summaryrefslogtreecommitdiffstats
path: root/crypto/sm4
diff options
context:
space:
mode:
authorzhangzhilei <zhangzhilei@360.cn>2022-02-24 12:26:57 +0800
committerTomas Mraz <tomas@openssl.org>2022-03-03 13:19:55 +0100
commit13ba91cb02479a91b0743d2bf5f5ec7ce42860d0 (patch)
treeed194fd8b129020d04a5e2eab5840ec027a19b2d /crypto/sm4
parent43135a5d2274c24e97f50e16ce492c22eb717ab2 (diff)
SM4 optimization for non-asm mode
This patch use table-lookup borrow from aes in crypto/aes/aes_core.c. Test on my PC(AMD Ryzen Threadripper 3990X 64-Core Processor), before and after optimization: debug mode: Before: type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes 16384 bytes SM4-CBC 40101.14k 41453.80k 42073.86k 42174.81k 42216.11k 42227.03k SM4-ECB 41222.60k 42074.88k 42673.66k 42868.05k 42896.04k 42844.16k SM4-CTR 35867.22k 36874.47k 37004.97k 37083.82k 37052.42k 37076.99k After: type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes 16384 bytes SM4-CBC 47273.51k 48957.40k 49665.19k 49810.77k 49859.24k 49834.67k SM4-ECB 48100.01k 49323.34k 50224.04k 50273.28k 50533.72k 50730.12k SM4-CTR 41352.64k 42621.29k 42971.22k 43061.59k 43089.92k 43100.84k non-debug mode: Before: type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes 16384 bytes SM4-CBC 141596.59k 145102.93k 146794.50k 146540.89k 146650.45k 146877.10k SM4-ECB 144774.71k 155106.28k 158166.36k 158279.00k 158520.66k 159280.97k SM4-CTR 138021.10k 141577.60k 142493.53k 142736.38k 142852.10k 143125.16k After: type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes 16384 bytes SM4-CBC 142016.95k 150068.48k 152238.25k 152773.97k 153094.83k 152027.14k SM4-ECB 148842.94k 159919.87k 163628.37k 164515.84k 164697.43k 164790.27k SM4-CTR 141774.23k 146206.89k 147470.25k 147816.28k 146770.60k 148346.20k Reviewed-by: Paul Dale <pauli@openssl.org> Reviewed-by: Tomas Mraz <tomas@openssl.org> (Merged from https://github.com/openssl/openssl/pull/17766)
Diffstat (limited to 'crypto/sm4')
-rw-r--r--crypto/sm4/sm4.c145
1 files changed, 140 insertions, 5 deletions
diff --git a/crypto/sm4/sm4.c b/crypto/sm4/sm4.c
index 1e11ee6506..492b670234 100644
--- a/crypto/sm4/sm4.c
+++ b/crypto/sm4/sm4.c
@@ -40,7 +40,7 @@ static const uint8_t SM4_S[256] = {
/*
* SM4_SBOX_T[j] == L(SM4_SBOX[j]).
