diff --git a/crypto/cipher_extra/e_aesgcmsiv.c b/crypto/cipher_extra/e_aesgcmsiv.c
index d717572..64febae 100644
--- a/crypto/cipher_extra/e_aesgcmsiv.c
+++ b/crypto/cipher_extra/e_aesgcmsiv.c
@@ -723,14 +723,6 @@
   }
 
   OPENSSL_memcpy(out_keys->auth_key, key_material, 16);
-  // Note the |ctr128_f| function uses a big-endian couner, while AES-GCM-SIV
-  // uses a little-endian counter. We ignore the return value and only use
-  // |block128_f|. This has a significant performance cost for the fallback
-  // bitsliced AES implementations (bsaes and aes_nohw).
-  //
-  // We currently do not consider AES-GCM-SIV to be performance-sensitive on
-  // client hardware. If this changes, we can write little-endian |ctr128_f|
-  // functions.
   aes_ctr_set_key(&out_keys->enc_key.ks, NULL, &out_keys->enc_block,
                   key_material + 16, gcm_siv_ctx->is_256 ? 32 : 16);
 }
diff --git a/crypto/fipsmodule/CMakeLists.txt b/crypto/fipsmodule/CMakeLists.txt
index a675fbd..3081a41 100644
--- a/crypto/fipsmodule/CMakeLists.txt
+++ b/crypto/fipsmodule/CMakeLists.txt
@@ -6,6 +6,7 @@
 
     aesni-gcm-x86_64.${ASM_EXT}
     aesni-x86_64.${ASM_EXT}
+    aes-x86_64.${ASM_EXT}
     ghash-ssse3-x86_64.${ASM_EXT}
     ghash-x86_64.${ASM_EXT}
     md5-x86_64.${ASM_EXT}
@@ -26,6 +27,7 @@
   set(
     BCM_ASM_SOURCES
 
+    aes-586.${ASM_EXT}
     aesni-x86.${ASM_EXT}
     bn-586.${ASM_EXT}
     co-586.${ASM_EXT}
@@ -44,6 +46,7 @@
   set(
     BCM_ASM_SOURCES
 
+    aes-armv4.${ASM_EXT}
     aesv8-armx.${ASM_EXT}
     armv4-mont.${ASM_EXT}
     bsaes-armv7.${ASM_EXT}
@@ -80,11 +83,14 @@
   )
 endif()
 
+perlasm(aes-586.${ASM_EXT} aes/asm/aes-586.pl)
+perlasm(aes-armv4.${ASM_EXT} aes/asm/aes-armv4.pl)
 perlasm(aesni-gcm-x86_64.${ASM_EXT} modes/asm/aesni-gcm-x86_64.pl)
 perlasm(aesni-x86_64.${ASM_EXT} aes/asm/aesni-x86_64.pl)
 perlasm(aesni-x86.${ASM_EXT} aes/asm/aesni-x86.pl)
 perlasm(aesp8-ppc.${ASM_EXT} aes/asm/aesp8-ppc.pl)
 perlasm(aesv8-armx.${ASM_EXT} aes/asm/aesv8-armx.pl)
+perlasm(aes-x86_64.${ASM_EXT} aes/asm/aes-x86_64.pl)
 perlasm(armv4-mont.${ASM_EXT} bn/asm/armv4-mont.pl)
 perlasm(armv8-mont.${ASM_EXT} bn/asm/armv8-mont.pl)
 perlasm(bn-586.${ASM_EXT} bn/asm/bn-586.pl)
diff --git a/crypto/fipsmodule/aes/aes.c b/crypto/fipsmodule/aes/aes.c
index f60281d..48d60ee 100644
--- a/crypto/fipsmodule/aes/aes.c
+++ b/crypto/fipsmodule/aes/aes.c
@@ -56,6 +56,758 @@
 #include "../modes/internal.h"
 
 
+#if defined(OPENSSL_NO_ASM) || \
+    (!defined(OPENSSL_X86) && !defined(OPENSSL_X86_64) && !defined(OPENSSL_ARM))
+
+// Te0[x] = S [x].[02, 01, 01, 03];
+// Te1[x] = S [x].[03, 02, 01, 01];
+// Te2[x] = S [x].[01, 03, 02, 01];
+// Te3[x] = S [x].[01, 01, 03, 02];
+//
+// Td0[x] = Si[x].[0e, 09, 0d, 0b];
+// Td1[x] = Si[x].[0b, 0e, 09, 0d];
+// Td2[x] = Si[x].[0d, 0b, 0e, 09];
+// Td3[x] = Si[x].[09, 0d, 0b, 0e];
+// Td4[x] = Si[x].[01];
+
+static const uint32_t Te0[256] = {
+    0xc66363a5U, 0xf87c7c84U, 0xee777799U, 0xf67b7b8dU, 0xfff2f20dU,
+    0xd66b6bbdU, 0xde6f6fb1U, 0x91c5c554U, 0x60303050U, 0x02010103U,
+    0xce6767a9U, 0x562b2b7dU, 0xe7fefe19U, 0xb5d7d762U, 0x4dababe6U,
+    0xec76769aU, 0x8fcaca45U, 0x1f82829dU, 0x89c9c940U, 0xfa7d7d87U,
+    0xeffafa15U, 0xb25959ebU, 0x8e4747c9U, 0xfbf0f00bU, 0x41adadecU,
+    0xb3d4d467U, 0x5fa2a2fdU, 0x45afafeaU, 0x239c9cbfU, 0x53a4a4f7U,
+    0xe4727296U, 0x9bc0c05bU, 0x75b7b7c2U, 0xe1fdfd1cU, 0x3d9393aeU,
+    0x4c26266aU, 0x6c36365aU, 0x7e3f3f41U, 0xf5f7f702U, 0x83cccc4fU,
+    0x6834345cU, 0x51a5a5f4U, 0xd1e5e534U, 0xf9f1f108U, 0xe2717193U,
+    0xabd8d873U, 0x62313153U, 0x2a15153fU, 0x0804040cU, 0x95c7c752U,
+    0x46232365U, 0x9dc3c35eU, 0x30181828U, 0x379696a1U, 0x0a05050fU,
+    0x2f9a9ab5U, 0x0e070709U, 0x24121236U, 0x1b80809bU, 0xdfe2e23dU,
+    0xcdebeb26U, 0x4e272769U, 0x7fb2b2cdU, 0xea75759fU, 0x1209091bU,
+    0x1d83839eU, 0x582c2c74U, 0x341a1a2eU, 0x361b1b2dU, 0xdc6e6eb2U,
+    0xb45a5aeeU, 0x5ba0a0fbU, 0xa45252f6U, 0x763b3b4dU, 0xb7d6d661U,
+    0x7db3b3ceU, 0x5229297bU, 0xdde3e33eU, 0x5e2f2f71U, 0x13848497U,
+    0xa65353f5U, 0xb9d1d168U, 0x00000000U, 0xc1eded2cU, 0x40202060U,
+    0xe3fcfc1fU, 0x79b1b1c8U, 0xb65b5bedU, 0xd46a6abeU, 0x8dcbcb46U,
+    0x67bebed9U, 0x7239394bU, 0x944a4adeU, 0x984c4cd4U, 0xb05858e8U,
+    0x85cfcf4aU, 0xbbd0d06bU, 0xc5efef2aU, 0x4faaaae5U, 0xedfbfb16U,
+    0x864343c5U, 0x9a4d4dd7U, 0x66333355U, 0x11858594U, 0x8a4545cfU,
+    0xe9f9f910U, 0x04020206U, 0xfe7f7f81U, 0xa05050f0U, 0x783c3c44U,
+    0x259f9fbaU, 0x4ba8a8e3U, 0xa25151f3U, 0x5da3a3feU, 0x804040c0U,
+    0x058f8f8aU, 0x3f9292adU, 0x219d9dbcU, 0x70383848U, 0xf1f5f504U,
+    0x63bcbcdfU, 0x77b6b6c1U, 0xafdada75U, 0x42212163U, 0x20101030U,
+    0xe5ffff1aU, 0xfdf3f30eU, 0xbfd2d26dU, 0x81cdcd4cU, 0x180c0c14U,
+    0x26131335U, 0xc3ecec2fU, 0xbe5f5fe1U, 0x359797a2U, 0x884444ccU,
+    0x2e171739U, 0x93c4c457U, 0x55a7a7f2U, 0xfc7e7e82U, 0x7a3d3d47U,
+    0xc86464acU, 0xba5d5de7U, 0x3219192bU, 0xe6737395U, 0xc06060a0U,
+    0x19818198U, 0x9e4f4fd1U, 0xa3dcdc7fU, 0x44222266U, 0x542a2a7eU,
+    0x3b9090abU, 0x0b888883U, 0x8c4646caU, 0xc7eeee29U, 0x6bb8b8d3U,
+    0x2814143cU, 0xa7dede79U, 0xbc5e5ee2U, 0x160b0b1dU, 0xaddbdb76U,
+    0xdbe0e03bU, 0x64323256U, 0x743a3a4eU, 0x140a0a1eU, 0x924949dbU,
+    0x0c06060aU, 0x4824246cU, 0xb85c5ce4U, 0x9fc2c25dU, 0xbdd3d36eU,
+    0x43acacefU, 0xc46262a6U, 0x399191a8U, 0x319595a4U, 0xd3e4e437U,
+    0xf279798bU, 0xd5e7e732U, 0x8bc8c843U, 0x6e373759U, 0xda6d6db7U,
+    0x018d8d8cU, 0xb1d5d564U, 0x9c4e4ed2U, 0x49a9a9e0U, 0xd86c6cb4U,
+    0xac5656faU, 0xf3f4f407U, 0xcfeaea25U, 0xca6565afU, 0xf47a7a8eU,
+    0x47aeaee9U, 0x10080818U, 0x6fbabad5U, 0xf0787888U, 0x4a25256fU,
+    0x5c2e2e72U, 0x381c1c24U, 0x57a6a6f1U, 0x73b4b4c7U, 0x97c6c651U,
+    0xcbe8e823U, 0xa1dddd7cU, 0xe874749cU, 0x3e1f1f21U, 0x964b4bddU,
+    0x61bdbddcU, 0x0d8b8b86U, 0x0f8a8a85U, 0xe0707090U, 0x7c3e3e42U,
+    0x71b5b5c4U, 0xcc6666aaU, 0x904848d8U, 0x06030305U, 0xf7f6f601U,
+    0x1c0e0e12U, 0xc26161a3U, 0x6a35355fU, 0xae5757f9U, 0x69b9b9d0U,
+    0x17868691U, 0x99c1c158U, 0x3a1d1d27U, 0x279e9eb9U, 0xd9e1e138U,
+    0xebf8f813U, 0x2b9898b3U, 0x22111133U, 0xd26969bbU, 0xa9d9d970U,
+    0x078e8e89U, 0x339494a7U, 0x2d9b9bb6U, 0x3c1e1e22U, 0x15878792U,
+    0xc9e9e920U, 0x87cece49U, 0xaa5555ffU, 0x50282878U, 0xa5dfdf7aU,
+    0x038c8c8fU, 0x59a1a1f8U, 0x09898980U, 0x1a0d0d17U, 0x65bfbfdaU,
+    0xd7e6e631U, 0x844242c6U, 0xd06868b8U, 0x824141c3U, 0x299999b0U,
+    0x5a2d2d77U, 0x1e0f0f11U, 0x7bb0b0cbU, 0xa85454fcU, 0x6dbbbbd6U,
+    0x2c16163aU, };
+
+static const uint32_t Te1[256] = {
+    0xa5c66363U, 0x84f87c7cU, 0x99ee7777U, 0x8df67b7bU, 0x0dfff2f2U,
+    0xbdd66b6bU, 0xb1de6f6fU, 0x5491c5c5U, 0x50603030U, 0x03020101U,
+    0xa9ce6767U, 0x7d562b2bU, 0x19e7fefeU, 0x62b5d7d7U, 0xe64dababU,
+    0x9aec7676U, 0x458fcacaU, 0x9d1f8282U, 0x4089c9c9U, 0x87fa7d7dU,
+    0x15effafaU, 0xebb25959U, 0xc98e4747U, 0x0bfbf0f0U, 0xec41adadU,
+    0x67b3d4d4U, 0xfd5fa2a2U, 0xea45afafU, 0xbf239c9cU, 0xf753a4a4U,
+    0x96e47272U, 0x5b9bc0c0U, 0xc275b7b7U, 0x1ce1fdfdU, 0xae3d9393U,
+    0x6a4c2626U, 0x5a6c3636U, 0x417e3f3fU, 0x02f5f7f7U, 0x4f83ccccU,
+    0x5c683434U, 0xf451a5a5U, 0x34d1e5e5U, 0x08f9f1f1U, 0x93e27171U,
+    0x73abd8d8U, 0x53623131U, 0x3f2a1515U, 0x0c080404U, 0x5295c7c7U,
+    0x65462323U, 0x5e9dc3c3U, 0x28301818U, 0xa1379696U, 0x0f0a0505U,
+    0xb52f9a9aU, 0x090e0707U, 0x36241212U, 0x9b1b8080U, 0x3ddfe2e2U,
+    0x26cdebebU, 0x694e2727U, 0xcd7fb2b2U, 0x9fea7575U, 0x1b120909U,
+    0x9e1d8383U, 0x74582c2cU, 0x2e341a1aU, 0x2d361b1bU, 0xb2dc6e6eU,
+    0xeeb45a5aU, 0xfb5ba0a0U, 0xf6a45252U, 0x4d763b3bU, 0x61b7d6d6U,
+    0xce7db3b3U, 0x7b522929U, 0x3edde3e3U, 0x715e2f2fU, 0x97138484U,
+    0xf5a65353U, 0x68b9d1d1U, 0x00000000U, 0x2cc1ededU, 0x60402020U,
+    0x1fe3fcfcU, 0xc879b1b1U, 0xedb65b5bU, 0xbed46a6aU, 0x468dcbcbU,
+    0xd967bebeU, 0x4b723939U, 0xde944a4aU, 0xd4984c4cU, 0xe8b05858U,
+    0x4a85cfcfU, 0x6bbbd0d0U, 0x2ac5efefU, 0xe54faaaaU, 0x16edfbfbU,
+    0xc5864343U, 0xd79a4d4dU, 0x55663333U, 0x94118585U, 0xcf8a4545U,
+    0x10e9f9f9U, 0x06040202U, 0x81fe7f7fU, 0xf0a05050U, 0x44783c3cU,
+    0xba259f9fU, 0xe34ba8a8U, 0xf3a25151U, 0xfe5da3a3U, 0xc0804040U,
+    0x8a058f8fU, 0xad3f9292U, 0xbc219d9dU, 0x48703838U, 0x04f1f5f5U,
+    0xdf63bcbcU, 0xc177b6b6U, 0x75afdadaU, 0x63422121U, 0x30201010U,
+    0x1ae5ffffU, 0x0efdf3f3U, 0x6dbfd2d2U, 0x4c81cdcdU, 0x14180c0cU,
+    0x35261313U, 0x2fc3ececU, 0xe1be5f5fU, 0xa2359797U, 0xcc884444U,
+    0x392e1717U, 0x5793c4c4U, 0xf255a7a7U, 0x82fc7e7eU, 0x477a3d3dU,
+    0xacc86464U, 0xe7ba5d5dU, 0x2b321919U, 0x95e67373U, 0xa0c06060U,
+    0x98198181U, 0xd19e4f4fU, 0x7fa3dcdcU, 0x66442222U, 0x7e542a2aU,
+    0xab3b9090U, 0x830b8888U, 0xca8c4646U, 0x29c7eeeeU, 0xd36bb8b8U,
+    0x3c281414U, 0x79a7dedeU, 0xe2bc5e5eU, 0x1d160b0bU, 0x76addbdbU,
+    0x3bdbe0e0U, 0x56643232U, 0x4e743a3aU, 0x1e140a0aU, 0xdb924949U,
+    0x0a0c0606U, 0x6c482424U, 0xe4b85c5cU, 0x5d9fc2c2U, 0x6ebdd3d3U,
+    0xef43acacU, 0xa6c46262U, 0xa8399191U, 0xa4319595U, 0x37d3e4e4U,
+    0x8bf27979U, 0x32d5e7e7U, 0x438bc8c8U, 0x596e3737U, 0xb7da6d6dU,
+    0x8c018d8dU, 0x64b1d5d5U, 0xd29c4e4eU, 0xe049a9a9U, 0xb4d86c6cU,
+    0xfaac5656U, 0x07f3f4f4U, 0x25cfeaeaU, 0xafca6565U, 0x8ef47a7aU,
+    0xe947aeaeU, 0x18100808U, 0xd56fbabaU, 0x88f07878U, 0x6f4a2525U,
+    0x725c2e2eU, 0x24381c1cU, 0xf157a6a6U, 0xc773b4b4U, 0x5197c6c6U,
+    0x23cbe8e8U, 0x7ca1ddddU, 0x9ce87474U, 0x213e1f1fU, 0xdd964b4bU,
+    0xdc61bdbdU, 0x860d8b8bU, 0x850f8a8aU, 0x90e07070U, 0x427c3e3eU,
+    0xc471b5b5U, 0xaacc6666U, 0xd8904848U, 0x05060303U, 0x01f7f6f6U,
+    0x121c0e0eU, 0xa3c26161U, 0x5f6a3535U, 0xf9ae5757U, 0xd069b9b9U,
+    0x91178686U, 0x5899c1c1U, 0x273a1d1dU, 0xb9279e9eU, 0x38d9e1e1U,
+    0x13ebf8f8U, 0xb32b9898U, 0x33221111U, 0xbbd26969U, 0x70a9d9d9U,
+    0x89078e8eU, 0xa7339494U, 0xb62d9b9bU, 0x223c1e1eU, 0x92158787U,
+    0x20c9e9e9U, 0x4987ceceU, 0xffaa5555U, 0x78502828U, 0x7aa5dfdfU,
+    0x8f038c8cU, 0xf859a1a1U, 0x80098989U, 0x171a0d0dU, 0xda65bfbfU,
+    0x31d7e6e6U, 0xc6844242U, 0xb8d06868U, 0xc3824141U, 0xb0299999U,
+    0x775a2d2dU, 0x111e0f0fU, 0xcb7bb0b0U, 0xfca85454U, 0xd66dbbbbU,
+    0x3a2c1616U, };
+
+static const uint32_t Te2[256] = {
+    0x63a5c663U, 0x7c84f87cU, 0x7799ee77U, 0x7b8df67bU, 0xf20dfff2U,
+    0x6bbdd66bU, 0x6fb1de6fU, 0xc55491c5U, 0x30506030U, 0x01030201U,
+    0x67a9ce67U, 0x2b7d562bU, 0xfe19e7feU, 0xd762b5d7U, 0xabe64dabU,
+    0x769aec76U, 0xca458fcaU, 0x829d1f82U, 0xc94089c9U, 0x7d87fa7dU,
+    0xfa15effaU, 0x59ebb259U, 0x47c98e47U, 0xf00bfbf0U, 0xadec41adU,
+    0xd467b3d4U, 0xa2fd5fa2U, 0xafea45afU, 0x9cbf239cU, 0xa4f753a4U,
+    0x7296e472U, 0xc05b9bc0U, 0xb7c275b7U, 0xfd1ce1fdU, 0x93ae3d93U,
+    0x266a4c26U, 0x365a6c36U, 0x3f417e3fU, 0xf702f5f7U, 0xcc4f83ccU,
+    0x345c6834U, 0xa5f451a5U, 0xe534d1e5U, 0xf108f9f1U, 0x7193e271U,
+    0xd873abd8U, 0x31536231U, 0x153f2a15U, 0x040c0804U, 0xc75295c7U,
+    0x23654623U, 0xc35e9dc3U, 0x18283018U, 0x96a13796U, 0x050f0a05U,
+    0x9ab52f9aU, 0x07090e07U, 0x12362412U, 0x809b1b80U, 0xe23ddfe2U,
+    0xeb26cdebU, 0x27694e27U, 0xb2cd7fb2U, 0x759fea75U, 0x091b1209U,
+    0x839e1d83U, 0x2c74582cU, 0x1a2e341aU, 0x1b2d361bU, 0x6eb2dc6eU,
+    0x5aeeb45aU, 0xa0fb5ba0U, 0x52f6a452U, 0x3b4d763bU, 0xd661b7d6U,
+    0xb3ce7db3U, 0x297b5229U, 0xe33edde3U, 0x2f715e2fU, 0x84971384U,
+    0x53f5a653U, 0xd168b9d1U, 0x00000000U, 0xed2cc1edU, 0x20604020U,
+    0xfc1fe3fcU, 0xb1c879b1U, 0x5bedb65bU, 0x6abed46aU, 0xcb468dcbU,
+    0xbed967beU, 0x394b7239U, 0x4ade944aU, 0x4cd4984cU, 0x58e8b058U,
+    0xcf4a85cfU, 0xd06bbbd0U, 0xef2ac5efU, 0xaae54faaU, 0xfb16edfbU,
+    0x43c58643U, 0x4dd79a4dU, 0x33556633U, 0x85941185U, 0x45cf8a45U,
+    0xf910e9f9U, 0x02060402U, 0x7f81fe7fU, 0x50f0a050U, 0x3c44783cU,
+    0x9fba259fU, 0xa8e34ba8U, 0x51f3a251U, 0xa3fe5da3U, 0x40c08040U,
+    0x8f8a058fU, 0x92ad3f92U, 0x9dbc219dU, 0x38487038U, 0xf504f1f5U,
+    0xbcdf63bcU, 0xb6c177b6U, 0xda75afdaU, 0x21634221U, 0x10302010U,
+    0xff1ae5ffU, 0xf30efdf3U, 0xd26dbfd2U, 0xcd4c81cdU, 0x0c14180cU,
+    0x13352613U, 0xec2fc3ecU, 0x5fe1be5fU, 0x97a23597U, 0x44cc8844U,
+    0x17392e17U, 0xc45793c4U, 0xa7f255a7U, 0x7e82fc7eU, 0x3d477a3dU,
+    0x64acc864U, 0x5de7ba5dU, 0x192b3219U, 0x7395e673U, 0x60a0c060U,
+    0x81981981U, 0x4fd19e4fU, 0xdc7fa3dcU, 0x22664422U, 0x2a7e542aU,
+    0x90ab3b90U, 0x88830b88U, 0x46ca8c46U, 0xee29c7eeU, 0xb8d36bb8U,
+    0x143c2814U, 0xde79a7deU, 0x5ee2bc5eU, 0x0b1d160bU, 0xdb76addbU,
+    0xe03bdbe0U, 0x32566432U, 0x3a4e743aU, 0x0a1e140aU, 0x49db9249U,
+    0x060a0c06U, 0x246c4824U, 0x5ce4b85cU, 0xc25d9fc2U, 0xd36ebdd3U,
+    0xacef43acU, 0x62a6c462U, 0x91a83991U, 0x95a43195U, 0xe437d3e4U,
+    0x798bf279U, 0xe732d5e7U, 0xc8438bc8U, 0x37596e37U, 0x6db7da6dU,
+    0x8d8c018dU, 0xd564b1d5U, 0x4ed29c4eU, 0xa9e049a9U, 0x6cb4d86cU,
+    0x56faac56U, 0xf407f3f4U, 0xea25cfeaU, 0x65afca65U, 0x7a8ef47aU,
+    0xaee947aeU, 0x08181008U, 0xbad56fbaU, 0x7888f078U, 0x256f4a25U,
+    0x2e725c2eU, 0x1c24381cU, 0xa6f157a6U, 0xb4c773b4U, 0xc65197c6U,
+    0xe823cbe8U, 0xdd7ca1ddU, 0x749ce874U, 0x1f213e1fU, 0x4bdd964bU,
+    0xbddc61bdU, 0x8b860d8bU, 0x8a850f8aU, 0x7090e070U, 0x3e427c3eU,
+    0xb5c471b5U, 0x66aacc66U, 0x48d89048U, 0x03050603U, 0xf601f7f6U,
+    0x0e121c0eU, 0x61a3c261U, 0x355f6a35U, 0x57f9ae57U, 0xb9d069b9U,
+    0x86911786U, 0xc15899c1U, 0x1d273a1dU, 0x9eb9279eU, 0xe138d9e1U,
+    0xf813ebf8U, 0x98b32b98U, 0x11332211U, 0x69bbd269U, 0xd970a9d9U,
+    0x8e89078eU, 0x94a73394U, 0x9bb62d9bU, 0x1e223c1eU, 0x87921587U,
+    0xe920c9e9U, 0xce4987ceU, 0x55ffaa55U, 0x28785028U, 0xdf7aa5dfU,
+    0x8c8f038cU, 0xa1f859a1U, 0x89800989U, 0x0d171a0dU, 0xbfda65bfU,
+    0xe631d7e6U, 0x42c68442U, 0x68b8d068U, 0x41c38241U, 0x99b02999U,
+    0x2d775a2dU, 0x0f111e0fU, 0xb0cb7bb0U, 0x54fca854U, 0xbbd66dbbU,
+    0x163a2c16U, };
+
+static const uint32_t Te3[256] = {
+    0x6363a5c6U, 0x7c7c84f8U, 0x777799eeU, 0x7b7b8df6U, 0xf2f20dffU,
+    0x6b6bbdd6U, 0x6f6fb1deU, 0xc5c55491U, 0x30305060U, 0x01010302U,
+    0x6767a9ceU, 0x2b2b7d56U, 0xfefe19e7U, 0xd7d762b5U, 0xababe64dU,
+    0x76769aecU, 0xcaca458fU, 0x82829d1fU, 0xc9c94089U, 0x7d7d87faU,
+    0xfafa15efU, 0x5959ebb2U, 0x4747c98eU, 0xf0f00bfbU, 0xadadec41U,
+    0xd4d467b3U, 0xa2a2fd5fU, 0xafafea45U, 0x9c9cbf23U, 0xa4a4f753U,
+    0x727296e4U, 0xc0c05b9bU, 0xb7b7c275U, 0xfdfd1ce1U, 0x9393ae3dU,
+    0x26266a4cU, 0x36365a6cU, 0x3f3f417eU, 0xf7f702f5U, 0xcccc4f83U,
+    0x34345c68U, 0xa5a5f451U, 0xe5e534d1U, 0xf1f108f9U, 0x717193e2U,
+    0xd8d873abU, 0x31315362U, 0x15153f2aU, 0x04040c08U, 0xc7c75295U,
+    0x23236546U, 0xc3c35e9dU, 0x18182830U, 0x9696a137U, 0x05050f0aU,
+    0x9a9ab52fU, 0x0707090eU, 0x12123624U, 0x80809b1bU, 0xe2e23ddfU,
+    0xebeb26cdU, 0x2727694eU, 0xb2b2cd7fU, 0x75759feaU, 0x09091b12U,
+    0x83839e1dU, 0x2c2c7458U, 0x1a1a2e34U, 0x1b1b2d36U, 0x6e6eb2dcU,
+    0x5a5aeeb4U, 0xa0a0fb5bU, 0x5252f6a4U, 0x3b3b4d76U, 0xd6d661b7U,
+    0xb3b3ce7dU, 0x29297b52U, 0xe3e33eddU, 0x2f2f715eU, 0x84849713U,
+    0x5353f5a6U, 0xd1d168b9U, 0x00000000U, 0xeded2cc1U, 0x20206040U,
+    0xfcfc1fe3U, 0xb1b1c879U, 0x5b5bedb6U, 0x6a6abed4U, 0xcbcb468dU,
+    0xbebed967U, 0x39394b72U, 0x4a4ade94U, 0x4c4cd498U, 0x5858e8b0U,
+    0xcfcf4a85U, 0xd0d06bbbU, 0xefef2ac5U, 0xaaaae54fU, 0xfbfb16edU,
+    0x4343c586U, 0x4d4dd79aU, 0x33335566U, 0x85859411U, 0x4545cf8aU,
+    0xf9f910e9U, 0x02020604U, 0x7f7f81feU, 0x5050f0a0U, 0x3c3c4478U,
+    0x9f9fba25U, 0xa8a8e34bU, 0x5151f3a2U, 0xa3a3fe5dU, 0x4040c080U,
+    0x8f8f8a05U, 0x9292ad3fU, 0x9d9dbc21U, 0x38384870U, 0xf5f504f1U,
+    0xbcbcdf63U, 0xb6b6c177U, 0xdada75afU, 0x21216342U, 0x10103020U,
+    0xffff1ae5U, 0xf3f30efdU, 0xd2d26dbfU, 0xcdcd4c81U, 0x0c0c1418U,
+    0x13133526U, 0xecec2fc3U, 0x5f5fe1beU, 0x9797a235U, 0x4444cc88U,
+    0x1717392eU, 0xc4c45793U, 0xa7a7f255U, 0x7e7e82fcU, 0x3d3d477aU,
+    0x6464acc8U, 0x5d5de7baU, 0x19192b32U, 0x737395e6U, 0x6060a0c0U,
+    0x81819819U, 0x4f4fd19eU, 0xdcdc7fa3U, 0x22226644U, 0x2a2a7e54U,
+    0x9090ab3bU, 0x8888830bU, 0x4646ca8cU, 0xeeee29c7U, 0xb8b8d36bU,
+    0x14143c28U, 0xdede79a7U, 0x5e5ee2bcU, 0x0b0b1d16U, 0xdbdb76adU,
+    0xe0e03bdbU, 0x32325664U, 0x3a3a4e74U, 0x0a0a1e14U, 0x4949db92U,
+    0x06060a0cU, 0x24246c48U, 0x5c5ce4b8U, 0xc2c25d9fU, 0xd3d36ebdU,
+    0xacacef43U, 0x6262a6c4U, 0x9191a839U, 0x9595a431U, 0xe4e437d3U,
+    0x79798bf2U, 0xe7e732d5U, 0xc8c8438bU, 0x3737596eU, 0x6d6db7daU,
+    0x8d8d8c01U, 0xd5d564b1U, 0x4e4ed29cU, 0xa9a9e049U, 0x6c6cb4d8U,
+    0x5656faacU, 0xf4f407f3U, 0xeaea25cfU, 0x6565afcaU, 0x7a7a8ef4U,
+    0xaeaee947U, 0x08081810U, 0xbabad56fU, 0x787888f0U, 0x25256f4aU,
+    0x2e2e725cU, 0x1c1c2438U, 0xa6a6f157U, 0xb4b4c773U, 0xc6c65197U,
+    0xe8e823cbU, 0xdddd7ca1U, 0x74749ce8U, 0x1f1f213eU, 0x4b4bdd96U,
+    0xbdbddc61U, 0x8b8b860dU, 0x8a8a850fU, 0x707090e0U, 0x3e3e427cU,
+    0xb5b5c471U, 0x6666aaccU, 0x4848d890U, 0x03030506U, 0xf6f601f7U,
+    0x0e0e121cU, 0x6161a3c2U, 0x35355f6aU, 0x5757f9aeU, 0xb9b9d069U,
+    0x86869117U, 0xc1c15899U, 0x1d1d273aU, 0x9e9eb927U, 0xe1e138d9U,
+    0xf8f813ebU, 0x9898b32bU, 0x11113322U, 0x6969bbd2U, 0xd9d970a9U,
+    0x8e8e8907U, 0x9494a733U, 0x9b9bb62dU, 0x1e1e223cU, 0x87879215U,
+    0xe9e920c9U, 0xcece4987U, 0x5555ffaaU, 0x28287850U, 0xdfdf7aa5U,
+    0x8c8c8f03U, 0xa1a1f859U, 0x89898009U, 0x0d0d171aU, 0xbfbfda65U,
+    0xe6e631d7U, 0x4242c684U, 0x6868b8d0U, 0x4141c382U, 0x9999b029U,
+    0x2d2d775aU, 0x0f0f111eU, 0xb0b0cb7bU, 0x5454fca8U, 0xbbbbd66dU,
+    0x16163a2cU, };
+
+static const uint32_t Td0[256] = {
+    0x51f4a750U, 0x7e416553U, 0x1a17a4c3U, 0x3a275e96U, 0x3bab6bcbU,
+    0x1f9d45f1U, 0xacfa58abU, 0x4be30393U, 0x2030fa55U, 0xad766df6U,
+    0x88cc7691U, 0xf5024c25U, 0x4fe5d7fcU, 0xc52acbd7U, 0x26354480U,
+    0xb562a38fU, 0xdeb15a49U, 0x25ba1b67U, 0x45ea0e98U, 0x5dfec0e1U,
+    0xc32f7502U, 0x814cf012U, 0x8d4697a3U, 0x6bd3f9c6U, 0x038f5fe7U,
+    0x15929c95U, 0xbf6d7aebU, 0x955259daU, 0xd4be832dU, 0x587421d3U,
+    0x49e06929U, 0x8ec9c844U, 0x75c2896aU, 0xf48e7978U, 0x99583e6bU,
+    0x27b971ddU, 0xbee14fb6U, 0xf088ad17U, 0xc920ac66U, 0x7dce3ab4U,
+    0x63df4a18U, 0xe51a3182U, 0x97513360U, 0x62537f45U, 0xb16477e0U,
+    0xbb6bae84U, 0xfe81a01cU, 0xf9082b94U, 0x70486858U, 0x8f45fd19U,
+    0x94de6c87U, 0x527bf8b7U, 0xab73d323U, 0x724b02e2U, 0xe31f8f57U,
+    0x6655ab2aU, 0xb2eb2807U, 0x2fb5c203U, 0x86c57b9aU, 0xd33708a5U,
+    0x302887f2U, 0x23bfa5b2U, 0x02036abaU, 0xed16825cU, 0x8acf1c2bU,
+    0xa779b492U, 0xf307f2f0U, 0x4e69e2a1U, 0x65daf4cdU, 0x0605bed5U,
+    0xd134621fU, 0xc4a6fe8aU, 0x342e539dU, 0xa2f355a0U, 0x058ae132U,
+    0xa4f6eb75U, 0x0b83ec39U, 0x4060efaaU, 0x5e719f06U, 0xbd6e1051U,
+    0x3e218af9U, 0x96dd063dU, 0xdd3e05aeU, 0x4de6bd46U, 0x91548db5U,
+    0x71c45d05U, 0x0406d46fU, 0x605015ffU, 0x1998fb24U, 0xd6bde997U,
+    0x894043ccU, 0x67d99e77U, 0xb0e842bdU, 0x07898b88U, 0xe7195b38U,
+    0x79c8eedbU, 0xa17c0a47U, 0x7c420fe9U, 0xf8841ec9U, 0x00000000U,
+    0x09808683U, 0x322bed48U, 0x1e1170acU, 0x6c5a724eU, 0xfd0efffbU,
+    0x0f853856U, 0x3daed51eU, 0x362d3927U, 0x0a0fd964U, 0x685ca621U,
+    0x9b5b54d1U, 0x24362e3aU, 0x0c0a67b1U, 0x9357e70fU, 0xb4ee96d2U,
+    0x1b9b919eU, 0x80c0c54fU, 0x61dc20a2U, 0x5a774b69U, 0x1c121a16U,
+    0xe293ba0aU, 0xc0a02ae5U, 0x3c22e043U, 0x121b171dU, 0x0e090d0bU,
+    0xf28bc7adU, 0x2db6a8b9U, 0x141ea9c8U, 0x57f11985U, 0xaf75074cU,
+    0xee99ddbbU, 0xa37f60fdU, 0xf701269fU, 0x5c72f5bcU, 0x44663bc5U,
+    0x5bfb7e34U, 0x8b432976U, 0xcb23c6dcU, 0xb6edfc68U, 0xb8e4f163U,
+    0xd731dccaU, 0x42638510U, 0x13972240U, 0x84c61120U, 0x854a247dU,
+    0xd2bb3df8U, 0xaef93211U, 0xc729a16dU, 0x1d9e2f4bU, 0xdcb230f3U,
+    0x0d8652ecU, 0x77c1e3d0U, 0x2bb3166cU, 0xa970b999U, 0x119448faU,
+    0x47e96422U, 0xa8fc8cc4U, 0xa0f03f1aU, 0x567d2cd8U, 0x223390efU,
+    0x87494ec7U, 0xd938d1c1U, 0x8ccaa2feU, 0x98d40b36U, 0xa6f581cfU,
+    0xa57ade28U, 0xdab78e26U, 0x3fadbfa4U, 0x2c3a9de4U, 0x5078920dU,
+    0x6a5fcc9bU, 0x547e4662U, 0xf68d13c2U, 0x90d8b8e8U, 0x2e39f75eU,
+    0x82c3aff5U, 0x9f5d80beU, 0x69d0937cU, 0x6fd52da9U, 0xcf2512b3U,
+    0xc8ac993bU, 0x10187da7U, 0xe89c636eU, 0xdb3bbb7bU, 0xcd267809U,
+    0x6e5918f4U, 0xec9ab701U, 0x834f9aa8U, 0xe6956e65U, 0xaaffe67eU,
+    0x21bccf08U, 0xef15e8e6U, 0xbae79bd9U, 0x4a6f36ceU, 0xea9f09d4U,
+    0x29b07cd6U, 0x31a4b2afU, 0x2a3f2331U, 0xc6a59430U, 0x35a266c0U,
+    0x744ebc37U, 0xfc82caa6U, 0xe090d0b0U, 0x33a7d815U, 0xf104984aU,
+    0x41ecdaf7U, 0x7fcd500eU, 0x1791f62fU, 0x764dd68dU, 0x43efb04dU,
+    0xccaa4d54U, 0xe49604dfU, 0x9ed1b5e3U, 0x4c6a881bU, 0xc12c1fb8U,
+    0x4665517fU, 0x9d5eea04U, 0x018c355dU, 0xfa877473U, 0xfb0b412eU,
+    0xb3671d5aU, 0x92dbd252U, 0xe9105633U, 0x6dd64713U, 0x9ad7618cU,
+    0x37a10c7aU, 0x59f8148eU, 0xeb133c89U, 0xcea927eeU, 0xb761c935U,
+    0xe11ce5edU, 0x7a47b13cU, 0x9cd2df59U, 0x55f2733fU, 0x1814ce79U,
+    0x73c737bfU, 0x53f7cdeaU, 0x5ffdaa5bU, 0xdf3d6f14U, 0x7844db86U,
+    0xcaaff381U, 0xb968c43eU, 0x3824342cU, 0xc2a3405fU, 0x161dc372U,
+    0xbce2250cU, 0x283c498bU, 0xff0d9541U, 0x39a80171U, 0x080cb3deU,
+    0xd8b4e49cU, 0x6456c190U, 0x7bcb8461U, 0xd532b670U, 0x486c5c74U,
+    0xd0b85742U, };
+
+static const uint32_t Td1[256] = {
+    0x5051f4a7U, 0x537e4165U, 0xc31a17a4U, 0x963a275eU, 0xcb3bab6bU,
+    0xf11f9d45U, 0xabacfa58U, 0x934be303U, 0x552030faU, 0xf6ad766dU,
+    0x9188cc76U, 0x25f5024cU, 0xfc4fe5d7U, 0xd7c52acbU, 0x80263544U,
+    0x8fb562a3U, 0x49deb15aU, 0x6725ba1bU, 0x9845ea0eU, 0xe15dfec0U,
+    0x02c32f75U, 0x12814cf0U, 0xa38d4697U, 0xc66bd3f9U, 0xe7038f5fU,
+    0x9515929cU, 0xebbf6d7aU, 0xda955259U, 0x2dd4be83U, 0xd3587421U,
+    0x2949e069U, 0x448ec9c8U, 0x6a75c289U, 0x78f48e79U, 0x6b99583eU,
+    0xdd27b971U, 0xb6bee14fU, 0x17f088adU, 0x66c920acU, 0xb47dce3aU,
+    0x1863df4aU, 0x82e51a31U, 0x60975133U, 0x4562537fU, 0xe0b16477U,
+    0x84bb6baeU, 0x1cfe81a0U, 0x94f9082bU, 0x58704868U, 0x198f45fdU,
+    0x8794de6cU, 0xb7527bf8U, 0x23ab73d3U, 0xe2724b02U, 0x57e31f8fU,
+    0x2a6655abU, 0x07b2eb28U, 0x032fb5c2U, 0x9a86c57bU, 0xa5d33708U,
+    0xf2302887U, 0xb223bfa5U, 0xba02036aU, 0x5ced1682U, 0x2b8acf1cU,
+    0x92a779b4U, 0xf0f307f2U, 0xa14e69e2U, 0xcd65daf4U, 0xd50605beU,
+    0x1fd13462U, 0x8ac4a6feU, 0x9d342e53U, 0xa0a2f355U, 0x32058ae1U,
+    0x75a4f6ebU, 0x390b83ecU, 0xaa4060efU, 0x065e719fU, 0x51bd6e10U,
+    0xf93e218aU, 0x3d96dd06U, 0xaedd3e05U, 0x464de6bdU, 0xb591548dU,
+    0x0571c45dU, 0x6f0406d4U, 0xff605015U, 0x241998fbU, 0x97d6bde9U,
+    0xcc894043U, 0x7767d99eU, 0xbdb0e842U, 0x8807898bU, 0x38e7195bU,
+    0xdb79c8eeU, 0x47a17c0aU, 0xe97c420fU, 0xc9f8841eU, 0x00000000U,
+    0x83098086U, 0x48322bedU, 0xac1e1170U, 0x4e6c5a72U, 0xfbfd0effU,
+    0x560f8538U, 0x1e3daed5U, 0x27362d39U, 0x640a0fd9U, 0x21685ca6U,
+    0xd19b5b54U, 0x3a24362eU, 0xb10c0a67U, 0x0f9357e7U, 0xd2b4ee96U,
+    0x9e1b9b91U, 0x4f80c0c5U, 0xa261dc20U, 0x695a774bU, 0x161c121aU,
+    0x0ae293baU, 0xe5c0a02aU, 0x433c22e0U, 0x1d121b17U, 0x0b0e090dU,
+    0xadf28bc7U, 0xb92db6a8U, 0xc8141ea9U, 0x8557f119U, 0x4caf7507U,
+    0xbbee99ddU, 0xfda37f60U, 0x9ff70126U, 0xbc5c72f5U, 0xc544663bU,
+    0x345bfb7eU, 0x768b4329U, 0xdccb23c6U, 0x68b6edfcU, 0x63b8e4f1U,
+    0xcad731dcU, 0x10426385U, 0x40139722U, 0x2084c611U, 0x7d854a24U,
+    0xf8d2bb3dU, 0x11aef932U, 0x6dc729a1U, 0x4b1d9e2fU, 0xf3dcb230U,
+    0xec0d8652U, 0xd077c1e3U, 0x6c2bb316U, 0x99a970b9U, 0xfa119448U,
+    0x2247e964U, 0xc4a8fc8cU, 0x1aa0f03fU, 0xd8567d2cU, 0xef223390U,
+    0xc787494eU, 0xc1d938d1U, 0xfe8ccaa2U, 0x3698d40bU, 0xcfa6f581U,
+    0x28a57adeU, 0x26dab78eU, 0xa43fadbfU, 0xe42c3a9dU, 0x0d507892U,
+    0x9b6a5fccU, 0x62547e46U, 0xc2f68d13U, 0xe890d8b8U, 0x5e2e39f7U,
+    0xf582c3afU, 0xbe9f5d80U, 0x7c69d093U, 0xa96fd52dU, 0xb3cf2512U,
+    0x3bc8ac99U, 0xa710187dU, 0x6ee89c63U, 0x7bdb3bbbU, 0x09cd2678U,
+    0xf46e5918U, 0x01ec9ab7U, 0xa8834f9aU, 0x65e6956eU, 0x7eaaffe6U,
+    0x0821bccfU, 0xe6ef15e8U, 0xd9bae79bU, 0xce4a6f36U, 0xd4ea9f09U,
+    0xd629b07cU, 0xaf31a4b2U, 0x312a3f23U, 0x30c6a594U, 0xc035a266U,
+    0x37744ebcU, 0xa6fc82caU, 0xb0e090d0U, 0x1533a7d8U, 0x4af10498U,
+    0xf741ecdaU, 0x0e7fcd50U, 0x2f1791f6U, 0x8d764dd6U, 0x4d43efb0U,
+    0x54ccaa4dU, 0xdfe49604U, 0xe39ed1b5U, 0x1b4c6a88U, 0xb8c12c1fU,
+    0x7f466551U, 0x049d5eeaU, 0x5d018c35U, 0x73fa8774U, 0x2efb0b41U,
+    0x5ab3671dU, 0x5292dbd2U, 0x33e91056U, 0x136dd647U, 0x8c9ad761U,
+    0x7a37a10cU, 0x8e59f814U, 0x89eb133cU, 0xeecea927U, 0x35b761c9U,
+    0xede11ce5U, 0x3c7a47b1U, 0x599cd2dfU, 0x3f55f273U, 0x791814ceU,
+    0xbf73c737U, 0xea53f7cdU, 0x5b5ffdaaU, 0x14df3d6fU, 0x867844dbU,
+    0x81caaff3U, 0x3eb968c4U, 0x2c382434U, 0x5fc2a340U, 0x72161dc3U,
+    0x0cbce225U, 0x8b283c49U, 0x41ff0d95U, 0x7139a801U, 0xde080cb3U,
+    0x9cd8b4e4U, 0x906456c1U, 0x617bcb84U, 0x70d532b6U, 0x74486c5cU,
+    0x42d0b857U, };
+
+static const uint32_t Td2[256] = {
+    0xa75051f4U, 0x65537e41U, 0xa4c31a17U, 0x5e963a27U, 0x6bcb3babU,
+    0x45f11f9dU, 0x58abacfaU, 0x03934be3U, 0xfa552030U, 0x6df6ad76U,
+    0x769188ccU, 0x4c25f502U, 0xd7fc4fe5U, 0xcbd7c52aU, 0x44802635U,
+    0xa38fb562U, 0x5a49deb1U, 0x1b6725baU, 0x0e9845eaU, 0xc0e15dfeU,
+    0x7502c32fU, 0xf012814cU, 0x97a38d46U, 0xf9c66bd3U, 0x5fe7038fU,
+    0x9c951592U, 0x7aebbf6dU, 0x59da9552U, 0x832dd4beU, 0x21d35874U,
+    0x692949e0U, 0xc8448ec9U, 0x896a75c2U, 0x7978f48eU, 0x3e6b9958U,
+    0x71dd27b9U, 0x4fb6bee1U, 0xad17f088U, 0xac66c920U, 0x3ab47dceU,
+    0x4a1863dfU, 0x3182e51aU, 0x33609751U, 0x7f456253U, 0x77e0b164U,
+    0xae84bb6bU, 0xa01cfe81U, 0x2b94f908U, 0x68587048U, 0xfd198f45U,
+    0x6c8794deU, 0xf8b7527bU, 0xd323ab73U, 0x02e2724bU, 0x8f57e31fU,
+    0xab2a6655U, 0x2807b2ebU, 0xc2032fb5U, 0x7b9a86c5U, 0x08a5d337U,
+    0x87f23028U, 0xa5b223bfU, 0x6aba0203U, 0x825ced16U, 0x1c2b8acfU,
+    0xb492a779U, 0xf2f0f307U, 0xe2a14e69U, 0xf4cd65daU, 0xbed50605U,
+    0x621fd134U, 0xfe8ac4a6U, 0x539d342eU, 0x55a0a2f3U, 0xe132058aU,
+    0xeb75a4f6U, 0xec390b83U, 0xefaa4060U, 0x9f065e71U, 0x1051bd6eU,
+    0x8af93e21U, 0x063d96ddU, 0x05aedd3eU, 0xbd464de6U, 0x8db59154U,
+    0x5d0571c4U, 0xd46f0406U, 0x15ff6050U, 0xfb241998U, 0xe997d6bdU,
+    0x43cc8940U, 0x9e7767d9U, 0x42bdb0e8U, 0x8b880789U, 0x5b38e719U,
+    0xeedb79c8U, 0x0a47a17cU, 0x0fe97c42U, 0x1ec9f884U, 0x00000000U,
+    0x86830980U, 0xed48322bU, 0x70ac1e11U, 0x724e6c5aU, 0xfffbfd0eU,
+    0x38560f85U, 0xd51e3daeU, 0x3927362dU, 0xd9640a0fU, 0xa621685cU,
+    0x54d19b5bU, 0x2e3a2436U, 0x67b10c0aU, 0xe70f9357U, 0x96d2b4eeU,
+    0x919e1b9bU, 0xc54f80c0U, 0x20a261dcU, 0x4b695a77U, 0x1a161c12U,
+    0xba0ae293U, 0x2ae5c0a0U, 0xe0433c22U, 0x171d121bU, 0x0d0b0e09U,
+    0xc7adf28bU, 0xa8b92db6U, 0xa9c8141eU, 0x198557f1U, 0x074caf75U,
+    0xddbbee99U, 0x60fda37fU, 0x269ff701U, 0xf5bc5c72U, 0x3bc54466U,
+    0x7e345bfbU, 0x29768b43U, 0xc6dccb23U, 0xfc68b6edU, 0xf163b8e4U,
+    0xdccad731U, 0x85104263U, 0x22401397U, 0x112084c6U, 0x247d854aU,
+    0x3df8d2bbU, 0x3211aef9U, 0xa16dc729U, 0x2f4b1d9eU, 0x30f3dcb2U,
+    0x52ec0d86U, 0xe3d077c1U, 0x166c2bb3U, 0xb999a970U, 0x48fa1194U,
+    0x642247e9U, 0x8cc4a8fcU, 0x3f1aa0f0U, 0x2cd8567dU, 0x90ef2233U,
+    0x4ec78749U, 0xd1c1d938U, 0xa2fe8ccaU, 0x0b3698d4U, 0x81cfa6f5U,
+    0xde28a57aU, 0x8e26dab7U, 0xbfa43fadU, 0x9de42c3aU, 0x920d5078U,
+    0xcc9b6a5fU, 0x4662547eU, 0x13c2f68dU, 0xb8e890d8U, 0xf75e2e39U,
+    0xaff582c3U, 0x80be9f5dU, 0x937c69d0U, 0x2da96fd5U, 0x12b3cf25U,
+    0x993bc8acU, 0x7da71018U, 0x636ee89cU, 0xbb7bdb3bU, 0x7809cd26U,
+    0x18f46e59U, 0xb701ec9aU, 0x9aa8834fU, 0x6e65e695U, 0xe67eaaffU,
+    0xcf0821bcU, 0xe8e6ef15U, 0x9bd9bae7U, 0x36ce4a6fU, 0x09d4ea9fU,
+    0x7cd629b0U, 0xb2af31a4U, 0x23312a3fU, 0x9430c6a5U, 0x66c035a2U,
+    0xbc37744eU, 0xcaa6fc82U, 0xd0b0e090U, 0xd81533a7U, 0x984af104U,
+    0xdaf741ecU, 0x500e7fcdU, 0xf62f1791U, 0xd68d764dU, 0xb04d43efU,
+    0x4d54ccaaU, 0x04dfe496U, 0xb5e39ed1U, 0x881b4c6aU, 0x1fb8c12cU,
+    0x517f4665U, 0xea049d5eU, 0x355d018cU, 0x7473fa87U, 0x412efb0bU,
+    0x1d5ab367U, 0xd25292dbU, 0x5633e910U, 0x47136dd6U, 0x618c9ad7U,
+    0x0c7a37a1U, 0x148e59f8U, 0x3c89eb13U, 0x27eecea9U, 0xc935b761U,
+    0xe5ede11cU, 0xb13c7a47U, 0xdf599cd2U, 0x733f55f2U, 0xce791814U,
+    0x37bf73c7U, 0xcdea53f7U, 0xaa5b5ffdU, 0x6f14df3dU, 0xdb867844U,
+    0xf381caafU, 0xc43eb968U, 0x342c3824U, 0x405fc2a3U, 0xc372161dU,
+    0x250cbce2U, 0x498b283cU, 0x9541ff0dU, 0x017139a8U, 0xb3de080cU,
+    0xe49cd8b4U, 0xc1906456U, 0x84617bcbU, 0xb670d532U, 0x5c74486cU,
+    0x5742d0b8U, };
+
+static const uint32_t Td3[256] = {
+    0xf4a75051U, 0x4165537eU, 0x17a4c31aU, 0x275e963aU, 0xab6bcb3bU,
+    0x9d45f11fU, 0xfa58abacU, 0xe303934bU, 0x30fa5520U, 0x766df6adU,
+    0xcc769188U, 0x024c25f5U, 0xe5d7fc4fU, 0x2acbd7c5U, 0x35448026U,
+    0x62a38fb5U, 0xb15a49deU, 0xba1b6725U, 0xea0e9845U, 0xfec0e15dU,
+    0x2f7502c3U, 0x4cf01281U, 0x4697a38dU, 0xd3f9c66bU, 0x8f5fe703U,
+    0x929c9515U, 0x6d7aebbfU, 0x5259da95U, 0xbe832dd4U, 0x7421d358U,
+    0xe0692949U, 0xc9c8448eU, 0xc2896a75U, 0x8e7978f4U, 0x583e6b99U,
+    0xb971dd27U, 0xe14fb6beU, 0x88ad17f0U, 0x20ac66c9U, 0xce3ab47dU,
+    0xdf4a1863U, 0x1a3182e5U, 0x51336097U, 0x537f4562U, 0x6477e0b1U,
+    0x6bae84bbU, 0x81a01cfeU, 0x082b94f9U, 0x48685870U, 0x45fd198fU,
+    0xde6c8794U, 0x7bf8b752U, 0x73d323abU, 0x4b02e272U, 0x1f8f57e3U,
+    0x55ab2a66U, 0xeb2807b2U, 0xb5c2032fU, 0xc57b9a86U, 0x3708a5d3U,
+    0x2887f230U, 0xbfa5b223U, 0x036aba02U, 0x16825cedU, 0xcf1c2b8aU,
+    0x79b492a7U, 0x07f2f0f3U, 0x69e2a14eU, 0xdaf4cd65U, 0x05bed506U,
+    0x34621fd1U, 0xa6fe8ac4U, 0x2e539d34U, 0xf355a0a2U, 0x8ae13205U,
+    0xf6eb75a4U, 0x83ec390bU, 0x60efaa40U, 0x719f065eU, 0x6e1051bdU,
+    0x218af93eU, 0xdd063d96U, 0x3e05aeddU, 0xe6bd464dU, 0x548db591U,
+    0xc45d0571U, 0x06d46f04U, 0x5015ff60U, 0x98fb2419U, 0xbde997d6U,
