Add aes-mips.pl assembler module.

1 files changed, 1606 insertions, 0 deletions

diff --git a/crypto/aes/asm/aes-mips.pl b/crypto/aes/asm/aes-mips.pl
new file mode 100644
index 0000000000..8431f9fcfb
--- /dev/null
+++ b/crypto/aes/asm/aes-mips.pl
@@ -0,0 +1,1606 @@
+#!/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/.
+# ====================================================================
+
+# AES for MIPS
+
+# October 2010
+#
+# Code uses 1K[+256B] S-box and on single-issue core [such as R5000]
+# spends ~68 cycles per byte processed with 128-bit key. This is ~16%
+# faster than gcc-generated code, which is not very impressive. But
+# recall that compressed S-box requires extra processing, namely
+# additional rotations. Rotations are implemented with lwl/lwr pairs,
+# which is normally used for loading unaligned data. Another cool
+# thing about this module is its endian neutrality, which means that
+# it processes data without ever changing byte order...
+
+######################################################################
+# There is a number of MIPS ABI in use, O32 and N32/64 are most
+# widely used. Then there is a new contender: NUBI. It appears that if
+# one picks the latter, it's possible to arrange code in ABI neutral
+# manner. Therefore let's stick to NUBI register layout:
+#
+($zero,$at,$t0,$t1,$t2)=map("\$$_",(0..2,24,25));
+($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
+($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7,$s8,$s9,$s10,$s11)=map("\$$_",(12..23));
+($gp,$tp,$sp,$fp,$ra)=map("\$$_",(3,28..31));
+#
+# The return value is placed in $a0. Following coding rules facilitate
+# interoperability:
+#
+# - never ever touch $tp, "thread pointer", former $gp;
+# - copy return value to $t0, former $v0 [or to $a0 if you're adapting
+#   old code];
+# - on O32 populate $a4-$a7 with 'lw $aN,4*N($sp)' if necessary;
+#
+# For reference here is register layout for N32/64 MIPS ABIs:
+#
+# ($zero,$at,$v0,$v1)=map("\$$_",(0..3));
+# ($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7)=map("\$$_",(4..11));
+# ($t0,$t1,$t2,$t3,$t8,$t9)=map("\$$_",(12..15,24,25));
+# ($s0,$s1,$s2,$s3,$s4,$s5,$s6,$s7)=map("\$$_",(16..23));
+# ($gp,$sp,$fp,$ra)=map("\$$_",(28..31));
+#
+$flavour = shift; # supported flavours are o32,n32,64,nubi32,nubi64
+
+if ($flavour =~ /64|n32/i) {
+	$PTR_ADD="dadd";	# incidentally works even on n32
+	$PTR_SUB="dsub";	# incidentally works even on n32
+	$REG_S="sd";
+	$REG_L="ld";
+	$PTR_SLL="dsll";	# incidentally works even on n32
+	$SZREG=8;
+} else {
+	$PTR_ADD="add";
+	$PTR_SUB="sub";
+	$REG_S="sw";
+	$REG_L="lw";
+	$PTR_SLL="sll";
+	$SZREG=4;
+}
+$pf = ($flavour =~ /nubi/i) ? $t0 : $t2;
+#
+# <appro@openssl.org>
+#
+######################################################################
+
+for (@ARGV) {	$big_endian=1 if (/\-DB_ENDIAN/);
+		$big_endian=0 if (/\-DL_ENDIAN/);
+		$output=$_ if (/^\w[\w\-]*\.\w+$/);	}
+open STDOUT,">$output";
+
+if (!defined($big_endian))
+{    $big_endian=(unpack('L',pack('N',1))==1);   }
+
+while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
+open STDOUT,">$output";
+
+my ($MSB,$LSB)=(0,3);	# automatically converted to little-endian
+
+$code.=<<___;
+.text
+
+.option	pic2
+.set	noat
+___
+
+{{{
+my $FRAMESIZE=16*$SZREG;
+my $SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc0fff008 : 0xc0ff0000;
+
+my ($inp,$out,$key,$Tbl,$s0,$s1,$s2,$s3)=($a0,$a1,$a2,$a3,$a4,$a5,$a6,$a7);
+my ($i0,$i1,$i2,$i3)=($at,$t0,$t1,$t2);
+my ($t0,$t1,$t2,$t3,$t4,$t5,$t6,$t7,$t8,$t9,$t10,$t11) = map("\$$_",(12..23));
+my ($key0,$cnt)=($gp,$fp);
+
+# instuction ordering is "stolen" from output from MIPSpro assembler
+# invoked with -mips3 -O3 arguments...
