diff options
Diffstat (limited to 'deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes')
3 files changed, 6725 insertions, 0 deletions
diff --git a/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aes-ppc.s b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aes-ppc.s new file mode 100644 index 0000000000..5f37cdc33c --- /dev/null +++ b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aes-ppc.s @@ -0,0 +1,1539 @@ +.machine "any" +.text + +.align 7 +.LAES_Te: + mflr 0 + bcl 20,31,$+4 + mflr 3 + addi 3,3,120 + mtlr 0 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.space 28 +.LAES_Td: + mflr 0 + bcl 20,31,$+4 + mflr 3 + addi 3,3,2360 + mtlr 0 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.space 28 +.long 0xc66363a5,0xc66363a5 +.long 0xf87c7c84,0xf87c7c84 +.long 0xee777799,0xee777799 +.long 0xf67b7b8d,0xf67b7b8d +.long 0xfff2f20d,0xfff2f20d +.long 0xd66b6bbd,0xd66b6bbd +.long 0xde6f6fb1,0xde6f6fb1 +.long 0x91c5c554,0x91c5c554 +.long 0x60303050,0x60303050 +.long 0x02010103,0x02010103 +.long 0xce6767a9,0xce6767a9 +.long 0x562b2b7d,0x562b2b7d +.long 0xe7fefe19,0xe7fefe19 +.long 0xb5d7d762,0xb5d7d762 +.long 0x4dababe6,0x4dababe6 +.long 0xec76769a,0xec76769a +.long 0x8fcaca45,0x8fcaca45 +.long 0x1f82829d,0x1f82829d +.long 0x89c9c940,0x89c9c940 +.long 0xfa7d7d87,0xfa7d7d87 +.long 0xeffafa15,0xeffafa15 +.long 0xb25959eb,0xb25959eb +.long 0x8e4747c9,0x8e4747c9 +.long 0xfbf0f00b,0xfbf0f00b +.long 0x41adadec,0x41adadec +.long 0xb3d4d467,0xb3d4d467 +.long 0x5fa2a2fd,0x5fa2a2fd +.long 0x45afafea,0x45afafea +.long 0x239c9cbf,0x239c9cbf +.long 0x53a4a4f7,0x53a4a4f7 +.long 0xe4727296,0xe4727296 +.long 0x9bc0c05b,0x9bc0c05b +.long 0x75b7b7c2,0x75b7b7c2 +.long 0xe1fdfd1c,0xe1fdfd1c +.long 0x3d9393ae,0x3d9393ae +.long 0x4c26266a,0x4c26266a +.long 0x6c36365a,0x6c36365a +.long 0x7e3f3f41,0x7e3f3f41 +.long 0xf5f7f702,0xf5f7f702 +.long 0x83cccc4f,0x83cccc4f +.long 0x6834345c,0x6834345c +.long 0x51a5a5f4,0x51a5a5f4 +.long 0xd1e5e534,0xd1e5e534 +.long 0xf9f1f108,0xf9f1f108 +.long 0xe2717193,0xe2717193 +.long 0xabd8d873,0xabd8d873 +.long 0x62313153,0x62313153 +.long 0x2a15153f,0x2a15153f +.long 0x0804040c,0x0804040c +.long 0x95c7c752,0x95c7c752 +.long 0x46232365,0x46232365 +.long 0x9dc3c35e,0x9dc3c35e +.long 0x30181828,0x30181828 +.long 0x379696a1,0x379696a1 +.long 0x0a05050f,0x0a05050f +.long 0x2f9a9ab5,0x2f9a9ab5 +.long 0x0e070709,0x0e070709 +.long 0x24121236,0x24121236 +.long 0x1b80809b,0x1b80809b +.long 0xdfe2e23d,0xdfe2e23d +.long 0xcdebeb26,0xcdebeb26 +.long 0x4e272769,0x4e272769 +.long 0x7fb2b2cd,0x7fb2b2cd +.long 0xea75759f,0xea75759f +.long 0x1209091b,0x1209091b +.long 0x1d83839e,0x1d83839e +.long 0x582c2c74,0x582c2c74 +.long 0x341a1a2e,0x341a1a2e +.long 0x361b1b2d,0x361b1b2d +.long 0xdc6e6eb2,0xdc6e6eb2 +.long 0xb45a5aee,0xb45a5aee +.long 0x5ba0a0fb,0x5ba0a0fb +.long 0xa45252f6,0xa45252f6 +.long 0x763b3b4d,0x763b3b4d +.long 0xb7d6d661,0xb7d6d661 +.long 0x7db3b3ce,0x7db3b3ce +.long 0x5229297b,0x5229297b +.long 0xdde3e33e,0xdde3e33e +.long 0x5e2f2f71,0x5e2f2f71 +.long 0x13848497,0x13848497 +.long 0xa65353f5,0xa65353f5 +.long 0xb9d1d168,0xb9d1d168 +.long 0x00000000,0x00000000 +.long 0xc1eded2c,0xc1eded2c +.long 0x40202060,0x40202060 +.long 0xe3fcfc1f,0xe3fcfc1f +.long 0x79b1b1c8,0x79b1b1c8 +.long 0xb65b5bed,0xb65b5bed +.long 0xd46a6abe,0xd46a6abe +.long 0x8dcbcb46,0x8dcbcb46 +.long 0x67bebed9,0x67bebed9 +.long 0x7239394b,0x7239394b +.long 0x944a4ade,0x944a4ade +.long 0x984c4cd4,0x984c4cd4 +.long 0xb05858e8,0xb05858e8 +.long 0x85cfcf4a,0x85cfcf4a +.long 0xbbd0d06b,0xbbd0d06b +.long 0xc5efef2a,0xc5efef2a +.long 0x4faaaae5,0x4faaaae5 +.long 0xedfbfb16,0xedfbfb16 +.long 0x864343c5,0x864343c5 +.long 0x9a4d4dd7,0x9a4d4dd7 +.long 0x66333355,0x66333355 +.long 0x11858594,0x11858594 +.long 0x8a4545cf,0x8a4545cf +.long 0xe9f9f910,0xe9f9f910 +.long 0x04020206,0x04020206 +.long 0xfe7f7f81,0xfe7f7f81 +.long 0xa05050f0,0xa05050f0 +.long 0x783c3c44,0x783c3c44 +.long 0x259f9fba,0x259f9fba +.long 0x4ba8a8e3,0x4ba8a8e3 +.long 0xa25151f3,0xa25151f3 +.long 0x5da3a3fe,0x5da3a3fe +.long 0x804040c0,0x804040c0 +.long 0x058f8f8a,0x058f8f8a +.long 0x3f9292ad,0x3f9292ad +.long 0x219d9dbc,0x219d9dbc +.long 0x70383848,0x70383848 +.long 0xf1f5f504,0xf1f5f504 +.long 0x63bcbcdf,0x63bcbcdf +.long 0x77b6b6c1,0x77b6b6c1 +.long 0xafdada75,0xafdada75 +.long 0x42212163,0x42212163 +.long 0x20101030,0x20101030 +.long 0xe5ffff1a,0xe5ffff1a +.long 0xfdf3f30e,0xfdf3f30e +.long 0xbfd2d26d,0xbfd2d26d +.long 0x81cdcd4c,0x81cdcd4c +.long 0x180c0c14,0x180c0c14 +.long 0x26131335,0x26131335 +.long 0xc3ecec2f,0xc3ecec2f +.long 0xbe5f5fe1,0xbe5f5fe1 +.long 0x359797a2,0x359797a2 +.long 0x884444cc,0x884444cc +.long 0x2e171739,0x2e171739 +.long 0x93c4c457,0x93c4c457 +.long 0x55a7a7f2,0x55a7a7f2 +.long 0xfc7e7e82,0xfc7e7e82 +.long 0x7a3d3d47,0x7a3d3d47 +.long 0xc86464ac,0xc86464ac +.long 0xba5d5de7,0xba5d5de7 +.long 0x3219192b,0x3219192b +.long 0xe6737395,0xe6737395 +.long 0xc06060a0,0xc06060a0 +.long 0x19818198,0x19818198 +.long 0x9e4f4fd1,0x9e4f4fd1 +.long 0xa3dcdc7f,0xa3dcdc7f +.long 0x44222266,0x44222266 +.long 0x542a2a7e,0x542a2a7e +.long 0x3b9090ab,0x3b9090ab +.long 0x0b888883,0x0b888883 +.long 0x8c4646ca,0x8c4646ca +.long 0xc7eeee29,0xc7eeee29 +.long 0x6bb8b8d3,0x6bb8b8d3 +.long 0x2814143c,0x2814143c +.long 0xa7dede79,0xa7dede79 +.long 0xbc5e5ee2,0xbc5e5ee2 +.long 0x160b0b1d,0x160b0b1d +.long 0xaddbdb76,0xaddbdb76 +.long 0xdbe0e03b,0xdbe0e03b +.long 0x64323256,0x64323256 +.long 0x743a3a4e,0x743a3a4e +.long 0x140a0a1e,0x140a0a1e +.long 0x924949db,0x924949db +.long 0x0c06060a,0x0c06060a +.long 0x4824246c,0x4824246c +.long 0xb85c5ce4,0xb85c5ce4 +.long 0x9fc2c25d,0x9fc2c25d +.long 0xbdd3d36e,0xbdd3d36e +.long 0x43acacef,0x43acacef +.long 0xc46262a6,0xc46262a6 +.long 0x399191a8,0x399191a8 +.long 0x319595a4,0x319595a4 +.long 0xd3e4e437,0xd3e4e437 +.long 0xf279798b,0xf279798b +.long 0xd5e7e732,0xd5e7e732 +.long 0x8bc8c843,0x8bc8c843 +.long 0x6e373759,0x6e373759 +.long 0xda6d6db7,0xda6d6db7 +.long 0x018d8d8c,0x018d8d8c +.long 0xb1d5d564,0xb1d5d564 +.long 0x9c4e4ed2,0x9c4e4ed2 +.long 0x49a9a9e0,0x49a9a9e0 +.long 0xd86c6cb4,0xd86c6cb4 +.long 0xac5656fa,0xac5656fa +.long 0xf3f4f407,0xf3f4f407 +.long 0xcfeaea25,0xcfeaea25 +.long 0xca6565af,0xca6565af +.long 0xf47a7a8e,0xf47a7a8e +.long 0x47aeaee9,0x47aeaee9 +.long 0x10080818,0x10080818 +.long 0x6fbabad5,0x6fbabad5 +.long 0xf0787888,0xf0787888 +.long 0x4a25256f,0x4a25256f +.long 0x5c2e2e72,0x5c2e2e72 +.long 0x381c1c24,0x381c1c24 +.long 0x57a6a6f1,0x57a6a6f1 +.long 0x73b4b4c7,0x73b4b4c7 +.long 0x97c6c651,0x97c6c651 +.long 0xcbe8e823,0xcbe8e823 +.long 0xa1dddd7c,0xa1dddd7c +.long 0xe874749c,0xe874749c +.long 0x3e1f1f21,0x3e1f1f21 +.long 0x964b4bdd,0x964b4bdd +.long 0x61bdbddc,0x61bdbddc +.long 0x0d8b8b86,0x0d8b8b86 +.long 0x0f8a8a85,0x0f8a8a85 +.long 0xe0707090,0xe0707090 +.long 0x7c3e3e42,0x7c3e3e42 +.long 0x71b5b5c4,0x71b5b5c4 +.long 0xcc6666aa,0xcc6666aa +.long 0x904848d8,0x904848d8 +.long 0x06030305,0x06030305 +.long 0xf7f6f601,0xf7f6f601 +.long 0x1c0e0e12,0x1c0e0e12 +.long 0xc26161a3,0xc26161a3 +.long 0x6a35355f,0x6a35355f +.long 0xae5757f9,0xae5757f9 +.long 0x69b9b9d0,0x69b9b9d0 +.long 0x17868691,0x17868691 +.long 0x99c1c158,0x99c1c158 +.long 0x3a1d1d27,0x3a1d1d27 +.long 0x279e9eb9,0x279e9eb9 +.long 0xd9e1e138,0xd9e1e138 +.long 0xebf8f813,0xebf8f813 +.long 0x2b9898b3,0x2b9898b3 +.long 0x22111133,0x22111133 +.long 0xd26969bb,0xd26969bb +.long 0xa9d9d970,0xa9d9d970 +.long 0x078e8e89,0x078e8e89 +.long 0x339494a7,0x339494a7 +.long 0x2d9b9bb6,0x2d9b9bb6 +.long 0x3c1e1e22,0x3c1e1e22 +.long 0x15878792,0x15878792 +.long 0xc9e9e920,0xc9e9e920 +.long 0x87cece49,0x87cece49 +.long 0xaa5555ff,0xaa5555ff +.long 0x50282878,0x50282878 +.long 0xa5dfdf7a,0xa5dfdf7a +.long 0x038c8c8f,0x038c8c8f +.long 0x59a1a1f8,0x59a1a1f8 +.long 0x09898980,0x09898980 +.long 0x1a0d0d17,0x1a0d0d17 +.long 0x65bfbfda,0x65bfbfda +.long 0xd7e6e631,0xd7e6e631 +.long 0x844242c6,0x844242c6 +.long 0xd06868b8,0xd06868b8 +.long 0x824141c3,0x824141c3 +.long 0x299999b0,0x299999b0 +.long 0x5a2d2d77,0x5a2d2d77 +.long 0x1e0f0f11,0x1e0f0f11 +.long 0x7bb0b0cb,0x7bb0b0cb +.long 0xa85454fc,0xa85454fc +.long 0x6dbbbbd6,0x6dbbbbd6 +.long 0x2c16163a,0x2c16163a +.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 +.long 0x51f4a750,0x51f4a750 +.long 0x7e416553,0x7e416553 +.long 0x1a17a4c3,0x1a17a4c3 +.long 0x3a275e96,0x3a275e96 +.long 0x3bab6bcb,0x3bab6bcb +.long 0x1f9d45f1,0x1f9d45f1 +.long 0xacfa58ab,0xacfa58ab +.long 0x4be30393,0x4be30393 +.long 0x2030fa55,0x2030fa55 +.long 0xad766df6,0xad766df6 +.long 0x88cc7691,0x88cc7691 +.long 0xf5024c25,0xf5024c25 +.long 0x4fe5d7fc,0x4fe5d7fc +.long 0xc52acbd7,0xc52acbd7 +.long 0x26354480,0x26354480 +.long 0xb562a38f,0xb562a38f +.long 0xdeb15a49,0xdeb15a49 +.long 0x25ba1b67,0x25ba1b67 +.long 0x45ea0e98,0x45ea0e98 +.long 0x5dfec0e1,0x5dfec0e1 +.long 0xc32f7502,0xc32f7502 +.long 0x814cf012,0x814cf012 +.long 0x8d4697a3,0x8d4697a3 +.long 0x6bd3f9c6,0x6bd3f9c6 +.long 0x038f5fe7,0x038f5fe7 +.long 0x15929c95,0x15929c95 +.long 0xbf6d7aeb,0xbf6d7aeb +.long 0x955259da,0x955259da +.long 0xd4be832d,0xd4be832d +.long 0x587421d3,0x587421d3 +.long 0x49e06929,0x49e06929 +.long 0x8ec9c844,0x8ec9c844 +.long 0x75c2896a,0x75c2896a +.long 0xf48e7978,0xf48e7978 +.long 0x99583e6b,0x99583e6b +.long 0x27b971dd,0x27b971dd +.long 0xbee14fb6,0xbee14fb6 +.long 0xf088ad17,0xf088ad17 +.long 0xc920ac66,0xc920ac66 +.long 0x7dce3ab4,0x7dce3ab4 +.long 0x63df4a18,0x63df4a18 +.long 0xe51a3182,0xe51a3182 +.long 0x97513360,0x97513360 +.long 0x62537f45,0x62537f45 +.long 0xb16477e0,0xb16477e0 +.long 0xbb6bae84,0xbb6bae84 +.long 0xfe81a01c,0xfe81a01c +.long 0xf9082b94,0xf9082b94 +.long 0x70486858,0x70486858 +.long 0x8f45fd19,0x8f45fd19 +.long 0x94de6c87,0x94de6c87 +.long 0x527bf8b7,0x527bf8b7 +.long 0xab73d323,0xab73d323 +.long 0x724b02e2,0x724b02e2 +.long 0xe31f8f57,0xe31f8f57 +.long 0x6655ab2a,0x6655ab2a +.long 0xb2eb2807,0xb2eb2807 +.long 0x2fb5c203,0x2fb5c203 +.long 0x86c57b9a,0x86c57b9a +.long 0xd33708a5,0xd33708a5 +.long 0x302887f2,0x302887f2 +.long 0x23bfa5b2,0x23bfa5b2 +.long 0x02036aba,0x02036aba +.long 0xed16825c,0xed16825c +.long 0x8acf1c2b,0x8acf1c2b +.long 0xa779b492,0xa779b492 +.long 0xf307f2f0,0xf307f2f0 +.long 0x4e69e2a1,0x4e69e2a1 +.long 0x65daf4cd,0x65daf4cd +.long 0x0605bed5,0x0605bed5 +.long 0xd134621f,0xd134621f +.long 0xc4a6fe8a,0xc4a6fe8a +.long 0x342e539d,0x342e539d +.long 0xa2f355a0,0xa2f355a0 +.long 0x058ae132,0x058ae132 +.long 0xa4f6eb75,0xa4f6eb75 +.long 0x0b83ec39,0x0b83ec39 +.long 0x4060efaa,0x4060efaa +.long 0x5e719f06,0x5e719f06 +.long 0xbd6e1051,0xbd6e1051 +.long 0x3e218af9,0x3e218af9 +.long 0x96dd063d,0x96dd063d +.long 0xdd3e05ae,0xdd3e05ae +.long 0x4de6bd46,0x4de6bd46 +.long 0x91548db5,0x91548db5 +.long 0x71c45d05,0x71c45d05 +.long 0x0406d46f,0x0406d46f +.long 0x605015ff,0x605015ff +.long 0x1998fb24,0x1998fb24 +.long 0xd6bde997,0xd6bde997 +.long 0x894043cc,0x894043cc +.long 0x67d99e77,0x67d99e77 +.long 0xb0e842bd,0xb0e842bd +.long 0x07898b88,0x07898b88 +.long 0xe7195b38,0xe7195b38 +.long 0x79c8eedb,0x79c8eedb +.long 0xa17c0a47,0xa17c0a47 +.long 0x7c420fe9,0x7c420fe9 +.long 0xf8841ec9,0xf8841ec9 +.long 0x00000000,0x00000000 +.long 0x09808683,0x09808683 +.long 0x322bed48,0x322bed48 +.long 0x1e1170ac,0x1e1170ac +.long 0x6c5a724e,0x6c5a724e +.long 0xfd0efffb,0xfd0efffb +.long 0x0f853856,0x0f853856 +.long 0x3daed51e,0x3daed51e +.long 0x362d3927,0x362d3927 +.long 0x0a0fd964,0x0a0fd964 +.long 0x685ca621,0x685ca621 +.long 0x9b5b54d1,0x9b5b54d1 +.long 0x24362e3a,0x24362e3a +.long 0x0c0a67b1,0x0c0a67b1 +.long 0x9357e70f,0x9357e70f +.long 0xb4ee96d2,0xb4ee96d2 +.long 0x1b9b919e,0x1b9b919e +.long 0x80c0c54f,0x80c0c54f +.long 0x61dc20a2,0x61dc20a2 +.long 0x5a774b69,0x5a774b69 +.long 0x1c121a16,0x1c121a16 +.long 0xe293ba0a,0xe293ba0a +.long 0xc0a02ae5,0xc0a02ae5 +.long 0x3c22e043,0x3c22e043 +.long 0x121b171d,0x121b171d +.long 0x0e090d0b,0x0e090d0b +.long 0xf28bc7ad,0xf28bc7ad +.long 0x2db6a8b9,0x2db6a8b9 +.long 0x141ea9c8,0x141ea9c8 +.long 0x57f11985,0x57f11985 +.long 0xaf75074c,0xaf75074c +.long 0xee99ddbb,0xee99ddbb +.long 0xa37f60fd,0xa37f60fd +.long 0xf701269f,0xf701269f +.long 0x5c72f5bc,0x5c72f5bc +.long 0x44663bc5,0x44663bc5 +.long 0x5bfb7e34,0x5bfb7e34 +.long 0x8b432976,0x8b432976 +.long 0xcb23c6dc,0xcb23c6dc +.long 0xb6edfc68,0xb6edfc68 +.long 0xb8e4f163,0xb8e4f163 +.long 0xd731dcca,0xd731dcca +.long 0x42638510,0x42638510 +.long 0x13972240,0x13972240 +.long 0x84c61120,0x84c61120 +.long 0x854a247d,0x854a247d +.long 0xd2bb3df8,0xd2bb3df8 +.long 0xaef93211,0xaef93211 +.long 0xc729a16d,0xc729a16d +.long 0x1d9e2f4b,0x1d9e2f4b +.long 0xdcb230f3,0xdcb230f3 +.long 0x0d8652ec,0x0d8652ec +.long 0x77c1e3d0,0x77c1e3d0 +.long 0x2bb3166c,0x2bb3166c +.long 0xa970b999,0xa970b999 +.long 0x119448fa,0x119448fa +.long 0x47e96422,0x47e96422 +.long 