+    0x4043cc89U, 0xd99e7767U, 0xe842bdb0U, 0x898b8807U, 0x195b38e7U,
+    0xc8eedb79U, 0x7c0a47a1U, 0x420fe97cU, 0x841ec9f8U, 0x00000000U,
+    0x80868309U, 0x2bed4832U, 0x1170ac1eU, 0x5a724e6cU, 0x0efffbfdU,
+    0x8538560fU, 0xaed51e3dU, 0x2d392736U, 0x0fd9640aU, 0x5ca62168U,
+    0x5b54d19bU, 0x362e3a24U, 0x0a67b10cU, 0x57e70f93U, 0xee96d2b4U,
+    0x9b919e1bU, 0xc0c54f80U, 0xdc20a261U, 0x774b695aU, 0x121a161cU,
+    0x93ba0ae2U, 0xa02ae5c0U, 0x22e0433cU, 0x1b171d12U, 0x090d0b0eU,
+    0x8bc7adf2U, 0xb6a8b92dU, 0x1ea9c814U, 0xf1198557U, 0x75074cafU,
+    0x99ddbbeeU, 0x7f60fda3U, 0x01269ff7U, 0x72f5bc5cU, 0x663bc544U,
+    0xfb7e345bU, 0x4329768bU, 0x23c6dccbU, 0xedfc68b6U, 0xe4f163b8U,
+    0x31dccad7U, 0x63851042U, 0x97224013U, 0xc6112084U, 0x4a247d85U,
+    0xbb3df8d2U, 0xf93211aeU, 0x29a16dc7U, 0x9e2f4b1dU, 0xb230f3dcU,
+    0x8652ec0dU, 0xc1e3d077U, 0xb3166c2bU, 0x70b999a9U, 0x9448fa11U,
+    0xe9642247U, 0xfc8cc4a8U, 0xf03f1aa0U, 0x7d2cd856U, 0x3390ef22U,
+    0x494ec787U, 0x38d1c1d9U, 0xcaa2fe8cU, 0xd40b3698U, 0xf581cfa6U,
+    0x7ade28a5U, 0xb78e26daU, 0xadbfa43fU, 0x3a9de42cU, 0x78920d50U,
+    0x5fcc9b6aU, 0x7e466254U, 0x8d13c2f6U, 0xd8b8e890U, 0x39f75e2eU,
+    0xc3aff582U, 0x5d80be9fU, 0xd0937c69U, 0xd52da96fU, 0x2512b3cfU,
+    0xac993bc8U, 0x187da710U, 0x9c636ee8U, 0x3bbb7bdbU, 0x267809cdU,
+    0x5918f46eU, 0x9ab701ecU, 0x4f9aa883U, 0x956e65e6U, 0xffe67eaaU,
+    0xbccf0821U, 0x15e8e6efU, 0xe79bd9baU, 0x6f36ce4aU, 0x9f09d4eaU,
+    0xb07cd629U, 0xa4b2af31U, 0x3f23312aU, 0xa59430c6U, 0xa266c035U,
+    0x4ebc3774U, 0x82caa6fcU, 0x90d0b0e0U, 0xa7d81533U, 0x04984af1U,
+    0xecdaf741U, 0xcd500e7fU, 0x91f62f17U, 0x4dd68d76U, 0xefb04d43U,
+    0xaa4d54ccU, 0x9604dfe4U, 0xd1b5e39eU, 0x6a881b4cU, 0x2c1fb8c1U,
+    0x65517f46U, 0x5eea049dU, 0x8c355d01U, 0x877473faU, 0x0b412efbU,
+    0x671d5ab3U, 0xdbd25292U, 0x105633e9U, 0xd647136dU, 0xd7618c9aU,
+    0xa10c7a37U, 0xf8148e59U, 0x133c89ebU, 0xa927eeceU, 0x61c935b7U,
+    0x1ce5ede1U, 0x47b13c7aU, 0xd2df599cU, 0xf2733f55U, 0x14ce7918U,
+    0xc737bf73U, 0xf7cdea53U, 0xfdaa5b5fU, 0x3d6f14dfU, 0x44db8678U,
+    0xaff381caU, 0x68c43eb9U, 0x24342c38U, 0xa3405fc2U, 0x1dc37216U,
+    0xe2250cbcU, 0x3c498b28U, 0x0d9541ffU, 0xa8017139U, 0x0cb3de08U,
+    0xb4e49cd8U, 0x56c19064U, 0xcb84617bU, 0x32b670d5U, 0x6c5c7448U,
+    0xb85742d0U, };
+
+static const uint8_t Td4[256] = {
+    0x52U, 0x09U, 0x6aU, 0xd5U, 0x30U, 0x36U, 0xa5U, 0x38U, 0xbfU, 0x40U, 0xa3U,
+    0x9eU, 0x81U, 0xf3U, 0xd7U, 0xfbU, 0x7cU, 0xe3U, 0x39U, 0x82U, 0x9bU, 0x2fU,
+    0xffU, 0x87U, 0x34U, 0x8eU, 0x43U, 0x44U, 0xc4U, 0xdeU, 0xe9U, 0xcbU, 0x54U,
+    0x7bU, 0x94U, 0x32U, 0xa6U, 0xc2U, 0x23U, 0x3dU, 0xeeU, 0x4cU, 0x95U, 0x0bU,
+    0x42U, 0xfaU, 0xc3U, 0x4eU, 0x08U, 0x2eU, 0xa1U, 0x66U, 0x28U, 0xd9U, 0x24U,
+    0xb2U, 0x76U, 0x5bU, 0xa2U, 0x49U, 0x6dU, 0x8bU, 0xd1U, 0x25U, 0x72U, 0xf8U,
+    0xf6U, 0x64U, 0x86U, 0x68U, 0x98U, 0x16U, 0xd4U, 0xa4U, 0x5cU, 0xccU, 0x5dU,
+    0x65U, 0xb6U, 0x92U, 0x6cU, 0x70U, 0x48U, 0x50U, 0xfdU, 0xedU, 0xb9U, 0xdaU,
+    0x5eU, 0x15U, 0x46U, 0x57U, 0xa7U, 0x8dU, 0x9dU, 0x84U, 0x90U, 0xd8U, 0xabU,
+    0x00U, 0x8cU, 0xbcU, 0xd3U, 0x0aU, 0xf7U, 0xe4U, 0x58U, 0x05U, 0xb8U, 0xb3U,
+    0x45U, 0x06U, 0xd0U, 0x2cU, 0x1eU, 0x8fU, 0xcaU, 0x3fU, 0x0fU, 0x02U, 0xc1U,
+    0xafU, 0xbdU, 0x03U, 0x01U, 0x13U, 0x8aU, 0x6bU, 0x3aU, 0x91U, 0x11U, 0x41U,
+    0x4fU, 0x67U, 0xdcU, 0xeaU, 0x97U, 0xf2U, 0xcfU, 0xceU, 0xf0U, 0xb4U, 0xe6U,
+    0x73U, 0x96U, 0xacU, 0x74U, 0x22U, 0xe7U, 0xadU, 0x35U, 0x85U, 0xe2U, 0xf9U,
+    0x37U, 0xe8U, 0x1cU, 0x75U, 0xdfU, 0x6eU, 0x47U, 0xf1U, 0x1aU, 0x71U, 0x1dU,
+    0x29U, 0xc5U, 0x89U, 0x6fU, 0xb7U, 0x62U, 0x0eU, 0xaaU, 0x18U, 0xbeU, 0x1bU,
+    0xfcU, 0x56U, 0x3eU, 0x4bU, 0xc6U, 0xd2U, 0x79U, 0x20U, 0x9aU, 0xdbU, 0xc0U,
+    0xfeU, 0x78U, 0xcdU, 0x5aU, 0xf4U, 0x1fU, 0xddU, 0xa8U, 0x33U, 0x88U, 0x07U,
+    0xc7U, 0x31U, 0xb1U, 0x12U, 0x10U, 0x59U, 0x27U, 0x80U, 0xecU, 0x5fU, 0x60U,
+    0x51U, 0x7fU, 0xa9U, 0x19U, 0xb5U, 0x4aU, 0x0dU, 0x2dU, 0xe5U, 0x7aU, 0x9fU,
+    0x93U, 0xc9U, 0x9cU, 0xefU, 0xa0U, 0xe0U, 0x3bU, 0x4dU, 0xaeU, 0x2aU, 0xf5U,
+    0xb0U, 0xc8U, 0xebU, 0xbbU, 0x3cU, 0x83U, 0x53U, 0x99U, 0x61U, 0x17U, 0x2bU,
+    0x04U, 0x7eU, 0xbaU, 0x77U, 0xd6U, 0x26U, 0xe1U, 0x69U, 0x14U, 0x63U, 0x55U,
+    0x21U, 0x0cU, 0x7dU, };
+
+static const uint32_t rcon[] = {
+    0x01000000, 0x02000000, 0x04000000, 0x08000000, 0x10000000,
+    0x20000000, 0x40000000, 0x80000000, 0x1B000000, 0x36000000,
+    // for 128-bit blocks, Rijndael never uses more than 10 rcon values
+};
+
+int aes_nohw_set_encrypt_key(const uint8_t *key, unsigned bits,
+                             AES_KEY *aeskey) {
+  uint32_t *rk;
+  int i = 0;
+  uint32_t temp;
+
+  if (!key || !aeskey) {
+    return -1;
+  }
+
+  switch (bits) {
+    case 128:
+      aeskey->rounds = 10;
+      break;
+    case 192:
+      aeskey->rounds = 12;
+      break;
+    case 256:
+      aeskey->rounds = 14;
+      break;
+    default:
+      return -2;
+  }
+
+  rk = aeskey->rd_key;
+
+  rk[0] = GETU32(key);
+  rk[1] = GETU32(key + 4);
+  rk[2] = GETU32(key + 8);
+  rk[3] = GETU32(key + 12);
+  if (bits == 128) {
+    while (1) {
+      temp = rk[3];
+      rk[4] = rk[0] ^ (Te2[(temp >> 16) & 0xff] & 0xff000000) ^
+              (Te3[(temp >> 8) & 0xff] & 0x00ff0000) ^
+              (Te0[(temp) & 0xff] & 0x0000ff00) ^
+              (Te1[(temp >> 24)] & 0x000000ff) ^ rcon[i];
+      rk[5] = rk[1] ^ rk[4];
+      rk[6] = rk[2] ^ rk[5];
+      rk[7] = rk[3] ^ rk[6];
+      if (++i == 10) {
+        return 0;
+      }
+      rk += 4;
+    }
+  }
+  rk[4] = GETU32(key + 16);
+  rk[5] = GETU32(key + 20);
+  if (bits == 192) {
+    while (1) {
+      temp = rk[5];
+      rk[6] = rk[0] ^ (Te2[(temp >> 16) & 0xff] & 0xff000000) ^
+              (Te3[(temp >> 8) & 0xff] & 0x00ff0000) ^
+              (Te0[(temp) & 0xff] & 0x0000ff00) ^
+              (Te1[(temp >> 24)] & 0x000000ff) ^ rcon[i];
+      rk[7] = rk[1] ^ rk[6];
+      rk[8] = rk[2] ^ rk[7];
+      rk[9] = rk[3] ^ rk[8];
+      if (++i == 8) {
+        return 0;
+      }
+      rk[10] = rk[4] ^ rk[9];
+      rk[11] = rk[5] ^ rk[10];
+      rk += 6;
+    }
+  }
+  rk[6] = GETU32(key + 24);
+  rk[7] = GETU32(key + 28);
+  if (bits == 256) {
+    while (1) {
+      temp = rk[7];
+      rk[8] = rk[0] ^ (Te2[(temp >> 16) & 0xff] & 0xff000000) ^
+              (Te3[(temp >> 8) & 0xff] & 0x00ff0000) ^
+              (Te0[(temp) & 0xff] & 0x0000ff00) ^
+              (Te1[(temp >> 24)] & 0x000000ff) ^ rcon[i];
+      rk[9] = rk[1] ^ rk[8];
+      rk[10] = rk[2] ^ rk[9];
+      rk[11] = rk[3] ^ rk[10];
+      if (++i == 7) {
+        return 0;
+      }
+      temp = rk[11];
+      rk[12] = rk[4] ^ (Te2[(temp >> 24)] & 0xff000000) ^
+               (Te3[(temp >> 16) & 0xff] & 0x00ff0000) ^
+               (Te0[(temp >> 8) & 0xff] & 0x0000ff00) ^
+               (Te1[(temp) & 0xff] & 0x000000ff);
+      rk[13] = rk[5] ^ rk[12];
+      rk[14] = rk[6] ^ rk[13];
+      rk[15] = rk[7] ^ rk[14];
+
+      rk += 8;
+    }
+  }
+  return 0;
+}
+
+int aes_nohw_set_decrypt_key(const uint8_t *key, unsigned bits,
+                             AES_KEY *aeskey) {
+  uint32_t *rk;
+  int i, j, status;
+  uint32_t temp;
+
+  // first, start with an encryption schedule
+  status = AES_set_encrypt_key(key, bits, aeskey);
+  if (status < 0) {
+    return status;
+  }
+
+  rk = aeskey->rd_key;
+
+  // invert the order of the round keys:
+  for (i = 0, j = 4 * aeskey->rounds; i < j; i += 4, j -= 4) {
+    temp = rk[i];
+    rk[i] = rk[j];
+    rk[j] = temp;
+    temp = rk[i + 1];
+    rk[i + 1] = rk[j + 1];
+    rk[j + 1] = temp;
+    temp = rk[i + 2];
+    rk[i + 2] = rk[j + 2];
+    rk[j + 2] = temp;
+    temp = rk[i + 3];
+    rk[i + 3] = rk[j + 3];
+    rk[j + 3] = temp;
+  }
+  // apply the inverse MixColumn transform to all round keys but the first and
+  // the last:
+  for (i = 1; i < (int)aeskey->rounds; i++) {
+    rk += 4;
+    rk[0] =
+        Td0[Te1[(rk[0] >> 24)] & 0xff] ^ Td1[Te1[(rk[0] >> 16) & 0xff] & 0xff] ^
+        Td2[Te1[(rk[0] >> 8) & 0xff] & 0xff] ^ Td3[Te1[(rk[0]) & 0xff] & 0xff];
+    rk[1] =
+        Td0[Te1[(rk[1] >> 24)] & 0xff] ^ Td1[Te1[(rk[1] >> 16) & 0xff] & 0xff] ^
+        Td2[Te1[(rk[1] >> 8) & 0xff] & 0xff] ^ Td3[Te1[(rk[1]) & 0xff] & 0xff];
+    rk[2] =
+        Td0[Te1[(rk[2] >> 24)] & 0xff] ^ Td1[Te1[(rk[2] >> 16) & 0xff] & 0xff] ^
+        Td2[Te1[(rk[2] >> 8) & 0xff] & 0xff] ^ Td3[Te1[(rk[2]) & 0xff] & 0xff];
+    rk[3] =
+        Td0[Te1[(rk[3] >> 24)] & 0xff] ^ Td1[Te1[(rk[3] >> 16) & 0xff] & 0xff] ^
+        Td2[Te1[(rk[3] >> 8) & 0xff] & 0xff] ^ Td3[Te1[(rk[3]) & 0xff] & 0xff];
+  }
+  return 0;
+}
+
+void aes_nohw_encrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key) {
+  const uint32_t *rk;
+  uint32_t s0, s1, s2, s3, t0, t1, t2, t3;
+  int r;
+
+  assert(in && out && key);
+  rk = key->rd_key;
+
+  // map byte array block to cipher state
+  // and add initial round key:
+  s0 = GETU32(in) ^ rk[0];
+  s1 = GETU32(in + 4) ^ rk[1];
+  s2 = GETU32(in + 8) ^ rk[2];
+  s3 = GETU32(in + 12) ^ rk[3];
+
+  // Nr - 1 full rounds:
+  r = key->rounds >> 1;
+  for (;;) {
+    t0 = Te0[(s0 >> 24)] ^ Te1[(s1 >> 16) & 0xff] ^ Te2[(s2 >> 8) & 0xff] ^
+         Te3[(s3) & 0xff] ^ rk[4];
+    t1 = Te0[(s1 >> 24)] ^ Te1[(s2 >> 16) & 0xff] ^ Te2[(s3 >> 8) & 0xff] ^
+         Te3[(s0) & 0xff] ^ rk[5];
+    t2 = Te0[(s2 >> 24)] ^ Te1[(s3 >> 16) & 0xff] ^ Te2[(s0 >> 8) & 0xff] ^
+         Te3[(s1) & 0xff] ^ rk[6];
+    t3 = Te0[(s3 >> 24)] ^ Te1[(s0 >> 16) & 0xff] ^ Te2[(s1 >> 8) & 0xff] ^
+         Te3[(s2) & 0xff] ^ rk[7];
+
+    rk += 8;
+    if (--r == 0) {
+      break;
+    }
+
+    s0 = Te0[(t0 >> 24)] ^ Te1[(t1 >> 16) & 0xff] ^ Te2[(t2 >> 8) & 0xff] ^
+         Te3[(t3) & 0xff] ^ rk[0];
+    s1 = Te0[(t1 >> 24)] ^ Te1[(t2 >> 16) & 0xff] ^ Te2[(t3 >> 8) & 0xff] ^
+         Te3[(t0) & 0xff] ^ rk[1];
+    s2 = Te0[(t2 >> 24)] ^ Te1[(t3 >> 16) & 0xff] ^ Te2[(t0 >> 8) & 0xff] ^
+         Te3[(t1) & 0xff] ^ rk[2];
+    s3 = Te0[(t3 >> 24)] ^ Te1[(t0 >> 16) & 0xff] ^ Te2[(t1 >> 8) & 0xff] ^
+         Te3[(t2) & 0xff] ^ rk[3];
+  }
+
+  //  apply last round and map cipher state to byte array block:
+  s0 = (Te2[(t0 >> 24)] & 0xff000000) ^ (Te3[(t1 >> 16) & 0xff] & 0x00ff0000) ^
+       (Te0[(t2 >> 8) & 0xff] & 0x0000ff00) ^ (Te1[(t3) & 0xff] & 0x000000ff) ^
+       rk[0];
+  PUTU32(out, s0);
+  s1 = (Te2[(t1 >> 24)] & 0xff000000) ^ (Te3[(t2 >> 16) & 0xff] & 0x00ff0000) ^
+       (Te0[(t3 >> 8) & 0xff] & 0x0000ff00) ^ (Te1[(t0) & 0xff] & 0x000000ff) ^
+       rk[1];
+  PUTU32(out + 4, s1);
+  s2 = (Te2[(t2 >> 24)] & 0xff000000) ^ (Te3[(t3 >> 16) & 0xff] & 0x00ff0000) ^
+       (Te0[(t0 >> 8) & 0xff] & 0x0000ff00) ^ (Te1[(t1) & 0xff] & 0x000000ff) ^
+       rk[2];
+  PUTU32(out + 8, s2);
+  s3 = (Te2[(t3 >> 24)] & 0xff000000) ^ (Te3[(t0 >> 16) & 0xff] & 0x00ff0000) ^
+       (Te0[(t1 >> 8) & 0xff] & 0x0000ff00) ^ (Te1[(t2) & 0xff] & 0x000000ff) ^
+       rk[3];
+  PUTU32(out + 12, s3);
+}
+
+void aes_nohw_decrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key) {
+  const uint32_t *rk;
+  uint32_t s0, s1, s2, s3, t0, t1, t2, t3;
+  int r;
+
+  assert(in && out && key);
+  rk = key->rd_key;
+
+  // map byte array block to cipher state
+  // and add initial round key:
+  s0 = GETU32(in) ^ rk[0];
+  s1 = GETU32(in + 4) ^ rk[1];
+  s2 = GETU32(in + 8) ^ rk[2];
+  s3 = GETU32(in + 12) ^ rk[3];
+
+  // Nr - 1 full rounds:
+  r = key->rounds >> 1;
+  for (;;) {
+    t0 = Td0[(s0 >> 24)] ^ Td1[(s3 >> 16) & 0xff] ^ Td2[(s2 >> 8) & 0xff] ^
+         Td3[(s1) & 0xff] ^ rk[4];
+    t1 = Td0[(s1 >> 24)] ^ Td1[(s0 >> 16) & 0xff] ^ Td2[(s3 >> 8) & 0xff] ^
+         Td3[(s2) & 0xff] ^ rk[5];
+    t2 = Td0[(s2 >> 24)] ^ Td1[(s1 >> 16) & 0xff] ^ Td2[(s0 >> 8) & 0xff] ^
+         Td3[(s3) & 0xff] ^ rk[6];
+    t3 = Td0[(s3 >> 24)] ^ Td1[(s2 >> 16) & 0xff] ^ Td2[(s1 >> 8) & 0xff] ^
+         Td3[(s0) & 0xff] ^ rk[7];
+
+    rk += 8;
+    if (--r == 0) {
+      break;
+    }
+
+    s0 = Td0[(t0 >> 24)] ^ Td1[(t3 >> 16) & 0xff] ^ Td2[(t2 >> 8) & 0xff] ^
+         Td3[(t1) & 0xff] ^ rk[0];
+    s1 = Td0[(t1 >> 24)] ^ Td1[(t0 >> 16) & 0xff] ^ Td2[(t3 >> 8) & 0xff] ^
+         Td3[(t2) & 0xff] ^ rk[1];
+    s2 = Td0[(t2 >> 24)] ^ Td1[(t1 >> 16) & 0xff] ^ Td2[(t0 >> 8) & 0xff] ^
+         Td3[(t3) & 0xff] ^ rk[2];
+    s3 = Td0[(t3 >> 24)] ^ Td1[(t2 >> 16) & 0xff] ^ Td2[(t1 >> 8) & 0xff] ^
+         Td3[(t0) & 0xff] ^ rk[3];
+  }
+
+  // apply last round and
+  // map cipher state to byte array block:
+  s0 = ((uint32_t)Td4[(t0 >> 24)] << 24) ^
+       ((uint32_t)Td4[(t3 >> 16) & 0xff] << 16) ^
+       ((uint32_t)Td4[(t2 >> 8) & 0xff] << 8) ^
+       ((uint32_t)Td4[(t1) & 0xff]) ^ rk[0];
+  PUTU32(out, s0);
+  s1 = ((uint32_t)Td4[(t1 >> 24)] << 24) ^
+       ((uint32_t)Td4[(t0 >> 16) & 0xff] << 16) ^
+       ((uint32_t)Td4[(t3 >> 8) & 0xff] << 8) ^
+       ((uint32_t)Td4[(t2) & 0xff]) ^ rk[1];
+  PUTU32(out + 4, s1);
+  s2 = ((uint32_t)Td4[(t2 >> 24)] << 24) ^
+       ((uint32_t)Td4[(t1 >> 16) & 0xff] << 16) ^
+       ((uint32_t)Td4[(t0 >> 8) & 0xff] << 8) ^
+       ((uint32_t)Td4[(t3) & 0xff]) ^ rk[2];
+  PUTU32(out + 8, s2);
+  s3 = ((uint32_t)Td4[(t3 >> 24)] << 24) ^
+       ((uint32_t)Td4[(t2 >> 16) & 0xff] << 16) ^
+       ((uint32_t)Td4[(t1 >> 8) & 0xff] << 8) ^
+       ((uint32_t)Td4[(t0) & 0xff]) ^ rk[3];
+  PUTU32(out + 12, s3);
+}
+
+#endif  // NO_ASM || (!X86 && !X86_64 && !ARM)
+
 // Be aware that different sets of AES functions use incompatible key
 // representations, varying in format of the key schedule, the |AES_KEY.rounds|
 // value, or both. Therefore they cannot mix. Also, on AArch64, the plain-C
diff --git a/crypto/fipsmodule/aes/aes_nohw.c b/crypto/fipsmodule/aes/aes_nohw.c
deleted file mode 100644
index e3e4e28..0000000
--- a/crypto/fipsmodule/aes/aes_nohw.c
+++ /dev/null
@@ -1,1272 +0,0 @@
-/* Copyright (c) 2019, Google Inc.
- *
- * Permission to use, copy, modify, and/or distribute this software for any
- * purpose with or without fee is hereby granted, provided that the above
- * copyright notice and this permission notice appear in all copies.
- *
- * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
- * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
- * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
- * SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
- * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
- * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */
-
-#include <openssl/aes.h>
-
-#include <assert.h>
-#include <string.h>
-
-#include "../../internal.h"
-
-#if defined(OPENSSL_SSE2)
-#include <emmintrin.h>
-#endif
-
-
-// This file contains a constant-time implementation of AES, bitsliced with
-// 32-bit, 64-bit, or 128-bit words, operating on two-, four-, and eight-block
-// batches, respectively. The 128-bit implementation requires SSE2 intrinsics.
-//
-// This implementation is based on the algorithms described in the following
-// references:
-// - https://bearssl.org/constanttime.html#aes
-// - https://eprint.iacr.org/2009/129.pdf
-// - https://eprint.iacr.org/2009/191.pdf
-
-
-// Word operations.
-//
-// An aes_word_t is the word used for this AES implementation. Throughout this
-// file, bits and bytes are ordered little-endian, though "left" and "right"
-// shifts match the operations themselves, which makes them reversed in a
-// little-endian, left-to-right reading.
-//
-// Eight |aes_word_t|s contain |AES_NOHW_BATCH_SIZE| blocks. The bits in an
-// |aes_word_t| are divided into 16 consecutive groups of |AES_NOHW_BATCH_SIZE|
-// bits each, each corresponding to a byte in an AES block in column-major
-// order (AES's byte order). We refer to these as "logical bytes". Note, in the
-// 32-bit and 64-bit implementations, they are smaller than a byte. (The
-// contents of a logical byte will be described later.)
-//
-// MSVC does not support C bit operators on |__m128i|, so the wrapper functions
-// |aes_nohw_and|, etc., should be used instead. Note |aes_nohw_shift_left| and
-// |aes_nohw_shift_right| measure the shift in logical bytes. That is, the shift
-// value ranges from 0 to 15 independent of |aes_word_t| and
-// |AES_NOHW_BATCH_SIZE|.
-//
-// This ordering is different from https://eprint.iacr.org/2009/129.pdf, which
-// uses row-major order. Matching the AES order was easier to reason about, and
-// we do not have PSHUFB available to arbitrarily permute bytes.
-
-#if defined(OPENSSL_SSE2)
-typedef __m128i aes_word_t;
-// AES_NOHW_WORD_SIZE is sizeof(aes_word_t). alignas(sizeof(T)) does not work in
-// MSVC, so we define a constant.
-#define AES_NOHW_WORD_SIZE 16
-#define AES_NOHW_BATCH_SIZE 8
-#define AES_NOHW_ROW0_MASK \
-  _mm_set_epi32(0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff)
-#define AES_NOHW_ROW1_MASK \
-  _mm_set_epi32(0x0000ff00, 0x0000ff00, 0x0000ff00, 0x0000ff00)
-#define AES_NOHW_ROW2_MASK \
-  _mm_set_epi32(0x00ff0000, 0x00ff0000, 0x00ff0000, 0x00ff0000)
-#define AES_NOHW_ROW3_MASK \
-  _mm_set_epi32(0xff000000, 0xff000000, 0xff000000, 0xff000000)
-#define AES_NOHW_COL01_MASK \
-  _mm_set_epi32(0x00000000, 0x00000000, 0xffffffff, 0xffffffff)
-#define AES_NOHW_COL2_MASK \
-  _mm_set_epi32(0x00000000, 0xffffffff, 0x00000000, 0x00000000)
-#define AES_NOHW_COL3_MASK \
-  _mm_set_epi32(0xffffffff, 0x00000000, 0x00000000, 0x00000000)
-
-static inline aes_word_t aes_nohw_and(aes_word_t a, aes_word_t b) {
-  return _mm_and_si128(a, b);
-}
-
-static inline aes_word_t aes_nohw_or(aes_word_t a, aes_word_t b) {
-  return _mm_or_si128(a, b);
-}
-
-static inline aes_word_t aes_nohw_xor(aes_word_t a, aes_word_t b) {
-  return _mm_xor_si128(a, b);
-}
-
-static inline aes_word_t aes_nohw_not(aes_word_t a) {
-  return _mm_xor_si128(
-      a, _mm_set_epi32(0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff));
-}
-
-// These are macros because parameters to |_mm_slli_si128| and |_mm_srli_si128|
-// must be constants.
-#define aes_nohw_shift_left(/* aes_word_t */ a, /* const */ i) \
-  _mm_slli_si128((a), (i))
-#define aes_nohw_shift_right(/* aes_word_t */ a, /* const */ i) \
-  _mm_srli_si128((a), (i))
-#else  // !OPENSSL_SSE2
-#if defined(OPENSSL_64_BIT)
-typedef uint64_t aes_word_t;
-#define AES_NOHW_WORD_SIZE 8
-#define AES_NOHW_BATCH_SIZE 4
-#define AES_NOHW_ROW0_MASK UINT64_C(0x000f000f000f000f)
-#define AES_NOHW_ROW1_MASK UINT64_C(0x00f000f000f000f0)
-#define AES_NOHW_ROW2_MASK UINT64_C(0x0f000f000f000f00)
-#define AES_NOHW_ROW3_MASK UINT64_C(0xf000f000f000f000)
-#define AES_NOHW_COL01_MASK UINT64_C(0x00000000ffffffff)
-#define AES_NOHW_COL2_MASK UINT64_C(0x0000ffff00000000)
-#define AES_NOHW_COL3_MASK UINT64_C(0xffff000000000000)
-#else  // !OPENSSL_64_BIT
-typedef uint32_t aes_word_t;
-#define AES_NOHW_WORD_SIZE 4
-#define AES_NOHW_BATCH_SIZE 2
-#define AES_NOHW_ROW0_MASK 0x03030303
-#define AES_NOHW_ROW1_MASK 0x0c0c0c0c
-#define AES_NOHW_ROW2_MASK 0x30303030
-#define AES_NOHW_ROW3_MASK 0xc0c0c0c0
-#define AES_NOHW_COL01_MASK 0x0000ffff
-#define AES_NOHW_COL2_MASK 0x00ff0000
-#define AES_NOHW_COL3_MASK 0xff000000
-#endif  // OPENSSL_64_BIT
-
-static inline aes_word_t aes_nohw_and(aes_word_t a, aes_word_t b) {
-  return a & b;
-}
-
-static inline aes_word_t aes_nohw_or(aes_word_t a, aes_word_t b) {
-  return a | b;
-}
-
-static inline aes_word_t aes_nohw_xor(aes_word_t a, aes_word_t b) {
-  return a ^ b;
-}
-
-static inline aes_word_t aes_nohw_not(aes_word_t a) { return ~a; }
-
-static inline aes_word_t aes_nohw_shift_left(aes_word_t a, aes_word_t i) {
-  return a << (i * AES_NOHW_BATCH_SIZE);
-}
-
-static inline aes_word_t aes_nohw_shift_right(aes_word_t a, aes_word_t i) {
-  return a >> (i * AES_NOHW_BATCH_SIZE);
-}
-#endif  // OPENSSL_SSE2
-
-OPENSSL_STATIC_ASSERT(AES_NOHW_BATCH_SIZE * 128 == 8 * 8 * sizeof(aes_word_t),
-                      "batch size does not match word size");
-OPENSSL_STATIC_ASSERT(AES_NOHW_WORD_SIZE == sizeof(aes_word_t),
-                      "AES_NOHW_WORD_SIZE is incorrect");
-
-
-// Block representations.
-//
-// This implementation uses three representations for AES blocks. First, the
-// public API represents blocks as uint8_t[16] in the usual way. Second, most
-// AES steps are evaluated in bitsliced form, stored in an |AES_NOHW_BATCH|.
-// This stores |AES_NOHW_BATCH_SIZE| blocks in bitsliced order. For 64-bit words
-// containing bitsliced blocks a, b, c, d, this would be as follows (vertical
-// bars divide logical bytes):
-//
-//   batch.w[0] = a0 b0 c0 d0 |  a8  b8  c8  d8 | a16 b16 c16 d16 ...
-//   batch.w[1] = a1 b1 c1 d1 |  a9  b9  c9  d9 | a17 b17 c17 d17 ...
-//   batch.w[2] = a2 b2 c2 d2 | a10 b10 c10 d10 | a18 b18 c18 d18 ...
-//   batch.w[3] = a3 b3 c3 d3 | a11 b11 c11 d11 | a19 b19 c19 d19 ...
-//   ...
-//
-// Finally, an individual block may be stored as an intermediate form in an
-// aes_word_t[AES_NOHW_BLOCK_WORDS]. In this form, we permute the bits in each
-// block, so that block[0]'s ith logical byte contains least-significant
-// |AES_NOHW_BATCH_SIZE| bits of byte i, block[1] contains the next group of
-// |AES_NOHW_BATCH_SIZE| bits, and so on. We refer to this transformation as
-// "compacting" the block. Note this is no-op with 128-bit words because then
-// |AES_NOHW_BLOCK_WORDS| is one and |AES_NOHW_BATCH_SIZE| is eight. For 64-bit
-// words, one block would be stored in two words:
-//
-//   block[0] = a0 a1 a2 a3 |  a8  a9 a10 a11 | a16 a17 a18 a19 ...
-//   block[1] = a4 a5 a6 a7 | a12 a13 a14 a15 | a20 a21 a22 a23 ...
-//
-// Observe that the distances between corresponding bits in bitsliced and
-// compact bit orders match. If we line up corresponding words of each block,
-// the bitsliced and compact representations may be converted by tranposing bits
-// in corresponding logical bytes. Continuing the 64-bit example:
-//
-//   block_a[0] = a0 a1 a2 a3 |  a8  a9 a10 a11 | a16 a17 a18 a19 ...
-//   block_b[0] = b0 b1 b2 b3 |  b8  b9 b10 b11 | b16 b17 b18 b19 ...
-//   block_c[0] = c0 c1 c2 c3 |  c8  c9 c10 c11 | c16 c17 c18 c19 ...
-//   block_d[0] = d0 d1 d2 d3 |  d8  d9 d10 d11 | d16 d17 d18 d19 ...
-//
-//   batch.w[0] = a0 b0 c0 d0 |  a8  b8  c8  d8 | a16 b16 c16 d16 ...
-//   batch.w[1] = a1 b1 c1 d1 |  a9  b9  c9  d9 | a17 b17 c17 d17 ...
-//   batch.w[2] = a2 b2 c2 d2 | a10 b10 c10 d10 | a18 b18 c18 d18 ...
-//   batch.w[3] = a3 b3 c3 d3 | a11 b11 c11 d11 | a19 b19 c19 d19 ...
-//
-// Note also that bitwise operations and (logical) byte permutations on an
-// |aes_word_t| work equally for the bitsliced and compact words.
-//
-// We use the compact form in the |AES_KEY| representation to save work
-// inflating round keys into |AES_NOHW_BATCH|. The compact form also exists
-// temporarily while moving blocks in or out of an |AES_NOHW_BATCH|, immediately
-// before or after |aes_nohw_transpose|.
-
-#define AES_NOHW_BLOCK_WORDS (16 / sizeof(aes_word_t))
-
-// An AES_NOHW_BATCH stores |AES_NOHW_BATCH_SIZE| blocks. Unless otherwise
-// specified, it is in bitsliced form.
-typedef struct {
-  aes_word_t w[8];
-} AES_NOHW_BATCH;
-
-// An AES_NOHW_SCHEDULE is an expanded bitsliced AES key schedule. It is
-// suitable for encryption or decryption. It is as large as |AES_NOHW_BATCH|
-// |AES_KEY|s so it should not be used as a long-term key representation.
-typedef struct {
-  // keys is an array of batches, one for each round key. Each batch stores
-  // |AES_NOHW_BATCH_SIZE| copies of the round key in bitsliced form.
-  AES_NOHW_BATCH keys[AES_MAXNR + 1];
-} AES_NOHW_SCHEDULE;
-
-// aes_nohw_batch_set sets the |i|th block of |batch| to |in|. |batch| is in
-// compact form.
-static inline void aes_nohw_batch_set(AES_NOHW_BATCH *batch,
-                                      const aes_word_t in[AES_NOHW_BLOCK_WORDS],
-                                      size_t i) {
-  // Note the words are interleaved. The order comes from |aes_nohw_transpose|.
-  // If |i| is zero and this is the 64-bit implementation, in[0] contains bits
-  // 0-3 and in[1] contains bits 4-7. We place in[0] at w[0] and in[1] at
-  // w[4] so that bits 0 and 4 are in the correct position. (In general, bits
-  // along diagonals of |AES_NOHW_BATCH_SIZE| by |AES_NOHW_BATCH_SIZE| squares
-  // will be correctly placed.)
-  assert(i < AES_NOHW_BATCH_SIZE);
-#if defined(OPENSSL_SSE2)
-  batch->w[i] = in[0];
-#elif defined(OPENSSL_64_BIT)
-  batch->w[i] = in[0];
-  batch->w[i + 4] = in[1];
-#else
-  batch->w[i] = in[0];
-  batch->w[i + 2] = in[1];
-  batch->w[i + 4] = in[2];
-  batch->w[i + 6] = in[3];
-#endif
-}
-
-// aes_nohw_batch_get writes the |i|th block of |batch| to |out|. |batch| is in
-// compact form.
-static inline void aes_nohw_batch_get(const AES_NOHW_BATCH *batch,
-                                      aes_word_t out[AES_NOHW_BLOCK_WORDS],
-                                      size_t i) {
-  assert(i < AES_NOHW_BATCH_SIZE);
-#if defined(OPENSSL_SSE2)
-  out[0] = batch->w[i];
-#elif defined(OPENSSL_64_BIT)
-  out[0] = batch->w[i];
-  out[1] = batch->w[i + 4];
-#else
-  out[0] = batch->w[i];
-  out[1] = batch->w[i + 2];
-  out[2] = batch->w[i + 4];
-  out[3] = batch->w[i + 6];
-#endif
-}
-
-#if !defined(OPENSSL_SSE2)
-// aes_nohw_delta_swap returns |a| with bits |a & mask| and
-// |a & (mask << shift)| swapped. |mask| and |mask << shift| may not overlap.
-static inline aes_word_t aes_nohw_delta_swap(aes_word_t a, aes_word_t mask,
-                                             aes_word_t shift) {
-  // See
-  // https://reflectionsonsecurity.wordpress.com/2014/05/11/efficient-bit-permutation-using-delta-swaps/
-  aes_word_t b = (a ^ (a >> shift)) & mask;
-  return a ^ b ^ (b << shift);
-}
-
-// In the 32-bit and 64-bit implementations, a block spans multiple words.
-// |aes_nohw_compact_block| must permute bits across different words. First we
-// implement |aes_nohw_compact_word| which performs a smaller version of the
-// transformation which stays within a single word.
-//
-// These transformations are generalizations of the output of
-// http://programming.sirrida.de/calcperm.php on smaller inputs.
-#if defined(OPENSSL_64_BIT)
-static inline uint64_t aes_nohw_compact_word(uint64_t a) {
-  // Numbering the 64/2 = 16 4-bit chunks, least to most significant, we swap
-  // quartets of those chunks:
-  //   0 1 2 3 | 4 5 6 7 | 8  9 10 11 | 12 13 14 15 =>
-  //   0 2 1 3 | 4 6 5 7 | 8 10  9 11 | 12 14 13 15
-  a = aes_nohw_delta_swap(a, UINT64_C(0x00f000f000f000f0), 4);
-  // Swap quartets of 8-bit chunks (still numbering by 4-bit chunks):
-  //   0 2 1 3 | 4 6 5 7 | 8 10  9 11 | 12 14 13 15 =>
-  //   0 2 4 6 | 1 3 5 7 | 8 10 12 14 |  9 11 13 15
-  a = aes_nohw_delta_swap(a, UINT64_C(0x0000ff000000ff00), 8);
-  // Swap quartets of 16-bit chunks (still numbering by 4-bit chunks):
-  //   0 2 4 6 | 1  3  5  7 | 8 10 12 14 | 9 11 13 15 =>
-  //   0 2 4 6 | 8 10 12 14 | 1  3  5  7 | 9 11 13 15
-  a = aes_nohw_delta_swap(a, UINT64_C(0x00000000ffff0000), 16);
-  return a;
-}
-
-static inline uint64_t aes_nohw_uncompact_word(uint64_t a) {
-  // Reverse the steps of |aes_nohw_uncompact_word|.
-  a = aes_nohw_delta_swap(a, UINT64_C(0x00000000ffff0000), 16);
-  a = aes_nohw_delta_swap(a, UINT64_C(0x0000ff000000ff00), 8);
-  a = aes_nohw_delta_swap(a, UINT64_C(0x00f000f000f000f0), 4);
-  return a;
-}
-#else   // !OPENSSL_64_BIT
-static inline uint32_t aes_nohw_compact_word(uint32_t a) {
-  // Numbering the 32/2 = 16 pairs of bits, least to most significant, we swap:
-  //   0 1 2 3 | 4 5 6 7 | 8  9 10 11 | 12 13 14 15 =>
-  //   0 4 2 6 | 1 5 3 7 | 8 12 10 14 |  9 13 11 15
-  // Note:  0x00cc = 0b0000_0000_1100_1100
-  //   0x00cc << 6 = 0b0011_0011_0000_0000
-  a = aes_nohw_delta_swap(a, 0x00cc00cc, 6);
-  // Now we swap groups of four bits (still numbering by pairs):
-  //   0 4 2  6 | 1 5 3  7 | 8 12 10 14 | 9 13 11 15 =>
-  //   0 4 8 12 | 1 5 9 13 | 2  6 10 14 | 3  7 11 15
-  // Note: 0x0000_f0f0 << 12 = 0x0f0f_0000
-  a = aes_nohw_delta_swap(a, 0x0000f0f0, 12);
-  return a;
-}
-
-static inline uint32_t aes_nohw_uncompact_word(uint32_t a) {
-  // Reverse the steps of |aes_nohw_uncompact_word|.
-  a = aes_nohw_delta_swap(a, 0x0000f0f0, 12);
-  a = aes_nohw_delta_swap(a, 0x00cc00cc, 6);
-  return a;
-}
-#endif  // OPENSSL_64_BIT
-#endif  // !OPENSSL_SSE2
-
-static inline void aes_nohw_compact_block(aes_word_t out[AES_NOHW_BLOCK_WORDS],
-                                          const uint8_t in[16]) {
-  memcpy(out, in, 16);
-#if defined(OPENSSL_SSE2)
-  // No conversions needed.
-#elif defined(OPENSSL_64_BIT)
-  uint64_t a0 = aes_nohw_compact_word(out[0]);
-  uint64_t a1 = aes_nohw_compact_word(out[1]);
-  out[0] = (a0 & UINT64_C(0x00000000ffffffff)) | (a1 << 32);
-  out[1] = (a1 & UINT64_C(0xffffffff00000000)) | (a0 >> 32);
-#else
-  uint32_t a0 = aes_nohw_compact_word(out[0]);
-  uint32_t a1 = aes_nohw_compact_word(out[1]);
-  uint32_t a2 = aes_nohw_compact_word(out[2]);
-  uint32_t a3 = aes_nohw_compact_word(out[3]);
-  out[0] = (a0 & 0x000000ff) | ((a1 & 0x000000ff) << 8) |
-           ((a2 & 0x000000ff) << 16) | ((a3 & 0x000000ff) << 24);
-  out[1] = ((a0 & 0x0000ff00) >> 8) | (a1 & 0x0000ff00) |
-           ((a2 & 0x0000ff00) << 8) | ((a3 & 0x0000ff00) << 16);
-  out[2] = ((a0 & 0x00ff0000) >> 16) | ((a1 & 0x00ff0000) >> 8) |
-           (a2 & 0x00ff0000) | ((a3 & 0x00ff0000) << 8);
-  out[3] = ((a0 & 0xff000000) >> 24) | ((a1 & 0xff000000) >> 16) |
-           ((a2 & 0xff000000) >> 8) | (a3 & 0xff000000);
-#endif
-}
-
-static inline void aes_nohw_uncompact_block(
-    uint8_t out[16], const aes_word_t in[AES_NOHW_BLOCK_WORDS]) {
-#if defined(OPENSSL_SSE2)
-  memcpy(out, in, 16);  // No conversions needed.
