#!/usr/bin/env perl

# ====================================================================
# Written by Andy Polyakov <appro@fy.chalmers.se> 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/.
# ====================================================================

# Needs more work: key setup, CBC routine...
#
# ppc_AES_[en|de]crypt perform at 18 cycles per byte processed with
# 128-bit key, which is ~40% better than 64-bit code generated by gcc
# 4.0. But these are not the ones currently used! Their "compact"
# counterparts are, for security reason. ppc_AES_encrypt_compact runs
# at 1/2 of ppc_AES_encrypt speed, while ppc_AES_decrypt_compact -
# at 1/3 of ppc_AES_decrypt.

# February 2010
#
# Rescheduling instructions to favour Power6 pipeline gave 10%
# performance improvement on the platform in question (and marginal
# improvement even on others). It should be noted that Power6 fails
# to process byte in 18 cycles, only in 23, because it fails to issue
# 4 load instructions in two cycles, only in 3. As result non-compact
# block subroutines are 25% slower than one would expect. Compact
# functions scale better, because they have pure computational part,
# which scales perfectly with clock frequency. To be specific
# ppc_AES_encrypt_compact operates at 42 cycles per byte, while
# ppc_AES_decrypt_compact - at 55 (in 64-bit build).

$flavour = shift;

if ($flavour =~ /64/) {
	$SIZE_T	=8;
	$LRSAVE	=2*$SIZE_T;
	$STU	="stdu";
	$POP	="ld";
	$PUSH	="std";
} elsif ($flavour =~ /32/) {
	$SIZE_T	=4;
	$LRSAVE	=$SIZE_T;
	$STU	="stwu";
	$POP	="lwz";
	$PUSH	="stw";
} else { die "nonsense $flavour"; }

$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}ppc-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/ppc-xlate.pl" and -f $xlate) or
die "can't locate ppc-xlate.pl";

open STDOUT,"| $^X $xlate $flavour ".shift || die "can't call $xlate: $!";

$FRAME=32*$SIZE_T;

sub _data_word()
{ my $i;
    while(defined($i=shift)) { $code.=sprintf"\t.long\t0x%08x,0x%08x\n",$i,$i; }
}

$sp="r1";
$toc="r2";
$inp="r3";
$out="r4";
$key="r5";

$Tbl0="r3";
$Tbl1="r6";
$Tbl2="r7";
$Tbl3="r2";

$s0="r8";
$s1="r9";
$s2="r10";
$s3="r11";

$t0="r12";
$t1="r13";
$t2="r14";
$t3="r15";

$acc00="r16";
$acc01="r17";
$acc02="r18";
$acc03="r19";

$acc04="r20";
$acc05="r21";
$acc06="r22";
$acc07="r23";

$acc08="r24";
$acc09="r25";
$acc10="r26";
$acc11="r27";

$acc12="r28";
$acc13="r29";
$acc14="r30";
$acc15="r31";

# stay away from TLS pointer
if ($SIZE_T==8)	{ die if ($t1 ne "r13");  $t1="r0";		}
else		{ die if ($Tbl3 ne "r2"); $Tbl3=$t0; $t0="r0";	}
$mask80=$Tbl2;
$mask1b=$Tbl3;