+$code.=<<___;
+.align	5
+.ent	_mips_AES_encrypt
+_mips_AES_encrypt:
+	.frame	$sp,0,$ra
+	.set	reorder
+	lw	$t0,0($key)
+	lw	$t1,4($key)
+	lw	$t2,8($key)
+	lw	$t3,12($key)
+	lw	$cnt,240($key)
+	$PTR_ADD $key0,$key,16
+
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+
+	sub	$cnt,1
+	_xtr	$i0,$s1,16-2
+.Loop_enc:
+	_xtr	$i1,$s2,16-2
+	_xtr	$i2,$s3,16-2
+	_xtr	$i3,$s0,16-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t0,3($i0)		# Te1[s1>>16]
+	lwl	$t1,3($i1)		# Te1[s2>>16]
+	lwl	$t2,3($i2)		# Te1[s3>>16]
+	lwl	$t3,3($i3)		# Te1[s0>>16]
+	lwr	$t0,2($i0)		# Te1[s1>>16]
+	lwr	$t1,2($i1)		# Te1[s2>>16]
+	lwr	$t2,2($i2)		# Te1[s3>>16]
+	lwr	$t3,2($i3)		# Te1[s0>>16]
+
+	_xtr	$i0,$s2,8-2
+	_xtr	$i1,$s3,8-2
+	_xtr	$i2,$s0,8-2
+	_xtr	$i3,$s1,8-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t4,2($i0)		# Te2[s2>>8]
+	lwl	$t5,2($i1)		# Te2[s3>>8]
+	lwl	$t6,2($i2)		# Te2[s0>>8]
+	lwl	$t7,2($i3)		# Te2[s1>>8]
+	lwr	$t4,1($i0)		# Te2[s2>>8]
+	lwr	$t5,1($i1)		# Te2[s3>>8]
+	lwr	$t6,1($i2)		# Te2[s0>>8]
+	lwr	$t7,1($i3)		# Te2[s1>>8]
+
+	_xtr	$i0,$s3,0-2
+	_xtr	$i1,$s0,0-2
+	_xtr	$i2,$s1,0-2
+	_xtr	$i3,$s2,0-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t8,1($i0)		# Te3[s3]
+	lwl	$t9,1($i1)		# Te3[s0]
+	lwl	$t10,1($i2)		# Te3[s1]
+	lwl	$t11,1($i3)		# Te3[s2]
+	lwr	$t8,0($i0)		# Te3[s3]
+	lwr	$t9,0($i1)		# Te3[s0]
+	lwr	$t10,0($i2)		# Te3[s1]
+	lwr	$t11,0($i3)		# Te3[s2]
+
+	_xtr	$i0,$s0,24-2
+	_xtr	$i1,$s1,24-2
+	_xtr	$i2,$s2,24-2
+	_xtr	$i3,$s3,24-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+	lw	$t4,0($i0)		# Te0[s0>>24]
+	lw	$t5,0($i1)		# Te0[s1>>24]
+	lw	$t6,0($i2)		# Te0[s2>>24]
+	lw	$t7,0($i3)		# Te0[s3>>24]
+
+	lw	$s0,0($key0)
+	lw	$s1,4($key0)
+	lw	$s2,8($key0)
+	lw	$s3,12($key0)
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	sub	$cnt,1
+	$PTR_ADD $key0,16
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+	.set	noreorder
+	bnez	$cnt,.Loop_enc
+	_xtr	$i0,$s1,16-2
+
+	.set	reorder
+	_xtr	$i1,$s2,16-2
+	_xtr	$i2,$s3,16-2
+	_xtr	$i3,$s0,16-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t0,2($i0)		# Te4[s1>>16]
+	lbu	$t1,2($i1)		# Te4[s2>>16]
+	lbu	$t2,2($i2)		# Te4[s3>>16]
+	lbu	$t3,2($i3)		# Te4[s0>>16]
+
+	_xtr	$i0,$s2,8-2
+	_xtr	$i1,$s3,8-2
+	_xtr	$i2,$s0,8-2
+	_xtr	$i3,$s1,8-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t4,2($i0)		# Te4[s2>>8]
+	lbu	$t5,2($i1)		# Te4[s3>>8]
+	lbu	$t6,2($i2)		# Te4[s0>>8]
+	lbu	$t7,2($i3)		# Te4[s1>>8]
+
+	_xtr	$i0,$s0,24-2
+	_xtr	$i1,$s1,24-2
+	_xtr	$i2,$s2,24-2
+	_xtr	$i3,$s3,24-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t8,2($i0)		# Te4[s0>>24]
+	lbu	$t9,2($i1)		# Te4[s1>>24]
+	lbu	$t10,2($i2)		# Te4[s2>>24]
+	lbu	$t11,2($i3)		# Te4[s3>>24]
+
+	_xtr	$i0,$s3,0-2
+	_xtr	$i1,$s0,0-2
+	_xtr	$i2,$s1,0-2
+	_xtr	$i3,$s2,0-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+
+	_ins	$t0,16
+	_ins	$t1,16
+	_ins	$t2,16
+	_ins	$t3,16
+
+	_ins	$t4,8
+	_ins	$t5,8
+	_ins	$t6,8
+	_ins	$t7,8
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t4,2($i0)		# Te4[s3]
+	lbu	$t5,2($i1)		# Te4[s0]
+	lbu	$t6,2($i2)		# Te4[s1]
+	lbu	$t7,2($i3)		# Te4[s2]
+
+	_ins	$t8,24
+	_ins	$t9,24
+	_ins	$t10,24
+	_ins	$t11,24
+
+	lw	$s0,0($key0)
+	lw	$s1,4($key0)
+	lw	$s2,8($key0)
+	lw	$s3,12($key0)