0xa8fc8cc4,0xa8fc8cc4 +.long 0xa0f03f1a,0xa0f03f1a +.long 0x567d2cd8,0x567d2cd8 +.long 0x223390ef,0x223390ef +.long 0x87494ec7,0x87494ec7 +.long 0xd938d1c1,0xd938d1c1 +.long 0x8ccaa2fe,0x8ccaa2fe +.long 0x98d40b36,0x98d40b36 +.long 0xa6f581cf,0xa6f581cf +.long 0xa57ade28,0xa57ade28 +.long 0xdab78e26,0xdab78e26 +.long 0x3fadbfa4,0x3fadbfa4 +.long 0x2c3a9de4,0x2c3a9de4 +.long 0x5078920d,0x5078920d +.long 0x6a5fcc9b,0x6a5fcc9b +.long 0x547e4662,0x547e4662 +.long 0xf68d13c2,0xf68d13c2 +.long 0x90d8b8e8,0x90d8b8e8 +.long 0x2e39f75e,0x2e39f75e +.long 0x82c3aff5,0x82c3aff5 +.long 0x9f5d80be,0x9f5d80be +.long 0x69d0937c,0x69d0937c +.long 0x6fd52da9,0x6fd52da9 +.long 0xcf2512b3,0xcf2512b3 +.long 0xc8ac993b,0xc8ac993b +.long 0x10187da7,0x10187da7 +.long 0xe89c636e,0xe89c636e +.long 0xdb3bbb7b,0xdb3bbb7b +.long 0xcd267809,0xcd267809 +.long 0x6e5918f4,0x6e5918f4 +.long 0xec9ab701,0xec9ab701 +.long 0x834f9aa8,0x834f9aa8 +.long 0xe6956e65,0xe6956e65 +.long 0xaaffe67e,0xaaffe67e +.long 0x21bccf08,0x21bccf08 +.long 0xef15e8e6,0xef15e8e6 +.long 0xbae79bd9,0xbae79bd9 +.long 0x4a6f36ce,0x4a6f36ce +.long 0xea9f09d4,0xea9f09d4 +.long 0x29b07cd6,0x29b07cd6 +.long 0x31a4b2af,0x31a4b2af +.long 0x2a3f2331,0x2a3f2331 +.long 0xc6a59430,0xc6a59430 +.long 0x35a266c0,0x35a266c0 +.long 0x744ebc37,0x744ebc37 +.long 0xfc82caa6,0xfc82caa6 +.long 0xe090d0b0,0xe090d0b0 +.long 0x33a7d815,0x33a7d815 +.long 0xf104984a,0xf104984a +.long 0x41ecdaf7,0x41ecdaf7 +.long 0x7fcd500e,0x7fcd500e +.long 0x1791f62f,0x1791f62f +.long 0x764dd68d,0x764dd68d +.long 0x43efb04d,0x43efb04d +.long 0xccaa4d54,0xccaa4d54 +.long 0xe49604df,0xe49604df +.long 0x9ed1b5e3,0x9ed1b5e3 +.long 0x4c6a881b,0x4c6a881b +.long 0xc12c1fb8,0xc12c1fb8 +.long 0x4665517f,0x4665517f +.long 0x9d5eea04,0x9d5eea04 +.long 0x018c355d,0x018c355d +.long 0xfa877473,0xfa877473 +.long 0xfb0b412e,0xfb0b412e +.long 0xb3671d5a,0xb3671d5a +.long 0x92dbd252,0x92dbd252 +.long 0xe9105633,0xe9105633 +.long 0x6dd64713,0x6dd64713 +.long 0x9ad7618c,0x9ad7618c +.long 0x37a10c7a,0x37a10c7a +.long 0x59f8148e,0x59f8148e +.long 0xeb133c89,0xeb133c89 +.long 0xcea927ee,0xcea927ee +.long 0xb761c935,0xb761c935 +.long 0xe11ce5ed,0xe11ce5ed +.long 0x7a47b13c,0x7a47b13c +.long 0x9cd2df59,0x9cd2df59 +.long 0x55f2733f,0x55f2733f +.long 0x1814ce79,0x1814ce79 +.long 0x73c737bf,0x73c737bf +.long 0x53f7cdea,0x53f7cdea +.long 0x5ffdaa5b,0x5ffdaa5b +.long 0xdf3d6f14,0xdf3d6f14 +.long 0x7844db86,0x7844db86 +.long 0xcaaff381,0xcaaff381 +.long 0xb968c43e,0xb968c43e +.long 0x3824342c,0x3824342c +.long 0xc2a3405f,0xc2a3405f +.long 0x161dc372,0x161dc372 +.long 0xbce2250c,0xbce2250c +.long 0x283c498b,0x283c498b +.long 0xff0d9541,0xff0d9541 +.long 0x39a80171,0x39a80171 +.long 0x080cb3de,0x080cb3de +.long 0xd8b4e49c,0xd8b4e49c +.long 0x6456c190,0x6456c190 +.long 0x7bcb8461,0x7bcb8461 +.long 0xd532b670,0xd532b670 +.long 0x486c5c74,0x486c5c74 +.long 0xd0b85742,0xd0b85742 +.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 +.type AES_encrypt,@function +.section ".opd","aw" +.align 3 +AES_encrypt: +.quad .AES_encrypt,.TOC.@tocbase,0 +.previous +.align 7 +.AES_encrypt: + stdu 1,-256(1) + mflr 0 + + std 4,104(1) + std 14,112(1) + std 15,120(1) + std 16,128(1) + std 17,136(1) + std 18,144(1) + std 19,152(1) + std 20,160(1) + std 21,168(1) + std 22,176(1) + std 23,184(1) + std 24,192(1) + std 25,200(1) + std 26,208(1) + std 27,216(1) + std 28,224(1) + std 29,232(1) + std 30,240(1) + std 31,248(1) + std 0,272(1) + + andi. 12,3,3 + andi. 0,4,3 + or. 12,12,0 + bne .Lenc_unaligned + +.Lenc_unaligned_ok: + lwz 8,0(3) + lwz 9,4(3) + lwz 10,8(3) + lwz 11,12(3) + bl .LAES_Te + bl .Lppc_AES_encrypt_compact + ld 4,104(1) + stw 8,0(4) + stw 9,4(4) + stw 10,8(4) + stw 11,12(4) + b .Lenc_done + +.Lenc_unaligned: + subfic 12,3,4096 + subfic 0,4,4096 + andi. 12,12,4096-16 + beq .Lenc_xpage + andi. 0,0,4096-16 + bne .Lenc_unaligned_ok + +.Lenc_xpage: + lbz 16,0(3) + lbz 17,1(3) + lbz 18,2(3) + lbz 8,3(3) + lbz 20,4(3) + lbz 21,5(3) + lbz 22,6(3) + lbz 9,7(3) + lbz 24,8(3) + lbz 25,9(3) + lbz 26,10(3) + insrwi 8,16,8,0 + lbz 10,11(3) + insrwi 9,20,8,0 + lbz 28,12(3) + insrwi 8,17,8,8 + lbz 29,13(3) + insrwi 9,21,8,8 + lbz 30,14(3) + insrwi 8,18,8,16 + lbz 11,15(3) + insrwi 9,22,8,16 + insrwi 10,24,8,0 + insrwi 11,28,8,0 + insrwi 10,25,8,8 + insrwi 11,29,8,8 + insrwi 10,26,8,16 + insrwi 11,30,8,16 + + bl .LAES_Te + bl .Lppc_AES_encrypt_compact + ld 4,104(1) + + extrwi 16,8,8,0 + extrwi 17,8,8,8 + stb 16,0(4) + extrwi 18,8,8,16 + stb 17,1(4) + stb 18,2(4) + extrwi 20,9,8,0 + stb 8,3(4) + extrwi 21,9,8,8 + stb 20,4(4) + extrwi 22,9,8,16 + stb 21,5(4) + stb 22,6(4) + extrwi 24,10,8,0 + stb 9,7(4) + extrwi 25,10,8,8 + stb 24,8(4) + extrwi 26,10,8,16 + stb 25,9(4) + stb 26,10(4) + extrwi 28,11,8,0 + stb 10,11(4) + extrwi 29,11,8,8 + stb 28,12(4) + extrwi 30,11,8,16 + stb 29,13(4) + stb 30,14(4) + stb 11,15(4) + +.Lenc_done: + ld 0,272(1) + ld 14,112(1) + ld 15,120(1) + ld 16,128(1) + ld 17,136(1) + ld 18,144(1) + ld 19,152(1) + ld 20,160(1) + ld 21,168(1) + ld 22,176(1) + ld 23,184(1) + ld 24,192(1) + ld 25,200(1) + ld 26,208(1) + ld 27,216(1) + ld 28,224(1) + ld 29,232(1) + ld 30,240(1) + ld 31,248(1) + mtlr 0 + addi 1,1,256 + blr +.long 0 +.byte 0,12,4,1,0x80,18,3,0 +.long 0 + +.align 5 +.Lppc_AES_encrypt: + lwz 16,240(5) + addi 6,3,3 + lwz 12,0(5) + addi 7,3,2 + lwz 0,4(5) + addi 4,3,1 + lwz 14,8(5) + addi 16,16,-1 + lwz 15,12(5) + addi 5,5,16 + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + mtctr 16 +.align 4 +.Lenc_loop: + rlwinm 16,8,11,21,28 + rlwinm 17,9,11,21,28 + rlwinm 18,10,11,21,28 + rlwinm 19,11,11,21,28 + lwz 12,0(5) + rlwinm 20,9,19,21,28 + lwz 0,4(5) + rlwinm 21,10,19,21,28 + lwz 14,8(5) + rlwinm 22,11,19,21,28 + lwz 15,12(5) + rlwinm 23,8,19,21,28 + lwzx 16,3,16 + rlwinm 24,10,27,21,28 + lwzx 17,3,17 + rlwinm 25,11,27,21,28 + lwzx 18,3,18 + rlwinm 26,8,27,21,28 + lwzx 19,3,19 + rlwinm 27,9,27,21,28 + lwzx 20,6,20 + rlwinm 28,11,3,21,28 + lwzx 21,6,21 + rlwinm 29,8,3,21,28 + lwzx 22,6,22 + rlwinm 30,9,3,21,28 + lwzx 23,6,23 + rlwinm 31,10,3,21,28 + lwzx 24,7,24 + xor 12,12,16 + lwzx 25,7,25 + xor 0,0,17 + lwzx 26,7,26 + xor 14,14,18 + lwzx 27,7,27 + xor 15,15,19 + lwzx 28,4,28 + xor 12,12,20 + lwzx 29,4,29 + xor 0,0,21 + lwzx 30,4,30 + xor 14,14,22 + lwzx 31,4,31 + xor 15,15,23 + xor 12,12,24 + xor 0,0,25 + xor 14,14,26 + xor 15,15,27 + xor 8,12,28 + xor 9,0,29 + xor 10,14,30 + xor 11,15,31 + addi 5,5,16 + bdnz .Lenc_loop + + addi 7,3,2048 + nop + lwz 12,0(5) + rlwinm 16,8,8,24,31 + lwz 0,4(5) + rlwinm 17,9,8,24,31 + lwz 14,8(5) + rlwinm 18,10,8,24,31 + lwz 15,12(5) + rlwinm 19,11,8,24,31 + lwz 24,2048(3) + rlwinm 20,9,16,24,31 + lwz 25,2080(3) + rlwinm 21,10,16,24,31 + lwz 26,2112(3) + rlwinm 22,11,16,24,31 + lwz 27,2144(3) + rlwinm 23,8,16,24,31 + lwz 28,2176(3) + rlwinm 24,10,24,24,31 + lwz 29,2208(3) + rlwinm 25,11,24,24,31 + lwz 30,2240(3) + rlwinm 26,8,24,24,31 + lwz 31,2272(3) + rlwinm 27,9,24,24,31 + lbzx 16,7,16 + rlwinm 28,11,0,24,31 + lbzx 17,7,17 + rlwinm 29,8,0,24,31 + lbzx 18,7,18 + rlwinm 30,9,0,24,31 + lbzx 19,7,19 + rlwinm 31,10,0,24,31 + lbzx 20,7,20 + rlwinm 8,16,24,0,7 + lbzx 21,7,21 + rlwinm 9,17,24,0,7 + lbzx 22,7,22 + rlwinm 10,18,24,0,7 + lbzx 23,7,23 + rlwinm 11,19,24,0,7 + lbzx 24,7,24 + rlwimi 8,20,16,8,15 + lbzx 25,7,25 + rlwimi 9,21,16,8,15 + lbzx 26,7,26 + rlwimi 10,22,16,8,15 + lbzx 27,7,27 + rlwimi 11,23,16,8,15 + lbzx 28,7,28 + rlwimi 8,24,8,16,23 + lbzx 29,7,29 + rlwimi 9,25,8,16,23 + lbzx 30,7,30 + rlwimi 10,26,8,16,23 + lbzx 31,7,31 + rlwimi 11,27,8,16,23 + or 8,8,28 + or 9,9,29 + or 10,10,30 + or 11,11,31 + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.align 4 +.Lppc_AES_encrypt_compact: + lwz 16,240(5) + addi 6,3,2048 + lwz 12,0(5) + lis 7,0x8080 + lwz 0,4(5) + lis 4,0x1b1b + lwz 14,8(5) + ori 7,7,0x8080 + lwz 15,12(5) + ori 4,4,0x1b1b + addi 5,5,16 + mtctr 16 +.align 4 +.Lenc_compact_loop: + xor 8,8,12 + xor 9,9,0 + rlwinm 16,8,8,24,31 + xor 10,10,14 + rlwinm 17,9,8,24,31 + xor 11,11,15 + rlwinm 18,10,8,24,31 + rlwinm 19,11,8,24,31 + rlwinm 20,9,16,24,31 + rlwinm 21,10,16,24,31 + rlwinm 22,11,16,24,31 + rlwinm 23,8,16,24,31 + lbzx 16,6,16 + rlwinm 24,10,24,24,31 + lbzx 17,6,17 + rlwinm 25,11,24,24,31 + lbzx 18,6,18 + rlwinm 26,8,24,24,31 + lbzx 19,6,19 + rlwinm 27,9,24,24,31 + lbzx 20,6,20 + rlwinm 28,11,0,24,31 + lbzx 21,6,21 + rlwinm 29,8,0,24,31 + lbzx 22,6,22 + rlwinm 30,9,0,24,31 + lbzx 23,6,23 + rlwinm 31,10,0,24,31 + lbzx 24,6,24 + rlwinm 8,16,24,0,7 + lbzx 25,6,25 + rlwinm 9,17,24,0,7 + lbzx 26,6,26 + rlwinm 10,18,24,0,7 + lbzx 27,6,27 + rlwinm 11,19,24,0,7 + lbzx 28,6,28 + rlwimi 8,20,16,8,15 + lbzx 29,6,29 + rlwimi 9,21,16,8,15 + lbzx 30,6,30 + rlwimi 10,22,16,8,15 + lbzx 31,6,31 + rlwimi 11,23,16,8,15 + rlwimi 8,24,8,16,23 + rlwimi 9,25,8,16,23 + rlwimi 10,26,8,16,23 + rlwimi 11,27,8,16,23 + lwz 12,0(5) + or 8,8,28 + lwz 0,4(5) + or 9,9,29 + lwz 14,8(5) + or 10,10,30 + lwz 15,12(5) + or 11,11,31 + + addi 5,5,16 + bdz .Lenc_compact_done + + and 16,8,7 + and 17,9,7 + and 18,10,7 + and 19,11,7 + srwi 20,16,7 + andc 24,8,7 + srwi 21,17,7 + andc 25,9,7 + srwi 22,18,7 + andc 26,10,7 + srwi 23,19,7 + andc 27,11,7 + sub 16,16,20 + sub 17,17,21 + sub 18,18,22 + sub 19,19,23 + add 24,24,24 + add 25,25,25 + add 26,26,26 + add 27,27,27 + and 16,16,4 + and 17,17,4 + and 18,18,4 + and 19,19,4 + xor 16,16,24 + xor 17,17,25 + rotlwi 28,8,16 + xor 18,18,26 + rotlwi 29,9,16 + xor 19,19,27 + rotlwi 30,10,16 + + xor 8,8,16 + rotlwi 31,11,16 + xor 9,9,17 + rotrwi 8,8,24 + xor 10,10,18 + rotrwi 9,9,24 + xor 11,11,19 + rotrwi 10,10,24 + xor 8,8,16 + rotrwi 11,11,24 + xor 9,9,17 + xor 10,10,18 + xor 11,11,19 + rotlwi 24,28,8 + xor 8,8,28 + rotlwi 25,29,8 + xor 9,9,29 + rotlwi 26,30,8 + xor 10,10,30 + rotlwi 27,31,8 + xor 11,11,31 + xor 8,8,24 + xor 9,9,25 + xor 10,10,26 + xor 11,11,27 + + b .Lenc_compact_loop +.align 4 +.Lenc_compact_done: + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.size .AES_encrypt,.-.AES_encrypt +.size AES_encrypt,.-.AES_encrypt + +.globl AES_decrypt +.type AES_decrypt,@function +.section ".opd","aw" +.align 3 +AES_decrypt: +.quad .AES_decrypt,.TOC.@tocbase,0 +.previous +.align 7 +.AES_decrypt: + stdu 1,-256(1) + mflr 0 + + std 4,104(1) + std 14,112(1) + std 15,120(1) + std 16,128(1) + std 17,136(1) + std 18,144(1) + std 19,152(1) + std 20,160(1) + std 21,168(1) + std 22,176(1) + std 23,184(1) + std 24,192(1) + std 25,200(1) + std 26,208(1) + std 27,216(1) + std 28,224(1) + std 29,232(1) + std 30,240(1) + std 31,248(1) + std 0,272(1) + + andi. 12,3,3 + andi. 0,4,3 + or. 12,12,0 + bne .Ldec_unaligned + +.Ldec_unaligned_ok: + lwz 8,0(3) + lwz 9,4(3) + lwz 10,8(3) + lwz 11,12(3) + bl .LAES_Td + bl .Lppc_AES_decrypt_compact + ld 4,104(1) + stw 8,0(4) + stw 9,4(4) + stw 10,8(4) + stw 11,12(4) + b .Ldec_done + +.Ldec_unaligned: + subfic 12,3,4096 + subfic 0,4,4096 + andi. 12,12,4096-16 + beq .Ldec_xpage + andi. 0,0,4096-16 + bne .Ldec_unaligned_ok + +.Ldec_xpage: + lbz 16,0(3) + lbz 17,1(3) + lbz 18,2(3) + lbz 8,3(3) + lbz 20,4(3) + lbz 21,5(3) + lbz 22,6(3) + lbz 9,7(3) + lbz 24,8(3) + lbz 25,9(3) + lbz 26,10(3) + insrwi 8,16,8,0 + lbz 10,11(3) + insrwi 9,20,8,0 + lbz 28,12(3) + insrwi 8,17,8,8 + lbz 29,13(3) + insrwi 9,21,8,8 + lbz 30,14(3) + insrwi 8,18,8,16 + lbz 11,15(3) + insrwi 9,22,8,16 + insrwi 10,24,8,0 + insrwi 11,28,8,0 + insrwi 10,25,8,8 + insrwi 11,29,8,8 + insrwi 10,26,8,16 + insrwi 11,30,8,16 + + bl .LAES_Td + bl .Lppc_AES_decrypt_compact + ld 4,104(1) + + extrwi 16,8,8,0 + extrwi 17,8,8,8 + stb 16,0(4) + extrwi 18,8,8,16 + stb 17,1(4) + stb 18,2(4) + extrwi 20,9,8,0 + stb 8,3(4) + extrwi 21,9,8,8 + stb 20,4(4) + extrwi 22,9,8,16 + stb 21,5(4) + stb 22,6(4) + extrwi 24,10,8,0 + stb 9,7(4) + extrwi 25,10,8,8 + stb 24,8(4) + extrwi 26,10,8,16 + stb 25,9(4) + stb 26,10(4) + extrwi 28,11,8,0 + stb 10,11(4) + extrwi 29,11,8,8 + stb 28,12(4) + extrwi 30,11,8,16 + stb 29,13(4) + stb 30,14(4) + stb 11,15(4) + +.Ldec_done: + ld 0,272(1) + ld 14,112(1) + ld 15,120(1) + ld 16,128(1) + ld 17,136(1) + ld 18,144(1) + ld 19,152(1) + ld 20,160(1) + ld 21,168(1) + ld 22,176(1) + ld 23,184(1) + ld 24,192(1) + ld 25,200(1) + ld 26,208(1) + ld 27,216(1) + ld 28,224(1) + ld 29,232(1) + ld 30,240(1) + ld 31,248(1) + mtlr 0 + addi 1,1,256 + blr +.long 0 +.byte 0,12,4,1,0x80,18,3,0 +.long 0 + +.align 5 +.Lppc_AES_decrypt: + lwz 16,240(5) + addi 6,3,3 + lwz 12,0(5) + addi 7,3,2 + lwz 0,4(5) + addi 4,3,1 + lwz 14,8(5) + addi 16,16,-1 + lwz 15,12(5) + addi 5,5,16 + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + mtctr 16 +.align 4 +.Ldec_loop: + rlwinm 16,8,11,21,28 + rlwinm 17,9,11,21,28 + rlwinm 18,10,11,21,28 + rlwinm 19,11,11,21,28 + lwz 12,0(5) + rlwinm 20,11,19,21,28 + lwz 0,4(5) + rlwinm 21,8,19,21,28 + lwz 14,8(5) + rlwinm 22,9,19,21,28 + lwz 15,12(5) + rlwinm 23,10,19,21,28 + lwzx 16,3,16 + rlwinm 24,10,27,21,28 + lwzx 17,3,17 + rlwinm 25,11,27,21,28 + lwzx 18,3,18 + rlwinm 26,8,27,21,28 + lwzx 19,3,19 + rlwinm 27,9,27,21,28 + lwzx 20,6,20 + rlwinm 28,9,3,21,28 + lwzx 21,6,21 + rlwinm 29,10,3,21,28 + lwzx 22,6,22 + rlwinm 30,11,3,21,28 + lwzx 23,6,23 + rlwinm 31,8,3,21,28 + lwzx 24,7,24 + xor 12,12,16 + lwzx 25,7,25 + xor 0,0,17 + lwzx 26,7,26 + xor 14,14,18 + lwzx 27,7,27 + xor 15,15,19 + lwzx 28,4,28 + xor 12,12,20 + lwzx 29,4,29 + xor 0,0,21 + lwzx 30,4,30 + xor 14,14,22 + lwzx 31,4,31 + xor 15,15,23 + xor 12,12,24 + xor 0,0,25 + xor 14,14,26 + xor 15,15,27 + xor 8,12,28 + xor 9,0,29 + xor 10,14,30 + xor 11,15,31 + addi 5,5,16 + bdnz .Ldec_loop + + addi 7,3,2048 + nop + lwz 12,0(5) + rlwinm 16,8,8,24,31 + lwz 0,4(5) + rlwinm 17,9,8,24,31 + lwz 14,8(5) + rlwinm 18,10,8,24,31 + lwz 15,12(5) + rlwinm 19,11,8,24,31 + lwz 24,2048(3) + rlwinm 20,11,16,24,31 + lwz 25,2080(3) + rlwinm 21,8,16,24,31 + lwz 26,2112(3) + lbzx 16,7,16 + lwz 27,2144(3) + lbzx 17,7,17 + lwz 28,2176(3) + rlwinm 22,9,16,24,31 + lwz 29,2208(3) + rlwinm 23,10,16,24,31 + lwz 30,2240(3) + rlwinm 24,10,24,24,31 + lwz 31,2272(3) + rlwinm 25,11,24,24,31 + lbzx 18,7,18 + rlwinm 26,8,24,24,31 + lbzx 19,7,19 + rlwinm 27,9,24,24,31 + lbzx 20,7,20 + rlwinm 28,9,0,24,31 + lbzx 21,7,21 + rlwinm 29,10,0,24,31 + lbzx 22,7,22 + rlwinm 30,11,0,24,31 + lbzx 23,7,23 + rlwinm 31,8,0,24,31 + lbzx 24,7,24 + rlwinm 8,16,24,0,7 + lbzx 25,7,25 + rlwinm 9,17,24,0,7 + lbzx 26,7,26 + rlwinm 10,18,24,0,7 + lbzx 27,7,27 + rlwinm 11,19,24,0,7 + lbzx 28,7,28 + rlwimi 8,20,16,8,15 + lbzx 29,7,29 + rlwimi 9,21,16,8,15 + lbzx 30,7,30 + rlwimi 10,22,16,8,15 + lbzx 31,7,31 + rlwimi 11,23,16,8,15 + rlwimi 8,24,8,16,23 + rlwimi 9,25,8,16,23 + rlwimi 10,26,8,16,23 + rlwimi 11,27,8,16,23 + or 8,8,28 + or 9,9,29 + or 10,10,30 + or 11,11,31 + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.align 4 +.Lppc_AES_decrypt_compact: + lwz 16,240(5) + addi 6,3,2048 + lwz 12,0(5) + lis 7,0x8080 + lwz 0,4(5) + lis 4,0x1b1b + lwz 14,8(5) + ori 7,7,0x8080 + lwz 15,12(5) + ori 4,4,0x1b1b + addi 5,5,16 + insrdi 7,7,32,0 + insrdi 4,4,32,0 + mtctr 16 +.align 4 +.Ldec_compact_loop: + xor 8,8,12 + xor 9,9,0 + rlwinm 16,8,8,24,31 + xor 10,10,14 + rlwinm 17,9,8,24,31 + xor 11,11,15 + rlwinm 18,10,8,24,31 + rlwinm 19,11,8,24,31 + rlwinm 20,11,16,24,31 + rlwinm 21,8,16,24,31 + rlwinm 22,9,16,24,31 + rlwinm 23,10,16,24,31 + lbzx 16,6,16 + rlwinm 24,10,24,24,31 + lbzx 17,6,17 + rlwinm 25,11,24,24,31 + lbzx 18,6,18 + rlwinm 26,8,24,24,31 + lbzx 19,6,19 + rlwinm 27,9,24,24,31 + lbzx 20,6,20 + rlwinm 28,9,0,24,31 + lbzx 21,6,21 + rlwinm 29,10,0,24,31 + lbzx 22,6,22 + rlwinm 30,11,0,24,31 + lbzx 23,6,23 + rlwinm 31,8,0,24,31 + lbzx 24,6,24 + rlwinm 8,16,24,0,7 + lbzx 25,6,25 + rlwinm 9,17,24,0,7 + lbzx 26,6,26 + rlwinm 10,18,24,0,7 + lbzx 27,6,27 + rlwinm 11,19,24,0,7 + lbzx 28,6,28 + rlwimi 8,20,16,8,15 + lbzx 29,6,29 + rlwimi 9,21,16,8,15 + lbzx 30,6,30 + rlwimi 10,22,16,8,15 + lbzx 31,6,31 + rlwimi 11,23,16,8,15 + rlwimi 8,24,8,16,23 + rlwimi 9,25,8,16,23 + rlwimi 10,26,8,16,23 + rlwimi 11,27,8,16,23 + lwz 12,0(5) + or 8,8,28 + lwz 0,4(5) + or 9,9,29 + lwz 14,8(5) + or 10,10,30 + lwz 15,12(5) + or 11,11,31 + + addi 5,5,16 + bdz .Ldec_compact_done + + insrdi 8,9,32,0 + insrdi 10,11,32,0 + + and 16,8,7 + and 18,10,7 + srdi 20,16,7 + srdi 22,18,7 + andc 24,8,7 + andc 26,10,7 + sub 16,16,20 + sub 18,18,22 + add 24,24,24 + add 26,26,26 + and 16,16,4 + and 18,18,4 + xor 16,16,24 + xor 18,18,26 + + and 20,16,7 + and 22,18,7 + srdi 24,20,7 + srdi 26,22,7 + andc 28,16,7 + andc 30,18,7 + sub 20,20,24 + sub 22,22,26 + add 28,28,28 + add 30,30,30 + and 20,20,4 + and 22,22,4 + xor 20,20,28 + xor 22,22,30 + + and 24,20,7 + and 26,22,7 + srdi 28,24,7 + srdi 30,26,7 + sub 24,24,28 + sub 26,26,30 + andc 28,20,7 + andc 30,22,7 + add 28,28,28 + add 30,30,30 + and 24,24,4 + and 26,26,4 + xor 24,24,28 + xor 26,26,30 + + xor 16,16,8 + xor 18,18,10 + xor 20,20,8 + xor 22,22,10 + + rldicl 17,16,32,32 + rldicl 19,18,32,32 + rldicl 21,20,32,32 + rldicl 23,22,32,32 + rldicl 25,24,32,32 + rldicl 27,26,32,32 + rotrwi 8,8,8 + rotrwi 9,9,8 + xor 8,8,16 + rotrwi 10,10,8 + xor 9,9,17 + rotrwi 11,11,8 + xor 10,10,18 + xor 11,11,19 + xor 16,16,24 + xor 17,17,25 + xor 18,18,26 + xor 19,19,27 + xor 8,8,20 + rotrwi 16,16,24 + xor 9,9,21 + rotrwi 17,17,24 + xor 10,10,22 + rotrwi 18,18,24 + xor 11,11,23 + rotrwi 19,19,24 + xor 20,20,24 + xor 21,21,25 + xor 22,22,26 + xor 23,23,27 + xor 8,8,24 + rotrwi 20,20,16 + xor 9,9,25 + rotrwi 21,21,16 + xor 10,10,26 + rotrwi 22,22,16 + xor 11,11,27 + rotrwi 23,23,16 + xor 8,8,16 + rotrwi 24,24,8 + xor 9,9,17 + rotrwi 25,25,8 + xor 10,10,18 + rotrwi 26,26,8 + xor 11,11,19 + rotrwi 27,27,8 + xor 8,8,20 + xor 9,9,21 + xor 10,10,22 + xor 11,11,23 + xor 8,8,24 + xor 9,9,25 + xor 10,10,26 + xor 11,11,27 + + b .Ldec_compact_loop +.align 4 +.Ldec_compact_done: + xor 8,8,12 + xor 9,9,0 + xor 10,10,14 + xor 11,11,15 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.size .AES_decrypt,.-.AES_decrypt +.size AES_decrypt,.-.AES_decrypt + +.byte 65,69,83,32,102,111,114,32,80,80,67,44,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0 +.align 2 +.align 7 diff --git a/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aesp8-ppc.s b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aesp8-ppc.s new file mode 100644 index 0000000000..b850d2226d --- /dev/null +++ b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/aesp8-ppc.s @@ -0,0 +1,3689 @@ +.machine "any" + +.text + +.align 7 +rcon: +.byte 0x01,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x01,0x00,0x00,0x00,0x01,0x00,0x00,0x00 +.byte 0x1b,0x00,0x00,0x00,0x1b,0x00,0x00,0x00,0x1b,0x00,0x00,0x00,0x1b,0x00,0x00,0x00 +.byte 0x0d,0x0e,0x0f,0x0c,0x0d,0x0e,0x0f,0x0c,0x0d,0x0e,0x0f,0x0c,0x0d,0x0e,0x0f,0x0c +.byte 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00 +.Lconsts: + mflr 0 + bcl 20,31,$+4 + mflr 6 + addi 6,6,-0x48 + mtlr 0 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.byte 65,69,83,32,102,111,114,32,80,111,119,101,114,73,83,65,32,50,46,48,55,44,67,82,89,80,84,79,71,65,77,83,32,98,121,32,60,97,112,112,114,111,64,111,112,101,110,115,115,108,46,111,114,103,62,0 +.align 2 + +.globl aes_p8_set_encrypt_key +.type aes_p8_set_encrypt_key,@function +.section ".opd","aw" +.align 3 +aes_p8_set_encrypt_key: +.quad .aes_p8_set_encrypt_key,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_set_encrypt_key: +.Lset_encrypt_key: + mflr 11 + std 11,16(1) + + li 6,-1 + cmpldi 3,0 + beq- .Lenc_key_abort + cmpldi 5,0 + beq- .Lenc_key_abort + li 6,-2 + cmpwi 4,128 + blt- .Lenc_key_abort + cmpwi 4,256 + bgt- .Lenc_key_abort + andi. 0,4,0x3f + bne- .Lenc_key_abort + + lis 0,0xfff0 + mfspr 12,256 + mtspr 256,0 + + bl .Lconsts + mtlr 11 + + neg 9,3 + lvx 1,0,3 + addi 3,3,15 + lvsr 3,0,9 + li 8,0x20 + cmpwi 4,192 + lvx 2,0,3 + + lvx 4,0,6 + + lvx 5,8,6 + addi 6,6,0x10 + vperm 1,1,2,3 + li 7,8 + vxor 0,0,0 + mtctr 7 + + lvsr 8,0,5 + vspltisb 9,-1 + lvx 10,0,5 + vperm 9,0,9,8 + + blt .Loop128 + addi 3,3,8 + beq .L192 + addi 3,3,8 + b .L256 + +.align 4 +.Loop128: + vperm 3,1,1,5 + vsldoi 6,0,1,12 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + .long 0x10632509 + stvx 7,0,5 + addi 5,5,16 + + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vadduwm 4,4,4 + vxor 1,1,3 + bdnz .Loop128 + + lvx 4,0,6 + + vperm 3,1,1,5 + vsldoi 6,0,1,12 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + .long 0x10632509 + stvx 7,0,5 + addi 5,5,16 + + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vadduwm 4,4,4 + vxor 1,1,3 + + vperm 3,1,1,5 + vsldoi 6,0,1,12 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + .long 0x10632509 + stvx 7,0,5 + addi 5,5,16 + + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vxor 1,1,3 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + stvx 7,0,5 + + addi 3,5,15 + addi 5,5,0x50 + + li 8,10 + b .Ldone + +.align 4 +.L192: + lvx 6,0,3 + li 7,4 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + stvx 7,0,5 + addi 5,5,16 + vperm 2,2,6,3 + vspltisb 3,8 + mtctr 7 + vsububm 5,5,3 + +.Loop192: + vperm 3,2,2,5 + vsldoi 6,0,1,12 + .long 0x10632509 + + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + + vsldoi 7,0,2,8 + vspltw 6,1,3 + vxor 6,6,2 + vsldoi 2,0,2,12 + vadduwm 4,4,4 + vxor 2,2,6 + vxor 1,1,3 + vxor 2,2,3 + vsldoi 7,7,1,8 + + vperm 3,2,2,5 + vsldoi 6,0,1,12 + vperm 11,7,7,8 + vsel 7,10,11,9 + vor 10,11,11 + .long 0x10632509 + stvx 7,0,5 + addi 5,5,16 + + vsldoi 7,1,2,8 + vxor 1,1,6 + vsldoi 6,0,6,12 + vperm 11,7,7,8 + vsel 7,10,11,9 + vor 10,11,11 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + stvx 7,0,5 + addi 5,5,16 + + vspltw 6,1,3 + vxor 6,6,2 + vsldoi 2,0,2,12 + vadduwm 4,4,4 + vxor 2,2,6 + vxor 1,1,3 + vxor 2,2,3 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + stvx 7,0,5 + addi 3,5,15 + addi 5,5,16 + bdnz .Loop192 + + li 8,12 + addi 5,5,0x20 + b .Ldone + +.align 4 +.L256: + lvx 6,0,3 + li 7,7 + li 8,14 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + stvx 7,0,5 + addi 5,5,16 + vperm 2,2,6,3 + mtctr 7 + +.Loop256: + vperm 3,2,2,5 + vsldoi 6,0,1,12 + vperm 11,2,2,8 + vsel 7,10,11,9 + vor 10,11,11 + .long 0x10632509 + stvx 7,0,5 + addi 5,5,16 + + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vsldoi 6,0,6,12 + vxor 1,1,6 + vadduwm 4,4,4 + vxor 1,1,3 + vperm 11,1,1,8 + vsel 7,10,11,9 + vor 10,11,11 + stvx 7,0,5 + addi 3,5,15 + addi 5,5,16 + bdz .Ldone + + vspltw 3,1,3 + vsldoi 6,0,2,12 + .long 0x106305C8 + + vxor 2,2,6 + vsldoi 6,0,6,12 + vxor 2,2,6 + vsldoi 6,0,6,12 + vxor 2,2,6 + + vxor 2,2,3 + b .Loop256 + +.align 4 +.Ldone: + lvx 2,0,3 + vsel 2,10,2,9 + stvx 2,0,3 + li 6,0 + mtspr 256,12 + stw 8,0(5) + +.Lenc_key_abort: + mr 3,6 + blr +.long 0 +.byte 0,12,0x14,1,0,0,3,0 +.long 0 +.size .aes_p8_set_encrypt_key,.-.aes_p8_set_encrypt_key +.size aes_p8_set_encrypt_key,.-.aes_p8_set_encrypt_key + +.globl aes_p8_set_decrypt_key +.type aes_p8_set_decrypt_key,@function +.section ".opd","aw" +.align 3 +aes_p8_set_decrypt_key: +.quad .aes_p8_set_decrypt_key,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_set_decrypt_key: + stdu 1,-64(1) + mflr 10 + std 10,64+16(1) + bl .Lset_encrypt_key + mtlr 10 + + cmpwi 3,0 + bne- .Ldec_key_abort + + slwi 7,8,4 + subi 3,5,240 + srwi 8,8,1 + add 5,3,7 + mtctr 8 + +.Ldeckey: + lwz 0, 0(3) + lwz 6, 4(3) + lwz 7, 8(3) + lwz 8, 12(3) + addi 3,3,16 + lwz 9, 0(5) + lwz 10,4(5) + lwz 11,8(5) + lwz 12,12(5) + stw 0, 0(5) + stw 6, 4(5) + stw 7, 8(5) + stw 8, 12(5) + subi 5,5,16 + stw 9, -16(3) + stw 10,-12(3) + stw 11,-8(3) + stw 12,-4(3) + bdnz .Ldeckey + + xor 3,3,3 +.Ldec_key_abort: + addi 1,1,64 + blr +.long 0 +.byte 0,12,4,1,0x80,0,3,0 +.long 0 +.size .aes_p8_set_decrypt_key,.-.aes_p8_set_decrypt_key +.size aes_p8_set_decrypt_key,.-.aes_p8_set_decrypt_key +.globl aes_p8_encrypt +.type aes_p8_encrypt,@function +.section ".opd","aw" +.align 3 +aes_p8_encrypt: +.quad .aes_p8_encrypt,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_encrypt: + lwz 6,240(5) + lis 0,0xfc00 + mfspr 12,256 + li 7,15 + mtspr 256,0 + + lvx 0,0,3 + neg 11,4 + lvx 1,7,3 + lvsl 2,0,3 + + lvsl 3,0,11 + + li 7,16 + vperm 0,0,1,2 + lvx 1,0,5 + lvsl 5,0,5 + srwi 6,6,1 + lvx 2,7,5 + addi 7,7,16 + subi 6,6,1 + vperm 1,1,2,5 + + vxor 0,0,1 + lvx 1,7,5 + addi 7,7,16 + mtctr 6 + +.Loop_enc: + vperm 2,2,1,5 + .long 0x10001508 + lvx 2,7,5 + addi 7,7,16 + vperm 1,1,2,5 + .long 0x10000D08 + lvx 1,7,5 + addi 7,7,16 + bdnz .Loop_enc + + vperm 2,2,1,5 + .long 0x10001508 + lvx 2,7,5 + vperm 1,1,2,5 + .long 0x10000D09 + + vspltisb 2,-1 + vxor 1,1,1 + li 7,15 + vperm 2,1,2,3 + + lvx 1,0,4 + vperm 0,0,0,3 + vsel 1,1,0,2 + lvx 4,7,4 + stvx 1,0,4 + vsel 0,0,4,2 + stvx 0,7,4 + + mtspr 256,12 + blr +.long 0 +.byte 0,12,0x14,0,0,0,3,0 +.long 0 +.size .aes_p8_encrypt,.-.aes_p8_encrypt +.size aes_p8_encrypt,.-.aes_p8_encrypt +.globl aes_p8_decrypt +.type aes_p8_decrypt,@function +.section ".opd","aw" +.align 3 +aes_p8_decrypt: +.quad .aes_p8_decrypt,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_decrypt: + lwz 6,240(5) + lis 0,0xfc00 + mfspr 12,256 + li 7,15 + mtspr 256,0 + + lvx 0,0,3 + neg 11,4 + lvx 1,7,3 + lvsl 2,0,3 + + lvsl 3,0,11 + + li 7,16 + vperm 0,0,1,2 + lvx 1,0,5 + lvsl 5,0,5 + srwi 6,6,1 + lvx 2,7,5 + addi 7,7,16 + subi 6,6,1 + vperm 1,1,2,5 + + vxor 0,0,1 + lvx 1,7,5 + addi 7,7,16 + mtctr 6 + +.Loop_dec: + vperm 2,2,1,5 + .long 0x10001548 + lvx 2,7,5 + addi 7,7,16 + vperm 1,1,2,5 + .long 0x10000D48 + lvx 1,7,5 + addi 7,7,16 + bdnz .Loop_dec + + vperm 2,2,1,5 + .long 0x10001548 + lvx 2,7,5 + vperm 1,1,2,5 + .long 0x10000D49 + + vspltisb 2,-1 + vxor 1,1,1 + li 7,15 + vperm 2,1,2,3 + + lvx 1,0,4 + vperm 0,0,0,3 + vsel 1,1,0,2 + lvx 4,7,4 + stvx 1,0,4 + vsel 0,0,4,2 + stvx 0,7,4 + + mtspr 256,12 + blr +.long 0 +.byte 0,12,0x14,0,0,0,3,0 +.long 0 +.size .aes_p8_decrypt,.