*/
-static const uint32_t SM4_SBOX_T[256] = {
+static const uint32_t SM4_SBOX_T0[256] = {
0x8ED55B5B, 0xD0924242, 0x4DEAA7A7, 0x06FDFBFB, 0xFCCF3333, 0x65E28787,
0xC93DF4F4, 0x6BB5DEDE, 0x4E165858, 0x6EB4DADA, 0x44145050, 0xCAC10B0B,
0x8828A0A0, 0x17F8EFEF, 0x9C2CB0B0, 0x11051414, 0x872BACAC, 0xFB669D9D,
@@ -85,6 +85,141 @@ static const uint32_t SM4_SBOX_T[256] = {
0x794C3535, 0xA0208080, 0x9D78E5E5, 0x56EDBBBB, 0x235E7D7D, 0xC63EF8F8,
0x8BD45F5F, 0xE7C82F2F, 0xDD39E4E4, 0x68492121 };
+static uint32_t SM4_SBOX_T1[256] = {
+ 0x5B8ED55B, 0x42D09242, 0xA74DEAA7, 0xFB06FDFB, 0x33FCCF33, 0x8765E287,
+ 0xF4C93DF4, 0xDE6BB5DE, 0x584E1658, 0xDA6EB4DA, 0x50441450, 0x0BCAC10B,
+ 0xA08828A0, 0xEF17F8EF, 0xB09C2CB0, 0x14110514, 0xAC872BAC, 0x9DFB669D,
+ 0x6AF2986A, 0xD9AE77D9, 0xA8822AA8, 0xFA46BCFA, 0x10140410, 0x0FCFC00F,
+ 0xAA02A8AA, 0x11544511, 0x4C5F134C, 0x98BE2698, 0x256D4825, 0x1A9E841A,
+ 0x181E0618, 0x66FD9B66, 0x72EC9E72, 0x094A4309, 0x41105141, 0xD324F7D3,
+ 0x46D59346, 0xBF53ECBF, 0x62F89A62, 0xE9927BE9, 0xCCFF33CC, 0x51045551,
+ 0x2C270B2C, 0x0D4F420D, 0xB759EEB7, 0x3FF3CC3F, 0xB21CAEB2, 0x89EA6389,
+ 0x9374E793, 0xCE7FB1CE, 0x706C1C70, 0xA60DABA6, 0x27EDCA27, 0x20280820,
+ 0xA348EBA3, 0x56C19756, 0x02808202, 0x7FA3DC7F, 0x52C49652, 0xEB12F9EB,
+ 0xD5A174D5, 0x3EB38D3E, 0xFCC33FFC, 0x9A3EA49A, 0x1D5B461D, 0x1C1B071C,
+ 0x9E3BA59E, 0xF30CFFF3, 0xCF3FF0CF, 0xCDBF72CD, 0x5C4B175C, 0xEA52B8EA,
+ 0x0E8F810E, 0x653D5865, 0xF0CC3CF0, 0x647D1964, 0x9B7EE59B, 0x16918716,
+ 0x3D734E3D, 0xA208AAA2, 0xA1C869A1, 0xADC76AAD, 0x06858306, 0xCA7AB0CA,
+ 0xC5B570C5, 0x91F46591, 0x6BB2D96B, 0x2EA7892E, 0xE318FBE3, 0xAF47E8AF,
+ 0x3C330F3C, 0x2D674A2D, 0xC1B071C1, 0x590E5759, 0x76E99F76, 0xD4E135D4,
+ 0x78661E78, 0x90B42490, 0x38360E38, 0x79265F79, 0x8DEF628D, 0x61385961,
+ 0x4795D247, 0x8A2AA08A, 0x94B12594, 0x88AA2288, 0xF18C7DF1, 0xECD73BEC,
+ 0x04050104, 0x84A52184, 0xE19879E1, 0x1E9B851E, 0x5384D753, 0x00000000,
+ 0x195E4719, 0x5D0B565D, 0x7EE39D7E, 0x4F9FD04F, 0x9CBB279C, 0x491A5349,
+ 0x317C4D31, 0xD8EE36D8, 0x080A0208, 0x9F7BE49F, 0x8220A282, 0x13D4C713,
+ 0x23E8CB23, 0x7AE69C7A, 0xAB42E9AB, 0xFE43BDFE, 0x2AA2882A, 0x4B9AD14B,