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	_ins	$t4,0
+	_ins	$t5,0
+	_ins	$t6,0
+	_ins	$t7,0
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+
+	jr	$ra
+.end	_mips_AES_encrypt
+
+.align	5
+.globl	AES_encrypt
+.ent	AES_encrypt
+AES_encrypt:
+	.frame	$sp,$FRAMESIZE,$ra
+	.mask	$SAVED_REGS_MASK,-$SZREG
+	.set	noreorder
+___
+$code.=<<___ if ($flavour =~ /o32/i);	# o32 PIC-ification
+	.cpload	$pf
+___
+$code.=<<___;
+	$PTR_SUB $sp,$FRAMESIZE
+	$REG_S	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_S	$fp,$FRAMESIZE-2*$SZREG($sp)
+	$REG_S	$s11,$FRAMESIZE-3*$SZREG($sp)
+	$REG_S	$s10,$FRAMESIZE-4*$SZREG($sp)
+	$REG_S	$s9,$FRAMESIZE-5*$SZREG($sp)
+	$REG_S	$s8,$FRAMESIZE-6*$SZREG($sp)
+	$REG_S	$s7,$FRAMESIZE-7*$SZREG($sp)
+	$REG_S	$s6,$FRAMESIZE-8*$SZREG($sp)
+	$REG_S	$s5,$FRAMESIZE-9*$SZREG($sp)
+	$REG_S	$s4,$FRAMESIZE-10*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
+	$REG_S	\$15,$FRAMESIZE-11*$SZREG($sp)
+	$REG_S	\$14,$FRAMESIZE-12*$SZREG($sp)
+	$REG_S	\$13,$FRAMESIZE-13*$SZREG($sp)
+	$REG_S	\$12,$FRAMESIZE-14*$SZREG($sp)
+	$REG_S	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
+	.cplocal	$Tbl
+	.cpsetup	$pf,$zero,AES_encrypt
+___
+$code.=<<___;
+	.set	reorder
+	la	$Tbl,AES_Te		# PIC-ified 'load address'
+
+	lwl	$s0,0+$MSB($inp)
+	lwl	$s1,4+$MSB($inp)
+	lwl	$s2,8+$MSB($inp)
+	lwl	$s3,12+$MSB($inp)
+	lwr	$s0,0+$LSB($inp)
+	lwr	$s1,4+$LSB($inp)
+	lwr	$s2,8+$LSB($inp)
+	lwr	$s3,12+$LSB($inp)
+
+	bal	_mips_AES_encrypt
+
+	swr	$s0,0+$LSB($out)
+	swr	$s1,4+$LSB($out)
+	swr	$s2,8+$LSB($out)
+	swr	$s3,12+$LSB($out)
+	swl	$s0,0+$MSB($out)
+	swl	$s1,4+$MSB($out)
+	swl	$s2,8+$MSB($out)
+	swl	$s3,12+$MSB($out)
+
+	.set	noreorder
+	$REG_L	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_L	$fp,$FRAMESIZE-2*$SZREG($sp)
+	$REG_L	$s11,$FRAMESIZE-3*$SZREG($sp)
+	$REG_L	$s10,$FRAMESIZE-4*$SZREG($sp)
+	$REG_L	$s9,$FRAMESIZE-5*$SZREG($sp)
+	$REG_L	$s8,$FRAMESIZE-6*$SZREG($sp)
+	$REG_L	$s7,$FRAMESIZE-7*$SZREG($sp)
+	$REG_L	$s6,$FRAMESIZE-8*$SZREG($sp)
+	$REG_L	$s5,$FRAMESIZE-9*$SZREG($sp)
+	$REG_L	$s4,$FRAMESIZE-10*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);
+	$REG_L	\$15,$FRAMESIZE-11*$SZREG($sp)
+	$REG_L	\$14,$FRAMESIZE-12*$SZREG($sp)
+	$REG_L	\$13,$FRAMESIZE-13*$SZREG($sp)
+	$REG_L	\$12,$FRAMESIZE-14*$SZREG($sp)
+	$REG_L	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___;
+	jr	$ra
+	$PTR_ADD $sp,$FRAMESIZE
+.end	AES_encrypt
+___
+
+$code.=<<___;
+.align	5
+.ent	_mips_AES_decrypt
+_mips_AES_decrypt:
+	.frame	$sp,0,$ra
+	.set	reorder
+	lw	$t0,0($key)
+	lw	$t1,4($key)
+	lw	$t2,8($key)
+	lw	$t3,12($key)
+	lw	$cnt,240($key)
+	$PTR_ADD $key0,$key,16
+
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+
+	sub	$cnt,1
+	_xtr	$i0,$s3,16-2
+.Loop_dec:
+	_xtr	$i1,$s0,16-2
+	_xtr	$i2,$s1,16-2
+	_xtr	$i3,$s2,16-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t0,3($i0)		# Td1[s3>>16]
+	lwl	$t1,3($i1)		# Td1[s0>>16]
+	lwl	$t2,3($i2)		# Td1[s1>>16]
+	lwl	$t3,3($i3)		# Td1[s2>>16]
+	lwr	$t0,2($i0)		# Td1[s3>>16]
+	lwr	$t1,2($i1)		# Td1[s0>>16]
+	lwr	$t2,2($i2)		# Td1[s1>>16]
+	lwr	$t3,2($i3)		# Td1[s2>>16]
+
+	_xtr	$i0,$s2,8-2
+	_xtr	$i1,$s3,8-2
+	_xtr	$i2,$s0,8-2
+	_xtr	$i3,$s1,8-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t4,2($i0)		# Td2[s2>>8]