$code.=<<___;
.machine	"any"
.text

.align	7
LAES_Te:
	mflr	r0
	bcl	20,31,\$+4
	mflr	$Tbl0	;    vvvvv "distance" between . and 1st data entry
	addi	$Tbl0,$Tbl0,`128-8`
	mtlr	r0
	blr
	.space	`64-12*4`
LAES_Td:
	mflr	r0
	bcl	20,31,\$+4
	mflr	$Tbl0	;    vvvvvvvv "distance" between . and 1st data entry
	addi	$Tbl0,$Tbl0,`128-64-8+2048+256`
	mtlr	r0
	blr
	.space	`128-64-12*4`
___
&_data_word(
	0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d,
	0xfff2f20d, 0xd66b6bbd, 0xde6f6fb1, 0x91c5c554,
	0x60303050, 0x02010103, 0xce6767a9, 0x562b2b7d,
	0xe7fefe19, 0xb5d7d762, 0x4dababe6, 0xec76769a,
	0x8fcaca45, 0x1f82829d, 0x89c9c940, 0xfa7d7d87,
	0xeffafa15, 0xb25959eb, 0x8e4747c9, 0xfbf0f00b,
	0x41adadec, 0xb3d4d467, 0x5fa2a2fd, 0x45afafea,
	0x239c9cbf, 0x53a4a4f7, 0xe4727296, 0x9bc0c05b,
	0x75b7b7c2, 0xe1fdfd1c, 0x3d9393ae, 0x4c26266a,
	0x6c36365a, 0x7e3f3f41, 0xf5f7f702, 0x83cccc4f,
	0x6834345c, 0x51a5a5f4, 0xd1e5e534, 0xf9f1f108,
	0xe2717193, 0xabd8d873, 0x62313153, 0x2a15153f,
	0x0804040c, 0x95c7c752, 0x46232365, 0x9dc3c35e,
	0x30181828, 0x379696a1, 0x0a05050f, 0x2f9a9ab5,
	0x0e070709, 0x24121236, 0x1b80809b, 0xdfe2e23d,
	0xcdebeb26, 0x4e272769, 0x7fb2b2cd, 0xea75759f,
	0x1209091b, 0x1d83839e, 0x582c2c74, 0x341a1a2e,
	0x361b1b2d, 0xdc6e6eb2, 0xb45a5aee, 0x5ba0a0fb,
	0xa45252f6, 0x763b3b4d, 0xb7d6d661, 0x7db3b3ce,
	0x5229297b, 0xdde3e33e, 0x5e2f2f71, 0x13848497,
	0xa65353f5, 0xb9d1d168, 0x00000000, 0xc1eded2c,
	0x40202060, 0xe3fcfc1f, 0x79b1b1c8, 0xb65b5bed,
	0xd46a6abe, 0x8dcbcb46, 0x67bebed9, 0x7239394b,
	0x944a4ade, 0x984c4cd4, 0xb05858e8, 0x85cfcf4a,
	0xbbd0d06b, 0xc5efef2a, 0x4faaaae5, 0xedfbfb16,
	0x864343c5, 0x9a4d4dd7, 0x66333355, 0x11858594,
	0x8a4545cf, 0xe9f9f910, 0x04020206, 0xfe7f7f81,
	0xa05050f0, 0x783c3c44, 0x259f9fba, 0x4ba8a8e3,
	0xa25151f3, 0x5da3a3fe, 0x804040c0, 0x058f8f8a,
	0x3f9292ad, 0x219d9dbc, 0x70383848, 0xf1f5f504,
	0x63bcbcdf, 0x77b6b6c1, 0xafdada75, 0x42212163,
	0x20101030, 0xe5ffff1a, 0xfdf3f30e, 0xbfd2d26d,
	0x81cdcd4c, 0x180c0c14, 0x26131335, 0xc3ecec2f,
	0xbe5f5fe1, 0x359797a2, 0x884444cc, 0x2e171739,
	0x93c4c457, 0x55a7a7f2, 0xfc7e7e82, 0x7a3d3d47,
	0xc86464ac, 0xba5d5de7, 0x3219192b, 0xe6737395,
	0xc06060a0, 0x19818198, 0x9e4f4fd1, 0xa3dcdc7f,
	0x44222266, 0x542a2a7e, 0x3b9090ab, 0x0b888883,
	0x8c4646ca, 0xc7eeee29, 0x6bb8b8d3, 0x2814143c,
	0xa7dede79, 0xbc5e5ee2, 0x160b0b1d, 0xaddbdb76,
	0xdbe0e03b, 0x64323256, 0x743a3a4e, 0x140a0a1e,
	0x924949db, 0x0c06060a, 0x4824246c, 0xb85c5ce4,
	0x9fc2c25d, 0xbdd3d36e, 0x43acacef, 0xc46262a6,
	0x399191a8, 0x319595a4, 0xd3e4e437, 0xf279798b,
	0xd5e7e732, 0x8bc8c843, 0x6e373759, 0xda6d6db7,
	0x018d8d8c, 0xb1d5d564, 0x9c4e4ed2, 0x49a9a9e0,
	0xd86c6cb4, 0xac5656fa, 0xf3f4f407, 0xcfeaea25,
	0xca6565af, 0xf47a7a8e, 0x47aeaee9, 0x10080818,
	0x6fbabad5, 0xf0787888, 0x4a25256f, 0x5c2e2e72,
	0x381c1c24, 0x57a6a6f1, 0x73b4b4c7, 0x97c6c651,
	0xcbe8e823, 0xa1dddd7c, 0xe874749c, 0x3e1f1f21,
	0x964b4bdd, 0x61bdbddc, 0x0d8b8b86, 0x0f8a8a85,
	0xe0707090, 0x7c3e3e42, 0x71b5b5c4, 0xcc6666aa,
	0x904848d8, 0x06030305, 0xf7f6f601, 0x1c0e0e12,
	0xc26161a3, 0x6a35355f, 0xae5757f9, 0x69b9b9d0,
	0x17868691, 0x99c1c158, 0x3a1d1d27, 0x279e9eb9,
	0xd9e1e138, 0xebf8f813, 0x2b9898b3, 0x22111133,
	0xd26969bb, 0xa9d9d970, 0x078e8e89, 0x339494a7,
	0x2d9b9bb6, 0x3c1e1e22, 0x15878792, 0xc9e9e920,
	0x87cece49, 0xaa5555ff, 0x50282878, 0xa5dfdf7a,
	0x038c8c8f, 0x59a1a1f8, 0x09898980, 0x1a0d0d17,
	0x65bfbfda, 0xd7e6e631, 0x844242c6, 0xd06868b8,
	0x824141c3, 0x299999b0, 0x5a2d2d77, 0x1e0f0f11,
	0x7bb0b0cb, 0xa85454fc, 0x6dbbbbd6, 0x2c16163a);
$code.=<<___;
.byte	0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5
.byte	0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76
.byte	0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0
.byte	0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0
.byte	0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc
.byte	0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15
.byte	0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a
.byte	0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75
.byte	0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0
.byte	0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84
.byte	0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b
.byte	0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf
.byte	0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85
.byte	0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8
.byte	0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5
.byte	0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2
.byte	0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17
.byte	0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73
.byte	0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88
.byte	0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb
.byte	0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c
.byte	0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79
.byte	0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9
.byte	0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08
.byte	0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6
.byte	0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a
.byte	0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e
.byte	0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e
.byte	0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94
.byte	0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf
.byte	0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68
.byte	0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16
___
&_data_word(
	0x51f4a750, 0x7e416553, 0x1a17a4c3, 0x3a275e96,
	0x3bab6bcb, 0x1f9d45f1, 0xacfa58ab, 0x4be30393,
	0x2030fa55, 0xad766df6, 0x88cc7691, 0xf5024c25,
	0x4fe5d7fc, 0xc52acbd7, 0x26354480, 0xb562a38f,
	0xdeb15a49, 0x25ba1b67, 0x45ea0e98, 0x5dfec0e1,
	0xc32f7502, 0x814cf012, 0x8d4697a3, 0x6bd3f9c6,
	0x038f5fe7, 0x15929c95, 0xbf6d7aeb, 0x955259da,
	0xd4be832d, 0x587421d3, 0x49e06929, 0x8ec9c844,
	0x75c2896a, 0xf48e7978, 0x99583e6b, 0x27b971dd,
	0xbee14fb6, 0xf088ad17, 0xc920ac66, 0x7dce3ab4,
	0x63df4a18, 0xe51a3182, 0x97513360, 0x62537f45,
	0xb16477e0, 0xbb6bae84, 0xfe81a01c, 0xf9082b94,
	0x70486858, 0x8f45fd19, 0x94de6c87, 0x527bf8b7,
	0xab73d323, 0x724b02e2, 0xe31f8f57, 0x6655ab2a,
	0xb2eb2807, 0x2fb5c203, 0x86c57b9a, 0xd33708a5,
	0x302887f2, 0x23bfa5b2, 0x02036aba, 0xed16825c,
	0x8acf1c2b, 0xa779b492, 0xf307f2f0, 0x4e69e2a1,
	0x65daf4cd, 0x0605bed5, 0xd134621f, 0xc4a6fe8a,
	0x342e539d, 0xa2f355a0, 0x058ae132, 0xa4f6eb75,
	0x0b83ec39, 0x4060efaa, 0x5e719f06, 0xbd6e1051,
	0x3e218af9, 0x96dd063d, 0xdd3e05ae, 0x4de6bd46,
	0x91548db5, 0x71c45d05, 0x0406d46f, 0x605015ff,
	0x1998fb24, 0xd6bde997, 0x894043cc, 0x67d99e77,
	0xb0e842bd, 0x07898b88, 0xe7195b38, 0x79c8eedb,
	0xa17c0a47, 0x7c420fe9, 0xf8841ec9, 0x00000000,
	0x09808683, 0x322bed48, 0x1e1170ac, 0x6c5a724e,
	0xfd0efffb, 0x0f853856, 0x3daed51e, 0x362d3927,
	0x0a0fd964, 0x685ca621, 0x9b5b54d1, 0x24362e3a,
	0x0c0a67b1, 0x9357e70f, 0xb4ee96d2, 0x1b9b919e,
	0x80c0c54f, 0x61dc20a2, 0x5a774b69, 0x1c121a16,
	0xe293ba0a, 0xc0a02ae5, 0x3c22e043, 0x121b171d,
	0x0e090d0b, 0xf28bc7ad, 0x2db6a8b9, 0x141ea9c8,
	0x57f11985, 0xaf75074c, 0xee99ddbb, 0xa37f60fd,
	0xf701269f, 0x5c72f5bc, 0x44663bc5, 0x5bfb7e34,
	0x8b432976, 0xcb23c6dc, 0xb6edfc68, 0xb8e4f163,
	0xd731dcca, 0x42638510, 0x13972240, 0x84c61120,
	0x854a247d, 0xd2bb3df8, 0xaef93211, 0xc729a16d,
	0x1d9e2f4b, 0xdcb230f3, 0x0d8652ec, 0x77c1e3d0,
	0x2bb3166c, 0xa970b999, 0x119448fa, 0x47e96422,
	0xa8fc8cc4, 0xa0f03f1a, 0x567d2cd8, 0x223390ef,
	0x87494ec7, 0xd938d1c1, 0x8ccaa2fe, 0x98d40b36,
	0xa6f581cf, 0xa57ade28, 0xdab78e26, 0x3fadbfa4,
	0x2c3a9de4, 0x5078920d, 0x6a5fcc9b, 0x547e4662,
	0xf68d13c2, 0x90d8b8e8, 0x2e39f75e, 0x82c3aff5,
	0x9f5d80be, 0x69d0937c, 0x6fd52da9, 0xcf2512b3,
	0xc8ac993b, 0x10187da7, 0xe89c636e, 0xdb3bbb7b,
	0xcd267809, 0x6e5918f4, 0xec9ab701, 0x834f9aa8,
	0xe6956e65, 0xaaffe67e, 0x21bccf08, 0xef15e8e6,
	0xbae79bd9, 0x4a6f36ce, 0xea9f09d4, 0x29b07cd6,
	0x31a4b2af, 0x2a3f2331, 0xc6a59430, 0x35a266c0,
	0x744ebc37, 0xfc82caa6, 0xe090d0b0, 0x33a7d815,
	0xf104984a, 0x41ecdaf7, 0x7fcd500e, 0x1791f62f,
	0x764dd68d, 0x43efb04d, 0xccaa4d54, 0xe49604df,
	0x9ed1b5e3, 0x4c6a881b, 0xc12c1fb8, 0x4665517f,
	0x9d5eea04, 0x018c355d, 0xfa877473, 0xfb0b412e,
	0xb3671d5a, 0x92dbd252, 0xe9105633, 0x6dd64713,
	0x9ad7618c, 0x37a10c7a, 0x59f8148e, 0xeb133c89,
	0xcea927ee, 0xb761c935, 0xe11ce5ed, 0x7a47b13c,
	0x9cd2df59, 0x55f2733f, 0x1814ce79, 0x73c737bf,
	0x53f7cdea, 0x5ffdaa5b, 0xdf3d6f14, 0x7844db86,
	0xcaaff381, 0xb968c43e, 0x3824342c, 0xc2a3405f,
	0x161dc372, 0xbce2250c, 0x283c498b, 0xff0d9541,
	0x39a80171, 0x080cb3de, 0xd8b4e49c, 0x6456c190,
	0x7bcb8461, 0xd532b670, 0x486c5c74, 0xd0b85742);
$code.=<<___;
.byte	0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38
.byte	0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb
.byte	0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87
.byte	0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb
.byte	0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d
.byte	0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e
.byte	0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2
.byte	0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25
.byte	0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16
.byte	0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92
.byte	0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda
.byte	0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84
.byte	0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a
.byte	0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06
.byte	0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02
.byte	0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b
.byte	0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea
.byte	0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73
.byte	0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85
.byte	0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e
.byte	0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89
.byte	0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b
.byte	0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20
.byte	0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4
.byte	0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31
.byte	0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f
.byte	0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d
.byte	0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef
.byte	0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0
.byte	0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61
.byte	0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26
.byte	0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d