-#elif defined(OPENSSL_64_BIT)
-  uint64_t a0 = in[0];
-  uint64_t a1 = in[1];
-  uint64_t b0 =
-      aes_nohw_uncompact_word((a0 & UINT64_C(0x00000000ffffffff)) | (a1 << 32));
-  uint64_t b1 =
-      aes_nohw_uncompact_word((a1 & UINT64_C(0xffffffff00000000)) | (a0 >> 32));
-  memcpy(out, &b0, 8);
-  memcpy(out + 8, &b1, 8);
-#else
-  uint32_t a0 = in[0];
-  uint32_t a1 = in[1];
-  uint32_t a2 = in[2];
-  uint32_t a3 = in[3];
-  uint32_t b0 = (a0 & 0x000000ff) | ((a1 & 0x000000ff) << 8) |
-                ((a2 & 0x000000ff) << 16) | ((a3 & 0x000000ff) << 24);
-  uint32_t b1 = ((a0 & 0x0000ff00) >> 8) | (a1 & 0x0000ff00) |
-                ((a2 & 0x0000ff00) << 8) | ((a3 & 0x0000ff00) << 16);
-  uint32_t b2 = ((a0 & 0x00ff0000) >> 16) | ((a1 & 0x00ff0000) >> 8) |
-                (a2 & 0x00ff0000) | ((a3 & 0x00ff0000) << 8);
-  uint32_t b3 = ((a0 & 0xff000000) >> 24) | ((a1 & 0xff000000) >> 16) |
-                ((a2 & 0xff000000) >> 8) | (a3 & 0xff000000);
-  b0 = aes_nohw_uncompact_word(b0);
-  b1 = aes_nohw_uncompact_word(b1);
-  b2 = aes_nohw_uncompact_word(b2);
-  b3 = aes_nohw_uncompact_word(b3);
-  memcpy(out, &b0, 4);
-  memcpy(out + 4, &b1, 4);
-  memcpy(out + 8, &b2, 4);
-  memcpy(out + 12, &b3, 4);
-#endif
-}
-
-// aes_nohw_swap_bits is a variation on a delta swap. It swaps the bits in
-// |*a & (mask << shift)| with the bits in |*b & mask|. |mask| and
-// |mask << shift| must not overlap. |mask| is specified as a |uint32_t|, but it
-// is repeated to the full width of |aes_word_t|.
-#if defined(OPENSSL_SSE2)
-// This must be a macro because |_mm_srli_epi32| and |_mm_slli_epi32| require
-// constant shift values.
-#define aes_nohw_swap_bits(/*__m128i* */ a, /*__m128i* */ b,              \
-                           /* uint32_t */ mask, /* const */ shift)        \
-  do {                                                                    \
-    __m128i swap =                                                        \
-        _mm_and_si128(_mm_xor_si128(_mm_srli_epi32(*(a), (shift)), *(b)), \
-                      _mm_set_epi32((mask), (mask), (mask), (mask)));     \
-    *(a) = _mm_xor_si128(*(a), _mm_slli_epi32(swap, (shift)));            \
-    *(b) = _mm_xor_si128(*(b), swap);                                     \
-                                                                          \
-  } while (0)
-#else
-static inline void aes_nohw_swap_bits(aes_word_t *a, aes_word_t *b,
-                                      uint32_t mask, aes_word_t shift) {
-#if defined(OPENSSL_64_BIT)
-  aes_word_t mask_w = (((uint64_t)mask) << 32) | mask;
-#else
-  aes_word_t mask_w = mask;
-#endif
-  // This is a variation on a delta swap.
-  aes_word_t swap = ((*a >> shift) ^ *b) & mask_w;
-  *a ^= swap << shift;
-  *b ^= swap;
-}
-#endif  // OPENSSL_SSE2
-
-// aes_nohw_transpose converts |batch| to and from bitsliced form. It divides
-// the 8 × word_size bits into AES_NOHW_BATCH_SIZE × AES_NOHW_BATCH_SIZE squares
-// and transposes each square.
-static void aes_nohw_transpose(AES_NOHW_BATCH *batch) {
-  // Swap bits with index 0 and 1 mod 2 (0x55 = 0b01010101).
-  aes_nohw_swap_bits(&batch->w[0], &batch->w[1], 0x55555555, 1);
-  aes_nohw_swap_bits(&batch->w[2], &batch->w[3], 0x55555555, 1);
-  aes_nohw_swap_bits(&batch->w[4], &batch->w[5], 0x55555555, 1);
-  aes_nohw_swap_bits(&batch->w[6], &batch->w[7], 0x55555555, 1);
-
-#if AES_NOHW_BATCH_SIZE >= 4
-  // Swap bits with index 0-1 and 2-3 mod 4 (0x33 = 0b00110011).
-  aes_nohw_swap_bits(&batch->w[0], &batch->w[2], 0x33333333, 2);
-  aes_nohw_swap_bits(&batch->w[1], &batch->w[3], 0x33333333, 2);
-  aes_nohw_swap_bits(&batch->w[4], &batch->w[6], 0x33333333, 2);
-  aes_nohw_swap_bits(&batch->w[5], &batch->w[7], 0x33333333, 2);
-#endif
-
-#if AES_NOHW_BATCH_SIZE >= 8
-  // Swap bits with index 0-3 and 4-7 mod 8 (0x0f = 0b00001111).
-  aes_nohw_swap_bits(&batch->w[0], &batch->w[4], 0x0f0f0f0f, 4);
-  aes_nohw_swap_bits(&batch->w[1], &batch->w[5], 0x0f0f0f0f, 4);
-  aes_nohw_swap_bits(&batch->w[2], &batch->w[6], 0x0f0f0f0f, 4);
-  aes_nohw_swap_bits(&batch->w[3], &batch->w[7], 0x0f0f0f0f, 4);
-#endif
-}
-
-// aes_nohw_to_batch initializes |out| with the |num_blocks| blocks from |in|.
-// |num_blocks| must be at most |AES_NOHW_BATCH|.
-static void aes_nohw_to_batch(AES_NOHW_BATCH *out, const uint8_t *in,
-                              size_t num_blocks) {
-  // Don't leave unused blocks unitialized.
-  memset(out, 0, sizeof(AES_NOHW_BATCH));
-  assert(num_blocks <= AES_NOHW_BATCH_SIZE);
-  for (size_t i = 0; i < num_blocks; i++) {
-    aes_word_t block[AES_NOHW_BLOCK_WORDS];
-    aes_nohw_compact_block(block, in + 16 * i);
-    aes_nohw_batch_set(out, block, i);
-  }
-
-  aes_nohw_transpose(out);
-}
-
-// aes_nohw_to_batch writes the first |num_blocks| blocks in |batch| to |out|.
-// |num_blocks| must be at most |AES_NOHW_BATCH|.
-static void aes_nohw_from_batch(uint8_t *out, size_t num_blocks,
-                                const AES_NOHW_BATCH *batch) {
-  AES_NOHW_BATCH copy = *batch;
-  aes_nohw_transpose(&copy);
-
-  assert(num_blocks <= AES_NOHW_BATCH_SIZE);
-  for (size_t i = 0; i < num_blocks; i++) {
-    aes_word_t block[AES_NOHW_BLOCK_WORDS];
-    aes_nohw_batch_get(&copy, block, i);
-    aes_nohw_uncompact_block(out + 16 * i, block);
-  }
-}
-
-
-// AES round steps.
-
-static void aes_nohw_add_round_key(AES_NOHW_BATCH *batch,
-                                   const AES_NOHW_BATCH *key) {
-  for (size_t i = 0; i < 8; i++) {
-    batch->w[i] = aes_nohw_xor(batch->w[i], key->w[i]);
-  }
-}
-
-static void aes_nohw_sub_bytes(AES_NOHW_BATCH *batch) {
-  // See https://eprint.iacr.org/2009/191.pdf, Appendix C.
-  aes_word_t x0 = batch->w[7];
-  aes_word_t x1 = batch->w[6];
-  aes_word_t x2 = batch->w[5];
-  aes_word_t x3 = batch->w[4];
-  aes_word_t x4 = batch->w[3];
-  aes_word_t x5 = batch->w[2];
-  aes_word_t x6 = batch->w[1];
-  aes_word_t x7 = batch->w[0];
-
-  // Figure 2, the top linear transformation.
-  aes_word_t y14 = aes_nohw_xor(x3, x5);
-  aes_word_t y13 = aes_nohw_xor(x0, x6);
-  aes_word_t y9 = aes_nohw_xor(x0, x3);
-  aes_word_t y8 = aes_nohw_xor(x0, x5);
-  aes_word_t t0 = aes_nohw_xor(x1, x2);
-  aes_word_t y1 = aes_nohw_xor(t0, x7);
-  aes_word_t y4 = aes_nohw_xor(y1, x3);
-  aes_word_t y12 = aes_nohw_xor(y13, y14);
-  aes_word_t y2 = aes_nohw_xor(y1, x0);
-  aes_word_t y5 = aes_nohw_xor(y1, x6);
-  aes_word_t y3 = aes_nohw_xor(y5, y8);
-  aes_word_t t1 = aes_nohw_xor(x4, y12);
-  aes_word_t y15 = aes_nohw_xor(t1, x5);
-  aes_word_t y20 = aes_nohw_xor(t1, x1);
-  aes_word_t y6 = aes_nohw_xor(y15, x7);
-  aes_word_t y10 = aes_nohw_xor(y15, t0);
-  aes_word_t y11 = aes_nohw_xor(y20, y9);
-  aes_word_t y7 = aes_nohw_xor(x7, y11);
-  aes_word_t y17 = aes_nohw_xor(y10, y11);
-  aes_word_t y19 = aes_nohw_xor(y10, y8);
-  aes_word_t y16 = aes_nohw_xor(t0, y11);
-  aes_word_t y21 = aes_nohw_xor(y13, y16);
-  aes_word_t y18 = aes_nohw_xor(x0, y16);
-
-  // Figure 3, the middle non-linear section.
-  aes_word_t t2 = aes_nohw_and(y12, y15);
-  aes_word_t t3 = aes_nohw_and(y3, y6);
-  aes_word_t t4 = aes_nohw_xor(t3, t2);
-  aes_word_t t5 = aes_nohw_and(y4, x7);
-  aes_word_t t6 = aes_nohw_xor(t5, t2);
-  aes_word_t t7 = aes_nohw_and(y13, y16);
-  aes_word_t t8 = aes_nohw_and(y5, y1);
-  aes_word_t t9 = aes_nohw_xor(t8, t7);
-  aes_word_t t10 = aes_nohw_and(y2, y7);
-  aes_word_t t11 = aes_nohw_xor(t10, t7);
-  aes_word_t t12 = aes_nohw_and(y9, y11);
-  aes_word_t t13 = aes_nohw_and(y14, y17);
-  aes_word_t t14 = aes_nohw_xor(t13, t12);
-  aes_word_t t15 = aes_nohw_and(y8, y10);
-  aes_word_t t16 = aes_nohw_xor(t15, t12);
-  aes_word_t t17 = aes_nohw_xor(t4, t14);
-  aes_word_t t18 = aes_nohw_xor(t6, t16);
-  aes_word_t t19 = aes_nohw_xor(t9, t14);
-  aes_word_t t20 = aes_nohw_xor(t11, t16);
-  aes_word_t t21 = aes_nohw_xor(t17, y20);
-  aes_word_t t22 = aes_nohw_xor(t18, y19);
-  aes_word_t t23 = aes_nohw_xor(t19, y21);
-  aes_word_t t24 = aes_nohw_xor(t20, y18);
-  aes_word_t t25 = aes_nohw_xor(t21, t22);
-  aes_word_t t26 = aes_nohw_and(t21, t23);
-  aes_word_t t27 = aes_nohw_xor(t24, t26);
-  aes_word_t t28 = aes_nohw_and(t25, t27);
-  aes_word_t t29 = aes_nohw_xor(t28, t22);
-  aes_word_t t30 = aes_nohw_xor(t23, t24);
-  aes_word_t t31 = aes_nohw_xor(t22, t26);
-  aes_word_t t32 = aes_nohw_and(t31, t30);
-  aes_word_t t33 = aes_nohw_xor(t32, t24);
-  aes_word_t t34 = aes_nohw_xor(t23, t33);
-  aes_word_t t35 = aes_nohw_xor(t27, t33);
-  aes_word_t t36 = aes_nohw_and(t24, t35);
-  aes_word_t t37 = aes_nohw_xor(t36, t34);
-  aes_word_t t38 = aes_nohw_xor(t27, t36);
-  aes_word_t t39 = aes_nohw_and(t29, t38);
-  aes_word_t t40 = aes_nohw_xor(t25, t39);
-  aes_word_t t41 = aes_nohw_xor(t40, t37);
-  aes_word_t t42 = aes_nohw_xor(t29, t33);
-  aes_word_t t43 = aes_nohw_xor(t29, t40);
-  aes_word_t t44 = aes_nohw_xor(t33, t37);
-  aes_word_t t45 = aes_nohw_xor(t42, t41);
-  aes_word_t z0 = aes_nohw_and(t44, y15);
-  aes_word_t z1 = aes_nohw_and(t37, y6);
-  aes_word_t z2 = aes_nohw_and(t33, x7);
-  aes_word_t z3 = aes_nohw_and(t43, y16);
-  aes_word_t z4 = aes_nohw_and(t40, y1);
-  aes_word_t z5 = aes_nohw_and(t29, y7);
-  aes_word_t z6 = aes_nohw_and(t42, y11);
-  aes_word_t z7 = aes_nohw_and(t45, y17);
-  aes_word_t z8 = aes_nohw_and(t41, y10);
-  aes_word_t z9 = aes_nohw_and(t44, y12);
-  aes_word_t z10 = aes_nohw_and(t37, y3);
-  aes_word_t z11 = aes_nohw_and(t33, y4);
-  aes_word_t z12 = aes_nohw_and(t43, y13);
-  aes_word_t z13 = aes_nohw_and(t40, y5);
-  aes_word_t z14 = aes_nohw_and(t29, y2);
-  aes_word_t z15 = aes_nohw_and(t42, y9);
-  aes_word_t z16 = aes_nohw_and(t45, y14);
-  aes_word_t z17 = aes_nohw_and(t41, y8);
-
-  // Figure 4, bottom linear transformation.
-  aes_word_t t46 = aes_nohw_xor(z15, z16);
-  aes_word_t t47 = aes_nohw_xor(z10, z11);
-  aes_word_t t48 = aes_nohw_xor(z5, z13);
-  aes_word_t t49 = aes_nohw_xor(z9, z10);
-  aes_word_t t50 = aes_nohw_xor(z2, z12);
-  aes_word_t t51 = aes_nohw_xor(z2, z5);
-  aes_word_t t52 = aes_nohw_xor(z7, z8);
-  aes_word_t t53 = aes_nohw_xor(z0, z3);
-  aes_word_t t54 = aes_nohw_xor(z6, z7);
-  aes_word_t t55 = aes_nohw_xor(z16, z17);
-  aes_word_t t56 = aes_nohw_xor(z12, t48);
-  aes_word_t t57 = aes_nohw_xor(t50, t53);
-  aes_word_t t58 = aes_nohw_xor(z4, t46);
-  aes_word_t t59 = aes_nohw_xor(z3, t54);
-  aes_word_t t60 = aes_nohw_xor(t46, t57);
-  aes_word_t t61 = aes_nohw_xor(z14, t57);
-  aes_word_t t62 = aes_nohw_xor(t52, t58);
-  aes_word_t t63 = aes_nohw_xor(t49, t58);
-  aes_word_t t64 = aes_nohw_xor(z4, t59);
-  aes_word_t t65 = aes_nohw_xor(t61, t62);
-  aes_word_t t66 = aes_nohw_xor(z1, t63);
-  aes_word_t s0 = aes_nohw_xor(t59, t63);
-  aes_word_t s6 = aes_nohw_xor(t56, aes_nohw_not(t62));
-  aes_word_t s7 = aes_nohw_xor(t48, aes_nohw_not(t60));
-  aes_word_t t67 = aes_nohw_xor(t64, t65);
-  aes_word_t s3 = aes_nohw_xor(t53, t66);
-  aes_word_t s4 = aes_nohw_xor(t51, t66);
-  aes_word_t s5 = aes_nohw_xor(t47, t65);
-  aes_word_t s1 = aes_nohw_xor(t64, aes_nohw_not(s3));
-  aes_word_t s2 = aes_nohw_xor(t55, aes_nohw_not(t67));
-
-  batch->w[0] = s7;
-  batch->w[1] = s6;
-  batch->w[2] = s5;
-  batch->w[3] = s4;
-  batch->w[4] = s3;
-  batch->w[5] = s2;
-  batch->w[6] = s1;
-  batch->w[7] = s0;
-}
-
-// aes_nohw_sub_bytes_inv_affine inverts the affine transform portion of the AES
-// S-box, defined in FIPS PUB 197, section 5.1.1, step 2.
-static void aes_nohw_sub_bytes_inv_affine(AES_NOHW_BATCH *batch) {
-  aes_word_t a0 = batch->w[0];
-  aes_word_t a1 = batch->w[1];
-  aes_word_t a2 = batch->w[2];
-  aes_word_t a3 = batch->w[3];
-  aes_word_t a4 = batch->w[4];
-  aes_word_t a5 = batch->w[5];
-  aes_word_t a6 = batch->w[6];
-  aes_word_t a7 = batch->w[7];
-
-  // Apply the circulant [0 0 1 0 0 1 0 1]. This is the inverse of the circulant
-  // [1 0 0 0 1 1 1 1].
-  aes_word_t b0 = aes_nohw_xor(a2, aes_nohw_xor(a5, a7));
-  aes_word_t b1 = aes_nohw_xor(a3, aes_nohw_xor(a6, a0));
-  aes_word_t b2 = aes_nohw_xor(a4, aes_nohw_xor(a7, a1));
-  aes_word_t b3 = aes_nohw_xor(a5, aes_nohw_xor(a0, a2));
-  aes_word_t b4 = aes_nohw_xor(a6, aes_nohw_xor(a1, a3));
-  aes_word_t b5 = aes_nohw_xor(a7, aes_nohw_xor(a2, a4));
-  aes_word_t b6 = aes_nohw_xor(a0, aes_nohw_xor(a3, a5));
-  aes_word_t b7 = aes_nohw_xor(a1, aes_nohw_xor(a4, a6));
-
-  // XOR 0x05. Equivalently, we could XOR 0x63 before applying the circulant,
-  // but 0x05 has lower Hamming weight. (0x05 is the circulant applied to 0x63.)
-  batch->w[0] = aes_nohw_not(b0);
-  batch->w[1] = b1;
-  batch->w[2] = aes_nohw_not(b2);
-  batch->w[3] = b3;
-  batch->w[4] = b4;
-  batch->w[5] = b5;
-  batch->w[6] = b6;
-  batch->w[7] = b7;
-}
-
-static void aes_nohw_inv_sub_bytes(AES_NOHW_BATCH *batch) {
-  // We implement the inverse S-box using the forwards implementation with the
-  // technique described in https://www.bearssl.org/constanttime.html#aes.
-  //
-  // The forwards S-box inverts its input and applies an affine transformation:
-  // S(x) = A(Inv(x)). Thus Inv(x) = InvA(S(x)). The inverse S-box is then:
-  //
-  //   InvS(x) = Inv(InvA(x)).
-  //           = InvA(S(InvA(x)))
-  aes_nohw_sub_bytes_inv_affine(batch);
-  aes_nohw_sub_bytes(batch);
-  aes_nohw_sub_bytes_inv_affine(batch);
-}
-
-// aes_nohw_rotate_cols_right returns |v| with the columns in each row rotated
-// to the right by |n|. This is a macro because |aes_nohw_shift_*| require
-// constant shift counts in the SSE2 implementation.
-#define aes_nohw_rotate_cols_right(/* aes_word_t */ v, /* const */ n) \
-  (aes_nohw_or(aes_nohw_shift_right((v), (n)*4),                      \
-               aes_nohw_shift_left((v), 16 - (n)*4)))
-
-static void aes_nohw_shift_rows(AES_NOHW_BATCH *batch) {
-  for (size_t i = 0; i < 8; i++) {
-    aes_word_t row0 = aes_nohw_and(batch->w[i], AES_NOHW_ROW0_MASK);
-    aes_word_t row1 = aes_nohw_and(batch->w[i], AES_NOHW_ROW1_MASK);
-    aes_word_t row2 = aes_nohw_and(batch->w[i], AES_NOHW_ROW2_MASK);
-    aes_word_t row3 = aes_nohw_and(batch->w[i], AES_NOHW_ROW3_MASK);
-    row1 = aes_nohw_rotate_cols_right(row1, 1);
-    row2 = aes_nohw_rotate_cols_right(row2, 2);
-    row3 = aes_nohw_rotate_cols_right(row3, 3);
-    batch->w[i] = aes_nohw_or(aes_nohw_or(row0, row1), aes_nohw_or(row2, row3));
-  }
-}
-
-static void aes_nohw_inv_shift_rows(AES_NOHW_BATCH *batch) {
-  for (size_t i = 0; i < 8; i++) {
-    aes_word_t row0 = aes_nohw_and(batch->w[i], AES_NOHW_ROW0_MASK);
-    aes_word_t row1 = aes_nohw_and(batch->w[i], AES_NOHW_ROW1_MASK);
-    aes_word_t row2 = aes_nohw_and(batch->w[i], AES_NOHW_ROW2_MASK);
-    aes_word_t row3 = aes_nohw_and(batch->w[i], AES_NOHW_ROW3_MASK);
-    row1 = aes_nohw_rotate_cols_right(row1, 3);
-    row2 = aes_nohw_rotate_cols_right(row2, 2);
-    row3 = aes_nohw_rotate_cols_right(row3, 1);
-    batch->w[i] = aes_nohw_or(aes_nohw_or(row0, row1), aes_nohw_or(row2, row3));
-  }
-}
-
-// aes_nohw_rotate_rows_down returns |v| with the rows in each column rotated
-// down by one.
-static inline aes_word_t aes_nohw_rotate_rows_down(aes_word_t v) {
-#if defined(OPENSSL_SSE2)
-  return _mm_or_si128(_mm_srli_epi32(v, 8), _mm_slli_epi32(v, 24));
-#elif defined(OPENSSL_64_BIT)
-  return ((v >> 4) & UINT64_C(0x0fff0fff0fff0fff)) |
-         ((v << 12) & UINT64_C(0xf000f000f000f000));
-#else
-  return ((v >> 2) & 0x3f3f3f3f) | ((v << 6) & 0xc0c0c0c0);
-#endif
-}
-
-// aes_nohw_rotate_rows_twice returns |v| with the rows in each column rotated
-// by two.
-static inline aes_word_t aes_nohw_rotate_rows_twice(aes_word_t v) {
-#if defined(OPENSSL_SSE2)
-  return _mm_or_si128(_mm_srli_epi32(v, 16), _mm_slli_epi32(v, 16));
-#elif defined(OPENSSL_64_BIT)
-  return ((v >> 8) & UINT64_C(0x00ff00ff00ff00ff)) |
-         ((v << 8) & UINT64_C(0xff00ff00ff00ff00));
-#else
-  return ((v >> 4) & 0x0f0f0f0f) | ((v << 4) & 0xf0f0f0f0);
-#endif
-}
-
-static void aes_nohw_mix_columns(AES_NOHW_BATCH *batch) {
-  // See https://eprint.iacr.org/2009/129.pdf, section 4.4 and appendix A.
-  aes_word_t a0 = batch->w[0];
-  aes_word_t a1 = batch->w[1];
-  aes_word_t a2 = batch->w[2];
-  aes_word_t a3 = batch->w[3];
-  aes_word_t a4 = batch->w[4];
-  aes_word_t a5 = batch->w[5];
-  aes_word_t a6 = batch->w[6];
-  aes_word_t a7 = batch->w[7];
-
-  aes_word_t r0 = aes_nohw_rotate_rows_down(a0);
-  aes_word_t a0_r0 = aes_nohw_xor(a0, r0);
-  aes_word_t r1 = aes_nohw_rotate_rows_down(a1);
-  aes_word_t a1_r1 = aes_nohw_xor(a1, r1);
-  aes_word_t r2 = aes_nohw_rotate_rows_down(a2);
-  aes_word_t a2_r2 = aes_nohw_xor(a2, r2);
-  aes_word_t r3 = aes_nohw_rotate_rows_down(a3);
-  aes_word_t a3_r3 = aes_nohw_xor(a3, r3);
-  aes_word_t r4 = aes_nohw_rotate_rows_down(a4);
-  aes_word_t a4_r4 = aes_nohw_xor(a4, r4);
-  aes_word_t r5 = aes_nohw_rotate_rows_down(a5);
-  aes_word_t a5_r5 = aes_nohw_xor(a5, r5);
-  aes_word_t r6 = aes_nohw_rotate_rows_down(a6);
-  aes_word_t a6_r6 = aes_nohw_xor(a6, r6);
-  aes_word_t r7 = aes_nohw_rotate_rows_down(a7);
-  aes_word_t a7_r7 = aes_nohw_xor(a7, r7);
-
-  batch->w[0] =
-      aes_nohw_xor(aes_nohw_xor(a7_r7, r0), aes_nohw_rotate_rows_twice(a0_r0));
-  batch->w[1] =
-      aes_nohw_xor(aes_nohw_xor(a0_r0, a7_r7),
-                   aes_nohw_xor(r1, aes_nohw_rotate_rows_twice(a1_r1)));
-  batch->w[2] =
-      aes_nohw_xor(aes_nohw_xor(a1_r1, r2), aes_nohw_rotate_rows_twice(a2_r2));
-  batch->w[3] =
-      aes_nohw_xor(aes_nohw_xor(a2_r2, a7_r7),
-                   aes_nohw_xor(r3, aes_nohw_rotate_rows_twice(a3_r3)));
-  batch->w[4] =
-      aes_nohw_xor(aes_nohw_xor(a3_r3, a7_r7),
-                   aes_nohw_xor(r4, aes_nohw_rotate_rows_twice(a4_r4)));
-  batch->w[5] =
-      aes_nohw_xor(aes_nohw_xor(a4_r4, r5), aes_nohw_rotate_rows_twice(a5_r5));
-  batch->w[6] =
-      aes_nohw_xor(aes_nohw_xor(a5_r5, r6), aes_nohw_rotate_rows_twice(a6_r6));
-  batch->w[7] =
-      aes_nohw_xor(aes_nohw_xor(a6_r6, r7), aes_nohw_rotate_rows_twice(a7_r7));
-}
-
-static void aes_nohw_inv_mix_columns(AES_NOHW_BATCH *batch) {
-  aes_word_t a0 = batch->w[0];
-  aes_word_t a1 = batch->w[1];
-  aes_word_t a2 = batch->w[2];
-  aes_word_t a3 = batch->w[3];
-  aes_word_t a4 = batch->w[4];
-  aes_word_t a5 = batch->w[5];
-  aes_word_t a6 = batch->w[6];
-  aes_word_t a7 = batch->w[7];
-
-  // bsaes-x86_64.pl describes the following decomposition of the inverse
-  // MixColumns matrix, credited to Jussi Kivilinna. This gives a much simpler
-  // multiplication.
-  //
-  // | 0e 0b 0d 09 |   | 02 03 01 01 |   | 05 00 04 00 |
-  // | 09 0e 0b 0d | = | 01 02 03 01 | x | 00 05 00 04 |
-  // | 0d 09 0e 0b |   | 01 01 02 03 |   | 04 00 05 00 |
-  // | 0b 0d 09 0e |   | 03 01 01 02 |   | 00 04 00 05 |
-  //
-  // First, apply the [5 0 4 0] matrix. Multiplying by 4 in F_(2^8) is described
-  // by the following bit equations:
-  //
-  //   b0 = a6
-  //   b1 = a6 ^ a7
-  //   b2 = a0 ^ a7
-  //   b3 = a1 ^ a6
-  //   b4 = a2 ^ a6 ^ a7
-  //   b5 = a3 ^ a7
-  //   b6 = a4
-  //   b7 = a5
-  //
-  // Each coefficient is given by:
-  //
-  //   b_ij = 05·a_ij ⊕ 04·a_i(j+2) = 04·(a_ij ⊕ a_i(j+2)) ⊕ a_ij
-  //
-  // We combine the two equations below. Note a_i(j+2) is a row rotation.
-  aes_word_t a0_r0 = aes_nohw_xor(a0, aes_nohw_rotate_rows_twice(a0));
-  aes_word_t a1_r1 = aes_nohw_xor(a1, aes_nohw_rotate_rows_twice(a1));
-  aes_word_t a2_r2 = aes_nohw_xor(a2, aes_nohw_rotate_rows_twice(a2));
-  aes_word_t a3_r3 = aes_nohw_xor(a3, aes_nohw_rotate_rows_twice(a3));
-  aes_word_t a4_r4 = aes_nohw_xor(a4, aes_nohw_rotate_rows_twice(a4));
-  aes_word_t a5_r5 = aes_nohw_xor(a5, aes_nohw_rotate_rows_twice(a5));
-  aes_word_t a6_r6 = aes_nohw_xor(a6, aes_nohw_rotate_rows_twice(a6));
-  aes_word_t a7_r7 = aes_nohw_xor(a7, aes_nohw_rotate_rows_twice(a7));
-
-  batch->w[0] = aes_nohw_xor(a0, a6_r6);
-  batch->w[1] = aes_nohw_xor(a1, aes_nohw_xor(a6_r6, a7_r7));
-  batch->w[2] = aes_nohw_xor(a2, aes_nohw_xor(a0_r0, a7_r7));
-  batch->w[3] = aes_nohw_xor(a3, aes_nohw_xor(a1_r1, a6_r6));
-  batch->w[4] =
-      aes_nohw_xor(aes_nohw_xor(a4, a2_r2), aes_nohw_xor(a6_r6, a7_r7));
-  batch->w[5] = aes_nohw_xor(a5, aes_nohw_xor(a3_r3, a7_r7));
-  batch->w[6] = aes_nohw_xor(a6, a4_r4);
-  batch->w[7] = aes_nohw_xor(a7, a5_r5);
-
-  // Apply the [02 03 01 01] matrix, which is just MixColumns.
-  aes_nohw_mix_columns(batch);
-}
-
-static void aes_nohw_encrypt_batch(const AES_NOHW_SCHEDULE *key,
-                                   size_t num_rounds, AES_NOHW_BATCH *batch) {
-  aes_nohw_add_round_key(batch, &key->keys[0]);
-  for (size_t i = 1; i < num_rounds; i++) {
-    aes_nohw_sub_bytes(batch);
-    aes_nohw_shift_rows(batch);
-    aes_nohw_mix_columns(batch);
-    aes_nohw_add_round_key(batch, &key->keys[i]);
-  }
-  aes_nohw_sub_bytes(batch);
-  aes_nohw_shift_rows(batch);
-  aes_nohw_add_round_key(batch, &key->keys[num_rounds]);
-}
-
-static void aes_nohw_decrypt_batch(const AES_NOHW_SCHEDULE *key,
-                                   size_t num_rounds, AES_NOHW_BATCH *batch) {
-  aes_nohw_add_round_key(batch, &key->keys[num_rounds]);
-  aes_nohw_inv_shift_rows(batch);
-  aes_nohw_inv_sub_bytes(batch);
-  for (size_t i = num_rounds - 1; i > 0; i--) {
-    aes_nohw_add_round_key(batch, &key->keys[i]);
-    aes_nohw_inv_mix_columns(batch);
-    aes_nohw_inv_shift_rows(batch);
-    aes_nohw_inv_sub_bytes(batch);
-  }
-  aes_nohw_add_round_key(batch, &key->keys[0]);
-}
-
-
-// Key schedule.
-
-static void aes_nohw_expand_round_keys(AES_NOHW_SCHEDULE *out,
-                                       const AES_KEY *key) {
-  for (size_t i = 0; i <= key->rounds; i++) {
-    // Copy the round key into each block in the batch.
-    for (size_t j = 0; j < AES_NOHW_BATCH_SIZE; j++) {
-      aes_word_t tmp[AES_NOHW_BLOCK_WORDS];
-      memcpy(tmp, key->rd_key + 4 * i, 16);
-      aes_nohw_batch_set(&out->keys[i], tmp, j);
-    }
-    aes_nohw_transpose(&out->keys[i]);
-  }
-}
-
-static const uint8_t aes_nohw_rcon[10] = {0x01, 0x02, 0x04, 0x08, 0x10,
-                                          0x20, 0x40, 0x80, 0x1b, 0x36};
-
-// aes_nohw_rcon_slice returns the |i|th group of |AES_NOHW_BATCH_SIZE| bits in
-// |rcon|, stored in a |aes_word_t|.
-static inline aes_word_t aes_nohw_rcon_slice(uint8_t rcon, size_t i) {
-  rcon = (rcon >> (i * AES_NOHW_BATCH_SIZE)) & ((1 << AES_NOHW_BATCH_SIZE) - 1);
-#if defined(OPENSSL_SSE2)
-  return _mm_set_epi32(0, 0, 0, rcon);
-#else
-  return ((aes_word_t)rcon);
-#endif
-}
-
-static void aes_nohw_sub_block(aes_word_t out[AES_NOHW_BLOCK_WORDS],
-                               const aes_word_t in[AES_NOHW_BLOCK_WORDS]) {
-  AES_NOHW_BATCH batch;
-  memset(&batch, 0, sizeof(batch));
-  aes_nohw_batch_set(&batch, in, 0);
-  aes_nohw_transpose(&batch);
-  aes_nohw_sub_bytes(&batch);
-  aes_nohw_transpose(&batch);
-  aes_nohw_batch_get(&batch, out, 0);
-}
-
-static void aes_nohw_setup_key_128(AES_KEY *key, const uint8_t in[16]) {
-  key->rounds = 10;
-
-  aes_word_t block[AES_NOHW_BLOCK_WORDS];
-  aes_nohw_compact_block(block, in);
-  memcpy(key->rd_key, block, 16);
-
-  for (size_t i = 1; i <= 10; i++) {
-    aes_word_t sub[AES_NOHW_BLOCK_WORDS];
-    aes_nohw_sub_block(sub, block);
-    uint8_t rcon = aes_nohw_rcon[i - 1];
-    for (size_t j = 0; j < AES_NOHW_BLOCK_WORDS; j++) {
-      // Incorporate |rcon| and the transformed word into the first word.
-      block[j] = aes_nohw_xor(block[j], aes_nohw_rcon_slice(rcon, j));
-      block[j] = aes_nohw_xor(
-          block[j],
-          aes_nohw_shift_right(aes_nohw_rotate_rows_down(sub[j]), 12));
-      // Propagate to the remaining words. Note this is reordered from the usual
-      // formulation to avoid needing masks.
-      aes_word_t v = block[j];
-      block[j] = aes_nohw_xor(block[j], aes_nohw_shift_left(v, 4));
-      block[j] = aes_nohw_xor(block[j], aes_nohw_shift_left(v, 8));
-      block[j] = aes_nohw_xor(block[j], aes_nohw_shift_left(v, 12));
-    }
-    memcpy(key->rd_key + 4 * i, block, 16);
-  }
-}
-
-static void aes_nohw_setup_key_192(AES_KEY *key, const uint8_t in[24]) {
-  key->rounds = 12;
-
-  aes_word_t storage1[AES_NOHW_BLOCK_WORDS], storage2[AES_NOHW_BLOCK_WORDS];
-  aes_word_t *block1 = storage1, *block2 = storage2;
-
-  // AES-192's key schedule is complex because each key schedule iteration
-  // produces six words, but we compute on blocks and each block is four words.
-  // We maintain a sliding window of two blocks, filled to 1.5 blocks at a time.
-  // We loop below every three blocks or two key schedule iterations.
-  //
-  // On entry to the loop, |block1| and the first half of |block2| contain the
-  // previous key schedule iteration. |block1| has been written to |key|, but
-  // |block2| has not as it is incomplete.
-  aes_nohw_compact_block(block1, in);
-  memcpy(key->rd_key, block1, 16);
-
-  uint8_t half_block[16] = {0};
-  memcpy(half_block, in + 16, 8);
-  aes_nohw_compact_block(block2, half_block);
-
-  for (size_t i = 0; i < 4; i++) {
-    aes_word_t sub[AES_NOHW_BLOCK_WORDS];
-    aes_nohw_sub_block(sub, block2);
-    uint8_t rcon = aes_nohw_rcon[2 * i];
-    for (size_t j = 0; j < AES_NOHW_BLOCK_WORDS; j++) {
-      // Compute the first two words of the next key schedule iteration, which
-      // go in the second half of |block2|. The first two words of the previous
-      // iteration are in the first half of |block1|. Apply |rcon| here too
-      // because the shifts match.
-      block2[j] = aes_nohw_or(
-          block2[j],
-          aes_nohw_shift_left(
-              aes_nohw_xor(block1[j], aes_nohw_rcon_slice(rcon, j)), 8));
-      // Incorporate the transformed word and propagate. Note the last word of
-      // the previous iteration corresponds to the second word of |copy|. This
-      // is incorporated into the first word of the next iteration, or the third
-      // word of |block2|.
-      block2[j] = aes_nohw_xor(
-          block2[j], aes_nohw_and(aes_nohw_shift_left(
-                                      aes_nohw_rotate_rows_down(sub[j]), 4),
-                                  AES_NOHW_COL2_MASK));
-      block2[j] = aes_nohw_xor(
-          block2[j],
-          aes_nohw_and(aes_nohw_shift_left(block2[j], 4), AES_NOHW_COL3_MASK));
-
-      // Compute the remaining four words, which fill |block1|. Begin by moving
-      // the corresponding words of the previous iteration: the second half of
-      // |block1| and the first half of |block2|.
-      block1[j] = aes_nohw_shift_right(block1[j], 8);
-      block1[j] = aes_nohw_or(block1[j], aes_nohw_shift_left(block2[j], 8));
-      // Incorporate the second word, computed previously in |block2|, and
-      // propagate.
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_right(block2[j], 12));
-      aes_word_t v = block1[j];
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 4));
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 8));
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 12));
-    }
-
-    // This completes two round keys. Note half of |block2| was computed in the
-    // previous loop iteration but was not yet output.
-    memcpy(key->rd_key + 4 * (3 * i + 1), block2, 16);
-    memcpy(key->rd_key + 4 * (3 * i + 2), block1, 16);
-
-    aes_nohw_sub_block(sub, block1);
-    rcon = aes_nohw_rcon[2 * i + 1];
-    for (size_t j = 0; j < AES_NOHW_BLOCK_WORDS; j++) {
-      // Compute the first four words of the next key schedule iteration in
-      // |block2|. Begin by moving the corresponding words of the previous
-      // iteration: the second half of |block2| and the first half of |block1|.
-      block2[j] = aes_nohw_shift_right(block2[j], 8);
-      block2[j] = aes_nohw_or(block2[j], aes_nohw_shift_left(block1[j], 8));
-      // Incorporate rcon and the transformed word. Note the last word of the
-      // previous iteration corresponds to the last word of |copy|.
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_rcon_slice(rcon, j));
-      block2[j] = aes_nohw_xor(
-          block2[j],
-          aes_nohw_shift_right(aes_nohw_rotate_rows_down(sub[j]), 12));
-      // Propagate to the remaining words.
-      aes_word_t v = block2[j];
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 4));
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 8));
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 12));
-
-      // Compute the last two words, which go in the first half of |block1|. The
-      // last two words of the previous iteration are in the second half of
-      // |block1|.
-      block1[j] = aes_nohw_shift_right(block1[j], 8);
-      // Propagate blocks and mask off the excess.
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_right(block2[j], 12));
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(block1[j], 4));
-      block1[j] = aes_nohw_and(block1[j], AES_NOHW_COL01_MASK);
-    }
-
-    // |block2| has a complete round key. |block1| will be completed in the next
-    // iteration.
-    memcpy(key->rd_key + 4 * (3 * i + 3), block2, 16);
-
-    // Swap blocks to restore the invariant.
-    aes_word_t *tmp = block1;
-    block1 = block2;
-    block2 = tmp;
-  }
-}
-
-static void aes_nohw_setup_key_256(AES_KEY *key, const uint8_t in[32]) {
-  key->rounds = 14;
-
-  // Each key schedule iteration produces two round keys.
-  aes_word_t block1[AES_NOHW_BLOCK_WORDS], block2[AES_NOHW_BLOCK_WORDS];
-  aes_nohw_compact_block(block1, in);
-  memcpy(key->rd_key, block1, 16);
-
-  aes_nohw_compact_block(block2, in + 16);
-  memcpy(key->rd_key + 4, block2, 16);
-
-  for (size_t i = 2; i <= 14; i += 2) {
-    aes_word_t sub[AES_NOHW_BLOCK_WORDS];
-    aes_nohw_sub_block(sub, block2);
-    uint8_t rcon = aes_nohw_rcon[i / 2 - 1];
-    for (size_t j = 0; j < AES_NOHW_BLOCK_WORDS; j++) {
-      // Incorporate |rcon| and the transformed word into the first word.
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_rcon_slice(rcon, j));
-      block1[j] = aes_nohw_xor(
-          block1[j],
-          aes_nohw_shift_right(aes_nohw_rotate_rows_down(sub[j]), 12));
-      // Propagate to the remaining words.
-      aes_word_t v = block1[j];
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 4));
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 8));
-      block1[j] = aes_nohw_xor(block1[j], aes_nohw_shift_left(v, 12));
-    }
-    memcpy(key->rd_key + 4 * i, block1, 16);
-
-    if (i == 14) {
-      break;
-    }
-
-    aes_nohw_sub_block(sub, block1);
-    for (size_t j = 0; j < AES_NOHW_BLOCK_WORDS; j++) {
-      // Incorporate the transformed word into the first word.
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_right(sub[j], 12));
-      // Propagate to the remaining words.
-      aes_word_t v = block2[j];
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 4));
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 8));
-      block2[j] = aes_nohw_xor(block2[j], aes_nohw_shift_left(v, 12));
-    }
-    memcpy(key->rd_key + 4 * (i + 1), block2, 16);
-  }
-}
-
-
-// External API.
-
-int aes_nohw_set_encrypt_key(const uint8_t *key, unsigned bits,
-                             AES_KEY *aeskey) {
-  switch (bits) {
-    case 128:
-      aes_nohw_setup_key_128(aeskey, key);
-      return 0;
-    case 192:
-      aes_nohw_setup_key_192(aeskey, key);
-      return 0;
-    case 256:
-      aes_nohw_setup_key_256(aeskey, key);
-      return 0;
-  }
-  return 1;
-}
-
-int aes_nohw_set_decrypt_key(const uint8_t *key, unsigned bits,
-                             AES_KEY *aeskey) {
-  return aes_nohw_set_encrypt_key(key, bits, aeskey);
-}
-
-void aes_nohw_encrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key) {
-  AES_NOHW_SCHEDULE sched;
-  aes_nohw_expand_round_keys(&sched, key);
-  AES_NOHW_BATCH batch;
-  aes_nohw_to_batch(&batch, in, /*num_blocks=*/1);
-  aes_nohw_encrypt_batch(&sched, key->rounds, &batch);
-  aes_nohw_from_batch(out, /*num_blocks=*/1, &batch);
-}
-
-void aes_nohw_decrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key) {
-  AES_NOHW_SCHEDULE sched;
-  aes_nohw_expand_round_keys(&sched, key);
-  AES_NOHW_BATCH batch;
-  aes_nohw_to_batch(&batch, in, /*num_blocks=*/1);
-  aes_nohw_decrypt_batch(&sched, key->rounds, &batch);
-  aes_nohw_from_batch(out, /*num_blocks=*/1, &batch);
-}
-
-static inline void aes_nohw_xor_block(uint8_t out[16], const uint8_t a[16],
-                                      const uint8_t b[16]) {
-  for (size_t i = 0; i < 16; i += sizeof(aes_word_t)) {
-    aes_word_t x, y;
-    memcpy(&x, a + i, sizeof(aes_word_t));
-    memcpy(&y, b + i, sizeof(aes_word_t));
-    x = aes_nohw_xor(x, y);
-    memcpy(out + i, &x, sizeof(aes_word_t));
-  }
-}
-
-void aes_nohw_ctr32_encrypt_blocks(const uint8_t *in, uint8_t *out,
-                                   size_t blocks, const AES_KEY *key,
-                                   const uint8_t ivec[16]) {
-  if (blocks == 0) {
-    return;
-  }
-
-  AES_NOHW_SCHEDULE sched;
-  aes_nohw_expand_round_keys(&sched, key);
-
-  // Make |AES_NOHW_BATCH_SIZE| copies of |ivec|.
-  alignas(AES_NOHW_WORD_SIZE) union {
-    uint32_t u32[AES_NOHW_BATCH_SIZE * 4];
-    uint8_t u8[AES_NOHW_BATCH_SIZE * 16];
-  } ivs, enc_ivs;
-  for (size_t i = 0; i < AES_NOHW_BATCH_SIZE; i++) {
-    memcpy(ivs.u8 + 16 * i, ivec, 16);
-  }
-
-  uint32_t ctr = CRYPTO_bswap4(ivs.u32[3]);
-  for (;;) {
-    // Update counters.
-    for (size_t i = 0; i < AES_NOHW_BATCH_SIZE; i++) {
-      ivs.u32[4 * i + 3] = CRYPTO_bswap4(ctr + i);
-    }
-
-    size_t todo = blocks >= AES_NOHW_BATCH_SIZE ? AES_NOHW_BATCH_SIZE : blocks;
-    AES_NOHW_BATCH batch;
-    aes_nohw_to_batch(&batch, ivs.u8, todo);
-    aes_nohw_encrypt_batch(&sched, key->rounds, &batch);
-    aes_nohw_from_batch(enc_ivs.u8, todo, &batch);
-
-    for (size_t i = 0; i < todo; i++) {
-      aes_nohw_xor_block(out + 16 * i, in + 16 * i, enc_ivs.u8 + 16 * i);
-    }
-
-    blocks -= todo;
-    if (blocks == 0) {
-      break;
-    }
-
-    in += 16 * AES_NOHW_BATCH_SIZE;
-    out += 16 * AES_NOHW_BATCH_SIZE;
-    ctr += AES_NOHW_BATCH_SIZE;
-  }
-}
-
-void aes_nohw_cbc_encrypt(const uint8_t *in, uint8_t *out, size_t len,
-                          const AES_KEY *key, uint8_t *ivec, const int enc) {
-  assert(len % 16 == 0);
-  size_t blocks = len / 16;
-  if (blocks == 0) {
-    return;
-  }
-
-  AES_NOHW_SCHEDULE sched;
-  aes_nohw_expand_round_keys(&sched, key);
-  alignas(AES_NOHW_WORD_SIZE) uint8_t iv[16];
-  memcpy(iv, ivec, 16);
-
-  if (enc) {
-    // CBC encryption is not parallelizable.
-    while (blocks > 0) {
-      aes_nohw_xor_block(iv, iv, in);
-
-      AES_NOHW_BATCH batch;
-      aes_nohw_to_batch(&batch, iv, /*num_blocks=*/1);
-      aes_nohw_encrypt_batch(&sched, key->rounds, &batch);
-      aes_nohw_from_batch(out, /*num_blocks=*/1, &batch);
-
-      memcpy(iv, out, 16);
-
-      in += 16;
-      out += 16;
-      blocks--;
-    }
-    memcpy(ivec, iv, 16);
-    return;
-  }
-
-  for (;;) {
-    size_t todo = blocks >= AES_NOHW_BATCH_SIZE ? AES_NOHW_BATCH_SIZE : blocks;
-    // Make a copy of the input so we can decrypt in-place.
-    alignas(AES_NOHW_WORD_SIZE) uint8_t copy[AES_NOHW_BATCH_SIZE * 16];
-    memcpy(copy, in, todo * 16);
-
-    AES_NOHW_BATCH batch;
-    aes_nohw_to_batch(&batch, in, todo);
-    aes_nohw_decrypt_batch(&sched, key->rounds, &batch);
-    aes_nohw_from_batch(out, todo, &batch);
-
-    aes_nohw_xor_block(out, out, iv);
-    for (size_t i = 1; i < todo; i++) {
-      aes_nohw_xor_block(out + 16 * i, out + 16 * i, copy + 16 * (i - 1));
-    }
-
-    // Save the last block as the IV.
-    memcpy(iv, copy + 16 * (todo - 1), 16);
-
-    blocks -= todo;
-    if (blocks == 0) {
-      break;
-    }
-
-    in += 16 * AES_NOHW_BATCH_SIZE;
-    out += 16 * AES_NOHW_BATCH_SIZE;
-  }
-
-  memcpy(ivec, iv, 16);
-}
diff --git a/crypto/fipsmodule/aes/aes_test.cc b/crypto/fipsmodule/aes/aes_test.cc
index fd7ce24..4c913d3 100644
--- a/crypto/fipsmodule/aes/aes_test.cc
+++ b/crypto/fipsmodule/aes/aes_test.cc
@@ -287,6 +287,26 @@
       block_counts = {0, 1, 8};
     }
 
+    CHECK_ABI(aes_nohw_set_encrypt_key, kKey, bits, &key);
+    CHECK_ABI(aes_nohw_encrypt, block, block, &key);
+#if defined(AES_NOHW_CBC)
+    for (size_t blocks : block_counts) {
+      SCOPED_TRACE(blocks);
+      CHECK_ABI(aes_nohw_cbc_encrypt, buf, buf, AES_BLOCK_SIZE * blocks, &key,
+                block, AES_ENCRYPT);
+    }
+#endif
+
+    CHECK_ABI(aes_nohw_set_decrypt_key, kKey, bits, &key);
+    CHECK_ABI(aes_nohw_decrypt, block, block, &key);
+#if defined(AES_NOHW_CBC)
+    for (size_t blocks : block_counts) {
+      SCOPED_TRACE(blocks);
+      CHECK_ABI(aes_nohw_cbc_encrypt, buf, buf, AES_BLOCK_SIZE * blocks, &key,
+                block, AES_DECRYPT);
+    }
+#endif
+
     if (bsaes_capable()) {
       vpaes_set_encrypt_key(kKey, bits, &key);
       CHECK_ABI(vpaes_encrypt_key_to_bsaes, &key, &key);
diff --git a/crypto/fipsmodule/aes/asm/aes-586.pl b/crypto/fipsmodule/aes/asm/aes-586.pl
new file mode 100755
index 0000000..9b373de
--- /dev/null
+++ b/crypto/fipsmodule/aes/asm/aes-586.pl
@@ -0,0 +1,3000 @@
+#! /usr/bin/env perl
+# Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# Version 4.3.