+	lwl	$t5,2($i1)		# Td2[s3>>8]
+	lwl	$t6,2($i2)		# Td2[s0>>8]
+	lwl	$t7,2($i3)		# Td2[s1>>8]
+	lwr	$t4,1($i0)		# Td2[s2>>8]
+	lwr	$t5,1($i1)		# Td2[s3>>8]
+	lwr	$t6,1($i2)		# Td2[s0>>8]
+	lwr	$t7,1($i3)		# Td2[s1>>8]
+
+	_xtr	$i0,$s1,0-2
+	_xtr	$i1,$s2,0-2
+	_xtr	$i2,$s3,0-2
+	_xtr	$i3,$s0,0-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lwl	$t8,1($i0)		# Td3[s1]
+	lwl	$t9,1($i1)		# Td3[s2]
+	lwl	$t10,1($i2)		# Td3[s3]
+	lwl	$t11,1($i3)		# Td3[s0]
+	lwr	$t8,0($i0)		# Td3[s1]
+	lwr	$t9,0($i1)		# Td3[s2]
+	lwr	$t10,0($i2)		# Td3[s3]
+	lwr	$t11,0($i3)		# Td3[s0]
+
+	_xtr	$i0,$s0,24-2
+	_xtr	$i1,$s1,24-2
+	_xtr	$i2,$s2,24-2
+	_xtr	$i3,$s3,24-2
+	and	$i0,0x3fc
+	and	$i1,0x3fc
+	and	$i2,0x3fc
+	and	$i3,0x3fc
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+
+	lw	$t4,0($i0)		# Td0[s0>>24]
+	lw	$t5,0($i1)		# Td0[s1>>24]
+	lw	$t6,0($i2)		# Td0[s2>>24]
+	lw	$t7,0($i3)		# Td0[s3>>24]
+
+	lw	$s0,0($key0)
+	lw	$s1,4($key0)
+	lw	$s2,8($key0)
+	lw	$s3,12($key0)
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	sub	$cnt,1
+	$PTR_ADD $key0,16
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+	.set	noreorder
+	bnez	$cnt,.Loop_dec
+	_xtr	$i0,$s3,16-2
+
+	.set	reorder
+	lw	$t4,1024($Tbl)		# prefetch Td4
+	lw	$t5,1024+32($Tbl)
+	lw	$t6,1024+64($Tbl)
+	lw	$t7,1024+96($Tbl)
+	lw	$t8,1024+128($Tbl)
+	lw	$t9,1024+160($Tbl)
+	lw	$t10,1024+192($Tbl)
+	lw	$t11,1024+224($Tbl)
+
+	_xtr	$i0,$s3,16
+	_xtr	$i1,$s0,16
+	_xtr	$i2,$s1,16
+	_xtr	$i3,$s2,16
+	and	$i0,0xff
+	and	$i1,0xff
+	and	$i2,0xff
+	and	$i3,0xff
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t0,1024($i0)		# Td4[s3>>16]
+	lbu	$t1,1024($i1)		# Td4[s0>>16]
+	lbu	$t2,1024($i2)		# Td4[s1>>16]
+	lbu	$t3,1024($i3)		# Td4[s2>>16]
+
+	_xtr	$i0,$s2,8
+	_xtr	$i1,$s3,8
+	_xtr	$i2,$s0,8
+	_xtr	$i3,$s1,8
+	and	$i0,0xff
+	and	$i1,0xff
+	and	$i2,0xff
+	and	$i3,0xff
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t4,1024($i0)		# Td4[s2>>8]
+	lbu	$t5,1024($i1)		# Td4[s3>>8]
+	lbu	$t6,1024($i2)		# Td4[s0>>8]
+	lbu	$t7,1024($i3)		# Td4[s1>>8]
+
+	_xtr	$i0,$s0,24
+	_xtr	$i1,$s1,24
+	_xtr	$i2,$s2,24
+	_xtr	$i3,$s3,24
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t8,1024($i0)		# Td4[s0>>24]
+	lbu	$t9,1024($i1)		# Td4[s1>>24]
+	lbu	$t10,1024($i2)		# Td4[s2>>24]
+	lbu	$t11,1024($i3)		# Td4[s3>>24]
+
+	_xtr	$i0,$s1,0
+	_xtr	$i1,$s2,0
+	_xtr	$i2,$s3,0
+	_xtr	$i3,$s0,0
+
+	_ins	$t0,16
+	_ins	$t1,16
+	_ins	$t2,16
+	_ins	$t3,16
+
+	_ins	$t4,8
+	_ins	$t5,8
+	_ins	$t6,8
+	_ins	$t7,8
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$t4,1024($i0)		# Td4[s1]
+	lbu	$t5,1024($i1)		# Td4[s2]
+	lbu	$t6,1024($i2)		# Td4[s3]
+	lbu	$t7,1024($i3)		# Td4[s0]
+
+	_ins	$t8,24
+	_ins	$t9,24
+	_ins	$t10,24
+	_ins	$t11,24
+
+	lw	$s0,0($key0)
+	lw	$s1,4($key0)
+	lw	$s2,8($key0)
+	lw	$s3,12($key0)
+
+	_ins	$t4,0
+	_ins	$t5,0
+	_ins	$t6,0
+	_ins	$t7,0
+
+
+	xor	$t0,$t8
+	xor	$t1,$t9
+	xor	$t2,$t10
+	xor	$t3,$t11
+
+	xor	$t0,$t4
+	xor	$t1,$t5
+	xor	$t2,$t6
+	xor	$t3,$t7
+
+	xor	$s0,$t0
+	xor	$s1,$t1
+	xor	$s2,$t2
+	xor	$s3,$t3
+
+	jr	$ra
+.end	_mips_AES_decrypt
+
+.align	5
+.globl	AES_decrypt
+.ent	AES_decrypt
+AES_decrypt:
+	.frame	$sp,$FRAMESIZE,$ra
+	.mask	$SAVED_REGS_MASK,-$SZREG
+	.set	noreorder
+___
