/* * (C) 2026 Jack Lloyd * * Botan is released under the Simplified BSD License (see license.txt) */ #include #include #include #include namespace Botan { namespace SEED_AVX512_GFNI { namespace { BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI SIMD_16x32 seed_g(const SIMD_16x32& X) { /* * SEED's two sboxes are both based on inversions in GF(2^8) modulo the polynomial * x^8+x^6+x^5+x+1 (0x163), followed by different affine transforms. * * GFNI uses AES's field (modulo 0x11B) so the pre-inversion matrix is a field isomorphism * that maps the inputs into the AES field. The post-inversion matrices then apply map * back to SEED's field and apply the appropriate linear transform. */ // Field isomorphism from SEED's field to AES field constexpr uint64_t seed_pre_a = gfni_matrix(R"( 1 1 0 1 0 0 0 0 0 0 1 1 0 0 1 1 0 0 0 0 1 1 0 1 0 1 1 1 0 1 0 0 0 1 1 0 1 0 0 0 0 0 0 1 1 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0 1 1 1 0 )"); // Field isomorphism from AES->SEED multiplied by S0's affine matrix constexpr uint64_t seed_s0_post_a = gfni_matrix(R"( 0 1 0 1 1 0 0 1 0 0 1 1 1 0 1 0 1 0 0 0 1 1 1 0 1 1 0 0 1 0 0 1 0 1 0 1 1 0 1 1 1 1 1 1 1 0 1 1 0 0 1 1 0 1 0 1 0 0 0 1 0 1 1 1 )"); // Field isomorphism from AES->SEED multiplied by S1's affine matrix constexpr uint64_t seed_s1_post_a = gfni_matrix(R"( 0 0 1 1 0 1 1 0 0 1 1 0 0 0 1 0 0 1 0 1 1 0 1 1 0 0 0 0 0 0 1 1 1 1 0 1 0 0 0 0 0 1 0 0 1 0 1 1 1 1 1 0 1 0 1 1 1 1 1 1 0 0 0 1 )"); constexpr uint8_t seed_s0_post_c = 0xA9; constexpr uint8_t seed_s1_post_c = 0x38; // Compute S0(x) and S1(x) for all bytes const auto pre = gf2p8affine(X); const auto s0 = gf2p8affineinv(pre); const auto s1 = gf2p8affineinv(pre); // Blend S0/S1 outputs by alternating bytes constexpr uint64_t blend_mask = 0xAAAAAAAAAAAAAAAA; // 0b1010.... const auto sbox = SIMD_16x32(_mm512_mask_blend_epi8(blend_mask, s0.raw(), s1.raw())); // Linear mixing layer const auto M0 = SIMD_16x32::splat(0x3FCFF3FC); const auto M1 = SIMD_16x32::splat(0xFC3FCFF3); const auto M2 = SIMD_16x32::splat(0xF3FC3FCF); const auto M3 = SIMD_16x32::splat(0xCFF3FC3F); // Masks for broadcasting each byte across the 32 bit word that contains it // clang-format off alignas(64) constexpr uint8_t SHUF_BYTE0[64] = { 0, 0, 0, 0, 4, 4, 4, 4, 8, 8, 8, 8, 12, 12, 12, 12, 0, 0, 0, 0, 4, 4, 4, 4, 8, 8, 8, 8, 12, 12, 12, 12, 0, 0, 0, 0, 4, 4, 4, 4, 8, 8, 8, 8, 12, 12, 12, 12, 0, 0, 0, 0, 4, 4, 4, 4, 8, 8, 8, 8, 12, 12, 12, 12, }; alignas(64) constexpr uint8_t