+#
+# You might fail to appreciate this module performance from the first
+# try. If compared to "vanilla" linux-ia32-icc target, i.e. considered
+# to be *the* best Intel C compiler without -KPIC, performance appears
+# to be virtually identical... But try to re-configure with shared
+# library support... Aha! Intel compiler "suddenly" lags behind by 30%
+# [on P4, more on others]:-) And if compared to position-independent
+# code generated by GNU C, this code performs *more* than *twice* as
+# fast! Yes, all this buzz about PIC means that unlike other hand-
+# coded implementations, this one was explicitly designed to be safe
+# to use even in shared library context... This also means that this
+# code isn't necessarily absolutely fastest "ever," because in order
+# to achieve position independence an extra register has to be
+# off-loaded to stack, which affects the benchmark result.
+#
+# Special note about instruction choice. Do you recall RC4_INT code
+# performing poorly on P4? It might be the time to figure out why.
+# RC4_INT code implies effective address calculations in base+offset*4
+# form. Trouble is that it seems that offset scaling turned to be
+# critical path... At least eliminating scaling resulted in 2.8x RC4
+# performance improvement [as you might recall]. As AES code is hungry
+# for scaling too, I [try to] avoid the latter by favoring off-by-2
+# shifts and masking the result with 0xFF<<2 instead of "boring" 0xFF.
+#
+# As was shown by Dean Gaudet, the above note turned out to be
+# void. Performance improvement with off-by-2 shifts was observed on
+# intermediate implementation, which was spilling yet another register
+# to stack... Final offset*4 code below runs just a tad faster on P4,
+# but exhibits up to 10% improvement on other cores.
+#
+# Second version is "monolithic" replacement for aes_core.c, which in
+# addition to AES_[de|en]crypt implements AES_set_[de|en]cryption_key.
+# This made it possible to implement little-endian variant of the
+# algorithm without modifying the base C code. Motivating factor for
+# the undertaken effort was that it appeared that in tight IA-32
+# register window little-endian flavor could achieve slightly higher
+# Instruction Level Parallelism, and it indeed resulted in up to 15%
+# better performance on most recent µ-archs...
+#
+# Third version adds AES_cbc_encrypt implementation, which resulted in
+# up to 40% performance improvement of CBC benchmark results. 40% was
+# observed on P4 core, where "overall" improvement coefficient, i.e. if
+# compared to PIC generated by GCC and in CBC mode, was observed to be
+# as large as 4x:-) CBC performance is virtually identical to ECB now
+# and on some platforms even better, e.g. 17.6 "small" cycles/byte on
+# Opteron, because certain function prologues and epilogues are
+# effectively taken out of the loop...
+#
+# Version 3.2 implements compressed tables and prefetch of these tables
+# in CBC[!] mode. Former means that 3/4 of table references are now
+# misaligned, which unfortunately has negative impact on elder IA-32
+# implementations, Pentium suffered 30% penalty, PIII - 10%.
+#
+# Version 3.3 avoids L1 cache aliasing between stack frame and
+# S-boxes, and 3.4 - L1 cache aliasing even between key schedule. The
+# latter is achieved by copying the key schedule to controlled place in
+# stack. This unfortunately has rather strong impact on small block CBC
+# performance, ~2x deterioration on 16-byte block if compared to 3.3.
+#
+# Version 3.5 checks if there is L1 cache aliasing between user-supplied
+# key schedule and S-boxes and abstains from copying the former if
+# there is no. This allows end-user to consciously retain small block
+# performance by aligning key schedule in specific manner.
+#
+# Version 3.6 compresses Td4 to 256 bytes and prefetches it in ECB.
+#
+# Current ECB performance numbers for 128-bit key in CPU cycles per
+# processed byte [measure commonly used by AES benchmarkers] are:
+#
+#		small footprint		fully unrolled
+# P4		24			22
+# AMD K8	20			19
+# PIII		25			23
+# Pentium	81			78
+#
+# Version 3.7 reimplements outer rounds as "compact." Meaning that
+# first and last rounds reference compact 256 bytes S-box. This means
+# that first round consumes a lot more CPU cycles and that encrypt
+# and decrypt performance becomes asymmetric. Encrypt performance
+# drops by 10-12%, while decrypt - by 20-25%:-( 256 bytes S-box is
+# aggressively pre-fetched.
+#
+# Version 4.0 effectively rolls back to 3.6 and instead implements
+# additional set of functions, _[x86|sse]_AES_[en|de]crypt_compact,
+# which use exclusively 256 byte S-box. These functions are to be
+# called in modes not concealing plain text, such as ECB, or when
+# we're asked to process smaller amount of data [or unconditionally
+# on hyper-threading CPU]. Currently it's called unconditionally from
+# AES_[en|de]crypt, which affects all modes, but CBC. CBC routine
+# still needs to be modified to switch between slower and faster
+# mode when appropriate... But in either case benchmark landscape
+# changes dramatically and below numbers are CPU cycles per processed
+# byte for 128-bit key.
+#
+#		ECB encrypt	ECB decrypt	CBC large chunk
+# P4		52[54]		83[95]		23
+# AMD K8	46[41]		66[70]		18
+# PIII		41[50]		60[77]		24
+# Core 2	31[36]		45[64]		18.5
+# Atom		76[100]		96[138]		60
+# Pentium	115		150		77
+#
+# Version 4.1 switches to compact S-box even in key schedule setup.
+#
+# Version 4.2 prefetches compact S-box in every SSE round or in other
+# words every cache-line is *guaranteed* to be accessed within ~50
+# cycles window. Why just SSE? Because it's needed on hyper-threading
+# CPU! Which is also why it's prefetched with 64 byte stride. Best
+# part is that it has no negative effect on performance:-)
+#
+# Version 4.3 implements switch between compact and non-compact block
+# functions in AES_cbc_encrypt depending on how much data was asked
+# to be processed in one stroke.
+#
+######################################################################
+# Timing attacks are classified in two classes: synchronous when
+# attacker consciously initiates cryptographic operation and collects
+# timing data of various character afterwards, and asynchronous when
+# malicious code is executed on same CPU simultaneously with AES,
+# instruments itself and performs statistical analysis of this data.
+#
+# As far as synchronous attacks go the root to the AES timing
+# vulnerability is twofold. Firstly, of 256 S-box elements at most 160
+# are referred to in single 128-bit block operation. Well, in C
+# implementation with 4 distinct tables it's actually as little as 40
+# references per 256 elements table, but anyway... Secondly, even
+# though S-box elements are clustered into smaller amount of cache-
+# lines, smaller than 160 and even 40, it turned out that for certain
+# plain-text pattern[s] or simply put chosen plain-text and given key
+# few cache-lines remain unaccessed during block operation. Now, if
+# attacker can figure out this access pattern, he can deduct the key
+# [or at least part of it]. The natural way to mitigate this kind of
+# attacks is to minimize the amount of cache-lines in S-box and/or
+# prefetch them to ensure that every one is accessed for more uniform
+# timing. But note that *if* plain-text was concealed in such way that
+# input to block function is distributed *uniformly*, then attack
+# wouldn't apply. Now note that some encryption modes, most notably
+# CBC, do mask the plain-text in this exact way [secure cipher output
+# is distributed uniformly]. Yes, one still might find input that
+# would reveal the information about given key, but if amount of
+# candidate inputs to be tried is larger than amount of possible key
+# combinations then attack becomes infeasible. This is why revised
+# AES_cbc_encrypt "dares" to switch to larger S-box when larger chunk
+# of data is to be processed in one stroke. The current size limit of
+# 512 bytes is chosen to provide same [diminishingly low] probability
+# for cache-line to remain untouched in large chunk operation with
+# large S-box as for single block operation with compact S-box and
+# surely needs more careful consideration...
+#
+# As for asynchronous attacks. There are two flavours: attacker code
+# being interleaved with AES on hyper-threading CPU at *instruction*
+# level, and two processes time sharing single core. As for latter.
+# Two vectors. 1. Given that attacker process has higher priority,
+# yield execution to process performing AES just before timer fires
+# off the scheduler, immediately regain control of CPU and analyze the
+# cache state. For this attack to be efficient attacker would have to
+# effectively slow down the operation by several *orders* of magnitude,
+# by ratio of time slice to duration of handful of AES rounds, which
+# unlikely to remain unnoticed. Not to mention that this also means
+# that he would spend correspondingly more time to collect enough
+# statistical data to mount the attack. It's probably appropriate to
+# say that if adversary reckons that this attack is beneficial and
+# risks to be noticed, you probably have larger problems having him
+# mere opportunity. In other words suggested code design expects you
+# to preclude/mitigate this attack by overall system security design.
+# 2. Attacker manages to make his code interrupt driven. In order for
+# this kind of attack to be feasible, interrupt rate has to be high
+# enough, again comparable to duration of handful of AES rounds. But
+# is there interrupt source of such rate? Hardly, not even 1Gbps NIC
+# generates interrupts at such raging rate...
+#
+# And now back to the former, hyper-threading CPU or more specifically
+# Intel P4. Recall that asynchronous attack implies that malicious
+# code instruments itself. And naturally instrumentation granularity
+# has be noticeably lower than duration of codepath accessing S-box.
+# Given that all cache-lines are accessed during that time that is.
+# Current implementation accesses *all* cache-lines within ~50 cycles
+# window, which is actually *less* than RDTSC latency on Intel P4!
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+push(@INC,"${dir}","${dir}../../../perlasm");
+require "x86asm.pl";
+
+$output = pop;
+open OUT,">$output";
+*STDOUT=*OUT;
+
+&asm_init($ARGV[0],$x86only = $ARGV[$#ARGV] eq "386");
+&static_label("AES_Te");
+&static_label("AES_Td");
+
+$s0="eax";
+$s1="ebx";
+$s2="ecx";
+$s3="edx";
+$key="edi";
+$acc="esi";
+$tbl="ebp";
+
+# stack frame layout in _[x86|sse]_AES_* routines, frame is allocated
+# by caller
+$__ra=&DWP(0,"esp");	# return address
+$__s0=&DWP(4,"esp");	# s0 backing store
+$__s1=&DWP(8,"esp");	# s1 backing store
+$__s2=&DWP(12,"esp");	# s2 backing store
+$__s3=&DWP(16,"esp");	# s3 backing store
+$__key=&DWP(20,"esp");	# pointer to key schedule
+$__end=&DWP(24,"esp");	# pointer to end of key schedule
+$__tbl=&DWP(28,"esp");	# %ebp backing store
+
+# stack frame layout in AES_[en|crypt] routines, which differs from
+# above by 4 and overlaps by %ebp backing store
+$_tbl=&DWP(24,"esp");
+$_esp=&DWP(28,"esp");
+
+sub _data_word() { my $i; while(defined($i=shift)) { &data_word($i,$i); } }
+
+$speed_limit=512;	# chunks smaller than $speed_limit are
+			# processed with compact routine in CBC mode
+$small_footprint=1;	# $small_footprint=1 code is ~5% slower [on
+			# recent µ-archs], but ~5 times smaller!
+			# I favor compact code to minimize cache
+			# contention and in hope to "collect" 5% back
+			# in real-life applications...
+
+$vertical_spin=0;	# shift "vertically" defaults to 0, because of
+			# its proof-of-concept status...
+# Note that there is no decvert(), as well as last encryption round is
+# performed with "horizontal" shifts. This is because this "vertical"
+# implementation [one which groups shifts on a given $s[i] to form a
+# "column," unlike "horizontal" one, which groups shifts on different
+# $s[i] to form a "row"] is work in progress. It was observed to run
+# few percents faster on Intel cores, but not AMD. On AMD K8 core it's
+# whole 12% slower:-( So we face a trade-off... Shall it be resolved
+# some day? Till then the code is considered experimental and by
+# default remains dormant...
+
+sub encvert()
+{ my ($te,@s) = @_;
+  my ($v0,$v1) = ($acc,$key);
+
+	&mov	($v0,$s[3]);				# copy s3
+	&mov	(&DWP(4,"esp"),$s[2]);			# save s2
+	&mov	($v1,$s[0]);				# copy s0
+	&mov	(&DWP(8,"esp"),$s[1]);			# save s1
+
+	&movz	($s[2],&HB($s[0]));
+	&and	($s[0],0xFF);
+	&mov	($s[0],&DWP(0,$te,$s[0],8));		# s0>>0
+	&shr	($v1,16);
+	&mov	($s[3],&DWP(3,$te,$s[2],8));		# s0>>8
+	&movz	($s[1],&HB($v1));
+	&and	($v1,0xFF);
+	&mov	($s[2],&DWP(2,$te,$v1,8));		# s0>>16
+	 &mov	($v1,$v0);
+	&mov	($s[1],&DWP(1,$te,$s[1],8));		# s0>>24
+
+	&and	($v0,0xFF);
+	&xor	($s[3],&DWP(0,$te,$v0,8));		# s3>>0
+	&movz	($v0,&HB($v1));
+	&shr	($v1,16);
+	&xor	($s[2],&DWP(3,$te,$v0,8));		# s3>>8
+	&movz	($v0,&HB($v1));
+	&and	($v1,0xFF);
+	&xor	($s[1],&DWP(2,$te,$v1,8));		# s3>>16
+	 &mov	($v1,&DWP(4,"esp"));			# restore s2
+	&xor	($s[0],&DWP(1,$te,$v0,8));		# s3>>24
+
+	&mov	($v0,$v1);
+	&and	($v1,0xFF);
+	&xor	($s[2],&DWP(0,$te,$v1,8));		# s2>>0
+	&movz	($v1,&HB($v0));
+	&shr	($v0,16);
+	&xor	($s[1],&DWP(3,$te,$v1,8));		# s2>>8
+	&movz	($v1,&HB($v0));
+	&and	($v0,0xFF);
+	&xor	($s[0],&DWP(2,$te,$v0,8));		# s2>>16
+	 &mov	($v0,&DWP(8,"esp"));			# restore s1
+	&xor	($s[3],&DWP(1,$te,$v1,8));		# s2>>24
+
+	&mov	($v1,$v0);
+	&and	($v0,0xFF);
+	&xor	($s[1],&DWP(0,$te,$v0,8));		# s1>>0
+	&movz	($v0,&HB($v1));
+	&shr	($v1,16);
+	&xor	($s[0],&DWP(3,$te,$v0,8));		# s1>>8
+	&movz	($v0,&HB($v1));
+	&and	($v1,0xFF);
+	&xor	($s[3],&DWP(2,$te,$v1,8));		# s1>>16
+	 &mov	($key,$__key);				# reincarnate v1 as key
+	&xor	($s[2],&DWP(1,$te,$v0,8));		# s1>>24
+}
+
+# Another experimental routine, which features "horizontal spin," but
+# eliminates one reference to stack. Strangely enough runs slower...
+sub enchoriz()
+{ my ($v0,$v1) = ($key,$acc);
+
+	&movz	($v0,&LB($s0));			#  3, 2, 1, 0*
+	&rotr	($s2,8);			#  8,11,10, 9
+	&mov	($v1,&DWP(0,$te,$v0,8));	#  0
+	&movz	($v0,&HB($s1));			#  7, 6, 5*, 4
+	&rotr	($s3,16);			# 13,12,15,14
+	&xor	($v1,&DWP(3,$te,$v0,8));	#  5
+	&movz	($v0,&HB($s2));			#  8,11,10*, 9
+	&rotr	($s0,16);			#  1, 0, 3, 2
+	&xor	($v1,&DWP(2,$te,$v0,8));	# 10
+	&movz	($v0,&HB($s3));			# 13,12,15*,14
+	&xor	($v1,&DWP(1,$te,$v0,8));	# 15, t[0] collected
+	&mov	($__s0,$v1);			# t[0] saved
+
+	&movz	($v0,&LB($s1));			#  7, 6, 5, 4*
+	&shr	($s1,16);			#  -, -, 7, 6
+	&mov	($v1,&DWP(0,$te,$v0,8));	#  4
+	&movz	($v0,&LB($s3));			# 13,12,15,14*
+	&xor	($v1,&DWP(2,$te,$v0,8));	# 14
+	&movz	($v0,&HB($s0));			#  1, 0, 3*, 2
+	&and	($s3,0xffff0000);		# 13,12, -, -
+	&xor	($v1,&DWP(1,$te,$v0,8));	#  3
+	&movz	($v0,&LB($s2));			#  8,11,10, 9*
+	&or	($s3,$s1);			# 13,12, 7, 6
+	&xor	($v1,&DWP(3,$te,$v0,8));	#  9, t[1] collected
+	&mov	($s1,$v1);			#  s[1]=t[1]
+
+	&movz	($v0,&LB($s0));			#  1, 0, 3, 2*
+	&shr	($s2,16);			#  -, -, 8,11
+	&mov	($v1,&DWP(2,$te,$v0,8));	#  2
+	&movz	($v0,&HB($s3));			# 13,12, 7*, 6
+	&xor	($v1,&DWP(1,$te,$v0,8));	#  7
+	&movz	($v0,&HB($s2));			#  -, -, 8*,11
+	&xor	($v1,&DWP(0,$te,$v0,8));	#  8
+	&mov	($v0,$s3);
+	&shr	($v0,24);			# 13
+	&xor	($v1,&DWP(3,$te,$v0,8));	# 13, t[2] collected
+
+	&movz	($v0,&LB($s2));			#  -, -, 8,11*
+	&shr	($s0,24);			#  1*
+	&mov	($s2,&DWP(1,$te,$v0,8));	# 11
+	&xor	($s2,&DWP(3,$te,$s0,8));	#  1
+	&mov	($s0,$__s0);			# s[0]=t[0]
+	&movz	($v0,&LB($s3));			# 13,12, 7, 6*
+	&shr	($s3,16);			#   ,  ,13,12
+	&xor	($s2,&DWP(2,$te,$v0,8));	#  6
+	&mov	($key,$__key);			# reincarnate v0 as key
+	&and	($s3,0xff);			#   ,  ,13,12*
+	&mov	($s3,&DWP(0,$te,$s3,8));	# 12
+	&xor	($s3,$s2);			# s[2]=t[3] collected
+	&mov	($s2,$v1);			# s[2]=t[2]
+}
+
+# More experimental code... SSE one... Even though this one eliminates
+# *all* references to stack, it's not faster...
+sub sse_encbody()
+{
+	&movz	($acc,&LB("eax"));		#  0
+	&mov	("ecx",&DWP(0,$tbl,$acc,8));	#  0
+	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
+	&movz	("edx",&HB("eax"));		#  1
+	&mov	("edx",&DWP(3,$tbl,"edx",8));	#  1
+	&shr	("eax",16);			#  5, 4
+
+	&movz	($acc,&LB("ebx"));		# 10
+	&xor	("ecx",&DWP(2,$tbl,$acc,8));	# 10
+	&pshufw	("mm6","mm4",0x08);		# 13,12, 9, 8
+	&movz	($acc,&HB("ebx"));		# 11
+	&xor	("edx",&DWP(1,$tbl,$acc,8));	# 11
+	&shr	("ebx",16);			# 15,14
+
+	&movz	($acc,&HB("eax"));		#  5
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	#  5
+	&movq	("mm3",QWP(16,$key));
+	&movz	($acc,&HB("ebx"));		# 15
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	# 15
+	&movd	("mm0","ecx");			# t[0] collected
+
+	&movz	($acc,&LB("eax"));		#  4
+	&mov	("ecx",&DWP(0,$tbl,$acc,8));	#  4
+	&movd	("eax","mm2");			#  7, 6, 3, 2
+	&movz	($acc,&LB("ebx"));		# 14
+	&xor	("ecx",&DWP(2,$tbl,$acc,8));	# 14
+	&movd	("ebx","mm6");			# 13,12, 9, 8
+
+	&movz	($acc,&HB("eax"));		#  3
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	#  3
+	&movz	($acc,&HB("ebx"));		#  9
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	#  9
+	&movd	("mm1","ecx");			# t[1] collected
+
+	&movz	($acc,&LB("eax"));		#  2
+	&mov	("ecx",&DWP(2,$tbl,$acc,8));	#  2
+	&shr	("eax",16);			#  7, 6
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+	&movz	($acc,&LB("ebx"));		#  8
+	&xor	("ecx",&DWP(0,$tbl,$acc,8));	#  8
+	&shr	("ebx",16);			# 13,12
+
+	&movz	($acc,&HB("eax"));		#  7
+	&xor	("ecx",&DWP(1,$tbl,$acc,8));	#  7
+	&pxor	("mm0","mm3");
+	&movz	("eax",&LB("eax"));		#  6
+	&xor	("edx",&DWP(2,$tbl,"eax",8));	#  6
+	&pshufw	("mm1","mm0",0x08);		#  5, 4, 1, 0
+	&movz	($acc,&HB("ebx"));		# 13
+	&xor	("ecx",&DWP(3,$tbl,$acc,8));	# 13
+	&xor	("ecx",&DWP(24,$key));		# t[2]
+	&movd	("mm4","ecx");			# t[2] collected
+	&movz	("ebx",&LB("ebx"));		# 12
+	&xor	("edx",&DWP(0,$tbl,"ebx",8));	# 12
+	&shr	("ecx",16);
+	&movd	("eax","mm1");			#  5, 4, 1, 0
+	&mov	("ebx",&DWP(28,$key));		# t[3]
+	&xor	("ebx","edx");
+	&movd	("mm5","ebx");			# t[3] collected
+	&and	("ebx",0xffff0000);
+	&or	("ebx","ecx");
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+######################################################################
+# "Compact" block function
+######################################################################
+
+sub enccompact()
+{ my $Fn = \&mov;
+  while ($#_>5) { pop(@_); $Fn=sub{}; }
+  my ($i,$te,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# $Fn is used in first compact round and its purpose is to
+	# void restoration of some values from stack, so that after
+	# 4xenccompact with extra argument $key value is left there...
+	if ($i==3)  {	&$Fn	($key,$__key);			}##%edx
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
+	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
+			&movz	($out,&BP(-128,$te,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}##%eax
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,8);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$__s0);		}##%ebx
+	else        {	&mov	($tmp,$s[2]);
+			&shr	($tmp,16);			}
+	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
+			&and	($tmp,0xFF);
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}##%ecx
+	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
+	else        {	&mov	($tmp,$s[3]);
+			&shr	($tmp,24);			}
+			&movz	($tmp,&BP(-128,$te,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$acc);			}
+	&comment();
+}
+
+sub enctransform()
+{ my @s = ($s0,$s1,$s2,$s3);
+  my $i = shift;
+  my $tmp = $tbl;
+  my $r2  = $key ;
+
+	&and	($tmp,$s[$i]);
+	&lea	($r2,&DWP(0,$s[$i],$s[$i]));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&and	($r2,0xfefefefe);
+	&sub	($acc,$tmp);
+	&mov	($tmp,$s[$i]);
+	&and	($acc,0x1b1b1b1b);
+	&rotr	($tmp,16);
+	&xor	($acc,$r2);	# r2
+	&mov	($r2,$s[$i]);
+
+	&xor	($s[$i],$acc);	# r0 ^ r2
+	&rotr	($r2,16+8);
+	&xor	($acc,$tmp);
+	&rotl	($s[$i],24);
+	&xor	($acc,$r2);
+	&mov	($tmp,0x80808080)	if ($i!=1);
+	&xor	($s[$i],$acc);	# ROTATE(r2^r0,24) ^ r2
+}
+
+&function_begin_B("_x86_AES_encrypt_compact");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($s0,&DWP(0,$key));		# xor with key
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	# prefetch Te4
+	&mov	($key,&DWP(0-128,$tbl));
+	&mov	($acc,&DWP(32-128,$tbl));
+	&mov	($key,&DWP(64-128,$tbl));
+	&mov	($acc,&DWP(96-128,$tbl));
+	&mov	($key,&DWP(128-128,$tbl));
+	&mov	($acc,&DWP(160-128,$tbl));
+	&mov	($key,&DWP(192-128,$tbl));
+	&mov	($acc,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+
+		&enccompact(0,$tbl,$s0,$s1,$s2,$s3,1);
+		&enccompact(1,$tbl,$s1,$s2,$s3,$s0,1);
+		&enccompact(2,$tbl,$s2,$s3,$s0,$s1,1);
+		&enccompact(3,$tbl,$s3,$s0,$s1,$s2,1);
+		&mov	($tbl,0x80808080);
+		&enctransform(2);
+		&enctransform(3);
+		&enctransform(0);
+		&enctransform(1);
+		&mov 	($key,$__key);
+		&mov	($tbl,$__tbl);
+		&add	($key,16);		# advance rd_key
+		&xor	($s0,&DWP(0,$key));
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+
+	&cmp	($key,$__end);
+	&mov	($__key,$key);
+	&jb	(&label("loop"));
+
+	&enccompact(0,$tbl,$s0,$s1,$s2,$s3);
+	&enccompact(1,$tbl,$s1,$s2,$s3,$s0);
+	&enccompact(2,$tbl,$s2,$s3,$s0,$s1);
+	&enccompact(3,$tbl,$s3,$s0,$s1,$s2);
+
+	&xor	($s0,&DWP(16,$key));
+	&xor	($s1,&DWP(20,$key));
+	&xor	($s2,&DWP(24,$key));
+	&xor	($s3,&DWP(28,$key));
+
+	&ret	();
+&function_end_B("_x86_AES_encrypt_compact");
+
+######################################################################
+# "Compact" SSE block function.
+######################################################################
+#
+# Performance is not actually extraordinary in comparison to pure
+# x86 code. In particular encrypt performance is virtually the same.
+# Decrypt performance on the other hand is 15-20% better on newer
+# µ-archs [but we're thankful for *any* improvement here], and ~50%
+# better on PIII:-) And additionally on the pros side this code
+# eliminates redundant references to stack and thus relieves/
+# minimizes the pressure on the memory bus.
+#
+# MMX register layout                           lsb
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |          mm4          |          mm0          |
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |     s3    |     s2    |     s1    |     s0    |
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+# |15|14|13|12|11|10| 9| 8| 7| 6| 5| 4| 3| 2| 1| 0|
+# +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
+#
+# Indexes translate as s[N/4]>>(8*(N%4)), e.g. 5 means s1>>8.
+# In this terms encryption and decryption "compact" permutation
+# matrices can be depicted as following:
+#
+# encryption              lsb	# decryption              lsb
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t0 || 15 | 10 |  5 |  0 |	# | t0 ||  7 | 10 | 13 |  0 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t1 ||  3 | 14 |  9 |  4 |	# | t1 || 11 | 14 |  1 |  4 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t2 ||  7 |  2 | 13 |  8 |	# | t2 || 15 |  2 |  5 |  8 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+# | t3 || 11 |  6 |  1 | 12 |	# | t3 ||  3 |  6 |  9 | 12 |
+# +----++----+----+----+----+	# +----++----+----+----+----+
+#
+######################################################################
+# Why not xmm registers? Short answer. It was actually tested and
+# was not any faster, but *contrary*, most notably on Intel CPUs.
+# Longer answer. Main advantage of using mm registers is that movd
+# latency is lower, especially on Intel P4. While arithmetic
+# instructions are twice as many, they can be scheduled every cycle
+# and not every second one when they are operating on xmm register,
+# so that "arithmetic throughput" remains virtually the same. And
+# finally the code can be executed even on elder SSE-only CPUs:-)
+
+sub sse_enccompact()
+{
+	&pshufw	("mm1","mm0",0x08);		#  5, 4, 1, 0
+	&pshufw	("mm5","mm4",0x0d);		# 15,14,11,10
+	&movd	("eax","mm1");			#  5, 4, 1, 0
+	&movd	("ebx","mm5");			# 15,14,11,10
+	&mov	($__key,$key);
+
+	&movz	($acc,&LB("eax"));		#  0
+	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x0d);		#  7, 6, 3, 2
+	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
+	&movz	($key,&LB("ebx"));		# 10
+	&movz	("edx",&BP(-128,$tbl,"edx",1));	#  1
+	&shr	("eax",16);			#  5, 4
+	&shl	("edx",8);			#  1
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 10
+	&movz	($key,&HB("ebx"));		# 11
+	&shl	($acc,16);			# 10
+	&pshufw	("mm6","mm4",0x08);		# 13,12, 9, 8
+	&or	("ecx",$acc);			# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  5
+	&shl	($acc,24);			# 11
+	&shr	("ebx",16);			# 15,14
+	&or	("edx",$acc);			# 11
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  5
+	&movz	($key,&HB("ebx"));		# 15
+	&shl	($acc,8);			#  5
+	&or	("ecx",$acc);			#  5
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 15
+	&movz	($key,&LB("eax"));		#  4
+	&shl	($acc,24);			# 15
+	&or	("ecx",$acc);			# 15
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
+	&movd	("eax","mm2");			#  7, 6, 3, 2
+	&movd	("mm0","ecx");			# t[0] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	# 14
+	&movz	($key,&HB("eax"));		#  3
+	&shl	("ecx",16);			# 14
+	&movd	("ebx","mm6");			# 13,12, 9, 8
+	&or	("ecx",$acc);			# 14
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  3
+	&movz	($key,&HB("ebx"));		#  9
+	&shl	($acc,24);			#  3
+	&or	("ecx",$acc);			#  3
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  9
+	&movz	($key,&LB("ebx"));		#  8
+	&shl	($acc,8);			#  9
+	&shr	("ebx",16);			# 13,12
+	&or	("ecx",$acc);			#  9
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  8
+	&movz	($key,&LB("eax"));		#  2
+	&shr	("eax",16);			#  7, 6
+	&movd	("mm1","ecx");			# t[1] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	#  2
+	&movz	($key,&HB("eax"));		#  7
+	&shl	("ecx",16);			#  2
+	&and	("eax",0xff);			#  6
+	&or	("ecx",$acc);			#  2
+
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
+	&shl	($acc,24);			#  7
+	&and	("ebx",0xff);			# 12
+	&movz	("eax",&BP(-128,$tbl,"eax",1));	#  6
+	&or	("ecx",$acc);			#  7
+	&shl	("eax",16);			#  6
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 13
+	&or	("edx","eax");			#  6
+	&shl	($acc,8);			# 13
+	&movz	("ebx",&BP(-128,$tbl,"ebx",1));	# 12
+	&or	("ecx",$acc);			# 13
+	&or	("edx","ebx");			# 12
+	&mov	($key,$__key);
+	&movd	("mm4","ecx");			# t[2] collected
+	&movd	("mm5","edx");			# t[3] collected
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+					if (!$x86only) {
+&function_begin_B("_sse_AES_encrypt_compact");
+	&pxor	("mm0",&QWP(0,$key));	#  7, 6, 5, 4, 3, 2, 1, 0
+	&pxor	("mm4",&QWP(8,$key));	# 15,14,13,12,11,10, 9, 8
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	&mov	($s0,0x1b1b1b1b);		# magic constant
+	&mov	(&DWP(8,"esp"),$s0);
+	&mov	(&DWP(12,"esp"),$s0);
+
+	# prefetch Te4
+	&mov	($s0,&DWP(0-128,$tbl));
+	&mov	($s1,&DWP(32-128,$tbl));
+	&mov	($s2,&DWP(64-128,$tbl));
+	&mov	($s3,&DWP(96-128,$tbl));
+	&mov	($s0,&DWP(128-128,$tbl));
+	&mov	($s1,&DWP(160-128,$tbl));
+	&mov	($s2,&DWP(192-128,$tbl));
+	&mov	($s3,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+		&sse_enccompact();
+		&add	($key,16);
+		&cmp	($key,$__end);
+		&ja	(&label("out"));
+
+		&movq	("mm2",&QWP(8,"esp"));
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&movq	("mm1","mm0");		&movq	("mm5","mm4");	# r0
+		&pcmpgtb("mm3","mm0");		&pcmpgtb("mm7","mm4");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&pshufw	("mm2","mm0",0xb1);	&pshufw	("mm6","mm4",0xb1);# ROTATE(r0,16)
+		&paddb	("mm0","mm0");		&paddb	("mm4","mm4");
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# = r2
+		&pshufw	("mm3","mm2",0xb1);	&pshufw	("mm7","mm6",0xb1);# r0
+		&pxor	("mm1","mm0");		&pxor	("mm5","mm4");	# r0^r2
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= ROTATE(r0,16)
+
+		&movq	("mm2","mm3");		&movq	("mm6","mm7");
+		&pslld	("mm3",8);		&pslld	("mm7",8);
+		&psrld	("mm2",24);		&psrld	("mm6",24);
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= r0<<8
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= r0>>24
+
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&movq	("mm2",&QWP(0,$key));	&movq	("mm6",&QWP(8,$key));
+		&psrld	("mm1",8);		&psrld	("mm5",8);
+		&mov	($s0,&DWP(0-128,$tbl));
+		&pslld	("mm3",24);		&pslld	("mm7",24);
+		&mov	($s1,&DWP(64-128,$tbl));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= (r2^r0)<<8
+		&mov	($s2,&DWP(128-128,$tbl));
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= (r2^r0)>>24
+		&mov	($s3,&DWP(192-128,$tbl));
+
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");
+	&jmp	(&label("loop"));
+
+	&set_label("out",16);
+	&pxor	("mm0",&QWP(0,$key));
+	&pxor	("mm4",&QWP(8,$key));
+
+	&ret	();
+&function_end_B("_sse_AES_encrypt_compact");
+					}
+
+######################################################################
+# Vanilla block function.
+######################################################################
+
+sub encstep()
+{ my ($i,$te,@s) = @_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# lines marked with #%e?x[i] denote "reordered" instructions...
+	if ($i==3)  {	&mov	($key,$__key);			}##%edx
+	else        {	&mov	($out,$s[0]);
+			&and	($out,0xFF);			}
+	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
+	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
+			&mov	($out,&DWP(0,$te,$out,8));
+
+	if ($i==3)  {	$tmp=$s[1];				}##%eax
+			&movz	($tmp,&HB($s[1]));
+			&xor	($out,&DWP(3,$te,$tmp,8));
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$__s0);		}##%ebx
+	else        {	&mov	($tmp,$s[2]);
+			&shr	($tmp,16);			}
+	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
+			&and	($tmp,0xFF);
+			&xor	($out,&DWP(2,$te,$tmp,8));
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}##%ecx
+	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
+	else        {	&mov	($tmp,$s[3]);
+			&shr	($tmp,24)			}
+			&xor	($out,&DWP(1,$te,$tmp,8));
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$acc);			}
+			&comment();
+}
+
+sub enclast()
+{ my ($i,$te,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	if ($i==3)  {	&mov	($key,$__key);			}##%edx
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+	if ($i==1)  {	&shr	($s[0],16);			}#%ebx[1]
+	if ($i==2)  {	&shr	($s[0],24);			}#%ecx[2]
+			&mov	($out,&DWP(2,$te,$out,8));
+			&and	($out,0x000000ff);
+
+	if ($i==3)  {	$tmp=$s[1];				}##%eax
+			&movz	($tmp,&HB($s[1]));
+			&mov	($tmp,&DWP(0,$te,$tmp,8));
+			&and	($tmp,0x0000ff00);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$__s0);		}##%ebx
+	else        {	&mov	($tmp,$s[2]);
+			&shr	($tmp,16);			}
+	if ($i==2)  {	&and	($s[1],0xFF);			}#%edx[2]
+			&and	($tmp,0xFF);
+			&mov	($tmp,&DWP(0,$te,$tmp,8));
+			&and	($tmp,0x00ff0000);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}##%ecx
+	elsif($i==2){	&movz	($tmp,&HB($s[3]));		}#%ebx[2]
+	else        {	&mov	($tmp,$s[3]);
+			&shr	($tmp,24);			}
+			&mov	($tmp,&DWP(2,$te,$tmp,8));
+			&and	($tmp,0xff000000);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$acc);			}
+}
+
+&function_begin_B("_x86_AES_encrypt");
+	if ($vertical_spin) {
+		# I need high parts of volatile registers to be accessible...
+		&exch	($s1="edi",$key="ebx");
+		&mov	($s2="esi",$acc="ecx");
+	}
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($s0,&DWP(0,$key));		# xor with key
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+
+	if ($small_footprint) {
+	    &lea	($acc,&DWP(-2,$acc,$acc));
+	    &lea	($acc,&DWP(0,$key,$acc,8));
+	    &mov	($__end,$acc);		# end of key schedule
+
+	    &set_label("loop",16);
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$s0,$s1,$s2);
+		}
+		&add	($key,16);		# advance rd_key
+		&xor	($s0,&DWP(0,$key));
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+	    &cmp	($key,$__end);
+	    &mov	($__key,$key);
+	    &jb		(&label("loop"));
+	}
+	else {
+	    &cmp	($acc,10);
+	    &jle	(&label("10rounds"));
+	    &cmp	($acc,12);
+	    &jle	(&label("12rounds"));
+
+	&set_label("14rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$s0,$s1,$s2);
+		}
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	    &add	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("12rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$s0,$s1,$s2);
+		}
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	    &add	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("10rounds",4);
+	    for ($i=1;$i<10;$i++) {
+		if ($vertical_spin) {
+		    &encvert($tbl,$s0,$s1,$s2,$s3);
+		} else {
+		    &encstep(0,$tbl,$s0,$s1,$s2,$s3);
+		    &encstep(1,$tbl,$s1,$s2,$s3,$s0);
+		    &encstep(2,$tbl,$s2,$s3,$s0,$s1);
+		    &encstep(3,$tbl,$s3,$s0,$s1,$s2);
+		}
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	}
+
+	if ($vertical_spin) {
+	    # "reincarnate" some registers for "horizontal" spin...
+	    &mov	($s1="ebx",$key="edi");
+	    &mov	($s2="ecx",$acc="esi");
+	}
+	&enclast(0,$tbl,$s0,$s1,$s2,$s3);
+	&enclast(1,$tbl,$s1,$s2,$s3,$s0);
+	&enclast(2,$tbl,$s2,$s3,$s0,$s1);
+	&enclast(3,$tbl,$s3,$s0,$s1,$s2);
+
+	&add	($key,$small_footprint?16:160);
+	&xor	($s0,&DWP(0,$key));
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&ret	();
+
+&set_label("AES_Te",64);	# Yes! I keep it in the code segment!
+	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
+	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
+	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
+	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
+	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
+	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
+	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
+	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
+	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
+	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
+	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
+	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
+	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
+	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
+	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
+	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
+	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
+	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
+	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
+	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
+	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
+	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
+	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
+	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
+	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
+	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
+	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
+	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
+	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
+	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
+	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
+	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
+	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
+	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
+	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
+	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
+	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
+	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
+	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
+	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
+	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
+	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
+	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
+	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
+	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
+	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
+	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
+	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
+	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
+	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
+	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
+	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
+	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
+	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
+	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
+	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
+	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
+	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
+	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
+	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
+	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
+	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
+	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
+	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
+
+#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+#rcon:
+	&data_word(0x00000001, 0x00000002, 0x00000004, 0x00000008);
+	&data_word(0x00000010, 0x00000020, 0x00000040, 0x00000080);
+	&data_word(0x0000001b, 0x00000036, 0x00000000, 0x00000000);
+	&data_word(0x00000000, 0x00000000, 0x00000000, 0x00000000);
+&function_end_B("_x86_AES_encrypt");
+
+# void aes_nohw_encrypt (const void *inp,void *out,const AES_KEY *key);
+&function_begin("aes_nohw_encrypt");
+	&mov	($acc,&wparam(0));		# load inp
+	&mov	($key,&wparam(2));		# load key
+
+	&mov	($s0,"esp");
+	&sub	("esp",36);
+	&and	("esp",-64);			# align to cache-line
+
+	# place stack frame just "above" the key schedule
+	&lea	($s1,&DWP(-64-63,$key));
+	&sub	($s1,"esp");
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	("esp",$s1);
+	&add	("esp",4);	# 4 is reserved for caller's return address
+	&mov	($_esp,$s0);			# save stack pointer
+
+	&call   (&label("pic_point"));          # make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if (!$x86only);
+	&lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+
+	# pick Te4 copy which can't "overlap" with stack frame or key schedule
+	&lea	($s1,&DWP(768-4,"esp"));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),25);	# check for SSE bit
+	&jnc	(&label("x86"));
+
+	&movq	("mm0",&QWP(0,$acc));
+	&movq	("mm4",&QWP(8,$acc));
+	&call	("_sse_AES_encrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&movq	(&QWP(0,$acc),"mm0");		# write output data
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&function_end_A();
+					}
+	&set_label("x86",16);
+	&mov	($_tbl,$tbl);
+	&mov	($s0,&DWP(0,$acc));		# load input data
+	&mov	($s1,&DWP(4,$acc));
+	&mov	($s2,&DWP(8,$acc));
+	&mov	($s3,&DWP(12,$acc));
+	&call	("_x86_AES_encrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&mov	(&DWP(0,$acc),$s0);		# write output data
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+&function_end("aes_nohw_encrypt");
+
+#--------------------------------------------------------------------#
+
+######################################################################
+# "Compact" block function
+######################################################################
+
+sub deccompact()
+{ my $Fn = \&mov;
+  while ($#_>5) { pop(@_); $Fn=sub{}; }
+  my ($i,$td,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# $Fn is used in first compact round and its purpose is to
+	# void restoration of some values from stack, so that after
+	# 4xdeccompact with extra argument $key, $s0 and $s1 values
+	# are left there...
+	if($i==3)   {	&$Fn	($key,$__key);			}
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+			&movz	($out,&BP(-128,$td,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(-128,$td,$tmp,1));
+			&shl	($tmp,8);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
+	else        {	mov	($tmp,$s[2]);			}
+			&shr	($tmp,16);
+			&and	($tmp,0xFF);
+			&movz	($tmp,&BP(-128,$td,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &$Fn ($s[2],$__s1);		}
+	else        {	&mov	($tmp,$s[3]);			}
+			&shr	($tmp,24);
+			&movz	($tmp,&BP(-128,$td,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&$Fn	($s[3],$__s0);			}
+}
+
+# must be called with 2,3,0,1 as argument sequence!!!
+sub dectransform()
+{ my @s = ($s0,$s1,$s2,$s3);
+  my $i = shift;
+  my $tmp = $key;
+  my $tp2 = @s[($i+2)%4]; $tp2 = @s[2] if ($i==1);
+  my $tp4 = @s[($i+3)%4]; $tp4 = @s[3] if ($i==1);
+  my $tp8 = $tbl;
+
+	&mov	($tmp,0x80808080);
+	&and	($tmp,$s[$i]);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp2,&DWP(0,$s[$i],$s[$i]));
+	&sub	($acc,$tmp);
+	&and	($tp2,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp2);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp4,&DWP(0,$tp2,$tp2));
+	&sub	($acc,$tmp);
+	&and	($tp4,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp2,$s[$i]);	# tp2^tp1
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp4);
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&lea	($tp8,&DWP(0,$tp4,$tp4));
+	&sub	($acc,$tmp);
+	&and	($tp8,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp4,$s[$i]);	# tp4^tp1
+	 &rotl	($s[$i],8);	# = ROTATE(tp1,8)
+	&xor	($tp8,$acc);
+
+	&xor	($s[$i],$tp2);
+	&xor	($tp2,$tp8);
+	&xor	($s[$i],$tp4);
+	&xor	($tp4,$tp8);
+	&rotl	($tp2,24);
+	&xor	($s[$i],$tp8);	# ^= tp8^(tp4^tp1)^(tp2^tp1)
+	&rotl	($tp4,16);
+	&xor	($s[$i],$tp2);	# ^= ROTATE(tp8^tp2^tp1,24)
+	&rotl	($tp8,8);
+	&xor	($s[$i],$tp4);	# ^= ROTATE(tp8^tp4^tp1,16)
+	 &mov	($s[0],$__s0)			if($i==2); #prefetch $s0
+	 &mov	($s[1],$__s1)			if($i==3); #prefetch $s1
+	 &mov	($s[2],$__s2)			if($i==1);
+	&xor	($s[$i],$tp8);	# ^= ROTATE(tp8,8)
+
+	&mov	($s[3],$__s3)			if($i==1);
+	&mov	(&DWP(4+4*$i,"esp"),$s[$i])	if($i>=2);
+}
+
+&function_begin_B("_x86_AES_decrypt_compact");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($s0,&DWP(0,$key));		# xor with key
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	# prefetch Td4
+	&mov	($key,&DWP(0-128,$tbl));
+	&mov	($acc,&DWP(32-128,$tbl));
+	&mov	($key,&DWP(64-128,$tbl));
+	&mov	($acc,&DWP(96-128,$tbl));
+	&mov	($key,&DWP(128-128,$tbl));
+	&mov	($acc,&DWP(160-128,$tbl));
+	&mov	($key,&DWP(192-128,$tbl));
+	&mov	($acc,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+
+		&deccompact(0,$tbl,$s0,$s3,$s2,$s1,1);
+		&deccompact(1,$tbl,$s1,$s0,$s3,$s2,1);
+		&deccompact(2,$tbl,$s2,$s1,$s0,$s3,1);
+		&deccompact(3,$tbl,$s3,$s2,$s1,$s0,1);
+		&dectransform(2);
+		&dectransform(3);
+		&dectransform(0);
+		&dectransform(1);
+		&mov 	($key,$__key);
+		&mov	($tbl,$__tbl);
+		&add	($key,16);		# advance rd_key
+		&xor	($s0,&DWP(0,$key));
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+
+	&cmp	($key,$__end);
+	&mov	($__key,$key);
+	&jb	(&label("loop"));
+
+	&deccompact(0,$tbl,$s0,$s3,$s2,$s1);
+	&deccompact(1,$tbl,$s1,$s0,$s3,$s2);
+	&deccompact(2,$tbl,$s2,$s1,$s0,$s3);
+	&deccompact(3,$tbl,$s3,$s2,$s1,$s0);
+
+	&xor	($s0,&DWP(16,$key));
+	&xor	($s1,&DWP(20,$key));
+	&xor	($s2,&DWP(24,$key));
+	&xor	($s3,&DWP(28,$key));
+
+	&ret	();
+&function_end_B("_x86_AES_decrypt_compact");
+
+######################################################################
+# "Compact" SSE block function.