+$code.=<<___ if ($flavour =~ /o32/i);	# o32 PIC-ification
+	.cpload	$pf
+___
+$code.=<<___;
+	$PTR_SUB $sp,$FRAMESIZE
+	$REG_S	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_S	$fp,$FRAMESIZE-2*$SZREG($sp)
+	$REG_S	$s11,$FRAMESIZE-3*$SZREG($sp)
+	$REG_S	$s10,$FRAMESIZE-4*$SZREG($sp)
+	$REG_S	$s9,$FRAMESIZE-5*$SZREG($sp)
+	$REG_S	$s8,$FRAMESIZE-6*$SZREG($sp)
+	$REG_S	$s7,$FRAMESIZE-7*$SZREG($sp)
+	$REG_S	$s6,$FRAMESIZE-8*$SZREG($sp)
+	$REG_S	$s5,$FRAMESIZE-9*$SZREG($sp)
+	$REG_S	$s4,$FRAMESIZE-10*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
+	$REG_S	\$15,$FRAMESIZE-11*$SZREG($sp)
+	$REG_S	\$14,$FRAMESIZE-12*$SZREG($sp)
+	$REG_S	\$13,$FRAMESIZE-13*$SZREG($sp)
+	$REG_S	\$12,$FRAMESIZE-14*$SZREG($sp)
+	$REG_S	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
+	.cplocal	$Tbl
+	.cpsetup	$pf,$zero,AES_decrypt
+___
+$code.=<<___;
+	.set	reorder
+	la	$Tbl,AES_Td		# PIC-ified 'load address'
+
+	lwl	$s0,0+$MSB($inp)
+	lwl	$s1,4+$MSB($inp)
+	lwl	$s2,8+$MSB($inp)
+	lwl	$s3,12+$MSB($inp)
+	lwr	$s0,0+$LSB($inp)
+	lwr	$s1,4+$LSB($inp)
+	lwr	$s2,8+$LSB($inp)
+	lwr	$s3,12+$LSB($inp)
+
+	bal	_mips_AES_decrypt
+
+	swr	$s0,0+$LSB($out)
+	swr	$s1,4+$LSB($out)
+	swr	$s2,8+$LSB($out)
+	swr	$s3,12+$LSB($out)
+	swl	$s0,0+$MSB($out)
+	swl	$s1,4+$MSB($out)
+	swl	$s2,8+$MSB($out)
+	swl	$s3,12+$MSB($out)
+
+	.set	noreorder
+	$REG_L	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_L	$fp,$FRAMESIZE-2*$SZREG($sp)
+	$REG_L	$s11,$FRAMESIZE-3*$SZREG($sp)
+	$REG_L	$s10,$FRAMESIZE-4*$SZREG($sp)
+	$REG_L	$s9,$FRAMESIZE-5*$SZREG($sp)
+	$REG_L	$s8,$FRAMESIZE-6*$SZREG($sp)
+	$REG_L	$s7,$FRAMESIZE-7*$SZREG($sp)
+	$REG_L	$s6,$FRAMESIZE-8*$SZREG($sp)
+	$REG_L	$s5,$FRAMESIZE-9*$SZREG($sp)
+	$REG_L	$s4,$FRAMESIZE-10*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);
+	$REG_L	\$15,$FRAMESIZE-11*$SZREG($sp)
+	$REG_L	\$14,$FRAMESIZE-12*$SZREG($sp)
+	$REG_L	\$13,$FRAMESIZE-13*$SZREG($sp)
+	$REG_L	\$12,$FRAMESIZE-14*$SZREG($sp)
+	$REG_L	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___;
+	jr	$ra
+	$PTR_ADD $sp,$FRAMESIZE
+.end	AES_decrypt
+___
+}}}
+
+{{{
+my $FRAMESIZE=8*$SZREG;
+my $SAVED_REGS_MASK = ($flavour =~ /nubi/i) ? 0xc000f008 : 0xc0000000;
+
+my ($inp,$bits,$key,$Tbl)=($a0,$a1,$a2,$a3);
+my ($rk0,$rk1,$rk2,$rk3,$rk4,$rk5,$rk6,$rk7)=($a4,$a5,$a6,$a7,$s0,$s1,$s2,$s3);
+my ($i0,$i1,$i2,$i3)=($at,$t0,$t1,$t2);
+my ($rcon,$cnt)=($gp,$fp);
+
+$code.=<<___;
+.align	5
+.ent	_mips_AES_set_encrypt_key
+_mips_AES_set_encrypt_key:
+	.frame	$sp,0,$ra
+	.set	noreorder
+	beqz	$inp,.Lekey_done
+	li	$t0,-1
+	beqz	$key,.Lekey_done
+	$PTR_ADD $rcon,$Tbl,1024+256
+
+	.set	reorder
+	lwl	$rk0,0+$MSB($inp)	# load 128 bits
+	lwl	$rk1,4+$MSB($inp)
+	lwl	$rk2,8+$MSB($inp)
+	lwl	$rk3,12+$MSB($inp)
+	li	$at,128
+	lwr	$rk0,0+$LSB($inp)
+	lwr	$rk1,4+$LSB($inp)
+	lwr	$rk2,8+$LSB($inp)
+	lwr	$rk3,12+$LSB($inp)
+	.set	noreorder
+	beq	$bits,$at,.L128bits
+	li	$cnt,10
+
+	.set	reorder
+	lwl	$rk4,16+$MSB($inp)	# load 192 bits
+	lwl	$rk5,20+$MSB($inp)
+	li	$at,192
+	lwr	$rk4,16+$LSB($inp)
+	lwr	$rk5,20+$LSB($inp)
+	.set	noreorder
+	beq	$bits,$at,.L192bits
+	li	$cnt,8
+
+	.set	reorder
+	lwl	$rk6,24+$MSB($inp)	# load 256 bits
+	lwl	$rk7,28+$MSB($inp)
+	li	$at,256
+	lwr	$rk6,24+$LSB($inp)
+	lwr	$rk7,28+$LSB($inp)
+	.set	noreorder