-.aes_p8_decrypt +.size aes_p8_decrypt,.-.aes_p8_decrypt +.globl aes_p8_cbc_encrypt +.type aes_p8_cbc_encrypt,@function +.section ".opd","aw" +.align 3 +aes_p8_cbc_encrypt: +.quad .aes_p8_cbc_encrypt,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_cbc_encrypt: + cmpldi 5,16 + .long 0x4dc00020 + + cmpwi 8,0 + lis 0,0xffe0 + mfspr 12,256 + mtspr 256,0 + + li 10,15 + vxor 0,0,0 + + + lvx 4,0,7 + lvsl 6,0,7 + lvx 5,10,7 + + vperm 4,4,5,6 + + neg 11,3 + lvsl 10,0,6 + lwz 9,240(6) + + lvsr 6,0,11 + lvx 5,0,3 + addi 3,3,15 + + + lvsr 8,0,4 + vspltisb 9,-1 + lvx 7,0,4 + vperm 9,0,9,8 + + + srwi 9,9,1 + li 10,16 + subi 9,9,1 + beq .Lcbc_dec + +.Lcbc_enc: + vor 2,5,5 + lvx 5,0,3 + addi 3,3,16 + mtctr 9 + subi 5,5,16 + + lvx 0,0,6 + vperm 2,2,5,6 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + vxor 2,2,0 + lvx 0,10,6 + addi 10,10,16 + vxor 2,2,4 + +.Loop_cbc_enc: + vperm 1,1,0,10 + .long 0x10420D08 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + .long 0x10420508 + lvx 0,10,6 + addi 10,10,16 + bdnz .Loop_cbc_enc + + vperm 1,1,0,10 + .long 0x10420D08 + lvx 1,10,6 + li 10,16 + vperm 0,0,1,10 + .long 0x10820509 + cmpldi 5,16 + + vperm 3,4,4,8 + vsel 2,7,3,9 + vor 7,3,3 + stvx 2,0,4 + addi 4,4,16 + bge .Lcbc_enc + + b .Lcbc_done + +.align 4 +.Lcbc_dec: + cmpldi 5,128 + bge _aesp8_cbc_decrypt8x + vor 3,5,5 + lvx 5,0,3 + addi 3,3,16 + mtctr 9 + subi 5,5,16 + + lvx 0,0,6 + vperm 3,3,5,6 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + vxor 2,3,0 + lvx 0,10,6 + addi 10,10,16 + +.Loop_cbc_dec: + vperm 1,1,0,10 + .long 0x10420D48 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + .long 0x10420548 + lvx 0,10,6 + addi 10,10,16 + bdnz .Loop_cbc_dec + + vperm 1,1,0,10 + .long 0x10420D48 + lvx 1,10,6 + li 10,16 + vperm 0,0,1,10 + .long 0x10420549 + cmpldi 5,16 + + vxor 2,2,4 + vor 4,3,3 + vperm 3,2,2,8 + vsel 2,7,3,9 + vor 7,3,3 + stvx 2,0,4 + addi 4,4,16 + bge .Lcbc_dec + +.Lcbc_done: + addi 4,4,-1 + lvx 2,0,4 + vsel 2,7,2,9 + stvx 2,0,4 + + neg 8,7 + li 10,15 + vxor 0,0,0 + vspltisb 9,-1 + + lvsl 8,0,8 + vperm 9,0,9,8 + + lvx 7,0,7 + vperm 4,4,4,8 + vsel 2,7,4,9 + lvx 5,10,7 + stvx 2,0,7 + vsel 2,4,5,9 + stvx 2,10,7 + + mtspr 256,12 + blr +.long 0 +.byte 0,12,0x14,0,0,0,6,0 +.long 0 +.align 5 +_aesp8_cbc_decrypt8x: + stdu 1,-448(1) + li 10,207 + li 11,223 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + li 0,-1 + stw 12,396(1) + li 8,0x10 + std 26,400(1) + li 26,0x20 + std 27,408(1) + li 27,0x30 + std 28,416(1) + li 28,0x40 + std 29,424(1) + li 29,0x50 + std 30,432(1) + li 30,0x60 + std 31,440(1) + li 31,0x70 + mtspr 256,0 + + subi 9,9,3 + subi 5,5,128 + + lvx 23,0,6 + lvx 30,8,6 + addi 6,6,0x20 + lvx 31,0,6 + vperm 23,23,30,10 + addi 11,1,64+15 + mtctr 9 + +.Load_cbc_dec_key: + vperm 24,30,31,10 + lvx 30,8,6 + addi 6,6,0x20 + stvx 24,0,11 + vperm 25,31,30,10 + lvx 31,0,6 + stvx 25,8,11 + addi 11,11,0x20 + bdnz .Load_cbc_dec_key + + lvx 26,8,6 + vperm 24,30,31,10 + lvx 27,26,6 + stvx 24,0,11 + vperm 25,31,26,10 + lvx 28,27,6 + stvx 25,8,11 + addi 11,1,64+15 + vperm 26,26,27,10 + lvx 29,28,6 + vperm 27,27,28,10 + lvx 30,29,6 + vperm 28,28,29,10 + lvx 31,30,6 + vperm 29,29,30,10 + lvx 14,31,6 + vperm 30,30,31,10 + lvx 24,0,11 + vperm 31,31,14,10 + lvx 25,8,11 + + + + subi 3,3,15 + + + .long 0x7C001E99 + + + .long 0x7C281E99 + + .long 0x7C5A1E99 + + .long 0x7C7B1E99 + + .long 0x7D5C1E99 + + vxor 14,0,23 + .long 0x7D7D1E99 + + vxor 15,1,23 + .long 0x7D9E1E99 + + vxor 16,2,23 + .long 0x7DBF1E99 + addi 3,3,0x80 + + vxor 17,3,23 + + vxor 18,10,23 + + vxor 19,11,23 + vxor 20,12,23 + vxor 21,13,23 + + mtctr 9 + b .Loop_cbc_dec8x +.align 5 +.Loop_cbc_dec8x: + .long 0x11CEC548 + .long 0x11EFC548 + .long 0x1210C548 + .long 0x1231C548 + .long 0x1252C548 + .long 0x1273C548 + .long 0x1294C548 + .long 0x12B5C548 + lvx 24,26,11 + addi 11,11,0x20 + + .long 0x11CECD48 + .long 0x11EFCD48 + .long 0x1210CD48 + .long 0x1231CD48 + .long 0x1252CD48 + .long 0x1273CD48 + .long 0x1294CD48 + .long 0x12B5CD48 + lvx 25,8,11 + bdnz .Loop_cbc_dec8x + + subic 5,5,128 + .long 0x11CEC548 + .long 0x11EFC548 + .long 0x1210C548 + .long 0x1231C548 + .long 0x1252C548 + .long 0x1273C548 + .long 0x1294C548 + .long 0x12B5C548 + + subfe. 0,0,0 + .long 0x11CECD48 + .long 0x11EFCD48 + .long 0x1210CD48 + .long 0x1231CD48 + .long 0x1252CD48 + .long 0x1273CD48 + .long 0x1294CD48 + .long 0x12B5CD48 + + and 0,0,5 + .long 0x11CED548 + .long 0x11EFD548 + .long 0x1210D548 + .long 0x1231D548 + .long 0x1252D548 + .long 0x1273D548 + .long 0x1294D548 + .long 0x12B5D548 + + add 3,3,0 + + + + .long 0x11CEDD48 + .long 0x11EFDD48 + .long 0x1210DD48 + .long 0x1231DD48 + .long 0x1252DD48 + .long 0x1273DD48 + .long 0x1294DD48 + .long 0x12B5DD48 + + addi 11,1,64+15 + .long 0x11CEE548 + .long 0x11EFE548 + .long 0x1210E548 + .long 0x1231E548 + .long 0x1252E548 + .long 0x1273E548 + .long 0x1294E548 + .long 0x12B5E548 + lvx 24,0,11 + + .long 0x11CEED48 + .long 0x11EFED48 + .long 0x1210ED48 + .long 0x1231ED48 + .long 0x1252ED48 + .long 0x1273ED48 + .long 0x1294ED48 + .long 0x12B5ED48 + lvx 25,8,11 + + .long 0x11CEF548 + vxor 4,4,31 + .long 0x11EFF548 + vxor 0,0,31 + .long 0x1210F548 + vxor 1,1,31 + .long 0x1231F548 + vxor 2,2,31 + .long 0x1252F548 + vxor 3,3,31 + .long 0x1273F548 + vxor 10,10,31 + .long 0x1294F548 + vxor 11,11,31 + .long 0x12B5F548 + vxor 12,12,31 + + .long 0x11CE2549 + .long 0x11EF0549 + .long 0x7C001E99 + .long 0x12100D49 + .long 0x7C281E99 + .long 0x12311549 + + .long 0x7C5A1E99 + .long 0x12521D49 + + .long 0x7C7B1E99 + .long 0x12735549 + + .long 0x7D5C1E99 + .long 0x12945D49 + + .long 0x7D7D1E99 + .long 0x12B56549 + + .long 0x7D9E1E99 + vor 4,13,13 + + .long 0x7DBF1E99 + addi 3,3,0x80 + + + + .long 0x7DC02799 + + vxor 14,0,23 + + .long 0x7DE82799 + + vxor 15,1,23 + + .long 0x7E1A2799 + vxor 16,2,23 + + .long 0x7E3B2799 + vxor 17,3,23 + + .long 0x7E5C2799 + vxor 18,10,23 + + .long 0x7E7D2799 + vxor 19,11,23 + + .long 0x7E9E2799 + vxor 20,12,23 + .long 0x7EBF2799 + addi 4,4,0x80 + vxor 21,13,23 + + mtctr 9 + beq .Loop_cbc_dec8x + + addic. 5,5,128 + beq .Lcbc_dec8x_done + nop + nop + +.Loop_cbc_dec8x_tail: + .long 0x11EFC548 + .long 0x1210C548 + .long 0x1231C548 + .long 0x1252C548 + .long 0x1273C548 + .long 0x1294C548 + .long 0x12B5C548 + lvx 24,26,11 + addi 11,11,0x20 + + .long 0x11EFCD48 + .long 0x1210CD48 + .long 0x1231CD48 + .long 0x1252CD48 + .long 0x1273CD48 + .long 0x1294CD48 + .long 0x12B5CD48 + lvx 25,8,11 + bdnz .Loop_cbc_dec8x_tail + + .long 0x11EFC548 + .long 0x1210C548 + .long 0x1231C548 + .long 0x1252C548 + .long 0x1273C548 + .long 0x1294C548 + .long 0x12B5C548 + + .long 0x11EFCD48 + .long 0x1210CD48 + .long 0x1231CD48 + .long 0x1252CD48 + .long 0x1273CD48 + .long 0x1294CD48 + .long 0x12B5CD48 + + .long 0x11EFD548 + .long 0x1210D548 + .long 0x1231D548 + .long 0x1252D548 + .long 0x1273D548 + .long 0x1294D548 + .long 0x12B5D548 + + .long 0x11EFDD48 + .long 0x1210DD48 + .long 0x1231DD48 + .long 0x1252DD48 + .long 0x1273DD48 + .long 0x1294DD48 + .long 0x12B5DD48 + + .long 0x11EFE548 + .long 0x1210E548 + .long 0x1231E548 + .long 0x1252E548 + .long 0x1273E548 + .long 0x1294E548 + .long 0x12B5E548 + + .long 0x11EFED48 + .long 0x1210ED48 + .long 0x1231ED48 + .long 0x1252ED48 + .long 0x1273ED48 + .long 0x1294ED48 + .long 0x12B5ED48 + + .long 0x11EFF548 + vxor 4,4,31 + .long 0x1210F548 + vxor 1,1,31 + .long 0x1231F548 + vxor 2,2,31 + .long 0x1252F548 + vxor 3,3,31 + .long 0x1273F548 + vxor 10,10,31 + .long 0x1294F548 + vxor 11,11,31 + .long 0x12B5F548 + vxor 12,12,31 + + cmplwi 5,32 + blt .Lcbc_dec8x_one + nop + beq .Lcbc_dec8x_two + cmplwi 5,64 + blt .Lcbc_dec8x_three + nop + beq .Lcbc_dec8x_four + cmplwi 5,96 + blt .Lcbc_dec8x_five + nop + beq .Lcbc_dec8x_six + +.Lcbc_dec8x_seven: + .long 0x11EF2549 + .long 0x12100D49 + .long 0x12311549 + .long 0x12521D49 + .long 0x12735549 + .long 0x12945D49 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + + .long 0x7E5B2799 + + .long 0x7E7C2799 + + .long 0x7E9D2799 + .long 0x7EBE2799 + addi 4,4,0x70 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_six: + .long 0x12102549 + .long 0x12311549 + .long 0x12521D49 + .long 0x12735549 + .long 0x12945D49 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7E002799 + + .long 0x7E282799 + + .long 0x7E5A2799 + + .long 0x7E7B2799 + + .long 0x7E9C2799 + .long 0x7EBD2799 + addi 4,4,0x60 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_five: + .long 0x12312549 + .long 0x12521D49 + .long 0x12735549 + .long 0x12945D49 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7E202799 + + .long 0x7E482799 + + .long 0x7E7A2799 + + .long 0x7E9B2799 + .long 0x7EBC2799 + addi 4,4,0x50 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_four: + .long 0x12522549 + .long 0x12735549 + .long 0x12945D49 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7E402799 + + .long 0x7E682799 + + .long 0x7E9A2799 + .long 0x7EBB2799 + addi 4,4,0x40 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_three: + .long 0x12732549 + .long 0x12945D49 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7E602799 + + .long 0x7E882799 + .long 0x7EBA2799 + addi 4,4,0x30 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_two: + .long 0x12942549 + .long 0x12B56549 + vor 4,13,13 + + + + .long 0x7E802799 + .long 0x7EA82799 + addi 4,4,0x20 + b .Lcbc_dec8x_done + +.align 5 +.Lcbc_dec8x_one: + .long 0x12B52549 + vor 4,13,13 + + + .long 0x7EA02799 + addi 4,4,0x10 + +.Lcbc_dec8x_done: + + .long 0x7C803F99 + + li 10,79 + li 11,95 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + + mtspr 256,12 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + ld 26,400(1) + ld 27,408(1) + ld 28,416(1) + ld 29,424(1) + ld 30,432(1) + ld 31,440(1) + addi 1,1,448 + blr +.long 0 +.byte 0,12,0x04,0,0x80,6,6,0 +.long 0 +.size .aes_p8_cbc_encrypt,.-.aes_p8_cbc_encrypt +.size aes_p8_cbc_encrypt,.-.aes_p8_cbc_encrypt +.globl aes_p8_ctr32_encrypt_blocks +.type aes_p8_ctr32_encrypt_blocks,@function +.section ".opd","aw" +.align 3 +aes_p8_ctr32_encrypt_blocks: +.quad .aes_p8_ctr32_encrypt_blocks,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_ctr32_encrypt_blocks: + cmpldi 5,1 + .long 0x4dc00020 + + lis 0,0xfff0 + mfspr 12,256 + mtspr 256,0 + + li 10,15 + vxor 0,0,0 + + + lvx 4,0,7 + lvsl 6,0,7 + lvx 5,10,7 + vspltisb 11,1 + + vperm 4,4,5,6 + vsldoi 11,0,11,1 + + neg 11,3 + lvsl 10,0,6 + lwz 9,240(6) + + lvsr 6,0,11 + lvx 5,0,3 + addi 3,3,15 + + + srwi 9,9,1 + li 10,16 + subi 9,9,1 + + cmpldi 5,8 + bge _aesp8_ctr32_encrypt8x + + lvsr 8,0,4 + vspltisb 9,-1 + lvx 7,0,4 + vperm 9,0,9,8 + + + lvx 0,0,6 + mtctr 9 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + vxor 2,4,0 + lvx 0,10,6 + addi 10,10,16 + b .Loop_ctr32_enc + +.align 5 +.Loop_ctr32_enc: + vperm 1,1,0,10 + .long 0x10420D08 + lvx 1,10,6 + addi 10,10,16 + vperm 0,0,1,10 + .long 0x10420508 + lvx 0,10,6 + addi 10,10,16 + bdnz .Loop_ctr32_enc + + vadduwm 4,4,11 + vor 3,5,5 + lvx 5,0,3 + addi 3,3,16 + subic. 5,5,1 + + vperm 1,1,0,10 + .long 0x10420D08 + lvx 1,10,6 + vperm 3,3,5,6 + li 10,16 + vperm 1,0,1,10 + lvx 0,0,6 + vxor 3,3,1 + .long 0x10421D09 + + lvx 1,10,6 + addi 10,10,16 + vperm 2,2,2,8 + vsel 3,7,2,9 + mtctr 9 + vperm 0,0,1,10 + vor 7,2,2 + vxor 2,4,0 + lvx 0,10,6 + addi 10,10,16 + stvx 3,0,4 + addi 4,4,16 + bne .Loop_ctr32_enc + + addi 4,4,-1 + lvx 2,0,4 + vsel 2,7,2,9 + stvx 2,0,4 + + mtspr 256,12 + blr +.long 0 +.byte 0,12,0x14,0,0,0,6,0 +.long 0 +.align 5 +_aesp8_ctr32_encrypt8x: + stdu 1,-448(1) + li 10,207 + li 11,223 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + li 0,-1 + stw 12,396(1) + li 8,0x10 + std 26,400(1) + li 26,0x20 + std 27,408(1) + li 27,0x30 + std 28,416(1) + li 28,0x40 + std 29,424(1) + li 29,0x50 + std 30,432(1) + li 30,0x60 + std 31,440(1) + li 31,0x70 + mtspr 256,0 + + subi 9,9,3 + + lvx 23,0,6 + lvx 30,8,6 + addi 6,6,0x20 + lvx 31,0,6 + vperm 23,23,30,10 + addi 11,1,64+15 + mtctr 9 + +.Load_ctr32_enc_key: + vperm 24,30,31,10 + lvx 30,8,6 + addi 6,6,0x20 + stvx 24,0,11 + vperm 25,31,30,10 + lvx 31,0,6 + stvx 25,8,11 + addi 11,11,0x20 + bdnz .Load_ctr32_enc_key + + lvx 26,8,6 + vperm 24,30,31,10 + lvx 27,26,6 + stvx 24,0,11 + vperm 25,31,26,10 + lvx 28,27,6 + stvx 25,8,11 + addi 11,1,64+15 + vperm 26,26,27,10 + lvx 29,28,6 + vperm 27,27,28,10 + lvx 30,29,6 + vperm 28,28,29,10 + lvx 31,30,6 + vperm 29,29,30,10 + lvx 15,31,6 + vperm 30,30,31,10 + lvx 24,0,11 + vperm 31,31,15,10 + lvx 25,8,11 + + vadduwm 7,11,11 + subi 3,3,15 + sldi 5,5,4 + + vadduwm 16,4,11 + vadduwm 17,4,7 + vxor 15,4,23 + + vadduwm 18,16,7 + vxor 16,16,23 + + vadduwm 19,17,7 + vxor 17,17,23 + + vadduwm 20,18,7 + vxor 18,18,23 + + vadduwm 21,19,7 + vxor 19,19,23 + vadduwm 22,20,7 + vxor 20,20,23 + vadduwm 4,21,7 + vxor 21,21,23 + vxor 22,22,23 + + mtctr 9 + b .Loop_ctr32_enc8x +.align 5 +.Loop_ctr32_enc8x: + .long 0x11EFC508 + .long 0x1210C508 + .long 0x1231C508 + .long 0x1252C508 + .long 0x1273C508 + .long 0x1294C508 + .long 0x12B5C508 + .long 0x12D6C508 +.Loop_ctr32_enc8x_middle: + lvx 24,26,11 + addi 11,11,0x20 + + .long 0x11EFCD08 + .long 0x1210CD08 + .long 0x1231CD08 + .long 0x1252CD08 + .long 0x1273CD08 + .long 0x1294CD08 + .long 0x12B5CD08 + .long 0x12D6CD08 + lvx 25,8,11 + bdnz .Loop_ctr32_enc8x + + subic 11,5,256 + .long 0x11EFC508 + .long 0x1210C508 + .long 0x1231C508 + .long 0x1252C508 + .long 0x1273C508 + .long 0x1294C508 + .long 0x12B5C508 + .long 0x12D6C508 + + subfe 0,0,0 + .long 0x11EFCD08 + .long 0x1210CD08 + .long 0x1231CD08 + .long 0x1252CD08 + .long 0x1273CD08 + .long 0x1294CD08 + .long 0x12B5CD08 + .long 0x12D6CD08 + + and 0,0,11 + addi 11,1,64+15 + .long 0x11EFD508 + .long 0x1210D508 + .long 0x1231D508 + .long 0x1252D508 + .long 0x1273D508 + .long 0x1294D508 + .long 0x12B5D508 + .long 0x12D6D508 + lvx 24,0,11 + + subic 5,5,129 + .long 0x11EFDD08 + addi 5,5,1 + .long 0x1210DD08 + .long 0x1231DD08 + .long 0x1252DD08 + .long 0x1273DD08 + .long 0x1294DD08 + .long 0x12B5DD08 + .long 0x12D6DD08 + lvx 25,8,11 + + .long 0x11EFE508 + .long 0x7C001E99 + .long 0x1210E508 + .long 0x7C281E99 + .long 0x1231E508 + .long 0x7C5A1E99 + .long 0x1252E508 + .long 0x7C7B1E99 + .long 0x1273E508 + .long 0x7D5C1E99 + .long 0x1294E508 + .long 0x7D9D1E99 + .long 0x12B5E508 + .long 0x7DBE1E99 + .long 0x12D6E508 + .long 0x7DDF1E99 + addi 3,3,0x80 + + .long 0x11EFED08 + + .long 0x1210ED08 + + .long 0x1231ED08 + + .long 0x1252ED08 + + .long 0x1273ED08 + + .long 0x1294ED08 + + .long 0x12B5ED08 + + .long 0x12D6ED08 + + + add 3,3,0 + + + + subfe. 