SHUF_BYTE1[64] = { 1, 1, 1, 1, 5, 5, 5, 5, 9, 9, 9, 9, 13, 13, 13, 13, 1, 1, 1, 1, 5, 5, 5, 5, 9, 9, 9, 9, 13, 13, 13, 13, 1, 1, 1, 1, 5, 5, 5, 5, 9, 9, 9, 9, 13, 13, 13, 13, 1, 1, 1, 1, 5, 5, 5, 5, 9, 9, 9, 9, 13, 13, 13, 13, }; alignas(64) constexpr uint8_t SHUF_BYTE2[64] = { 2, 2, 2, 2, 6, 6, 6, 6, 10, 10, 10, 10, 14, 14, 14, 14, 2, 2, 2, 2, 6, 6, 6, 6, 10, 10, 10, 10, 14, 14, 14, 14, 2, 2, 2, 2, 6, 6, 6, 6, 10, 10, 10, 10, 14, 14, 14, 14, 2, 2, 2, 2, 6, 6, 6, 6, 10, 10, 10, 10, 14, 14, 14, 14, }; alignas(64) constexpr uint8_t SHUF_BYTE3[64] = { 3, 3, 3, 3, 7, 7, 7, 7, 11, 11, 11, 11, 15, 15, 15, 15, 3, 3, 3, 3, 7, 7, 7, 7, 11, 11, 11, 11, 15, 15, 15, 15, 3, 3, 3, 3, 7, 7, 7, 7, 11, 11, 11, 11, 15, 15, 15, 15, 3, 3, 3, 3, 7, 7, 7, 7, 11, 11, 11, 11, 15, 15, 15, 15, }; // clang-format on const auto b0 = SIMD_16x32(_mm512_shuffle_epi8(sbox.raw(), _mm512_load_si512(SHUF_BYTE0))); const auto b1 = SIMD_16x32(_mm512_shuffle_epi8(sbox.raw(), _mm512_load_si512(SHUF_BYTE1))); const auto b2 = SIMD_16x32(_mm512_shuffle_epi8(sbox.raw(), _mm512_load_si512(SHUF_BYTE2))); const auto b3 = SIMD_16x32(_mm512_shuffle_epi8(sbox.raw(), _mm512_load_si512(SHUF_BYTE3))); // Return (b0 & M0) ^ (b1 & M1) ^ (b2 & M2) ^ (b3 & M3) // ternlogd 0x78 is a ^ (b & c) auto result = SIMD_16x32(b0) & M0; result = SIMD_16x32::ternary_fn<0x78>(result, b1, M1); result = SIMD_16x32::ternary_fn<0x78>(result, b2, M2); result = SIMD_16x32::ternary_fn<0x78>(result, b3, M3); return SIMD_16x32(result); } BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI void seed_round( SIMD_16x32& B0, SIMD_16x32& B1, SIMD_16x32& B2, SIMD_16x32& B3, uint32_t K0, uint32_t K1, uint32_t K2, uint32_t K3) { auto T0 = B2 ^ SIMD_16x32::splat(K0); auto T1 = seed_g(B2 ^ B3 ^ SIMD_16x32::splat(K1)); T0 = seed_g(T1 + T0); T1 = seed_g(T1 + T0); B1 ^= T1; B0 ^= T0 + T1; T0 = B0 ^ SIMD_16x32::splat(K2); T1 = seed_g(B0 ^ B1 ^ SIMD_16x32::splat(K3)); T0 = seed_g(T1 + T0); T1 = seed_g(T1 + T0); B3 ^= T1; B2 ^= T0 + T1; } BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI void encrypt(const uint8_t ptext[16 * 4 * 4], uint8_t ctext[16 * 4 * 4], std::span RK) { SIMD_16x32 B0 = SIMD_16x32::load_be(ptext + 16 * 4 * 0); SIMD_16x32 B1 = SIMD_16x32::load_be(ptext + 16 * 4 * 1); SIMD_16x32 B2 = SIMD_16x32::load_be(ptext + 16 * 4 * 2); SIMD_16x32 B3 = SIMD_16x32::load_be(ptext + 16 * 4 * 3); SIMD_16x32::transpose(B0, B1, B2, B3); for(size_t j = 0; j != 8; ++j) { const uint32_t K0 = RK[4 * j]; const uint32_t K1 = RK[4 * j + 1]; const uint32_t K2 = RK[4 * j + 2]; const uint32_t K3 = RK[4 * j + 3]; seed_round(B0, B1, B2, B3, K0, K1, K2, K3); } // Output order is B2, B3, B0, B1 SIMD_16x32::transpose(B2, B3, B0, B1); B2.store_be(ctext + 16 * 4 * 0); B3.store_be(ctext + 16 * 4 * 1); B0.store_be(ctext + 16 * 4 * 2); B1.store_be(ctext + 16 * 4 * 3); } BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI void decrypt(const uint8_t ctext[16 * 4 * 4], uint8_t ptext[16 * 4 * 4], std::span RK) { SIMD_16x32 B0 = SIMD_16x32::load_be(ctext + 16 * 4 * 0); SIMD_16x32 B1 = SIMD_16x32::load_be(ctext + 16 * 4 * 1); SIMD_16x32 B2 = SIMD_16x32::load_be(ctext + 16 * 4 * 2); SIMD_16x32 B3 = SIMD_16x32::load_be(ctext + 16 * 4 * 3); SIMD_16x32::transpose(B0, B1, B2, B3); for(size_t j = 0; j != 8; ++j) { const uint32_t K0 = RK[30 - 4 * j]; const uint32_t K1 = RK[31 - 4 * j]; const uint32_t K2 = RK[28 - 4 * j]; const uint32_t K3 = RK[29 - 4 * j]; seed_round(B0, B1, B2, B3, K0, K1, K2, K3); } SIMD_16x32::transpose(B2, B3, B0, B1); B2.store_be(ptext + 16 * 4 * 0); B3.store_be(ptext + 16 * 4 * 1); B0.store_be(ptext + 16 * 4 * 2); B1.store_be(ptext + 16 * 4 * 3); } BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI void encrypt_x2(const uint8_t ptext[32 * 4 * 4], uint8_t ctext[32 * 4 * 4], std::span RK) { SIMD_16x32 B0 = SIMD_16x32::load_be(ptext + 16 * 4 * 0); SIMD_16x32 B1 = SIMD_16x32::load_be(ptext + 16 * 4 * 1); SIMD_16x32 B2 = SIMD_16x32::load_be(ptext + 16 * 4 * 2); SIMD_16x32 B3 = SIMD_16x32::load_be(ptext + 16 * 4 * 3); SIMD_16x32 B4 = SIMD_16x32::load_be(ptext + 16 * 4 * 4); SIMD_16x32 B5 = SIMD_16x32::load_be(ptext + 16 * 4 * 5); SIMD_16x32 B6 = SIMD_16x32::load_be(ptext + 16 * 4 * 6); SIMD_16x32 B7 = SIMD_16x32::load_be(ptext + 16 * 4 * 7); SIMD_16x32::transpose(B0, B1, B2, B3); SIMD_16x32::transpose(B4, B5, B6, B7); for(size_t j = 0; j != 8; ++j) { const uint32_t K0 = RK[4 * j]; const uint32_t K1 = RK[4 * j + 1]; const uint32_t K2 = RK[4 * j + 2]; const uint32_t K3 = RK[4 * j + 3]; seed_round(B0, B1, B2, B3, K0, K1, K2, K3); seed_round(B4, B5, B6, B7, K0, K1, K2, K3); } SIMD_16x32::transpose(B2, B3, B0, B1); SIMD_16x32::transpose(B6, B7, B4, B5); B2.store_be(ctext + 16 * 4 * 0); B3.store_be(ctext + 16 * 4 * 1); B0.store_be(ctext + 16 * 4 * 2); B1.store_be(ctext + 16 * 4 * 3); B6.store_be(ctext + 16 * 4 * 4); B7.store_be(ctext + 16 * 4 * 5); B4.store_be(ctext + 16 * 4 * 6); B5.store_be(ctext + 16 * 4 * 7); } BOTAN_FORCE_INLINE BOTAN_FN_ISA_AVX512_GFNI void