+	beq	$bits,$at,.L256bits
+	li	$cnt,7
+
+	b	.Lekey_done
+	li	$t0,-2
+
+.align	4
+.L128bits:
+	.set	reorder
+	srl	$i0,$rk3,16
+	srl	$i1,$rk3,8
+	and	$i0,0xff
+	and	$i1,0xff
+	and	$i2,$rk3,0xff
+	srl	$i3,$rk3,24
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$i0,1024($i0)
+	lbu	$i1,1024($i1)
+	lbu	$i2,1024($i2)
+	lbu	$i3,1024($i3)
+
+	sw	$rk0,0($key)
+	sw	$rk1,4($key)
+	sw	$rk2,8($key)
+	sw	$rk3,12($key)
+	sub	$cnt,1
+	$PTR_ADD $key,16
+
+	_bias	$i0,24
+	_bias	$i1,16
+	_bias	$i2,8
+	_bias	$i3,0
+
+	xor	$rk0,$i0
+	lw	$i0,0($rcon)
+	xor	$rk0,$i1
+	xor	$rk0,$i2
+	xor	$rk0,$i3
+	xor	$rk0,$i0
+
+	xor	$rk1,$rk0
+	xor	$rk2,$rk1
+	xor	$rk3,$rk2
+
+	.set	noreorder
+	bnez	$cnt,.L128bits
+	$PTR_ADD $rcon,4
+
+	sw	$rk0,0($key)
+	sw	$rk1,4($key)
+	sw	$rk2,8($key)
+	li	$cnt,10
+	sw	$rk3,12($key)
+	li	$t0,0
+	sw	$cnt,80($key)
+	b	.Lekey_done
+	$PTR_SUB $key,10*16
+
+.align	4
+.L192bits:
+	.set	reorder
+	srl	$i0,$rk5,16
+	srl	$i1,$rk5,8
+	and	$i0,0xff
+	and	$i1,0xff
+	and	$i2,$rk5,0xff
+	srl	$i3,$rk5,24
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$i0,1024($i0)
+	lbu	$i1,1024($i1)
+	lbu	$i2,1024($i2)
+	lbu	$i3,1024($i3)
+
+	sw	$rk0,0($key)
+	sw	$rk1,4($key)
+	sw	$rk2,8($key)
+	sw	$rk3,12($key)
+	sw	$rk4,16($key)
+	sw	$rk5,20($key)
+	sub	$cnt,1
+	$PTR_ADD $key,24
+
+	_bias	$i0,24
+	_bias	$i1,16
+	_bias	$i2,8
+	_bias	$i3,0
+
+	xor	$rk0,$i0
+	lw	$i0,0($rcon)
+	xor	$rk0,$i1
+	xor	$rk0,$i2
+	xor	$rk0,$i3
+	xor	$rk0,$i0
+
+	xor	$rk1,$rk0
+	xor	$rk2,$rk1
+	xor	$rk3,$rk2
+	xor	$rk4,$rk3
+	xor	$rk5,$rk4
+
+	.set	noreorder
+	bnez	$cnt,.L192bits
+	$PTR_ADD $rcon,4
+
+	sw	$rk0,0($key)
+	sw	$rk1,4($key)
+	sw	$rk2,8($key)
+	li	$cnt,12
+	sw	$rk3,12($key)
+	li	$t0,0
+	sw	$cnt,48($key)
+	b	.Lekey_done
+	$PTR_SUB $key,12*16
+
+.align	4
+.L256bits:
+	.set	reorder
+	srl	$i0,$rk7,16
+	srl	$i1,$rk7,8
+	and	$i0,0xff
+	and	$i1,0xff
+	and	$i2,$rk7,0xff
+	srl	$i3,$rk7,24
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$i0,1024($i0)
+	lbu	$i1,1024($i1)
+	lbu	$i2,1024($i2)
+	lbu	$i3,1024($i3)
+
+	sw	$rk0,0($key)
+	sw	$rk1,4($key)
+	sw	$rk2,8($key)
+	sw	$rk3,12($key)
+	sw	$rk4,16($key)
+	sw	$rk5,20($key)
+	sw	$rk6,24($key)
+	sw	$rk7,28($key)
+	sub	$cnt,1
+
+	_bias	$i0,24
+	_bias	$i1,16
+	_bias	$i2,8
+	_bias	$i3,0
+
+	xor	$rk0,$i0
+	lw	$i0,0($rcon)
+	xor	$rk0,$i1
+	xor	$rk0,$i2
+	xor	$rk0,$i3
+	xor	$rk0,$i0
+
+	xor	$rk1,$rk0
+	xor	$rk2,$rk1
+	xor	$rk3,$rk2
+	beqz	$cnt,.L256bits_done
+
+	srl	$i0,$rk3,24
+	srl	$i1,$rk3,16
+	srl	$i2,$rk3,8
+	and	$i3,$rk3,0xff
+	and	$i1,0xff
+	and	$i2,0xff
+	$PTR_ADD $i0,$Tbl
+	$PTR_ADD $i1,$Tbl
+	$PTR_ADD $i2,$Tbl
+	$PTR_ADD $i3,$Tbl
+	lbu	$i0,1024($i0)
+	lbu	$i1,1024($i1)
+	lbu	$i2,1024($i2)
+	lbu	$i3,1024($i3)
+	sll	$i0,24
+	sll	$i1,16
+	sll	$i2,8
+
+	xor	$rk4,$i0
+	xor	$rk4,$i1
+	xor	$rk4,$i2
+	xor	$rk4,$i3
+
+	xor	$rk5,$rk4
+	xor	$rk6,$rk5
+	xor	$rk7,$rk6
+
+	$PTR_ADD $key,32
+	.set	noreorder
+	b	.L256bits
+	$PTR_ADD $rcon,4
+
+.L256bits_done:
+	sw	$rk0,32($key)
+	sw	$rk1,36($key)
+	sw	$rk2,40($key)
+	li	$cnt,14
+	sw	$rk3,44($key)
+	li	$t0,0
+	sw	$cnt,48($key)
+	$PTR_SUB $key,12*16
+
+.Lekey_done:
+	jr	$ra
+	nop
+.end	_mips_AES_set_encrypt_key
+
+.globl	AES_set_encrypt_key
+.ent	AES_set_encrypt_key
+AES_set_encrypt_key:
+	.frame	$sp,$FRAMESIZE,$ra
