diff options
| author | lloyd <[email protected]> | 2009-10-29 07:54:52 +0000 |
|---|---|---|
| committer | lloyd <[email protected]> | 2009-10-29 07:54:52 +0000 |
| commit | 077ad13c6f97ab79ecb585c0f584bb8963096098 (patch) | |
| tree | 6129c91e2944f831098ae7cbf52e85ff32e7620c | |
| parent | 0cf1ad7aca8506bbd541641cb1ebc0c73c62bcd2 (diff) | |
| parent | 1419961fbc1ef832940d0f5184c3d71797270ef9 (diff) | |
propagate from branch 'net.randombit.botan' (head 4fd7eb9630271d3c1dfed21987ef864680d4ce7b)
to branch 'net.randombit.botan.general-simd' (head 91df868149cdc4754d340e6103028acc82182609)
30 files changed, 1723 insertions, 964 deletions
diff --git a/doc/examples/cpuid.cpp b/doc/examples/cpuid.cpp index 59940b500..1bdee787c 100644 --- a/doc/examples/cpuid.cpp +++ b/doc/examples/cpuid.cpp @@ -12,4 +12,6 @@ int main() printf("SSSE3 %d\n", CPUID::has_ssse3()); printf("SSE41 %d\n", CPUID::has_sse41()); printf("SSE42 %d\n", CPUID::has_sse42()); + + printf("AltiVec %d\n", CPUID::has_altivec()); } diff --git a/doc/log.txt b/doc/log.txt index 1d219eb1c..84379c3e8 100644 --- a/doc/log.txt +++ b/doc/log.txt @@ -1,5 +1,7 @@ * 1.9.2-dev, ????-??-?? + - Add SIMD version of XTEA + - Support both SSE2 and AltiVec SIMD for Serpent and XTEA * 1.9.1, 2009-10-23 - Better support for Python and Perl wrappers diff --git a/src/algo_factory/prov_weight.cpp b/src/algo_factory/prov_weight.cpp index a55a8b1e6..d7e84a323 100644 --- a/src/algo_factory/prov_weight.cpp +++ b/src/algo_factory/prov_weight.cpp @@ -22,7 +22,7 @@ u32bit static_provider_weight(const std::string& prov_name) if(prov_name == "core") return 5; if(prov_name == "ia32") return 6; if(prov_name == "amd64") return 7; - if(prov_name == "sse2") return 8; + if(prov_name == "simd") return 8; if(prov_name == "openssl") return 2; if(prov_name == "gmp") return 1; diff --git a/src/block/serpent_simd/info.txt b/src/block/serpent_simd/info.txt new file mode 100644 index 000000000..d65b41235 --- /dev/null +++ b/src/block/serpent_simd/info.txt @@ -0,0 +1,7 @@ +define SERPENT_SIMD + +<requires> +serpent +simd_32 +simd_engine +</requires> diff --git a/src/block/serpent_sse2/serp_sse2.cpp b/src/block/serpent_simd/serp_simd.cpp index c51bb69ab..b394b0c26 100644 --- a/src/block/serpent_sse2/serp_sse2.cpp +++ b/src/block/serpent_simd/serp_simd.cpp @@ -1,99 +1,71 @@ /* -* Serpent (SSE2) +* Serpent (SIMD) * (C) 2009 Jack Lloyd * * Distributed under the terms of the Botan license */ -#include <botan/serp_sse2.h> -#include <botan/serp_sse2_sbox.h> +#include <botan/serp_simd.h> +#include <botan/serp_simd_sbox.h> +#include <botan/simd_32.h> #include <botan/loadstor.h> -#include <emmintrin.h> namespace Botan { namespace { -#define key_xor(round, B0, B1, B2, B3) \ - do { \ - __m128i key = _mm_loadu_si128(keys + round); \ - B0 = _mm_xor_si128(B0, _mm_shuffle_epi32(key, _MM_SHUFFLE(0,0,0,0))); \ - B1 = _mm_xor_si128(B1, _mm_shuffle_epi32(key, _MM_SHUFFLE(1,1,1,1))); \ - B2 = _mm_xor_si128(B2, _mm_shuffle_epi32(key, _MM_SHUFFLE(2,2,2,2))); \ - B3 = _mm_xor_si128(B3, _mm_shuffle_epi32(key, _MM_SHUFFLE(3,3,3,3))); \ +#define key_xor(round, B0, B1, B2, B3) \ + do { \ + B0 ^= SIMD_32(keys[4*round ]); \ + B1 ^= SIMD_32(keys[4*round+1]); \ + B2 ^= SIMD_32(keys[4*round+2]); \ + B3 ^= SIMD_32(keys[4*round+3]); \ } while(0); /* * Serpent's linear transformations */ -#define rotate_left_m128(vec, rot) \ - _mm_or_si128(_mm_slli_epi32(vec, rot), _mm_srli_epi32(vec, 32-rot)) - -#define rotate_right_m128(vec, rot) \ - _mm_or_si128(_mm_srli_epi32(vec, rot), _mm_slli_epi32(vec, 32-rot)) - -#define transform(B0, B1, B2, B3) \ - do { \ - B0 = rotate_left_m128(B0, 13); \ - B2 = rotate_left_m128(B2, 3); \ - B1 = _mm_xor_si128(B1, _mm_xor_si128(B0, B2)); \ - B3 = _mm_xor_si128(B3, _mm_xor_si128(B2, _mm_slli_epi32(B0, 3))); \ - B1 = rotate_left_m128(B1, 1); \ - B3 = rotate_left_m128(B3, 7); \ - B0 = _mm_xor_si128(B0, _mm_xor_si128(B1, B3)); \ - B2 = _mm_xor_si128(B2, _mm_xor_si128(B3, _mm_slli_epi32(B1, 7))); \ - B0 = rotate_left_m128(B0, 5); \ - B2 = rotate_left_m128(B2, 22); \ +#define transform(B0, B1, B2, B3) \ + do { \ + B0.rotate_left(13); \ + B2.rotate_left(3); \ + B1 ^= B0 ^ B2; \ + B3 ^= B2 ^ (B0 << 3); \ + B1.rotate_left(1); \ + B3.rotate_left(7); \ + B0 ^= B1 ^ B3; \ + B2 ^= B3 ^ (B1 << 7); \ + B0.rotate_left(5); \ + B2.rotate_left(22); \ } while(0); -#define i_transform(B0, B1, B2, B3) \ - do { \ - B2 = rotate_right_m128(B2, 22); \ - B0 = rotate_right_m128(B0, 5); \ - B2 = _mm_xor_si128(B2, _mm_xor_si128(B3, _mm_slli_epi32(B1, 7))); \ - B0 = _mm_xor_si128(B0, _mm_xor_si128(B1, B3)); \ - B3 = rotate_right_m128(B3, 7); \ - B1 = rotate_right_m128(B1, 1); \ - B3 = _mm_xor_si128(B3, _mm_xor_si128(B2, _mm_slli_epi32(B0, 3))); \ - B1 = _mm_xor_si128(B1, _mm_xor_si128(B0, B2)); \ - B2 = rotate_right_m128(B2, 3); \ - B0 = rotate_right_m128(B0, 13); \ +#define i_transform(B0, B1, B2, B3) \ + do { \ + B2.rotate_right(22); \ + B0.rotate_right(5); \ + B2 ^= B3 ^ (B1 << 7); \ + B0 ^= B1 ^ B3; \ + B3.rotate_right(7); \ + B1.rotate_right(1); \ + B3 ^= B2 ^ (B0 << 3); \ + B1 ^= B0 ^ B2; \ + B2.rotate_right(3); \ + B0.rotate_right(13); \ } while(0); /* -* 4x4 SSE2 integer matrix transpose -*/ -#define transpose(B0, B1, B2, B3) \ - do { \ - __m128i T0 = _mm_unpacklo_epi32(B0, B1); \ - __m128i T1 = _mm_unpacklo_epi32(B2, B3); \ - __m128i T2 = _mm_unpackhi_epi32(B0, B1); \ - __m128i T3 = _mm_unpackhi_epi32(B2, B3); \ - B0 = _mm_unpacklo_epi64(T0, T1); \ - B1 = _mm_unpackhi_epi64(T0, T1); \ - B2 = _mm_unpacklo_epi64(T2, T3); \ - B3 = _mm_unpackhi_epi64(T2, T3); \ - } while(0); - -/* -* SSE2 Serpent Encryption of 4 blocks in parallel +* SIMD Serpent Encryption of 4 blocks in parallel */ void serpent_encrypt_4(const byte in[64], byte out[64], - const u32bit keys_32[132]) + const u32bit keys[132]) { - const __m128i all_ones = _mm_set1_epi8(0xFF); + SIMD_32 B0 = SIMD_32::load_le(in); + SIMD_32 B1 = SIMD_32::load_le(in + 16); + SIMD_32 B2 = SIMD_32::load_le(in + 32); + SIMD_32 B3 = SIMD_32::load_le(in + 48); - const __m128i* keys = (const __m128i*)(keys_32); - __m128i* out_mm = (__m128i*)(out); - __m128i* in_mm = (__m128i*)(in); - - __m128i B0 = _mm_loadu_si128(in_mm); - __m128i B1 = _mm_loadu_si128(in_mm + 1); - __m128i B2 = _mm_loadu_si128(in_mm + 2); - __m128i B3 = _mm_loadu_si128(in_mm + 3); - - transpose(B0, B1, B2, B3); + SIMD_32::transpose(B0, B1, B2, B3); key_xor( 0,B0,B1,B2,B3); SBoxE1(B0,B1,B2,B3); transform(B0,B1,B2,B3); key_xor( 1,B0,B1,B2,B3); SBoxE2(B0,B1,B2,B3); transform(B0,B1,B2,B3); @@ -131,33 +103,27 @@ void serpent_encrypt_4(const byte in[64], key_xor(30,B0,B1,B2,B3); SBoxE7(B0,B1,B2,B3); transform(B0,B1,B2,B3); key_xor(31,B0,B1,B2,B3); SBoxE8(B0,B1,B2,B3); key_xor(32,B0,B1,B2,B3); - transpose(B0, B1, B2, B3); + SIMD_32::transpose(B0, B1, B2, B3); - _mm_storeu_si128(out_mm , B0); - _mm_storeu_si128(out_mm + 1, B1); - _mm_storeu_si128(out_mm + 2, B2); - _mm_storeu_si128(out_mm + 3, B3); + B0.store_le(out); + B1.store_le(out + 16); + B2.store_le(out + 32); + B3.store_le(out + 48); } /* -* SSE2 Serpent Decryption of 4 blocks in parallel +* SIMD Serpent Decryption of 4 blocks in parallel */ void serpent_decrypt_4(const byte in[64], byte out[64], - const u32bit keys_32[132]) + const u32bit keys[132]) { - const __m128i all_ones = _mm_set1_epi8(0xFF); - - const __m128i* keys = (const __m128i*)(keys_32); - __m128i* out_mm = (__m128i*)(out); - __m128i* in_mm = (__m128i*)(in); - - __m128i B0 = _mm_loadu_si128(in_mm); - __m128i B1 = _mm_loadu_si128(in_mm + 1); - __m128i B2 = _mm_loadu_si128(in_mm + 2); - __m128i B3 = _mm_loadu_si128(in_mm + 3); + SIMD_32 B0 = SIMD_32::load_le(in); + SIMD_32 B1 = SIMD_32::load_le(in + 