing
+		    $_=($1|$1>>5)&31;
+		}
+	    }
+	    return  sprintf ".word\t0x%08x !%s",
+			    2<<30|0b011001<<19|$opf<<5|$args[0]<<14|$args[1]|$args[2]<<9|$args[3]<<25,
+			    $ref;
+	} elsif ($mnemonic eq "des_kexpand") {	# 3-arg
+	    foreach (@args[0..2]) {
+		return $ref if (!/(%f)?([0-9]{1,2})/);
+		$_=$2;
+		if ($2>=32) {
+		    return $ref if ($2&1);
+		    # re-encode for upper double register addressing
+		    $_=($2|$2>>5)&31;
+		}
+	    }
+	    return  sprintf ".word\t0x%08x !%s",
+			    2<<30|0b110110<<19|$opf<<5|$args[0]<<14|$args[1]|$args[2]<<25,
+			    $ref;
+	} else {				# 2-arg
+	    foreach (@args[0..1]) {
+		return $ref if (!/%f([0-9]{1,2})/);
+		$_=$1;
+		if ($1>=32) {
+		    return $ref if ($2&1);
+		    # re-encode for upper double register addressing
+		    $_=($1|$1>>5)&31;
+		}
+	    }
+	    return  sprintf ".word\t0x%08x !%s",
+			    2<<30|0b110110<<19|$opf<<5|$args[0]<<14|$args[1]<<25,
+			    $ref;
+	}
+    } else {
+	return $ref;
+    }
+}
+
 sub emit_assembler {
     foreach (split("\n",$::code)) {
 	s/\`([^\`]*)\`/eval $1/ge;
@@ -1600,6 +1657,9 @@ sub emit_assembler {
 	s/\b(camellia_[^s]+)\s+(%f[0-9]{1,2}),\s*(%f[0-9]{1,2}),\s*(%f[0-9]{1,2})/
 		&uncamellia3($1,$2,$3,$4)
 	 /geo or
+	s/\b(des_\w+)\s+(?<rs1>%f[0-9]{1,2}),\s*(?<rs2>[%fx0-9]+)(,\s*(?<rs3>%f[0-9]{1,2})(,\s*(?<rs4>%f[0-9]{1,2}))?)?/
+		&undes($1,$+{rs1},$+{rs2},$+{rs3},$+{rs4})
+	 /geo or
 	s/\b(mov[ds]to\w+)\s+(%f[0-9]{1,2}),\s*(%[goli][0-7])/
 		&unmovxtox($1,$2,$3)
 	 /geo or
diff --git a/test/test_t4 b/test/test_t4
index 826e1fd779..5cecb56394 100755
--- a/test/test_t4
+++ b/test/test_t4
@@ -18,7 +18,8 @@ if [ 1 ]; then
 
     HASH=`cat $PROG | $PROG dgst -hex`
 
-    AES_ALGS="	camellia-128-cbc camellia-128-cfb \
+    AES_ALGS="	des-cbc des-ede-cbc des-ede3-cbc \
+		camellia-128-cbc camellia-128-cfb \
 		camellia-192-cbc camellia-192-cfb \
 		camellia-256-cbc camellia-256-cfb \
 		aes-128-ctr aes-128-cbc aes-128-cfb aes-128-ofb \