0,0,0 + .long 0x11EFF508 + vxor 0,0,31 + .long 0x1210F508 + vxor 1,1,31 + .long 0x1231F508 + vxor 2,2,31 + .long 0x1252F508 + vxor 3,3,31 + .long 0x1273F508 + vxor 10,10,31 + .long 0x1294F508 + vxor 12,12,31 + .long 0x12B5F508 + vxor 13,13,31 + .long 0x12D6F508 + vxor 14,14,31 + + bne .Lctr32_enc8x_break + + .long 0x100F0509 + .long 0x10300D09 + vadduwm 16,4,11 + .long 0x10511509 + vadduwm 17,4,7 + vxor 15,4,23 + .long 0x10721D09 + vadduwm 18,16,7 + vxor 16,16,23 + .long 0x11535509 + vadduwm 19,17,7 + vxor 17,17,23 + .long 0x11946509 + vadduwm 20,18,7 + vxor 18,18,23 + .long 0x11B56D09 + vadduwm 21,19,7 + vxor 19,19,23 + .long 0x11D67509 + vadduwm 22,20,7 + vxor 20,20,23 + + vadduwm 4,21,7 + vxor 21,21,23 + + vxor 22,22,23 + mtctr 9 + + .long 0x11EFC508 + .long 0x7C002799 + + .long 0x1210C508 + .long 0x7C282799 + + .long 0x1231C508 + .long 0x7C5A2799 + + .long 0x1252C508 + .long 0x7C7B2799 + + .long 0x1273C508 + .long 0x7D5C2799 + + .long 0x1294C508 + .long 0x7D9D2799 + + .long 0x12B5C508 + .long 0x7DBE2799 + .long 0x12D6C508 + .long 0x7DDF2799 + addi 4,4,0x80 + + b .Loop_ctr32_enc8x_middle + +.align 5 +.Lctr32_enc8x_break: + cmpwi 5,-0x60 + blt .Lctr32_enc8x_one + nop + beq .Lctr32_enc8x_two + cmpwi 5,-0x40 + blt .Lctr32_enc8x_three + nop + beq .Lctr32_enc8x_four + cmpwi 5,-0x20 + blt .Lctr32_enc8x_five + nop + beq .Lctr32_enc8x_six + cmpwi 5,0x00 + blt .Lctr32_enc8x_seven + +.Lctr32_enc8x_eight: + .long 0x11EF0509 + .long 0x12100D09 + .long 0x12311509 + .long 0x12521D09 + .long 0x12735509 + .long 0x12946509 + .long 0x12B56D09 + .long 0x12D67509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + + .long 0x7E5B2799 + + .long 0x7E7C2799 + + .long 0x7E9D2799 + + .long 0x7EBE2799 + .long 0x7EDF2799 + addi 4,4,0x80 + b .Lctr32_enc8x_done + +.align 5 +.Lctr32_enc8x_seven: + .long 0x11EF0D09 + .long 0x12101509 + .long 0x12311D09 + .long 0x12525509 + .long 0x12736509 + .long 0x12946D09 + .long 0x12B57509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + + .long 0x7E5B2799 + + .long 0x7E7C2799 + + .long 0x7E9D2799 + .long 0x7EBE2799 + addi 4,4,0x70 + b .Lctr32_enc8x_done + +.align 5 +.Lctr32_enc8x_six: + .long 0x11EF1509 + .long 0x12101D09 + .long 0x12315509 + .long 0x12526509 + .long 0x12736D09 + .long 0x12947509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + + .long 0x7E5B2799 + + .long 0x7E7C2799 + .long 0x7E9D2799 + addi 4,4,0x60 + b .Lctr32_enc8x_done + +.align 5 +.Lctr32_enc8x_five: + .long 0x11EF1D09 + .long 0x12105509 + .long 0x12316509 + .long 0x12526D09 + .long 0x12737509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + + .long 0x7E5B2799 + .long 0x7E7C2799 + addi 4,4,0x50 + b .Lctr32_enc8x_done + +.align 5 +.Lctr32_enc8x_four: + .long 0x11EF5509 + .long 0x12106509 + .long 0x12316D09 + .long 0x12527509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + + .long 0x7E3A2799 + .long 0x7E5B2799 + addi 4,4,0x40 + b .Lctr32_enc8x_done + +.align 5 +.Lctr32_enc8x_three: + .long 0x11EF6509 + .long 0x12106D09 + .long 0x12317509 + + + + .long 0x7DE02799 + + .long 0x7E082799 + .long 0x7E3A2799 + addi 4,4,0x30 + b .Lcbc_dec8x_done + +.align 5 +.Lctr32_enc8x_two: + .long 0x11EF6D09 + .long 0x12107509 + + + + .long 0x7DE02799 + .long 0x7E082799 + addi 4,4,0x20 + b .Lcbc_dec8x_done + +.align 5 +.Lctr32_enc8x_one: + .long 0x11EF7509 + + + .long 0x7DE02799 + addi 4,4,0x10 + +.Lctr32_enc8x_done: + li 10,79 + li 11,95 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + stvx 6,10,1 + addi 10,10,32 + stvx 6,11,1 + addi 11,11,32 + + mtspr 256,12 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + ld 26,400(1) + ld 27,408(1) + ld 28,416(1) + ld 29,424(1) + ld 30,432(1) + ld 31,440(1) + addi 1,1,448 + blr +.long 0 +.byte 0,12,0x04,0,0x80,6,6,0 +.long 0 +.size .aes_p8_ctr32_encrypt_blocks,.-.aes_p8_ctr32_encrypt_blocks +.size aes_p8_ctr32_encrypt_blocks,.-.aes_p8_ctr32_encrypt_blocks +.globl aes_p8_xts_encrypt +.type aes_p8_xts_encrypt,@function +.section ".opd","aw" +.align 3 +aes_p8_xts_encrypt: +.quad .aes_p8_xts_encrypt,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_xts_encrypt: + mr 10,3 + li 3,-1 + cmpldi 5,16 + .long 0x4dc00020 + + lis 0,0xfff0 + mfspr 12,256 + li 11,0 + mtspr 256,0 + + vspltisb 9,0x07 + + + + + li 3,15 + lvx 8,0,8 + lvsl 5,0,8 + lvx 4,3,8 + + vperm 8,8,4,5 + + neg 11,10 + lvsr 5,0,11 + lvx 2,0,10 + addi 10,10,15 + + + cmpldi 7,0 + beq .Lxts_enc_no_key2 + + lvsl 7,0,7 + lwz 9,240(7) + srwi 9,9,1 + subi 9,9,1 + li 3,16 + + lvx 0,0,7 + lvx 1,3,7 + addi 3,3,16 + vperm 0,0,1,7 + vxor 8,8,0 + lvx 0,3,7 + addi 3,3,16 + mtctr 9 + +.Ltweak_xts_enc: + vperm 1,1,0,7 + .long 0x11080D08 + lvx 1,3,7 + addi 3,3,16 + vperm 0,0,1,7 + .long 0x11080508 + lvx 0,3,7 + addi 3,3,16 + bdnz .Ltweak_xts_enc + + vperm 1,1,0,7 + .long 0x11080D08 + lvx 1,3,7 + vperm 0,0,1,7 + .long 0x11080509 + + li 8,0 + b .Lxts_enc + +.Lxts_enc_no_key2: + li 3,-16 + and 5,5,3 + + +.Lxts_enc: + lvx 4,0,10 + addi 10,10,16 + + lvsl 7,0,6 + lwz 9,240(6) + srwi 9,9,1 + subi 9,9,1 + li 3,16 + + vslb 10,9,9 + vor 10,10,9 + vspltisb 11,1 + vsldoi 10,10,11,15 + + cmpldi 5,96 + bge _aesp8_xts_encrypt6x + + andi. 7,5,15 + subic 0,5,32 + subi 7,7,16 + subfe 0,0,0 + and 0,0,7 + add 10,10,0 + + lvx 0,0,6 + lvx 1,3,6 + addi 3,3,16 + vperm 2,2,4,5 + vperm 0,0,1,7 + vxor 2,2,8 + vxor 2,2,0 + lvx 0,3,6 + addi 3,3,16 + mtctr 9 + b .Loop_xts_enc + +.align 5 +.Loop_xts_enc: + vperm 1,1,0,7 + .long 0x10420D08 + lvx 1,3,6 + addi 3,3,16 + vperm 0,0,1,7 + .long 0x10420508 + lvx 0,3,6 + addi 3,3,16 + bdnz .Loop_xts_enc + + vperm 1,1,0,7 + .long 0x10420D08 + lvx 1,3,6 + li 3,16 + vperm 0,0,1,7 + vxor 0,0,8 + .long 0x10620509 + + + nop + + .long 0x7C602799 + addi 4,4,16 + + subic. 5,5,16 + beq .Lxts_enc_done + + vor 2,4,4 + lvx 4,0,10 + addi 10,10,16 + lvx 0,0,6 + lvx 1,3,6 + addi 3,3,16 + + subic 0,5,32 + subfe 0,0,0 + and 0,0,7 + add 10,10,0 + + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 8,8,11 + + vperm 2,2,4,5 + vperm 0,0,1,7 + vxor 2,2,8 + vxor 3,3,0 + vxor 2,2,0 + lvx 0,3,6 + addi 3,3,16 + + mtctr 9 + cmpldi 5,16 + bge .Loop_xts_enc + + vxor 3,3,8 + lvsr 5,0,5 + vxor 4,4,4 + vspltisb 11,-1 + vperm 4,4,11,5 + vsel 2,2,3,4 + + subi 11,4,17 + subi 4,4,16 + mtctr 5 + li 5,16 +.Loop_xts_enc_steal: + lbzu 0,1(11) + stb 0,16(11) + bdnz .Loop_xts_enc_steal + + mtctr 9 + b .Loop_xts_enc + +.Lxts_enc_done: + cmpldi 8,0 + beq .Lxts_enc_ret + + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 8,8,11 + + + .long 0x7D004799 + +.Lxts_enc_ret: + mtspr 256,12 + li 3,0 + blr +.long 0 +.byte 0,12,0x04,0,0x80,6,6,0 +.long 0 +.size .aes_p8_xts_encrypt,.-.aes_p8_xts_encrypt +.size aes_p8_xts_encrypt,.-.aes_p8_xts_encrypt + +.globl aes_p8_xts_decrypt +.type aes_p8_xts_decrypt,@function +.section ".opd","aw" +.align 3 +aes_p8_xts_decrypt: +.quad .aes_p8_xts_decrypt,.TOC.@tocbase,0 +.previous +.align 5 +.aes_p8_xts_decrypt: + mr 10,3 + li 3,-1 + cmpldi 5,16 + .long 0x4dc00020 + + lis 0,0xfff8 + mfspr 12,256 + li 11,0 + mtspr 256,0 + + andi. 0,5,15 + neg 0,0 + andi. 0,0,16 + sub 5,5,0 + + vspltisb 9,0x07 + + + + + li 3,15 + lvx 8,0,8 + lvsl 5,0,8 + lvx 4,3,8 + + vperm 8,8,4,5 + + neg 11,10 + lvsr 5,0,11 + lvx 2,0,10 + addi 10,10,15 + + + cmpldi 7,0 + beq .Lxts_dec_no_key2 + + lvsl 7,0,7 + lwz 9,240(7) + srwi 9,9,1 + subi 9,9,1 + li 3,16 + + lvx 0,0,7 + lvx 1,3,7 + addi 3,3,16 + vperm 0,0,1,7 + vxor 8,8,0 + lvx 0,3,7 + addi 3,3,16 + mtctr 9 + +.Ltweak_xts_dec: + vperm 1,1,0,7 + .long 0x11080D08 + lvx 1,3,7 + addi 3,3,16 + vperm 0,0,1,7 + .long 0x11080508 + lvx 0,3,7 + addi 3,3,16 + bdnz .Ltweak_xts_dec + + vperm 1,1,0,7 + .long 0x11080D08 + lvx 1,3,7 + vperm 0,0,1,7 + .long 0x11080509 + + li 8,0 + b .Lxts_dec + +.Lxts_dec_no_key2: + neg 3,5 + andi. 3,3,15 + add 5,5,3 + + +.Lxts_dec: + lvx 4,0,10 + addi 10,10,16 + + lvsl 7,0,6 + lwz 9,240(6) + srwi 9,9,1 + subi 9,9,1 + li 3,16 + + vslb 10,9,9 + vor 10,10,9 + vspltisb 11,1 + vsldoi 10,10,11,15 + + cmpldi 5,96 + bge _aesp8_xts_decrypt6x + + lvx 0,0,6 + lvx 1,3,6 + addi 3,3,16 + vperm 2,2,4,5 + vperm 0,0,1,7 + vxor 2,2,8 + vxor 2,2,0 + lvx 0,3,6 + addi 3,3,16 + mtctr 9 + + cmpldi 5,16 + blt .Ltail_xts_dec + b .Loop_xts_dec + +.align 5 +.Loop_xts_dec: + vperm 1,1,0,7 + .long 0x10420D48 + lvx 1,3,6 + addi 3,3,16 + vperm 0,0,1,7 + .long 0x10420548 + lvx 0,3,6 + addi 3,3,16 + bdnz .Loop_xts_dec + + vperm 1,1,0,7 + .long 0x10420D48 + lvx 1,3,6 + li 3,16 + vperm 0,0,1,7 + vxor 0,0,8 + .long 0x10620549 + + + nop + + .long 0x7C602799 + addi 4,4,16 + + subic. 5,5,16 + beq .Lxts_dec_done + + vor 2,4,4 + lvx 4,0,10 + addi 10,10,16 + lvx 0,0,6 + lvx 1,3,6 + addi 3,3,16 + + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 8,8,11 + + vperm 2,2,4,5 + vperm 0,0,1,7 + vxor 2,2,8 + vxor 2,2,0 + lvx 0,3,6 + addi 3,3,16 + + mtctr 9 + cmpldi 5,16 + bge .Loop_xts_dec + +.Ltail_xts_dec: + vsrab 11,8,9 + vaddubm 12,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 12,12,11 + + subi 10,10,16 + add 10,10,5 + + vxor 2,2,8 + vxor 2,2,12 + +.Loop_xts_dec_short: + vperm 1,1,0,7 + .long 0x10420D48 + lvx 1,3,6 + addi 3,3,16 + vperm 0,0,1,7 + .long 0x10420548 + lvx 0,3,6 + addi 3,3,16 + bdnz .Loop_xts_dec_short + + vperm 1,1,0,7 + .long 0x10420D48 + lvx 1,3,6 + li 3,16 + vperm 0,0,1,7 + vxor 0,0,12 + .long 0x10620549 + + + nop + + .long 0x7C602799 + + vor 2,4,4 + lvx 4,0,10 + + lvx 0,0,6 + lvx 1,3,6 + addi 3,3,16 + vperm 2,2,4,5 + vperm 0,0,1,7 + + lvsr 5,0,5 + vxor 4,4,4 + vspltisb 11,-1 + vperm 4,4,11,5 + vsel 2,2,3,4 + + vxor 0,0,8 + vxor 2,2,0 + lvx 0,3,6 + addi 3,3,16 + + subi 11,4,1 + mtctr 5 + li 5,16 +.Loop_xts_dec_steal: + lbzu 0,1(11) + stb 0,16(11) + bdnz .Loop_xts_dec_steal + + mtctr 9 + b .Loop_xts_dec + +.Lxts_dec_done: + cmpldi 8,0 + beq .Lxts_dec_ret + + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 8,8,11 + + + .long 0x7D004799 + +.Lxts_dec_ret: + mtspr 256,12 + li 3,0 + blr +.long 0 +.byte 0,12,0x04,0,0x80,6,6,0 +.long 0 +.size .aes_p8_xts_decrypt,.-.aes_p8_xts_decrypt +.size aes_p8_xts_decrypt,.