16); + SIMD_32 B2 = SIMD_32::load_le(in + 32); + SIMD_32 B3 = SIMD_32::load_le(in + 48); - transpose(B0, B1, B2, B3); + SIMD_32::transpose(B0, B1, B2, B3); key_xor(32,B0,B1,B2,B3); SBoxD8(B0,B1,B2,B3); key_xor(31,B0,B1,B2,B3); i_transform(B0,B1,B2,B3); SBoxD7(B0,B1,B2,B3); key_xor(30,B0,B1,B2,B3); @@ -195,12 +161,12 @@ void serpent_decrypt_4(const byte in[64], i_transform(B0,B1,B2,B3); SBoxD2(B0,B1,B2,B3); key_xor( 1,B0,B1,B2,B3); i_transform(B0,B1,B2,B3); SBoxD1(B0,B1,B2,B3); key_xor( 0,B0,B1,B2,B3); - transpose(B0, B1, B2, B3); + SIMD_32::transpose(B0, B1, B2, B3); - _mm_storeu_si128(out_mm , B0); - _mm_storeu_si128(out_mm + 1, B1); - _mm_storeu_si128(out_mm + 2, B2); - _mm_storeu_si128(out_mm + 3, B3); + B0.store_le(out); + B1.store_le(out + 16); + B2.store_le(out + 32); + B3.store_le(out + 48); } } @@ -208,7 +174,7 @@ void serpent_decrypt_4(const byte in[64], /* * Serpent Encryption */ -void Serpent_SSE2::encrypt_n(const byte in[], byte out[], u32bit blocks) const +void Serpent_SIMD::encrypt_n(const byte in[], byte out[], u32bit blocks) const { while(blocks >= 4) { @@ -224,7 +190,7 @@ void Serpent_SSE2::encrypt_n(const byte in[], byte out[], u32bit blocks) const /* * Serpent Decryption */ -void Serpent_SSE2::decrypt_n(const byte in[], byte out[], u32bit blocks) const +void Serpent_SIMD::decrypt_n(const byte in[], byte out[], u32bit blocks) const { while(blocks >= 4) { diff --git a/src/block/serpent_sse2/serp_sse2.h b/src/block/serpent_simd/serp_simd.h index f1e5c2028..1ecb70159 100644 --- a/src/block/serpent_sse2/serp_sse2.h +++ b/src/block/serpent_simd/serp_simd.h @@ -1,12 +1,12 @@ /* -* Serpent (SSE2) +* Serpent (SIMD) * (C) 2009 Jack Lloyd * * Distributed under the terms of the Botan license */ -#ifndef BOTAN_SERPENT_SSE2_H__ -#define BOTAN_SERPENT_SSE2_H__ +#ifndef BOTAN_SERPENT_SIMD_H__ +#define BOTAN_SERPENT_SIMD_H__ #include <botan/serpent.h> @@ -15,13 +15,13 @@ namespace Botan { /* * Serpent */ -class BOTAN_DLL Serpent_SSE2 : public Serpent +class BOTAN_DLL Serpent_SIMD : public Serpent { public: void encrypt_n(const byte in[], byte out[], u32bit blocks) const; void decrypt_n(const byte in[], byte out[], u32bit blocks) const; - BlockCipher* clone() const { return new Serpent_SSE2; } + BlockCipher* clone() const { return new Serpent_SIMD; } }; } diff --git a/src/block/serpent_simd/serp_simd_sbox.h b/src/block/serpent_simd/serp_simd_sbox.h new file mode 100644 index 000000000..6e3da7359 --- /dev/null +++ b/src/block/serpent_simd/serp_simd_sbox.h @@ -0,0 +1,426 @@ +/* +* Serpent Sboxes in SIMD form +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef SERPENT_SIMD_SBOXES_H__ +#define SERPENT_SIMD_SBOXES_H__ + +#define SBoxE1(B0, B1, B2, B3) \ + do { \ + B3 ^= B0; \ + SIMD_32 B4 = B1; \ + B1 &= B3; \ + B4 ^= B2; \ + B1 ^= B0; \ + B0 |= B3; \ + B0 ^= B4; \ + B4 ^= B3; \ + B3 ^= B2; \ + B2 |= B1; \ + B2 ^= B4; \ + B4 = ~B4; \ + B4 |= B1; \ + B1 ^= B3; \ + B1 ^= B4; \ + B3 |= B0; \ + B1 ^= B3; \ + B4 ^= B3; \ + B3 = B0; \ + B0 = B1; \ + B1 = B4; \ + } while(0); + +#define SBoxE2(B0, B1, B2, B3) \ + do { \ + B0 = ~B0; \ + B2 = ~B2; \ + SIMD_32 B4 = B0; \ + B0 &= B1; \ + B2 ^= B0; \ + B0 |= B3; \ + B3 ^= B2; \ + B1 ^= B0; \ + B0 ^= B4; \ + B4 |= B1; \ + B1 ^= B3; \ + B2 |= B0; \ + B2 &= B4; \ + B0 ^= B1; \ + B1 &= B2; \ + B1 ^= B0; \ + B0 &= B2; \ + B4 ^= B0; \ + B0 = B2; \ + B2 = B3; \ + B3 = B1; \ + B1 = B4; \ + } while(0); + +#define SBoxE3(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B0; \ + B0 &= B2; \ + B0 ^= B3; \ + B2 ^= B1; \ + B2 ^= B0; \ + B3 |= B4; \ + B3 ^= B1; \ + B4 ^= B2; \ + B1 = B3; \ + B3 |= B4; \ + B3 ^= B0; \ + B0 &= B1; \ + B4 ^= B0; \ + B1 ^= B3; \ + B1 ^= B4; \ + B4 = ~B4; \ + B0 = B2; \ + B2 = B1; \ + B1 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxE4(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B0; \ + B0 |= B3; \ + B3 ^= B1; \ + B1 &= B4; \ + B4 ^= B2; \ + B2 ^= B3; \ + B3 &= B0; \ + B4 |= B1; \ + B3 ^= B4; \ + B0 ^= B1; \ + B4 &= B0; \ + B1 ^= B3; \ + B4 ^= B2; \ + B1 |= B0; \ + B1 ^= B2; \ + B0 ^= B3; \ + B2 = B1; \ + B1 |= B3; \ + B0 ^= B1; \ + B1 = B2; \ + B2 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxE5(B0, B1, B2, B3) \ + do { \ + B1 ^= B3; \ + B3 = ~B3; \ + B2 ^= B3; \ + B3 ^= B0; \ + SIMD_32 B4 = B1; \ + B1 &= B3; \ + B1 ^= B2; \ + B4 ^= B3; \ + B0 ^= B4; \ + B2 &= B4; \ + B2 ^= B0; \ + B0 &= B1; \ + B3 ^= B0; \ + B4 |= B1; \ + B4 ^= B0; \ + B0 |= B3; \ + B0 ^= B2; \ + B2 &= B3; \ + B0 = ~B0; \ + B4 ^= B2; \ + B2 = B0; \ + B0 = B1; \ + B1 = B4; \ + } while(0); + +#define SBoxE6(B0, B1, B2, B3) \ + do { \ + B0 ^= B1; \ + B1 ^= B3; \ + B3 = ~B3; \ + SIMD_32 B4 = B1; \ + B1 &= B0; \ + B2 ^= B3; \ + B1 ^= B2; \ + B2 |= B4; \ + B4 ^= B3; \ + B3 &= B1; \ + B3 ^= B0; \ + B4 ^= B1; \ + B4 ^= B2; \ + B2 ^= B0; \ + B0 &= B3; \ + B2 = ~B2; \ + B0 ^= B4; \ + B4 |= B3; \ + B4 ^= B2; \ + B2 = B0; \ + B0 = B1; \ + B1 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxE7(B0, B1, B2, B3) \ + do { \ + B2 = ~B2; \ + SIMD_32 B4 = B3; \ + B3 &= B0; \ + B0 ^= B4; \ + B3 ^= B2; \ + B2 |= B4; \ + B1 ^= B3; \ + B2 ^= B0; \ + B0 |= B1; \ + B2 ^= B1; \ + B4 ^= B0; \ + B0 |= B3; \ + B0 ^= B2; \ + B4 ^= B3; \ + B4 ^= B0; \ + B3 = ~B3; \ + B2 &= B4; \ + B3 ^= B2; \ + B2 = B4; \ + } while(0); + +#define SBoxE8(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B1; \ + B1 |= B2; \ + B1 ^= B3; \ + B4 ^= B2; \ + B2 ^= B1; \ + B3 |= B4; \ + B3 &= B0; \ + B4 ^= B2; \ + B3 ^= B1; \ + B1 |= B4; \ + B1 ^= B0; \ + B0 |= B4; \ + B0 ^= B2; \ + B1 ^= B4; \ + B2 ^= B1; \ + B1 &= B0; \ + B1 ^= B4; \ + B2 = ~B2; \ + B2 |= B0; \ + B4 ^= B2; \ + B2 = B1; \ + B1 = B3; \ + B3 = B0; \ + B0 = B4; \ + } while(0); + +#define SBoxD1(B0, B1, B2, B3) \ + do { \ + B2 = ~B2; \ + SIMD_32 B4 = B1; \ + B1 |= B0; \ + B4 = ~B4; \ + B1 ^= B2; \ + B2 |= B4; \ + B1 ^= B3; \ + B0 ^= B4; \ + B2 ^= B0; \ + B0 &= B3; \ + B4 ^= B0; \ + B0 |= B1; \ + B0 ^= B2; \ + B3 ^= B4; \ + B2 ^= B1; \ + B3 ^= B0; \ + B3 ^= B1; \ + B2 &= B3; \ + B4 ^= B2; \ + B2 = B1; \ + B1 = B4; \ + } while(0); + +#define SBoxD2(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B1; \ + B1 ^= B3; \ + B3 &= B1; \ + B4 ^= B2; \ + B3 ^= B0; \ + B0 |= B1; \ + B2 ^= B3; \ + B0 ^= B4; \ + B0 |= B2; \ + B1 ^= B3; \ + B0 ^= B1; \ + B1 |= B3; \ + B1 ^= B0; \ + B4 = ~B4; \ + B4 ^= B1; \ + B1 |= B0; \ + B1 ^= B0; \ + B1 |= B4; \ + B3 ^= B1; \ + B1 = B0; \ + B0 = B4; \ + B4 = B2; \ + B2 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxD3(B0, B1, B2, B3) \ + do { \ + B2 ^= B3; \ + B3 ^= B0; \ + SIMD_32 B4 = B3; \ + B3 &= B2; \ + B3 ^= B1; \ + B1 |= B2; \ + B1 ^= B4; \ + B4 &= B3; \ + B2 ^= B3; \ + B4 &= B0; \ + B4 ^= B2; \ + B2 &= B1; \ + B2 |= B0; \ + B3 = ~B3; \ + B2 ^= B3; \ + B0 ^= B3; \ + B0 &= B1; \ + B3 ^= B4; \ + B3 ^= B0; \ + B0 = B1; \ + B1 = B4; \ + } while(0); + +#define SBoxD4(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B2; \ + B2 ^= B1; \ + B0 ^= B2; \ + B4 &= B2; \ + B4 ^= B0; \ + B0 &= B1; \ + B1 ^= B3; \ + B3 |= B4; \ + B2 ^= B3; \ + B0 ^= B3; \ + B1 ^= B4; \ + B3 &= B2; \ + B3 ^= B1; \ + B1 ^= B0; \ + B1 |= B2; \ + B0 ^= B3; \ + B1 ^= B4; \ + B0 ^= B1; \ + B4 = B0; \ + B0 = B2; \ + B2 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxD5(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B2; \ + B2 &= B3; \ + B2 ^= B1; \ + B1 |= B3; \ + B1 &= B0; \ + B4 ^= B2; \ + B4 ^= B1; \ + B1 &= B2; \ + B0 = ~B0; \ + B3 ^= B4; \ + B1 ^= B3; \ + B3 &= B0; \ + B3 ^= B2; \ + B0 ^= B1; \ + B2 &= B0; \ + B3 ^= B0; \ + B2 ^= B4; \ + B2 |= B3; \ + B3 ^= B0; \ + B2 ^= B1; \ + B1 = B3; \ + B3 = B4; \ + } while(0); + +#define SBoxD6(B0, B1, B2, B3) \ + do { \ + B1 = ~B1; \ + SIMD_32 B4 = B3; \ + B2 ^= B1; \ + B3 |= B0; \ + B3 ^= B2; \ + B2 |= B1; \ + B2 &= B0; \ + B4 ^= B3; \ + B2 ^= B4; \ + B4 |= B0; \ + B4 ^= B1; \ + B1 &= B2; \ + B1 ^= B3; \ + B4 ^= B2; \ + B3 &= B4; \ + B4 ^= B1; \ + B3 ^= B4; \ + B4 = ~B4; \ + B3 ^= B0; \ + B0 = B1; \ + B1 = B4; \ + B4 = B3; \ + B3 = B2; \ + B2 = B4; \ + } while(0); + +#define SBoxD7(B0, B1, B2, B3) \ + do { \ + B0 ^= B2; \ + SIMD_32 B4 = B2; \ + B2 &= B0; \ + B4 ^= B3; \ + B2 = ~B2; \ + B3 ^= B1; \ + B2 ^= B3; \ + B4 |= B0; \ + B0 ^= B2; \ + B3 ^= B4; \ + B4 ^= B1; \ + B1 &= B3; \ + B1 ^= B0; \ + B0 ^= B3; \ + B0 |= B2; \ + B3 ^= B1; \ + B4 ^= B0; \ + B0 = B1; \ + B1 = B2; \ + B2 = B4; \ + } while(0); + +#define SBoxD8(B0, B1, B2, B3) \ + do { \ + SIMD_32 B4 = B2; \ + B2 ^= B0; \ + B0 &= B3; \ + B4 |= B3; \ + B2 = ~B2; \ + B3 ^= B1; \ + B1 |= B0; \ + B0 ^= B2; \ + B2 &= B4; \ + B3 &= B4; \ + B1 ^= B2; \ + B2 ^= B0; \ + B0 |= B2; \ + B4 ^= B1; \ + B0 ^= B3; \ + B3 ^= B4; \ + B4 |= B0; \ + B3 ^= B2; \ + B4 ^= B2; \ + B2 = B1; \ + B1 = B0; \ + B0 = B3; \ + B3 = B4; \ + } while(0); + +#endif diff --git a/src/block/serpent_sse2/info.txt b/src/block/serpent_sse2/info.txt deleted file mode 100644 index a4ec561a8..000000000 --- a/src/block/serpent_sse2/info.txt +++ /dev/null @@ -1,6 +0,0 @@ -define SERPENT_SSE2 - -<requires> -serpent -sse2_eng -</requires> diff --git a/src/block/serpent_sse2/serp_sse2_sbox.h b/src/block/serpent_sse2/serp_sse2_sbox.h deleted file mode 100644 index 40c552e87..000000000 --- a/src/block/serpent_sse2/serp_sse2_sbox.h +++ /dev/null @@ -1,434 +0,0 @@ -/* -* Serpent Sboxes in SSE2 form -* (C) 2009 Jack Lloyd -* -* Distributed under the terms of the Botan license -*/ - -#ifndef SERPENT_SSE2_SBOXES_H__ -#define SERPENT_SSE2_SBOXES_H__ - -#define SBoxE1(B0, B1, B2, B3) \ - do { \ - B3 = _mm_xor_si128(B3, B0); \ - __m128i B4 = B1; \ - B1 = _mm_and_si128(B1, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B1 = _mm_xor_si128(B1, B0); \ - B0 = _mm_or_si128(B0, B3); \ - B0 = _mm_xor_si128(B0, B4); \ - B4 = _mm_xor_si128(B4, B3); \ - B3 = _mm_xor_si128(B3, B2); \ - B2 = _mm_or_si128(B2, B1); \ - B2 = _mm_xor_si128(B2, B4); \ - B4 = _mm_xor_si128(B4, all_ones); \ - B4 = _mm_or_si128(B4, B1); \ - B1 = _mm_xor_si128(B1, B3); \ - B1 = _mm_xor_si128(B1, B4); \ - B3 = _mm_or_si128(B3, B0); \ - B1 = _mm_xor_si128(B1, B3); \ - B4 = _mm_xor_si128(B4, B3); \ - B3 = B0; \ - B0 = B1; \ - B1 = B4; \ - } while(0); - -#define SBoxE2(B0, B1, B2, B3) \ - do { \ - B0 = _mm_xor_si128(B0, all_ones); \ - B2 = _mm_xor_si128(B2, all_ones); \ - __m128i B4 = B0; \ - B0 = _mm_and_si128(B0, B1); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_or_si128(B0, B3); \ - B3 = _mm_xor_si128(B3, B2); \ - B1 = _mm_xor_si128(B1, B0); \ - B0 = _mm_xor_si128(B0, B4); \ - B4 = _mm_or_si128(B4, B1); \ - B1 = _mm_xor_si128(B1, B3); \ - B2 = _mm_or_si128(B2, B0); \ - B2 = _mm_and_si128(B2, B4); \ - B0 = _mm_xor_si128(B0, B1); \ - B1 = _mm_and_si128(B1, B2); \ - B1 = _mm_xor_si128(B1, B0); \ - B0 = _mm_and_si128(B0, B2); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = B2; \ - B2 = B3; \ - B3 = B1; \ - B1 = B4; \ - } while(0); - -#define SBoxE3(B0, B1, B2, B3) \ - do { \ - __m128i B4 = B0; \ - B0 = _mm_and_si128(B0, B2); \ - B0 = _mm_xor_si128(B0, B3); \ - B2 = _mm_xor_si128(B2, B1); \ - B2 = _mm_xor_si128(B2, B0); \ - B3 = _mm_or_si128(B3, B4); \ - B3 = _mm_xor_si128(B3, B1); \ - B4 = _mm_xor_si128(B4, B2); \ - B1 = B3; \ - B3 = _mm_or_si128(B3, B4); \ - B3 = _mm_xor_si128(B3, B0); \ - B0 = _mm_and_si128(B0, B1); \ - B4 = _mm_xor_si128(B4, B0); \ - B1 = _mm_xor_si128(B1, B3); \ - B1 = _mm_xor_si128(B1, B4); \ - B4 = _mm_xor_si128(B4, all_ones); \ - B0 = B2; \ - B2 = B1; \ - B1 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxE4(B0, B1, B2, B3) \ - do { \ - __m128i B4 = B0; \ - B0 = _mm_or_si128(B0, B3); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = _mm_and_si128(B1, B4); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = _mm_xor_si128(B2, B3); \ - B3 = _mm_and_si128(B3, B0); \ - B4 = _mm_or_si128(B4, B1); \ - B3 = _mm_xor_si128(B3, B4); \ - B0 = _mm_xor_si128(B0, B1); \ - B4 = _mm_and_si128(B4, B0); \ - B1 = _mm_xor_si128(B1, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B1 = _mm_or_si128(B1, B0); \ - B1 = _mm_xor_si128(B1, B2); \ - B0 = _mm_xor_si128(B0, B3); \ - B2 = B1; \ - B1 = _mm_or_si128(B1, B3); \ - B0 = _mm_xor_si128(B0, B1); \ - B1 = B2; \ - B2 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxE5(B0, B1, B2, B3) \ - do { \ - B1 = _mm_xor_si128(B1, B3); \ - B3 = _mm_xor_si128(B3, all_ones); \ - B2 = _mm_xor_si128(B2, B3); \ - B3 = _mm_xor_si128(B3, B0); \ - __m128i B4 = B1; \ - B1 = _mm_and_si128(B1, B3); \ - B1 = _mm_xor_si128(B1, B2); \ - B4 = _mm_xor_si128(B4, B3); \ - B0 = _mm_xor_si128(B0, B4); \ - B2 = _mm_and_si128(B2, B4); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_and_si128(B0, B1); \ - B3 = _mm_xor_si128(B3, B0); \ - B4 = _mm_or_si128(B4, B1); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = _mm_or_si128(B0, B3); \ - B0 = _mm_xor_si128(B0, B2); \ - B2 = _mm_and_si128(B2, B3); \ - B0 = _mm_xor_si128(B0, all_ones); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = B0; \ - B0 = B1; \ - B1 = B4; \ - } while(0); - -#define SBoxE6(B0, B1, B2, B3) \ - do { \ - B0 = _mm_xor_si128(B0, B1); \ - B1 = _mm_xor_si128(B1, B3); \ - B3 = _mm_xor_si128(B3, all_ones); \ - __m128i B4 = B1; \ - B1 = _mm_and_si128(B1, B0); \ - B2 = _mm_xor_si128(B2, B3); \ - B1 = _mm_xor_si128(B1, B2); \ - B2 = _mm_or_si128(B2, B4); \ - B4 = _mm_xor_si128(B4, B3); \ - B3 = _mm_and_si128(B3, B1); \ - B3 = _mm_xor_si128(B3, B0); \ - B4 = _mm_xor_si128(B4, B1); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_and_si128(B0, B3); \ - B2 = _mm_xor_si128(B2, all_ones); \ - B0 = _mm_xor_si128(B0, B4); \ - B4 = _mm_or_si128(B4, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = B0; \ - B0 = B1; \ - B1 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxE7(B0, B1, B2, B3) \ - do { \ - B2 = _mm_xor_si128(B2, all_ones); \ - __m128i B4 = B3; \ - B3 = _mm_and_si128(B3, B0); \ - B0 = _mm_xor_si128(B0, B4); \ - B3 = _mm_xor_si128(B3, B2); \ - B2 = _mm_or_si128(B2, B4); \ - B1 = _mm_xor_si128(B1, B3); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_or_si128(B0, B1); \ - B2 = _mm_xor_si128(B2, B1); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = _mm_or_si128(B0, B3); \ - B0 = _mm_xor_si128(B0, B2); \ - B4 = _mm_xor_si128(B4, B3); \ - B4 = _mm_xor_si128(B4, B0); \ - B3 = _mm_xor_si128(B3, all_ones); \ - B2 = _mm_and_si128(B2, B4); \ - B3 = _mm_xor_si128(B3, B2); \ - B2 = B4; \ - } while(0); - -#define SBoxE8(B0, B1, B2, B3) \ - do { \ - __m128i B4 = B1; \ - B1 = _mm_or_si128(B1, B2); \ - B1 = _mm_xor_si128(B1, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = _mm_xor_si128(B2, B1); \ - B3 = _mm_or_si128(B3, B4); \ - B3 = _mm_and_si128(B3, B0); \ - B4 = _mm_xor_si128(B4, B2); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = _mm_or_si128(B1, B4); \ - B1 = _mm_xor_si128(B1, B0); \ - B0 = _mm_or_si128(B0, B4); \ - B0 = _mm_xor_si128(B0, B2); \ - B1 = _mm_xor_si128(B1, B4); \ - B2 = _mm_xor_si128(B2, B1); \ - B1 = _mm_and_si128(B1, B0); \ - B1 = _mm_xor_si128(B1, B4); \ - B2 = _mm_xor_si128(B2, all_ones); \ - B2 = _mm_or_si128(B2, B0); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = B1; \ - B1 = B3; \ - B3 = B0; \ - B0 = B4; \ - } while(0); - -#define SBoxD1(B0, B1, B2, B3) \ - do \ - { \ - B2 = _mm_xor_si128(B2, all_ones); \ - __m128i B4 = B1; \ - B1 = _mm_or_si128(B1, B0); \ - B4 = _mm_xor_si128(B4, all_ones); \ - B1 = _mm_xor_si128(B1, B2); \ - B2 = _mm_or_si128(B2, B4); \ - B1 = _mm_xor_si128(B1, B3); \ - B0 = _mm_xor_si128(B0, B4); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_and_si128(B0, B3); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = _mm_or_si128(B0, B1); \ - B0 = _mm_xor_si128(B0, B2); \ - B3 = _mm_xor_si128(B3, B4); \ - B2 = _mm_xor_si128(B2, B1); \ - B3 = _mm_xor_si128(B3, B0); \ - B3 = _mm_xor_si128(B3, B1); \ - B2 = _mm_and_si128(B2, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = B1; \ - B1 = B4; \ - } while(0); - -#define SBoxD2(B0, B1, B2, B3) \ - do \ - { \ - __m128i B4 = B1; \ - B1 = _mm_xor_si128(B1, B3); \ - B3 = _mm_and_si128(B3, B1); \ - B4 = _mm_xor_si128(B4, B2); \ - B3 = _mm_xor_si128(B3, B0); \ - B0 = _mm_or_si128(B0, B1); \ - B2 = _mm_xor_si128(B2, B3); \ - B0 = _mm_xor_si128(B0, B4); \ - B0 = _mm_or_si128(B0, B2); \ - B1 = _mm_xor_si128(B1, B3); \ - B0 = _mm_xor_si128(B0, B1); \ - B1 = _mm_or_si128(B1, B3); \ - B1 = _mm_xor_si128(B1, B0); \ - B4 = _mm_xor_si128(B4, all_ones); \ - B4 = _mm_xor_si128(B4, B1); \ - B1 = _mm_or_si128(B1, B0); \ - B1 = _mm_xor_si128(B1, B0); \ - B1 = _mm_or_si128(B1, B4); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = B0; \ - B0 = B4; \ - B4 = B2; \ - B2 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxD3(B0, B1, B2, B3) \ - do \ - { \ - B2 = _mm_xor_si128(B2, B3); \ - B3 = _mm_xor_si128(B3, B0); \ - __m128i B4 = B3; \ - B3 = _mm_and_si128(B3, B2); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = _mm_or_si128(B1, B2); \ - B1 = _mm_xor_si128(B1, B4); \ - B4 = _mm_and_si128(B4, B3); \ - B2 = _mm_xor_si128(B2, B3); \ - B4 = _mm_and_si128(B4, B0); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = _mm_and_si128(B2, B1); \ - B2 = _mm_or_si128(B2, B0); \ - B3 = _mm_xor_si128(B3, all_ones); \ - B2 = _mm_xor_si128(B2, B3); \ - B0 = _mm_xor_si128(B0, B3); \ - B0 = _mm_and_si128(B0, B1); \ - B3 = _mm_xor_si128(B3, B4); \ - B3 = _mm_xor_si128(B3, B0); \ - B0 = B1; \ - B1 = B4; \ - } while(0); - -#define SBoxD4(B0, B1, B2, B3) \ - do \ - { \ - __m128i B4 = B2; \ - B2 = _mm_xor_si128(B2, B1); \ - B0 = _mm_xor_si128(B0, B2); \ - B4 = _mm_and_si128(B4, B2); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = _mm_and_si128(B0, B1); \ - B1 = _mm_xor_si128(B1, B3); \ - B3 = _mm_or_si128(B3, B4); \ - B2 = _mm_xor_si128(B2, B3); \ - B0 = _mm_xor_si128(B0, B3); \ - B1 = _mm_xor_si128(B1, B4); \ - B3 = _mm_and_si128(B3, B2); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = _mm_xor_si128(B1, B0); \ - B1 = _mm_or_si128(B1, B2); \ - B0 = _mm_xor_si128(B0, B3); \ - B1 = _mm_xor_si128(B1, B4); \ - B0 = _mm_xor_si128(B0, B1); \ - B4 = B0; \ - B0 = B2; \ - B2 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxD5(B0, B1, B2, B3) \ - do \ - { \ - __m128i B4 = B2; \ - B2 = _mm_and_si128(B2, B3); \ - B2 = _mm_xor_si128(B2, B1); \ - B1 = _mm_or_si128(B1, B3); \ - B1 = _mm_and_si128(B1, B0); \ - B4 = _mm_xor_si128(B4, B2); \ - B4 = _mm_xor_si128(B4, B1); \ - B1 = _mm_and_si128(B1, B2); \ - B0 = _mm_xor_si128(B0, all_ones); \ - B3 = _mm_xor_si128(B3, B4); \ - B1 = _mm_xor_si128(B1, B3); \ - B3 = _mm_and_si128(B3, B0); \ - B3 = _mm_xor_si128(B3, B2); \ - B0 = _mm_xor_si128(B0, B1); \ - B2 = _mm_and_si128(B2, B0); \ - B3 = _mm_xor_si128(B3, B0); \ - B2 = _mm_xor_si128(B2, B4); \ - B2 = _mm_or_si128(B2, B3); \ - B3 = _mm_xor_si128(B3, B0); \ - B2 = _mm_xor_si128(B2, B1); \ - B1 = B3; \ - B3 = B4; \ - } while(0); - -#define SBoxD6(B0, B1, B2, B3) \ - do \ - { \ - B1 = _mm_xor_si128(B1, all_ones); \ - __m128i B4 = B3; \ - B2 = _mm_xor_si128(B2, B1); \ - B3 = _mm_or_si128(B3, B0); \ - B3 = _mm_xor_si128(B3, B2); \ - B2 = _mm_or_si128(B2, B1); \ - B2 = _mm_and_si128(B2, B0); \ - B4 = _mm_xor_si128(B4, B3); \ - B2 = _mm_xor_si128(B2, B4); \ - B4 = _mm_or_si128(B4, B0); \ - B4 = _mm_xor_si128(B4, B1); \ - B1 = _mm_and_si128(B1, B2); \ - B1 = _mm_xor_si128(B1, B3); \ - B4 = _mm_xor_si128(B4, B2); \ - B3 = _mm_and_si128(B3, B4); \ - B4 = _mm_xor_si128(B4, B1); \ - B3 = _mm_xor_si128(B3, B4); \ - B4 = _mm_xor_si128(B4, all_ones); \ - B3 = _mm_xor_si128(B3, B0); \ - B0 = B1; \ - B1 = B4; \ - B4 = B3; \ - B3 = B2; \ - B2 = B4; \ - } while(0); - -#define SBoxD7(B0, B1, B2, B3) \ - do \ - { \ - B0 = _mm_xor_si128(B0, B2); \ - __m128i B4 = B2; \ - B2 = _mm_and_si128(B2, B0); \ - B4 = _mm_xor_si128(B4, B3); \ - B2 = _mm_xor_si128(B2, all_ones); \ - B3 = _mm_xor_si128(B3, B1); \ - B2 = _mm_xor_si128(B2, B3); \ - B4 = _mm_or_si128(B4, B0); \ - B0 = _mm_xor_si128(B0, B2); \ - B3 = _mm_xor_si128(B3, B4); \ - B4 = _mm_xor_si128(B4, B1); \ - B1 = _mm_and_si128(B1, B3); \ - B1 = _mm_xor_si128(B1, B0); \ - B0 = _mm_xor_si128(B0, B3); \ - B0 = _mm_or_si128(B0, B2); \ - B3 = _mm_xor_si128(B3, B1); \ - B4 = _mm_xor_si128(B4, B0); \ - B0 = B1; \ - B1 = B2; \ - B2 = B4; \ - } while(0); - -#define SBoxD8(B0, B1, B2, B3) \ - do \ - { \ - __m128i B4 = B2; \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_and_si128(B0, B3); \ - B4 = _mm_or_si128(B4, B3); \ - B2 = _mm_xor_si128(B2, all_ones); \ - B3 = _mm_xor_si128(B3, B1); \ - B1 = _mm_or_si128(B1, B0); \ - B0 = _mm_xor_si128(B0, B2); \ - B2 = _mm_and_si128(B2, B4); \ - B3 = _mm_and_si128(B3, B4); \ - B1 = _mm_xor_si128(B1, B2); \ - B2 = _mm_xor_si128(B2, B0); \ - B0 = _mm_or_si128(B0, B2); \ - B4 = _mm_xor_si128(B4, B1); \ - B0 = _mm_xor_si128(B0, B3); \ - B3 = _mm_xor_si128(B3, B4); \ - B4 = _mm_or_si128(B4, B0); \ - B3 = _mm_xor_si128(B3, B2); \ - B4 = _mm_xor_si128(B4, B2); \ - B2 = B1; \ - B1 = B0; \ - B0 = B3; \ - B3 = B4; \ - } while(0); - -#endif diff --git a/src/block/xtea/xtea.h b/src/block/xtea/xtea.h index f3b554edb..9982d0712 100644 --- a/src/block/xtea/xtea.h +++ b/src/block/xtea/xtea.h @@ -26,7 +26,7 @@ class BOTAN_DLL XTEA : public BlockCipher BlockCipher* clone() const { return new XTEA; } XTEA() : BlockCipher(8, 16) {} - private: + protected: void key_schedule(const byte[], u32bit); SecureBuffer<u32bit, 64> EK; }; diff --git a/src/block/xtea_simd/info.txt b/src/block/xtea_simd/info.txt new file mode 100644 index 000000000..98a6e941f --- /dev/null +++ b/src/block/xtea_simd/info.txt @@ -0,0 +1,14 @@ +define XTEA_SIMD + +load_on auto + +<add> +xtea_simd.cpp +xtea_simd.h +</add> + +<requires> +xtea +simd_32 +simd_engine +</requires> diff --git a/src/block/xtea_simd/xtea_simd.cpp b/src/block/xtea_simd/xtea_simd.cpp new file mode 100644 index 000000000..6151c355c --- /dev/null +++ b/src/block/xtea_simd/xtea_simd.cpp @@ -0,0 +1,124 @@ +/* +* XTEA in SIMD +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#include <botan/xtea_simd.h> +#include <botan/loadstor.h> +#include <botan/simd_32.h> + +namespace Botan { + +namespace { + +void xtea_encrypt_8(const byte in[64], byte out[64], const u32bit EK[64]) + { + SIMD_32 L0 = SIMD_32::load_be(in ); + SIMD_32 R0 = SIMD_32::load_be(in + 16); + SIMD_32 L1 = SIMD_32::load_be(in + 32); + SIMD_32 R1 = SIMD_32::load_be(in + 48); + + SIMD_32::transpose(L0, R0, L1, R1); + + for(u32bit i = 0; i != 32; i += 2) + { + SIMD_32 K0(EK[2*i ]); + SIMD_32 K1(EK[2*i+1]); + SIMD_32 K2(EK[2*i+2]); + SIMD_32 K3(EK[2*i+3]); + + L0 += (((R0 << 4) ^ (R0 >> 5)) + R0) ^ K0; + L1 += (((R1 << 4) ^ (R1 >> 5)) + R1) ^ K0; + + R0 += (((L0 << 4) ^ (L0 >> 5)) + L0) ^ K1; + R1 += (((L1 << 4) ^ (L1 >> 5)) + L1) ^ K1; + + L0 += (((R0 << 4) ^ (R0 >> 5)) + R0) ^ K2; + L1 += (((R1 << 4) ^ (R1 >> 5)) + R1) ^ K2; + + R0 += (((L0 << 4) ^ (L0 >> 5)) + L0) ^ K3; + R1 += (((L1 << 4) ^ (L1 >> 5)) + L1) ^ K3; + } + + SIMD_32::transpose(L0, R0, L1, R1); + + L0.store_be(out); + R0.store_be(out + 16); + L1.store_be(out + 32); + R1.store_be(out + 48); + } + +void xtea_decrypt_8(const byte in[64], byte out[64], const u32bit EK[64]) + { + SIMD_32 L0 = SIMD_32::load_be(in ); + SIMD_32 R0 = SIMD_32::load_be(in + 16); + SIMD_32 L1 = SIMD_32::load_be(in + 