+	.mask	$SAVED_REGS_MASK,-$SZREG
+	.set	noreorder
+___
+$code.=<<___ if ($flavour =~ /o32/i);	# o32 PIC-ification
+	.cpload	$pf
+___
+$code.=<<___;
+	$PTR_SUB $sp,$FRAMESIZE
+	$REG_S	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_S	$fp,$FRAMESIZE-2*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
+	$REG_S	$s3,$FRAMESIZE-3*$SZREG($sp)
+	$REG_S	$s2,$FRAMESIZE-4*$SZREG($sp)
+	$REG_S	$s1,$FRAMESIZE-5*$SZREG($sp)
+	$REG_S	$s0,$FRAMESIZE-6*$SZREG($sp)
+	$REG_S	$gp,$FRAMESIZE-7*$SZREG($sp)
+___
+$code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
+	.cplocal	$Tbl
+	.cpsetup	$pf,$zero,AES_set_encrypt_key
+___
+$code.=<<___;
+	.set	reorder
+	la	$Tbl,AES_Te		# PIC-ified 'load address'
+
+	bal	_mips_AES_set_encrypt_key
+
+	.set	noreorder
+	move	$a0,$t0
+	$REG_L	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_L	$fp,$FRAMESIZE-2*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);
+	$REG_L	$s3,$FRAMESIZE-11*$SZREG($sp)
+	$REG_L	$s2,$FRAMESIZE-12*$SZREG($sp)
+	$REG_L	$s1,$FRAMESIZE-13*$SZREG($sp)
+	$REG_L	$s0,$FRAMESIZE-14*$SZREG($sp)
+	$REG_L	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___;
+	jr	$ra
+	$PTR_ADD $sp,$FRAMESIZE
+.end	AES_set_encrypt_key
+___
+
+my ($head,$tail)=($inp,$bits);
+my ($tp1,$tp2,$tp4,$tp8,$tp9,$tpb,$tpd,$tpe)=($a4,$a5,$a6,$a7,$s0,$s1,$s2,$s3);
+my ($m,$x80808080,$x7f7f7f7f,$x1b1b1b1b)=($at,$t0,$t1,$t2);
+$code.=<<___;
+.align	5
+.globl	AES_set_decrypt_key
+.ent	AES_set_decrypt_key
+AES_set_decrypt_key:
+	.frame	$sp,$FRAMESIZE,$ra
+	.mask	$SAVED_REGS_MASK,-$SZREG
+	.set	noreorder
+___
+$code.=<<___ if ($flavour =~ /o32/i);	# o32 PIC-ification
+	.cpload	$pf
+___
+$code.=<<___;
+	$PTR_SUB $sp,$FRAMESIZE
+	$REG_S	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_S	$fp,$FRAMESIZE-2*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);	# optimize non-nubi prologue
+	$REG_S	$s3,$FRAMESIZE-3*$SZREG($sp)
+	$REG_S	$s2,$FRAMESIZE-4*$SZREG($sp)
+	$REG_S	$s1,$FRAMESIZE-5*$SZREG($sp)
+	$REG_S	$s0,$FRAMESIZE-6*$SZREG($sp)
+	$REG_S	$gp,$FRAMESIZE-7*$SZREG($sp)
+___
+$code.=<<___ if ($flavour !~ /o32/i);	# non-o32 PIC-ification
+	.cplocal	$Tbl
+	.cpsetup	$pf,$zero,AES_set_decrypt_key
+___
+$code.=<<___;
+	.set	reorder
+	la	$Tbl,AES_Te		# PIC-ified 'load address'
+
+	bal	_mips_AES_set_encrypt_key
+
+	bltz	$t0,.Ldkey_done
+
+	sll	$at,$cnt,4
+	$PTR_ADD $head,$key,0
+	$PTR_ADD $tail,$key,$at
+.align	4
+.Lswap:
+	lw	$rk0,0($head)
+	lw	$rk1,4($head)
+	lw	$rk2,8($head)
+	lw	$rk3,12($head)
+	lw	$rk4,0($tail)
+	lw	$rk5,4($tail)
+	lw	$rk6,8($tail)
+	lw	$rk7,12($tail)
+	sw	$rk0,0($tail)
+	sw	$rk1,4($tail)
+	sw	$rk2,8($tail)
+	sw	$rk3,12($tail)
+	$PTR_ADD $head,16
+	$PTR_SUB $tail,16
+	sw	$rk4,-16($head)
+	sw	$rk5,-12($head)
+	sw	$rk6,-8($head)
+	sw	$rk7,-4($head)
+	bne	$head,$tail,.Lswap
+
+	lw	$tp1,16($key)		# modulo-scheduled
+	lui	$x80808080,0x8080
+	sub	$cnt,1
+	or	$x80808080,0x8080
+	sll	$cnt,2
+	$PTR_ADD $key,16
+	lui	$x1b1b1b1b,0x1b1b
+	nor	$x7f7f7f7f,$zero,$x80808080
+	or	$x1b1b1b1b,0x1b1b
+.align	4
+.Lmix:
+	and	$m,$tp1,$x80808080
+	and	$tp2,$tp1,$x7f7f7f7f
+	srl	$tp4,$m,7
+	addu	$tp2,$tp2		# tp2<<1
+	subu	$m,$tp4
+	and	$m,$x1b1b1b1b
+	xor	$tp2,$m
+
+	and	$m,$tp2,$x80808080
+	and	$tp4,$tp2,$x7f7f7f7f
+	srl	$tp8,$m,7
+	addu	$tp4,$tp4		# tp4<<1