+ 0x01404101, 0x1FDBC41F, 0xE0D838E0, 0xD661B7D6, 0x8E2FA18E, 0xDF2BF4DF,
+ 0xCB3AF1CB, 0x3BF6CD3B, 0xE71DFAE7, 0x85E56085, 0x54411554, 0x8625A386,
+ 0x8360E383, 0xBA16ACBA, 0x75295C75, 0x9234A692, 0x6EF7996E, 0xD0E434D0,
+ 0x68721A68, 0x55015455, 0xB619AFB6, 0x4EDF914E, 0xC8FA32C8, 0xC0F030C0,
+ 0xD721F6D7, 0x32BC8E32, 0xC675B3C6, 0x8F6FE08F, 0x74691D74, 0xDB2EF5DB,
+ 0x8B6AE18B, 0xB8962EB8, 0x0A8A800A, 0x99FE6799, 0x2BE2C92B, 0x81E06181,
+ 0x03C0C303, 0xA48D29A4, 0x8CAF238C, 0xAE07A9AE, 0x34390D34, 0x4D1F524D,
+ 0x39764F39, 0xBDD36EBD, 0x5781D657, 0x6FB7D86F, 0xDCEB37DC, 0x15514415,
+ 0x7BA6DD7B, 0xF709FEF7, 0x3AB68C3A, 0xBC932FBC, 0x0C0F030C, 0xFF03FCFF,
+ 0xA9C26BA9, 0xC9BA73C9, 0xB5D96CB5, 0xB1DC6DB1, 0x6D375A6D, 0x45155045,
+ 0x36B98F36, 0x6C771B6C, 0xBE13ADBE, 0x4ADA904A, 0xEE57B9EE, 0x77A9DE77,
+ 0xF24CBEF2, 0xFD837EFD, 0x44551144, 0x67BDDA67, 0x712C5D71, 0x05454005,
+ 0x7C631F7C, 0x40501040, 0x69325B69, 0x63B8DB63, 0x28220A28, 0x07C5C207,
+ 0xC4F531C4, 0x22A88A22, 0x9631A796, 0x37F9CE37, 0xED977AED, 0xF649BFF6,
+ 0xB4992DB4, 0xD1A475D1, 0x4390D343, 0x485A1248, 0xE258BAE2, 0x9771E697,
+ 0xD264B6D2, 0xC270B2C2, 0x26AD8B26, 0xA5CD68A5, 0x5ECB955E, 0x29624B29,
+ 0x303C0C30, 0x5ACE945A, 0xDDAB76DD, 0xF9867FF9, 0x95F16495, 0xE65DBBE6,
+ 0xC735F2C7, 0x242D0924, 0x17D1C617, 0xB9D66FB9, 0x1BDEC51B, 0x12948612,
+ 0x60781860, 0xC330F3C3, 0xF5897CF5, 0xB35CEFB3, 0xE8D23AE8, 0x73ACDF73,
+ 0x35794C35, 0x80A02080, 0xE59D78E5, 0xBB56EDBB, 0x7D235E7D, 0xF8C63EF8,
+ 0x5F8BD45F, 0x2FE7C82F, 0xE4DD39E4, 0x21684921};
+
+static uint32_t SM4_SBOX_T2[256] = {
+ 0x5B5B8ED5, 0x4242D092, 0xA7A74DEA, 0xFBFB06FD, 0x3333FCCF, 0x878765E2,
+ 0xF4F4C93D, 0xDEDE6BB5, 0x58584E16, 0xDADA6EB4, 0x50504414, 0x0B0BCAC1,
+ 0xA0A08828, 0xEFEF17F8, 0xB0B09C2C, 0x14141105, 0xACAC872B, 0x9D9DFB66,
+ 0x6A6AF298, 0xD9D9AE77, 0xA8A8822A, 0xFAFA46BC, 0x10101404, 0x0F0FCFC0,
+ 0xAAAA02A8, 0x11115445, 0x4C4C5F13, 0x9898BE26, 0x25256D48, 0x1A1A9E84,
+ 0x18181E06, 0x6666FD9B, 0x7272EC9E, 0x09094A43, 0x41411051, 0xD3D324F7,
+ 0x4646D593, 0xBFBF53EC, 0x6262F89A, 0xE9E9927B, 0xCCCCFF33, 0x51510455,