.globl	.AES_encrypt
.align	7
.AES_encrypt:
	$STU	$sp,-$FRAME($sp)
	mflr	r0

	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
	$PUSH	r17,`$FRAME-$SIZE_T*15`($sp)
	$PUSH	r18,`$FRAME-$SIZE_T*14`($sp)
	$PUSH	r19,`$FRAME-$SIZE_T*13`($sp)
	$PUSH	r20,`$FRAME-$SIZE_T*12`($sp)
	$PUSH	r21,`$FRAME-$SIZE_T*11`($sp)
	$PUSH	r22,`$FRAME-$SIZE_T*10`($sp)
	$PUSH	r23,`$FRAME-$SIZE_T*9`($sp)
	$PUSH	r24,`$FRAME-$SIZE_T*8`($sp)
	$PUSH	r25,`$FRAME-$SIZE_T*7`($sp)
	$PUSH	r26,`$FRAME-$SIZE_T*6`($sp)
	$PUSH	r27,`$FRAME-$SIZE_T*5`($sp)
	$PUSH	r28,`$FRAME-$SIZE_T*4`($sp)
	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
	$PUSH	r0,`$FRAME+$LRSAVE`($sp)

	andi.	$t0,$inp,3
	andi.	$t1,$out,3
	or.	$t0,$t0,$t1
	bne	Lenc_unaligned

Lenc_unaligned_ok:
	lwz	$s0,0($inp)
	lwz	$s1,4($inp)
	lwz	$s2,8($inp)
	lwz	$s3,12($inp)
	bl	LAES_Te
	bl	Lppc_AES_encrypt_compact
	stw	$s0,0($out)
	stw	$s1,4($out)
	stw	$s2,8($out)
	stw	$s3,12($out)
	b	Lenc_done

Lenc_unaligned:
	subfic	$t0,$inp,4096
	subfic	$t1,$out,4096
	andi.	$t0,$t0,4096-16
	beq	Lenc_xpage
	andi.	$t1,$t1,4096-16
	bne	Lenc_unaligned_ok

Lenc_xpage:
	lbz	$acc00,0($inp)
	lbz	$acc01,1($inp)
	lbz	$acc02,2($inp)
	lbz	$s0,3($inp)
	lbz	$acc04,4($inp)
	lbz	$acc05,5($inp)
	lbz	$acc06,6($inp)
	lbz	$s1,7($inp)
	lbz	$acc08,8($inp)
	lbz	$acc09,9($inp)
	lbz	$acc10,10($inp)
	insrwi	$s0,$acc00,8,0
	lbz	$s2,11($inp)
	insrwi	$s1,$acc04,8,0
	lbz	$acc12,12($inp)
	insrwi	$s0,$acc01,8,8
	lbz	$acc13,13($inp)
	insrwi	$s1,$acc05,8,8
	lbz	$acc14,14($inp)
	insrwi	$s0,$acc02,8,16
	lbz	$s3,15($inp)
	insrwi	$s1,$acc06,8,16
	insrwi	$s2,$acc08,8,0
	insrwi	$s3,$acc12,8,0
	insrwi	$s2,$acc09,8,8
	insrwi	$s3,$acc13,8,8
	insrwi	$s2,$acc10,8,16
	insrwi	$s3,$acc14,8,16

	bl	LAES_Te
	bl	Lppc_AES_encrypt_compact

	extrwi	$acc00,$s0,8,0
	extrwi	$acc01,$s0,8,8
	stb	$acc00,0($out)
	extrwi	$acc02,$s0,8,16
	stb	$acc01,1($out)
	stb	$acc02,2($out)
	extrwi	$acc04,$s1,8,0
	stb	$s0,3($out)
	extrwi	$acc05,$s1,8,8
	stb	$acc04,4($out)
	extrwi	$acc06,$s1,8,16
	stb	$acc05,5($out)
	stb	$acc06,6($out)
	extrwi	$acc08,$s2,8,0
	stb	$s1,7($out)
	extrwi	$acc09,$s2,8,8
	stb	$acc08,8($out)
	extrwi	$acc10,$s2,8,16
	stb	$acc09,9($out)
	stb	$acc10,10($out)
	extrwi	$acc12,$s3,8,0
	stb	$s2,11($out)
	extrwi	$acc13,$s3,8,8
	stb	$acc12,12($out)
	extrwi	$acc14,$s3,8,16
	stb	$acc13,13($out)
	stb	$acc14,14($out)
	stb	$s3,15($out)

Lenc_done:
	$POP	r0,`$FRAME+$LRSAVE`($sp)
	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
	mtlr	r0
	addi	$sp,$sp,$FRAME
	blr

.align	5
Lppc_AES_encrypt:
	lwz	$acc00,240($key)
	addi	$Tbl1,$Tbl0,3
	lwz	$t0,0($key)
	addi	$Tbl2,$Tbl0,2
	lwz	$t1,4($key)
	addi	$Tbl3,$Tbl0,1
	lwz	$t2,8($key)
	addi	$acc00,$acc00,-1
	lwz	$t3,12($key)
	addi	$key,$key,16
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	mtctr	$acc00
.align	4
Lenc_loop:
	rlwinm	$acc00,$s0,`32-24+3`,21,28
	rlwinm	$acc01,$s1,`32-24+3`,21,28
	rlwinm	$acc02,$s2,`32-24+3`,21,28
	rlwinm	$acc03,$s3,`32-24+3`,21,28
	lwz	$t0,0($key)
	rlwinm	$acc04,$s1,`32-16+3`,21,28
	lwz	$t1,4($key)
	rlwinm	$acc05,$s2,`32-16+3`,21,28
	lwz	$t2,8($key)
	rlwinm	$acc06,$s3,`32-16+3`,21,28
	lwz	$t3,12($key)
	rlwinm	$acc07,$s0,`32-16+3`,21,28
	lwzx	$acc00,$Tbl0,$acc00
	rlwinm	$acc08,$s2,`32-8+3`,21,28
	lwzx	$acc01,$Tbl0,$acc01
	rlwinm	$acc09,$s3,`32-8+3`,21,28
	lwzx	$acc02,$Tbl0,$acc02
	rlwinm	$acc10,$s0,`32-8+3`,21,28
	lwzx	$acc03,$Tbl0,$acc03
	rlwinm	$acc11,$s1,`32-8+3`,21,28
	lwzx	$acc04,$Tbl1,$acc04
	rlwinm	$acc12,$s3,`0+3`,21,28
	lwzx	$acc05,$Tbl1,$acc05
	rlwinm	$acc13,$s0,`0+3`,21,28
	lwzx	$acc06,$Tbl1,$acc06
	rlwinm	$acc14,$s1,`0+3`,21,28
	lwzx	$acc07,$Tbl1,$acc07
	rlwinm	$acc15,$s2,`0+3`,21,28
	lwzx	$acc08,$Tbl2,$acc08
	xor	$t0,$t0,$acc00
	lwzx	$acc09,$Tbl2,$acc09
	xor	$t1,$t1,$acc01
	lwzx	$acc10,$Tbl2,$acc10
	xor	$t2,$t2,$acc02
	lwzx	$acc11,$Tbl2,$acc11
	xor	$t3,$t3,$acc03
	lwzx	$acc12,$Tbl3,$acc12
	xor	$t0,$t0,$acc04
	lwzx	$acc13,$Tbl3,$acc13
	xor	$t1,$t1,$acc05
	lwzx	$acc14,$Tbl3,$acc14
	xor	$t2,$t2,$acc06
	lwzx	$acc15,$Tbl3,$acc15
	xor	$t3,$t3,$acc07
	xor	$t0,$t0,$acc08
	xor	$t1,$t1,$acc09
	xor	$t2,$t2,$acc10
	xor	$t3,$t3,$acc11
	xor	$s0,$t0,$acc12
	xor	$s1,$t1,$acc13
	xor	$s2,$t2,$acc14
	xor	$s3,$t3,$acc15
	addi	$key,$key,16
	bdnz-	Lenc_loop