+	subu	$m,$tp8
+	and	$m,$x1b1b1b1b
+	xor	$tp4,$m
+
+	and	$m,$tp4,$x80808080
+	and	$tp8,$tp4,$x7f7f7f7f
+	srl	$tp9,$m,7
+	addu	$tp8,$tp8		# tp8<<1
+	subu	$m,$tp9
+	and	$m,$x1b1b1b1b
+	xor	$tp8,$m
+
+	xor	$tp9,$tp8,$tp1
+	xor	$tpe,$tp8,$tp4
+	xor	$tpb,$tp9,$tp2
+	xor	$tpd,$tp9,$tp4
+
+	_ror	$tp1,$tpd,16
+	 xor	$tpe,$tp2
+	_ror	$tp2,$tpd,-16
+	xor	$tpe,$tp1
+	_ror	$tp1,$tp9,8
+	xor	$tpe,$tp2
+	_ror	$tp2,$tp9,-24
+	xor	$tpe,$tp1
+	_ror	$tp1,$tpb,24
+	xor	$tpe,$tp2
+	_ror	$tp2,$tpb,-8
+	xor	$tpe,$tp1
+	lw	$tp1,4($key)		# modulo-scheduled
+	xor	$tpe,$tp2
+	sub	$cnt,1
+	sw	$tpe,0($key)
+	$PTR_ADD $key,4
+	bnez	$cnt,.Lmix
+
+	li	$t0,0
+.Ldkey_done:
+	.set	noreorder
+	move	$a0,$t0
+	$REG_L	$ra,$FRAMESIZE-1*$SZREG($sp)
+	$REG_L	$fp,$FRAMESIZE-2*$SZREG($sp)
+___
+$code.=<<___ if ($flavour =~ /nubi/i);
+	$REG_L	$s3,$FRAMESIZE-11*$SZREG($sp)
+	$REG_L	$s2,$FRAMESIZE-12*$SZREG($sp)
+	$REG_L	$s1,$FRAMESIZE-13*$SZREG($sp)
+	$REG_L	$s0,$FRAMESIZE-14*$SZREG($sp)
+	$REG_L	$gp,$FRAMESIZE-15*$SZREG($sp)
+___
+$code.=<<___;
+	jr	$ra
+	$PTR_ADD $sp,$FRAMESIZE
+.end	AES_set_decrypt_key
+___
+}}}
+
+######################################################################
+# Tables are kept in endian-neutral manner
+$code.=<<___;
+.rdata
+.align	6
+AES_Te:
+.byte	0xc6,0x63,0x63,0xa5,	0xf8,0x7c,0x7c,0x84	# Te0
+.byte	0xee,0x77,0x77,0x99,	0xf6,0x7b,0x7b,0x8d
+.byte	0xff,0xf2,0xf2,0x0d,	0xd6,0x6b,0x6b,0xbd
+.byte	0xde,0x6f,0x6f,0xb1,	0x91,0xc5,0xc5,0x54
+.byte	0x60,0x30,0x30,0x50,	0x02,0x01,0x01,0x03
+.byte	0xce,0x67,0x67,0xa9,	0x56,0x2b,0x2b,0x7d
+.byte	0xe7,0xfe,0xfe,0x19,	0xb5,0xd7,0xd7,0x62
+.byte	0x4d,0xab,0xab,0xe6,	0xec,0x76,0x76,0x9a
+.byte	0x8f,0xca,0xca,0x45,	0x1f,0x82,0x82,0x9d
+.byte	0x89,0xc9,0xc9,0x40,	0xfa,0x7d,0x7d,0x87
+.byte	0xef,0xfa,0xfa,0x15,	0xb2,0x59,0x59,0xeb
+.byte	0x8e,0x47,0x47,0xc9,	0xfb,0xf0,0xf0,0x0b
+.byte	0x41,0xad,0xad,0xec,	0xb3,0xd4,0xd4,0x67
+.byte	0x5f,0xa2,0xa2,0xfd,	0x45,0xaf,0xaf,0xea
+.byte	0x23,0x9c,0x9c,0xbf,	0x53,0xa4,0xa4,0xf7
+.byte	0xe4,0x72,0x72,0x96,	0x9b,0xc0,0xc0,0x5b
+.byte	0x75,0xb7,0xb7,0xc2,	0xe1,0xfd,0xfd,0x1c
+.byte	0x3d,0x93,0x93,0xae,	0x4c,0x26,0x26,0x6a
+.byte	0x6c,0x36,0x36,0x5a,	0x7e,0x3f,0x3f,0x41
+.byte	0xf5,0xf7,0xf7,0x02,	0x83,0xcc,0xcc,0x4f
+.byte	0x68,0x34,0x34,0x5c,	0x51,0xa5,0xa5,0xf4
+.byte	0xd1,0xe5,0xe5,0x34,	0xf9,0xf1,0xf1,0x08
+.byte	0xe2,0x71,0x71,0x93,	0xab,0xd8,0xd8,0x73
+.byte	0x62,0x31,0x31,0x53,	0x2a,0x15,0x15,0x3f
+.byte	0x08,0x04,0x04,0x0c,	0x95,0xc7,0xc7,0x52
+.byte	0x46,0x23,0x23,0x65,	0x9d,0xc3,0xc3,0x5e
+.byte	0x30,0x18,0x18,0x28,	0x37,0x96,0x96,0xa1
+.byte	0x0a,0x05,0x05,0x0f,	0x2f,0x9a,0x9a,0xb5
+.byte	0x0e,0x07,0x07,0x09,	0x24,0x12,0x12,0x36
+.byte	0x1b,0x80,0x80,0x9b,	0xdf,0xe2,0xe2,0x3d
+.byte	0xcd,0xeb,0xeb,0x26,	0x4e,0x27,0x27,0x69
+.byte	0x7f,0xb2,0xb2,0xcd,	0xea,0x75,0x75,0x9f
+.byte	0x12,0x09,0x09,0x1b,	0x1d,0x83,0x83,0x9e
+.byte	0x58,0x2c,0x2c,0x74,	0x34,0x1a,0x1a,0x2e
+.byte	0x36,0x1b,0x1b,0x2d,	0xdc,0x6e,0x6e,0xb2
+.byte	0xb4,0x5a,0x5a,0xee,	0x5b,0xa0,0xa0,0xfb
+.byte	0xa4,0x52,0x52,0xf6,	0x76,0x3b,0x3b,0x4d
+.byte	0xb7,0xd6,0xd6,0x61,	0x7d,0xb3,0xb3,0xce
+.byte	0x52,0x29,0x29,0x7b,	0xdd,0xe3,0xe3,0x3e
+.byte	0x5e,0x2f,0x2f,0x71,	0x13,0x84,0x84,0x97
+.byte	0xa6,0x53,0x53,0xf5,	0xb9,0xd1,0xd1,0x