decrypt_x2(const uint8_t ctext[32 * 4 * 4], uint8_t ptext[32 * 4 * 4], std::span RK) { SIMD_16x32 B0 = SIMD_16x32::load_be(ctext + 16 * 4 * 0); SIMD_16x32 B1 = SIMD_16x32::load_be(ctext + 16 * 4 * 1); SIMD_16x32 B2 = SIMD_16x32::load_be(ctext + 16 * 4 * 2); SIMD_16x32 B3 = SIMD_16x32::load_be(ctext + 16 * 4 * 3); SIMD_16x32 B4 = SIMD_16x32::load_be(ctext + 16 * 4 * 4); SIMD_16x32 B5 = SIMD_16x32::load_be(ctext + 16 * 4 * 5); SIMD_16x32 B6 = SIMD_16x32::load_be(ctext + 16 * 4 * 6); SIMD_16x32 B7 = SIMD_16x32::load_be(ctext + 16 * 4 * 7); SIMD_16x32::transpose(B0, B1, B2, B3); SIMD_16x32::transpose(B4, B5, B6, B7); for(size_t j = 0; j != 8; ++j) { const uint32_t K0 = RK[30 - 4 * j]; const uint32_t K1 = RK[31 - 4 * j]; const uint32_t K2 = RK[28 - 4 * j]; const uint32_t K3 = RK[29 - 4 * j]; seed_round(B0, B1, B2, B3, K0, K1, K2, K3); seed_round(B4, B5, B6, B7, K0, K1, K2, K3); } SIMD_16x32::transpose(B2, B3, B0, B1); SIMD_16x32::transpose(B6, B7, B4, B5); B2.store_be(ptext + 16 * 4 * 0); B3.store_be(ptext + 16 * 4 * 1); B0.store_be(ptext + 16 * 4 * 2); B1.store_be(ptext + 16 * 4 * 3); B6.store_be(ptext + 16 * 4 * 4); B7.store_be(ptext + 16 * 4 * 5); B4.store_be(ptext + 16 * 4 * 6); B5.store_be(ptext + 16 * 4 * 7); } } // namespace } // namespace SEED_AVX512_GFNI void BOTAN_FN_ISA_AVX512_GFNI SEED::avx512_gfni_encrypt(const uint8_t ptext[], uint8_t ctext[], size_t blocks) const { while(blocks >= 32) { SEED_AVX512_GFNI::encrypt_x2(ptext, ctext, m_K); ptext += 16 * 32; ctext += 16 * 32; blocks -= 32; } while(blocks >= 16) { SEED_AVX512_GFNI::encrypt(ptext, ctext, m_K); ptext += 16 * 16; ctext += 16 * 16; blocks -= 16; } if(blocks > 0) { BOTAN_ASSERT_NOMSG(blocks < 16); uint8_t pbuf[16 * 16] = {0}; uint8_t cbuf[16 * 16] = {0}; copy_mem(pbuf, ptext, blocks * 16); SEED_AVX512_GFNI::encrypt(pbuf, cbuf, m_K); copy_mem(ctext, cbuf, blocks * 16); } } void BOTAN_FN_ISA_AVX512_GFNI SEED::avx512_gfni_decrypt(const uint8_t ctext[], uint8_t ptext[], size_t blocks) const { while(blocks >= 32) { SEED_AVX512_GFNI::decrypt_x2(ctext, ptext, m_K); ptext += 16 * 32; ctext += 16 * 32; blocks -= 32; } while(blocks >= 16) { SEED_AVX512_GFNI::decrypt(ctext, ptext, m_K); ptext += 16 * 16; ctext += 16 * 16; blocks -= 16; } if(blocks > 0) { BOTAN_ASSERT_NOMSG(blocks < 16); uint8_t pbuf[16 * 16] = {0}; uint8_t cbuf[16 * 16] = {0}; copy_mem(cbuf, ctext, blocks * 16); SEED_AVX512_GFNI::decrypt(cbuf, pbuf, m_K); copy_mem(ptext, pbuf, blocks * 16); } } } // namespace Botan