	addi	$Tbl2,$Tbl0,2048
	nop
	lwz	$t0,0($key)
	rlwinm	$acc00,$s0,`32-24`,24,31
	lwz	$t1,4($key)
	rlwinm	$acc01,$s1,`32-24`,24,31
	lwz	$t2,8($key)
	rlwinm	$acc02,$s2,`32-24`,24,31
	lwz	$t3,12($key)
	rlwinm	$acc03,$s3,`32-24`,24,31
	lwz	$acc08,`2048+0`($Tbl0)	! prefetch Te4
	rlwinm	$acc04,$s1,`32-16`,24,31
	lwz	$acc09,`2048+32`($Tbl0)
	rlwinm	$acc05,$s2,`32-16`,24,31
	lwz	$acc10,`2048+64`($Tbl0)
	rlwinm	$acc06,$s3,`32-16`,24,31
	lwz	$acc11,`2048+96`($Tbl0)
	rlwinm	$acc07,$s0,`32-16`,24,31
	lwz	$acc12,`2048+128`($Tbl0)
	rlwinm	$acc08,$s2,`32-8`,24,31
	lwz	$acc13,`2048+160`($Tbl0)
	rlwinm	$acc09,$s3,`32-8`,24,31
	lwz	$acc14,`2048+192`($Tbl0)
	rlwinm	$acc10,$s0,`32-8`,24,31
	lwz	$acc15,`2048+224`($Tbl0)
	rlwinm	$acc11,$s1,`32-8`,24,31
	lbzx	$acc00,$Tbl2,$acc00
	rlwinm	$acc12,$s3,`0`,24,31
	lbzx	$acc01,$Tbl2,$acc01
	rlwinm	$acc13,$s0,`0`,24,31
	lbzx	$acc02,$Tbl2,$acc02
	rlwinm	$acc14,$s1,`0`,24,31
	lbzx	$acc03,$Tbl2,$acc03
	rlwinm	$acc15,$s2,`0`,24,31
	lbzx	$acc04,$Tbl2,$acc04
	rlwinm	$s0,$acc00,24,0,7
	lbzx	$acc05,$Tbl2,$acc05
	rlwinm	$s1,$acc01,24,0,7
	lbzx	$acc06,$Tbl2,$acc06
	rlwinm	$s2,$acc02,24,0,7
	lbzx	$acc07,$Tbl2,$acc07
	rlwinm	$s3,$acc03,24,0,7
	lbzx	$acc08,$Tbl2,$acc08
	rlwimi	$s0,$acc04,16,8,15
	lbzx	$acc09,$Tbl2,$acc09
	rlwimi	$s1,$acc05,16,8,15
	lbzx	$acc10,$Tbl2,$acc10
	rlwimi	$s2,$acc06,16,8,15
	lbzx	$acc11,$Tbl2,$acc11
	rlwimi	$s3,$acc07,16,8,15
	lbzx	$acc12,$Tbl2,$acc12
	rlwimi	$s0,$acc08,8,16,23
	lbzx	$acc13,$Tbl2,$acc13
	rlwimi	$s1,$acc09,8,16,23
	lbzx	$acc14,$Tbl2,$acc14
	rlwimi	$s2,$acc10,8,16,23
	lbzx	$acc15,$Tbl2,$acc15
	rlwimi	$s3,$acc11,8,16,23
	or	$s0,$s0,$acc12
	or	$s1,$s1,$acc13
	or	$s2,$s2,$acc14
	or	$s3,$s3,$acc15
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	blr

.align	4
Lppc_AES_encrypt_compact:
	lwz	$acc00,240($key)
	addi	$Tbl1,$Tbl0,2048
	lwz	$t0,0($key)
	lis	$mask80,0x8080
	lwz	$t1,4($key)
	lis	$mask1b,0x1b1b
	lwz	$t2,8($key)
	ori	$mask80,$mask80,0x8080
	lwz	$t3,12($key)
	ori	$mask1b,$mask1b,0x1b1b
	addi	$key,$key,16
	mtctr	$acc00
.align	4
Lenc_compact_loop:
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	rlwinm	$acc00,$s0,`32-24`,24,31
	xor	$s2,$s2,$t2
	rlwinm	$acc01,$s1,`32-24`,24,31
	xor	$s3,$s3,$t3
	rlwinm	$acc02,$s2,`32-24`,24,31
	rlwinm	$acc03,$s3,`32-24`,24,31
	rlwinm	$acc04,$s1,`32-16`,24,31
	rlwinm	$acc05,$s2,`32-16`,24,31
	rlwinm	$acc06,$s3,`32-16`,24,31
	rlwinm	$acc07,$s0,`32-16`,24,31
	lbzx	$acc00,$Tbl1,$acc00
	rlwinm	$acc08,$s2,`32-8`,24,31
	lbzx	$acc01,$Tbl1,$acc01
	rlwinm	$acc09,$s3,`32-8`,24,31
	lbzx	$acc02,$Tbl1,$acc02
	rlwinm	$acc10,$s0,`32-8`,24,31
	lbzx	$acc03,$Tbl1,$acc03
	rlwinm	$acc11,$s1,`32-8`,24,31
	lbzx	$acc04,$Tbl1,$acc04
	rlwinm	$acc12,$s3,`0`,24,31
	lbzx	$acc05,$Tbl1,$acc05
	rlwinm	$acc13,$s0,`0`,24,31
	lbzx	$acc06,$Tbl1,$acc06
	rlwinm	$acc14,$s1,`0`,24,31
	lbzx	$acc07,$Tbl1,$acc07
	rlwinm	$acc15,$s2,`0`,24,31
	lbzx	$acc08,$Tbl1,$acc08
	rlwinm	$s0,$acc00,24,0,7
	lbzx	$acc09,$Tbl1,$acc09
	rlwinm	$s1,$acc01,24,0,7
	lbzx	$acc10,$Tbl1,$acc10
	rlwinm	$s2,$acc02,24,0,7
	lbzx	$acc11,$Tbl1,$acc11
	rlwinm	$s3,$acc03,24,0,7
	lbzx	$acc12,$Tbl1,$acc12
	rlwimi	$s0,$acc04,16,8,15
	lbzx	$acc13,$Tbl1,$acc13
	rlwimi	$s1,$acc05,16,8,15
	lbzx	$acc14,$Tbl1,$acc14
	rlwimi	$s2,$acc06,16,8,15
	lbzx	$acc15,$Tbl1,$acc15
	rlwimi	$s3,$acc07,16,8,15
	rlwimi	$s0,$acc08,8,16,23
	rlwimi	$s1,$acc09,8,16,23
	rlwimi	$s2,$acc10,8,16,23
	rlwimi	$s3,$acc11,8,16,23
	lwz	$t0,0($key)
	or	$s0,$s0,$acc12
	lwz	$t1,4($key)
	or	$s1,$s1,$acc13
	lwz	$t2,8($key)
	or	$s2,$s2,$acc14
	lwz	$t3,12($key)
	or	$s3,$s3,$acc15