+######################################################################
+
+sub sse_deccompact()
+{
+	&pshufw	("mm1","mm0",0x0c);		#  7, 6, 1, 0
+	&pshufw	("mm5","mm4",0x09);		# 13,12,11,10
+	&movd	("eax","mm1");			#  7, 6, 1, 0
+	&movd	("ebx","mm5");			# 13,12,11,10
+	&mov	($__key,$key);
+
+	&movz	($acc,&LB("eax"));		#  0
+	&movz	("edx",&HB("eax"));		#  1
+	&pshufw	("mm2","mm0",0x06);		#  3, 2, 5, 4
+	&movz	("ecx",&BP(-128,$tbl,$acc,1));	#  0
+	&movz	($key,&LB("ebx"));		# 10
+	&movz	("edx",&BP(-128,$tbl,"edx",1));	#  1
+	&shr	("eax",16);			#  7, 6
+	&shl	("edx",8);			#  1
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 10
+	&movz	($key,&HB("ebx"));		# 11
+	&shl	($acc,16);			# 10
+	&pshufw	("mm6","mm4",0x03);		# 9, 8,15,14
+	&or	("ecx",$acc);			# 10
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 11
+	&movz	($key,&HB("eax"));		#  7
+	&shl	($acc,24);			# 11
+	&shr	("ebx",16);			# 13,12
+	&or	("edx",$acc);			# 11
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  7
+	&movz	($key,&HB("ebx"));		# 13
+	&shl	($acc,24);			#  7
+	&or	("ecx",$acc);			#  7
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 13
+	&movz	($key,&LB("eax"));		#  6
+	&shl	($acc,8);			# 13
+	&movd	("eax","mm2");			#  3, 2, 5, 4
+	&or	("ecx",$acc);			# 13
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  6
+	&movz	($key,&LB("ebx"));		# 12
+	&shl	($acc,16);			#  6
+	&movd	("ebx","mm6");			#  9, 8,15,14
+	&movd	("mm0","ecx");			# t[0] collected
+	&movz	("ecx",&BP(-128,$tbl,$key,1));	# 12
+	&movz	($key,&LB("eax"));		#  4
+	&or	("ecx",$acc);			# 12
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  4
+	&movz	($key,&LB("ebx"));		# 14
+	&or	("edx",$acc);			#  4
+	&movz	($acc,&BP(-128,$tbl,$key,1));	# 14
+	&movz	($key,&HB("eax"));		#  5
+	&shl	($acc,16);			# 14
+	&shr	("eax",16);			#  3, 2
+	&or	("edx",$acc);			# 14
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  5
+	&movz	($key,&HB("ebx"));		# 15
+	&shr	("ebx",16);			#  9, 8
+	&shl	($acc,8);			#  5
+	&movd	("mm1","edx");			# t[1] collected
+	&movz	("edx",&BP(-128,$tbl,$key,1));	# 15
+	&movz	($key,&HB("ebx"));		#  9
+	&shl	("edx",24);			# 15
+	&and	("ebx",0xff);			#  8
+	&or	("edx",$acc);			# 15
+
+	&punpckldq	("mm0","mm1");		# t[0,1] collected
+
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  9
+	&movz	($key,&LB("eax"));		#  2
+	&shl	($acc,8);			#  9
+	&movz	("eax",&HB("eax"));		#  3
+	&movz	("ebx",&BP(-128,$tbl,"ebx",1));	#  8
+	&or	("ecx",$acc);			#  9
+	&movz	($acc,&BP(-128,$tbl,$key,1));	#  2
+	&or	("edx","ebx");			#  8
+	&shl	($acc,16);			#  2
+	&movz	("eax",&BP(-128,$tbl,"eax",1));	#  3
+	&or	("edx",$acc);			#  2
+	&shl	("eax",24);			#  3
+	&or	("ecx","eax");			#  3
+	&mov	($key,$__key);
+	&movd	("mm4","edx");			# t[2] collected
+	&movd	("mm5","ecx");			# t[3] collected
+
+	&punpckldq	("mm4","mm5");		# t[2,3] collected
+}
+
+					if (!$x86only) {
+&function_begin_B("_sse_AES_decrypt_compact");
+	&pxor	("mm0",&QWP(0,$key));	#  7, 6, 5, 4, 3, 2, 1, 0
+	&pxor	("mm4",&QWP(8,$key));	# 15,14,13,12,11,10, 9, 8
+
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	($__end,$acc);			# end of key schedule
+
+	&mov	($s0,0x1b1b1b1b);		# magic constant
+	&mov	(&DWP(8,"esp"),$s0);
+	&mov	(&DWP(12,"esp"),$s0);
+
+	# prefetch Td4
+	&mov	($s0,&DWP(0-128,$tbl));
+	&mov	($s1,&DWP(32-128,$tbl));
+	&mov	($s2,&DWP(64-128,$tbl));
+	&mov	($s3,&DWP(96-128,$tbl));
+	&mov	($s0,&DWP(128-128,$tbl));
+	&mov	($s1,&DWP(160-128,$tbl));
+	&mov	($s2,&DWP(192-128,$tbl));
+	&mov	($s3,&DWP(224-128,$tbl));
+
+	&set_label("loop",16);
+		&sse_deccompact();
+		&add	($key,16);
+		&cmp	($key,$__end);
+		&ja	(&label("out"));
+
+		# ROTATE(x^y,N) == ROTATE(x,N)^ROTATE(y,N)
+		&movq	("mm3","mm0");		&movq	("mm7","mm4");
+		&movq	("mm2","mm0",1);	&movq	("mm6","mm4",1);
+		&movq	("mm1","mm0");		&movq	("mm5","mm4");
+		&pshufw	("mm0","mm0",0xb1);	&pshufw	("mm4","mm4",0xb1);# = ROTATE(tp0,16)
+		&pslld	("mm2",8);		&pslld	("mm6",8);
+		&psrld	("mm3",8);		&psrld	("mm7",8);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp0<<8
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp0>>8
+		&pslld	("mm2",16);		&pslld	("mm6",16);
+		&psrld	("mm3",16);		&psrld	("mm7",16);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp0<<24
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp0>>24
+
+		&movq	("mm3",&QWP(8,"esp"));
+		&pxor	("mm2","mm2");		&pxor	("mm6","mm6");
+		&pcmpgtb("mm2","mm1");		&pcmpgtb("mm6","mm5");
+		&pand	("mm2","mm3");		&pand	("mm6","mm3");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm2");		&pxor	("mm5","mm6");	# tp2
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&movq	("mm2","mm1");		&movq	("mm6","mm5");
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp2
+		&pslld	("mm3",24);		&pslld	("mm7",24);
+		&psrld	("mm2",8);		&psrld	("mm6",8);
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp2<<24
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= tp2>>8
+
+		&movq	("mm2",&QWP(8,"esp"));
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&pcmpgtb("mm3","mm1");		&pcmpgtb("mm7","mm5");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm3");		&pxor	("mm5","mm7");	# tp4
+		&pshufw	("mm3","mm1",0xb1);	&pshufw	("mm7","mm5",0xb1);
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp4
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= ROTATE(tp4,16)
+
+		&pxor	("mm3","mm3");		&pxor	("mm7","mm7");
+		&pcmpgtb("mm3","mm1");		&pcmpgtb("mm7","mm5");
+		&pand	("mm3","mm2");		&pand	("mm7","mm2");
+		&paddb	("mm1","mm1");		&paddb	("mm5","mm5");
+		&pxor	("mm1","mm3");		&pxor	("mm5","mm7");	# tp8
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8
+		&movq	("mm3","mm1");		&movq	("mm7","mm5");
+		&pshufw	("mm2","mm1",0xb1);	&pshufw	("mm6","mm5",0xb1);
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");	# ^= ROTATE(tp8,16)
+		&pslld	("mm1",8);		&pslld	("mm5",8);
+		&psrld	("mm3",8);		&psrld	("mm7",8);
+		&movq	("mm2",&QWP(0,$key));	&movq	("mm6",&QWP(8,$key));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8<<8
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp8>>8
+		&mov	($s0,&DWP(0-128,$tbl));
+		&pslld	("mm1",16);		&pslld	("mm5",16);
+		&mov	($s1,&DWP(64-128,$tbl));
+		&psrld	("mm3",16);		&psrld	("mm7",16);
+		&mov	($s2,&DWP(128-128,$tbl));
+		&pxor	("mm0","mm1");		&pxor	("mm4","mm5");	# ^= tp8<<24
+		&mov	($s3,&DWP(192-128,$tbl));
+		&pxor	("mm0","mm3");		&pxor	("mm4","mm7");	# ^= tp8>>24
+
+		&pxor	("mm0","mm2");		&pxor	("mm4","mm6");
+	&jmp	(&label("loop"));
+
+	&set_label("out",16);
+	&pxor	("mm0",&QWP(0,$key));
+	&pxor	("mm4",&QWP(8,$key));
+
+	&ret	();
+&function_end_B("_sse_AES_decrypt_compact");
+					}
+
+######################################################################
+# Vanilla block function.
+######################################################################
+
+sub decstep()
+{ my ($i,$td,@s) = @_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	# no instructions are reordered, as performance appears
+	# optimal... or rather that all attempts to reorder didn't
+	# result in better performance [which by the way is not a
+	# bit lower than encryption].
+	if($i==3)   {	&mov	($key,$__key);			}
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+			&mov	($out,&DWP(0,$td,$out,8));
+
+	if ($i==3)  {	$tmp=$s[1];				}
+			&movz	($tmp,&HB($s[1]));
+			&xor	($out,&DWP(3,$td,$tmp,8));
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
+	else        {	&mov	($tmp,$s[2]);			}
+			&shr	($tmp,16);
+			&and	($tmp,0xFF);
+			&xor	($out,&DWP(2,$td,$tmp,8));
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}
+	else        {	&mov	($tmp,$s[3]);			}
+			&shr	($tmp,24);
+			&xor	($out,&DWP(1,$td,$tmp,8));
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$__s0);			}
+			&comment();
+}
+
+sub declast()
+{ my ($i,$td,@s)=@_;
+  my $tmp = $key;
+  my $out = $i==3?$s[0]:$acc;
+
+	if($i==0)   {	&lea	($td,&DWP(2048+128,$td));
+			&mov	($tmp,&DWP(0-128,$td));
+			&mov	($acc,&DWP(32-128,$td));
+			&mov	($tmp,&DWP(64-128,$td));
+			&mov	($acc,&DWP(96-128,$td));
+			&mov	($tmp,&DWP(128-128,$td));
+			&mov	($acc,&DWP(160-128,$td));
+			&mov	($tmp,&DWP(192-128,$td));
+			&mov	($acc,&DWP(224-128,$td));
+			&lea	($td,&DWP(-128,$td));		}
+	if($i==3)   {	&mov	($key,$__key);			}
+	else        {	&mov	($out,$s[0]);			}
+			&and	($out,0xFF);
+			&movz	($out,&BP(0,$td,$out,1));
+
+	if ($i==3)  {	$tmp=$s[1];				}
+			&movz	($tmp,&HB($s[1]));
+			&movz	($tmp,&BP(0,$td,$tmp,1));
+			&shl	($tmp,8);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[2]; &mov ($s[1],$acc);		}
+	else        {	mov	($tmp,$s[2]);			}
+			&shr	($tmp,16);
+			&and	($tmp,0xFF);
+			&movz	($tmp,&BP(0,$td,$tmp,1));
+			&shl	($tmp,16);
+			&xor	($out,$tmp);
+
+	if ($i==3)  {	$tmp=$s[3]; &mov ($s[2],$__s1);		}
+	else        {	&mov	($tmp,$s[3]);			}
+			&shr	($tmp,24);
+			&movz	($tmp,&BP(0,$td,$tmp,1));
+			&shl	($tmp,24);
+			&xor	($out,$tmp);
+	if ($i<2)   {	&mov	(&DWP(4+4*$i,"esp"),$out);	}
+	if ($i==3)  {	&mov	($s[3],$__s0);
+			&lea	($td,&DWP(-2048,$td));		}
+}
+
+&function_begin_B("_x86_AES_decrypt");
+	# note that caller is expected to allocate stack frame for me!
+	&mov	($__key,$key);			# save key
+
+	&xor	($s0,&DWP(0,$key));		# xor with key
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&mov	($acc,&DWP(240,$key));		# load key->rounds
+
+	if ($small_footprint) {
+	    &lea	($acc,&DWP(-2,$acc,$acc));
+	    &lea	($acc,&DWP(0,$key,$acc,8));
+	    &mov	($__end,$acc);		# end of key schedule
+	    &set_label("loop",16);
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$s2,$s1,$s0);
+		&add	($key,16);		# advance rd_key
+		&xor	($s0,&DWP(0,$key));
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+	    &cmp	($key,$__end);
+	    &mov	($__key,$key);
+	    &jb		(&label("loop"));
+	}
+	else {
+	    &cmp	($acc,10);
+	    &jle	(&label("10rounds"));
+	    &cmp	($acc,12);
+	    &jle	(&label("12rounds"));
+
+	&set_label("14rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$s2,$s1,$s0);
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	    &add	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("12rounds",4);
+	    for ($i=1;$i<3;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$s2,$s1,$s0);
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	    &add	($key,32);
+	    &mov	($__key,$key);		# advance rd_key
+	&set_label("10rounds",4);
+	    for ($i=1;$i<10;$i++) {
+		&decstep(0,$tbl,$s0,$s3,$s2,$s1);
+		&decstep(1,$tbl,$s1,$s0,$s3,$s2);
+		&decstep(2,$tbl,$s2,$s1,$s0,$s3);
+		&decstep(3,$tbl,$s3,$s2,$s1,$s0);
+		&xor	($s0,&DWP(16*$i+0,$key));
+		&xor	($s1,&DWP(16*$i+4,$key));
+		&xor	($s2,&DWP(16*$i+8,$key));
+		&xor	($s3,&DWP(16*$i+12,$key));
+	    }
+	}
+
+	&declast(0,$tbl,$s0,$s3,$s2,$s1);
+	&declast(1,$tbl,$s1,$s0,$s3,$s2);
+	&declast(2,$tbl,$s2,$s1,$s0,$s3);
+	&declast(3,$tbl,$s3,$s2,$s1,$s0);
+
+	&add	($key,$small_footprint?16:160);
+	&xor	($s0,&DWP(0,$key));
+	&xor	($s1,&DWP(4,$key));
+	&xor	($s2,&DWP(8,$key));
+	&xor	($s3,&DWP(12,$key));
+
+	&ret	();
+
+&set_label("AES_Td",64);	# Yes! I keep it in the code segment!
+	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
+	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
+	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
+	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
+	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
+	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
+	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
+	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
+	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
+	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
+	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
+	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
+	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
+	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
+	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
+	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
+	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
+	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
+	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
+	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
+	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
+	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
+	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
+	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
+	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
+	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
+	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
+	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
+	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
+	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
+	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
+	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
+	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
+	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
+	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
+	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
+	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
+	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
+	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
+	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
+	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
+	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
+	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
+	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
+	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
+	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
+	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
+	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
+	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
+	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
+	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
+	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
+	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
+	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
+	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
+	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
+	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
+	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
+	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
+	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
+	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
+	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
+	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
+	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
+
+#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+&function_end_B("_x86_AES_decrypt");
+
+# void aes_nohw_decrypt (const void *inp,void *out,const AES_KEY *key);
+&function_begin("aes_nohw_decrypt");
+	&mov	($acc,&wparam(0));		# load inp
+	&mov	($key,&wparam(2));		# load key
+
+	&mov	($s0,"esp");
+	&sub	("esp",36);
+	&and	("esp",-64);			# align to cache-line
+
+	# place stack frame just "above" the key schedule
+	&lea	($s1,&DWP(-64-63,$key));
+	&sub	($s1,"esp");
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	("esp",$s1);
+	&add	("esp",4);	# 4 is reserved for caller's return address
+	&mov	($_esp,$s0);	# save stack pointer
+
+	&call   (&label("pic_point"));          # make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
+	&lea    ($tbl,&DWP(&label("AES_Td")."-".&label("pic_point"),$tbl));
+
+	# pick Td4 copy which can't "overlap" with stack frame or key schedule
+	&lea	($s1,&DWP(768-4,"esp"));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),25);	# check for SSE bit
+	&jnc	(&label("x86"));
+
+	&movq	("mm0",&QWP(0,$acc));
+	&movq	("mm4",&QWP(8,$acc));
+	&call	("_sse_AES_decrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&movq	(&QWP(0,$acc),"mm0");		# write output data
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&function_end_A();
+					}
+	&set_label("x86",16);
+	&mov	($_tbl,$tbl);
+	&mov	($s0,&DWP(0,$acc));		# load input data
+	&mov	($s1,&DWP(4,$acc));
+	&mov	($s2,&DWP(8,$acc));
+	&mov	($s3,&DWP(12,$acc));
+	&call	("_x86_AES_decrypt_compact");
+	&mov	("esp",$_esp);			# restore stack pointer
+	&mov	($acc,&wparam(1));		# load out
+	&mov	(&DWP(0,$acc),$s0);		# write output data
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+&function_end("aes_nohw_decrypt");
+
+# void aes_nohw_cbc_encrypt (const void char *inp, unsigned char *out,
+#			    size_t length, const AES_KEY *key,
+#			    unsigned char *ivp,const int enc);
+{
+# stack frame layout
+#             -4(%esp)		# return address	 0(%esp)
+#              0(%esp)		# s0 backing store	 4(%esp)
+#              4(%esp)		# s1 backing store	 8(%esp)
+#              8(%esp)		# s2 backing store	12(%esp)
+#             12(%esp)		# s3 backing store	16(%esp)
+#             16(%esp)		# key backup		20(%esp)
+#             20(%esp)		# end of key schedule	24(%esp)
+#             24(%esp)		# %ebp backup		28(%esp)
+#             28(%esp)		# %esp backup
+my $_inp=&DWP(32,"esp");	# copy of wparam(0)
+my $_out=&DWP(36,"esp");	# copy of wparam(1)
+my $_len=&DWP(40,"esp");	# copy of wparam(2)
+my $_key=&DWP(44,"esp");	# copy of wparam(3)
+my $_ivp=&DWP(48,"esp");	# copy of wparam(4)
+my $_tmp=&DWP(52,"esp");	# volatile variable
+#
+my $ivec=&DWP(60,"esp");	# ivec[16]
+my $aes_key=&DWP(76,"esp");	# copy of aes_key
+my $mark=&DWP(76+240,"esp");	# copy of aes_key->rounds
+
+&function_begin("aes_nohw_cbc_encrypt");
+	&mov	($s2 eq "ecx"? $s2 : "",&wparam(2));	# load len
+	&cmp	($s2,0);
+	&je	(&label("drop_out"));
+
+	&call   (&label("pic_point"));		# make it PIC!
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&picmeup($s0,"OPENSSL_ia32cap_P",$tbl,&label("pic_point")) if(!$x86only);
+
+	&cmp	(&wparam(5),0);
+	&lea    ($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+	&jne	(&label("picked_te"));
+	&lea	($tbl,&DWP(&label("AES_Td")."-".&label("AES_Te"),$tbl));
+	&set_label("picked_te");
+
+	# one can argue if this is required
+	&pushf	();
+	&cld	();
+
+	&cmp	($s2,$speed_limit);
+	&jb	(&label("slow_way"));
+	&test	($s2,15);
+	&jnz	(&label("slow_way"));
+					if (!$x86only) {
+	&bt	(&DWP(0,$s0),28);	# check for hyper-threading bit
+	&jc	(&label("slow_way"));
+					}
+	# pre-allocate aligned stack frame...
+	&lea	($acc,&DWP(-80-244,"esp"));
+	&and	($acc,-64);
+
+	# ... and make sure it doesn't alias with $tbl modulo 4096
+	&mov	($s0,$tbl);
+	&lea	($s1,&DWP(2048+256,$tbl));
+	&mov	($s3,$acc);
+	&and	($s0,0xfff);		# s = %ebp&0xfff
+	&and	($s1,0xfff);		# e = (%ebp+2048+256)&0xfff
+	&and	($s3,0xfff);		# p = %esp&0xfff
+
+	&cmp	($s3,$s1);		# if (p>=e) %esp =- (p-e);
+	&jb	(&label("tbl_break_out"));
+	&sub	($s3,$s1);
+	&sub	($acc,$s3);
+	&jmp	(&label("tbl_ok"));
+	&set_label("tbl_break_out",4);	# else %esp -= (p-s)&0xfff + framesz;
+	&sub	($s3,$s0);
+	&and	($s3,0xfff);
+	&add	($s3,384);
+	&sub	($acc,$s3);
+	&set_label("tbl_ok",4);
+
+	&lea	($s3,&wparam(0));	# obtain pointer to parameter block
+	&exch	("esp",$acc);		# allocate stack frame
+	&add	("esp",4);		# reserve for return address!
+	&mov	($_tbl,$tbl);		# save %ebp
+	&mov	($_esp,$acc);		# save %esp
+
+	&mov	($s0,&DWP(0,$s3));	# load inp
+	&mov	($s1,&DWP(4,$s3));	# load out
+	#&mov	($s2,&DWP(8,$s3));	# load len
+	&mov	($key,&DWP(12,$s3));	# load key
+	&mov	($acc,&DWP(16,$s3));	# load ivp
+	&mov	($s3,&DWP(20,$s3));	# load enc flag
+
+	&mov	($_inp,$s0);		# save copy of inp
+	&mov	($_out,$s1);		# save copy of out
+	&mov	($_len,$s2);		# save copy of len
+	&mov	($_key,$key);		# save copy of key
+	&mov	($_ivp,$acc);		# save copy of ivp
+
+	&mov	($mark,0);		# copy of aes_key->rounds = 0;
+	# do we copy key schedule to stack?
+	&mov	($s1 eq "ebx" ? $s1 : "",$key);
+	&mov	($s2 eq "ecx" ? $s2 : "",244/4);
+	&sub	($s1,$tbl);
+	&mov	("esi",$key);
+	&and	($s1,0xfff);
+	&lea	("edi",$aes_key);
+	&cmp	($s1,2048+256);
+	&jb	(&label("do_copy"));
+	&cmp	($s1,4096-244);
+	&jb	(&label("skip_copy"));
+	&set_label("do_copy",4);
+		&mov	($_key,"edi");
+		&data_word(0xA5F3F689);	# rep movsd
+	&set_label("skip_copy");
+
+	&mov	($key,16);
+	&set_label("prefetch_tbl",4);
+		&mov	($s0,&DWP(0,$tbl));
+		&mov	($s1,&DWP(32,$tbl));
+		&mov	($s2,&DWP(64,$tbl));
+		&mov	($acc,&DWP(96,$tbl));
+		&lea	($tbl,&DWP(128,$tbl));
+		&sub	($key,1);
+	&jnz	(&label("prefetch_tbl"));
+	&sub	($tbl,2048);
+
+	&mov	($acc,$_inp);
+	&mov	($key,$_ivp);
+
+	&cmp	($s3,0);
+	&je	(&label("fast_decrypt"));
+
+#----------------------------- ENCRYPT -----------------------------#
+	&mov	($s0,&DWP(0,$key));		# load iv
+	&mov	($s1,&DWP(4,$key));
+
+	&set_label("fast_enc_loop",16);
+		&mov	($s2,&DWP(8,$key));
+		&mov	($s3,&DWP(12,$key));
+
+		&xor	($s0,&DWP(0,$acc));	# xor input data
+		&xor	($s1,&DWP(4,$acc));
+		&xor	($s2,&DWP(8,$acc));
+		&xor	($s3,&DWP(12,$acc));
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_encrypt");
+
+		&mov	($acc,$_inp);		# load inp
+		&mov	($key,$_out);		# load out
+
+		&mov	(&DWP(0,$key),$s0);	# save output data
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($s2,$_len);		# load len
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_enc_loop"));
+	&mov	($acc,$_ivp);		# load ivp
+	&mov	($s2,&DWP(8,$key));	# restore last 2 dwords
+	&mov	($s3,&DWP(12,$key));
+	&mov	(&DWP(0,$acc),$s0);	# save ivec
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+
+	&cmp	($mark,0);		# was the key schedule copied?
+	&mov	("edi",$_key);
+	&je	(&label("skip_ezero"));
+	# zero copy of key schedule
+	&mov	("ecx",240/4);
+	&xor	("eax","eax");
+	&align	(4);
+	&data_word(0xABF3F689);		# rep stosd
+	&set_label("skip_ezero");
+	&mov	("esp",$_esp);
+	&popf	();
+    &set_label("drop_out");
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+#----------------------------- DECRYPT -----------------------------#
+&set_label("fast_decrypt",16);
+
+	&cmp	($acc,$_out);
+	&je	(&label("fast_dec_in_place"));	# in-place processing...
+
+	&mov	($_tmp,$key);
+
+	&align	(4);
+	&set_label("fast_dec_loop",16);
+		&mov	($s0,&DWP(0,$acc));	# read input
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_decrypt");
+
+		&mov	($key,$_tmp);		# load ivp
+		&mov	($acc,$_len);		# load len
+		&xor	($s0,&DWP(0,$key));	# xor iv
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+
+		&mov	($key,$_out);		# load out
+		&mov	($acc,$_inp);		# load inp
+
+		&mov	(&DWP(0,$key),$s0);	# write output
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($s2,$_len);		# load len
+		&mov	($_tmp,$acc);		# save ivp
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($key,&DWP(16,$key));	# advance out
+		&mov	($_out,$key);		# save out
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_dec_loop"));
+	&mov	($key,$_tmp);		# load temp ivp
+	&mov	($acc,$_ivp);		# load user ivp
+	&mov	($s0,&DWP(0,$key));	# load iv
+	&mov	($s1,&DWP(4,$key));
+	&mov	($s2,&DWP(8,$key));
+	&mov	($s3,&DWP(12,$key));
+	&mov	(&DWP(0,$acc),$s0);	# copy back to user
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+	&jmp	(&label("fast_dec_out"));
+
+    &set_label("fast_dec_in_place",16);
+	&set_label("fast_dec_in_place_loop");
+		&mov	($s0,&DWP(0,$acc));	# read input
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&lea	($key,$ivec);
+		&mov	(&DWP(0,$key),$s0);	# copy to temp
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_decrypt");
+
+		&mov	($key,$_ivp);		# load ivp
+		&mov	($acc,$_out);		# load out
+		&xor	($s0,&DWP(0,$key));	# xor iv
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+
+		&mov	(&DWP(0,$acc),$s0);	# write output
+		&mov	(&DWP(4,$acc),$s1);
+		&mov	(&DWP(8,$acc),$s2);
+		&mov	(&DWP(12,$acc),$s3);
+
+		&lea	($acc,&DWP(16,$acc));	# advance out
+		&mov	($_out,$acc);		# save out
+
+		&lea	($acc,$ivec);
+		&mov	($s0,&DWP(0,$acc));	# read temp
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&mov	(&DWP(0,$key),$s0);	# copy iv
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($acc,$_inp);		# load inp
+		&mov	($s2,$_len);		# load len
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&sub	($s2,16);		# decrease len
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("fast_dec_in_place_loop"));
+
+    &set_label("fast_dec_out",4);
+	&cmp	($mark,0);		# was the key schedule copied?
+	&mov	("edi",$_key);
+	&je	(&label("skip_dzero"));
+	# zero copy of key schedule
+	&mov	("ecx",240/4);
+	&xor	("eax","eax");
+	&align	(4);
+	&data_word(0xABF3F689);		# rep stosd
+	&set_label("skip_dzero");
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+#--------------------------- SLOW ROUTINE ---------------------------#
+&set_label("slow_way",16);
+
+	&mov	($s0,&DWP(0,$s0)) if (!$x86only);# load OPENSSL_ia32cap
+	&mov	($key,&wparam(3));	# load key
+
+	# pre-allocate aligned stack frame...
+	&lea	($acc,&DWP(-80,"esp"));
+	&and	($acc,-64);
+
+	# ... and make sure it doesn't alias with $key modulo 1024
+	&lea	($s1,&DWP(-80-63,$key));
+	&sub	($s1,$acc);
+	&neg	($s1);
+	&and	($s1,0x3C0);	# modulo 1024, but aligned to cache-line
+	&sub	($acc,$s1);
+
+	# pick S-box copy which can't overlap with stack frame or $key
+	&lea	($s1,&DWP(768,$acc));
+	&sub	($s1,$tbl);
+	&and	($s1,0x300);
+	&lea	($tbl,&DWP(2048+128,$tbl,$s1));
+
+	&lea	($s3,&wparam(0));	# pointer to parameter block
+
+	&exch	("esp",$acc);
+	&add	("esp",4);		# reserve for return address!
+	&mov	($_tbl,$tbl);		# save %ebp
+	&mov	($_esp,$acc);		# save %esp
+	&mov	($_tmp,$s0);		# save OPENSSL_ia32cap
+
+	&mov	($s0,&DWP(0,$s3));	# load inp
+	&mov	($s1,&DWP(4,$s3));	# load out
+	#&mov	($s2,&DWP(8,$s3));	# load len
+	#&mov	($key,&DWP(12,$s3));	# load key
+	&mov	($acc,&DWP(16,$s3));	# load ivp
+	&mov	($s3,&DWP(20,$s3));	# load enc flag
+
+	&mov	($_inp,$s0);		# save copy of inp
+	&mov	($_out,$s1);		# save copy of out
+	&mov	($_len,$s2);		# save copy of len
+	&mov	($_key,$key);		# save copy of key
+	&mov	($_ivp,$acc);		# save copy of ivp
+
+	&mov	($key,$acc);
+	&mov	($acc,$s0);
+
+	&cmp	($s3,0);
+	&je	(&label("slow_decrypt"));
+
+#--------------------------- SLOW ENCRYPT ---------------------------#
+	&cmp	($s2,16);
+	&mov	($s3,$s1);
+	&jb	(&label("slow_enc_tail"));
+
+					if (!$x86only) {
+	&bt	($_tmp,25);		# check for SSE bit
+	&jnc	(&label("slow_enc_x86"));
+
+	&movq	("mm0",&QWP(0,$key));	# load iv
+	&movq	("mm4",&QWP(8,$key));
+
+	&set_label("slow_enc_loop_sse",16);
+		&pxor	("mm0",&QWP(0,$acc));	# xor input data
+		&pxor	("mm4",&QWP(8,$acc));
+
+		&mov	($key,$_key);
+		&call	("_sse_AES_encrypt_compact");
+
+		&mov	($acc,$_inp);		# load inp
+		&mov	($key,$_out);		# load out
+		&mov	($s2,$_len);		# load len
+
+		&movq	(&QWP(0,$key),"mm0");	# save output data
+		&movq	(&QWP(8,$key),"mm4");
+
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&cmp	($s2,16);
+		&mov	($_len,$s2);		# save len
+	&jae	(&label("slow_enc_loop_sse"));
+	&test	($s2,15);
+	&jnz	(&label("slow_enc_tail"));
+	&mov	($acc,$_ivp);		# load ivp
+	&movq	(&QWP(0,$acc),"mm0");	# save ivec
+	&movq	(&QWP(8,$acc),"mm4");
+	&emms	();
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+					}
+    &set_label("slow_enc_x86",16);
+	&mov	($s0,&DWP(0,$key));	# load iv
+	&mov	($s1,&DWP(4,$key));
+
+	&set_label("slow_enc_loop_x86",4);
+		&mov	($s2,&DWP(8,$key));
+		&mov	($s3,&DWP(12,$key));
+
+		&xor	($s0,&DWP(0,$acc));	# xor input data
+		&xor	($s1,&DWP(4,$acc));
+		&xor	($s2,&DWP(8,$acc));
+		&xor	($s3,&DWP(12,$acc));
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_encrypt_compact");
+
+		&mov	($acc,$_inp);		# load inp
+		&mov	($key,$_out);		# load out
+
+		&mov	(&DWP(0,$key),$s0);	# save output data
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($s2,$_len);		# load len
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&lea	($s3,&DWP(16,$key));	# advance out
+		&mov	($_out,$s3);		# save out
+		&sub	($s2,16);		# decrease len
+		&cmp	($s2,16);
+		&mov	($_len,$s2);		# save len
+	&jae	(&label("slow_enc_loop_x86"));
+	&test	($s2,15);
+	&jnz	(&label("slow_enc_tail"));
+	&mov	($acc,$_ivp);		# load ivp
+	&mov	($s2,&DWP(8,$key));	# restore last dwords
+	&mov	($s3,&DWP(12,$key));
+	&mov	(&DWP(0,$acc),$s0);	# save ivec
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_enc_tail",16);
+	&emms	()	if (!$x86only);
+	&mov	($key eq "edi"? $key:"",$s3);	# load out to edi
+	&mov	($s1,16);
+	&sub	($s1,$s2);
+	&cmp	($key,$acc eq "esi"? $acc:"");	# compare with inp
+	&je	(&label("enc_in_place"));
+	&align	(4);
+	&data_word(0xA4F3F689);	# rep movsb	# copy input
+	&jmp	(&label("enc_skip_in_place"));
+    &set_label("enc_in_place");
+	&lea	($key,&DWP(0,$key,$s2));
+    &set_label("enc_skip_in_place");
+	&mov	($s2,$s1);
+	&xor	($s0,$s0);
+	&align	(4);
+	&data_word(0xAAF3F689);	# rep stosb	# zero tail
+
+	&mov	($key,$_ivp);			# restore ivp
+	&mov	($acc,$s3);			# output as input
+	&mov	($s0,&DWP(0,$key));
+	&mov	($s1,&DWP(4,$key));
+	&mov	($_len,16);			# len=16
+	&jmp	(&label("slow_enc_loop_x86"));	# one more spin...
+
+#--------------------------- SLOW DECRYPT ---------------------------#
+&set_label("slow_decrypt",16);
+					if (!$x86only) {
+	&bt	($_tmp,25);		# check for SSE bit
+	&jnc	(&label("slow_dec_loop_x86"));
+
+	&set_label("slow_dec_loop_sse",4);
+		&movq	("mm0",&QWP(0,$acc));	# read input
+		&movq	("mm4",&QWP(8,$acc));
+
+		&mov	($key,$_key);
+		&call	("_sse_AES_decrypt_compact");
+
+		&mov	($acc,$_inp);		# load inp
+		&lea	($s0,$ivec);
+		&mov	($s1,$_out);		# load out
+		&mov	($s2,$_len);		# load len
+		&mov	($key,$_ivp);		# load ivp
+
+		&movq	("mm1",&QWP(0,$acc));	# re-read input
+		&movq	("mm5",&QWP(8,$acc));
+
+		&pxor	("mm0",&QWP(0,$key));	# xor iv
+		&pxor	("mm4",&QWP(8,$key));
+
+		&movq	(&QWP(0,$key),"mm1");	# copy input to iv
+		&movq	(&QWP(8,$key),"mm5");
+
+		&sub	($s2,16);		# decrease len
+		&jc	(&label("slow_dec_partial_sse"));
+
+		&movq	(&QWP(0,$s1),"mm0");	# write output
+		&movq	(&QWP(8,$s1),"mm4");
+
+		&lea	($s1,&DWP(16,$s1));	# advance out
+		&mov	($_out,$s1);		# save out
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+		&mov	($_len,$s2);		# save len
+	&jnz	(&label("slow_dec_loop_sse"));
+	&emms	();
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_dec_partial_sse",16);
+	&movq	(&QWP(0,$s0),"mm0");	# save output to temp
+	&movq	(&QWP(8,$s0),"mm4");
+	&emms	();
+
+	&add	($s2 eq "ecx" ? "ecx":"",16);
+	&mov	("edi",$s1);		# out
+	&mov	("esi",$s0);		# temp
+	&align	(4);
+	&data_word(0xA4F3F689);		# rep movsb # copy partial output
+
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+					}
+	&set_label("slow_dec_loop_x86",16);
+		&mov	($s0,&DWP(0,$acc));	# read input
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&lea	($key,$ivec);
+		&mov	(&DWP(0,$key),$s0);	# copy to temp
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($key,$_key);		# load key
+		&call	("_x86_AES_decrypt_compact");
+
+		&mov	($key,$_ivp);		# load ivp
+		&mov	($acc,$_len);		# load len
+		&xor	($s0,&DWP(0,$key));	# xor iv
+		&xor	($s1,&DWP(4,$key));
+		&xor	($s2,&DWP(8,$key));
+		&xor	($s3,&DWP(12,$key));
+
+		&sub	($acc,16);
+		&jc	(&label("slow_dec_partial_x86"));
+
+		&mov	($_len,$acc);		# save len
+		&mov	($acc,$_out);		# load out
+
+		&mov	(&DWP(0,$acc),$s0);	# write output
+		&mov	(&DWP(4,$acc),$s1);
+		&mov	(&DWP(8,$acc),$s2);
+		&mov	(&DWP(12,$acc),$s3);
+
+		&lea	($acc,&DWP(16,$acc));	# advance out
+		&mov	($_out,$acc);		# save out
+
+		&lea	($acc,$ivec);
+		&mov	($s0,&DWP(0,$acc));	# read temp
+		&mov	($s1,&DWP(4,$acc));
+		&mov	($s2,&DWP(8,$acc));
+		&mov	($s3,&DWP(12,$acc));
+
+		&mov	(&DWP(0,$key),$s0);	# copy it to iv
+		&mov	(&DWP(4,$key),$s1);
+		&mov	(&DWP(8,$key),$s2);
+		&mov	(&DWP(12,$key),$s3);
+
+		&mov	($acc,$_inp);		# load inp
+		&lea	($acc,&DWP(16,$acc));	# advance inp
+		&mov	($_inp,$acc);		# save inp
+	&jnz	(&label("slow_dec_loop_x86"));
+	&mov	("esp",$_esp);
+	&popf	();
+	&function_end_A();
+	&pushf	();			# kludge, never executed
+
+    &set_label("slow_dec_partial_x86",16);
+	&lea	($acc,$ivec);
+	&mov	(&DWP(0,$acc),$s0);	# save output to temp
+	&mov	(&DWP(4,$acc),$s1);
+	&mov	(&DWP(8,$acc),$s2);
+	&mov	(&DWP(12,$acc),$s3);
+
+	&mov	($acc,$_inp);
+	&mov	($s0,&DWP(0,$acc));	# re-read input
+	&mov	($s1,&DWP(4,$acc));
+	&mov	($s2,&DWP(8,$acc));
+	&mov	($s3,&DWP(12,$acc));
+
+	&mov	(&DWP(0,$key),$s0);	# copy it to iv
+	&mov	(&DWP(4,$key),$s1);
+	&mov	(&DWP(8,$key),$s2);
+	&mov	(&DWP(12,$key),$s3);
+
+	&mov	("ecx",$_len);
+	&mov	("edi",$_out);
+	&lea	("esi",$ivec);
+	&align	(4);
+	&data_word(0xA4F3F689);		# rep movsb # copy partial output
+
+	&mov	("esp",$_esp);
+	&popf	();
+&function_end("aes_nohw_cbc_encrypt");
+}
+
+#------------------------------------------------------------------#
+
+sub enckey()
+{
+	&movz	("esi",&LB("edx"));		# rk[i]>>0
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&movz	("esi",&HB("edx"));		# rk[i]>>8
+	&shl	("ebx",24);
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&shr	("edx",16);
+	&movz	("esi",&LB("edx"));		# rk[i]>>16
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&movz	("esi",&HB("edx"));		# rk[i]>>24
+	&shl	("ebx",8);
+	&xor	("eax","ebx");
+
+	&movz	("ebx",&BP(-128,$tbl,"esi",1));
+	&shl	("ebx",16);
+	&xor	("eax","ebx");
+
+	&xor	("eax",&DWP(1024-128,$tbl,"ecx",4));	# rcon
+}
+
+&function_begin("_x86_AES_set_encrypt_key");
+	&mov	("esi",&wparam(1));		# user supplied key
+	&mov	("edi",&wparam(3));		# private key schedule
+
+	&test	("esi",-1);
+	&jz	(&label("badpointer"));
+	&test	("edi",-1);
+	&jz	(&label("badpointer"));
+
+	&call	(&label("pic_point"));
+	&set_label("pic_point");
+	&blindpop($tbl);
+	&lea	($tbl,&DWP(&label("AES_Te")."-".&label("pic_point"),$tbl));
+	&lea	($tbl,&DWP(2048+128,$tbl));
+
+	# prefetch Te4
+	&mov	("eax",&DWP(0-128,$tbl));
+	&mov	("ebx",&DWP(32-128,$tbl));
+	&mov	("ecx",&DWP(64-128,$tbl));
+	&mov	("edx",&DWP(96-128,$tbl));
+	&mov	("eax",&DWP(128-128,$tbl));
+	&mov	("ebx",&DWP(160-128,$tbl));
+	&mov	("ecx",&DWP(192-128,$tbl));
+	&mov	("edx",&DWP(224-128,$tbl));
+
+	&mov	("ecx",&wparam(2));		# number of bits in key
+	&cmp	("ecx",128);
+	&je	(&label("10rounds"));
+	&cmp	("ecx",192);
+	&je	(&label("12rounds"));
+	&cmp	("ecx",256);
+	&je	(&label("14rounds"));
+	&mov	("eax",-2);			# invalid number of bits
+	&jmp	(&label("exit"));
+
+    &set_label("10rounds");
+	&mov	("eax",&DWP(0,"esi"));		# copy first 4 dwords
+	&mov	("ebx",&DWP(4,"esi"));
+	&mov	("ecx",&DWP(8,"esi"));
+	&mov	("edx",&DWP(12,"esi"));
+	&mov	(&DWP(0,"edi"),"eax");
+	&mov	(&DWP(4,"edi"),"ebx");
+	&mov	(&DWP(8,"edi"),"ecx");
+	&mov	(&DWP(12,"edi"),"edx");
+
+	&xor	("ecx","ecx");
+	&jmp	(&label("10shortcut"));
+
+	&align	(4);
+	&set_label("10loop");
+		&mov	("eax",&DWP(0,"edi"));		# rk[0]
+		&mov	("edx",&DWP(12,"edi"));		# rk[3]
+	&set_label("10shortcut");
+		&enckey	();
+
+		&mov	(&DWP(16,"edi"),"eax");		# rk[4]
+		&xor	("eax",&DWP(4,"edi"));
+		&mov	(&DWP(20,"edi"),"eax");		# rk[5]
+		&xor	("eax",&DWP(8,"edi"));
+		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
+		&xor	("eax",&DWP(12,"edi"));
+		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
+		&inc	("ecx");
+		&add	("edi",16);
+		&cmp	("ecx",10);
+	&jl	(&label("10loop"));
+
+	&mov	(&DWP(80,"edi"),10);		# setup number of rounds
+	&xor	("eax","eax");
+	&jmp	(&label("exit"));
+
+    &set_label("12rounds");
+	&mov	("eax",&DWP(0,"esi"));		# copy first 6 dwords
+	&mov	("ebx",&DWP(4,"esi"));
+	&mov	("ecx",&DWP(8,"esi"));
+	&mov	("edx",&DWP(12,"esi"));
+	&mov	(&DWP(0,"edi"),"eax");
+	&mov	(&DWP(4,"edi"),"ebx");
+	&mov	(&DWP(8,"edi"),"ecx");
+	&mov	(&DWP(12,"edi"),"edx");
+	&mov	("ecx",&DWP(16,"esi"));
+	&mov	("edx",&DWP(20,"esi"));
+	&mov	(&DWP(16,"edi"),"ecx");
+	&mov	(&DWP(20,"edi"),"edx");
+
+	&xor	("ecx","ecx");
+	&jmp	(&label("12shortcut"));
+
+	&align	(4);
+	&set_label("12loop");
+		&mov	("eax",&DWP(0,"edi"));		# rk[0]
+		&mov	("edx",&DWP(20,"edi"));		# rk[5]
+	&set_label("12shortcut");
+		&enckey	();
+
+		&mov	(&DWP(24,"edi"),"eax");		# rk[6]
+		&xor	("eax",&DWP(4,"edi"));
+		&mov	(&DWP(28,"edi"),"eax");		# rk[7]
+		&xor	("eax",&DWP(8,"edi"));
+		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
+		&xor	("eax",&DWP(12,"edi"));
+		&mov	(&DWP(36,"edi"),"eax");		# rk[9]
+
+		&cmp	("ecx",7);
+		&je	(&label("12break"));
+		&inc	("ecx");
+
+		&xor	("eax",&DWP(16,"edi"));
+		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
+		&xor	("eax",&DWP(20,"edi"));
+		&mov	(&DWP(44,"edi"),"eax");		# rk[11]
+
+		&add	("edi",24);
+	&jmp	(&label("12loop"));
+
+	&set_label("12break");
+	&mov	(&DWP(72,"edi"),12);		# setup number of rounds
+	&xor	("eax","eax");
+	&jmp	(&label("exit"));
+
+    &set_label("14rounds");
+	&mov	("eax",&DWP(0,"esi"));		# copy first 8 dwords
+	&mov	("ebx",&DWP(4,"esi"));
+	&mov	("ecx",&DWP(8,"esi"));
+	&mov	("edx",&DWP(12,"esi"));
+	&mov	(&DWP(0,"edi"),"eax");
+	&mov	(&DWP(4,"edi"),"ebx");
+	&mov	(&DWP(8,"edi"),"ecx");
+	&mov	(&DWP(12,"edi"),"edx");
+	&mov	("eax",&DWP(16,"esi"));
+	&mov	("ebx",&DWP(20,"esi"));
+	&mov	("ecx",&DWP(24,"esi"));
+	&mov	("edx",&DWP(28,"esi"));
+	&mov	(&DWP(16,"edi"),"eax");
+	&mov	(&DWP(20,"edi"),"ebx");
+	&mov	(&DWP(24,"edi"),"ecx");
+	&mov	(&DWP(28,"edi"),"edx");
+
+	&xor	("ecx","ecx");
+	&jmp	(&label("14shortcut"));
+
+	&align	(4);
+	&set_label("14loop");
+		&mov	("edx",&DWP(28,"edi"));		# rk[7]
+	&set_label("14shortcut");
+		&mov	("eax",&DWP(0,"edi"));		# rk[0]
+
+		&enckey	();
+
+		&mov	(&DWP(32,"edi"),"eax");		# rk[8]
+		&xor	("eax",&DWP(4,"edi"));
+		&mov	(&DWP(36,"edi"),"eax");		# rk[9]
+		&xor	("eax",&DWP(8,"edi"));
+		&mov	(&DWP(40,"edi"),"eax");		# rk[10]
+		&xor	("eax",&DWP(12,"edi"));
+		&mov	(&DWP(44,"edi"),"eax");		# rk[11]
+
+		&cmp	("ecx",6);
+		&je	(&label("14break"));
+		&inc	("ecx");
+
+		&mov	("edx","eax");
+		&mov	("eax",&DWP(16,"edi"));		# rk[4]
+		&movz	("esi",&LB("edx"));		# rk[11]>>0
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&movz	("esi",&HB("edx"));		# rk[11]>>8
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&shr	("edx",16);
+		&shl	("ebx",8);
+		&movz	("esi",&LB("edx"));		# rk[11]>>16
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&movz	("esi",&HB("edx"));		# rk[11]>>24
+		&shl	("ebx",16);
+		&xor	("eax","ebx");
+
+		&movz	("ebx",&BP(-128,$tbl,"esi",1));
+		&shl	("ebx",24);
+		&xor	("eax","ebx");
+
+		&mov	(&DWP(48,"edi"),"eax");		# rk[12]
+		&xor	("eax",&DWP(20,"edi"));
+		&mov	(&DWP(52,"edi"),"eax");		# rk[13]
+		&xor	("eax",&DWP(24,"edi"));
+		&mov	(&DWP(56,"edi"),"eax");		# rk[14]
+		&xor	("eax",&DWP(28,"edi"));
+		&mov	(&DWP(60,"edi"),"eax");		# rk[15]
+
+		&add	("edi",32);
+	&jmp	(&label("14loop"));
+
+	&set_label("14break");
+	&mov	(&DWP(48,"edi"),14);		# setup number of rounds
+	&xor	("eax","eax");
+	&jmp	(&label("exit"));
+
+    &set_label("badpointer");
+	&mov	("eax",-1);
+    &set_label("exit");
+&function_end("_x86_AES_set_encrypt_key");
+
+# int aes_nohw_set_encrypt_key(const unsigned char *userKey, const int bits,
+#                              AES_KEY *key)
+&function_begin_B("aes_nohw_set_encrypt_key");
+	&call	("_x86_AES_set_encrypt_key");
+	&ret	();
+&function_end_B("aes_nohw_set_encrypt_key");
+
+sub deckey()
+{ my ($i,$key,$tp1,$tp2,$tp4,$tp8) = @_;
+  my $tmp = $tbl;
+
+	&mov	($tmp,0x80808080);
+	&and	($tmp,$tp1);
+	&lea	($tp2,&DWP(0,$tp1,$tp1));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&sub	($acc,$tmp);
+	&and	($tp2,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	&xor	($tp2,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp2);
+	&lea	($tp4,&DWP(0,$tp2,$tp2));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	&sub	($acc,$tmp);
+	&and	($tp4,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &xor	($tp2,$tp1);	# tp2^tp1
+	&xor	($tp4,$acc);
+	&mov	($tmp,0x80808080);
+
+	&and	($tmp,$tp4);
+	&lea	($tp8,&DWP(0,$tp4,$tp4));
+	&mov	($acc,$tmp);
+	&shr	($tmp,7);
+	 &xor	($tp4,$tp1);	# tp4^tp1
+	&sub	($acc,$tmp);
+	&and	($tp8,0xfefefefe);
+	&and	($acc,0x1b1b1b1b);
+	 &rotl	($tp1,8);	# = ROTATE(tp1,8)
+	&xor	($tp8,$acc);
+
+	&mov	($tmp,&DWP(4*($i+1),$key));	# modulo-scheduled load
+
+	&xor	($tp1,$tp2);
+	&xor	($tp2,$tp8);
+	&xor	($tp1,$tp4);
+	&rotl	($tp2,24);
+	&xor	($tp4,$tp8);
+	&xor	($tp1,$tp8);	# ^= tp8^(tp4^tp1)^(tp2^tp1)
+	&rotl	($tp4,16);
+	&xor	($tp1,$tp2);	# ^= ROTATE(tp8^tp2^tp1,24)
+	&rotl	($tp8,8);
+	&xor	($tp1,$tp4);	# ^= ROTATE(tp8^tp4^tp1,16)
+	&mov	($tp2,$tmp);
+	&xor	($tp1,$tp8);	# ^= ROTATE(tp8,8)
+
+	&mov	(&DWP(4*$i,$key),$tp1);
+}
+
+# int aes_nohw_set_decrypt_key(const unsigned char *userKey, const int bits,
+#                              AES_KEY *key)
+&function_begin_B("aes_nohw_set_decrypt_key");
+	&call	("_x86_AES_set_encrypt_key");
+	&cmp	("eax",0);
+	&je	(&label("proceed"));
+	&ret	();
+
+    &set_label("proceed");
+	&push	("ebp");
+	&push	("ebx");
+	&push	("esi");
+	&push	("edi");
+
+	&mov	("esi",&wparam(2));
+	&mov	("ecx",&DWP(240,"esi"));	# pull number of rounds
+	&lea	("ecx",&DWP(0,"","ecx",4));
+	&lea	("edi",&DWP(0,"esi","ecx",4));	# pointer to last chunk
+
+	&set_label("invert",4);			# invert order of chunks
+		&mov	("eax",&DWP(0,"esi"));
+		&mov	("ebx",&DWP(4,"esi"));
+		&mov	("ecx",&DWP(0,"edi"));
+		&mov	("edx",&DWP(4,"edi"));
+		&mov	(&DWP(0,"edi"),"eax");
+		&mov	(&DWP(4,"edi"),"ebx");
+		&mov	(&DWP(0,"esi"),"ecx");
+		&mov	(&DWP(4,"esi"),"edx");
+		&mov	("eax",&DWP(8,"esi"));
+		&mov	("ebx",&DWP(12,"esi"));
+		&mov	("ecx",&DWP(8,"edi"));
+		&mov	("edx",&DWP(12,"edi"));
+		&mov	(&DWP(8,"edi"),"eax");
+		&mov	(&DWP(12,"edi"),"ebx");
+		&mov	(&DWP(8,"esi"),"ecx");
+		&mov	(&DWP(12,"esi"),"edx");
+		&add	("esi",16);
+		&sub	("edi",16);
+		&cmp	("esi","edi");
+	&jne	(&label("invert"));
+
+	&mov	($key,&wparam(2));
+	&mov	($acc,&DWP(240,$key));		# pull number of rounds
+	&lea	($acc,&DWP(-2,$acc,$acc));
+	&lea	($acc,&DWP(0,$key,$acc,8));
+	&mov	(&wparam(2),$acc);
+
+	&mov	($s0,&DWP(16,$key));		# modulo-scheduled load
+	&set_label("permute",4);		# permute the key schedule
+		&add	($key,16);
+		&deckey	(0,$key,$s0,$s1,$s2,$s3);
+		&deckey	(1,$key,$s1,$s2,$s3,$s0);
+		&deckey	(2,$key,$s2,$s3,$s0,$s1);
+		&deckey	(3,$key,$s3,$s0,$s1,$s2);
+		&cmp	($key,&wparam(2));
+	&jb	(&label("permute"));
+
+	&xor	("eax","eax");			# return success
+&function_end("aes_nohw_set_decrypt_key");
+&asciz("AES for x86, CRYPTOGAMS by <appro\@openssl.org>");
+
+&asm_finish();
+
+close STDOUT or die "error closing STDOUT";
diff --git a/crypto/fipsmodule/aes/asm/aes-armv4.pl b/crypto/fipsmodule/aes/asm/aes-armv4.pl
new file mode 100644
index 0000000..fbb1995
--- /dev/null
+++ b/crypto/fipsmodule/aes/asm/aes-armv4.pl
@@ -0,0 +1,1252 @@
+#! /usr/bin/env perl
+# Copyright 2007-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
+
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+
+# AES for ARMv4
+
+# January 2007.