+ 0x2C2C270B, 0x0D0D4F42, 0xB7B759EE, 0x3F3FF3CC, 0xB2B21CAE, 0x8989EA63,
+ 0x939374E7, 0xCECE7FB1, 0x70706C1C, 0xA6A60DAB, 0x2727EDCA, 0x20202808,
+ 0xA3A348EB, 0x5656C197, 0x02028082, 0x7F7FA3DC, 0x5252C496, 0xEBEB12F9,
+ 0xD5D5A174, 0x3E3EB38D, 0xFCFCC33F, 0x9A9A3EA4, 0x1D1D5B46, 0x1C1C1B07,
+ 0x9E9E3BA5, 0xF3F30CFF, 0xCFCF3FF0, 0xCDCDBF72, 0x5C5C4B17, 0xEAEA52B8,
+ 0x0E0E8F81, 0x65653D58, 0xF0F0CC3C, 0x64647D19, 0x9B9B7EE5, 0x16169187,
+ 0x3D3D734E, 0xA2A208AA, 0xA1A1C869, 0xADADC76A, 0x06068583, 0xCACA7AB0,
+ 0xC5C5B570, 0x9191F465, 0x6B6BB2D9, 0x2E2EA789, 0xE3E318FB, 0xAFAF47E8,
+ 0x3C3C330F, 0x2D2D674A, 0xC1C1B071, 0x59590E57, 0x7676E99F, 0xD4D4E135,
+ 0x7878661E, 0x9090B424, 0x3838360E, 0x7979265F, 0x8D8DEF62, 0x61613859,
+ 0x474795D2, 0x8A8A2AA0, 0x9494B125, 0x8888AA22, 0xF1F18C7D, 0xECECD73B,
+ 0x04040501, 0x8484A521, 0xE1E19879, 0x1E1E9B85, 0x535384D7, 0x00000000,
+ 0x19195E47, 0x5D5D0B56, 0x7E7EE39D, 0x4F4F9FD0, 0x9C9CBB27, 0x49491A53,
+ 0x31317C4D, 0xD8D8EE36, 0x08080A02, 0x9F9F7BE4, 0x828220A2, 0x1313D4C7,
+ 0x2323E8CB, 0x7A7AE69C, 0xABAB42E9, 0xFEFE43BD, 0x2A2AA288, 0x4B4B9AD1,
+ 0x01014041, 0x1F1FDBC4, 0xE0E0D838, 0xD6D661B7, 0x8E8E2FA1, 0xDFDF2BF4,
+ 0xCBCB3AF1, 0x3B3BF6CD, 0xE7E71DFA, 0x8585E560, 0x54544115, 0x868625A3,
+ 0x838360E3, 0xBABA16AC, 0x7575295C, 0x929234A6, 0x6E6EF799, 0xD0D0E434,
+ 0x6868721A, 0x55550154, 0xB6B619AF, 0x4E4EDF91, 0xC8C8FA32, 0xC0C0F030,
+ 0xD7D721F6, 0x3232BC8E, 0xC6C675B3, 0x8F8F6FE0, 0x7474691D, 0xDBDB2EF5,
+ 0x8B8B6AE1, 0xB8B8962E, 0x0A0A8A80, 0x9999FE67, 0x2B2BE2C9, 0x8181E061,
+ 0x0303C0C3, 0xA4A48D29, 0x8C8CAF23, 0xAEAE07A9, 0x3434390D, 0x4D4D1F52,
+ 0x3939764F, 0xBDBDD36E, 0x575781D6, 0x6F6FB7D8, 0xDCDCEB37, 0x15155144,
+ 0x7B7BA6DD, 0xF7F709FE, 0x3A3AB68C, 0xBCBC932F, 0x0C0C0F03, 0xFFFF03FC,
+ 0xA9A9C26B, 0xC9C9BA73, 0xB5B5D96C, 0xB1B1DC6D, 0x6D6D375A, 0x45451550,
+ 0x3636B98F, 0x6C6C771B, 0xBEBE13AD, 0x4A4ADA90, 0xEEEE57B9, 0x7777A9DE,
+ 0xF2F24CBE, 0xFDFD837E, 0x44445511, 0x6767BDDA, 0x71712C5D, 0x05054540,
+ 0x7C7C631F, 0x40405010, 0x6969325B, 0x6363B8DB, 0x2828220A, 0x0707C5C2,
+ 0xC4C4F531, 0x2222A88A, 0x969631A7, 0x3737F9CE, 0xEDED977A, 0xF6F649BF,