	addi	$key,$key,16
	bdz	Lenc_compact_done

	and	$acc00,$s0,$mask80	# r1=r0&0x80808080
	and	$acc01,$s1,$mask80
	and	$acc02,$s2,$mask80
	and	$acc03,$s3,$mask80
	srwi	$acc04,$acc00,7		# r1>>7
	andc	$acc08,$s0,$mask80	# r0&0x7f7f7f7f
	srwi	$acc05,$acc01,7
	andc	$acc09,$s1,$mask80
	srwi	$acc06,$acc02,7
	andc	$acc10,$s2,$mask80
	srwi	$acc07,$acc03,7
	andc	$acc11,$s3,$mask80
	sub	$acc00,$acc00,$acc04	# r1-(r1>>7)
	sub	$acc01,$acc01,$acc05
	sub	$acc02,$acc02,$acc06
	sub	$acc03,$acc03,$acc07
	add	$acc08,$acc08,$acc08	# (r0&0x7f7f7f7f)<<1
	add	$acc09,$acc09,$acc09
	add	$acc10,$acc10,$acc10
	add	$acc11,$acc11,$acc11
	and	$acc00,$acc00,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc01,$acc01,$mask1b
	and	$acc02,$acc02,$mask1b
	and	$acc03,$acc03,$mask1b
	xor	$acc00,$acc00,$acc08	# r2
	xor	$acc01,$acc01,$acc09
	 rotlwi	$acc12,$s0,16		# ROTATE(r0,16)
	xor	$acc02,$acc02,$acc10
	 rotlwi	$acc13,$s1,16
	xor	$acc03,$acc03,$acc11
	 rotlwi	$acc14,$s2,16

	xor	$s0,$s0,$acc00		# r0^r2
	rotlwi	$acc15,$s3,16
	xor	$s1,$s1,$acc01
	rotrwi	$s0,$s0,24		# ROTATE(r2^r0,24)
	xor	$s2,$s2,$acc02
	rotrwi	$s1,$s1,24
	xor	$s3,$s3,$acc03
	rotrwi	$s2,$s2,24
	xor	$s0,$s0,$acc00		# ROTATE(r2^r0,24)^r2
	rotrwi	$s3,$s3,24
	xor	$s1,$s1,$acc01
	xor	$s2,$s2,$acc02
	xor	$s3,$s3,$acc03
	rotlwi	$acc08,$acc12,8		# ROTATE(r0,24)
	xor	$s0,$s0,$acc12		#
	rotlwi	$acc09,$acc13,8
	xor	$s1,$s1,$acc13
	rotlwi	$acc10,$acc14,8
	xor	$s2,$s2,$acc14
	rotlwi	$acc11,$acc15,8
	xor	$s3,$s3,$acc15
	xor	$s0,$s0,$acc08		#
	xor	$s1,$s1,$acc09
	xor	$s2,$s2,$acc10
	xor	$s3,$s3,$acc11

	b	Lenc_compact_loop
.align	4
Lenc_compact_done:
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	blr

.globl	.AES_decrypt
.align	7
.AES_decrypt:
	$STU	$sp,-$FRAME($sp)
	mflr	r0

	$PUSH	$toc,`$FRAME-$SIZE_T*20`($sp)
	$PUSH	r13,`$FRAME-$SIZE_T*19`($sp)
	$PUSH	r14,`$FRAME-$SIZE_T*18`($sp)
	$PUSH	r15,`$FRAME-$SIZE_T*17`($sp)
	$PUSH	r16,`$FRAME-$SIZE_T*16`($sp)
	$PUSH	r17,`$FRAME-$SIZE_T*15`($sp)
	$PUSH	r18,`$FRAME-$SIZE_T*14`($sp)
	$PUSH	r19,`$FRAME-$SIZE_T*13`($sp)
	$PUSH	r20,`$FRAME-$SIZE_T*12`($sp)
	$PUSH	r21,`$FRAME-$SIZE_T*11`($sp)
	$PUSH	r22,`$FRAME-$SIZE_T*10`($sp)
	$PUSH	r23,`$FRAME-$SIZE_T*9`($sp)
	$PUSH	r24,`$FRAME-$SIZE_T*8`($sp)
	$PUSH	r25,`$FRAME-$SIZE_T*7`($sp)
	$PUSH	r26,`$FRAME-$SIZE_T*6`($sp)
	$PUSH	r27,`$FRAME-$SIZE_T*5`($sp)
	$PUSH	r28,`$FRAME-$SIZE_T*4`($sp)
	$PUSH	r29,`$FRAME-$SIZE_T*3`($sp)
	$PUSH	r30,`$FRAME-$SIZE_T*2`($sp)
	$PUSH	r31,`$FRAME-$SIZE_T*1`($sp)
	$PUSH	r0,`$FRAME+$LRSAVE`($sp)

	andi.	$t0,$inp,3
	andi.	$t1,$out,3
	or.	$t0,$t0,$t1
	bne	Ldec_unaligned

Ldec_unaligned_ok:
	lwz	$s0,0($inp)
	lwz	$s1,4($inp)
	lwz	$s2,8($inp)
	lwz	$s3,12($inp)
	bl	LAES_Td
	bl	Lppc_AES_decrypt_compact
	stw	$s0,0($out)
	stw	$s1,4($out)
	stw	$s2,8($out)
	stw	$s3,12($out)
	b	Ldec_done

Ldec_unaligned:
	subfic	$t0,$inp,4096
	subfic	$t1,$out,4096
	andi.	$t0,$t0,4096-16
	beq	Ldec_xpage
	andi.	$t1,$t1,4096-16
	bne	Ldec_unaligned_ok

Ldec_xpage:
	lbz	$acc00,0($inp)
	lbz	$acc01,1($inp)
	lbz	$acc02,2($inp)
	lbz	$s0,3($inp)
	lbz	$acc04,4($inp)
	lbz	$acc05,5($inp)
	lbz	$acc06,6($inp)
	lbz	$s1,7($inp)
	lbz	$acc08,8($inp)
	lbz	$acc09,9($inp)
	lbz	$acc10,10($inp)
	insrwi	$s0,$acc00,8,0
	lbz	$s2,11($inp)
	insrwi	$s1,$acc04,8,0
	lbz	$acc12,12($inp)
	insrwi	$s0,$acc01,8,8
	lbz	$acc13,13($inp)
	insrwi	$s1,$acc05,8,8
	lbz	$acc14,14($inp)
	insrwi	$s0,$acc02,8,16
	lbz	$s3,15($inp)
	insrwi	$s1,$acc06,8,16
	insrwi	$s2,$acc08,8,0
	insrwi	$s3,$acc12,8,0
	insrwi	$s2,$acc09,8,8
	insrwi	$s3,$acc13,8,8
	insrwi	$s2,$acc10,8,16
	insrwi	$s3,$acc14,8,16

	bl	LAES_Td
	bl	Lppc_AES_decrypt_compact

	extrwi	$acc00,$s0,8,0
	extrwi	$acc01,$s0,8,8
	stb	$acc00,0($out)
	extrwi	$acc02,$s0,8,16
	stb	$acc01,1($out)
	stb	$acc02,2($out)
	extrwi	$acc04,$s1,8,0
	stb	$s0,3($out)
	extrwi	$acc05,$s1,8,8
	stb	$acc04,4($out)
	extrwi	$acc06,$s1,8,16
	stb	$acc05,5($out)
	stb	$acc06,6($out)
	extrwi	$acc08,$s2,8,0
	stb	$s1,7($out)
	extrwi	$acc09,$s2,8,8
	stb	$acc08,8($out)
	extrwi	$acc10,$s2,8,16
	stb	$acc09,9($out)
	stb	$acc10,10($out)
	extrwi	$acc12,$s3,8,0
	stb	$s2,11($out)
	extrwi	$acc13,$s3,8,8
	stb	$acc12,12($out)
	extrwi	$acc14,$s3,8,16
	stb	$acc13,13($out)
	stb	$acc14,14($out)
	stb	$s3,15($out)