+#
+# Code uses single 1K S-box and is >2 times faster than code generated
+# by gcc-3.4.1. This is thanks to unique feature of ARMv4 ISA, which
+# allows to merge logical or arithmetic operation with shift or rotate
+# in one instruction and emit combined result every cycle. The module
+# is endian-neutral. The performance is ~42 cycles/byte for 128-bit
+# key [on single-issue Xscale PXA250 core].
+
+# May 2007.
+#
+# AES_set_[en|de]crypt_key is added.
+
+# July 2010.
+#
+# Rescheduling for dual-issue pipeline resulted in 12% improvement on
+# Cortex A8 core and ~25 cycles per byte processed with 128-bit key.
+
+# February 2011.
+#
+# Profiler-assisted and platform-specific optimization resulted in 16%
+# improvement on Cortex A8 core and ~21.5 cycles per byte.
+
+$flavour = shift;
+if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
+else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
+
+if ($flavour && $flavour ne "void") {
+    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+    ( $xlate="${dir}../../../perlasm/arm-xlate.pl" and -f $xlate) or
+    die "can't locate arm-xlate.pl";
+
+    open OUT,"| \"$^X\" $xlate $flavour $output";
+    *STDOUT=*OUT;
+} else {
+    open OUT,">$output";
+    *STDOUT=*OUT;
+}
+
+$s0="r0";
+$s1="r1";
+$s2="r2";
+$s3="r3";
+$t1="r4";
+$t2="r5";
+$t3="r6";
+$i1="r7";
+$i2="r8";
+$i3="r9";
+
+$tbl="r10";
+$key="r11";
+$rounds="r12";
+
+$code=<<___;
+#ifndef __KERNEL__
+# include <openssl/arm_arch.h>
+#else
+# define __ARM_ARCH__ __LINUX_ARM_ARCH__
+#endif
+
+@ Silence ARMv8 deprecated IT instruction warnings. This file is used by both
+@ ARMv7 and ARMv8 processors and does not use ARMv8 instructions. (ARMv8 AES
+@ instructions are in aesv8-armx.pl.)
+.arch  armv7-a
+
+.text
+#if defined(__thumb2__) && !defined(__APPLE__)
+.syntax	unified
+.thumb
+#else
+.code	32
+#undef __thumb2__
+#endif
+
+.type	AES_Te,%object
+.align	5
+AES_Te:
+.word	0xc66363a5, 0xf87c7c84, 0xee777799, 0xf67b7b8d
+.word	0xfff2f20d, 0xd66b6bbd, 0xde6f6fb1, 0x91c5c554
+.word	0x60303050, 0x02010103, 0xce6767a9, 0x562b2b7d
+.word	0xe7fefe19, 0xb5d7d762, 0x4dababe6, 0xec76769a
+.word	0x8fcaca45, 0x1f82829d, 0x89c9c940, 0xfa7d7d87
+.word	0xeffafa15, 0xb25959eb, 0x8e4747c9, 0xfbf0f00b
+.word	0x41adadec, 0xb3d4d467, 0x5fa2a2fd, 0x45afafea
+.word	0x239c9cbf, 0x53a4a4f7, 0xe4727296, 0x9bc0c05b
+.word	0x75b7b7c2, 0xe1fdfd1c, 0x3d9393ae, 0x4c26266a
+.word	0x6c36365a, 0x7e3f3f41, 0xf5f7f702, 0x83cccc4f
+.word	0x6834345c, 0x51a5a5f4, 0xd1e5e534, 0xf9f1f108
+.word	0xe2717193, 0xabd8d873, 0x62313153, 0x2a15153f
+.word	0x0804040c, 0x95c7c752, 0x46232365, 0x9dc3c35e
+.word	0x30181828, 0x379696a1, 0x0a05050f, 0x2f9a9ab5
+.word	0x0e070709, 0x24121236, 0x1b80809b, 0xdfe2e23d
+.word	0xcdebeb26, 0x4e272769, 0x7fb2b2cd, 0xea75759f
+.word	0x1209091b, 0x1d83839e, 0x582c2c74, 0x341a1a2e
+.word	0x361b1b2d, 0xdc6e6eb2, 0xb45a5aee, 0x5ba0a0fb
+.word	0xa45252f6, 0x763b3b4d, 0xb7d6d661, 0x7db3b3ce
+.word	0x5229297b, 0xdde3e33e, 0x5e2f2f71, 0x13848497
+.word	0xa65353f5, 0xb9d1d168, 0x00000000, 0xc1eded2c
+.word	0x40202060, 0xe3fcfc1f, 0x79b1b1c8, 0xb65b5bed
+.word	0xd46a6abe, 0x8dcbcb46, 0x67bebed9, 0x7239394b
+.word	0x944a4ade, 0x984c4cd4, 0xb05858e8, 0x85cfcf4a
+.word	0xbbd0d06b, 0xc5efef2a, 0x4faaaae5, 0xedfbfb16
+.word	0x864343c5, 0x9a4d4dd7, 0x66333355, 0x11858594
+.word	0x8a4545cf, 0xe9f9f910, 0x04020206, 0xfe7f7f81
+.word	0xa05050f0, 0x783c3c44, 0x259f9fba, 0x4ba8a8e3
+.word	0xa25151f3, 0x5da3a3fe, 0x804040c0, 0x058f8f8a
+.word	0x3f9292ad, 0x219d9dbc, 0x70383848, 0xf1f5f504
+.word	0x63bcbcdf, 0x77b6b6c1, 0xafdada75, 0x42212163
+.word	0x20101030, 0xe5ffff1a, 0xfdf3f30e, 0xbfd2d26d
+.word	0x81cdcd4c, 0x180c0c14, 0x26131335, 0xc3ecec2f
+.word	0xbe5f5fe1, 0x359797a2, 0x884444cc, 0x2e171739
+.word	0x93c4c457, 0x55a7a7f2, 0xfc7e7e82, 0x7a3d3d47
+.word	0xc86464ac, 0xba5d5de7, 0x3219192b, 0xe6737395
+.word	0xc06060a0, 0x19818198, 0x9e4f4fd1, 0xa3dcdc7f
+.word	0x44222266, 0x542a2a7e, 0x3b9090ab, 0x0b888883
+.word	0x8c4646ca, 0xc7eeee29, 0x6bb8b8d3, 0x2814143c
+.word	0xa7dede79, 0xbc5e5ee2, 0x160b0b1d, 0xaddbdb76
+.word	0xdbe0e03b, 0x64323256, 0x743a3a4e, 0x140a0a1e
+.word	0x924949db, 0x0c06060a, 0x4824246c, 0xb85c5ce4
+.word	0x9fc2c25d, 0xbdd3d36e, 0x43acacef, 0xc46262a6
+.word	0x399191a8, 0x319595a4, 0xd3e4e437, 0xf279798b
+.word	0xd5e7e732, 0x8bc8c843, 0x6e373759, 0xda6d6db7
+.word	0x018d8d8c, 0xb1d5d564, 0x9c4e4ed2, 0x49a9a9e0
+.word	0xd86c6cb4, 0xac5656fa, 0xf3f4f407, 0xcfeaea25
+.word	0xca6565af, 0xf47a7a8e, 0x47aeaee9, 0x10080818
+.word	0x6fbabad5, 0xf0787888, 0x4a25256f, 0x5c2e2e72
+.word	0x381c1c24, 0x57a6a6f1, 0x73b4b4c7, 0x97c6c651
+.word	0xcbe8e823, 0xa1dddd7c, 0xe874749c, 0x3e1f1f21
+.word	0x964b4bdd, 0x61bdbddc, 0x0d8b8b86, 0x0f8a8a85
+.word	0xe0707090, 0x7c3e3e42, 0x71b5b5c4, 0xcc6666aa
+.word	0x904848d8, 0x06030305, 0xf7f6f601, 0x1c0e0e12
+.word	0xc26161a3, 0x6a35355f, 0xae5757f9, 0x69b9b9d0
+.word	0x17868691, 0x99c1c158, 0x3a1d1d27, 0x279e9eb9
+.word	0xd9e1e138, 0xebf8f813, 0x2b9898b3, 0x22111133
+.word	0xd26969bb, 0xa9d9d970, 0x078e8e89, 0x339494a7
+.word	0x2d9b9bb6, 0x3c1e1e22, 0x15878792, 0xc9e9e920
+.word	0x87cece49, 0xaa5555ff, 0x50282878, 0xa5dfdf7a
+.word	0x038c8c8f, 0x59a1a1f8, 0x09898980, 0x1a0d0d17
+.word	0x65bfbfda, 0xd7e6e631, 0x844242c6, 0xd06868b8
+.word	0x824141c3, 0x299999b0, 0x5a2d2d77, 0x1e0f0f11
+.word	0x7bb0b0cb, 0xa85454fc, 0x6dbbbbd6, 0x2c16163a
+@ Te4[256]
+.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
+@ rcon[]
+.word	0x01000000, 0x02000000, 0x04000000, 0x08000000
+.word	0x10000000, 0x20000000, 0x40000000, 0x80000000
+.word	0x1B000000, 0x36000000, 0, 0, 0, 0, 0, 0
+.size	AES_Te,.-AES_Te
+
+@ void aes_nohw_encrypt(const unsigned char *in, unsigned char *out,
+@ 		                  const AES_KEY *key) {
+.global aes_nohw_encrypt
+.type   aes_nohw_encrypt,%function
+.align	5
+aes_nohw_encrypt:
+#ifndef	__thumb2__
+	sub	r3,pc,#8		@ aes_nohw_encrypt
+#else
+	adr	r3,.
+#endif
+	stmdb   sp!,{r1,r4-r12,lr}
+#if defined(__thumb2__) || defined(__APPLE__)
+	adr	$tbl,AES_Te
+#else
+	sub	$tbl,r3,#aes_nohw_encrypt-AES_Te	@ Te
+#endif
+	mov	$rounds,r0		@ inp
+	mov	$key,r2
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	orr	$s3,$s3,$t2,lsl#16
+	orr	$s3,$s3,$t3,lsl#24
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
+	bl	_armv4_AES_encrypt
+
+	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
+	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
+	mov	$t2,$s0,lsr#16		@ manner...
+	mov	$t3,$s0,lsr#8
+	strb	$t1,[$rounds,#0]
+	strb	$t2,[$rounds,#1]
+	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
+	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
+	mov	$t3,$s1,lsr#8
+	strb	$t1,[$rounds,#4]
+	strb	$t2,[$rounds,#5]
+	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
+	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
+	mov	$t3,$s2,lsr#8
+	strb	$t1,[$rounds,#8]
+	strb	$t2,[$rounds,#9]
+	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
+	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
+	mov	$t3,$s3,lsr#8
+	strb	$t1,[$rounds,#12]
+	strb	$t2,[$rounds,#13]
+	strb	$t3,[$rounds,#14]
+	strb	$s3,[$rounds,#15]
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	aes_nohw_encrypt,.-aes_nohw_encrypt
+
+.type   _armv4_AES_encrypt,%function
+.align	2
+_armv4_AES_encrypt:
+	str	lr,[sp,#-4]!		@ push lr
+	ldmia	$key!,{$t1-$i1}
+	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
+	eor	$s1,$s1,$t2
+	eor	$s2,$s2,$t3
+	eor	$s3,$s3,$i1
+	sub	$rounds,$rounds,#1
+	mov	lr,#255
+
+	and	$i1,lr,$s0
+	and	$i2,lr,$s0,lsr#8
+	and	$i3,lr,$s0,lsr#16
+	mov	$s0,$s0,lsr#24
+.Lenc_loop:
+	ldr	$t1,[$tbl,$i1,lsl#2]	@ Te3[s0>>0]
+	and	$i1,lr,$s1,lsr#16	@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Te2[s0>>8]
+	and	$i2,lr,$s1
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Te1[s0>>16]
+	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Te0[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te1[s1>>16]
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te3[s1>>0]
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te2[s1>>8]
+	eor	$s0,$s0,$i1,ror#8
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Te0[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,ror#8
+	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$t3,$i3,ror#8
+	and	$i3,lr,$s2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te2[s2>>8]
+	eor	$s1,$s1,$t1,ror#24
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te1[s2>>16]
+	mov	$s2,$s2,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te3[s2>>0]
+	eor	$s0,$s0,$i1,ror#16
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Te0[s2>>24]
+	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,ror#8
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$t3,$i3,ror#16
+	and	$i3,lr,$s3,lsr#16	@ i2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Te3[s3>>0]
+	eor	$s2,$s2,$t2,ror#16
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Te2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Te1[s3>>16]
+	eor	$s0,$s0,$i1,ror#24
+	ldr	$i1,[$key],#16
+	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Te0[s3>>24]
+	eor	$s2,$s2,$i3,ror#8
+	ldr	$t1,[$key,#-12]
+	eor	$s3,$s3,$t3,ror#8
+
+	ldr	$t2,[$key,#-8]
+	eor	$s0,$s0,$i1
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0,lsr#16
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
+
+	subs	$rounds,$rounds,#1
+	bne	.Lenc_loop
+
+	add	$tbl,$tbl,#2
+
+	ldrb	$t1,[$tbl,$i1,lsl#2]	@ Te4[s0>>0]
+	and	$i1,lr,$s1,lsr#16	@ i0
+	ldrb	$t2,[$tbl,$i2,lsl#2]	@ Te4[s0>>8]
+	and	$i2,lr,$s1
+	ldrb	$t3,[$tbl,$i3,lsl#2]	@ Te4[s0>>16]
+	and	$i3,lr,$s1,lsr#8
+	ldrb	$s0,[$tbl,$s0,lsl#2]	@ Te4[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s1>>16]
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s1>>0]
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s1>>8]
+	eor	$s0,$i1,$s0,lsl#8
+	ldrb	$s1,[$tbl,$s1,lsl#2]	@ Te4[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,lsl#8
+	and	$i2,lr,$s2,lsr#16	@ i1
+	eor	$t3,$i3,$t3,lsl#8
+	and	$i3,lr,$s2
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s2>>8]
+	eor	$s1,$t1,$s1,lsl#24
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s2>>16]
+	mov	$s2,$s2,lsr#24
+
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s2>>0]
+	eor	$s0,$i1,$s0,lsl#8
+	ldrb	$s2,[$tbl,$s2,lsl#2]	@ Te4[s2>>24]
+	and	$i1,lr,$s3		@ i0
+	eor	$s1,$s1,$i2,lsl#16
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,lsl#8
+	and	$i3,lr,$s3,lsr#16	@ i2
+	ldrb	$i1,[$tbl,$i1,lsl#2]	@ Te4[s3>>0]
+	eor	$s2,$t2,$s2,lsl#24
+	ldrb	$i2,[$tbl,$i2,lsl#2]	@ Te4[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldrb	$i3,[$tbl,$i3,lsl#2]	@ Te4[s3>>16]
+	eor	$s0,$i1,$s0,lsl#8
+	ldr	$i1,[$key,#0]
+	ldrb	$s3,[$tbl,$s3,lsl#2]	@ Te4[s3>>24]
+	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
+	eor	$s2,$s2,$i3,lsl#16
+	ldr	$t2,[$key,#8]
+	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
+
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
+
+	sub	$tbl,$tbl,#2
+	ldr	pc,[sp],#4		@ pop and return
+.size	_armv4_AES_encrypt,.-_armv4_AES_encrypt
+
+.global aes_nohw_set_encrypt_key
+.type   aes_nohw_set_encrypt_key,%function
+.align	5
+aes_nohw_set_encrypt_key:
+_armv4_AES_set_encrypt_key:
+#ifndef	__thumb2__
+	sub	r3,pc,#8		@ aes_nohw_set_encrypt_key
+#else
+	adr	r3,.
+#endif
+	teq	r0,#0
+#ifdef	__thumb2__
+	itt	eq			@ Thumb2 thing, sanity check in ARM
+#endif
+	moveq	r0,#-1
+	beq	.Labrt
+	teq	r2,#0
+#ifdef	__thumb2__
+	itt	eq			@ Thumb2 thing, sanity check in ARM
+#endif
+	moveq	r0,#-1
+	beq	.Labrt
+
+	teq	r1,#128
+	beq	.Lok
+	teq	r1,#192
+	beq	.Lok
+	teq	r1,#256
+#ifdef	__thumb2__
+	itt	ne			@ Thumb2 thing, sanity check in ARM
+#endif
+	movne	r0,#-1
+	bne	.Labrt
+
+.Lok:	stmdb   sp!,{r4-r12,lr}
+	mov	$rounds,r0		@ inp
+	mov	lr,r1			@ bits
+	mov	$key,r2			@ key
+
+#if defined(__thumb2__) || defined(__APPLE__)
+	adr	$tbl,AES_Te+1024				@ Te4
+#else
+	sub	$tbl,r3,#_armv4_AES_set_encrypt_key-AES_Te-1024	@ Te4
+#endif
+
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	str	$s0,[$key],#16
+	orr	$s3,$s3,$t2,lsl#16
+	str	$s1,[$key,#-12]
+	orr	$s3,$s3,$t3,lsl#24
+	str	$s2,[$key,#-8]
+	str	$s3,[$key,#-4]
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$key],#16
+	str	$s1,[$key,#-12]
+	str	$s2,[$key,#-8]
+	str	$s3,[$key,#-4]
+#endif
+
+	teq	lr,#128
+	bne	.Lnot128
+	mov	$rounds,#10
+	str	$rounds,[$key,#240-16]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+
+.L128_loop:
+	and	$t2,lr,$s3,lsr#24
+	and	$i1,lr,$s3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$t2,$t2,$t1
+	eor	$s0,$s0,$t2			@ rk[4]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[5]=rk[1]^rk[4]
+	str	$s0,[$key],#16
+	eor	$s2,$s2,$s1			@ rk[6]=rk[2]^rk[5]
+	str	$s1,[$key,#-12]
+	eor	$s3,$s3,$s2			@ rk[7]=rk[3]^rk[6]
+	str	$s2,[$key,#-8]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-4]
+	bne	.L128_loop
+	sub	r2,$key,#176
+	b	.Ldone
+
+.Lnot128:
+#if __ARM_ARCH__<7
+	ldrb	$i2,[$rounds,#19]
+	ldrb	$t1,[$rounds,#18]
+	ldrb	$t2,[$rounds,#17]
+	ldrb	$t3,[$rounds,#16]
+	orr	$i2,$i2,$t1,lsl#8
+	ldrb	$i3,[$rounds,#23]
+	orr	$i2,$i2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#22]
+	orr	$i2,$i2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#21]
+	ldrb	$t3,[$rounds,#20]
+	orr	$i3,$i3,$t1,lsl#8
+	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
+	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#16]
+	ldr	$i3,[$rounds,#20]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
+	str	$i2,[$key],#8
+	str	$i3,[$key,#-4]
+#endif
+
+	teq	lr,#192
+	bne	.Lnot192
+	mov	$rounds,#12
+	str	$rounds,[$key,#240-24]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+	mov	$rounds,#8
+
+.L192_loop:
+	and	$t2,lr,$i3,lsr#24
+	and	$i1,lr,$i3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$i3,$t2,$t1
+	eor	$s0,$s0,$i3			@ rk[6]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[7]=rk[1]^rk[6]
+	str	$s0,[$key],#24
+	eor	$s2,$s2,$s1			@ rk[8]=rk[2]^rk[7]
+	str	$s1,[$key,#-20]
+	eor	$s3,$s3,$s2			@ rk[9]=rk[3]^rk[8]
+	str	$s2,[$key,#-16]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-12]
+#ifdef	__thumb2__
+	itt	eq				@ Thumb2 thing, sanity check in ARM
+#endif
+	subeq	r2,$key,#216
+	beq	.Ldone
+
+	ldr	$i1,[$key,#-32]
+	ldr	$i2,[$key,#-28]
+	eor	$i1,$i1,$s3			@ rk[10]=rk[4]^rk[9]
+	eor	$i3,$i2,$i1			@ rk[11]=rk[5]^rk[10]
+	str	$i1,[$key,#-8]
+	str	$i3,[$key,#-4]
+	b	.L192_loop
+
+.Lnot192:
+#if __ARM_ARCH__<7
+	ldrb	$i2,[$rounds,#27]
+	ldrb	$t1,[$rounds,#26]
+	ldrb	$t2,[$rounds,#25]
+	ldrb	$t3,[$rounds,#24]
+	orr	$i2,$i2,$t1,lsl#8
+	ldrb	$i3,[$rounds,#31]
+	orr	$i2,$i2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#30]
+	orr	$i2,$i2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#29]
+	ldrb	$t3,[$rounds,#28]
+	orr	$i3,$i3,$t1,lsl#8
+	orr	$i3,$i3,$t2,lsl#16
+	str	$i2,[$key],#8
+	orr	$i3,$i3,$t3,lsl#24
+	str	$i3,[$key,#-4]
+#else
+	ldr	$i2,[$rounds,#24]
+	ldr	$i3,[$rounds,#28]
+#ifdef __ARMEL__
+	rev	$i2,$i2
+	rev	$i3,$i3
+#endif
+	str	$i2,[$key],#8
+	str	$i3,[$key,#-4]
+#endif
+
+	mov	$rounds,#14
+	str	$rounds,[$key,#240-32]
+	add	$t3,$tbl,#256			@ rcon
+	mov	lr,#255
+	mov	$rounds,#7
+
+.L256_loop:
+	and	$t2,lr,$i3,lsr#24
+	and	$i1,lr,$i3,lsr#16
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$i3,lsr#8
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$i3
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#24
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$t3],#4			@ rcon[i++]
+	orr	$t2,$t2,$i3,lsl#8
+	eor	$i3,$t2,$t1
+	eor	$s0,$s0,$i3			@ rk[8]=rk[0]^...
+	eor	$s1,$s1,$s0			@ rk[9]=rk[1]^rk[8]
+	str	$s0,[$key],#32
+	eor	$s2,$s2,$s1			@ rk[10]=rk[2]^rk[9]
+	str	$s1,[$key,#-28]
+	eor	$s3,$s3,$s2			@ rk[11]=rk[3]^rk[10]
+	str	$s2,[$key,#-24]
+	subs	$rounds,$rounds,#1
+	str	$s3,[$key,#-20]
+#ifdef	__thumb2__
+	itt	eq				@ Thumb2 thing, sanity check in ARM
+#endif
+	subeq	r2,$key,#256
+	beq	.Ldone
+
+	and	$t2,lr,$s3
+	and	$i1,lr,$s3,lsr#8
+	ldrb	$t2,[$tbl,$t2]
+	and	$i2,lr,$s3,lsr#16
+	ldrb	$i1,[$tbl,$i1]
+	and	$i3,lr,$s3,lsr#24
+	ldrb	$i2,[$tbl,$i2]
+	orr	$t2,$t2,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]
+	orr	$t2,$t2,$i2,lsl#16
+	ldr	$t1,[$key,#-48]
+	orr	$t2,$t2,$i3,lsl#24
+
+	ldr	$i1,[$key,#-44]
+	ldr	$i2,[$key,#-40]
+	eor	$t1,$t1,$t2			@ rk[12]=rk[4]^...
+	ldr	$i3,[$key,#-36]
+	eor	$i1,$i1,$t1			@ rk[13]=rk[5]^rk[12]
+	str	$t1,[$key,#-16]
+	eor	$i2,$i2,$i1			@ rk[14]=rk[6]^rk[13]
+	str	$i1,[$key,#-12]
+	eor	$i3,$i3,$i2			@ rk[15]=rk[7]^rk[14]
+	str	$i2,[$key,#-8]
+	str	$i3,[$key,#-4]
+	b	.L256_loop
+
+.align	2
+.Ldone:	mov	r0,#0
+	ldmia   sp!,{r4-r12,lr}
+.Labrt:
+#if __ARM_ARCH__>=5
+	ret				@ bx lr
+#else
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	aes_nohw_set_encrypt_key,.-aes_nohw_set_encrypt_key
+
+.global aes_nohw_set_decrypt_key
+.type   aes_nohw_set_decrypt_key,%function
+.align	5
+aes_nohw_set_decrypt_key:
+	str	lr,[sp,#-4]!            @ push lr
+	bl	_armv4_AES_set_encrypt_key
+	teq	r0,#0
+	ldr	lr,[sp],#4              @ pop lr
+	bne	.Labrt
+
+	mov	r0,r2			@ aes_nohw_set_encrypt_key preserves r2,
+	mov	r1,r2			@ which is AES_KEY *key
+	b	_armv4_AES_set_enc2dec_key
+.size	aes_nohw_set_decrypt_key,.-aes_nohw_set_decrypt_key
+
+@ void AES_set_enc2dec_key(const AES_KEY *inp,AES_KEY *out)
+.global	AES_set_enc2dec_key
+.type	AES_set_enc2dec_key,%function
+.align	5
+AES_set_enc2dec_key:
+_armv4_AES_set_enc2dec_key:
+	stmdb   sp!,{r4-r12,lr}
+
+	ldr	$rounds,[r0,#240]
+	mov	$i1,r0			@ input
+	add	$i2,r0,$rounds,lsl#4
+	mov	$key,r1			@ output
+	add	$tbl,r1,$rounds,lsl#4
+	str	$rounds,[r1,#240]
+
+.Linv:	ldr	$s0,[$i1],#16
+	ldr	$s1,[$i1,#-12]
+	ldr	$s2,[$i1,#-8]
+	ldr	$s3,[$i1,#-4]
+	ldr	$t1,[$i2],#-16
+	ldr	$t2,[$i2,#16+4]
+	ldr	$t3,[$i2,#16+8]
+	ldr	$i3,[$i2,#16+12]
+	str	$s0,[$tbl],#-16
+	str	$s1,[$tbl,#16+4]
+	str	$s2,[$tbl,#16+8]
+	str	$s3,[$tbl,#16+12]
+	str	$t1,[$key],#16
+	str	$t2,[$key,#-12]
+	str	$t3,[$key,#-8]
+	str	$i3,[$key,#-4]
+	teq	$i1,$i2
+	bne	.Linv
+
+	ldr	$s0,[$i1]
+	ldr	$s1,[$i1,#4]
+	ldr	$s2,[$i1,#8]
+	ldr	$s3,[$i1,#12]
+	str	$s0,[$key]
+	str	$s1,[$key,#4]
+	str	$s2,[$key,#8]
+	str	$s3,[$key,#12]
+	sub	$key,$key,$rounds,lsl#3
+___
+$mask80=$i1;
+$mask1b=$i2;
+$mask7f=$i3;
+$code.=<<___;
+	ldr	$s0,[$key,#16]!		@ prefetch tp1
+	mov	$mask80,#0x80
+	mov	$mask1b,#0x1b
+	orr	$mask80,$mask80,#0x8000
+	orr	$mask1b,$mask1b,#0x1b00
+	orr	$mask80,$mask80,$mask80,lsl#16
+	orr	$mask1b,$mask1b,$mask1b,lsl#16
+	sub	$rounds,$rounds,#1
+	mvn	$mask7f,$mask80
+	mov	$rounds,$rounds,lsl#2	@ (rounds-1)*4
+
+.Lmix:	and	$t1,$s0,$mask80
+	and	$s1,$s0,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s1,$t1,$s1,lsl#1	@ tp2
+
+	and	$t1,$s1,$mask80
+	and	$s2,$s1,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s2,$t1,$s2,lsl#1	@ tp4
+
+	and	$t1,$s2,$mask80
+	and	$s3,$s2,$mask7f
+	sub	$t1,$t1,$t1,lsr#7
+	and	$t1,$t1,$mask1b
+	eor	$s3,$t1,$s3,lsl#1	@ tp8
+
+	eor	$t1,$s1,$s2
+	eor	$t2,$s0,$s3		@ tp9
+	eor	$t1,$t1,$s3		@ tpe
+	eor	$t1,$t1,$s1,ror#24
+	eor	$t1,$t1,$t2,ror#24	@ ^= ROTATE(tpb=tp9^tp2,8)
+	eor	$t1,$t1,$s2,ror#16
+	eor	$t1,$t1,$t2,ror#16	@ ^= ROTATE(tpd=tp9^tp4,16)
+	eor	$t1,$t1,$t2,ror#8	@ ^= ROTATE(tp9,24)
+
+	ldr	$s0,[$key,#4]		@ prefetch tp1
+	str	$t1,[$key],#4
+	subs	$rounds,$rounds,#1
+	bne	.Lmix
+
+	mov	r0,#0
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	AES_set_enc2dec_key,.-AES_set_enc2dec_key
+
+.type	AES_Td,%object
+.align	5
+AES_Td:
+.word	0x51f4a750, 0x7e416553, 0x1a17a4c3, 0x3a275e96
+.word	0x3bab6bcb, 0x1f9d45f1, 0xacfa58ab, 0x4be30393
+.word	0x2030fa55, 0xad766df6, 0x88cc7691, 0xf5024c25
+.word	0x4fe5d7fc, 0xc52acbd7, 0x26354480, 0xb562a38f
+.word	0xdeb15a49, 0x25ba1b67, 0x45ea0e98, 0x5dfec0e1
+.word	0xc32f7502, 0x814cf012, 0x8d4697a3, 0x6bd3f9c6
+.word	0x038f5fe7, 0x15929c95, 0xbf6d7aeb, 0x955259da
+.word	0xd4be832d, 0x587421d3, 0x49e06929, 0x8ec9c844
+.word	0x75c2896a, 0xf48e7978, 0x99583e6b, 0x27b971dd
+.word	0xbee14fb6, 0xf088ad17, 0xc920ac66, 0x7dce3ab4
+.word	0x63df4a18, 0xe51a3182, 0x97513360, 0x62537f45
+.word	0xb16477e0, 0xbb6bae84, 0xfe81a01c, 0xf9082b94
+.word	0x70486858, 0x8f45fd19, 0x94de6c87, 0x527bf8b7
+.word	0xab73d323, 0x724b02e2, 0xe31f8f57, 0x6655ab2a
+.word	0xb2eb2807, 0x2fb5c203, 0x86c57b9a, 0xd33708a5
+.word	0x302887f2, 0x23bfa5b2, 0x02036aba, 0xed16825c
+.word	0x8acf1c2b, 0xa779b492, 0xf307f2f0, 0x4e69e2a1
+.word	0x65daf4cd, 0x0605bed5, 0xd134621f, 0xc4a6fe8a
+.word	0x342e539d, 0xa2f355a0, 0x058ae132, 0xa4f6eb75
+.word	0x0b83ec39, 0x4060efaa, 0x5e719f06, 0xbd6e1051
+.word	0x3e218af9, 0x96dd063d, 0xdd3e05ae, 0x4de6bd46
+.word	0x91548db5, 0x71c45d05, 0x0406d46f, 0x605015ff
+.word	0x1998fb24, 0xd6bde997, 0x894043cc, 0x67d99e77
+.word	0xb0e842bd, 0x07898b88, 0xe7195b38, 0x79c8eedb
+.word	0xa17c0a47, 0x7c420fe9, 0xf8841ec9, 0x00000000
+.word	0x09808683, 0x322bed48, 0x1e1170ac, 0x6c5a724e
+.word	0xfd0efffb, 0x0f853856, 0x3daed51e, 0x362d3927
+.word	0x0a0fd964, 0x685ca621, 0x9b5b54d1, 0x24362e3a
+.word	0x0c0a67b1, 0x9357e70f, 0xb4ee96d2, 0x1b9b919e
+.word	0x80c0c54f, 0x61dc20a2, 0x5a774b69, 0x1c121a16
+.word	0xe293ba0a, 0xc0a02ae5, 0x3c22e043, 0x121b171d
+.word	0x0e090d0b, 0xf28bc7ad, 0x2db6a8b9, 0x141ea9c8
+.word	0x57f11985, 0xaf75074c, 0xee99ddbb, 0xa37f60fd
+.word	0xf701269f, 0x5c72f5bc, 0x44663bc5, 0x5bfb7e34
+.word	0x8b432976, 0xcb23c6dc, 0xb6edfc68, 0xb8e4f163
+.word	0xd731dcca, 0x42638510, 0x13972240, 0x84c61120
+.word	0x854a247d, 0xd2bb3df8, 0xaef93211, 0xc729a16d
+.word	0x1d9e2f4b, 0xdcb230f3, 0x0d8652ec, 0x77c1e3d0
+.word	0x2bb3166c, 0xa970b999, 0x119448fa, 0x47e96422
+.word	0xa8fc8cc4, 0xa0f03f1a, 0x567d2cd8, 0x223390ef
+.word	0x87494ec7, 0xd938d1c1, 0x8ccaa2fe, 0x98d40b36
+.word	0xa6f581cf, 0xa57ade28, 0xdab78e26, 0x3fadbfa4
+.word	0x2c3a9de4, 0x5078920d, 0x6a5fcc9b, 0x547e4662
+.word	0xf68d13c2, 0x90d8b8e8, 0x2e39f75e, 0x82c3aff5
+.word	0x9f5d80be, 0x69d0937c, 0x6fd52da9, 0xcf2512b3
+.word	0xc8ac993b, 0x10187da7, 0xe89c636e, 0xdb3bbb7b
+.word	0xcd267809, 0x6e5918f4, 0xec9ab701, 0x834f9aa8
+.word	0xe6956e65, 0xaaffe67e, 0x21bccf08, 0xef15e8e6
+.word	0xbae79bd9, 0x4a6f36ce, 0xea9f09d4, 0x29b07cd6
+.word	0x31a4b2af, 0x2a3f2331, 0xc6a59430, 0x35a266c0
+.word	0x744ebc37, 0xfc82caa6, 0xe090d0b0, 0x33a7d815
+.word	0xf104984a, 0x41ecdaf7, 0x7fcd500e, 0x1791f62f
+.word	0x764dd68d, 0x43efb04d, 0xccaa4d54, 0xe49604df
+.word	0x9ed1b5e3, 0x4c6a881b, 0xc12c1fb8, 0x4665517f
+.word	0x9d5eea04, 0x018c355d, 0xfa877473, 0xfb0b412e
+.word	0xb3671d5a, 0x92dbd252, 0xe9105633, 0x6dd64713
+.word	0x9ad7618c, 0x37a10c7a, 0x59f8148e, 0xeb133c89
+.word	0xcea927ee, 0xb761c935, 0xe11ce5ed, 0x7a47b13c
+.word	0x9cd2df59, 0x55f2733f, 0x1814ce79, 0x73c737bf
+.word	0x53f7cdea, 0x5ffdaa5b, 0xdf3d6f14, 0x7844db86
+.word	0xcaaff381, 0xb968c43e, 0x3824342c, 0xc2a3405f
+.word	0x161dc372, 0xbce2250c, 0x283c498b, 0xff0d9541
+.word	0x39a80171, 0x080cb3de, 0xd8b4e49c, 0x6456c190
+.word	0x7bcb8461, 0xd532b670, 0x486c5c74, 0xd0b85742
+@ Td4[256]
+.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
+.size	AES_Td,.-AES_Td
+
+@ void aes_nohw_decrypt(const unsigned char *in, unsigned char *out,
+@ 		                  const AES_KEY *key) {
+.global aes_nohw_decrypt
+.type   aes_nohw_decrypt,%function
+.align	5
+aes_nohw_decrypt:
+#ifndef	__thumb2__
+	sub	r3,pc,#8		@ aes_nohw_decrypt
+#else
+	adr	r3,.
+#endif
+	stmdb   sp!,{r1,r4-r12,lr}
+#if defined(__thumb2__) || defined(__APPLE__)
+	adr	$tbl,AES_Td
+#else
+	sub	$tbl,r3,#aes_nohw_decrypt-AES_Td	@ Td
+#endif
+	mov	$rounds,r0		@ inp
+	mov	$key,r2
+#if __ARM_ARCH__<7
+	ldrb	$s0,[$rounds,#3]	@ load input data in endian-neutral
+	ldrb	$t1,[$rounds,#2]	@ manner...
+	ldrb	$t2,[$rounds,#1]
+	ldrb	$t3,[$rounds,#0]
+	orr	$s0,$s0,$t1,lsl#8
+	ldrb	$s1,[$rounds,#7]
+	orr	$s0,$s0,$t2,lsl#16
+	ldrb	$t1,[$rounds,#6]
+	orr	$s0,$s0,$t3,lsl#24
+	ldrb	$t2,[$rounds,#5]
+	ldrb	$t3,[$rounds,#4]
+	orr	$s1,$s1,$t1,lsl#8
+	ldrb	$s2,[$rounds,#11]
+	orr	$s1,$s1,$t2,lsl#16
+	ldrb	$t1,[$rounds,#10]
+	orr	$s1,$s1,$t3,lsl#24
+	ldrb	$t2,[$rounds,#9]
+	ldrb	$t3,[$rounds,#8]
+	orr	$s2,$s2,$t1,lsl#8
+	ldrb	$s3,[$rounds,#15]
+	orr	$s2,$s2,$t2,lsl#16
+	ldrb	$t1,[$rounds,#14]
+	orr	$s2,$s2,$t3,lsl#24
+	ldrb	$t2,[$rounds,#13]
+	ldrb	$t3,[$rounds,#12]
+	orr	$s3,$s3,$t1,lsl#8
+	orr	$s3,$s3,$t2,lsl#16
+	orr	$s3,$s3,$t3,lsl#24
+#else
+	ldr	$s0,[$rounds,#0]
+	ldr	$s1,[$rounds,#4]
+	ldr	$s2,[$rounds,#8]
+	ldr	$s3,[$rounds,#12]
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+#endif
+	bl	_armv4_AES_decrypt
+
+	ldr	$rounds,[sp],#4		@ pop out
+#if __ARM_ARCH__>=7
+#ifdef __ARMEL__
+	rev	$s0,$s0
+	rev	$s1,$s1
+	rev	$s2,$s2
+	rev	$s3,$s3
+#endif
+	str	$s0,[$rounds,#0]
+	str	$s1,[$rounds,#4]
+	str	$s2,[$rounds,#8]
+	str	$s3,[$rounds,#12]
+#else
+	mov	$t1,$s0,lsr#24		@ write output in endian-neutral
+	mov	$t2,$s0,lsr#16		@ manner...
+	mov	$t3,$s0,lsr#8
+	strb	$t1,[$rounds,#0]
+	strb	$t2,[$rounds,#1]
+	mov	$t1,$s1,lsr#24
+	strb	$t3,[$rounds,#2]
+	mov	$t2,$s1,lsr#16
+	strb	$s0,[$rounds,#3]
+	mov	$t3,$s1,lsr#8
+	strb	$t1,[$rounds,#4]
+	strb	$t2,[$rounds,#5]
+	mov	$t1,$s2,lsr#24
+	strb	$t3,[$rounds,#6]
+	mov	$t2,$s2,lsr#16
+	strb	$s1,[$rounds,#7]
+	mov	$t3,$s2,lsr#8
+	strb	$t1,[$rounds,#8]
+	strb	$t2,[$rounds,#9]
+	mov	$t1,$s3,lsr#24
+	strb	$t3,[$rounds,#10]
+	mov	$t2,$s3,lsr#16
+	strb	$s2,[$rounds,#11]
+	mov	$t3,$s3,lsr#8
+	strb	$t1,[$rounds,#12]
+	strb	$t2,[$rounds,#13]
+	strb	$t3,[$rounds,#14]
+	strb	$s3,[$rounds,#15]
+#endif
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r12,pc}
+#else
+	ldmia   sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	bx	lr			@ interoperable with Thumb ISA:-)
+#endif
+.size	aes_nohw_decrypt,.-aes_nohw_decrypt
+
+.type   _armv4_AES_decrypt,%function
+.align	2
+_armv4_AES_decrypt:
+	str	lr,[sp,#-4]!		@ push lr
+	ldmia	$key!,{$t1-$i1}
+	eor	$s0,$s0,$t1
+	ldr	$rounds,[$key,#240-16]
+	eor	$s1,$s1,$t2
+	eor	$s2,$s2,$t3
+	eor	$s3,$s3,$i1
+	sub	$rounds,$rounds,#1
+	mov	lr,#255
+
+	and	$i1,lr,$s0,lsr#16
+	and	$i2,lr,$s0,lsr#8
+	and	$i3,lr,$s0
+	mov	$s0,$s0,lsr#24
+.Ldec_loop:
+	ldr	$t1,[$tbl,$i1,lsl#2]	@ Td1[s0>>16]
+	and	$i1,lr,$s1		@ i0
+	ldr	$t2,[$tbl,$i2,lsl#2]	@ Td2[s0>>8]
+	and	$i2,lr,$s1,lsr#16
+	ldr	$t3,[$tbl,$i3,lsl#2]	@ Td3[s0>>0]
+	and	$i3,lr,$s1,lsr#8
+	ldr	$s0,[$tbl,$s0,lsl#2]	@ Td0[s0>>24]
+	mov	$s1,$s1,lsr#24
+
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td3[s1>>0]
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td1[s1>>16]
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td2[s1>>8]
+	eor	$s0,$s0,$i1,ror#24
+	ldr	$s1,[$tbl,$s1,lsl#2]	@ Td0[s1>>24]
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$i2,$t2,ror#8
+	and	$i2,lr,$s2		@ i1
+	eor	$t3,$i3,$t3,ror#8
+	and	$i3,lr,$s2,lsr#16
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td2[s2>>8]
+	eor	$s1,$s1,$t1,ror#8
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td3[s2>>0]
+	mov	$s2,$s2,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td1[s2>>16]
+	eor	$s0,$s0,$i1,ror#16
+	ldr	$s2,[$tbl,$s2,lsl#2]	@ Td0[s2>>24]
+	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s1,$s1,$i2,ror#24
+	and	$i2,lr,$s3,lsr#8	@ i1
+	eor	$t3,$i3,$t3,ror#8
+	and	$i3,lr,$s3		@ i2
+	ldr	$i1,[$tbl,$i1,lsl#2]	@ Td1[s3>>16]
+	eor	$s2,$s2,$t2,ror#8
+	ldr	$i2,[$tbl,$i2,lsl#2]	@ Td2[s3>>8]
+	mov	$s3,$s3,lsr#24
+
+	ldr	$i3,[$tbl,$i3,lsl#2]	@ Td3[s3>>0]
+	eor	$s0,$s0,$i1,ror#8
+	ldr	$i1,[$key],#16
+	eor	$s1,$s1,$i2,ror#16
+	ldr	$s3,[$tbl,$s3,lsl#2]	@ Td0[s3>>24]
+	eor	$s2,$s2,$i3,ror#24
+
+	ldr	$t1,[$key,#-12]
+	eor	$s0,$s0,$i1
+	ldr	$t2,[$key,#-8]
+	eor	$s3,$s3,$t3,ror#8
+	ldr	$t3,[$key,#-4]
+	and	$i1,lr,$s0,lsr#16
+	eor	$s1,$s1,$t1
+	and	$i2,lr,$s0,lsr#8
+	eor	$s2,$s2,$t2
+	and	$i3,lr,$s0
+	eor	$s3,$s3,$t3
+	mov	$s0,$s0,lsr#24
+
+	subs	$rounds,$rounds,#1
+	bne	.Ldec_loop
+
+	add	$tbl,$tbl,#1024
+
+	ldr	$t2,[$tbl,#0]		@ prefetch Td4
+	ldr	$t3,[$tbl,#32]
+	ldr	$t1,[$tbl,#64]
+	ldr	$t2,[$tbl,#96]
+	ldr	$t3,[$tbl,#128]
+	ldr	$t1,[$tbl,#160]
+	ldr	$t2,[$tbl,#192]
+	ldr	$t3,[$tbl,#224]
+
+	ldrb	$s0,[$tbl,$s0]		@ Td4[s0>>24]
+	ldrb	$t1,[$tbl,$i1]		@ Td4[s0>>16]
+	and	$i1,lr,$s1		@ i0
+	ldrb	$t2,[$tbl,$i2]		@ Td4[s0>>8]
+	and	$i2,lr,$s1,lsr#16
+	ldrb	$t3,[$tbl,$i3]		@ Td4[s0>>0]
+	and	$i3,lr,$s1,lsr#8
+
+	add	$s1,$tbl,$s1,lsr#24
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s1>>0]
+	ldrb	$s1,[$s1]		@ Td4[s1>>24]
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s1>>16]
+	eor	$s0,$i1,$s0,lsl#24
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s1>>8]
+	eor	$s1,$t1,$s1,lsl#8
+	and	$i1,lr,$s2,lsr#8	@ i0
+	eor	$t2,$t2,$i2,lsl#8
+	and	$i2,lr,$s2		@ i1
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s2>>8]
+	eor	$t3,$t3,$i3,lsl#8
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s2>>0]
+	and	$i3,lr,$s2,lsr#16
+
+	add	$s2,$tbl,$s2,lsr#24
+	ldrb	$s2,[$s2]		@ Td4[s2>>24]
+	eor	$s0,$s0,$i1,lsl#8
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s2>>16]
+	eor	$s1,$i2,$s1,lsl#16
+	and	$i1,lr,$s3,lsr#16	@ i0
+	eor	$s2,$t2,$s2,lsl#16
+	and	$i2,lr,$s3,lsr#8	@ i1
+	ldrb	$i1,[$tbl,$i1]		@ Td4[s3>>16]
+	eor	$t3,$t3,$i3,lsl#16
+	ldrb	$i2,[$tbl,$i2]		@ Td4[s3>>8]
+	and	$i3,lr,$s3		@ i2
+
+	add	$s3,$tbl,$s3,lsr#24
+	ldrb	$i3,[$tbl,$i3]		@ Td4[s3>>0]
+	ldrb	$s3,[$s3]		@ Td4[s3>>24]
+	eor	$s0,$s0,$i1,lsl#16
+	ldr	$i1,[$key,#0]
+	eor	$s1,$s1,$i2,lsl#8
+	ldr	$t1,[$key,#4]
+	eor	$s2,$i3,$s2,lsl#8
+	ldr	$t2,[$key,#8]
+	eor	$s3,$t3,$s3,lsl#24
+	ldr	$t3,[$key,#12]
+
+	eor	$s0,$s0,$i1
+	eor	$s1,$s1,$t1
+	eor	$s2,$s2,$t2
+	eor	$s3,$s3,$t3
+
+	sub	$tbl,$tbl,#1024
+	ldr	pc,[sp],#4		@ pop and return
+.size	_armv4_AES_decrypt,.-_armv4_AES_decrypt
+.asciz	"AES for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
+.align	2
+___
+
+$code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm;	# make it possible to compile with -march=armv4
+$code =~ s/\bret\b/bx\tlr/gm;
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+	last if (!s/^#/@/ and !/^$/);
+	print;
+}
+close SELF;
+
+print $code;
+close STDOUT or die "error closing STDOUT";	# enforce flush
diff --git a/crypto/fipsmodule/aes/asm/aes-x86_64.pl b/crypto/fipsmodule/aes/asm/aes-x86_64.pl
new file mode 100755
index 0000000..5b95785
--- /dev/null
+++ b/crypto/fipsmodule/aes/asm/aes-x86_64.pl
@@ -0,0 +1,2909 @@
+#! /usr/bin/env perl
+# Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
+#
+# Licensed under the OpenSSL license (the "License").  You may not use
+# this file except in compliance with the License.  You can obtain a copy
+# in the file LICENSE in the source distribution or at
+# https://www.openssl.org/source/license.html
+
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# Version 2.1.
+#
+# aes-*-cbc benchmarks are improved by >70% [compared to gcc 3.3.2 on
+# Opteron 240 CPU] plus all the bells-n-whistles from 32-bit version
+# [you'll notice a lot of resemblance], such as compressed S-boxes
+# in little-endian byte order, prefetch of these tables in CBC mode,
+# as well as avoiding L1 cache aliasing between stack frame and key
+# schedule and already mentioned tables, compressed Td4...