-.aes_p8_xts_decrypt +.align 5 +_aesp8_xts_encrypt6x: + stdu 1,-448(1) + mflr 11 + li 7,207 + li 3,223 + std 11,464(1) + stvx 20,7,1 + addi 7,7,32 + stvx 21,3,1 + addi 3,3,32 + stvx 22,7,1 + addi 7,7,32 + stvx 23,3,1 + addi 3,3,32 + stvx 24,7,1 + addi 7,7,32 + stvx 25,3,1 + addi 3,3,32 + stvx 26,7,1 + addi 7,7,32 + stvx 27,3,1 + addi 3,3,32 + stvx 28,7,1 + addi 7,7,32 + stvx 29,3,1 + addi 3,3,32 + stvx 30,7,1 + stvx 31,3,1 + li 0,-1 + stw 12,396(1) + li 3,0x10 + std 26,400(1) + li 26,0x20 + std 27,408(1) + li 27,0x30 + std 28,416(1) + li 28,0x40 + std 29,424(1) + li 29,0x50 + std 30,432(1) + li 30,0x60 + std 31,440(1) + li 31,0x70 + mtspr 256,0 + + subi 9,9,3 + + lvx 23,0,6 + lvx 30,3,6 + addi 6,6,0x20 + lvx 31,0,6 + vperm 23,23,30,7 + addi 7,1,64+15 + mtctr 9 + +.Load_xts_enc_key: + vperm 24,30,31,7 + lvx 30,3,6 + addi 6,6,0x20 + stvx 24,0,7 + vperm 25,31,30,7 + lvx 31,0,6 + stvx 25,3,7 + addi 7,7,0x20 + bdnz .Load_xts_enc_key + + lvx 26,3,6 + vperm 24,30,31,7 + lvx 27,26,6 + stvx 24,0,7 + vperm 25,31,26,7 + lvx 28,27,6 + stvx 25,3,7 + addi 7,1,64+15 + vperm 26,26,27,7 + lvx 29,28,6 + vperm 27,27,28,7 + lvx 30,29,6 + vperm 28,28,29,7 + lvx 31,30,6 + vperm 29,29,30,7 + lvx 22,31,6 + vperm 30,30,31,7 + lvx 24,0,7 + vperm 31,31,22,7 + lvx 25,3,7 + + vperm 0,2,4,5 + subi 10,10,31 + vxor 17,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 7,0,17 + vxor 8,8,11 + + .long 0x7C235699 + vxor 18,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 12,1,18 + vxor 8,8,11 + + .long 0x7C5A5699 + andi. 31,5,15 + vxor 19,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 13,2,19 + vxor 8,8,11 + + .long 0x7C7B5699 + sub 5,5,31 + vxor 20,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 14,3,20 + vxor 8,8,11 + + .long 0x7C9C5699 + subi 5,5,0x60 + vxor 21,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 15,4,21 + vxor 8,8,11 + + .long 0x7CBD5699 + addi 10,10,0x60 + vxor 22,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 16,5,22 + vxor 8,8,11 + + vxor 31,31,23 + mtctr 9 + b .Loop_xts_enc6x + +.align 5 +.Loop_xts_enc6x: + .long 0x10E7C508 + .long 0x118CC508 + .long 0x11ADC508 + .long 0x11CEC508 + .long 0x11EFC508 + .long 0x1210C508 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD08 + .long 0x118CCD08 + .long 0x11ADCD08 + .long 0x11CECD08 + .long 0x11EFCD08 + .long 0x1210CD08 + lvx 25,3,7 + bdnz .Loop_xts_enc6x + + subic 5,5,96 + vxor 0,17,31 + .long 0x10E7C508 + .long 0x118CC508 + vsrab 11,8,9 + vxor 17,8,23 + vaddubm 8,8,8 + .long 0x11ADC508 + .long 0x11CEC508 + vsldoi 11,11,11,15 + .long 0x11EFC508 + .long 0x1210C508 + + subfe. 0,0,0 + vand 11,11,10 + .long 0x10E7CD08 + .long 0x118CCD08 + vxor 8,8,11 + .long 0x11ADCD08 + .long 0x11CECD08 + vxor 1,18,31 + vsrab 11,8,9 + vxor 18,8,23 + .long 0x11EFCD08 + .long 0x1210CD08 + + and 0,0,5 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x10E7D508 + .long 0x118CD508 + vand 11,11,10 + .long 0x11ADD508 + .long 0x11CED508 + vxor 8,8,11 + .long 0x11EFD508 + .long 0x1210D508 + + add 10,10,0 + + + + vxor 2,19,31 + vsrab 11,8,9 + vxor 19,8,23 + vaddubm 8,8,8 + .long 0x10E7DD08 + .long 0x118CDD08 + vsldoi 11,11,11,15 + .long 0x11ADDD08 + .long 0x11CEDD08 + vand 11,11,10 + .long 0x11EFDD08 + .long 0x1210DD08 + + addi 7,1,64+15 + vxor 8,8,11 + .long 0x10E7E508 + .long 0x118CE508 + vxor 3,20,31 + vsrab 11,8,9 + vxor 20,8,23 + .long 0x11ADE508 + .long 0x11CEE508 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x11EFE508 + .long 0x1210E508 + lvx 24,0,7 + vand 11,11,10 + + .long 0x10E7ED08 + .long 0x118CED08 + vxor 8,8,11 + .long 0x11ADED08 + .long 0x11CEED08 + vxor 4,21,31 + vsrab 11,8,9 + vxor 21,8,23 + .long 0x11EFED08 + .long 0x1210ED08 + lvx 25,3,7 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + .long 0x10E7F508 + .long 0x118CF508 + vand 11,11,10 + .long 0x11ADF508 + .long 0x11CEF508 + vxor 8,8,11 + .long 0x11EFF508 + .long 0x1210F508 + vxor 5,22,31 + vsrab 11,8,9 + vxor 22,8,23 + + .long 0x10E70509 + .long 0x7C005699 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x118C0D09 + .long 0x7C235699 + .long 0x11AD1509 + + .long 0x7C5A5699 + vand 11,11,10 + .long 0x11CE1D09 + + .long 0x7C7B5699 + .long 0x11EF2509 + + .long 0x7C9C5699 + vxor 8,8,11 + .long 0x11702D09 + + + .long 0x7CBD5699 + addi 10,10,0x60 + + + + + + .long 0x7CE02799 + vxor 7,0,17 + + .long 0x7D832799 + vxor 12,1,18 + + .long 0x7DBA2799 + vxor 13,2,19 + + .long 0x7DDB2799 + vxor 14,3,20 + + .long 0x7DFC2799 + vxor 15,4,21 + + .long 0x7D7D2799 + vxor 16,5,22 + addi 4,4,0x60 + + mtctr 9 + beq .Loop_xts_enc6x + + addic. 5,5,0x60 + beq .Lxts_enc6x_zero + cmpwi 5,0x20 + blt .Lxts_enc6x_one + nop + beq .Lxts_enc6x_two + cmpwi 5,0x40 + blt .Lxts_enc6x_three + nop + beq .Lxts_enc6x_four + +.Lxts_enc6x_five: + vxor 7,1,17 + vxor 12,2,18 + vxor 13,3,19 + vxor 14,4,20 + vxor 15,5,21 + + bl _aesp8_xts_enc5x + + + vor 17,22,22 + + .long 0x7CE02799 + + .long 0x7D832799 + + .long 0x7DBA2799 + vxor 11,15,22 + + .long 0x7DDB2799 + .long 0x7DFC2799 + addi 4,4,0x50 + bne .Lxts_enc6x_steal + b .Lxts_enc6x_done + +.align 4 +.Lxts_enc6x_four: + vxor 7,2,17 + vxor 12,3,18 + vxor 13,4,19 + vxor 14,5,20 + vxor 15,15,15 + + bl _aesp8_xts_enc5x + + + vor 17,21,21 + + .long 0x7CE02799 + + .long 0x7D832799 + vxor 11,14,21 + + .long 0x7DBA2799 + .long 0x7DDB2799 + addi 4,4,0x40 + bne .Lxts_enc6x_steal + b .Lxts_enc6x_done + +.align 4 +.Lxts_enc6x_three: + vxor 7,3,17 + vxor 12,4,18 + vxor 13,5,19 + vxor 14,14,14 + vxor 15,15,15 + + bl _aesp8_xts_enc5x + + + vor 17,20,20 + + .long 0x7CE02799 + vxor 11,13,20 + + .long 0x7D832799 + .long 0x7DBA2799 + addi 4,4,0x30 + bne .Lxts_enc6x_steal + b .Lxts_enc6x_done + +.align 4 +.Lxts_enc6x_two: + vxor 7,4,17 + vxor 12,5,18 + vxor 13,13,13 + vxor 14,14,14 + vxor 15,15,15 + + bl _aesp8_xts_enc5x + + + vor 17,19,19 + vxor 11,12,19 + + .long 0x7CE02799 + .long 0x7D832799 + addi 4,4,0x20 + bne .Lxts_enc6x_steal + b .Lxts_enc6x_done + +.align 4 +.Lxts_enc6x_one: + vxor 7,5,17 + nop +.Loop_xts_enc1x: + .long 0x10E7C508 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD08 + lvx 25,3,7 + bdnz .Loop_xts_enc1x + + add 10,10,31 + cmpwi 31,0 + .long 0x10E7C508 + + subi 10,10,16 + .long 0x10E7CD08 + + lvsr 5,0,31 + .long 0x10E7D508 + + .long 0x7C005699 + .long 0x10E7DD08 + + addi 7,1,64+15 + .long 0x10E7E508 + lvx 24,0,7 + + .long 0x10E7ED08 + lvx 25,3,7 + vxor 17,17,31 + + + .long 0x10E7F508 + + vperm 0,0,0,5 + .long 0x10E78D09 + + vor 17,18,18 + vxor 11,7,18 + + .long 0x7CE02799 + addi 4,4,0x10 + bne .Lxts_enc6x_steal + b .Lxts_enc6x_done + +.align 4 +.Lxts_enc6x_zero: + cmpwi 31,0 + beq .Lxts_enc6x_done + + add 10,10,31 + subi 10,10,16 + .long 0x7C005699 + lvsr 5,0,31 + + vperm 0,0,0,5 + vxor 11,11,17 +.Lxts_enc6x_steal: + vxor 0,0,17 + vxor 7,7,7 + vspltisb 12,-1 + vperm 7,7,12,5 + vsel 7,0,11,7 + + subi 30,4,17 + subi 4,4,16 + mtctr 31 +.Loop_xts_enc6x_steal: + lbzu 0,1(30) + stb 0,16(30) + bdnz .Loop_xts_enc6x_steal + + li 31,0 + mtctr 9 + b .Loop_xts_enc1x + +.align 4 +.Lxts_enc6x_done: + cmpldi 8,0 + beq .Lxts_enc6x_ret + + vxor 8,17,23 + + .long 0x7D004799 + +.Lxts_enc6x_ret: + mtlr 11 + li 10,79 + li 11,95 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + + mtspr 256,12 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + ld 26,400(1) + ld 27,408(1) + ld 28,416(1) + ld 29,424(1) + ld 30,432(1) + ld 31,440(1) + addi 1,1,448 + blr +.long 0 +.byte 0,12,0x04,1,0x80,6,6,0 +.long 0 + +.align 5 +_aesp8_xts_enc5x: + .long 0x10E7C508 + .long 0x118CC508 + .long 0x11ADC508 + .long 0x11CEC508 + .long 0x11EFC508 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD08 + .long 0x118CCD08 + .long 0x11ADCD08 + .long 0x11CECD08 + .long 0x11EFCD08 + lvx 25,3,7 + bdnz _aesp8_xts_enc5x + + add 10,10,31 + cmpwi 31,0 + .long 0x10E7C508 + .long 0x118CC508 + .long 0x11ADC508 + .long 0x11CEC508 + .long 0x11EFC508 + + subi 10,10,16 + .long 0x10E7CD08 + .long 0x118CCD08 + .long 0x11ADCD08 + .long 0x11CECD08 + .long 0x11EFCD08 + vxor 17,17,31 + + .long 0x10E7D508 + lvsr 5,0,31 + .long 0x118CD508 + .long 0x11ADD508 + .long 0x11CED508 + .long 0x11EFD508 + vxor 1,18,31 + + .long 0x10E7DD08 + .long 0x7C005699 + .long 0x118CDD08 + .long 0x11ADDD08 + .long 0x11CEDD08 + .long 0x11EFDD08 + vxor 2,19,31 + + addi 7,1,64+15 + .long 0x10E7E508 + .long 0x118CE508 + .long 0x11ADE508 + .long 0x11CEE508 + .long 0x11EFE508 + lvx 24,0,7 + vxor 3,20,31 + + .long 0x10E7ED08 + + .long 0x118CED08 + .long 0x11ADED08 + .long 0x11CEED08 + .long 0x11EFED08 + lvx 25,3,7 + vxor 4,21,31 + + .long 0x10E7F508 + vperm 0,0,0,5 + .long 0x118CF508 + .long 0x11ADF508 + .long 0x11CEF508 + .long 0x11EFF508 + + .long 0x10E78D09 + .long 0x118C0D09 + .long 0x11AD1509 + .long 0x11CE1D09 + .long 0x11EF2509 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.align 5 +_aesp8_xts_decrypt6x: + stdu 1,-448(1) + mflr 11 + li 7,207 + li 3,223 + std 11,464(1) + stvx 20,7,1 + addi 7,7,32 + stvx 21,3,1 + addi 3,3,32 + stvx 22,7,1 + addi 7,7,32 + stvx 23,3,1 + addi 3,3,32 + stvx 24,7,1 + addi 7,7,32 + stvx 25,3,1 + addi 3,3,32 + stvx 26,7,1 + addi 7,7,32 + stvx 27,3,1 + addi 3,3,32 + stvx 28,7,1 + addi 7,7,32 + stvx 29,3,1 + addi 3,3,32 + stvx 30,7,1 + stvx 31,3,1 + li 0,-1 + stw 12,396(1) + li 3,0x10 + std 26,400(1) + li 26,0x20 + std 27,408(1) + li 27,0x30 + std 28,416(1) + li 28,0x40 + std 29,424(1) + li 29,0x50 + std 30,432(1) + li 30,0x60 + std 31,440(1) + li 31,0x70 + mtspr 256,0 + + subi 9,9,3 + + lvx 23,0,6 + lvx 30,3,6 + addi 6,6,0x20 + lvx 31,0,6 + vperm 23,23,30,7 + addi 7,1,64+15 + mtctr 9 + +.Load_xts_dec_key: + vperm 24,30,31,7 + lvx 30,3,6 + addi 6,6,0x20 + stvx 24,0,7 + vperm 25,31,30,7 + lvx 31,0,6 + stvx 25,3,7 + addi 7,7,0x20 + bdnz .Load_xts_dec_key + + lvx 26,3,6 + vperm 24,30,31,7 + lvx 27,26,6 + stvx 24,0,7 + vperm 25,31,26,7 + lvx 28,27,6 + stvx 25,3,7 + addi 7,1,64+15 + vperm 26,26,27,7 + lvx 29,28,6 + vperm 27,27,28,7 + lvx 30,29,6 + vperm 28,28,29,7 + lvx 31,30,6 + vperm 29,29,30,7 + lvx 22,31,6 + vperm 30,30,31,7 + lvx 24,0,7 + vperm 31,31,22,7 + lvx 25,3,7 + + vperm 0,2,4,5 + subi 10,10,31 + vxor 17,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + vand 11,11,10 + vxor 7,0,17 + vxor 8,8,11 + + .long 0x7C235699 + vxor 18,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 12,1,18 + vxor 8,8,11 + + .long 0x7C5A5699 + andi. 31,5,15 + vxor 19,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 13,2,19 + vxor 8,8,11 + + .long 0x7C7B5699 + sub 5,5,31 + vxor 20,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 14,3,20 + vxor 8,8,11 + + .long 0x7C9C5699 + subi 5,5,0x60 + vxor 21,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 15,4,21 + vxor 8,8,11 + + .long 0x7CBD5699 + addi 10,10,0x60 + vxor 22,8,23 + vsrab 11,8,9 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + vand 11,11,10 + vxor 16,5,22 + vxor 8,8,11 + + vxor 31,31,23 + mtctr 9 + b .Loop_xts_dec6x + +.align 5 +.Loop_xts_dec6x: + .long 0x10E7C548 + .long 0x118CC548 + .long 0x11ADC548 + .long 0x11CEC548 + .long 0x11EFC548 + .long 0x1210C548 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD48 + .long 0x118CCD48 + .long 0x11ADCD48 + .long 0x11CECD48 + .long 0x11EFCD48 + .long 0x1210CD48 + lvx 25,3,7 + bdnz .Loop_xts_dec6x + + subic 5,5,96 + vxor 0,17,31 + .long 0x10E7C548 + .long 0x118CC548 + vsrab 11,8,9 + vxor 17,8,23 + vaddubm 8,8,8 + .long 0x11ADC548 + .long 0x11CEC548 + vsldoi 11,11,11,15 + .long 0x11EFC548 + .long 0x1210C548 + + subfe. 0,0,0 + vand 11,11,10 + .long 0x10E7CD48 + .long 0x118CCD48 + vxor 8,8,11 + .long 0x11ADCD48 + .long 0x11CECD48 + vxor 1,18,31 + vsrab 11,8,9 + vxor 18,8,23 + .long 0x11EFCD48 + .long 0x1210CD48 + + and 0,0,5 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x10E7D548 + .long 0x118CD548 + vand 11,11,10 + .long 0x11ADD548 + .long 0x11CED548 + vxor 8,8,11 + .long 0x11EFD548 + .long 0x1210D548 + + add 10,10,0 + + + + vxor 2,19,31 + vsrab 11,8,9 + vxor 19,8,23 + vaddubm 8,8,8 + .long 0x10E7DD48 + .long 0x118CDD48 + vsldoi 11,11,11,15 + .long 0x11ADDD48 + .long 0x11CEDD48 + vand 11,11,10 + .long 0x11EFDD48 + .long 0x1210DD48 + + addi 7,1,64+15 + vxor 8,8,11 + .long 0x10E7E548 + .long 0x118CE548 + vxor 3,20,31 + vsrab 11,8,9 + vxor 20,8,23 + .long 0x11ADE548 + .long 0x11CEE548 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x11EFE548 + .long 0x1210E548 + lvx 24,0,7 + vand 11,11,10 + + .long 0x10E7ED48 + .long 0x118CED48 + vxor 8,8,11 + .long 0x11ADED48 + .long 0x11CEED48 + vxor 4,21,31 + vsrab 11,8,9 + vxor 21,8,23 + .long 0x11EFED48 + .long 0x1210ED48 + lvx 25,3,7 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + + .long 0x10E7F548 + .long 0x118CF548 + vand 11,11,10 + .long 0x11ADF548 + .long 0x11CEF548 + vxor 8,8,11 + .long 0x11EFF548 + .long 0x1210F548 + vxor 5,22,31 + vsrab 11,8,9 + vxor 22,8,23 + + .long 0x10E70549 + .long 0x7C005699 + vaddubm 8,8,8 + vsldoi 11,11,11,15 + .long 0x118C0D49 + .long 0x7C235699 + .long 0x11AD1549 + + .long 0x7C5A5699 + vand 11,11,10 + .long 0x11CE1D49 + + .long 0x7C7B5699 + .long 0x11EF2549 + + .long 0x7C9C5699 + vxor 8,8,11 + .long 0x12102D49 + + .long 0x7CBD5699 + addi 10,10,0x60 + + + + + + .long 0x7CE02799 + vxor 7,0,17 + + .long 0x7D832799 + vxor 12,1,18 + + .long 0x7DBA2799 + vxor 13,2,19 + + .long 0x7DDB2799 + vxor 14,3,20 + + .long 0x7DFC2799 + vxor 15,4,21 + .long 0x7E1D2799 + vxor 16,5,22 + addi 4,4,0x60 + + mtctr 9 + beq .Loop_xts_dec6x + + addic. 