Ldec_done:
	$POP	r0,`$FRAME+$LRSAVE`($sp)
	$POP	$toc,`$FRAME-$SIZE_T*20`($sp)
	$POP	r13,`$FRAME-$SIZE_T*19`($sp)
	$POP	r14,`$FRAME-$SIZE_T*18`($sp)
	$POP	r15,`$FRAME-$SIZE_T*17`($sp)
	$POP	r16,`$FRAME-$SIZE_T*16`($sp)
	$POP	r17,`$FRAME-$SIZE_T*15`($sp)
	$POP	r18,`$FRAME-$SIZE_T*14`($sp)
	$POP	r19,`$FRAME-$SIZE_T*13`($sp)
	$POP	r20,`$FRAME-$SIZE_T*12`($sp)
	$POP	r21,`$FRAME-$SIZE_T*11`($sp)
	$POP	r22,`$FRAME-$SIZE_T*10`($sp)
	$POP	r23,`$FRAME-$SIZE_T*9`($sp)
	$POP	r24,`$FRAME-$SIZE_T*8`($sp)
	$POP	r25,`$FRAME-$SIZE_T*7`($sp)
	$POP	r26,`$FRAME-$SIZE_T*6`($sp)
	$POP	r27,`$FRAME-$SIZE_T*5`($sp)
	$POP	r28,`$FRAME-$SIZE_T*4`($sp)
	$POP	r29,`$FRAME-$SIZE_T*3`($sp)
	$POP	r30,`$FRAME-$SIZE_T*2`($sp)
	$POP	r31,`$FRAME-$SIZE_T*1`($sp)
	mtlr	r0
	addi	$sp,$sp,$FRAME
	blr

.align	5
Lppc_AES_decrypt:
	lwz	$acc00,240($key)
	addi	$Tbl1,$Tbl0,3
	lwz	$t0,0($key)
	addi	$Tbl2,$Tbl0,2
	lwz	$t1,4($key)
	addi	$Tbl3,$Tbl0,1
	lwz	$t2,8($key)
	addi	$acc00,$acc00,-1
	lwz	$t3,12($key)
	addi	$key,$key,16
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	mtctr	$acc00
.align	4
Ldec_loop:
	rlwinm	$acc00,$s0,`32-24+3`,21,28
	rlwinm	$acc01,$s1,`32-24+3`,21,28
	rlwinm	$acc02,$s2,`32-24+3`,21,28
	rlwinm	$acc03,$s3,`32-24+3`,21,28
	lwz	$t0,0($key)
	rlwinm	$acc04,$s3,`32-16+3`,21,28
	lwz	$t1,4($key)
	rlwinm	$acc05,$s0,`32-16+3`,21,28
	lwz	$t2,8($key)
	rlwinm	$acc06,$s1,`32-16+3`,21,28
	lwz	$t3,12($key)
	rlwinm	$acc07,$s2,`32-16+3`,21,28
	lwzx	$acc00,$Tbl0,$acc00
	rlwinm	$acc08,$s2,`32-8+3`,21,28
	lwzx	$acc01,$Tbl0,$acc01
	rlwinm	$acc09,$s3,`32-8+3`,21,28
	lwzx	$acc02,$Tbl0,$acc02
	rlwinm	$acc10,$s0,`32-8+3`,21,28
	lwzx	$acc03,$Tbl0,$acc03
	rlwinm	$acc11,$s1,`32-8+3`,21,28
	lwzx	$acc04,$Tbl1,$acc04
	rlwinm	$acc12,$s1,`0+3`,21,28
	lwzx	$acc05,$Tbl1,$acc05
	rlwinm	$acc13,$s2,`0+3`,21,28
	lwzx	$acc06,$Tbl1,$acc06
	rlwinm	$acc14,$s3,`0+3`,21,28
	lwzx	$acc07,$Tbl1,$acc07
	rlwinm	$acc15,$s0,`0+3`,21,28
	lwzx	$acc08,$Tbl2,$acc08
	xor	$t0,$t0,$acc00
	lwzx	$acc09,$Tbl2,$acc09
	xor	$t1,$t1,$acc01
	lwzx	$acc10,$Tbl2,$acc10
	xor	$t2,$t2,$acc02
	lwzx	$acc11,$Tbl2,$acc11
	xor	$t3,$t3,$acc03
	lwzx	$acc12,$Tbl3,$acc12
	xor	$t0,$t0,$acc04
	lwzx	$acc13,$Tbl3,$acc13
	xor	$t1,$t1,$acc05
	lwzx	$acc14,$Tbl3,$acc14
	xor	$t2,$t2,$acc06
	lwzx	$acc15,$Tbl3,$acc15
	xor	$t3,$t3,$acc07
	xor	$t0,$t0,$acc08
	xor	$t1,$t1,$acc09
	xor	$t2,$t2,$acc10
	xor	$t3,$t3,$acc11
	xor	$s0,$t0,$acc12
	xor	$s1,$t1,$acc13
	xor	$s2,$t2,$acc14
	xor	$s3,$t3,$acc15
	addi	$key,$key,16
	bdnz-	Ldec_loop

	addi	$Tbl2,$Tbl0,2048
	nop
	lwz	$t0,0($key)
	rlwinm	$acc00,$s0,`32-24`,24,31
	lwz	$t1,4($key)
	rlwinm	$acc01,$s1,`32-24`,24,31
	lwz	$t2,8($key)
	rlwinm	$acc02,$s2,`32-24`,24,31
	lwz	$t3,12($key)
	rlwinm	$acc03,$s3,`32-24`,24,31
	lwz	$acc08,`2048+0`($Tbl0)	! prefetch Td4
	rlwinm	$acc04,$s3,`32-16`,24,31
	lwz	$acc09,`2048+32`($Tbl0)
	rlwinm	$acc05,$s0,`32-16`,24,31
	lwz	$acc10,`2048+64`($Tbl0)
	lbzx	$acc00,$Tbl2,$acc00
	lwz	$acc11,`2048+96`($Tbl0)
	lbzx	$acc01,$Tbl2,$acc01
	lwz	$acc12,`2048+128`($Tbl0)
	rlwinm	$acc06,$s1,`32-16`,24,31
	lwz	$acc13,`2048+160`($Tbl0)
	rlwinm	$acc07,$s2,`32-16`,24,31
	lwz	$acc14,`2048+192`($Tbl0)
	rlwinm	$acc08,$s2,`32-8`,24,31
	lwz	$acc15,`2048+224`($Tbl0)
	rlwinm	$acc09,$s3,`32-8`,24,31
	lbzx	$acc02,$Tbl2,$acc02
	rlwinm	$acc10,$s0,`32-8`,24,31
	lbzx	$acc03,$Tbl2,$acc03
	rlwinm	$acc11,$s1,`32-8`,24,31
	lbzx	$acc04,$Tbl2,$acc04
	rlwinm	$acc12,$s1,`0`,24,31
	lbzx	$acc05,$Tbl2,$acc05
	rlwinm	$acc13,$s2,`0`,24,31
	lbzx	$acc06,$Tbl2,$acc06
	rlwinm	$acc14,$s3,`0`,24,31
	lbzx	$acc07,$Tbl2,$acc07
	rlwinm	$acc15,$s0,`0`,24,31
	lbzx	$acc08,$Tbl2,$acc08
	rlwinm	$s0,$acc00,24,0,7
	lbzx	$acc09,$Tbl2,$acc09
	rlwinm	$s1,$acc01,24,0,7
	lbzx	$acc10,$Tbl2,$acc10
	rlwinm	$s2,$acc02,24,0,7
	lbzx	$acc11,$Tbl2,$acc11
	rlwinm	$s3,$acc03,24,0,7
	lbzx	$acc12,$Tbl2,$acc12
	rlwimi	$s0,$acc04,16,8,15
	lbzx	$acc13,$Tbl2,$acc13
	rlwimi	$s1,$acc05,16,8,15
	lbzx	$acc14,$Tbl2,$acc14
	rlwimi	$s2,$acc06,16,8,15
	lbzx	$acc15,$Tbl2,$acc15
	rlwimi	$s3,$acc07,16,8,15
	rlwimi	$s0,$acc08,8,16,23
	rlwimi	$s1,$acc09,8,16,23
	rlwimi	$s2,$acc10,8,16,23
	rlwimi	$s3,$acc11,8,16,23
	or	$s0,$s0,$acc12
	or	$s1,$s1,$acc13
	or	$s2,$s2,$acc14
	or	$s3,$s3,$acc15
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	blr