+ 0xB4B4992D, 0xD1D1A475, 0x434390D3, 0x48485A12, 0xE2E258BA, 0x979771E6,
+ 0xD2D264B6, 0xC2C270B2, 0x2626AD8B, 0xA5A5CD68, 0x5E5ECB95, 0x2929624B,
+ 0x30303C0C, 0x5A5ACE94, 0xDDDDAB76, 0xF9F9867F, 0x9595F164, 0xE6E65DBB,
+ 0xC7C735F2, 0x24242D09, 0x1717D1C6, 0xB9B9D66F, 0x1B1BDEC5, 0x12129486,
+ 0x60607818, 0xC3C330F3, 0xF5F5897C, 0xB3B35CEF, 0xE8E8D23A, 0x7373ACDF,
+ 0x3535794C, 0x8080A020, 0xE5E59D78, 0xBBBB56ED, 0x7D7D235E, 0xF8F8C63E,
+ 0x5F5F8BD4, 0x2F2FE7C8, 0xE4E4DD39, 0x21216849};
+
+static uint32_t SM4_SBOX_T3[256] = {
+ 0xD55B5B8E, 0x924242D0, 0xEAA7A74D, 0xFDFBFB06, 0xCF3333FC, 0xE2878765,
+ 0x3DF4F4C9, 0xB5DEDE6B, 0x1658584E, 0xB4DADA6E, 0x14505044, 0xC10B0BCA,
+ 0x28A0A088, 0xF8EFEF17, 0x2CB0B09C, 0x05141411, 0x2BACAC87, 0x669D9DFB,
+ 0x986A6AF2, 0x77D9D9AE, 0x2AA8A882, 0xBCFAFA46, 0x04101014, 0xC00F0FCF,
+ 0xA8AAAA02, 0x45111154, 0x134C4C5F, 0x269898BE, 0x4825256D, 0x841A1A9E,
+ 0x0618181E, 0x9B6666FD, 0x9E7272EC, 0x4309094A, 0x51414110, 0xF7D3D324,
+ 0x934646D5, 0xECBFBF53, 0x9A6262F8, 0x7BE9E992, 0x33CCCCFF, 0x55515104,
+ 0x0B2C2C27, 0x420D0D4F, 0xEEB7B759, 0xCC3F3FF3, 0xAEB2B21C, 0x638989EA,
+ 0xE7939374, 0xB1CECE7F, 0x1C70706C, 0xABA6A60D, 0xCA2727ED, 0x08202028,
+ 0xEBA3A348, 0x975656C1, 0x82020280, 0xDC7F7FA3, 0x965252C4, 0xF9EBEB12,
+ 0x74D5D5A1, 0x8D3E3EB3, 0x3FFCFCC3, 0xA49A9A3E, 0x461D1D5B, 0x071C1C1B,
+ 0xA59E9E3B, 0xFFF3F30C, 0xF0CFCF3F, 0x72CDCDBF, 0x175C5C4B, 0xB8EAEA52,
+ 0x810E0E8F, 0x5865653D, 0x3CF0F0CC, 0x1964647D, 0xE59B9B7E, 0x87161691,
+ 0x4E3D3D73, 0xAAA2A208, 0x69A1A1C8, 0x6AADADC7, 0x83060685, 0xB0CACA7A,
+ 0x70C5C5B5, 0x659191F4, 0xD96B6BB2, 0x892E2EA7, 0xFBE3E318, 0xE8AFAF47,
+ 0x0F3C3C33, 0x4A2D2D67, 0x71C1C1B0, 0x5759590E, 0x9F7676E9, 0x35D4D4E1,
+ 0x1E787866, 0x249090B4, 0x0E383836, 0x5F797926, 0x628D8DEF, 0x59616138,
+ 0xD2474795, 0xA08A8A2A, 0x259494B1, 0x228888AA, 0x7DF1F18C, 0x3BECECD7,
+ 0x01040405, 0x218484A5, 0x79E1E198, 0x851E1E9B, 0xD7535384, 0x00000000,
+ 0x4719195E, 0x565D5D0B, 0x9D7E7EE3, 0xD04F4F9F, 0x279C9CBB, 0x5349491A,
+ 0x4D31317C, 0x36D8D8EE, 0x0208080A, 0xE49F9F7B, 0xA2828220, 0xC71313D4,
+ 0xCB2323E8, 0x9C7A7AE6, 0xE9ABAB42, 0xBDFEFE43, 0x882A2AA2, 0xD14B4B9A,