32); + SIMD_32 R1 = SIMD_32::load_be(in + 48); + + SIMD_32::transpose(L0, R0, L1, R1); + + for(u32bit i = 0; i != 32; i += 2) + { + SIMD_32 K0(EK[63 - 2*i]); + SIMD_32 K1(EK[62 - 2*i]); + SIMD_32 K2(EK[61 - 2*i]); + SIMD_32 K3(EK[60 - 2*i]); + + R0 -= (((L0 << 4) ^ (L0 >> 5)) + L0) ^ K0; + R1 -= (((L1 << 4) ^ (L1 >> 5)) + L1) ^ K0; + + L0 -= (((R0 << 4) ^ (R0 >> 5)) + R0) ^ K1; + L1 -= (((R1 << 4) ^ (R1 >> 5)) + R1) ^ K1; + + R0 -= (((L0 << 4) ^ (L0 >> 5)) + L0) ^ K2; + R1 -= (((L1 << 4) ^ (L1 >> 5)) + L1) ^ K2; + + L0 -= (((R0 << 4) ^ (R0 >> 5)) + R0) ^ K3; + L1 -= (((R1 << 4) ^ (R1 >> 5)) + R1) ^ K3; + } + + SIMD_32::transpose(L0, R0, L1, R1); + + L0.store_be(out); + R0.store_be(out + 16); + L1.store_be(out + 32); + R1.store_be(out + 48); + } + +} + +/* +* XTEA Encryption +*/ +void XTEA_SIMD::encrypt_n(const byte in[], byte out[], u32bit blocks) const + { + while(blocks >= 8) + { + xtea_encrypt_8(in, out, this->EK); + in += 8 * BLOCK_SIZE; + out += 8 * BLOCK_SIZE; + blocks -= 8; + } + + XTEA::encrypt_n(in, out, blocks); + } + +/* +* XTEA Decryption +*/ +void XTEA_SIMD::decrypt_n(const byte in[], byte out[], u32bit blocks) const + { + while(blocks >= 8) + { + xtea_decrypt_8(in, out, this->EK); + in += 8 * BLOCK_SIZE; + out += 8 * BLOCK_SIZE; + blocks -= 8; + } + + XTEA::decrypt_n(in, out, blocks); + } + +} diff --git a/src/block/xtea_simd/xtea_simd.h b/src/block/xtea_simd/xtea_simd.h new file mode 100644 index 000000000..e4ce734ed --- /dev/null +++ b/src/block/xtea_simd/xtea_simd.h @@ -0,0 +1,28 @@ +/* +* XTEA in SIMD +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef BOTAN_XTEA_SIMD_H__ +#define BOTAN_XTEA_SIMD_H__ + +#include <botan/xtea.h> + +namespace Botan { + +/* +* XTEA (SIMD variant) +*/ +class BOTAN_DLL XTEA_SIMD : public XTEA + { + public: + void encrypt_n(const byte in[], byte out[], u32bit blocks) const; + void decrypt_n(const byte in[], byte out[], u32bit blocks) const; + BlockCipher* clone() const { return new XTEA_SIMD; } + }; + +} + +#endif diff --git a/src/engine/simd_engine/info.txt b/src/engine/simd_engine/info.txt new file mode 100644 index 000000000..b0523285f --- /dev/null +++ b/src/engine/simd_engine/info.txt @@ -0,0 +1,3 @@ +define ENGINE_SIMD + +load_on dep diff --git a/src/engine/simd_engine/simd_engine.cpp b/src/engine/simd_engine/simd_engine.cpp new file mode 100644 index 000000000..7e15f9ec1 --- /dev/null +++ b/src/engine/simd_engine/simd_engine.cpp @@ -0,0 +1,54 @@ +/** +* SIMD Engine +* (C) 1999-2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#include <botan/simd_engine.h> +#include <botan/cpuid.h> + +#if defined(BOTAN_HAS_SERPENT_SIMD) + #include <botan/serp_simd.h> +#endif + +#if defined(BOTAN_HAS_XTEA_SIMD) + #include <botan/xtea_simd.h> +#endif + +#if defined(BOTAN_HAS_SHA1_SSE2) + #include <botan/sha1_sse2.h> +#endif + +namespace Botan { + +BlockCipher* +SIMD_Engine::find_block_cipher(const SCAN_Name& request, + Algorithm_Factory&) const + { +#if defined(BOTAN_HAS_SERPENT_SIMD) + if(request.algo_name() == "Serpent") + return new Serpent_SIMD; +#endif + +#if defined(BOTAN_HAS_XTEA_SIMD) + if(request.algo_name() == "XTEA") + return new XTEA_SIMD; +#endif + + return 0; + } + +HashFunction* +SIMD_Engine::find_hash(const SCAN_Name& request, + Algorithm_Factory&) const + { +#if defined(BOTAN_HAS_SHA1_SSE2) + if(request.algo_name() == "SHA-160" && CPUID::has_sse2()) + return new SHA_160_SSE2; +#endif + + return 0; + } + +} diff --git a/src/engine/sse2_eng/eng_sse2.h b/src/engine/simd_engine/simd_engine.h index c6b0ce889..22a58e203 100644 --- a/src/engine/sse2_eng/eng_sse2.h +++ b/src/engine/simd_engine/simd_engine.h @@ -1,21 +1,21 @@ /** -* SSE2 Assembly Engine +* SIMD Assembly Engine * (C) 1999-2009 Jack Lloyd * * Distributed under the terms of the Botan license */ -#ifndef BOTAN_SSE2_ASM_ENGINE_H__ -#define BOTAN_SSE2_ASM_ENGINE_H__ +#ifndef BOTAN_SIMD_ENGINE_H__ +#define BOTAN_SIMD_ENGINE_H__ #include <botan/engine.h> namespace Botan { -class BOTAN_DLL SSE2_Assembler_Engine : public Engine +class BOTAN_DLL SIMD_Engine : public Engine { public: - std::string provider_name() const { return "sse2"; } + std::string provider_name() const { return "simd"; } private: BlockCipher* find_block_cipher(const SCAN_Name&, Algorithm_Factory&) const; diff --git a/src/engine/sse2_eng/eng_sse2.cpp b/src/engine/sse2_eng/eng_sse2.cpp deleted file mode 100644 index 07c625c7c..000000000 --- a/src/engine/sse2_eng/eng_sse2.cpp +++ /dev/null @@ -1,51 +0,0 @@ -/** -* SSE2 Assembly Engine -* (C) 1999-2009 Jack Lloyd -* -* Distributed under the terms of the Botan license -*/ - -#include <botan/eng_sse2.h> -#include <botan/cpuid.h> - -#if defined(BOTAN_HAS_SHA1_SSE2) - #include <botan/sha1_sse2.h> -#endif - -#if defined(BOTAN_HAS_SERPENT_SSE2) - #include <botan/serp_sse2.h> -#endif - -namespace Botan { - -BlockCipher* -SSE2_Assembler_Engine::find_block_cipher(const SCAN_Name& request, - Algorithm_Factory&) const - { - if(!CPUID::has_sse2()) - return 0; - -#if defined(BOTAN_HAS_SERPENT_SSE2) - if(request.algo_name() == "Serpent") - return new Serpent_SSE2; -#endif - - return 0; - } - -HashFunction* -SSE2_Assembler_Engine::find_hash(const SCAN_Name& request, - Algorithm_Factory&) const - { - if(!CPUID::has_sse2()) - return 0; - -#if defined(BOTAN_HAS_SHA1_SSE2) - if(request.algo_name() == "SHA-160") - return new SHA_160_SSE2; -#endif - - return 0; - } - -} diff --git a/src/engine/sse2_eng/info.txt b/src/engine/sse2_eng/info.txt deleted file mode 100644 index 43df92343..000000000 --- a/src/engine/sse2_eng/info.txt +++ /dev/null @@ -1,21 +0,0 @@ -define ENGINE_SSE2_ASSEMBLER - -load_on dep - -<add> -eng_sse2.cpp -eng_sse2.h -</add> - -<arch> -pentium-m -pentium4 -prescott -amd64 -</arch> - -<cc> -gcc -icc -msvc -</cc> diff --git a/src/hash/sha1/sha160.cpp b/src/hash/sha1/sha160.cpp index 92f4f5f11..88f2161e2 100644 --- a/src/hash/sha1/sha160.cpp +++ b/src/hash/sha1/sha160.cpp @@ -82,37 +82,49 @@ void SHA_160::compress_n(const byte input[], u32bit blocks) W[j+7] = rotate_left((W[j+4] ^ W[j-1] ^ W[j- 7] ^ W[j- 9]), 1); } - F1(A,B,C,D,E,W[ 0]); F1(E,A,B,C,D,W[ 1]); F1(D,E,A,B,C,W[ 2]); - F1(C,D,E,A,B,W[ 3]); F1(B,C,D,E,A,W[ 4]); F1(A,B,C,D,E,W[ 5]); - F1(E,A,B,C,D,W[ 6]); F1(D,E,A,B,C,W[ 7]); F1(C,D,E,A,B,W[ 8]); - F1(B,C,D,E,A,W[ 9]); F1(A,B,C,D,E,W[10]); F1(E,A,B,C,D,W[11]); - F1(D,E,A,B,C,W[12]); F1(C,D,E,A,B,W[13]); F1(B,C,D,E,A,W[14]); - F1(A,B,C,D,E,W[15]); F1(E,A,B,C,D,W[16]); F1(D,E,A,B,C,W[17]); - F1(C,D,E,A,B,W[18]); F1(B,C,D,E,A,W[19]); - - F2(A,B,C,D,E,W[20]); F2(E,A,B,C,D,W[21]); F2(D,E,A,B,C,W[22]); - F2(C,D,E,A,B,W[23]); F2(B,C,D,E,A,W[24]); F2(A,B,C,D,E,W[25]); - F2(E,A,B,C,D,W[26]); F2(D,E,A,B,C,W[27]); F2(C,D,E,A,B,W[28]); - F2(B,C,D,E,A,W[29]); F2(A,B,C,D,E,W[30]); F2(E,A,B,C,D,W[31]); - F2(D,E,A,B,C,W[32]); F2(C,D,E,A,B,W[33]); F2(B,C,D,E,A,W[34]); - F2(A,B,C,D,E,W[35]); F2(E,A,B,C,D,W[36]); F2(D,E,A,B,C,W[37]); - F2(C,D,E,A,B,W[38]); F2(B,C,D,E,A,W[39]); - - F3(A,B,C,D,E,W[40]); F3(E,A,B,C,D,W[41]); F3(D,E,A,B,C,W[42]); - F3(C,D,E,A,B,W[43]); F3(B,C,D,E,A,W[44]); F3(A,B,C,D,E,W[45]); - F3(E,A,B,C,D,W[46]); F3(D,E,A,B,C,W[47]); F3(C,D,E,A,B,W[48]); - F3(B,C,D,E,A,W[49]); F3(A,B,C,D,E,W[50]); F3(E,A,B,C,D,W[51]); - F3(D,E,A,B,C,W[52]); F3(C,D,E,A,B,W[53]); F3(B,C,D,E,A,W[54]); - F3(A,B,C,D,E,W[55]); F3(E,A,B,C,D,W[56]); F3(D,E,A,B,C,W[57]); - F3(C,D,E,A,B,W[58]); F3(B,C,D,E,A,W[59]); - - F4(A,B,C,D,E,W[60]); F4(E,A,B,C,D,W[61]); F4(D,E,A,B,C,W[62]); - F4(C,D,E,A,B,W[63]); F4(B,C,D,E,A,W[64]); F4(A,B,C,D,E,W[65]); - F4(E,A,B,C,D,W[66]); F4(D,E,A,B,C,W[67]); F4(C,D,E,A,B,W[68]); - F4(B,C,D,E,A,W[69]); F4(A,B,C,D,E,W[70]); F4(E,A,B,C,D,W[71]); - F4(D,E,A,B,C,W[72]); F4(C,D,E,A,B,W[73]); F4(B,C,D,E,A,W[74]); - F4(A,B,C,D,E,W[75]); F4(E,A,B,C,D,W[76]); F4(D,E,A,B,C,W[77]); - F4(C,D,E,A,B,W[78]); F4(B,C,D,E,A,W[79]); + F1(A, B, C, D, E, W[ 