.align	4
Lppc_AES_decrypt_compact:
	lwz	$acc00,240($key)
	addi	$Tbl1,$Tbl0,2048
	lwz	$t0,0($key)
	lis	$mask80,0x8080
	lwz	$t1,4($key)
	lis	$mask1b,0x1b1b
	lwz	$t2,8($key)
	ori	$mask80,$mask80,0x8080
	lwz	$t3,12($key)
	ori	$mask1b,$mask1b,0x1b1b
	addi	$key,$key,16
___
$code.=<<___ if ($SIZE_T==8);
	insrdi	$mask80,$mask80,32,0
	insrdi	$mask1b,$mask1b,32,0
___
$code.=<<___;
	mtctr	$acc00
.align	4
Ldec_compact_loop:
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	rlwinm	$acc00,$s0,`32-24`,24,31
	xor	$s2,$s2,$t2
	rlwinm	$acc01,$s1,`32-24`,24,31
	xor	$s3,$s3,$t3
	rlwinm	$acc02,$s2,`32-24`,24,31
	rlwinm	$acc03,$s3,`32-24`,24,31
	rlwinm	$acc04,$s3,`32-16`,24,31
	rlwinm	$acc05,$s0,`32-16`,24,31
	rlwinm	$acc06,$s1,`32-16`,24,31
	rlwinm	$acc07,$s2,`32-16`,24,31
	lbzx	$acc00,$Tbl1,$acc00
	rlwinm	$acc08,$s2,`32-8`,24,31
	lbzx	$acc01,$Tbl1,$acc01
	rlwinm	$acc09,$s3,`32-8`,24,31
	lbzx	$acc02,$Tbl1,$acc02
	rlwinm	$acc10,$s0,`32-8`,24,31
	lbzx	$acc03,$Tbl1,$acc03
	rlwinm	$acc11,$s1,`32-8`,24,31
	lbzx	$acc04,$Tbl1,$acc04
	rlwinm	$acc12,$s1,`0`,24,31
	lbzx	$acc05,$Tbl1,$acc05
	rlwinm	$acc13,$s2,`0`,24,31
	lbzx	$acc06,$Tbl1,$acc06
	rlwinm	$acc14,$s3,`0`,24,31
	lbzx	$acc07,$Tbl1,$acc07
	rlwinm	$acc15,$s0,`0`,24,31
	lbzx	$acc08,$Tbl1,$acc08
	rlwinm	$s0,$acc00,24,0,7
	lbzx	$acc09,$Tbl1,$acc09
	rlwinm	$s1,$acc01,24,0,7
	lbzx	$acc10,$Tbl1,$acc10
	rlwinm	$s2,$acc02,24,0,7
	lbzx	$acc11,$Tbl1,$acc11
	rlwinm	$s3,$acc03,24,0,7
	lbzx	$acc12,$Tbl1,$acc12
	rlwimi	$s0,$acc04,16,8,15
	lbzx	$acc13,$Tbl1,$acc13
	rlwimi	$s1,$acc05,16,8,15
	lbzx	$acc14,$Tbl1,$acc14
	rlwimi	$s2,$acc06,16,8,15
	lbzx	$acc15,$Tbl1,$acc15
	rlwimi	$s3,$acc07,16,8,15
	rlwimi	$s0,$acc08,8,16,23
	rlwimi	$s1,$acc09,8,16,23
	rlwimi	$s2,$acc10,8,16,23
	rlwimi	$s3,$acc11,8,16,23
	lwz	$t0,0($key)
	or	$s0,$s0,$acc12
	lwz	$t1,4($key)
	or	$s1,$s1,$acc13
	lwz	$t2,8($key)
	or	$s2,$s2,$acc14
	lwz	$t3,12($key)
	or	$s3,$s3,$acc15

	addi	$key,$key,16
	bdz	Ldec_compact_done
___
$code.=<<___ if ($SIZE_T==8);
	# vectorized permutation improves decrypt performance by 10%
	insrdi	$s0,$s1,32,0
	insrdi	$s2,$s3,32,0

	and	$acc00,$s0,$mask80	# r1=r0&0x80808080
	and	$acc02,$s2,$mask80
	srdi	$acc04,$acc00,7		# r1>>7
	srdi	$acc06,$acc02,7
	andc	$acc08,$s0,$mask80	# r0&0x7f7f7f7f
	andc	$acc10,$s2,$mask80
	sub	$acc00,$acc00,$acc04	# r1-(r1>>7)
	sub	$acc02,$acc02,$acc06
	add	$acc08,$acc08,$acc08	# (r0&0x7f7f7f7f)<<1
	add	$acc10,$acc10,$acc10
	and	$acc00,$acc00,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc02,$acc02,$mask1b
	xor	$acc00,$acc00,$acc08	# r2
	xor	$acc02,$acc02,$acc10

	and	$acc04,$acc00,$mask80	# r1=r2&0x80808080
	and	$acc06,$acc02,$mask80
	srdi	$acc08,$acc04,7		# r1>>7
	srdi	$acc10,$acc06,7
	andc	$acc12,$acc00,$mask80	# r2&0x7f7f7f7f
	andc	$acc14,$acc02,$mask80
	sub	$acc04,$acc04,$acc08	# r1-(r1>>7)
	sub	$acc06,$acc06,$acc10
	add	$acc12,$acc12,$acc12	# (r2&0x7f7f7f7f)<<1
	add	$acc14,$acc14,$acc14
	and	$acc04,$acc04,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc06,$acc06,$mask1b
	xor	$acc04,$acc04,$acc12	# r4
	xor	$acc06,$acc06,$acc14

	and	$acc08,$acc04,$mask80	# r1=r4&0x80808080
	and	$acc10,$acc06,$mask80
	srdi	$acc12,$acc08,7		# r1>>7
	srdi	$acc14,$acc10,7
	sub	$acc08,$acc08,$acc12	# r1-(r1>>7)
	sub	$acc10,$acc10,$acc14
	andc	$acc12,$acc04,$mask80	# r4&0x7f7f7f7f
	andc	$acc14,$acc06,$mask80
	add	$acc12,$acc12,$acc12	# (r4&0x7f7f7f7f)<<1
	add	$acc14,$acc14,$acc14
	and	$acc08,$acc08,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc10,$acc10,$mask1b
	xor	$acc08,$acc08,$acc12	# r8
	xor	$acc10,$acc10,$acc14

	xor	$acc00,$acc00,$s0	# r2^r0
	xor	$acc02,$acc02,$s2
	xor	$acc04,$acc04,$s0	# r4^r0
	xor	$acc06,$acc06,$s2

