aes-ppc.pl 37.1 KB
Newer Older
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365
	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
	.long	0
	.byte	0,12,0x14,0,0,0,0,0

.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
	.long	0
	.byte	0,12,0x14,0,0,0,0,0

.asciz	"AES for PPC, CRYPTOGAMS by <appro\@openssl.org>"
.align	7
___

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