+#
+# Performance in number of cycles per processed byte for 128-bit key:
+#
+#		ECB encrypt	ECB decrypt	CBC large chunk
+# AMD64		33		43		13.0
+# EM64T		38		56		18.6(*)
+# Core 2	30		42		14.5(*)
+# Atom		65		86		32.1(*)
+#
+# (*) with hyper-threading off
+
+$flavour = shift;
+$output  = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
+
+$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
+( $xlate="${dir}../../../perlasm/x86_64-xlate.pl" and -f $xlate) or
+die "can't locate x86_64-xlate.pl";
+
+open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
+*STDOUT=*OUT;
+
+$verticalspin=1;	# unlike 32-bit version $verticalspin performs
+			# ~15% better on both AMD and Intel cores
+$speed_limit=512;	# see aes-586.pl for details
+
+$code=".text\n";
+
+$s0="%eax";
+$s1="%ebx";
+$s2="%ecx";
+$s3="%edx";
+$acc0="%esi";	$mask80="%rsi";
+$acc1="%edi";	$maskfe="%rdi";
+$acc2="%ebp";	$mask1b="%rbp";
+$inp="%r8";
+$out="%r9";
+$t0="%r10d";
+$t1="%r11d";
+$t2="%r12d";
+$rnds="%r13d";
+$sbox="%r14";
+$key="%r15";
+
+sub hi() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1h/;	$r; }
+sub lo() { my $r=shift;	$r =~ s/%[er]([a-d])x/%\1l/;
+			$r =~ s/%[er]([sd]i)/%\1l/;
+			$r =~ s/%(r[0-9]+)[d]?/%\1b/;	$r; }
+sub LO() { my $r=shift; $r =~ s/%r([a-z]+)/%e\1/;
+			$r =~ s/%r([0-9]+)/%r\1d/;	$r; }
+sub _data_word()
+{ my $i;
+    while(defined($i=shift)) { $code.=sprintf".long\t0x%08x,0x%08x\n",$i,$i; }
+}
+sub data_word()
+{ my $i;
+  my $last=pop(@_);
+    $code.=".long\t";
+    while(defined($i=shift)) { $code.=sprintf"0x%08x,",$i; }
+    $code.=sprintf"0x%08x\n",$last;
+}
+
+sub data_byte()
+{ my $i;
+  my $last=pop(@_);
+    $code.=".byte\t";
+    while(defined($i=shift)) { $code.=sprintf"0x%02x,",$i&0xff; }
+    $code.=sprintf"0x%02x\n",$last&0xff;
+}
+
+sub encvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	# favor 3-way issue Opteron pipeline...
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	mov	0($sbox,$acc0,8),$t0
+	mov	0($sbox,$acc1,8),$t1
+	mov	0($sbox,$acc2,8),$t2
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	movzb	`&lo("$s3")`,$acc2
+	xor	3($sbox,$acc0,8),$t0
+	xor	3($sbox,$acc1,8),$t1
+	mov	0($sbox,$acc2,8),$t3
+
+	movzb	`&hi("$s3")`,$acc0
+	shr	\$16,$s2
+	movzb	`&hi("$s0")`,$acc2
+	xor	3($sbox,$acc0,8),$t2
+	shr	\$16,$s3
+	xor	3($sbox,$acc2,8),$t3
+
+	shr	\$16,$s1
+	lea	16($key),$key
+	shr	\$16,$s0
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	xor	2($sbox,$acc0,8),$t0
+	xor	2($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t2
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	movzb	`&lo("$s1")`,$acc2
+	xor	1($sbox,$acc0,8),$t0
+	xor	1($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t3
+
+	mov	12($key),$s3
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	mov	0($key),$s0
+	xor	1($sbox,$acc1,8),$t2
+	xor	1($sbox,$acc2,8),$t3
+
+	mov	4($key),$s1
+	mov	8($key),$s2
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub enclastvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	movzb	2($sbox,$acc0,8),$t0
+	movzb	2($sbox,$acc1,8),$t1
+	movzb	2($sbox,$acc2,8),$t2
+
+	movzb	`&lo("$s3")`,$acc0
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	movzb	2($sbox,$acc0,8),$t3
+	mov	0($sbox,$acc1,8),$acc1	#$t0
+	mov	0($sbox,$acc2,8),$acc2	#$t1
+
+	and	\$0x0000ff00,$acc1
+	and	\$0x0000ff00,$acc2
+
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+	shr	\$16,$s2
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	shr	\$16,$s3
+	mov	0($sbox,$acc0,8),$acc0	#$t2
+	mov	0($sbox,$acc1,8),$acc1	#$t3
+
+	and	\$0x0000ff00,$acc0
+	and	\$0x0000ff00,$acc1
+	shr	\$16,$s1
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+	shr	\$16,$s0
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	mov	0($sbox,$acc0,8),$acc0	#$t0
+	mov	0($sbox,$acc1,8),$acc1	#$t1
+	mov	0($sbox,$acc2,8),$acc2	#$t2
+
+	and	\$0x00ff0000,$acc0
+	and	\$0x00ff0000,$acc1
+	and	\$0x00ff0000,$acc2
+
+	xor	$acc0,$t0
+	xor	$acc1,$t1
+	xor	$acc2,$t2
+
+	movzb	`&lo("$s1")`,$acc0
+	movzb	`&hi("$s3")`,$acc1
+	movzb	`&hi("$s0")`,$acc2
+	mov	0($sbox,$acc0,8),$acc0	#$t3
+	mov	2($sbox,$acc1,8),$acc1	#$t0
+	mov	2($sbox,$acc2,8),$acc2	#$t1
+
+	and	\$0x00ff0000,$acc0
+	and	\$0xff000000,$acc1
+	and	\$0xff000000,$acc2
+
+	xor	$acc0,$t3
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	mov	16+12($key),$s3
+	mov	2($sbox,$acc0,8),$acc0	#$t2
+	mov	2($sbox,$acc1,8),$acc1	#$t3
+	mov	16+0($key),$s0
+
+	and	\$0xff000000,$acc0
+	and	\$0xff000000,$acc1
+
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+
+	mov	16+4($key),$s1
+	mov	16+8($key),$s2
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub encstep()
+{ my ($i,@s) = @_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	if ($i==3) {
+		$tmp0=$s[1];
+		$tmp1=$s[2];
+		$tmp2=$s[3];
+	}
+	$code.="	movzb	".&lo($s[0]).",$out\n";
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	lea	16($key),$key\n"	if ($i==0);
+
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	mov	0($sbox,$out,8),$out\n";
+
+	$code.="	shr	\$16,$tmp1\n";
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+	$code.="	xor	3($sbox,$tmp0,8),$out\n";
+
+	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+	$code.="	xor	4*$i($key),$out\n";
+
+	$code.="	xor	2($sbox,$tmp1,8),$out\n";
+	$code.="	xor	1($sbox,$tmp2,8),$out\n";
+
+	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+sub enclast()
+{ my ($i,@s)=@_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	if ($i==3) {
+		$tmp0=$s[1];
+		$tmp1=$s[2];
+		$tmp2=$s[3];
+	}
+	$code.="	movzb	".&lo($s[0]).",$out\n";
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+
+	$code.="	mov	2($sbox,$out,8),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+	$code.="	and	\$0x000000ff,$out\n";
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	movzb	".&lo($tmp1).",$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	mov	0($sbox,$tmp0,8),$tmp0\n";
+	$code.="	mov	0($sbox,$tmp1,8),$tmp1\n";
+	$code.="	mov	2($sbox,$tmp2,8),$tmp2\n";
+
+	$code.="	and	\$0x0000ff00,$tmp0\n";
+	$code.="	and	\$0x00ff0000,$tmp1\n";
+	$code.="	and	\$0xff000000,$tmp2\n";
+
+	$code.="	xor	$tmp0,$out\n";
+	$code.="	mov	$t0,$s[1]\n"		if ($i==3);
+	$code.="	xor	$tmp1,$out\n";
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	xor	$tmp2,$out\n";
+	$code.="	mov	$t2,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+$code.=<<___;
+.type	_x86_64_AES_encrypt,\@abi-omnipotent
+.align	16
+_x86_64_AES_encrypt:
+	xor	0($key),$s0			# xor with key
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+
+	mov	240($key),$rnds			# load key->rounds
+	sub	\$1,$rnds
+	jmp	.Lenc_loop
+.align	16
+.Lenc_loop:
+___
+	if ($verticalspin) { &encvert(); }
+	else {	&encstep(0,$s0,$s1,$s2,$s3);
+		&encstep(1,$s1,$s2,$s3,$s0);
+		&encstep(2,$s2,$s3,$s0,$s1);
+		&encstep(3,$s3,$s0,$s1,$s2);
+	}
+$code.=<<___;
+	sub	\$1,$rnds
+	jnz	.Lenc_loop
+___
+	if ($verticalspin) { &enclastvert(); }
+	else {	&enclast(0,$s0,$s1,$s2,$s3);
+		&enclast(1,$s1,$s2,$s3,$s0);
+		&enclast(2,$s2,$s3,$s0,$s1);
+		&enclast(3,$s3,$s0,$s1,$s2);
+		$code.=<<___;
+		xor	16+0($key),$s0		# xor with key
+		xor	16+4($key),$s1
+		xor	16+8($key),$s2
+		xor	16+12($key),$s3
+___
+	}
+$code.=<<___;
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_encrypt,.-_x86_64_AES_encrypt
+___
+
+# it's possible to implement this by shifting tN by 8, filling least
+# significant byte with byte load and finally bswap-ing at the end,
+# but such partial register load kills Core 2...
+sub enccompactvert()
+{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$t0
+	movzb	`&lo("$s1")`,$t1
+	movzb	`&lo("$s2")`,$t2
+	movzb	`&lo("$s3")`,$t3
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	shr	\$16,$s2
+	movzb	`&hi("$s3")`,$acc2
+	movzb	($sbox,$t0,1),$t0
+	movzb	($sbox,$t1,1),$t1
+	movzb	($sbox,$t2,1),$t2
+	movzb	($sbox,$t3,1),$t3
+
+	movzb	($sbox,$acc0,1),$t4	#$t0
+	movzb	`&hi("$s0")`,$acc0
+	movzb	($sbox,$acc1,1),$t5	#$t1
+	movzb	`&lo("$s2")`,$acc1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+
+	shl	\$8,$t4
+	shr	\$16,$s3
+	shl	\$8,$t5
+	xor	$t4,$t0
+	shr	\$16,$s0
+	movzb	`&lo("$s3")`,$t4
+	shr	\$16,$s1
+	xor	$t5,$t1
+	shl	\$8,$acc2
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	xor	$acc2,$t2
+
+	shl	\$8,$acc0
+	movzb	`&lo("$s1")`,$acc2
+	shl	\$16,$acc1
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s3")`,$acc0
+	movzb	($sbox,$t5,1),$t5	#$t2
+	xor	$acc1,$t0
+
+	shr	\$8,$s2
+	movzb	`&hi("$s0")`,$acc1
+	shl	\$16,$t4
+	shr	\$8,$s1
+	shl	\$16,$t5
+	xor	$t4,$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t3
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$acc1	#$t1
+	movzb	($sbox,$s2,1),$s3	#$t3
+	movzb	($sbox,$s1,1),$s2	#$t2
+
+	shl	\$16,$acc2
+	xor	$t5,$t2
+	shl	\$24,$acc0
+	xor	$acc2,$t3
+	shl	\$24,$acc1
+	xor	$acc0,$t0
+	shl	\$24,$s3
+	xor	$acc1,$t1
+	shl	\$24,$s2
+	mov	$t0,$s0
+	mov	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub enctransform_ref()
+{ my $sn = shift;
+  my ($acc,$r2,$tmp)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	mov	$sn,$acc
+	and	\$0x80808080,$acc
+	mov	$acc,$tmp
+	shr	\$7,$tmp
+	lea	($sn,$sn),$r2
+	sub	$tmp,$acc
+	and	\$0xfefefefe,$r2
+	and	\$0x1b1b1b1b,$acc
+	mov	$sn,$tmp
+	xor	$acc,$r2
+
+	xor	$r2,$sn
+	rol	\$24,$sn
+	xor	$r2,$sn
+	ror	\$16,$tmp
+	xor	$tmp,$sn
+	ror	\$8,$tmp
+	xor	$tmp,$sn
+___
+}
+
+# unlike decrypt case it does not pay off to parallelize enctransform
+sub enctransform()
+{ my ($t3,$r20,$r21)=($acc2,"%r8d","%r9d");
+
+$code.=<<___;
+	mov	\$0x80808080,$t0
+	mov	\$0x80808080,$t1
+	and	$s0,$t0
+	and	$s1,$t1
+	mov	$t0,$acc0
+	mov	$t1,$acc1
+	shr	\$7,$t0
+	lea	($s0,$s0),$r20
+	shr	\$7,$t1
+	lea	($s1,$s1),$r21
+	sub	$t0,$acc0
+	sub	$t1,$acc1
+	and	\$0xfefefefe,$r20
+	and	\$0xfefefefe,$r21
+	and	\$0x1b1b1b1b,$acc0
+	and	\$0x1b1b1b1b,$acc1
+	mov	$s0,$t0
+	mov	$s1,$t1
+	xor	$acc0,$r20
+	xor	$acc1,$r21
+
+	xor	$r20,$s0
+	xor	$r21,$s1
+	 mov	\$0x80808080,$t2
+	rol	\$24,$s0
+	 mov	\$0x80808080,$t3
+	rol	\$24,$s1
+	 and	$s2,$t2
+	 and	$s3,$t3
+	xor	$r20,$s0
+	xor	$r21,$s1
+	 mov	$t2,$acc0
+	ror	\$16,$t0
+	 mov	$t3,$acc1
+	ror	\$16,$t1
+	 lea	($s2,$s2),$r20
+	 shr	\$7,$t2
+	xor	$t0,$s0
+	 shr	\$7,$t3
+	xor	$t1,$s1
+	ror	\$8,$t0
+	 lea	($s3,$s3),$r21
+	ror	\$8,$t1
+	 sub	$t2,$acc0
+	 sub	$t3,$acc1
+	xor	$t0,$s0
+	xor	$t1,$s1
+
+	and	\$0xfefefefe,$r20
+	and	\$0xfefefefe,$r21
+	and	\$0x1b1b1b1b,$acc0
+	and	\$0x1b1b1b1b,$acc1
+	mov	$s2,$t2
+	mov	$s3,$t3
+	xor	$acc0,$r20
+	xor	$acc1,$r21
+
+	ror	\$16,$t2
+	xor	$r20,$s2
+	ror	\$16,$t3
+	xor	$r21,$s3
+	rol	\$24,$s2
+	mov	0($sbox),$acc0			# prefetch Te4
+	rol	\$24,$s3
+	xor	$r20,$s2
+	mov	64($sbox),$acc1
+	xor	$r21,$s3
+	mov	128($sbox),$r20
+	xor	$t2,$s2
+	ror	\$8,$t2
+	xor	$t3,$s3
+	ror	\$8,$t3
+	xor	$t2,$s2
+	mov	192($sbox),$r21
+	xor	$t3,$s3
+___
+}
+
+$code.=<<___;
+.type	_x86_64_AES_encrypt_compact,\@abi-omnipotent
+.align	16
+_x86_64_AES_encrypt_compact:
+.cfi_startproc
+	lea	128($sbox),$inp			# size optimization
+	mov	0-128($inp),$acc1		# prefetch Te4
+	mov	32-128($inp),$acc2
+	mov	64-128($inp),$t0
+	mov	96-128($inp),$t1
+	mov	128-128($inp),$acc1
+	mov	160-128($inp),$acc2
+	mov	192-128($inp),$t0
+	mov	224-128($inp),$t1
+	jmp	.Lenc_loop_compact
+.align	16
+.Lenc_loop_compact:
+		xor	0($key),$s0		# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+		lea	16($key),$key
+___
+		&enccompactvert();
+$code.=<<___;
+		cmp	16(%rsp),$key
+		je	.Lenc_compact_done
+___
+		&enctransform();
+$code.=<<___;
+	jmp	.Lenc_loop_compact
+.align	16
+.Lenc_compact_done:
+	xor	0($key),$s0
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
+.size	_x86_64_AES_encrypt_compact,.-_x86_64_AES_encrypt_compact
+___
+
+# void aes_nohw_encrypt (const void *inp,void *out,const AES_KEY *key);
+$code.=<<___;
+.align	16
+.globl	aes_nohw_encrypt
+.type	aes_nohw_encrypt,\@function,3
+.hidden	aes_nohw_encrypt
+aes_nohw_encrypt:
+.cfi_startproc
+	mov	%rsp,%rax
+.cfi_def_cfa_register	%rax
+	push	%rbx
+.cfi_push	%rbx
+	push	%rbp
+.cfi_push	%rbp
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+
+	# allocate frame "above" key schedule
+	lea	-63(%rdx),%rcx	# %rdx is key argument
+	and	\$-64,%rsp
+	sub	%rsp,%rcx
+	neg	%rcx
+	and	\$0x3c0,%rcx
+	sub	%rcx,%rsp
+	sub	\$32,%rsp
+
+	mov	%rsi,16(%rsp)	# save out
+	mov	%rax,24(%rsp)	# save original stack pointer
+.cfi_cfa_expression	%rsp+24,deref,+8
+.Lenc_prologue:
+
+	mov	%rdx,$key
+	mov	240($key),$rnds	# load rounds
+
+	mov	0(%rdi),$s0	# load input vector
+	mov	4(%rdi),$s1
+	mov	8(%rdi),$s2
+	mov	12(%rdi),$s3
+
+	shl	\$4,$rnds
+	lea	($key,$rnds),%rbp
+	mov	$key,(%rsp)	# key schedule
+	mov	%rbp,8(%rsp)	# end of key schedule
+
+	# pick Te4 copy which can't "overlap" with stack frame or key schedule
+	lea	.LAES_Te+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
+	sub	$sbox,%rbp
+	and	\$0x300,%rbp
+	lea	($sbox,%rbp),$sbox
+
+	call	_x86_64_AES_encrypt_compact
+
+	mov	16(%rsp),$out	# restore out
+	mov	24(%rsp),%rsi	# restore saved stack pointer
+.cfi_def_cfa	%rsi,8
+	mov	$s0,0($out)	# write output vector
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	mov	-48(%rsi),%r15
+.cfi_restore	%r15
+	mov	-40(%rsi),%r14
+.cfi_restore	%r14
+	mov	-32(%rsi),%r13
+.cfi_restore	%r13
+	mov	-24(%rsi),%r12
+.cfi_restore	%r12
+	mov	-16(%rsi),%rbp
+.cfi_restore	%rbp
+	mov	-8(%rsi),%rbx
+.cfi_restore	%rbx
+	lea	(%rsi),%rsp
+.cfi_def_cfa_register	%rsp
+.Lenc_epilogue:
+	ret
+.cfi_endproc
+.size	aes_nohw_encrypt,.-aes_nohw_encrypt
+___
+
+#------------------------------------------------------------------#
+
+sub decvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	# favor 3-way issue Opteron pipeline...
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	mov	0($sbox,$acc0,8),$t0
+	mov	0($sbox,$acc1,8),$t1
+	mov	0($sbox,$acc2,8),$t2
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	movzb	`&lo("$s3")`,$acc2
+	xor	3($sbox,$acc0,8),$t0
+	xor	3($sbox,$acc1,8),$t1
+	mov	0($sbox,$acc2,8),$t3
+
+	movzb	`&hi("$s1")`,$acc0
+	shr	\$16,$s0
+	movzb	`&hi("$s2")`,$acc2
+	xor	3($sbox,$acc0,8),$t2
+	shr	\$16,$s3
+	xor	3($sbox,$acc2,8),$t3
+
+	shr	\$16,$s1
+	lea	16($key),$key
+	shr	\$16,$s2
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	xor	2($sbox,$acc0,8),$t0
+	xor	2($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t2
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	movzb	`&lo("$s1")`,$acc2
+	xor	1($sbox,$acc0,8),$t0
+	xor	1($sbox,$acc1,8),$t1
+	xor	2($sbox,$acc2,8),$t3
+
+	movzb	`&hi("$s3")`,$acc0
+	mov	12($key),$s3
+	movzb	`&hi("$s0")`,$acc2
+	xor	1($sbox,$acc0,8),$t2
+	mov	0($key),$s0
+	xor	1($sbox,$acc2,8),$t3
+
+	xor	$t0,$s0
+	mov	4($key),$s1
+	mov	8($key),$s2
+	xor	$t2,$s2
+	xor	$t1,$s1
+	xor	$t3,$s3
+___
+}
+
+sub declastvert()
+{ my $t3="%r8d";	# zaps $inp!
+
+$code.=<<___;
+	lea	2048($sbox),$sbox	# size optimization
+	movzb	`&lo("$s0")`,$acc0
+	movzb	`&lo("$s1")`,$acc1
+	movzb	`&lo("$s2")`,$acc2
+	movzb	($sbox,$acc0,1),$t0
+	movzb	($sbox,$acc1,1),$t1
+	movzb	($sbox,$acc2,1),$t2
+
+	movzb	`&lo("$s3")`,$acc0
+	movzb	`&hi("$s3")`,$acc1
+	movzb	`&hi("$s0")`,$acc2
+	movzb	($sbox,$acc0,1),$t3
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	movzb	($sbox,$acc2,1),$acc2	#$t1
+
+	shl	\$8,$acc1
+	shl	\$8,$acc2
+
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+	shr	\$16,$s3
+
+	movzb	`&hi("$s1")`,$acc0
+	movzb	`&hi("$s2")`,$acc1
+	shr	\$16,$s0
+	movzb	($sbox,$acc0,1),$acc0	#$t2
+	movzb	($sbox,$acc1,1),$acc1	#$t3
+
+	shl	\$8,$acc0
+	shl	\$8,$acc1
+	shr	\$16,$s1
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+	shr	\$16,$s2
+
+	movzb	`&lo("$s2")`,$acc0
+	movzb	`&lo("$s3")`,$acc1
+	movzb	`&lo("$s0")`,$acc2
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$acc1	#$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+
+	shl	\$16,$acc0
+	shl	\$16,$acc1
+	shl	\$16,$acc2
+
+	xor	$acc0,$t0
+	xor	$acc1,$t1
+	xor	$acc2,$t2
+
+	movzb	`&lo("$s1")`,$acc0
+	movzb	`&hi("$s1")`,$acc1
+	movzb	`&hi("$s2")`,$acc2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	movzb	($sbox,$acc2,1),$acc2	#$t1
+
+	shl	\$16,$acc0
+	shl	\$24,$acc1
+	shl	\$24,$acc2
+
+	xor	$acc0,$t3
+	xor	$acc1,$t0
+	xor	$acc2,$t1
+
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	mov	16+12($key),$s3
+	movzb	($sbox,$acc0,1),$acc0	#$t2
+	movzb	($sbox,$acc1,1),$acc1	#$t3
+	mov	16+0($key),$s0
+
+	shl	\$24,$acc0
+	shl	\$24,$acc1
+
+	xor	$acc0,$t2
+	xor	$acc1,$t3
+
+	mov	16+4($key),$s1
+	mov	16+8($key),$s2
+	lea	-2048($sbox),$sbox
+	xor	$t0,$s0
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+sub decstep()
+{ my ($i,@s) = @_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	$code.="	mov	$s[0],$out\n"		if ($i!=3);
+			$tmp1=$s[2]			if ($i==3);
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	and	\$0xFF,$out\n";
+
+	$code.="	mov	0($sbox,$out,8),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+			$tmp2=$s[3]			if ($i==3);
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+			$tmp0=$s[1]			if ($i==3);
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	and	\$0xFF,$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	xor	3($sbox,$tmp0,8),$out\n";
+	$code.="	xor	2($sbox,$tmp1,8),$out\n";
+	$code.="	xor	1($sbox,$tmp2,8),$out\n";
+
+	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+sub declast()
+{ my ($i,@s)=@_;
+  my $tmp0=$acc0;
+  my $tmp1=$acc1;
+  my $tmp2=$acc2;
+  my $out=($t0,$t1,$t2,$s[0])[$i];
+
+	$code.="	mov	$s[0],$out\n"		if ($i!=3);
+			$tmp1=$s[2]			if ($i==3);
+	$code.="	mov	$s[2],$tmp1\n"		if ($i!=3);
+	$code.="	and	\$0xFF,$out\n";
+
+	$code.="	movzb	2048($sbox,$out,1),$out\n";
+	$code.="	shr	\$16,$tmp1\n";
+			$tmp2=$s[3]			if ($i==3);
+	$code.="	mov	$s[3],$tmp2\n"		if ($i!=3);
+
+			$tmp0=$s[1]			if ($i==3);
+	$code.="	movzb	".&hi($s[1]).",$tmp0\n";
+	$code.="	and	\$0xFF,$tmp1\n";
+	$code.="	shr	\$24,$tmp2\n";
+
+	$code.="	movzb	2048($sbox,$tmp0,1),$tmp0\n";
+	$code.="	movzb	2048($sbox,$tmp1,1),$tmp1\n";
+	$code.="	movzb	2048($sbox,$tmp2,1),$tmp2\n";
+
+	$code.="	shl	\$8,$tmp0\n";
+	$code.="	shl	\$16,$tmp1\n";
+	$code.="	shl	\$24,$tmp2\n";
+
+	$code.="	xor	$tmp0,$out\n";
+	$code.="	mov	$t2,$s[1]\n"		if ($i==3);
+	$code.="	xor	$tmp1,$out\n";
+	$code.="	mov	$t1,$s[2]\n"		if ($i==3);
+	$code.="	xor	$tmp2,$out\n";
+	$code.="	mov	$t0,$s[3]\n"		if ($i==3);
+	$code.="\n";
+}
+
+$code.=<<___;
+.type	_x86_64_AES_decrypt,\@abi-omnipotent
+.align	16
+_x86_64_AES_decrypt:
+	xor	0($key),$s0			# xor with key
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+
+	mov	240($key),$rnds			# load key->rounds
+	sub	\$1,$rnds
+	jmp	.Ldec_loop
+.align	16
+.Ldec_loop:
+___
+	if ($verticalspin) { &decvert(); }
+	else {	&decstep(0,$s0,$s3,$s2,$s1);
+		&decstep(1,$s1,$s0,$s3,$s2);
+		&decstep(2,$s2,$s1,$s0,$s3);
+		&decstep(3,$s3,$s2,$s1,$s0);
+		$code.=<<___;
+		lea	16($key),$key
+		xor	0($key),$s0			# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+___
+	}
+$code.=<<___;
+	sub	\$1,$rnds
+	jnz	.Ldec_loop
+___
+	if ($verticalspin) { &declastvert(); }
+	else {	&declast(0,$s0,$s3,$s2,$s1);
+		&declast(1,$s1,$s0,$s3,$s2);
+		&declast(2,$s2,$s1,$s0,$s3);
+		&declast(3,$s3,$s2,$s1,$s0);
+		$code.=<<___;
+		xor	16+0($key),$s0			# xor with key
+		xor	16+4($key),$s1
+		xor	16+8($key),$s2
+		xor	16+12($key),$s3
+___
+	}
+$code.=<<___;
+	.byte	0xf3,0xc3			# rep ret
+.size	_x86_64_AES_decrypt,.-_x86_64_AES_decrypt
+___
+
+sub deccompactvert()
+{ my ($t3,$t4,$t5)=("%r8d","%r9d","%r13d");
+
+$code.=<<___;
+	movzb	`&lo("$s0")`,$t0
+	movzb	`&lo("$s1")`,$t1
+	movzb	`&lo("$s2")`,$t2
+	movzb	`&lo("$s3")`,$t3
+	movzb	`&hi("$s3")`,$acc0
+	movzb	`&hi("$s0")`,$acc1
+	shr	\$16,$s3
+	movzb	`&hi("$s1")`,$acc2
+	movzb	($sbox,$t0,1),$t0
+	movzb	($sbox,$t1,1),$t1
+	movzb	($sbox,$t2,1),$t2
+	movzb	($sbox,$t3,1),$t3
+
+	movzb	($sbox,$acc0,1),$t4	#$t0
+	movzb	`&hi("$s2")`,$acc0
+	movzb	($sbox,$acc1,1),$t5	#$t1
+	movzb	($sbox,$acc2,1),$acc2	#$t2
+	movzb	($sbox,$acc0,1),$acc0	#$t3
+
+	shr	\$16,$s2
+	shl	\$8,$t5
+	shl	\$8,$t4
+	movzb	`&lo("$s2")`,$acc1
+	shr	\$16,$s0
+	xor	$t4,$t0
+	shr	\$16,$s1
+	movzb	`&lo("$s3")`,$t4
+
+	shl	\$8,$acc2
+	xor	$t5,$t1
+	shl	\$8,$acc0
+	movzb	`&lo("$s0")`,$t5
+	movzb	($sbox,$acc1,1),$acc1	#$t0
+	xor	$acc2,$t2
+	movzb	`&lo("$s1")`,$acc2
+
+	shl	\$16,$acc1
+	xor	$acc0,$t3
+	movzb	($sbox,$t4,1),$t4	#$t1
+	movzb	`&hi("$s1")`,$acc0
+	movzb	($sbox,$acc2,1),$acc2	#$t3
+	xor	$acc1,$t0
+	movzb	($sbox,$t5,1),$t5	#$t2
+	movzb	`&hi("$s2")`,$acc1
+
+	shl	\$16,$acc2
+	shl	\$16,$t4
+	shl	\$16,$t5
+	xor	$acc2,$t3
+	movzb	`&hi("$s3")`,$acc2
+	xor	$t4,$t1
+	shr	\$8,$s0
+	xor	$t5,$t2
+
+	movzb	($sbox,$acc0,1),$acc0	#$t0
+	movzb	($sbox,$acc1,1),$s1	#$t1
+	movzb	($sbox,$acc2,1),$s2	#$t2
+	movzb	($sbox,$s0,1),$s3	#$t3
+
+	mov	$t0,$s0
+	shl	\$24,$acc0
+	shl	\$24,$s1
+	shl	\$24,$s2
+	xor	$acc0,$s0
+	shl	\$24,$s3
+	xor	$t1,$s1
+	xor	$t2,$s2
+	xor	$t3,$s3
+___
+}
+
+# parallelized version! input is pair of 64-bit values: %rax=s1.s0
+# and %rcx=s3.s2, output is four 32-bit values in %eax=s0, %ebx=s1,
+# %ecx=s2 and %edx=s3.
+sub dectransform()
+{ my ($tp10,$tp20,$tp40,$tp80,$acc0)=("%rax","%r8", "%r9", "%r10","%rbx");
+  my ($tp18,$tp28,$tp48,$tp88,$acc8)=("%rcx","%r11","%r12","%r13","%rdx");
+  my $prefetch = shift;
+
+$code.=<<___;
+	mov	$mask80,$tp40
+	mov	$mask80,$tp48
+	and	$tp10,$tp40
+	and	$tp18,$tp48
+	mov	$tp40,$acc0
+	mov	$tp48,$acc8
+	shr	\$7,$tp40
+	lea	($tp10,$tp10),$tp20
+	shr	\$7,$tp48
+	lea	($tp18,$tp18),$tp28
+	sub	$tp40,$acc0
+	sub	$tp48,$acc8
+	and	$maskfe,$tp20
+	and	$maskfe,$tp28
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp20
+	xor	$acc8,$tp28
+	mov	$mask80,$tp80
+	mov	$mask80,$tp88
+
+	and	$tp20,$tp80
+	and	$tp28,$tp88
+	mov	$tp80,$acc0
+	mov	$tp88,$acc8
+	shr	\$7,$tp80
+	lea	($tp20,$tp20),$tp40
+	shr	\$7,$tp88
+	lea	($tp28,$tp28),$tp48
+	sub	$tp80,$acc0
+	sub	$tp88,$acc8
+	and	$maskfe,$tp40
+	and	$maskfe,$tp48
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp40
+	xor	$acc8,$tp48
+	mov	$mask80,$tp80
+	mov	$mask80,$tp88
+
+	and	$tp40,$tp80
+	and	$tp48,$tp88
+	mov	$tp80,$acc0
+	mov	$tp88,$acc8
+	shr	\$7,$tp80
+	 xor	$tp10,$tp20		# tp2^=tp1
+	shr	\$7,$tp88
+	 xor	$tp18,$tp28		# tp2^=tp1
+	sub	$tp80,$acc0
+	sub	$tp88,$acc8
+	lea	($tp40,$tp40),$tp80
+	lea	($tp48,$tp48),$tp88
+	 xor	$tp10,$tp40		# tp4^=tp1
+	 xor	$tp18,$tp48		# tp4^=tp1
+	and	$maskfe,$tp80
+	and	$maskfe,$tp88
+	and	$mask1b,$acc0
+	and	$mask1b,$acc8
+	xor	$acc0,$tp80
+	xor	$acc8,$tp88
+
+	xor	$tp80,$tp10		# tp1^=tp8
+	xor	$tp88,$tp18		# tp1^=tp8
+	xor	$tp80,$tp20		# tp2^tp1^=tp8
+	xor	$tp88,$tp28		# tp2^tp1^=tp8
+	mov	$tp10,$acc0
+	mov	$tp18,$acc8
+	xor	$tp80,$tp40		# tp4^tp1^=tp8
+	shr	\$32,$acc0
+	xor	$tp88,$tp48		# tp4^tp1^=tp8
+	shr	\$32,$acc8
+	xor	$tp20,$tp80		# tp8^=tp8^tp2^tp1=tp2^tp1
+	rol	\$8,`&LO("$tp10")`	# ROTATE(tp1^tp8,8)
+	xor	$tp28,$tp88		# tp8^=tp8^tp2^tp1=tp2^tp1
+	rol	\$8,`&LO("$tp18")`	# ROTATE(tp1^tp8,8)
+	xor	$tp40,$tp80		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
+	rol	\$8,`&LO("$acc0")`	# ROTATE(tp1^tp8,8)
+	xor	$tp48,$tp88		# tp2^tp1^=tp8^tp4^tp1=tp8^tp4^tp2
+
+	rol	\$8,`&LO("$acc8")`	# ROTATE(tp1^tp8,8)
+	xor	`&LO("$tp80")`,`&LO("$tp10")`
+	shr	\$32,$tp80
+	xor	`&LO("$tp88")`,`&LO("$tp18")`
+	shr	\$32,$tp88
+	xor	`&LO("$tp80")`,`&LO("$acc0")`
+	xor	`&LO("$tp88")`,`&LO("$acc8")`
+
+	mov	$tp20,$tp80
+	rol	\$24,`&LO("$tp20")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp28,$tp88
+	rol	\$24,`&LO("$tp28")`	# ROTATE(tp2^tp1^tp8,24)
+	shr	\$32,$tp80
+	xor	`&LO("$tp20")`,`&LO("$tp10")`
+	shr	\$32,$tp88
+	xor	`&LO("$tp28")`,`&LO("$tp18")`
+	rol	\$24,`&LO("$tp80")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp40,$tp20
+	rol	\$24,`&LO("$tp88")`	# ROTATE(tp2^tp1^tp8,24)
+	mov	$tp48,$tp28
+	shr	\$32,$tp20
+	xor	`&LO("$tp80")`,`&LO("$acc0")`
+	shr	\$32,$tp28
+	xor	`&LO("$tp88")`,`&LO("$acc8")`
+
+	`"mov	0($sbox),$mask80"	if ($prefetch)`
+	rol	\$16,`&LO("$tp40")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	64($sbox),$maskfe"	if ($prefetch)`
+	rol	\$16,`&LO("$tp48")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	128($sbox),$mask1b"	if ($prefetch)`
+	rol	\$16,`&LO("$tp20")`	# ROTATE(tp4^tp1^tp8,16)
+	`"mov	192($sbox),$tp80"	if ($prefetch)`
+	xor	`&LO("$tp40")`,`&LO("$tp10")`
+	rol	\$16,`&LO("$tp28")`	# ROTATE(tp4^tp1^tp8,16)
+	xor	`&LO("$tp48")`,`&LO("$tp18")`
+	`"mov	256($sbox),$tp88"	if ($prefetch)`
+	xor	`&LO("$tp20")`,`&LO("$acc0")`
+	xor	`&LO("$tp28")`,`&LO("$acc8")`
+___
+}
+
+$code.=<<___;
+.type	_x86_64_AES_decrypt_compact,\@abi-omnipotent
+.align	16
+_x86_64_AES_decrypt_compact:
+.cfi_startproc
+	lea	128($sbox),$inp			# size optimization
+	mov	0-128($inp),$acc1		# prefetch Td4
+	mov	32-128($inp),$acc2
+	mov	64-128($inp),$t0
+	mov	96-128($inp),$t1
+	mov	128-128($inp),$acc1
+	mov	160-128($inp),$acc2
+	mov	192-128($inp),$t0
+	mov	224-128($inp),$t1
+	jmp	.Ldec_loop_compact
+
+.align	16
+.Ldec_loop_compact:
+		xor	0($key),$s0		# xor with key
+		xor	4($key),$s1
+		xor	8($key),$s2
+		xor	12($key),$s3
+		lea	16($key),$key
+___
+		&deccompactvert();
+$code.=<<___;
+		cmp	16(%rsp),$key
+		je	.Ldec_compact_done
+
+		mov	256+0($sbox),$mask80
+		shl	\$32,%rbx
+		shl	\$32,%rdx
+		mov	256+8($sbox),$maskfe
+		or	%rbx,%rax
+		or	%rdx,%rcx
+		mov	256+16($sbox),$mask1b
+___
+		&dectransform(1);
+$code.=<<___;
+	jmp	.Ldec_loop_compact
+.align	16
+.Ldec_compact_done:
+	xor	0($key),$s0
+	xor	4($key),$s1
+	xor	8($key),$s2
+	xor	12($key),$s3
+	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
+.size	_x86_64_AES_decrypt_compact,.-_x86_64_AES_decrypt_compact
+___
+
+# void aes_nohw_decrypt (const void *inp,void *out,const AES_KEY *key);
+$code.=<<___;
+.align	16
+.globl	aes_nohw_decrypt
+.type	aes_nohw_decrypt,\@function,3
+.hidden	aes_nohw_decrypt
+aes_nohw_decrypt:
+.cfi_startproc
+	mov	%rsp,%rax
+.cfi_def_cfa_register	%rax
+	push	%rbx
+.cfi_push	%rbx
+	push	%rbp
+.cfi_push	%rbp
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+
+	# allocate frame "above" key schedule
+	lea	-63(%rdx),%rcx	# %rdx is key argument
+	and	\$-64,%rsp
+	sub	%rsp,%rcx
+	neg	%rcx
+	and	\$0x3c0,%rcx
+	sub	%rcx,%rsp
+	sub	\$32,%rsp
+
+	mov	%rsi,16(%rsp)	# save out
+	mov	%rax,24(%rsp)	# save original stack pointer
+.cfi_cfa_expression	%rsp+24,deref,+8
+.Ldec_prologue:
+
+	mov	%rdx,$key
+	mov	240($key),$rnds	# load rounds
+
+	mov	0(%rdi),$s0	# load input vector
+	mov	4(%rdi),$s1
+	mov	8(%rdi),$s2
+	mov	12(%rdi),$s3
+
+	shl	\$4,$rnds
+	lea	($key,$rnds),%rbp
+	mov	$key,(%rsp)	# key schedule
+	mov	%rbp,8(%rsp)	# end of key schedule
+
+	# pick Td4 copy which can't "overlap" with stack frame or key schedule
+	lea	.LAES_Td+2048(%rip),$sbox
+	lea	768(%rsp),%rbp
+	sub	$sbox,%rbp
+	and	\$0x300,%rbp
+	lea	($sbox,%rbp),$sbox
+	shr	\$3,%rbp	# recall "magic" constants!
+	add	%rbp,$sbox
+
+	call	_x86_64_AES_decrypt_compact
+
+	mov	16(%rsp),$out	# restore out
+	mov	24(%rsp),%rsi	# restore saved stack pointer
+.cfi_def_cfa	%rsi,8
+	mov	$s0,0($out)	# write output vector
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	mov	-48(%rsi),%r15
+.cfi_restore	%r15
+	mov	-40(%rsi),%r14
+.cfi_restore	%r14
+	mov	-32(%rsi),%r13
+.cfi_restore	%r13
+	mov	-24(%rsi),%r12
+.cfi_restore	%r12
+	mov	-16(%rsi),%rbp
+.cfi_restore	%rbp
+	mov	-8(%rsi),%rbx
+.cfi_restore	%rbx
+	lea	(%rsi),%rsp
+.cfi_def_cfa_register	%rsp
+.Ldec_epilogue:
+	ret
+.cfi_endproc
+.size	aes_nohw_decrypt,.-aes_nohw_decrypt
+___
+#------------------------------------------------------------------#
+
+sub enckey()
+{
+$code.=<<___;
+	movz	%dl,%esi		# rk[i]>>0
+	movzb	-128(%rbp,%rsi),%ebx
+	movz	%dh,%esi		# rk[i]>>8
+	shl	\$24,%ebx
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	shr	\$16,%edx
+	movz	%dl,%esi		# rk[i]>>16
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	movz	%dh,%esi		# rk[i]>>24
+	shl	\$8,%ebx
+	xor	%ebx,%eax
+
+	movzb	-128(%rbp,%rsi),%ebx
+	shl	\$16,%ebx
+	xor	%ebx,%eax
+
+	xor	1024-128(%rbp,%rcx,4),%eax		# rcon
+___
+}
+
+# int aes_nohw_set_encrypt_key(const unsigned char *userKey, const int bits, AES_KEY *key)
+$code.=<<___;
+.align	16
+.globl aes_nohw_set_encrypt_key
+.type  aes_nohw_set_encrypt_key,\@function,3
+aes_nohw_set_encrypt_key:
+.cfi_startproc
+	push	%rbx
+.cfi_push	%rbx
+	push	%rbp
+.cfi_push	%rbp
+	push	%r12			# redundant, but allows to share
+.cfi_push	%r12
+	push	%r13			# exception handler...
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+	sub	\$8,%rsp
+.cfi_adjust_cfa_offset	8
+.Lenc_key_prologue:
+
+	call	_x86_64_AES_set_encrypt_key
+
+	mov	40(%rsp),%rbp
+.cfi_restore	%rbp
+	mov	48(%rsp),%rbx
+.cfi_restore	%rbx
+	add	\$56,%rsp
+.cfi_adjust_cfa_offset	-56
+.Lenc_key_epilogue:
+	ret
+.cfi_endproc
+.size aes_nohw_set_encrypt_key,.-aes_nohw_set_encrypt_key
+
+.type	_x86_64_AES_set_encrypt_key,\@abi-omnipotent
+.align	16
+_x86_64_AES_set_encrypt_key:
+.cfi_startproc
+	mov	%esi,%ecx			# %ecx=bits
+	mov	%rdi,%rsi			# %rsi=userKey
+	mov	%rdx,%rdi			# %rdi=key
+
+	test	\$-1,%rsi
+	jz	.Lbadpointer
+	test	\$-1,%rdi
+	jz	.Lbadpointer
+
+	lea	.LAES_Te(%rip),%rbp
+	lea	2048+128(%rbp),%rbp
+
+	# prefetch Te4
+	mov	0-128(%rbp),%eax
+	mov	32-128(%rbp),%ebx
+	mov	64-128(%rbp),%r8d
+	mov	96-128(%rbp),%edx
+	mov	128-128(%rbp),%eax
+	mov	160-128(%rbp),%ebx
+	mov	192-128(%rbp),%r8d
+	mov	224-128(%rbp),%edx
+
+	cmp	\$128,%ecx
+	je	.L10rounds
+	cmp	\$192,%ecx
+	je	.L12rounds
+	cmp	\$256,%ecx
+	je	.L14rounds
+	mov	\$-2,%rax			# invalid number of bits
+	jmp	.Lexit
+
+.L10rounds:
+	mov	0(%rsi),%rax			# copy first 4 dwords
+	mov	8(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rdx,8(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L10shortcut
+.align	4
+.L10loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	12(%rdi),%edx			# rk[3]
+.L10shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,16(%rdi)			# rk[4]
+		xor	4(%rdi),%eax
+		mov	%eax,20(%rdi)			# rk[5]
+		xor	8(%rdi),%eax
+		mov	%eax,24(%rdi)			# rk[6]
+		xor	12(%rdi),%eax
+		mov	%eax,28(%rdi)			# rk[7]
+		add	\$1,%ecx
+		lea	16(%rdi),%rdi
+		cmp	\$10,%ecx
+	jl	.L10loop
+
+	movl	\$10,80(%rdi)			# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.L12rounds:
+	mov	0(%rsi),%rax			# copy first 6 dwords
+	mov	8(%rsi),%rbx
+	mov	16(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rbx,8(%rdi)
+	mov	%rdx,16(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L12shortcut
+.align	4
+.L12loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	20(%rdi),%edx			# rk[5]
+.L12shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,24(%rdi)			# rk[6]
+		xor	4(%rdi),%eax
+		mov	%eax,28(%rdi)			# rk[7]
+		xor	8(%rdi),%eax
+		mov	%eax,32(%rdi)			# rk[8]
+		xor	12(%rdi),%eax
+		mov	%eax,36(%rdi)			# rk[9]
+
+		cmp	\$7,%ecx
+		je	.L12break
+		add	\$1,%ecx
+
+		xor	16(%rdi),%eax
+		mov	%eax,40(%rdi)			# rk[10]
+		xor	20(%rdi),%eax
+		mov	%eax,44(%rdi)			# rk[11]
+
+		lea	24(%rdi),%rdi
+	jmp	.L12loop
+.L12break:
+	movl	\$12,72(%rdi)		# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.L14rounds:
+	mov	0(%rsi),%rax			# copy first 8 dwords
+	mov	8(%rsi),%rbx
+	mov	16(%rsi),%rcx
+	mov	24(%rsi),%rdx
+	mov	%rax,0(%rdi)
+	mov	%rbx,8(%rdi)
+	mov	%rcx,16(%rdi)
+	mov	%rdx,24(%rdi)
+
+	shr	\$32,%rdx
+	xor	%ecx,%ecx
+	jmp	.L14shortcut
+.align	4
+.L14loop:
+		mov	0(%rdi),%eax			# rk[0]
+		mov	28(%rdi),%edx			# rk[4]
+.L14shortcut:
+___
+		&enckey	();
+$code.=<<___;
+		mov	%eax,32(%rdi)			# rk[8]
+		xor	4(%rdi),%eax
+		mov	%eax,36(%rdi)			# rk[9]
+		xor	8(%rdi),%eax
+		mov	%eax,40(%rdi)			# rk[10]
+		xor	12(%rdi),%eax
+		mov	%eax,44(%rdi)			# rk[11]
+
+		cmp	\$6,%ecx
+		je	.L14break
+		add	\$1,%ecx
+
+		mov	%eax,%edx
+		mov	16(%rdi),%eax			# rk[4]
+		movz	%dl,%esi			# rk[11]>>0
+		movzb	-128(%rbp,%rsi),%ebx
+		movz	%dh,%esi			# rk[11]>>8
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		shr	\$16,%edx
+		shl	\$8,%ebx
+		movz	%dl,%esi			# rk[11]>>16
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		movz	%dh,%esi			# rk[11]>>24
+		shl	\$16,%ebx
+		xor	%ebx,%eax
+
+		movzb	-128(%rbp,%rsi),%ebx
+		shl	\$24,%ebx
+		xor	%ebx,%eax
+
+		mov	%eax,48(%rdi)			# rk[12]
+		xor	20(%rdi),%eax
+		mov	%eax,52(%rdi)			# rk[13]
+		xor	24(%rdi),%eax
+		mov	%eax,56(%rdi)			# rk[14]
+		xor	28(%rdi),%eax
+		mov	%eax,60(%rdi)			# rk[15]
+
+		lea	32(%rdi),%rdi
+	jmp	.L14loop
+.L14break:
+	movl	\$14,48(%rdi)		# setup number of rounds
+	xor	%rax,%rax
+	jmp	.Lexit
+
+.Lbadpointer:
+	mov	\$-1,%rax
+.Lexit:
+	.byte	0xf3,0xc3			# rep ret
+.cfi_endproc
+.size	_x86_64_AES_set_encrypt_key,.-_x86_64_AES_set_encrypt_key
+___
+
+sub deckey_ref()
+{ my ($i,$ptr,$te,$td) = @_;
+  my ($tp1,$tp2,$tp4,$tp8,$acc)=("%eax","%ebx","%edi","%edx","%r8d");
+$code.=<<___;
+	mov	$i($ptr),$tp1
+	mov	$tp1,$acc
+	and	\$0x80808080,$acc
+	mov	$acc,$tp4
+	shr	\$7,$tp4
+	lea	0($tp1,$tp1),$tp2
+	sub	$tp4,$acc
+	and	\$0xfefefefe,$tp2
+	and	\$0x1b1b1b1b,$acc
+	xor	$tp2,$acc
+	mov	$acc,$tp2
+
+	and	\$0x80808080,$acc
+	mov	$acc,$tp8
+	shr	\$7,$tp8
+	lea	0($tp2,$tp2),$tp4
+	sub	$tp8,$acc
+	and	\$0xfefefefe,$tp4
+	and	\$0x1b1b1b1b,$acc
+	 xor	$tp1,$tp2		# tp2^tp1
+	xor	$tp4,$acc
+	mov	$acc,$tp4
+
+	and	\$0x80808080,$acc
+	mov	$acc,$tp8
+	shr	\$7,$tp8
+	sub	$tp8,$acc
+	lea	0($tp4,$tp4),$tp8
+	 xor	$tp1,$tp4		# tp4^tp1
+	and	\$0xfefefefe,$tp8
+	and	\$0x1b1b1b1b,$acc
+	xor	$acc,$tp8
+
+	xor	$tp8,$tp1		# tp1^tp8
+	rol	\$8,$tp1		# ROTATE(tp1^tp8,8)
+	xor	$tp8,$tp2		# tp2^tp1^tp8
+	xor	$tp8,$tp4		# tp4^tp1^tp8
+	xor	$tp2,$tp8
+	xor	$tp4,$tp8		# tp8^(tp8^tp4^tp1)^(tp8^tp2^tp1)=tp8^tp4^tp2
+
+	xor	$tp8,$tp1
+	rol	\$24,$tp2		# ROTATE(tp2^tp1^tp8,24)
+	xor	$tp2,$tp1
+	rol	\$16,$tp4		# ROTATE(tp4^tp1^tp8,16)
+	xor	$tp4,$tp1
+
+	mov	$tp1,$i($ptr)
+___
+}
+
+# int aes_nohw_set_decrypt_key(const unsigned char *userKey, const int bits, AES_KEY *key)
+$code.=<<___;
+.align	16
+.globl aes_nohw_set_decrypt_key
+.type  aes_nohw_set_decrypt_key,\@function,3
+aes_nohw_set_decrypt_key:
+.cfi_startproc
+	push	%rbx
+.cfi_push	%rbx
+	push	%rbp
+.cfi_push	%rbp
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+	push	%rdx			# save key schedule
+.cfi_adjust_cfa_offset	8
+.Ldec_key_prologue:
+
+	call	_x86_64_AES_set_encrypt_key
+	mov	(%rsp),%r8		# restore key schedule
+	cmp	\$0,%eax
+	jne	.Labort
+
+	mov	240(%r8),%r14d		# pull number of rounds
+	xor	%rdi,%rdi
+	lea	(%rdi,%r14d,4),%rcx
+	mov	%r8,%rsi
+	lea	(%r8,%rcx,4),%rdi	# pointer to last chunk
+.align	4
+.Linvert:
+		mov	0(%rsi),%rax
+		mov	8(%rsi),%rbx
+		mov	0(%rdi),%rcx
+		mov	8(%rdi),%rdx
+		mov	%rax,0(%rdi)
+		mov	%rbx,8(%rdi)
+		mov	%rcx,0(%rsi)
+		mov	%rdx,8(%rsi)
+		lea	16(%rsi),%rsi
+		lea	-16(%rdi),%rdi
+		cmp	%rsi,%rdi
+	jne	.Linvert
+
+	lea	.LAES_Te+2048+1024(%rip),%rax	# rcon
+
+	mov	40(%rax),$mask80
+	mov	48(%rax),$maskfe
+	mov	56(%rax),$mask1b
+
+	mov	%r8,$key
+	sub	\$1,%r14d
+.align	4
+.Lpermute:
+		lea	16($key),$key
+		mov	0($key),%rax
+		mov	8($key),%rcx
+___
+		&dectransform ();
+$code.=<<___;
+		mov	%eax,0($key)
+		mov	%ebx,4($key)
+		mov	%ecx,8($key)
+		mov	%edx,12($key)
+		sub	\$1,%r14d
+	jnz	.Lpermute
+
+	xor	%rax,%rax
+.Labort:
+	mov	8(%rsp),%r15
+.cfi_restore	%r15
+	mov	16(%rsp),%r14
+.cfi_restore	%r14
+	mov	24(%rsp),%r13
+.cfi_restore	%r13
+	mov	32(%rsp),%r12
+.cfi_restore	%r12
+	mov	40(%rsp),%rbp
+.cfi_restore	%rbp
+	mov	48(%rsp),%rbx
+.cfi_restore	%rbx
+	add	\$56,%rsp
+.cfi_adjust_cfa_offset	-56
+.Ldec_key_epilogue:
+	ret
+.cfi_endproc
+.size	aes_nohw_set_decrypt_key,.-aes_nohw_set_decrypt_key
+___
+
+# void aes_nohw_cbc_encrypt (const void char *inp, unsigned char *out,
+#			                       size_t length, const AES_KEY *key,
+#			                       unsigned char *ivp,const int enc);
+{
+# stack frame layout
+# -8(%rsp)		return address
+my $keyp="0(%rsp)";		# one to pass as $key
+my $keyend="8(%rsp)";		# &(keyp->rd_key[4*keyp->rounds])
+my $_rsp="16(%rsp)";		# saved %rsp
+my $_inp="24(%rsp)";		# copy of 1st parameter, inp
+my $_out="32(%rsp)";		# copy of 2nd parameter, out
+my $_len="40(%rsp)";		# copy of 3rd parameter, length
+my $_key="48(%rsp)";		# copy of 4th parameter, key
+my $_ivp="56(%rsp)";		# copy of 5th parameter, ivp
+my $ivec="64(%rsp)";		# ivec[16]
+my $aes_key="80(%rsp)";		# copy of aes_key
+my $mark="80+240(%rsp)";	# copy of aes_key->rounds
+
+$code.=<<___;
+.align	16
+.globl	aes_nohw_cbc_encrypt
+.type	aes_nohw_cbc_encrypt,\@function,6
+.extern	OPENSSL_ia32cap_P
+.hidden	aes_nohw_cbc_encrypt
+aes_nohw_cbc_encrypt:
+.cfi_startproc
+	cmp	\$0,%rdx	# check length
+	je	.Lcbc_epilogue
+	pushfq
+# This could be .cfi_push 49, but libunwind fails on registers it does not
+# recognize. See https://bugzilla.redhat.com/show_bug.cgi?id=217087.