5,5,0x60 + beq .Lxts_dec6x_zero + cmpwi 5,0x20 + blt .Lxts_dec6x_one + nop + beq .Lxts_dec6x_two + cmpwi 5,0x40 + blt .Lxts_dec6x_three + nop + beq .Lxts_dec6x_four + +.Lxts_dec6x_five: + vxor 7,1,17 + vxor 12,2,18 + vxor 13,3,19 + vxor 14,4,20 + vxor 15,5,21 + + bl _aesp8_xts_dec5x + + + vor 17,22,22 + vxor 18,8,23 + + .long 0x7CE02799 + vxor 7,0,18 + + .long 0x7D832799 + + .long 0x7DBA2799 + + .long 0x7DDB2799 + .long 0x7DFC2799 + addi 4,4,0x50 + bne .Lxts_dec6x_steal + b .Lxts_dec6x_done + +.align 4 +.Lxts_dec6x_four: + vxor 7,2,17 + vxor 12,3,18 + vxor 13,4,19 + vxor 14,5,20 + vxor 15,15,15 + + bl _aesp8_xts_dec5x + + + vor 17,21,21 + vor 18,22,22 + + .long 0x7CE02799 + vxor 7,0,22 + + .long 0x7D832799 + + .long 0x7DBA2799 + .long 0x7DDB2799 + addi 4,4,0x40 + bne .Lxts_dec6x_steal + b .Lxts_dec6x_done + +.align 4 +.Lxts_dec6x_three: + vxor 7,3,17 + vxor 12,4,18 + vxor 13,5,19 + vxor 14,14,14 + vxor 15,15,15 + + bl _aesp8_xts_dec5x + + + vor 17,20,20 + vor 18,21,21 + + .long 0x7CE02799 + vxor 7,0,21 + + .long 0x7D832799 + .long 0x7DBA2799 + addi 4,4,0x30 + bne .Lxts_dec6x_steal + b .Lxts_dec6x_done + +.align 4 +.Lxts_dec6x_two: + vxor 7,4,17 + vxor 12,5,18 + vxor 13,13,13 + vxor 14,14,14 + vxor 15,15,15 + + bl _aesp8_xts_dec5x + + + vor 17,19,19 + vor 18,20,20 + + .long 0x7CE02799 + vxor 7,0,20 + .long 0x7D832799 + addi 4,4,0x20 + bne .Lxts_dec6x_steal + b .Lxts_dec6x_done + +.align 4 +.Lxts_dec6x_one: + vxor 7,5,17 + nop +.Loop_xts_dec1x: + .long 0x10E7C548 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD48 + lvx 25,3,7 + bdnz .Loop_xts_dec1x + + subi 0,31,1 + .long 0x10E7C548 + + andi. 0,0,16 + cmpwi 31,0 + .long 0x10E7CD48 + + sub 10,10,0 + .long 0x10E7D548 + + .long 0x7C005699 + .long 0x10E7DD48 + + addi 7,1,64+15 + .long 0x10E7E548 + lvx 24,0,7 + + .long 0x10E7ED48 + lvx 25,3,7 + vxor 17,17,31 + + + .long 0x10E7F548 + + mtctr 9 + .long 0x10E78D49 + + vor 17,18,18 + vor 18,19,19 + + .long 0x7CE02799 + addi 4,4,0x10 + vxor 7,0,19 + bne .Lxts_dec6x_steal + b .Lxts_dec6x_done + +.align 4 +.Lxts_dec6x_zero: + cmpwi 31,0 + beq .Lxts_dec6x_done + + .long 0x7C005699 + + vxor 7,0,18 +.Lxts_dec6x_steal: + .long 0x10E7C548 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD48 + lvx 25,3,7 + bdnz .Lxts_dec6x_steal + + add 10,10,31 + .long 0x10E7C548 + + cmpwi 31,0 + .long 0x10E7CD48 + + .long 0x7C005699 + .long 0x10E7D548 + + lvsr 5,0,31 + .long 0x10E7DD48 + + addi 7,1,64+15 + .long 0x10E7E548 + lvx 24,0,7 + + .long 0x10E7ED48 + lvx 25,3,7 + vxor 18,18,31 + + + .long 0x10E7F548 + + vperm 0,0,0,5 + .long 0x11679549 + + + + .long 0x7D602799 + + vxor 7,7,7 + vspltisb 12,-1 + vperm 7,7,12,5 + vsel 7,0,11,7 + vxor 7,7,17 + + subi 30,4,1 + mtctr 31 +.Loop_xts_dec6x_steal: + lbzu 0,1(30) + stb 0,16(30) + bdnz .Loop_xts_dec6x_steal + + li 31,0 + mtctr 9 + b .Loop_xts_dec1x + +.align 4 +.Lxts_dec6x_done: + cmpldi 8,0 + beq .Lxts_dec6x_ret + + vxor 8,17,23 + + .long 0x7D004799 + +.Lxts_dec6x_ret: + mtlr 11 + li 10,79 + li 11,95 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + stvx 9,10,1 + addi 10,10,32 + stvx 9,11,1 + addi 11,11,32 + + mtspr 256,12 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + ld 26,400(1) + ld 27,408(1) + ld 28,416(1) + ld 29,424(1) + ld 30,432(1) + ld 31,440(1) + addi 1,1,448 + blr +.long 0 +.byte 0,12,0x04,1,0x80,6,6,0 +.long 0 + +.align 5 +_aesp8_xts_dec5x: + .long 0x10E7C548 + .long 0x118CC548 + .long 0x11ADC548 + .long 0x11CEC548 + .long 0x11EFC548 + lvx 24,26,7 + addi 7,7,0x20 + + .long 0x10E7CD48 + .long 0x118CCD48 + .long 0x11ADCD48 + .long 0x11CECD48 + .long 0x11EFCD48 + lvx 25,3,7 + bdnz _aesp8_xts_dec5x + + subi 0,31,1 + .long 0x10E7C548 + .long 0x118CC548 + .long 0x11ADC548 + .long 0x11CEC548 + .long 0x11EFC548 + + andi. 0,0,16 + cmpwi 31,0 + .long 0x10E7CD48 + .long 0x118CCD48 + .long 0x11ADCD48 + .long 0x11CECD48 + .long 0x11EFCD48 + vxor 17,17,31 + + sub 10,10,0 + .long 0x10E7D548 + .long 0x118CD548 + .long 0x11ADD548 + .long 0x11CED548 + .long 0x11EFD548 + vxor 1,18,31 + + .long 0x10E7DD48 + .long 0x7C005699 + .long 0x118CDD48 + .long 0x11ADDD48 + .long 0x11CEDD48 + .long 0x11EFDD48 + vxor 2,19,31 + + addi 7,1,64+15 + .long 0x10E7E548 + .long 0x118CE548 + .long 0x11ADE548 + .long 0x11CEE548 + .long 0x11EFE548 + lvx 24,0,7 + vxor 3,20,31 + + .long 0x10E7ED48 + + .long 0x118CED48 + .long 0x11ADED48 + .long 0x11CEED48 + .long 0x11EFED48 + lvx 25,3,7 + vxor 4,21,31 + + .long 0x10E7F548 + .long 0x118CF548 + .long 0x11ADF548 + .long 0x11CEF548 + .long 0x11EFF548 + + .long 0x10E78D49 + .long 0x118C0D49 + .long 0x11AD1549 + .long 0x11CE1D49 + .long 0x11EF2549 + mtctr 9 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 diff --git a/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/vpaes-ppc.s b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/vpaes-ppc.s new file mode 100644 index 0000000000..439da0fe81 --- /dev/null +++ b/deps/openssl/config/archs/linux-ppc64/asm_avx2/crypto/aes/vpaes-ppc.s @@ -0,0 +1,1497 @@ +.machine "any" + +.text + +.align 7 +_vpaes_consts: +.Lk_mc_forward: +.byte 0x01,0x02,0x03,0x00,0x05,0x06,0x07,0x04,0x09,0x0a,0x0b,0x08,0x0d,0x0e,0x0f,0x0c +.byte 0x05,0x06,0x07,0x04,0x09,0x0a,0x0b,0x08,0x0d,0x0e,0x0f,0x0c,0x01,0x02,0x03,0x00 +.byte 0x09,0x0a,0x0b,0x08,0x0d,0x0e,0x0f,0x0c,0x01,0x02,0x03,0x00,0x05,0x06,0x07,0x04 +.byte 0x0d,0x0e,0x0f,0x0c,0x01,0x02,0x03,0x00,0x05,0x06,0x07,0x04,0x09,0x0a,0x0b,0x08 +.Lk_mc_backward: +.byte 0x03,0x00,0x01,0x02,0x07,0x04,0x05,0x06,0x0b,0x08,0x09,0x0a,0x0f,0x0c,0x0d,0x0e +.byte 0x0f,0x0c,0x0d,0x0e,0x03,0x00,0x01,0x02,0x07,0x04,0x05,0x06,0x0b,0x08,0x09,0x0a +.byte 0x0b,0x08,0x09,0x0a,0x0f,0x0c,0x0d,0x0e,0x03,0x00,0x01,0x02,0x07,0x04,0x05,0x06 +.byte 0x07,0x04,0x05,0x06,0x0b,0x08,0x09,0x0a,0x0f,0x0c,0x0d,0x0e,0x03,0x00,0x01,0x02 +.Lk_sr: +.byte 0x00,0x01,0x02,0x03,0x04,0x05,0x06,0x07,0x08,0x09,0x0a,0x0b,0x0c,0x0d,0x0e,0x0f +.byte 0x00,0x05,0x0a,0x0f,0x04,0x09,0x0e,0x03,0x08,0x0d,0x02,0x07,0x0c,0x01,0x06,0x0b +.byte 0x00,0x09,0x02,0x0b,0x04,0x0d,0x06,0x0f,0x08,0x01,0x0a,0x03,0x0c,0x05,0x0e,0x07 +.byte 0x00,0x0d,0x0a,0x07,0x04,0x01,0x0e,0x0b,0x08,0x05,0x02,0x0f,0x0c,0x09,0x06,0x03 + + + + +.Lk_inv: +.byte 0xf0,0x01,0x08,0x0d,0x0f,0x06,0x05,0x0e,0x02,0x0c,0x0b,0x0a,0x09,0x03,0x07,0x04 +.byte 0xf0,0x07,0x0b,0x0f,0x06,0x0a,0x04,0x01,0x09,0x08,0x05,0x02,0x0c,0x0e,0x0d,0x03 +.Lk_ipt: +.byte 0x00,0x70,0x2a,0x5a,0x98,0xe8,0xb2,0xc2,0x08,0x78,0x22,0x52,0x90,0xe0,0xba,0xca +.byte 0x00,0x4d,0x7c,0x31,0x7d,0x30,0x01,0x4c,0x81,0xcc,0xfd,0xb0,0xfc,0xb1,0x80,0xcd +.Lk_sbo: +.byte 0x00,0xc7,0xbd,0x6f,0x17,0x6d,0xd2,0xd0,0x78,0xa8,0x02,0xc5,0x7a,0xbf,0xaa,0x15 +.byte 0x00,0x6a,0xbb,0x5f,0xa5,0x74,0xe4,0xcf,0xfa,0x35,0x2b,0x41,0xd1,0x90,0x1e,0x8e +.Lk_sb1: +.byte 0x00,0x23,0xe2,0xfa,0x15,0xd4,0x18,0x36,0xef,0xd9,0x2e,0x0d,0xc1,0xcc,0xf7,0x3b +.byte 0x00,0x3e,0x50,0xcb,0x8f,0xe1,0x9b,0xb1,0x44,0xf5,0x2a,0x14,0x6e,0x7a,0xdf,0xa5 +.Lk_sb2: +.byte 0x00,0x29,0xe1,0x0a,0x40,0x88,0xeb,0x69,0x4a,0x23,0x82,0xab,0xc8,0x63,0xa1,0xc2 +.byte 0x00,0x24,0x71,0x0b,0xc6,0x93,0x7a,0xe2,0xcd,0x2f,0x98,0xbc,0x55,0xe9,0xb7,0x5e + + + + +.Lk_dipt: +.byte 0x00,0x5f,0x54,0x0b,0x04,0x5b,0x50,0x0f,0x1a,0x45,0x4e,0x11,0x1e,0x41,0x4a,0x15 +.byte 0x00,0x65,0x05,0x60,0xe6,0x83,0xe3,0x86,0x94,0xf1,0x91,0xf4,0x72,0x17,0x77,0x12 +.Lk_dsbo: +.byte 0x00,0x40,0xf9,0x7e,0x53,0xea,0x87,0x13,0x2d,0x3e,0x94,0xd4,0xb9,0x6d,0xaa,0xc7 +.byte 0x00,0x1d,0x44,0x93,0x0f,0x56,0xd7,0x12,0x9c,0x8e,0xc5,0xd8,0x59,0x81,0x4b,0xca +.Lk_dsb9: +.byte 0x00,0xd6,0x86,0x9a,0x53,0x03,0x1c,0x85,0xc9,0x4c,0x99,0x4f,0x50,0x1f,0xd5,0xca +.byte 0x00,0x49,0xd7,0xec,0x89,0x17,0x3b,0xc0,0x65,0xa5,0xfb,0xb2,0x9e,0x2c,0x5e,0x72 +.Lk_dsbd: +.byte 0x00,0xa2,0xb1,0xe6,0xdf,0xcc,0x57,0x7d,0x39,0x44,0x2a,0x88,0x13,0x9b,0x6e,0xf5 +.byte 0x00,0xcb,0xc6,0x24,0xf7,0xfa,0xe2,0x3c,0xd3,0xef,0xde,0x15,0x0d,0x18,0x31,0x29 +.Lk_dsbb: +.byte 0x00,0x42,0xb4,0x96,0x92,0x64,0x22,0xd0,0x04,0xd4,0xf2,0xb0,0xf6,0x46,0x26,0x60 +.byte 0x00,0x67,0x59,0xcd,0xa6,0x98,0x94,0xc1,0x6b,0xaa,0x55,0x32,0x3e,0x0c,0xff,0xf3 +.Lk_dsbe: +.byte 0x00,0xd0,0xd4,0x26,0x96,0x92,0xf2,0x46,0xb0,0xf6,0xb4,0x64,0x04,0x60,0x42,0x22 +.byte 0x00,0xc1,0xaa,0xff,0xcd,0xa6,0x55,0x0c,0x32,0x3e,0x59,0x98,0x6b,0xf3,0x67,0x94 + + + + +.Lk_dksd: +.byte 0x00,0x47,0xe4,0xa3,0x5d,0x1a,0xb9,0xfe,0xf9,0xbe,0x1d,0x5a,0xa4,0xe3,0x40,0x07 +.byte 0x00,0x83,0x36,0xb5,0xf4,0x77,0xc2,0x41,0x1e,0x9d,0x28,0xab,0xea,0x69,0xdc,0x5f +.Lk_dksb: +.byte 0x00,0xd5,0x50,0x85,0x1f,0xca,0x4f,0x9a,0x99,0x4c,0xc9,0x1c,0x86,0x53,0xd6,0x03 +.byte 0x00,0x4a,0xfc,0xb6,0xa7,0xed,0x5b,0x11,0xc8,0x82,0x34,0x7e,0x6f,0x25,0x93,0xd9 +.Lk_dkse: +.byte 0x00,0xd6,0xc9,0x1f,0xca,0x1c,0x03,0xd5,0x86,0x50,0x4f,0x99,0x4c,0x9a,0x85,0x53 +.byte 0xe8,0x7b,0xdc,0x4f,0x05,0x96,0x31,0xa2,0x87,0x14,0xb3,0x20,0x6a,0xf9,0x5e,0xcd +.Lk_dks9: +.byte 0x00,0xa7,0xd9,0x7e,0xc8,0x6f,0x11,0xb6,0xfc,0x5b,0x25,0x82,0x34,0x93,0xed,0x4a +.byte 0x00,0x33,0x14,0x27,0x62,0x51,0x76,0x45,0xce,0xfd,0xda,0xe9,0xac,0x9f,0xb8,0x8b + +.Lk_rcon: +.byte 0xb6,0xee,0x9d,0xaf,0xb9,0x91,0x83,0x1f,0x81,0x7d,0x7c,0x4d,0x08,0x98,0x2a,0x70 +.Lk_s63: +.byte 0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b,0x5b + +.Lk_opt: +.byte 0x00,0x60,0xb6,0xd6,0x29,0x49,0x9f,0xff,0x08,0x68,0xbe,0xde,0x21,0x41,0x97,0xf7 +.byte 0x00,0xec,0xbc,0x50,0x51,0xbd,0xed,0x01,0xe0,0x0c,0x5c,0xb0,0xb1,0x5d,0x0d,0xe1 +.Lk_deskew: +.byte 0x00,0xe3,0xa4,0x47,0x40,0xa3,0xe4,0x07,0x1a,0xf9,0xbe,0x5d,0x5a,0xb9,0xfe,0x1d +.byte 0x00,0x69,0xea,0x83,0xdc,0xb5,0x36,0x5f,0x77,0x1e,0x9d,0xf4,0xab,0xc2,0x41,0x28 +.align 5 +.Lconsts: + mflr 0 + bcl 20,31,$+4 + mflr 12 + addi 12,12,-0x308 + mtlr 0 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 +.byte 86,101,99,116,111,114,32,80,101,114,109,117,116,97,116,105,111,110,32,65,69,83,32,102,111,114,32,65,108,116,105,86,101,99,44,77,105,107,101,32,72,97,109,98,117,114,103,32,40,83,116,97,110,102,111,114,100,32,85,110,105,118,101,114,115,105,116,121,41,0 +.align 2 +.align 6 + + + + + + +.align 4 +_vpaes_encrypt_preheat: + mflr 8 + bl .Lconsts + mtlr 8 + li 11, 0xc0 + li 10, 0xd0 + li 9, 0xe0 + li 8, 0xf0 + vxor 7, 7, 7 + vspltisb 8,4 + vspltisb 9,0x0f + lvx 10, 12, 11 + li 11, 0x100 + lvx 11, 12, 10 + li 10, 0x110 + lvx 12, 12, 9 + li 9, 0x120 + lvx 13, 12, 8 + li 8, 0x130 + lvx 14, 12, 11 + li 11, 0x140 + lvx 15, 12, 10 + li 10, 0x150 + lvx 16, 12, 9 + lvx 17, 12, 8 + lvx 18, 12, 11 + lvx 19, 12, 10 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + + + + + + + + + + +.align 5 +_vpaes_encrypt_core: + lwz 8, 240(5) + li 9, 16 + lvx 5, 0, 5 + li 11, 0x10 + lvx 6, 9, 5 + addi 9, 9, 16 + vperm 5, 5, 6, 31 + addi 10, 11, 0x40 + vsrb 1, 0, 8 + vperm 0, 12, 12, 0 + vperm 1, 13, 13, 1 + vxor 0, 0, 5 + vxor 0, 0, 1 + mtctr 8 + b .Lenc_entry + +.align 4 +.Lenc_loop: + + vperm 4, 17, 7, 2 + lvx 1, 12, 11 + addi 11, 11, 16 + vperm 0, 16, 7, 3 + vxor 4, 4, 5 + andi. 11, 11, 0x30 + vperm 5, 19, 7, 2 + vxor 0, 0, 4 + vperm 2, 18, 7, 3 + lvx 4, 12, 10 + addi 10, 11, 0x40 + vperm 3, 0, 7, 1 + vxor 2, 2, 5 + vperm 0, 0, 7, 4 + vxor 3, 3, 2 + vperm 4, 3, 7, 1 + vxor 0, 0, 3 + vxor 0, 0, 4 + +.Lenc_entry: + + vsrb 1, 0, 8 + vperm 5, 11, 11, 0 + vxor 0, 0, 1 + vperm 3, 10, 10, 1 + vperm 4, 10, 10, 0 + vand 0, 0, 9 + vxor 3, 3, 5 + vxor 4, 4, 5 + vperm 2, 10, 7, 3 + vor 5,6,6 + lvx 6, 9, 5 + vperm 3, 10, 7, 4 + addi 9, 9, 16 + vxor 2, 2, 0 + vperm 5, 5, 6, 31 + vxor 3, 3, 1 + bdnz .Lenc_loop + + + addi 10, 11, 0x80 + + + vperm 4, 14, 7, 2 + lvx 1, 12, 10 + vperm 0, 15, 7, 3 + vxor 4, 4, 5 + vxor 0, 0, 4 + vperm 0, 0, 7, 1 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.globl vpaes_encrypt +.type vpaes_encrypt,@function +.section ".opd","aw" +.align 3 +vpaes_encrypt: +.quad .vpaes_encrypt,.TOC.@tocbase,0 +.previous +.align 5 +.vpaes_encrypt: + stdu 1,-256(1) + li 10,63 + li 11,79 + mflr 6 + mfspr 7,256 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + stw 7,252(1) + li 0, -1 + std 6,272(1) + mtspr 256,0 + + bl _vpaes_encrypt_preheat + + lvsl 27, 0, 3 + lvx 0, 0, 3 + addi 3, 3, 15 + lvsr 29, 0, 4 + lvsl 31, 0, 5 + lvx 26, 0, 3 + vperm 0, 0, 26, 27 + + bl _vpaes_encrypt_core + + andi. 8, 4, 15 + li 9, 16 + beq .Lenc_out_aligned + + vperm 0, 0, 0, 29 + mtctr 9 +.Lenc_out_unaligned: + stvebx 0, 0, 4 + addi 4, 4, 1 + bdnz .Lenc_out_unaligned + b .Lenc_done + +.align 4 +.Lenc_out_aligned: + stvx 0, 0, 4 +.Lenc_done: + + li 10,63 + li 11,79 + mtlr 6 + mtspr 256,7 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + addi 1,1,256 + blr +.long 0 +.byte 0,12,0x04,1,0x80,0,3,0 +.long 0 +.size .vpaes_encrypt,.-.vpaes_encrypt +.size vpaes_encrypt,.-.vpaes_encrypt + +.align 4 +_vpaes_decrypt_preheat: + mflr 8 + bl .Lconsts + mtlr 8 + li 11, 0xc0 + li 10, 0xd0 + li 9, 0x160 + li 8, 0x170 + vxor 7, 7, 7 + vspltisb 8,4 + vspltisb 9,0x0f + lvx 10, 12, 11 + li 11, 0x180 + lvx 11, 12, 10 + li 10, 0x190 + lvx 12, 12, 9 + li 9, 0x1a0 + lvx 13, 12, 8 + li 8, 0x1b0 + lvx 14, 12, 11 + li 11, 0x1c0 + lvx 15, 12, 10 + li 10, 0x1d0 + lvx 16, 12, 9 + li 9, 0x1e0 + lvx 17, 12, 8 + li 8, 0x1f0 + lvx 18, 12, 11 + li 11, 0x200 + lvx 19, 12, 10 + li 10, 0x210 + lvx 20, 12, 9 + lvx 21, 12, 8 + lvx 22, 12, 11 + lvx 23, 12, 10 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + +.align 4 +_vpaes_decrypt_core: + lwz 8, 240(5) + li 9, 16 + lvx 5, 0, 5 + li 11, 0x30 + lvx 6, 9, 5 + addi 9, 9, 16 + vperm 5, 5, 6, 31 + vsrb 1, 0, 8 + vperm 0, 12, 12, 0 + vperm 1, 13, 13, 1 + vxor 0, 0, 5 + vxor 0, 0, 1 + mtctr 8 + b .Ldec_entry + +.align 4 +.Ldec_loop: + + + + lvx 0, 12, 11 + + + vperm 4, 16, 7, 2 + subi 11, 11, 16 + vperm 1, 17, 7, 3 + andi. 11, 11, 0x30 + vxor 5, 5, 4 + + vxor 5, 5, 1 + + + vperm 4, 18, 7, 2 + vperm 5, 5, 7, 0 + vperm 1, 19, 7, 3 + vxor 5, 5, 4 + + vxor 5, 5, 1 + + + vperm 4, 20, 7, 2 + vperm 5, 5, 7, 0 + vperm 1, 21, 7, 3 + vxor 5, 5, 4 + + vxor 5, 5, 1 + + + vperm 4, 22, 7, 2 + vperm 5, 5, 7, 0 + vperm 1, 23, 7, 3 + vxor 0, 5, 4 + vxor 0, 0, 1 + +.Ldec_entry: + + vsrb 1, 0, 8 + vperm 2, 11, 11, 0 + vxor 0, 0, 1 + vperm 3, 10, 10, 1 + vperm 4, 10, 10, 0 + vand 0, 0, 9 + vxor 3, 3, 2 + vxor 4, 4, 2 + vperm 2, 10, 7, 3 + vor 5,6,6 + lvx 6, 9, 5 + vperm 3, 10, 7, 4 + addi 9, 9, 16 + vxor 2, 2, 0 + vperm 5, 5, 6, 31 + vxor 3, 3, 1 + bdnz .Ldec_loop + + + addi 10, 11, 0x80 + + vperm 4, 14, 7, 2 + + lvx 2, 12, 10 + vperm 1, 15, 7, 3 + vxor 4, 4, 5 + vxor 0, 1, 4 + vperm 0, 0, 7, 2 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.globl vpaes_decrypt +.type vpaes_decrypt,@function +.section ".opd","aw" +.align 3 +vpaes_decrypt: +.quad .vpaes_decrypt,.TOC.