	extrdi	$acc01,$acc00,32,0
	extrdi	$acc03,$acc02,32,0
	extrdi	$acc05,$acc04,32,0
	extrdi	$acc07,$acc06,32,0
	extrdi	$acc09,$acc08,32,0
	extrdi	$acc11,$acc10,32,0
___
$code.=<<___ if ($SIZE_T==4);
	and	$acc00,$s0,$mask80	# r1=r0&0x80808080
	and	$acc01,$s1,$mask80
	and	$acc02,$s2,$mask80
	and	$acc03,$s3,$mask80
	srwi	$acc04,$acc00,7		# r1>>7
	andc	$acc08,$s0,$mask80	# r0&0x7f7f7f7f
	srwi	$acc05,$acc01,7
	andc	$acc09,$s1,$mask80
	srwi	$acc06,$acc02,7
	andc	$acc10,$s2,$mask80
	srwi	$acc07,$acc03,7
	andc	$acc11,$s3,$mask80
	sub	$acc00,$acc00,$acc04	# r1-(r1>>7)
	sub	$acc01,$acc01,$acc05
	sub	$acc02,$acc02,$acc06
	sub	$acc03,$acc03,$acc07
	add	$acc08,$acc08,$acc08	# (r0&0x7f7f7f7f)<<1
	add	$acc09,$acc09,$acc09
	add	$acc10,$acc10,$acc10
	add	$acc11,$acc11,$acc11
	and	$acc00,$acc00,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc01,$acc01,$mask1b
	and	$acc02,$acc02,$mask1b
	and	$acc03,$acc03,$mask1b
	xor	$acc00,$acc00,$acc08	# r2
	xor	$acc01,$acc01,$acc09
	xor	$acc02,$acc02,$acc10
	xor	$acc03,$acc03,$acc11

	and	$acc04,$acc00,$mask80	# r1=r2&0x80808080
	and	$acc05,$acc01,$mask80
	and	$acc06,$acc02,$mask80
	and	$acc07,$acc03,$mask80
	srwi	$acc08,$acc04,7		# r1>>7
	andc	$acc12,$acc00,$mask80	# r2&0x7f7f7f7f
	srwi	$acc09,$acc05,7
	andc	$acc13,$acc01,$mask80
	srwi	$acc10,$acc06,7
	andc	$acc14,$acc02,$mask80
	srwi	$acc11,$acc07,7
	andc	$acc15,$acc03,$mask80
	sub	$acc04,$acc04,$acc08	# r1-(r1>>7)
	sub	$acc05,$acc05,$acc09
	sub	$acc06,$acc06,$acc10
	sub	$acc07,$acc07,$acc11
	add	$acc12,$acc12,$acc12	# (r2&0x7f7f7f7f)<<1
	add	$acc13,$acc13,$acc13
	add	$acc14,$acc14,$acc14
	add	$acc15,$acc15,$acc15
	and	$acc04,$acc04,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc05,$acc05,$mask1b
	and	$acc06,$acc06,$mask1b
	and	$acc07,$acc07,$mask1b
	xor	$acc04,$acc04,$acc12	# r4
	xor	$acc05,$acc05,$acc13
	xor	$acc06,$acc06,$acc14
	xor	$acc07,$acc07,$acc15

	and	$acc08,$acc04,$mask80	# r1=r4&0x80808080
	and	$acc09,$acc05,$mask80
	srwi	$acc12,$acc08,7		# r1>>7
	and	$acc10,$acc06,$mask80
	srwi	$acc13,$acc09,7
	and	$acc11,$acc07,$mask80
	srwi	$acc14,$acc10,7
	sub	$acc08,$acc08,$acc12	# r1-(r1>>7)
	srwi	$acc15,$acc11,7
	sub	$acc09,$acc09,$acc13
	sub	$acc10,$acc10,$acc14
	sub	$acc11,$acc11,$acc15
	andc	$acc12,$acc04,$mask80	# r4&0x7f7f7f7f
	andc	$acc13,$acc05,$mask80
	andc	$acc14,$acc06,$mask80
	andc	$acc15,$acc07,$mask80
	add	$acc12,$acc12,$acc12	# (r4&0x7f7f7f7f)<<1
	add	$acc13,$acc13,$acc13
	add	$acc14,$acc14,$acc14
	add	$acc15,$acc15,$acc15
	and	$acc08,$acc08,$mask1b	# (r1-(r1>>7))&0x1b1b1b1b
	and	$acc09,$acc09,$mask1b
	and	$acc10,$acc10,$mask1b
	and	$acc11,$acc11,$mask1b
	xor	$acc08,$acc08,$acc12	# r8
	xor	$acc09,$acc09,$acc13
	xor	$acc10,$acc10,$acc14
	xor	$acc11,$acc11,$acc15

	xor	$acc00,$acc00,$s0	# r2^r0
	xor	$acc01,$acc01,$s1
	xor	$acc02,$acc02,$s2
	xor	$acc03,$acc03,$s3
	xor	$acc04,$acc04,$s0	# r4^r0
	xor	$acc05,$acc05,$s1
	xor	$acc06,$acc06,$s2
	xor	$acc07,$acc07,$s3
___
$code.=<<___;
	rotrwi	$s0,$s0,8		# = ROTATE(r0,8)
	rotrwi	$s1,$s1,8
	xor	$s0,$s0,$acc00		# ^= r2^r0
	rotrwi	$s2,$s2,8
	xor	$s1,$s1,$acc01
	rotrwi	$s3,$s3,8
	xor	$s2,$s2,$acc02
	xor	$s3,$s3,$acc03
	xor	$acc00,$acc00,$acc08
	xor	$acc01,$acc01,$acc09
	xor	$acc02,$acc02,$acc10
	xor	$acc03,$acc03,$acc11
	xor	$s0,$s0,$acc04		# ^= r4^r0
	rotrwi	$acc00,$acc00,24
	xor	$s1,$s1,$acc05
	rotrwi	$acc01,$acc01,24
	xor	$s2,$s2,$acc06
	rotrwi	$acc02,$acc02,24
	xor	$s3,$s3,$acc07
	rotrwi	$acc03,$acc03,24
	xor	$acc04,$acc04,$acc08
	xor	$acc05,$acc05,$acc09
	xor	$acc06,$acc06,$acc10
	xor	$acc07,$acc07,$acc11
	xor	$s0,$s0,$acc08		# ^= r8 [^((r4^r0)^(r2^r0)=r4^r2)]
	rotrwi	$acc04,$acc04,16
	xor	$s1,$s1,$acc09
	rotrwi	$acc05,$acc05,16
	xor	$s2,$s2,$acc10
	rotrwi	$acc06,$acc06,16
	xor	$s3,$s3,$acc11
	rotrwi	$acc07,$acc07,16
	xor	$s0,$s0,$acc00		# ^= ROTATE(r8^r2^r0,24)
	rotrwi	$acc08,$acc08,8
	xor	$s1,$s1,$acc01
	rotrwi	$acc09,$acc09,8
	xor	$s2,$s2,$acc02
	rotrwi	$acc10,$acc10,8
	xor	$s3,$s3,$acc03
	rotrwi	$acc11,$acc11,8
	xor	$s0,$s0,$acc04		# ^= ROTATE(r8^r4^r0,16)
	xor	$s1,$s1,$acc05
	xor	$s2,$s2,$acc06
	xor	$s3,$s3,$acc07
	xor	$s0,$s0,$acc08		# ^= ROTATE(r8,8)	
	xor	$s1,$s1,$acc09	
	xor	$s2,$s2,$acc10	
	xor	$s3,$s3,$acc11	

	b	Ldec_compact_loop
.align	4
Ldec_compact_done:
	xor	$s0,$s0,$t0
	xor	$s1,$s1,$t1
	xor	$s2,$s2,$t2
	xor	$s3,$s3,$t3
	blr
___

$code =~ s/\`([^\`]*)\`/eval $1/gem;
print $code;
close STDOUT;