0]); F1(E, A, B, C, D, W[ 1]); + F1(D, E, A, B, C, W[ 2]); F1(C, D, E, A, B, W[ 3]); + F1(B, C, D, E, A, W[ 4]); F1(A, B, C, D, E, W[ 5]); + F1(E, A, B, C, D, W[ 6]); F1(D, E, A, B, C, W[ 7]); + F1(C, D, E, A, B, W[ 8]); F1(B, C, D, E, A, W[ 9]); + F1(A, B, C, D, E, W[10]); F1(E, A, B, C, D, W[11]); + F1(D, E, A, B, C, W[12]); F1(C, D, E, A, B, W[13]); + F1(B, C, D, E, A, W[14]); F1(A, B, C, D, E, W[15]); + F1(E, A, B, C, D, W[16]); F1(D, E, A, B, C, W[17]); + F1(C, D, E, A, B, W[18]); F1(B, C, D, E, A, W[19]); + + F2(A, B, C, D, E, W[20]); F2(E, A, B, C, D, W[21]); + F2(D, E, A, B, C, W[22]); F2(C, D, E, A, B, W[23]); + F2(B, C, D, E, A, W[24]); F2(A, B, C, D, E, W[25]); + F2(E, A, B, C, D, W[26]); F2(D, E, A, B, C, W[27]); + F2(C, D, E, A, B, W[28]); F2(B, C, D, E, A, W[29]); + F2(A, B, C, D, E, W[30]); F2(E, A, B, C, D, W[31]); + F2(D, E, A, B, C, W[32]); F2(C, D, E, A, B, W[33]); + F2(B, C, D, E, A, W[34]); F2(A, B, C, D, E, W[35]); + F2(E, A, B, C, D, W[36]); F2(D, E, A, B, C, W[37]); + F2(C, D, E, A, B, W[38]); F2(B, C, D, E, A, W[39]); + + F3(A, B, C, D, E, W[40]); F3(E, A, B, C, D, W[41]); + F3(D, E, A, B, C, W[42]); F3(C, D, E, A, B, W[43]); + F3(B, C, D, E, A, W[44]); F3(A, B, C, D, E, W[45]); + F3(E, A, B, C, D, W[46]); F3(D, E, A, B, C, W[47]); + F3(C, D, E, A, B, W[48]); F3(B, C, D, E, A, W[49]); + F3(A, B, C, D, E, W[50]); F3(E, A, B, C, D, W[51]); + F3(D, E, A, B, C, W[52]); F3(C, D, E, A, B, W[53]); + F3(B, C, D, E, A, W[54]); F3(A, B, C, D, E, W[55]); + F3(E, A, B, C, D, W[56]); F3(D, E, A, B, C, W[57]); + F3(C, D, E, A, B, W[58]); F3(B, C, D, E, A, W[59]); + + F4(A, B, C, D, E, W[60]); F4(E, A, B, C, D, W[61]); + F4(D, E, A, B, C, W[62]); F4(C, D, E, A, B, W[63]); + F4(B, C, D, E, A, W[64]); F4(A, B, C, D, E, W[65]); + F4(E, A, B, C, D, W[66]); F4(D, E, A, B, C, W[67]); + F4(C, D, E, A, B, W[68]); F4(B, C, D, E, A, W[69]); + F4(A, B, C, D, E, W[70]); F4(E, A, B, C, D, W[71]); + F4(D, E, A, B, C, W[72]); F4(C, D, E, A, B, W[73]); + F4(B, C, D, E, A, W[74]); F4(A, B, C, D, E, W[75]); + F4(E, A, B, C, D, W[76]); F4(D, E, A, B, C, W[77]); + F4(C, D, E, A, B, W[78]); F4(B, C, D, E, A, W[79]); A = (digest[0] += A); B = (digest[1] += B); diff --git a/src/hash/sha1_sse2/info.txt b/src/hash/sha1_sse2/info.txt index ad61aa5fa..ee61076b4 100644 --- a/src/hash/sha1_sse2/info.txt +++ b/src/hash/sha1_sse2/info.txt @@ -2,5 +2,18 @@ define SHA1_SSE2 <requires> sha1 -sse2_eng +simd_engine </requires> + +<arch> +pentium-m +pentium4 +prescott +amd64 +</arch> + +<cc> +gcc +icc +msvc +</cc> diff --git a/src/hash/sha1_sse2/sha1_sse2.cpp b/src/hash/sha1_sse2/sha1_sse2.cpp index dddc06b7b..fc6466dd0 100644 --- a/src/hash/sha1_sse2/sha1_sse2.cpp +++ b/src/hash/sha1_sse2/sha1_sse2.cpp @@ -1,23 +1,278 @@ /* -* SHA-160 (SSE2) -* (C) 1999-2007 Jack Lloyd +* SHA-1 using SSE2 +* (C) 2009 Jack Lloyd * * Distributed under the terms of the Botan license +* +* Based on public domain code by Dean Gaudet <[email protected]> +* Source - http://arctic.org/~dean/crypto/sha1.html */ #include <botan/sha1_sse2.h> +#include <botan/rotate.h> +#include <emmintrin.h> namespace Botan { +namespace { + +/* +First 16 bytes just need byte swapping. Preparing just means +adding in the round constants. +*/ + +#define prep00_15(P, W) \ + do { \ + W = _mm_shufflehi_epi16(W, _MM_SHUFFLE(2, 3, 0, 1)); \ + W = _mm_shufflelo_epi16(W, _MM_SHUFFLE(2, 3, 0, 1)); \ + W = _mm_or_si128(_mm_slli_epi16(W, 8), \ + _mm_srli_epi16(W, 8)); \ + P.u128 = _mm_add_epi32(W, K00_19); \ + } while(0) + +/* +for each multiple of 4, t, we want to calculate this: + +W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1); +W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1); +W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1); +W[t+3] = rol(W[t] ^ W[t-5] ^ W[t-11] ^ W[t-13], 1); + +we'll actually calculate this: + +W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1); +W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1); +W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1); +W[t+3] = rol( 0 ^ W[t-5] ^ W[t-11] ^ W[t-13], 1); +W[t+3] ^= rol(W[t+0], 1); + +the parameters are: + +W0 = &W[t-16]; +W1 = &W[t-12]; +W2 = &W[t- 8]; +W3 = &W[t- 4]; + +and on output: +prepared = W0 + K +W0 = W[t]..W[t+3] +*/ + +/* note that there is a step here where i want to do a rol by 1, which +* normally would look like this: +* +* r1 = psrld r0,$31 +* r0 = pslld r0,$1 +* r0 = por r0,r1 +* +* but instead i do this: +* +* r1 = pcmpltd r0,zero +* r0 = paddd r0,r0 +* r0 = psub r0,r1 +* +* because pcmpltd and paddd are availabe in both MMX units on +* efficeon, pentium-m, and opteron but shifts are available in +* only one unit. +*/ +#define prep(prep, XW0, XW1, XW2, XW3, K) \ + do { \ + __m128i r0, r1, r2, r3; \ + \ + /* load W[t-4] 16-byte aligned, and shift */ \ + r3 = _mm_srli_si128((XW3), 4); \ + r0 = (XW0); \ + /* get high 64-bits of XW0 into low 64-bits */ \ + r1 = _mm_shuffle_epi32((XW0), _MM_SHUFFLE(1,0,3,2)); \ + /* load high 64-bits of r1 */ \ + r1 = _mm_unpacklo_epi64(r1, (XW1)); \ + r2 = (XW2); \ + \ + r0 = _mm_xor_si128(r1, r0); \ + r2 = _mm_xor_si128(r3, r2); \ + r0 = _mm_xor_si128(r2, r0); \ + /* unrotated W[t]..W[t+2] in r0 ... still need W[t+3] */ \ + \ + r2 = _mm_slli_si128(r0, 12); \ + r1 = _mm_cmplt_epi32(r0, _mm_setzero_si128()); \ + r0 = _mm_add_epi32(r0, r0); /* shift left by 1 */ \ + r0 = _mm_sub_epi32(r0, r1); /* r0 has W[t]..W[t+2] */ \ + \ + r3 = _mm_srli_epi32(r2, 30); \ + r2 = _mm_slli_epi32(r2, 2); \ + \ + r0 = _mm_xor_si128(r0, r3); \ + r0 = _mm_xor_si128(r0, r2); /* r0 now has W[t+3] */ \ + \ + (XW0) = r0; \ + (prep).u128 = _mm_add_epi32(r0, K); \ + } while(0) + +/* +* SHA-160 F1 Function +*/ +inline void F1(u32bit A, u32bit& B, u32bit C, u32bit D, u32bit& E, u32bit msg) + { + E += (D ^ (B & (C ^ D))) + msg + rotate_left(A, 5); + B = rotate_left(B, 30); + } + /* -* SHA-160 Compression Function +* SHA-160 F2 Function */ -void SHA_160_SSE2::compress_n(const byte input[], u32bit blocks) +inline void F2(u32bit A, u32bit& B, u32bit C, u32bit D, u32bit& E, u32bit msg) { + E += (B ^ C ^ D) + msg + rotate_left(A, 5); + B = rotate_left(B, 30); + } + +/* +* SHA-160 F3 Function +*/ +inline void F3(u32bit A, u32bit& B, u32bit C, u32bit D, u32bit& E, u32bit msg) + { + E += ((B & C) | ((B | C) & D)) + msg + rotate_left(A, 5); + B = rotate_left(B, 30); + } + +/* +* SHA-160 F4 Function +*/ +inline void F4(u32bit A, u32bit& B, u32bit C, u32bit D, u32bit& E, u32bit msg) + { + E += (B ^ C ^ D) + msg + rotate_left(A, 5); + B = rotate_left(B, 30); + } + +} + +/* +* SHA-160 Compression Function using SSE for message expansion +*/ +void SHA_160_SSE2::compress_n(const byte input_bytes[], u32bit blocks) + { + const __m128i K00_19 = _mm_set1_epi32(0x5A827999); + const __m128i K20_39 = _mm_set1_epi32(0x6ED9EBA1); + const __m128i K40_59 = _mm_set1_epi32(0x8F1BBCDC); + const __m128i K60_79 = _mm_set1_epi32(0xCA62C1D6); + + u32bit A = digest[0], B = digest[1], C = digest[2], + D = digest[3], E = digest[4]; + + const __m128i* input = (const __m128i *)input_bytes; + for(u32bit i = 0; i != blocks; ++i) { - botan_sha1_sse2_compress(digest, reinterpret_cast<const u32bit*>(input)); - input += HASH_BLOCK_SIZE; + + /* I've tried arranging the SSE2 code to be 4, 8, 12, and 16 + * steps ahead of the integer code. 12 steps ahead seems to + * produce the best performance. -dean + * + * Todo: check this is still true on Barcelona and Core2 -Jack + */ + + union v4si { + u32bit u32[4]; + __m128i u128; + }; + + v4si