@tocbase,0 +.previous +.align 5 +.vpaes_decrypt: + stdu 1,-256(1) + li 10,63 + li 11,79 + mflr 6 + mfspr 7,256 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + stw 7,252(1) + li 0, -1 + std 6,272(1) + mtspr 256,0 + + bl _vpaes_decrypt_preheat + + lvsl 27, 0, 3 + lvx 0, 0, 3 + addi 3, 3, 15 + lvsr 29, 0, 4 + lvsl 31, 0, 5 + lvx 26, 0, 3 + vperm 0, 0, 26, 27 + + bl _vpaes_decrypt_core + + andi. 8, 4, 15 + li 9, 16 + beq .Ldec_out_aligned + + vperm 0, 0, 0, 29 + mtctr 9 +.Ldec_out_unaligned: + stvebx 0, 0, 4 + addi 4, 4, 1 + bdnz .Ldec_out_unaligned + b .Ldec_done + +.align 4 +.Ldec_out_aligned: + stvx 0, 0, 4 +.Ldec_done: + + li 10,63 + li 11,79 + mtlr 6 + mtspr 256,7 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + addi 1,1,256 + blr +.long 0 +.byte 0,12,0x04,1,0x80,0,3,0 +.long 0 +.size .vpaes_decrypt,.-.vpaes_decrypt +.size vpaes_decrypt,.-.vpaes_decrypt + +.globl vpaes_cbc_encrypt +.type vpaes_cbc_encrypt,@function +.section ".opd","aw" +.align 3 +vpaes_cbc_encrypt: +.quad .vpaes_cbc_encrypt,.TOC.@tocbase,0 +.previous +.align 5 +.vpaes_cbc_encrypt: + cmpldi 5,16 + .long 0x4dc00020 + + stdu 1,-272(1) + mflr 0 + li 10,63 + li 11,79 + mfspr 12,256 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + stw 12,252(1) + std 30,256(1) + std 31,264(1) + li 9, -16 + std 0, 288(1) + + and 30, 5, 9 + andi. 9, 4, 15 + mr 5, 6 + mr 31, 7 + li 6, -1 + mcrf 1, 0 + mr 7, 12 + mtspr 256,6 + + lvx 24, 0, 31 + li 9, 15 + lvsl 27, 0, 31 + lvx 25, 9, 31 + vperm 24, 24, 25, 27 + + cmpwi 8, 0 + neg 8, 3 + vxor 7, 7, 7 + lvsl 31, 0, 5 + lvsr 29, 0, 4 + lvsr 27, 0, 8 + vnor 30, 7, 7 + lvx 26, 0, 3 + vperm 30, 7, 30, 29 + addi 3, 3, 15 + + beq .Lcbc_decrypt + + bl _vpaes_encrypt_preheat + li 0, 16 + + beq 1, .Lcbc_enc_loop + + vor 0,26,26 + lvx 26, 0, 3 + addi 3, 3, 16 + vperm 0, 0, 26, 27 + vxor 0, 0, 24 + + bl _vpaes_encrypt_core + + andi. 8, 4, 15 + vor 24,0,0 + sub 9, 4, 8 + vperm 28, 0, 0, 29 + +.Lcbc_enc_head: + stvebx 28, 8, 9 + cmpwi 8, 15 + addi 8, 8, 1 + bne .Lcbc_enc_head + + sub. 30, 30, 0 + addi 4, 4, 16 + beq .Lcbc_unaligned_done + +.Lcbc_enc_loop: + vor 0,26,26 + lvx 26, 0, 3 + addi 3, 3, 16 + vperm 0, 0, 26, 27 + vxor 0, 0, 24 + + bl _vpaes_encrypt_core + + vor 24,0,0 + sub. 30, 30, 0 + vperm 0, 0, 0, 29 + vsel 1, 28, 0, 30 + vor 28,0,0 + stvx 1, 0, 4 + addi 4, 4, 16 + bne .Lcbc_enc_loop + + b .Lcbc_done + +.align 5 +.Lcbc_decrypt: + bl _vpaes_decrypt_preheat + li 0, 16 + + beq 1, .Lcbc_dec_loop + + vor 0,26,26 + lvx 26, 0, 3 + addi 3, 3, 16 + vperm 0, 0, 26, 27 + vor 25,0,0 + + bl _vpaes_decrypt_core + + andi. 8, 4, 15 + vxor 0, 0, 24 + vor 24,25,25 + sub 9, 4, 8 + vperm 28, 0, 0, 29 + +.Lcbc_dec_head: + stvebx 28, 8, 9 + cmpwi 8, 15 + addi 8, 8, 1 + bne .Lcbc_dec_head + + sub. 30, 30, 0 + addi 4, 4, 16 + beq .Lcbc_unaligned_done + +.Lcbc_dec_loop: + vor 0,26,26 + lvx 26, 0, 3 + addi 3, 3, 16 + vperm 0, 0, 26, 27 + vor 25,0,0 + + bl _vpaes_decrypt_core + + vxor 0, 0, 24 + vor 24,25,25 + sub. 30, 30, 0 + vperm 0, 0, 0, 29 + vsel 1, 28, 0, 30 + vor 28,0,0 + stvx 1, 0, 4 + addi 4, 4, 16 + bne .Lcbc_dec_loop + +.Lcbc_done: + beq 1, .Lcbc_write_iv + +.Lcbc_unaligned_done: + andi. 8, 4, 15 + sub 4, 4, 8 + li 9, 0 +.Lcbc_tail: + stvebx 28, 9, 4 + addi 9, 9, 1 + cmpw 9, 8 + bne .Lcbc_tail + +.Lcbc_write_iv: + neg 8, 31 + li 10, 4 + lvsl 29, 0, 8 + li 11, 8 + li 12, 12 + vperm 24, 24, 24, 29 + stvewx 24, 0, 31 + stvewx 24, 10, 31 + stvewx 24, 11, 31 + stvewx 24, 12, 31 + + mtspr 256,7 + li 10,63 + li 11,79 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 +.Lcbc_abort: + ld 0, 288(1) + ld 30,256(1) + ld 31,264(1) + mtlr 0 + addi 1,1,272 + blr +.long 0 +.byte 0,12,0x04,1,0x80,2,6,0 +.long 0 +.size .vpaes_cbc_encrypt,.-.vpaes_cbc_encrypt +.size vpaes_cbc_encrypt,.-.vpaes_cbc_encrypt + + + + + +.align 4 +_vpaes_key_preheat: + mflr 8 + bl .Lconsts + mtlr 8 + li 11, 0xc0 + li 10, 0xd0 + li 9, 0xe0 + li 8, 0xf0 + + vspltisb 8,4 + vxor 9,9,9 + lvx 10, 12, 11 + li 11, 0x120 + lvx 11, 12, 10 + li 10, 0x130 + lvx 12, 12, 9 + li 9, 0x220 + lvx 13, 12, 8 + li 8, 0x230 + + lvx 14, 12, 11 + li 11, 0x240 + lvx 15, 12, 10 + li 10, 0x250 + + lvx 16, 12, 9 + li 9, 0x260 + lvx 17, 12, 8 + li 8, 0x270 + lvx 18, 12, 11 + li 11, 0x280 + lvx 19, 12, 10 + li 10, 0x290 + lvx 20, 12, 9 + li 9, 0x2a0 + lvx 21, 12, 8 + li 8, 0x2b0 + lvx 22, 12, 11 + lvx 23, 12, 10 + + lvx 24, 12, 9 + lvx 25, 0, 12 + lvx 26, 12, 8 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.align 4 +_vpaes_schedule_core: + mflr 7 + + bl _vpaes_key_preheat + + + neg 8, 3 + lvx 0, 0, 3 + addi 3, 3, 15 + lvsr 27, 0, 8 + lvx 6, 0, 3 + addi 3, 3, 8 + vperm 0, 0, 6, 27 + + + vor 3,0,0 + bl _vpaes_schedule_transform + vor 7,0,0 + + bne 1, .Lschedule_am_decrypting + + + li 8, 0x30 + li 9, 4 + li 10, 8 + li 11, 12 + + lvsr 29, 0, 5 + vnor 30, 9, 9 + vperm 30, 9, 30, 29 + + + vperm 28, 0, 0, 29 + stvewx 28, 0, 5 + stvewx 28, 9, 5 + stvewx 28, 10, 5 + addi 10, 12, 0x80 + stvewx 28, 11, 5 + b .Lschedule_go + +.Lschedule_am_decrypting: + srwi 8, 4, 1 + andi. 8, 8, 32 + xori 8, 8, 32 + addi 10, 12, 0x80 + + lvx 1, 8, 10 + li 9, 4 + li 10, 8 + li 11, 12 + vperm 4, 3, 3, 1 + + neg 0, 5 + lvsl 29, 0, 0 + vnor 30, 9, 9 + vperm 30, 30, 9, 29 + + + vperm 28, 4, 4, 29 + stvewx 28, 0, 5 + stvewx 28, 9, 5 + stvewx 28, 10, 5 + addi 10, 12, 0x80 + stvewx 28, 11, 5 + addi 5, 5, 15 + xori 8, 8, 0x30 + +.Lschedule_go: + cmplwi 4, 192 + bgt .Lschedule_256 + beq .Lschedule_192 + + + + + + + + + + +.Lschedule_128: + li 0, 10 + mtctr 0 + +.Loop_schedule_128: + bl _vpaes_schedule_round + bdz .Lschedule_mangle_last + bl _vpaes_schedule_mangle + b .Loop_schedule_128 + + + + + + + + + + + + + + + + +.align 4 +.Lschedule_192: + li 0, 4 + lvx 0, 0, 3 + vperm 0, 6, 0, 27 + vsldoi 0, 3, 0, 8 + bl _vpaes_schedule_transform + vsldoi 6, 0, 9, 8 + vsldoi 6, 9, 6, 8 + mtctr 0 + +.Loop_schedule_192: + bl _vpaes_schedule_round + vsldoi 0, 6, 0, 8 + bl _vpaes_schedule_mangle + bl _vpaes_schedule_192_smear + bl _vpaes_schedule_mangle + bl _vpaes_schedule_round + bdz .Lschedule_mangle_last + bl _vpaes_schedule_mangle + bl _vpaes_schedule_192_smear + b .Loop_schedule_192 + + + + + + + + + + + +.align 4 +.Lschedule_256: + li 0, 7 + addi 3, 3, 8 + lvx 0, 0, 3 + vperm 0, 6, 0, 27 + bl _vpaes_schedule_transform + mtctr 0 + +.Loop_schedule_256: + bl _vpaes_schedule_mangle + vor 6,0,0 + + + bl _vpaes_schedule_round + bdz .Lschedule_mangle_last + bl _vpaes_schedule_mangle + + + vspltw 0, 0, 3 + vor 5,7,7 + vor 7,6,6 + bl _vpaes_schedule_low_round + vor 7,5,5 + + b .Loop_schedule_256 + + + + + + + + + + +.align 4 +.Lschedule_mangle_last: + + li 11, 0x2e0 + li 9, 0x2f0 + bne 1, .Lschedule_mangle_last_dec + + + lvx 1, 8, 10 + li 11, 0x2c0 + li 9, 0x2d0 + vperm 0, 0, 0, 1 + + lvx 12, 11, 12 + lvx 13, 9, 12 + addi 5, 5, 16 + vxor 0, 0, 26 + bl _vpaes_schedule_transform + + + vperm 0, 0, 0, 29 + li 10, 4 + vsel 2, 28, 0, 30 + li 11, 8 + stvx 2, 0, 5 + li 12, 12 + stvewx 0, 0, 5 + stvewx 0, 10, 5 + stvewx 0, 11, 5 + stvewx 0, 12, 5 + b .Lschedule_mangle_done + +.align 4 +.Lschedule_mangle_last_dec: + lvx 12, 11, 12 + lvx 13, 9, 12 + addi 5, 5, -16 + vxor 0, 0, 26 + bl _vpaes_schedule_transform + + + addi 9, 5, -15 + vperm 0, 0, 0, 29 + li 10, 4 + vsel 2, 28, 0, 30 + li 11, 8 + stvx 2, 0, 5 + li 12, 12 + stvewx 0, 0, 9 + stvewx 0, 10, 9 + stvewx 0, 11, 9 + stvewx 0, 12, 9 + + +.Lschedule_mangle_done: + mtlr 7 + + vxor 0, 0, 0 + vxor 1, 1, 1 + vxor 2, 2, 2 + vxor 3, 3, 3 + vxor 4, 4, 4 + vxor 5, 5, 5 + vxor 6, 6, 6 + vxor 7, 7, 7 + + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + + + + + + + + + + +.align 4 +_vpaes_schedule_192_smear: + vspltw 0, 7, 3 + vsldoi 1, 9, 6, 12 + vsldoi 0, 7, 0, 8 + vxor 6, 6, 1 + vxor 6, 6, 0 + vor 0,6,6 + vsldoi 6, 6, 9, 8 + vsldoi 6, 9, 6, 8 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + + + + + + + + + + + + + + +.align 4 +_vpaes_schedule_round: + + + vsldoi 1, 24, 9, 15 + vsldoi 24, 24, 24, 15 + vxor 7, 7, 1 + + + vspltw 0, 0, 3 + vsldoi 0, 0, 0, 1 + + + + +_vpaes_schedule_low_round: + + vsldoi 1, 9, 7, 12 + vxor 7, 7, 1 + vspltisb 1, 0x0f + vsldoi 4, 9, 7, 8 + + + vand 1, 1, 0 + vsrb 0, 0, 8 + vxor 7, 7, 4 + vperm 2, 11, 9, 1 + vxor 1, 1, 0 + vperm 3, 10, 9, 0 + vxor 3, 3, 2 + vperm 4, 10, 9, 1 + vxor 7, 7, 26 + vperm 3, 10, 9, 3 + vxor 4, 4, 2 + vperm 2, 10, 9, 4 + vxor 3, 3, 1 + vxor 2, 2, 0 + vperm 4, 15, 9, 3 + vperm 1, 14, 9, 2 + vxor 1, 1, 4 + + + vxor 0, 1, 7 + vxor 7, 1, 7 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + + + + + +.align 4 +_vpaes_schedule_transform: + + vsrb 2, 0, 8 + + vperm 0, 12, 12, 0 + + vperm 2, 13, 13, 2 + vxor 0, 0, 2 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + + + + + + + + + + + + + + + + + + + + + + + + +.align 4 +_vpaes_schedule_mangle: + + + bne 1, .Lschedule_mangle_dec + + + vxor 4, 0, 26 + addi 5, 5, 16 + vperm 4, 4, 4, 25 + vperm 1, 4, 4, 25 + vperm 3, 1, 1, 25 + vxor 4, 4, 1 + lvx 1, 8, 10 + vxor 3, 3, 4 + + vperm 3, 3, 3, 1 + addi 8, 8, -16 + andi. 8, 8, 0x30 + + + vperm 1, 3, 3, 29 + vsel 2, 28, 1, 30 + vor 28,1,1 + stvx 2, 0, 5 + blr + +.align 4 +.Lschedule_mangle_dec: + + + vsrb 1, 0, 8 + + + + vperm 2, 16, 16, 0 + + vperm 3, 17, 17, 1 + vxor 3, 3, 2 + vperm 3, 3, 9, 25 + + + vperm 2, 18, 18, 0 + vxor 2, 2, 3 + + vperm 3, 19, 19, 1 + vxor 3, 3, 2 + vperm 3, 3, 9, 25 + + + vperm 2, 20, 20, 0 + vxor 2, 2, 3 + + vperm 3, 21, 21, 1 + vxor 3, 3, 2 + + + vperm 2, 22, 22, 0 + vperm 3, 3, 9, 25 + + vperm 4, 23, 23, 1 + lvx 1, 8, 10 + vxor 2, 2, 3 + vxor 3, 4, 2 + + addi 5, 5, -16 + + vperm 3, 3, 3, 1 + addi 8, 8, -16 + andi. 8, 8, 0x30 + + + vperm 1, 3, 3, 29 + vsel 2, 28, 1, 30 + vor 28,1,1 + stvx 2, 0, 5 + blr +.long 0 +.byte 0,12,0x14,0,0,0,0,0 + +.globl vpaes_set_encrypt_key +.type vpaes_set_encrypt_key,@function +.section ".opd","aw" +.align 3 +vpaes_set_encrypt_key: +.quad .vpaes_set_encrypt_key,.TOC.@tocbase,0 +.previous +.align 5 +.vpaes_set_encrypt_key: + stdu 1,-256(1) + li 10,63 + li 11,79 + mflr 0 + mfspr 6,256 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + stw 6,252(1) + li 7, -1 + std 0, 272(1) + mtspr 256,7 + + srwi 9, 4, 5 + addi 9, 9, 6 + stw 9, 240(5) + + cmplw 1,4,4 + li 8, 0x30 + bl _vpaes_schedule_core + + ld 0, 272(1) + li 10,63 + li 11,79 + mtspr 256,6 + mtlr 0 + xor 3, 3, 3 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + addi 1,1,256 + blr +.long 0 +.byte 0,12,0x04,1,0x80,0,3,0 +.long 0 +.size .vpaes_set_encrypt_key,.-.vpaes_set_encrypt_key +.size vpaes_set_encrypt_key,.-.vpaes_set_encrypt_key + +.globl vpaes_set_decrypt_key +.type vpaes_set_decrypt_key,@function +.section ".opd","aw" +.align 3 +vpaes_set_decrypt_key: +.quad .vpaes_set_decrypt_key,.TOC.@tocbase,0 +.previous +.align 4 +.vpaes_set_decrypt_key: + stdu 1,-256(1) + li 10,63 + li 11,79 + mflr 0 + mfspr 6,256 + stvx 20,10,1 + addi 10,10,32 + stvx 21,11,1 + addi 11,11,32 + stvx 22,10,1 + addi 10,10,32 + stvx 23,11,1 + addi 11,11,32 + stvx 24,10,1 + addi 10,10,32 + stvx 25,11,1 + addi 11,11,32 + stvx 26,10,1 + addi 10,10,32 + stvx 27,11,1 + addi 11,11,32 + stvx 28,10,1 + addi 10,10,32 + stvx 29,11,1 + addi 11,11,32 + stvx 30,10,1 + stvx 31,11,1 + stw 6,252(1) + li 7, -1 + std 0, 272(1) + mtspr 256,7 + + srwi 9, 4, 5 + addi 9, 9, 6 + stw 9, 240(5) + + slwi 9, 9, 4 + add 5, 5, 9 + + cmplwi 1, 4, 0 + srwi 8, 4, 1 + andi. 8, 8, 32 + xori 8, 8, 32 + bl _vpaes_schedule_core + + ld 0, 272(1) + li 10,63 + li 11,79 + mtspr 256,6 + mtlr 0 + xor 3, 3, 3 + lvx 20,10,1 + addi 10,10,32 + lvx 21,11,1 + addi 11,11,32 + lvx 22,10,1 + addi 10,10,32 + lvx 23,11,1 + addi 11,11,32 + lvx 24,10,1 + addi 10,10,32 + lvx 25,11,1 + addi 11,11,32 + lvx 26,10,1 + addi 10,10,32 + lvx 27,11,1 + addi 11,11,32 + lvx 28,10,1 + addi 10,10,32 + lvx 29,11,1 + addi 11,11,32 + lvx 30,10,1 + lvx 31,11,1 + addi 1,1,256 + blr +.long 0 +.byte 0,12,0x04,1,0x80,0,3,0 +.long 0 +.size .vpaes_set_decrypt_key,.-.vpaes_set_decrypt_key +.size vpaes_set_decrypt_key,.-.vpaes_set_decrypt_key |