+.cfi_adjust_cfa_offset	8
+	push	%rbx
+.cfi_push	%rbx
+	push	%rbp
+.cfi_push	%rbp
+	push	%r12
+.cfi_push	%r12
+	push	%r13
+.cfi_push	%r13
+	push	%r14
+.cfi_push	%r14
+	push	%r15
+.cfi_push	%r15
+.Lcbc_prologue:
+
+	cld
+	mov	%r9d,%r9d	# clear upper half of enc
+
+	lea	.LAES_Te(%rip),$sbox
+	lea	.LAES_Td(%rip),%r10
+	cmp	\$0,%r9
+	cmoveq	%r10,$sbox
+
+.cfi_remember_state
+	leaq	OPENSSL_ia32cap_P(%rip),%r10
+	mov	(%r10), %r10d
+	cmp	\$$speed_limit,%rdx
+	jb	.Lcbc_slow_prologue
+	test	\$15,%rdx
+	jnz	.Lcbc_slow_prologue
+	bt	\$28,%r10d
+	jc	.Lcbc_slow_prologue
+
+	# allocate aligned stack frame...
+	lea	-88-248(%rsp),$key
+	and	\$-64,$key
+
+	# ... and make sure it doesn't alias with AES_T[ed] modulo 4096
+	mov	$sbox,%r10
+	lea	2304($sbox),%r11
+	mov	$key,%r12
+	and	\$0xFFF,%r10	# s = $sbox&0xfff
+	and	\$0xFFF,%r11	# e = ($sbox+2048)&0xfff
+	and	\$0xFFF,%r12	# p = %rsp&0xfff
+
+	cmp	%r11,%r12	# if (p=>e) %rsp =- (p-e);
+	jb	.Lcbc_te_break_out
+	sub	%r11,%r12
+	sub	%r12,$key
+	jmp	.Lcbc_te_ok
+.Lcbc_te_break_out:		# else %rsp -= (p-s)&0xfff + framesz
+	sub	%r10,%r12
+	and	\$0xFFF,%r12
+	add	\$320,%r12
+	sub	%r12,$key
+.align	4
+.Lcbc_te_ok:
+
+	xchg	%rsp,$key
+.cfi_def_cfa_register	$key
+	#add	\$8,%rsp	# reserve for return address!
+	mov	$key,$_rsp	# save %rsp
+.cfi_cfa_expression	$_rsp,deref,+64
+.Lcbc_fast_body:
+	mov	%rdi,$_inp	# save copy of inp
+	mov	%rsi,$_out	# save copy of out
+	mov	%rdx,$_len	# save copy of len
+	mov	%rcx,$_key	# save copy of key
+	mov	%r8,$_ivp	# save copy of ivp
+	movl	\$0,$mark	# copy of aes_key->rounds = 0;
+	mov	%r8,%rbp	# rearrange input arguments
+	mov	%r9,%rbx
+	mov	%rsi,$out
+	mov	%rdi,$inp
+	mov	%rcx,$key
+
+	mov	240($key),%eax		# key->rounds
+	# do we copy key schedule to stack?
+	mov	$key,%r10
+	sub	$sbox,%r10
+	and	\$0xfff,%r10
+	cmp	\$2304,%r10
+	jb	.Lcbc_do_ecopy
+	cmp	\$4096-248,%r10
+	jb	.Lcbc_skip_ecopy
+.align	4
+.Lcbc_do_ecopy:
+		mov	$key,%rsi
+		lea	$aes_key,%rdi
+		lea	$aes_key,$key
+		mov	\$240/8,%ecx
+		.long	0x90A548F3	# rep movsq
+		mov	%eax,(%rdi)	# copy aes_key->rounds
+.Lcbc_skip_ecopy:
+	mov	$key,$keyp	# save key pointer
+
+	mov	\$18,%ecx
+.align	4
+.Lcbc_prefetch_te:
+		mov	0($sbox),%r10
+		mov	32($sbox),%r11
+		mov	64($sbox),%r12
+		mov	96($sbox),%r13
+		lea	128($sbox),$sbox
+		sub	\$1,%ecx
+	jnz	.Lcbc_prefetch_te
+	lea	-2304($sbox),$sbox
+
+	cmp	\$0,%rbx
+	je	.LFAST_DECRYPT
+
+#----------------------------- ENCRYPT -----------------------------#
+	mov	0(%rbp),$s0		# load iv
+	mov	4(%rbp),$s1
+	mov	8(%rbp),$s2
+	mov	12(%rbp),$s3
+
+.align	4
+.Lcbc_fast_enc_loop:
+		xor	0($inp),$s0
+		xor	4($inp),$s1
+		xor	8($inp),$s2
+		xor	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_encrypt
+
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10
+		mov	$s0,0($out)
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		sub	\$16,%r10
+		test	\$-16,%r10
+		mov	%r10,$_len
+	jnz	.Lcbc_fast_enc_loop
+	mov	$_ivp,%rbp	# restore ivp
+	mov	$s0,0(%rbp)	# save ivec
+	mov	$s1,4(%rbp)
+	mov	$s2,8(%rbp)
+	mov	$s3,12(%rbp)
+
+	jmp	.Lcbc_fast_cleanup
+
+#----------------------------- DECRYPT -----------------------------#
+.align	16
+.LFAST_DECRYPT:
+	cmp	$inp,$out
+	je	.Lcbc_fast_dec_in_place
+
+	mov	%rbp,$ivec
+.align	4
+.Lcbc_fast_dec_loop:
+		mov	0($inp),$s0	# read input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_decrypt
+
+		mov	$ivec,%rbp	# load ivp
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10	# load len
+		xor	0(%rbp),$s0	# xor iv
+		xor	4(%rbp),$s1
+		xor	8(%rbp),$s2
+		xor	12(%rbp),$s3
+		mov	$inp,%rbp	# current input, next iv
+
+		sub	\$16,%r10
+		mov	%r10,$_len	# update len
+		mov	%rbp,$ivec	# update ivp
+
+		mov	$s0,0($out)	# write output
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+	jnz	.Lcbc_fast_dec_loop
+	mov	$_ivp,%r12		# load user ivp
+	mov	0(%rbp),%r10		# load iv
+	mov	8(%rbp),%r11
+	mov	%r10,0(%r12)		# copy back to user
+	mov	%r11,8(%r12)
+	jmp	.Lcbc_fast_cleanup
+
+.align	16
+.Lcbc_fast_dec_in_place:
+	mov	0(%rbp),%r10		# copy iv to stack
+	mov	8(%rbp),%r11
+	mov	%r10,0+$ivec
+	mov	%r11,8+$ivec
+.align	4
+.Lcbc_fast_dec_in_place_loop:
+		mov	0($inp),$s0	# load input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# if ($verticalspin) save inp
+
+		call	_x86_64_AES_decrypt
+
+		mov	$_inp,$inp	# if ($verticalspin) restore inp
+		mov	$_len,%r10
+		xor	0+$ivec,$s0
+		xor	4+$ivec,$s1
+		xor	8+$ivec,$s2
+		xor	12+$ivec,$s3
+
+		mov	0($inp),%r11	# load input
+		mov	8($inp),%r12
+		sub	\$16,%r10
+		jz	.Lcbc_fast_dec_in_place_done
+
+		mov	%r11,0+$ivec	# copy input to iv
+		mov	%r12,8+$ivec
+
+		mov	$s0,0($out)	# save output [zaps input]
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		mov	%r10,$_len
+	jmp	.Lcbc_fast_dec_in_place_loop
+.Lcbc_fast_dec_in_place_done:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)	# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0($out)	# save output [zaps input]
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+.align	4
+.Lcbc_fast_cleanup:
+	cmpl	\$0,$mark	# was the key schedule copied?
+	lea	$aes_key,%rdi
+	je	.Lcbc_exit
+		mov	\$240/8,%ecx
+		xor	%rax,%rax
+		.long	0x90AB48F3	# rep stosq
+
+	jmp	.Lcbc_exit
+
+#--------------------------- SLOW ROUTINE ---------------------------#
+.align	16
+.Lcbc_slow_prologue:
+.cfi_restore_state
+	# allocate aligned stack frame...
+	lea	-88(%rsp),%rbp
+	and	\$-64,%rbp
+	# ... just "above" key schedule
+	lea	-88-63(%rcx),%r10
+	sub	%rbp,%r10
+	neg	%r10
+	and	\$0x3c0,%r10
+	sub	%r10,%rbp
+
+	xchg	%rsp,%rbp
+.cfi_def_cfa_register	%rbp
+	#add	\$8,%rsp	# reserve for return address!
+	mov	%rbp,$_rsp	# save %rsp
+.cfi_cfa_expression	$_rsp,deref,+64
+.Lcbc_slow_body:
+	#mov	%rdi,$_inp	# save copy of inp
+	#mov	%rsi,$_out	# save copy of out
+	#mov	%rdx,$_len	# save copy of len
+	#mov	%rcx,$_key	# save copy of key
+	mov	%r8,$_ivp	# save copy of ivp
+	mov	%r8,%rbp	# rearrange input arguments
+	mov	%r9,%rbx
+	mov	%rsi,$out
+	mov	%rdi,$inp
+	mov	%rcx,$key
+	mov	%rdx,%r10
+
+	mov	240($key),%eax
+	mov	$key,$keyp	# save key pointer
+	shl	\$4,%eax
+	lea	($key,%rax),%rax
+	mov	%rax,$keyend
+
+	# pick Te4 copy which can't "overlap" with stack frame or key schedule
+	lea	2048($sbox),$sbox
+	lea	768-8(%rsp),%rax
+	sub	$sbox,%rax
+	and	\$0x300,%rax
+	lea	($sbox,%rax),$sbox
+
+	cmp	\$0,%rbx
+	je	.LSLOW_DECRYPT
+
+#--------------------------- SLOW ENCRYPT ---------------------------#
+	test	\$-16,%r10		# check upon length
+	mov	0(%rbp),$s0		# load iv
+	mov	4(%rbp),$s1
+	mov	8(%rbp),$s2
+	mov	12(%rbp),$s3
+	jz	.Lcbc_slow_enc_tail	# short input...
+
+.align	4
+.Lcbc_slow_enc_loop:
+		xor	0($inp),$s0
+		xor	4($inp),$s1
+		xor	8($inp),$s2
+		xor	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# save inp
+		mov	$out,$_out	# save out
+		mov	%r10,$_len	# save len
+
+		call	_x86_64_AES_encrypt_compact
+
+		mov	$_inp,$inp	# restore inp
+		mov	$_out,$out	# restore out
+		mov	$_len,%r10	# restore len
+		mov	$s0,0($out)
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+		sub	\$16,%r10
+		test	\$-16,%r10
+	jnz	.Lcbc_slow_enc_loop
+	test	\$15,%r10
+	jnz	.Lcbc_slow_enc_tail
+	mov	$_ivp,%rbp	# restore ivp
+	mov	$s0,0(%rbp)	# save ivec
+	mov	$s1,4(%rbp)
+	mov	$s2,8(%rbp)
+	mov	$s3,12(%rbp)
+
+	jmp	.Lcbc_exit
+
+.align	4
+.Lcbc_slow_enc_tail:
+	mov	%rax,%r11
+	mov	%rcx,%r12
+	mov	%r10,%rcx
+	mov	$inp,%rsi
+	mov	$out,%rdi
+	.long	0x9066A4F3		# rep movsb
+	mov	\$16,%rcx		# zero tail
+	sub	%r10,%rcx
+	xor	%rax,%rax
+	.long	0x9066AAF3		# rep stosb
+	mov	$out,$inp		# this is not a mistake!
+	mov	\$16,%r10		# len=16
+	mov	%r11,%rax
+	mov	%r12,%rcx
+	jmp	.Lcbc_slow_enc_loop	# one more spin...
+#--------------------------- SLOW DECRYPT ---------------------------#
+.align	16
+.LSLOW_DECRYPT:
+	shr	\$3,%rax
+	add	%rax,$sbox		# recall "magic" constants!
+
+	mov	0(%rbp),%r11		# copy iv to stack
+	mov	8(%rbp),%r12
+	mov	%r11,0+$ivec
+	mov	%r12,8+$ivec
+
+.align	4
+.Lcbc_slow_dec_loop:
+		mov	0($inp),$s0	# load input
+		mov	4($inp),$s1
+		mov	8($inp),$s2
+		mov	12($inp),$s3
+		mov	$keyp,$key	# restore key
+		mov	$inp,$_inp	# save inp
+		mov	$out,$_out	# save out
+		mov	%r10,$_len	# save len
+
+		call	_x86_64_AES_decrypt_compact
+
+		mov	$_inp,$inp	# restore inp
+		mov	$_out,$out	# restore out
+		mov	$_len,%r10
+		xor	0+$ivec,$s0
+		xor	4+$ivec,$s1
+		xor	8+$ivec,$s2
+		xor	12+$ivec,$s3
+
+		mov	0($inp),%r11	# load input
+		mov	8($inp),%r12
+		sub	\$16,%r10
+		jc	.Lcbc_slow_dec_partial
+		jz	.Lcbc_slow_dec_done
+
+		mov	%r11,0+$ivec	# copy input to iv
+		mov	%r12,8+$ivec
+
+		mov	$s0,0($out)	# save output [can zap input]
+		mov	$s1,4($out)
+		mov	$s2,8($out)
+		mov	$s3,12($out)
+
+		lea	16($inp),$inp
+		lea	16($out),$out
+	jmp	.Lcbc_slow_dec_loop
+.Lcbc_slow_dec_done:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)		# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0($out)		# save output [can zap input]
+	mov	$s1,4($out)
+	mov	$s2,8($out)
+	mov	$s3,12($out)
+
+	jmp	.Lcbc_exit
+
+.align	4
+.Lcbc_slow_dec_partial:
+	mov	$_ivp,%rdi
+	mov	%r11,0(%rdi)		# copy iv back to user
+	mov	%r12,8(%rdi)
+
+	mov	$s0,0+$ivec		# save output to stack
+	mov	$s1,4+$ivec
+	mov	$s2,8+$ivec
+	mov	$s3,12+$ivec
+
+	mov	$out,%rdi
+	lea	$ivec,%rsi
+	lea	16(%r10),%rcx
+	.long	0x9066A4F3	# rep movsb
+	jmp	.Lcbc_exit
+
+.align	16
+.Lcbc_exit:
+	mov	$_rsp,%rsi
+.cfi_def_cfa	%rsi,64
+	mov	(%rsi),%r15
+.cfi_restore	%r15
+	mov	8(%rsi),%r14
+.cfi_restore	%r14
+	mov	16(%rsi),%r13
+.cfi_restore	%r13
+	mov	24(%rsi),%r12
+.cfi_restore	%r12
+	mov	32(%rsi),%rbp
+.cfi_restore	%rbp
+	mov	40(%rsi),%rbx
+.cfi_restore	%rbx
+	lea	48(%rsi),%rsp
+.cfi_def_cfa	%rsp,16
+.Lcbc_popfq:
+	popfq
+# This could be .cfi_pop 49, but libunwind fails on registers it does not
+# recognize. See https://bugzilla.redhat.com/show_bug.cgi?id=217087.
+.cfi_adjust_cfa_offset	-8
+.Lcbc_epilogue:
+	ret
+.cfi_endproc
+.size	aes_nohw_cbc_encrypt,.-aes_nohw_cbc_encrypt
+___
+}
+
+$code.=<<___;
+.align	64
+.LAES_Te:
+___
+	&_data_word(0xa56363c6, 0x847c7cf8, 0x997777ee, 0x8d7b7bf6);
+	&_data_word(0x0df2f2ff, 0xbd6b6bd6, 0xb16f6fde, 0x54c5c591);
+	&_data_word(0x50303060, 0x03010102, 0xa96767ce, 0x7d2b2b56);
+	&_data_word(0x19fefee7, 0x62d7d7b5, 0xe6abab4d, 0x9a7676ec);
+	&_data_word(0x45caca8f, 0x9d82821f, 0x40c9c989, 0x877d7dfa);
+	&_data_word(0x15fafaef, 0xeb5959b2, 0xc947478e, 0x0bf0f0fb);
+	&_data_word(0xecadad41, 0x67d4d4b3, 0xfda2a25f, 0xeaafaf45);
+	&_data_word(0xbf9c9c23, 0xf7a4a453, 0x967272e4, 0x5bc0c09b);
+	&_data_word(0xc2b7b775, 0x1cfdfde1, 0xae93933d, 0x6a26264c);
+	&_data_word(0x5a36366c, 0x413f3f7e, 0x02f7f7f5, 0x4fcccc83);
+	&_data_word(0x5c343468, 0xf4a5a551, 0x34e5e5d1, 0x08f1f1f9);
+	&_data_word(0x937171e2, 0x73d8d8ab, 0x53313162, 0x3f15152a);
+	&_data_word(0x0c040408, 0x52c7c795, 0x65232346, 0x5ec3c39d);
+	&_data_word(0x28181830, 0xa1969637, 0x0f05050a, 0xb59a9a2f);
+	&_data_word(0x0907070e, 0x36121224, 0x9b80801b, 0x3de2e2df);
+	&_data_word(0x26ebebcd, 0x6927274e, 0xcdb2b27f, 0x9f7575ea);
+	&_data_word(0x1b090912, 0x9e83831d, 0x742c2c58, 0x2e1a1a34);
+	&_data_word(0x2d1b1b36, 0xb26e6edc, 0xee5a5ab4, 0xfba0a05b);
+	&_data_word(0xf65252a4, 0x4d3b3b76, 0x61d6d6b7, 0xceb3b37d);
+	&_data_word(0x7b292952, 0x3ee3e3dd, 0x712f2f5e, 0x97848413);
+	&_data_word(0xf55353a6, 0x68d1d1b9, 0x00000000, 0x2cededc1);
+	&_data_word(0x60202040, 0x1ffcfce3, 0xc8b1b179, 0xed5b5bb6);
+	&_data_word(0xbe6a6ad4, 0x46cbcb8d, 0xd9bebe67, 0x4b393972);
+	&_data_word(0xde4a4a94, 0xd44c4c98, 0xe85858b0, 0x4acfcf85);
+	&_data_word(0x6bd0d0bb, 0x2aefefc5, 0xe5aaaa4f, 0x16fbfbed);
+	&_data_word(0xc5434386, 0xd74d4d9a, 0x55333366, 0x94858511);
+	&_data_word(0xcf45458a, 0x10f9f9e9, 0x06020204, 0x817f7ffe);
+	&_data_word(0xf05050a0, 0x443c3c78, 0xba9f9f25, 0xe3a8a84b);
+	&_data_word(0xf35151a2, 0xfea3a35d, 0xc0404080, 0x8a8f8f05);
+	&_data_word(0xad92923f, 0xbc9d9d21, 0x48383870, 0x04f5f5f1);
+	&_data_word(0xdfbcbc63, 0xc1b6b677, 0x75dadaaf, 0x63212142);
+	&_data_word(0x30101020, 0x1affffe5, 0x0ef3f3fd, 0x6dd2d2bf);
+	&_data_word(0x4ccdcd81, 0x140c0c18, 0x35131326, 0x2fececc3);
+	&_data_word(0xe15f5fbe, 0xa2979735, 0xcc444488, 0x3917172e);
+	&_data_word(0x57c4c493, 0xf2a7a755, 0x827e7efc, 0x473d3d7a);
+	&_data_word(0xac6464c8, 0xe75d5dba, 0x2b191932, 0x957373e6);
+	&_data_word(0xa06060c0, 0x98818119, 0xd14f4f9e, 0x7fdcdca3);
+	&_data_word(0x66222244, 0x7e2a2a54, 0xab90903b, 0x8388880b);
+	&_data_word(0xca46468c, 0x29eeeec7, 0xd3b8b86b, 0x3c141428);
+	&_data_word(0x79dedea7, 0xe25e5ebc, 0x1d0b0b16, 0x76dbdbad);
+	&_data_word(0x3be0e0db, 0x56323264, 0x4e3a3a74, 0x1e0a0a14);
+	&_data_word(0xdb494992, 0x0a06060c, 0x6c242448, 0xe45c5cb8);
+	&_data_word(0x5dc2c29f, 0x6ed3d3bd, 0xefacac43, 0xa66262c4);
+	&_data_word(0xa8919139, 0xa4959531, 0x37e4e4d3, 0x8b7979f2);
+	&_data_word(0x32e7e7d5, 0x43c8c88b, 0x5937376e, 0xb76d6dda);
+	&_data_word(0x8c8d8d01, 0x64d5d5b1, 0xd24e4e9c, 0xe0a9a949);
+	&_data_word(0xb46c6cd8, 0xfa5656ac, 0x07f4f4f3, 0x25eaeacf);
+	&_data_word(0xaf6565ca, 0x8e7a7af4, 0xe9aeae47, 0x18080810);
+	&_data_word(0xd5baba6f, 0x887878f0, 0x6f25254a, 0x722e2e5c);
+	&_data_word(0x241c1c38, 0xf1a6a657, 0xc7b4b473, 0x51c6c697);
+	&_data_word(0x23e8e8cb, 0x7cdddda1, 0x9c7474e8, 0x211f1f3e);
+	&_data_word(0xdd4b4b96, 0xdcbdbd61, 0x868b8b0d, 0x858a8a0f);
+	&_data_word(0x907070e0, 0x423e3e7c, 0xc4b5b571, 0xaa6666cc);
+	&_data_word(0xd8484890, 0x05030306, 0x01f6f6f7, 0x120e0e1c);
+	&_data_word(0xa36161c2, 0x5f35356a, 0xf95757ae, 0xd0b9b969);
+	&_data_word(0x91868617, 0x58c1c199, 0x271d1d3a, 0xb99e9e27);
+	&_data_word(0x38e1e1d9, 0x13f8f8eb, 0xb398982b, 0x33111122);
+	&_data_word(0xbb6969d2, 0x70d9d9a9, 0x898e8e07, 0xa7949433);
+	&_data_word(0xb69b9b2d, 0x221e1e3c, 0x92878715, 0x20e9e9c9);
+	&_data_word(0x49cece87, 0xff5555aa, 0x78282850, 0x7adfdfa5);
+	&_data_word(0x8f8c8c03, 0xf8a1a159, 0x80898909, 0x170d0d1a);
+	&_data_word(0xdabfbf65, 0x31e6e6d7, 0xc6424284, 0xb86868d0);
+	&_data_word(0xc3414182, 0xb0999929, 0x772d2d5a, 0x110f0f1e);
+	&_data_word(0xcbb0b07b, 0xfc5454a8, 0xd6bbbb6d, 0x3a16162c);
+
+#Te4	# four copies of Te4 to choose from to avoid L1 aliasing
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+
+	&data_byte(0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5);
+	&data_byte(0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76);
+	&data_byte(0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0);
+	&data_byte(0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0);
+	&data_byte(0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc);
+	&data_byte(0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15);
+	&data_byte(0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a);
+	&data_byte(0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75);
+	&data_byte(0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0);
+	&data_byte(0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84);
+	&data_byte(0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b);
+	&data_byte(0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf);
+	&data_byte(0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85);
+	&data_byte(0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8);
+	&data_byte(0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5);
+	&data_byte(0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2);
+	&data_byte(0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17);
+	&data_byte(0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73);
+	&data_byte(0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88);
+	&data_byte(0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb);
+	&data_byte(0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c);
+	&data_byte(0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79);
+	&data_byte(0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9);
+	&data_byte(0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08);
+	&data_byte(0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6);
+	&data_byte(0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a);
+	&data_byte(0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e);
+	&data_byte(0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e);
+	&data_byte(0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94);
+	&data_byte(0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf);
+	&data_byte(0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68);
+	&data_byte(0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16);
+#rcon:
+$code.=<<___;
+	.long	0x00000001, 0x00000002, 0x00000004, 0x00000008
+	.long	0x00000010, 0x00000020, 0x00000040, 0x00000080
+	.long	0x0000001b, 0x00000036, 0x80808080, 0x80808080
+	.long	0xfefefefe, 0xfefefefe, 0x1b1b1b1b, 0x1b1b1b1b
+___
+$code.=<<___;
+.align	64
+.LAES_Td:
+___
+	&_data_word(0x50a7f451, 0x5365417e, 0xc3a4171a, 0x965e273a);
+	&_data_word(0xcb6bab3b, 0xf1459d1f, 0xab58faac, 0x9303e34b);
+	&_data_word(0x55fa3020, 0xf66d76ad, 0x9176cc88, 0x254c02f5);
+	&_data_word(0xfcd7e54f, 0xd7cb2ac5, 0x80443526, 0x8fa362b5);
+	&_data_word(0x495ab1de, 0x671bba25, 0x980eea45, 0xe1c0fe5d);
+	&_data_word(0x02752fc3, 0x12f04c81, 0xa397468d, 0xc6f9d36b);
+	&_data_word(0xe75f8f03, 0x959c9215, 0xeb7a6dbf, 0xda595295);
+	&_data_word(0x2d83bed4, 0xd3217458, 0x2969e049, 0x44c8c98e);
+	&_data_word(0x6a89c275, 0x78798ef4, 0x6b3e5899, 0xdd71b927);
+	&_data_word(0xb64fe1be, 0x17ad88f0, 0x66ac20c9, 0xb43ace7d);
+	&_data_word(0x184adf63, 0x82311ae5, 0x60335197, 0x457f5362);
+	&_data_word(0xe07764b1, 0x84ae6bbb, 0x1ca081fe, 0x942b08f9);
+	&_data_word(0x58684870, 0x19fd458f, 0x876cde94, 0xb7f87b52);
+	&_data_word(0x23d373ab, 0xe2024b72, 0x578f1fe3, 0x2aab5566);
+	&_data_word(0x0728ebb2, 0x03c2b52f, 0x9a7bc586, 0xa50837d3);
+	&_data_word(0xf2872830, 0xb2a5bf23, 0xba6a0302, 0x5c8216ed);
+	&_data_word(0x2b1ccf8a, 0x92b479a7, 0xf0f207f3, 0xa1e2694e);
+	&_data_word(0xcdf4da65, 0xd5be0506, 0x1f6234d1, 0x8afea6c4);
+	&_data_word(0x9d532e34, 0xa055f3a2, 0x32e18a05, 0x75ebf6a4);
+	&_data_word(0x39ec830b, 0xaaef6040, 0x069f715e, 0x51106ebd);
+	&_data_word(0xf98a213e, 0x3d06dd96, 0xae053edd, 0x46bde64d);
+	&_data_word(0xb58d5491, 0x055dc471, 0x6fd40604, 0xff155060);
+	&_data_word(0x24fb9819, 0x97e9bdd6, 0xcc434089, 0x779ed967);
+	&_data_word(0xbd42e8b0, 0x888b8907, 0x385b19e7, 0xdbeec879);
+	&_data_word(0x470a7ca1, 0xe90f427c, 0xc91e84f8, 0x00000000);
+	&_data_word(0x83868009, 0x48ed2b32, 0xac70111e, 0x4e725a6c);
+	&_data_word(0xfbff0efd, 0x5638850f, 0x1ed5ae3d, 0x27392d36);
+	&_data_word(0x64d90f0a, 0x21a65c68, 0xd1545b9b, 0x3a2e3624);
+	&_data_word(0xb1670a0c, 0x0fe75793, 0xd296eeb4, 0x9e919b1b);
+	&_data_word(0x4fc5c080, 0xa220dc61, 0x694b775a, 0x161a121c);
+	&_data_word(0x0aba93e2, 0xe52aa0c0, 0x43e0223c, 0x1d171b12);
+	&_data_word(0x0b0d090e, 0xadc78bf2, 0xb9a8b62d, 0xc8a91e14);
+	&_data_word(0x8519f157, 0x4c0775af, 0xbbdd99ee, 0xfd607fa3);
+	&_data_word(0x9f2601f7, 0xbcf5725c, 0xc53b6644, 0x347efb5b);
+	&_data_word(0x7629438b, 0xdcc623cb, 0x68fcedb6, 0x63f1e4b8);
+	&_data_word(0xcadc31d7, 0x10856342, 0x40229713, 0x2011c684);
+	&_data_word(0x7d244a85, 0xf83dbbd2, 0x1132f9ae, 0x6da129c7);
+	&_data_word(0x4b2f9e1d, 0xf330b2dc, 0xec52860d, 0xd0e3c177);
+	&_data_word(0x6c16b32b, 0x99b970a9, 0xfa489411, 0x2264e947);
+	&_data_word(0xc48cfca8, 0x1a3ff0a0, 0xd82c7d56, 0xef903322);
+	&_data_word(0xc74e4987, 0xc1d138d9, 0xfea2ca8c, 0x360bd498);
+	&_data_word(0xcf81f5a6, 0x28de7aa5, 0x268eb7da, 0xa4bfad3f);
+	&_data_word(0xe49d3a2c, 0x0d927850, 0x9bcc5f6a, 0x62467e54);
+	&_data_word(0xc2138df6, 0xe8b8d890, 0x5ef7392e, 0xf5afc382);
+	&_data_word(0xbe805d9f, 0x7c93d069, 0xa92dd56f, 0xb31225cf);
+	&_data_word(0x3b99acc8, 0xa77d1810, 0x6e639ce8, 0x7bbb3bdb);
+	&_data_word(0x097826cd, 0xf418596e, 0x01b79aec, 0xa89a4f83);
+	&_data_word(0x656e95e6, 0x7ee6ffaa, 0x08cfbc21, 0xe6e815ef);
+	&_data_word(0xd99be7ba, 0xce366f4a, 0xd4099fea, 0xd67cb029);
+	&_data_word(0xafb2a431, 0x31233f2a, 0x3094a5c6, 0xc066a235);
+	&_data_word(0x37bc4e74, 0xa6ca82fc, 0xb0d090e0, 0x15d8a733);
+	&_data_word(0x4a9804f1, 0xf7daec41, 0x0e50cd7f, 0x2ff69117);
+	&_data_word(0x8dd64d76, 0x4db0ef43, 0x544daacc, 0xdf0496e4);
+	&_data_word(0xe3b5d19e, 0x1b886a4c, 0xb81f2cc1, 0x7f516546);
+	&_data_word(0x04ea5e9d, 0x5d358c01, 0x737487fa, 0x2e410bfb);
+	&_data_word(0x5a1d67b3, 0x52d2db92, 0x335610e9, 0x1347d66d);
+	&_data_word(0x8c61d79a, 0x7a0ca137, 0x8e14f859, 0x893c13eb);
+	&_data_word(0xee27a9ce, 0x35c961b7, 0xede51ce1, 0x3cb1477a);
+	&_data_word(0x59dfd29c, 0x3f73f255, 0x79ce1418, 0xbf37c773);
+	&_data_word(0xeacdf753, 0x5baafd5f, 0x146f3ddf, 0x86db4478);
+	&_data_word(0x81f3afca, 0x3ec468b9, 0x2c342438, 0x5f40a3c2);
+	&_data_word(0x72c31d16, 0x0c25e2bc, 0x8b493c28, 0x41950dff);
+	&_data_word(0x7101a839, 0xdeb30c08, 0x9ce4b4d8, 0x90c15664);
+	&_data_word(0x6184cb7b, 0x70b632d5, 0x745c6c48, 0x4257b8d0);
+
+#Td4:	# four copies of Td4 to choose from to avoid L1 aliasing
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+___
+	&data_byte(0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38);
+	&data_byte(0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb);
+	&data_byte(0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87);
+	&data_byte(0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb);
+	&data_byte(0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d);
+	&data_byte(0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e);
+	&data_byte(0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2);
+	&data_byte(0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25);
+	&data_byte(0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16);
+	&data_byte(0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92);
+	&data_byte(0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda);
+	&data_byte(0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84);
+	&data_byte(0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a);
+	&data_byte(0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06);
+	&data_byte(0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02);
+	&data_byte(0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b);
+	&data_byte(0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea);
+	&data_byte(0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73);
+	&data_byte(0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85);
+	&data_byte(0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e);
+	&data_byte(0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89);
+	&data_byte(0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b);
+	&data_byte(0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20);
+	&data_byte(0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4);
+	&data_byte(0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31);
+	&data_byte(0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f);
+	&data_byte(0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d);
+	&data_byte(0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef);
+	&data_byte(0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0);
+	&data_byte(0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61);
+	&data_byte(0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26);
+	&data_byte(0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d);
+$code.=<<___;
+	.long	0x80808080, 0x80808080, 0xfefefefe, 0xfefefefe
+	.long	0x1b1b1b1b, 0x1b1b1b1b, 0, 0
+.asciz  "AES for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
+.align	64
+___
+
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+#		CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern	__imp_RtlVirtualUnwind
+.type	block_se_handler,\@abi-omnipotent
+.align	16
+block_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_block_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_block_prologue
+
+	mov	24(%rax),%rax		# pull saved real stack pointer
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_block_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	jmp	.Lcommon_seh_exit
+.size	block_se_handler,.-block_se_handler
+
+.type	key_se_handler,\@abi-omnipotent
+.align	16
+key_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	mov	8($disp),%rsi		# disp->ImageBase
+	mov	56($disp),%r11		# disp->HandlerData
+
+	mov	0(%r11),%r10d		# HandlerData[0]
+	lea	(%rsi,%r10),%r10	# prologue label
+	cmp	%r10,%rbx		# context->Rip<prologue label
+	jb	.Lin_key_prologue
+
+	mov	152($context),%rax	# pull context->Rsp
+
+	mov	4(%r11),%r10d		# HandlerData[1]
+	lea	(%rsi,%r10),%r10	# epilogue label
+	cmp	%r10,%rbx		# context->Rip>=epilogue label
+	jae	.Lin_key_prologue
+
+	lea	56(%rax),%rax
+
+	mov	-8(%rax),%rbx
+	mov	-16(%rax),%rbp
+	mov	-24(%rax),%r12
+	mov	-32(%rax),%r13
+	mov	-40(%rax),%r14
+	mov	-48(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_key_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+	jmp	.Lcommon_seh_exit
+.size	key_se_handler,.-key_se_handler
+
+.type	cbc_se_handler,\@abi-omnipotent
+.align	16
+cbc_se_handler:
+	push	%rsi
+	push	%rdi
+	push	%rbx
+	push	%rbp
+	push	%r12
+	push	%r13
+	push	%r14
+	push	%r15
+	pushfq
+	sub	\$64,%rsp
+
+	mov	120($context),%rax	# pull context->Rax
+	mov	248($context),%rbx	# pull context->Rip
+
+	lea	.Lcbc_prologue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_prologue
+	jb	.Lin_cbc_prologue
+
+	lea	.Lcbc_fast_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_fast_body
+	jb	.Lin_cbc_frame_setup
+
+	lea	.Lcbc_slow_prologue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_prologue
+	jb	.Lin_cbc_body
+
+	lea	.Lcbc_slow_body(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip<.Lcbc_slow_body
+	jb	.Lin_cbc_frame_setup
+
+.Lin_cbc_body:
+	mov	152($context),%rax	# pull context->Rsp
+
+	lea	.Lcbc_epilogue(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>=.Lcbc_epilogue
+	jae	.Lin_cbc_prologue
+
+	lea	8(%rax),%rax
+
+	lea	.Lcbc_popfq(%rip),%r10
+	cmp	%r10,%rbx		# context->Rip>=.Lcbc_popfq
+	jae	.Lin_cbc_prologue
+
+	mov	`16-8`(%rax),%rax	# biased $_rsp
+	lea	56(%rax),%rax
+
+.Lin_cbc_frame_setup:
+	mov	-16(%rax),%rbx
+	mov	-24(%rax),%rbp
+	mov	-32(%rax),%r12
+	mov	-40(%rax),%r13
+	mov	-48(%rax),%r14
+	mov	-56(%rax),%r15
+	mov	%rbx,144($context)	# restore context->Rbx
+	mov	%rbp,160($context)	# restore context->Rbp
+	mov	%r12,216($context)	# restore context->R12
+	mov	%r13,224($context)	# restore context->R13
+	mov	%r14,232($context)	# restore context->R14
+	mov	%r15,240($context)	# restore context->R15
+
+.Lin_cbc_prologue:
+	mov	8(%rax),%rdi
+	mov	16(%rax),%rsi
+	mov	%rax,152($context)	# restore context->Rsp
+	mov	%rsi,168($context)	# restore context->Rsi
+	mov	%rdi,176($context)	# restore context->Rdi
+
+.Lcommon_seh_exit:
+
+	mov	40($disp),%rdi		# disp->ContextRecord
+	mov	$context,%rsi		# context
+	mov	\$`1232/8`,%ecx		# sizeof(CONTEXT)
+	.long	0xa548f3fc		# cld; rep movsq
+
+	mov	$disp,%rsi
+	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
+	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
+	mov	0(%rsi),%r8		# arg3, disp->ControlPc
+	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
+	mov	40(%rsi),%r10		# disp->ContextRecord
+	lea	56(%rsi),%r11		# &disp->HandlerData
+	lea	24(%rsi),%r12		# &disp->EstablisherFrame
+	mov	%r10,32(%rsp)		# arg5
+	mov	%r11,40(%rsp)		# arg6
+	mov	%r12,48(%rsp)		# arg7
+	mov	%rcx,56(%rsp)		# arg8, (NULL)
+	call	*__imp_RtlVirtualUnwind(%rip)
+
+	mov	\$1,%eax		# ExceptionContinueSearch
+	add	\$64,%rsp
+	popfq
+	pop	%r15
+	pop	%r14
+	pop	%r13
+	pop	%r12
+	pop	%rbp
+	pop	%rbx
+	pop	%rdi
+	pop	%rsi
+	ret
+.size	cbc_se_handler,.-cbc_se_handler
+
+.section	.pdata
+.align	4
+	.rva	.LSEH_begin_aes_nohw_encrypt
+	.rva	.LSEH_end_aes_nohw_encrypt
+	.rva	.LSEH_info_aes_nohw_encrypt
+
+	.rva	.LSEH_begin_aes_nohw_decrypt
+	.rva	.LSEH_end_aes_nohw_decrypt
+	.rva	.LSEH_info_aes_nohw_decrypt
+
+	.rva	.LSEH_begin_aes_nohw_set_encrypt_key
+	.rva	.LSEH_end_aes_nohw_set_encrypt_key
+	.rva	.LSEH_info_aes_nohw_set_encrypt_key
+
+	.rva	.LSEH_begin_aes_nohw_set_decrypt_key
+	.rva	.LSEH_end_aes_nohw_set_decrypt_key
+	.rva	.LSEH_info_aes_nohw_set_decrypt_key
+
+	.rva	.LSEH_begin_aes_nohw_cbc_encrypt
+	.rva	.LSEH_end_aes_nohw_cbc_encrypt
+	.rva	.LSEH_info_aes_nohw_cbc_encrypt
+
+.section	.xdata
+.align	8
+.LSEH_info_aes_nohw_encrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Lenc_prologue,.Lenc_epilogue	# HandlerData[]
+.LSEH_info_aes_nohw_decrypt:
+	.byte	9,0,0,0
+	.rva	block_se_handler
+	.rva	.Ldec_prologue,.Ldec_epilogue	# HandlerData[]
+.LSEH_info_aes_nohw_set_encrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Lenc_key_prologue,.Lenc_key_epilogue	# HandlerData[]
+.LSEH_info_aes_nohw_set_decrypt_key:
+	.byte	9,0,0,0
+	.rva	key_se_handler
+	.rva	.Ldec_key_prologue,.Ldec_key_epilogue	# HandlerData[]
+.LSEH_info_aes_nohw_cbc_encrypt:
+	.byte	9,0,0,0
+	.rva	cbc_se_handler
+___
+}
+
+$code =~ s/\`([^\`]*)\`/eval($1)/gem;
+
+print $code;
+
+close STDOUT or die "error closing STDOUT";
diff --git a/crypto/fipsmodule/aes/internal.h b/crypto/fipsmodule/aes/internal.h
index 5b80695..99d509a 100644
--- a/crypto/fipsmodule/aes/internal.h
+++ b/crypto/fipsmodule/aes/internal.h
@@ -218,17 +218,19 @@
 #endif  // !VPAES
 
 
+void aes_nohw_encrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key);
+void aes_nohw_decrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key);
 int aes_nohw_set_encrypt_key(const uint8_t *key, unsigned bits,
                              AES_KEY *aeskey);
 int aes_nohw_set_decrypt_key(const uint8_t *key, unsigned bits,
                              AES_KEY *aeskey);
-void aes_nohw_encrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key);
-void aes_nohw_decrypt(const uint8_t *in, uint8_t *out, const AES_KEY *key);
-void aes_nohw_ctr32_encrypt_blocks(const uint8_t *in, uint8_t *out,
-                                   size_t blocks, const AES_KEY *key,
-                                   const uint8_t ivec[16]);
+
+#if !defined(OPENSSL_NO_ASM) && \
+    (defined(OPENSSL_X86_64) || defined(OPENSSL_X86))
+#define AES_NOHW_CBC
 void aes_nohw_cbc_encrypt(const uint8_t *in, uint8_t *out, size_t len,
                           const AES_KEY *key, uint8_t *ivec, const int enc);
+#endif
 
 
 #if defined(__cplusplus)
diff --git a/crypto/fipsmodule/aes/mode_wrappers.c b/crypto/fipsmodule/aes/mode_wrappers.c
index 206fcfd..ae8a91b 100644
--- a/crypto/fipsmodule/aes/mode_wrappers.c
+++ b/crypto/fipsmodule/aes/mode_wrappers.c
@@ -79,10 +79,12 @@
     return;
   }
 
+#if defined(AES_NOHW_CBC)
   if (!vpaes_capable()) {
     aes_nohw_cbc_encrypt(in, out, len, key, ivec, enc);
     return;
   }
+#endif
   if (enc) {
     CRYPTO_cbc128_encrypt(in, out, len, key, ivec, AES_encrypt);
   } else {
diff --git a/crypto/fipsmodule/bcm.c b/crypto/fipsmodule/bcm.c
index 567a0cd..7485f6c 100644
--- a/crypto/fipsmodule/bcm.c
+++ b/crypto/fipsmodule/bcm.c
@@ -31,7 +31,6 @@
 #include "../internal.h"
 
 #include "aes/aes.c"
-#include "aes/aes_nohw.c"
 #include "aes/key_wrap.c"
 #include "aes/mode_wrappers.c"
 #include "bn/add.c"
diff --git a/crypto/fipsmodule/cipher/e_aes.c b/crypto/fipsmodule/cipher/e_aes.c
index 8f4907f..c6dd973 100644
--- a/crypto/fipsmodule/cipher/e_aes.c
+++ b/crypto/fipsmodule/cipher/e_aes.c
@@ -173,9 +173,11 @@
       ret = aes_nohw_set_decrypt_key(key, ctx->key_len * 8, &dat->ks.ks);
       dat->block = aes_nohw_decrypt;
       dat->stream.cbc = NULL;
+#if defined(AES_NOHW_CBC)
       if (mode == EVP_CIPH_CBC_MODE) {
         dat->stream.cbc = aes_nohw_cbc_encrypt;
       }
+#endif
     }
   } else if (hwaes_capable()) {
     ret = aes_hw_set_encrypt_key(key, ctx->key_len * 8, &dat->ks.ks);
@@ -207,9 +209,11 @@
     ret = aes_nohw_set_encrypt_key(key, ctx->key_len * 8, &dat->ks.ks);
     dat->block = aes_nohw_encrypt;
     dat->stream.cbc = NULL;
+#if defined(AES_NOHW_CBC)
     if (mode == EVP_CIPH_CBC_MODE) {
       dat->stream.cbc = aes_nohw_cbc_encrypt;
     }
+#endif
   }
 
   if (ret < 0) {
@@ -314,7 +318,7 @@
   if (out_block) {
     *out_block = aes_nohw_encrypt;
   }
-  return aes_nohw_ctr32_encrypt_blocks;
+  return NULL;
 }
 
 #if defined(OPENSSL_32_BIT)