+ 0x41010140, 0xC41F1FDB, 0x38E0E0D8, 0xB7D6D661, 0xA18E8E2F, 0xF4DFDF2B,
+ 0xF1CBCB3A, 0xCD3B3BF6, 0xFAE7E71D, 0x608585E5, 0x15545441, 0xA3868625,
+ 0xE3838360, 0xACBABA16, 0x5C757529, 0xA6929234, 0x996E6EF7, 0x34D0D0E4,
+ 0x1A686872, 0x54555501, 0xAFB6B619, 0x914E4EDF, 0x32C8C8FA, 0x30C0C0F0,
+ 0xF6D7D721, 0x8E3232BC, 0xB3C6C675, 0xE08F8F6F, 0x1D747469, 0xF5DBDB2E,
+ 0xE18B8B6A, 0x2EB8B896, 0x800A0A8A, 0x679999FE, 0xC92B2BE2, 0x618181E0,
+ 0xC30303C0, 0x29A4A48D, 0x238C8CAF, 0xA9AEAE07, 0x0D343439, 0x524D4D1F,
+ 0x4F393976, 0x6EBDBDD3, 0xD6575781, 0xD86F6FB7, 0x37DCDCEB, 0x44151551,
+ 0xDD7B7BA6, 0xFEF7F709, 0x8C3A3AB6, 0x2FBCBC93, 0x030C0C0F, 0xFCFFFF03,
+ 0x6BA9A9C2, 0x73C9C9BA, 0x6CB5B5D9, 0x6DB1B1DC, 0x5A6D6D37, 0x50454515,
+ 0x8F3636B9, 0x1B6C6C77, 0xADBEBE13, 0x904A4ADA, 0xB9EEEE57, 0xDE7777A9,
+ 0xBEF2F24C, 0x7EFDFD83, 0x11444455, 0xDA6767BD, 0x5D71712C, 0x40050545,
+ 0x1F7C7C63, 0x10404050, 0x5B696932, 0xDB6363B8, 0x0A282822, 0xC20707C5,
+ 0x31C4C4F5, 0x8A2222A8, 0xA7969631, 0xCE3737F9, 0x7AEDED97, 0xBFF6F649,
+ 0x2DB4B499, 0x75D1D1A4, 0xD3434390, 0x1248485A, 0xBAE2E258, 0xE6979771,
+ 0xB6D2D264, 0xB2C2C270, 0x8B2626AD, 0x68A5A5CD, 0x955E5ECB, 0x4B292962,
+ 0x0C30303C, 0x945A5ACE, 0x76DDDDAB, 0x7FF9F986, 0x649595F1, 0xBBE6E65D,
+ 0xF2C7C735, 0x0924242D, 0xC61717D1, 0x6FB9B9D6, 0xC51B1BDE, 0x86121294,
+ 0x18606078, 0xF3C3C330, 0x7CF5F589, 0xEFB3B35C, 0x3AE8E8D2, 0xDF7373AC,
+ 0x4C353579, 0x208080A0, 0x78E5E59D, 0xEDBBBB56, 0x5E7D7D23, 0x3EF8F8C6,
+ 0xD45F5F8B, 0xC82F2FE7, 0x39E4E4DD, 0x49212168};
+
static ossl_inline uint32_t rotl(uint32_t a, uint8_t n)
{
return (a << n) | (a >> (32 - n));
@@ -123,10 +258,10 @@ static ossl_inline uint32_t SM4_T_slow(uint32_t X)
static ossl_inline uint32_t SM4_T(uint32_t X)
{
- return SM4_SBOX_T[(uint8_t)(X >> 24)] ^
- rotl(SM4_SBOX_T[(uint8_t)(X >> 16)], 24) ^
- rotl(SM4_SBOX_T[(uint8_t)(X >> 8)], 16) ^
- rotl(SM4_SBOX_T[(uint8_t)X], 8);
+ return SM4_SBOX_T0[(uint8_t)(X >> 24)] ^
+ SM4_SBOX_T1[(uint8_t)(X >> 16)] ^
+ SM4_SBOX_T2[(uint8_t)(X >> 8)] ^
+ SM4_SBOX_T3[(uint8_t)X];
}
int ossl_sm4_set_key(const uint8_t *key, SM4_KEY *ks)