P0, P1, P2; + + __m128i W0 = _mm_loadu_si128(&input[0]); + prep00_15(P0, W0); + + __m128i W1 = _mm_loadu_si128(&input[1]); + prep00_15(P1, W1); + + __m128i W2 = _mm_loadu_si128(&input[2]); + prep00_15(P2, W2); + + __m128i W3 = _mm_loadu_si128(&input[3]); + + F1(A, B, C, D, E, P0.u32[0]); F1(E, A, B, C, D, P0.u32[1]); + F1(D, E, A, B, C, P0.u32[2]); F1(C, D, E, A, B, P0.u32[3]); + prep00_15(P0, W3); + + F1(B, C, D, E, A, P1.u32[0]); F1(A, B, C, D, E, P1.u32[1]); + F1(E, A, B, C, D, P1.u32[2]); F1(D, E, A, B, C, P1.u32[3]); + prep(P1, W0, W1, W2, W3, K00_19); + + F1(C, D, E, A, B, P2.u32[0]); F1(B, C, D, E, A, P2.u32[1]); + F1(A, B, C, D, E, P2.u32[2]); F1(E, A, B, C, D, P2.u32[3]); + prep(P2, W1, W2, W3, W0, K20_39); + + F1(D, E, A, B, C, P0.u32[0]); F1(C, D, E, A, B, P0.u32[1]); + F1(B, C, D, E, A, P0.u32[2]); F1(A, B, C, D, E, P0.u32[3]); + prep(P0, W2, W3, W0, W1, K20_39); + + F1(E, A, B, C, D, P1.u32[0]); F1(D, E, A, B, C, P1.u32[1]); + F1(C, D, E, A, B, P1.u32[2]); F1(B, C, D, E, A, P1.u32[3]); + prep(P1, W3, W0, W1, W2, K20_39); + + F2(A, B, C, D, E, P2.u32[0]); F2(E, A, B, C, D, P2.u32[1]); + F2(D, E, A, B, C, P2.u32[2]); F2(C, D, E, A, B, P2.u32[3]); + prep(P2, W0, W1, W2, W3, K20_39); + + F2(B, C, D, E, A, P0.u32[0]); F2(A, B, C, D, E, P0.u32[1]); + F2(E, A, B, C, D, P0.u32[2]); F2(D, E, A, B, C, P0.u32[3]); + prep(P0, W1, W2, W3, W0, K20_39); + + F2(C, D, E, A, B, P1.u32[0]); F2(B, C, D, E, A, P1.u32[1]); + F2(A, B, C, D, E, P1.u32[2]); F2(E, A, B, C, D, P1.u32[3]); + prep(P1, W2, W3, W0, W1, K40_59); + + F2(D, E, A, B, C, P2.u32[0]); F2(C, D, E, A, B, P2.u32[1]); + F2(B, C, D, E, A, P2.u32[2]); F2(A, B, C, D, E, P2.u32[3]); + prep(P2, W3, W0, W1, W2, K40_59); + + F2(E, A, B, C, D, P0.u32[0]); F2(D, E, A, B, C, P0.u32[1]); + F2(C, D, E, A, B, P0.u32[2]); F2(B, C, D, E, A, P0.u32[3]); + prep(P0, W0, W1, W2, W3, K40_59); + + F3(A, B, C, D, E, P1.u32[0]); F3(E, A, B, C, D, P1.u32[1]); + F3(D, E, A, B, C, P1.u32[2]); F3(C, D, E, A, B, P1.u32[3]); + prep(P1, W1, W2, W3, W0, K40_59); + + F3(B, C, D, E, A, P2.u32[0]); F3(A, B, C, D, E, P2.u32[1]); + F3(E, A, B, C, D, P2.u32[2]); F3(D, E, A, B, C, P2.u32[3]); + prep(P2, W2, W3, W0, W1, K40_59); + + F3(C, D, E, A, B, P0.u32[0]); F3(B, C, D, E, A, P0.u32[1]); + F3(A, B, C, D, E, P0.u32[2]); F3(E, A, B, C, D, P0.u32[3]); + prep(P0, W3, W0, W1, W2, K60_79); + + F3(D, E, A, B, C, P1.u32[0]); F3(C, D, E, A, B, P1.u32[1]); + F3(B, C, D, E, A, P1.u32[2]); F3(A, B, C, D, E, P1.u32[3]); + prep(P1, W0, W1, W2, W3, K60_79); + + F3(E, A, B, C, D, P2.u32[0]); F3(D, E, A, B, C, P2.u32[1]); + F3(C, D, E, A, B, P2.u32[2]); F3(B, C, D, E, A, P2.u32[3]); + prep(P2, W1, W2, W3, W0, K60_79); + + F4(A, B, C, D, E, P0.u32[0]); F4(E, A, B, C, D, P0.u32[1]); + F4(D, E, A, B, C, P0.u32[2]); F4(C, D, E, A, B, P0.u32[3]); + prep(P0, W2, W3, W0, W1, K60_79); + + F4(B, C, D, E, A, P1.u32[0]); F4(A, B, C, D, E, P1.u32[1]); + F4(E, A, B, C, D, P1.u32[2]); F4(D, E, A, B, C, P1.u32[3]); + prep(P1, W3, W0, W1, W2, K60_79); + + F4(C, D, E, A, B, P2.u32[0]); F4(B, C, D, E, A, P2.u32[1]); + F4(A, B, C, D, E, P2.u32[2]); F4(E, A, B, C, D, P2.u32[3]); + + F4(D, E, A, B, C, P0.u32[0]); F4(C, D, E, A, B, P0.u32[1]); + F4(B, C, D, E, A, P0.u32[2]); F4(A, B, C, D, E, P0.u32[3]); + + F4(E, A, B, C, D, P1.u32[0]); F4(D, E, A, B, C, P1.u32[1]); + F4(C, D, E, A, B, P1.u32[2]); F4(B, C, D, E, A, P1.u32[3]); + + A = (digest[0] += A); + B = (digest[1] += B); + C = (digest[2] += C); + D = (digest[3] += D); + E = (digest[4] += E); + + input += (HASH_BLOCK_SIZE / 16); } } diff --git a/src/hash/sha1_sse2/sha1_sse2_imp.cpp b/src/hash/sha1_sse2/sha1_sse2_imp.cpp deleted file mode 100644 index 5ee222cca..000000000 --- a/src/hash/sha1_sse2/sha1_sse2_imp.cpp +++ /dev/null @@ -1,304 +0,0 @@ -/* -* SHA-1 using SSE2 -* (C) 2009 Jack Lloyd -* -* Distributed under the terms of the Botan license -*/ - -/* -* Based on public domain code by Dean Gaudet <[email protected]> -* http://arctic.org/~dean/crypto/sha1.html -*/ - -#include <botan/sha1_sse2.h> -#include <botan/rotate.h> -#include <emmintrin.h> - -namespace Botan { - -namespace { - -typedef union { - u32bit u32[4]; - __m128i u128; - } v4si; - -static const v4si K00_19 = { { 0x5a827999, 0x5a827999, 0x5a827999, 0x5a827999 } }; -static const v4si K20_39 = { { 0x6ed9eba1, 0x6ed9eba1, 0x6ed9eba1, 0x6ed9eba1 } }; -static const v4si K40_59 = { { 0x8f1bbcdc, 0x8f1bbcdc, 0x8f1bbcdc, 0x8f1bbcdc } }; -static const v4si K60_79 = { { 0xca62c1d6, 0xca62c1d6, 0xca62c1d6, 0xca62c1d6 } }; - -/* -the first 16 bytes only need byte swapping - -prepared points to 4x u32bit, 16-byte aligned - -W points to the 4 dwords which need preparing -- -and is overwritten with the swapped bytes -*/ -#define prep00_15(prep, W) do { \ - __m128i r1, r2; \ - \ - r1 = (W); \ - r1 = _mm_shufflehi_epi16(r1, _MM_SHUFFLE(2, 3, 0, 1)); \ - r1 = _mm_shufflelo_epi16(r1, _MM_SHUFFLE(2, 3, 0, 1)); \ - r2 = _mm_slli_epi16(r1, 8); \ - r1 = _mm_srli_epi16(r1, 8); \ - r1 = _mm_or_si128(r1, r2); \ - (W) = r1; \ - (prep).u128 = _mm_add_epi32(K00_19.u128, r1); \ - } while(0) - -/* -for each multiple of 4, t, we want to calculate this: - -W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1); -W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1); -W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1); -W[t+3] = rol(W[t] ^ W[t-5] ^ W[t-11] ^ W[t-13], 1); - -we'll actually calculate this: - -W[t+0] = rol(W[t-3] ^ W[t-8] ^ W[t-14] ^ W[t-16], 1); -W[t+1] = rol(W[t-2] ^ W[t-7] ^ W[t-13] ^ W[t-15], 1); -W[t+2] = rol(W[t-1] ^ W[t-6] ^ W[t-12] ^ W[t-14], 1); -W[t+3] = rol( 0 ^ W[t-5] ^ W[t-11] ^ W[t-13], 1); -W[t+3] ^= rol(W[t+0], 1); - -the parameters are: - -W0 = &W[t-16]; -W1 = &W[t-12]; -W2 = &W[t- 8]; -W3 = &W[t- 4]; - -and on output: -prepared = W0 + K -W0 = W[t]..W[t+3] -*/ - -/* note that there is a step here where i want to do a rol by 1, which -* normally would look like this: -* -* r1 = psrld r0,$31 -* r0 = pslld r0,$1 -* r0 = por r0,r1 -* -* but instead i do this: -* -* r1 = pcmpltd r0,zero -* r0 = paddd r0,r0 -* r0 = psub r0,r1 -* -* because pcmpltd and paddd are availabe in both MMX units on -* efficeon, pentium-m, and opteron but shifts are available in -* only one unit. -*/ -#define prep(prep, XW0, XW1, XW2, XW3, K) \ - do { \ - __m128i r0, r1, r2, r3; \ - \ - /* load W[t-4] 16-byte aligned, and shift */ \ - r3 = _mm_srli_si128((XW3), 4); \ - r0 = (XW0); \ - /* get high 64-bits of XW0 into low 64-bits */ \ - r1 = _mm_shuffle_epi32((XW0), _MM_SHUFFLE(1,0,3,2)); \ - /* load high 64-bits of r1 */ \ - r1 = _mm_unpacklo_epi64(r1, (XW1)); \ - r2 = (XW2); \ - \ - r0 = _mm_xor_si128(r1, r0); \ - r2 = _mm_xor_si128(r3, r2); \ - r0 = _mm_xor_si128(r2, r0); \ - /* unrotated W[t]..W[t+2] in r0 ... still need W[t+3] */ \ - \ - r2 = _mm_slli_si128(r0, 12); \ - r1 = _mm_cmplt_epi32(r0, _mm_setzero_si128()); \ - r0 = _mm_add_epi32(r0, r0); /* shift left by 1 */ \ - r0 = _mm_sub_epi32(r0, r1); /* r0 has W[t]..W[t+2] */ \ - \ - r3 = _mm_srli_epi32(r2, 30); \ - r2 = _mm_slli_epi32(r2, 2); \ - \ - r0 = _mm_xor_si128(r0, r3); \ - r0 = _mm_xor_si128(r0, r2); /* r0 now has W[t+3] */ \ - \ - (XW0) = r0; \ - (prep).u128 = _mm_add_epi32(r0, (K).u128); \ - } while(0) - -static inline u32bit f00_19(u32bit x, u32bit y, u32bit z) - { - /* FIPS 180-2 says this: (x & y) ^ (~x & z) - * but we can calculate it in fewer steps. - */ - return ((y ^ z) & x) ^ z; - } - - -static inline u32bit f20_39(u32bit x, u32bit y, u32bit z) - { - return (x ^ z) ^ y; - } - - -static inline u32bit f40_59(u32bit x, u32bit y, u32bit z) - { - /* FIPS 180-2 says this: (x & y) ^ (x & z) ^ (y & z) - * but we can calculate it in fewer steps. - */ - return (x & z) | ((x | z) & y); - } - - -static inline u32bit f60_79(u32bit x, u32bit y, u32bit z) - { - return f20_39(x, y, z); - } - -#define step(nn_mm, xa, xb, xc, xd, xe, xt, input) \ - do { \ - (xt) = (input) + f##nn_mm((xb), (xc), (xd)); \ - (xb) = rotate_left((xb), 30); \ - (xt) += ((xe) + rotate_left((xa), 5)); \ - } while(0) - -} - -extern "C" void botan_sha1_sse2_compress(u32bit H[5], - const u32bit* inputu) - { - const __m128i * input = (const __m128i *)inputu; - __m128i W0, W1, W2, W3; - v4si prep0, prep1, prep2; - u32bit a, b, c, d, e, t; - - a = H[0]; - b = H[1]; - c = H[2]; - d = H[3]; - e = H[4]; - - /* i've tried arranging the SSE2 code to be 4, 8, 12, and 16 - * steps ahead of the integer code. 12 steps ahead seems - * to produce the best performance. -dean - */ - W0 = _mm_loadu_si128(&input[0]); - prep00_15(prep0, W0); /* prepare for 00 through 03 */ - W1 = _mm_loadu_si128(&input[1]); - prep00_15(prep1, W1); /* prepare for 04 through 07 */ - W2 = _mm_loadu_si128(&input[2]); - prep00_15(prep2, W2); /* prepare for 08 through 11 */ - - W3 = _mm_loadu_si128(&input[3]); - step(00_19, a, b, c, d, e, t, prep0.u32[0]); /* 00 */ - step(00_19, t, a, b, c, d, e, prep0.u32[1]); /* 01 */ - step(00_19, e, t, a, b, c, d, prep0.u32[2]); /* 02 */ - step(00_19, d, e, t, a, b, c, prep0.u32[3]); /* 03 */ - prep00_15(prep0, W3); - step(00_19, c, d, e, t, a, b, prep1.u32[0]); /* 04 */ - step(00_19, b, c, d, e, t, a, prep1.u32[1]); /* 05 */ - step(00_19, a, b, c, d, e, t, prep1.u32[2]); /* 06 */ - step(00_19, t, a, b, c, d, e, prep1.u32[3]); /* 07 */ - prep(prep1, W0, W1, W2, W3, K00_19); /* prepare for 16 through 19 */ - step(00_19, e, t, a, b, c, d, prep2.u32[0]); /* 08 */ - step(00_19, d, e, t, a, b, c, prep2.u32[1]); /* 09 */ - step(00_19, c, d, e, t, a, b, prep2.u32[2]); /* 10 */ - step(00_19, b, c, d, e, t, a, prep2.u32[3]); /* 11 */ - prep(prep2, W1, W2, W3, W0, K20_39); /* prepare for 20 through 23 */ - step(00_19, a, b, c, d, e, t, prep0.u32[0]); /* 12 */ - step(00_19, t, a, b, c, d, e, prep0.u32[1]); /* 13 */ - step(00_19, e, t, a, b, c, d, prep0.u32[2]); /* 14 */ - step(00_19, d, e, t, a, b, c, prep0.u32[3]); /* 15 */ - prep(prep0, W2, W3, W0, W1, K20_39); - step(00_19, c, d, e, t, a, b, prep1.u32[0]); /* 16 */ - step(00_19, b, c, d, e, t, a, prep1.u32[1]); /* 17 */ - step(00_19, a, b, c, d, e, t, prep1.u32[2]); /* 18 */ - step(00_19, t, a, b, c, d, e, prep1.u32[3]); /* 19 */ - - prep(prep1, W3, W0, W1, W2, K20_39); - step(20_39, e, t, a, b, c, d, prep2.u32[0]); /* 20 */ - step(20_39, d, e, t, a, b, c, prep2.u32[1]); /* 21 */ - step(20_39, c, d, e, t, a, b, prep2.u32[2]); /* 22 */ - step(20_39, b, c, d, e, t, a, prep2.u32[3]); /* 23 */ - prep(prep2, W0, W1, W2, W3, K20_39); - step(20_39, a, b, c, d, e, t, prep0.u32[0]); /* 24 */ - step(20_39, t, a, b, c, d, e, prep0.u32[1]); /* 25 */ - step(20_39, e, t, a, b, c, d, prep0.u32[2]); /* 26 */ - step(20_39, d, e, t, a, b, c, prep0.u32[3]); /* 27 */ - prep(prep0, W1, W2, W3, W0, K20_39); - step(20_39, c, d, e, t, a, b, prep1.u32[0]); /* 28 */ - step(20_39, b, c, d, e, t, a, prep1.u32[1]); /* 29 */ - step(20_39, a, b, c, d, e, t, prep1.u32[2]); /* 30 */ - step(20_39, t, a, b, c, d, e, prep1.u32[3]); /* 31 */ - prep(prep1, W2, W3, W0, W1, K40_59); - step(20_39, e, t, a, b, c, d, prep2.u32[0]); /* 32 */ - step(20_39, d, e, t, a, b, c, prep2.u32[1]); /* 33 */ - step(20_39, c, d, e, t, a, b, prep2.u32[2]); /* 34 */ - step(20_39, b, c, d, e, t, a, prep2.u32[3]); /* 35 */ - prep(prep2, W3, W0, W1, W2, K40_59); - step(20_39, a, b, c, d, e, t, prep0.u32[0]); /* 36 */ - step(20_39, t, a, b, c, d, e, prep0.u32[1]); /* 37 */ - step(20_39, e, t, a, b, c, d, prep0.u32[2]); /* 38 */ - step(20_39, d, e, t, a, b, c, prep0.u32[3]); /* 39 */ - - prep(prep0, W0, W1, W2, W3, K40_59); - step(40_59, c, d, e, t, a, b, prep1.u32[0]); /* 40 */ - step(40_59, b, c, d, e, t, a, prep1.u32[1]); /* 41 */ - step(40_59, a, b, c, d, e, t, prep1.u32[2]); /* 42 */ - step(40_59, t, a, b, c, d, e, prep1.u32[3]); /* 43 */ - prep(prep1, W1, W2, W3, W0, K40_59); - step(40_59, e, t, a, b, c, d, prep2.u32[0]); /* 44 */ - step(40_59, d, e, t, a, b, c, prep2.u32[1]); /* 45 */ - step(40_59, c, d, e, t, a, b, prep2.u32[2]); /* 46 */ - step(40_59, b, c, d, e, t, a, prep2.u32[3]); /* 47 */ - prep(prep2, W2, W3, W0, W1, K40_59); - step(40_59, a, b, c, d, e, t, prep0.u32[0]); /* 48 */ - step(40_59, t, a, b, c, d, e, prep0.u32[1]); /* 49 */ - step(40_59, e, t, a, b, c, d, prep0.u32[2]); /* 50 */ - step(40_59, d, e, t, a, b, c, prep0.u32[3]); /* 51 */ - prep(prep0, W3, W0, W1, W2, K60_79); - step(40_59, c, d, e, t, a, b, prep1.u32[0]); /* 52 */ - step(40_59, b, c, d, e, t, a, prep1.u32[1]); /* 53 */ - step(40_59, a, b, c, d, e, t, prep1.u32[2]); /* 54 */ - step(40_59, t, a, b, c, d, e, prep1.u32[3]); /* 55 */ - prep(prep1, W0, W1, W2, W3, K60_79); - step(40_59, e, t, a, b, c, d, prep2.u32[0]); /* 56 */ - step(40_59, d, e, t, a, b, c, prep2.u32[1]); /* 57 */ - step(40_59, c, d, e, t, a, b, prep2.u32[2]); /* 58 */ - step(40_59, b, c, d, e, t, a, prep2.u32[3]); /* 59 */ - - prep(prep2, W1, W2, W3, W0, K60_79); - step(60_79, a, b, c, d, e, t, prep0.u32[0]); /* 60 */ - step(60_79, t, a, b, c, d, e, prep0.u32[1]); /* 61 */ - step(60_79, e, t, a, b, c, d, prep0.u32[2]); /* 62 */ - step(60_79, d, e, t, a, b, c, prep0.u32[3]); /* 63 */ - prep(prep0, W2, W3, W0, W1, K60_79); - step(60_79, c, d, e, t, a, b, prep1.u32[0]); /* 64 */ - step(60_79, b, c, d, e, t, a, prep1.u32[1]); /* 65 */ - step(60_79, a, b, c, d, e, t, prep1.u32[2]); /* 66 */ - step(60_79, t, a, b, c, d, e, prep1.u32[3]); /* 67 */ - prep(prep1, W3, W0, W1, W2, K60_79); - step(60_79, e, t, a, b, c, d, prep2.u32[0]); /* 68 */ - step(60_79, d, e, t, a, b, c, prep2.u32[1]); /* 69 */ - step(60_79, c, d, e, t, a, b, prep2.u32[2]); /* 70 */ - step(60_79, b, c, d, e, t, a, prep2.u32[3]); /* 71 */ - - step(60_79, a, b, c, d, e, t, prep0.u32[0]); /* 72 */ - step(60_79, t, a, b, c, d, e, prep0.u32[1]); /* 73 */ - step(60_79, e, t, a, b, c, d, prep0.u32[2]); /* 74 */ - step(60_79, d, e, t, a, b, c, prep0.u32[3]); /* 75 */ - /* no more input to prepare */ - step(60_79, c, d, e, t, a, b, prep1.u32[0]); /* 76 */ - step(60_79, b, c, d, e, t, a, prep1.u32[1]); /* 77 */ - step(60_79, a, b, c, d, e, t, prep1.u32[2]); /* 78 */ - step(60_79, t, a, b, c, d, e, prep1.u32[3]); /* 79 */ - /* e, t, a, b, c, d */ - H[0] += e; - H[1] += t; - H[2] += a; - H[3] += b; - H[4] += c; - } - -} diff --git a/src/libstate/libstate.cpp b/src/libstate/libstate.cpp index ca454458b..c78bce62d 100644 --- a/src/libstate/libstate.cpp +++ b/src/libstate/libstate.cpp @@ -37,8 +37,8 @@ #include <botan/eng_amd64.h> #endif -#if defined(BOTAN_HAS_ENGINE_SSE2_ASSEMBLER) - #include <botan/eng_sse2.h> +#if defined(BOTAN_HAS_ENGINE_SIMD) + #include <botan/simd_engine.h> #endif #if defined(BOTAN_HAS_ENGINE_GNU_MP) @@ -288,8 +288,8 @@ void Library_State::initialize(bool thread_safe) engines.push_back(new OpenSSL_Engine); #endif -#if defined(BOTAN_HAS_ENGINE_SSE2_ASSEMBLER) - engines.push_back(new SSE2_Assembler_Engine); +#if defined(BOTAN_HAS_ENGINE_SIMD) + engines.push_back(new SIMD_Engine); #endif #if defined(BOTAN_HAS_ENGINE_AMD64_ASSEMBLER) diff --git a/src/utils/cpuid.cpp b/src/utils/cpuid.cpp index f79e3a912..2d3b5d92c 100644 --- a/src/utils/cpuid.cpp +++ b/src/utils/cpuid.cpp @@ -98,4 +98,64 @@ u32bit CPUID::cache_line_size() return cl_size; } +bool CPUID::has_altivec() + { + static bool first_time = true; + static bool altivec_capable = false; + + if(first_time) + { +#if defined(BOTAN_TARGET_ARCH_IS_PPC) || defined(BOTAN_TARGET_ARCH_IS_PPC64) + + /* + PVR identifiers for various AltiVec enabled CPUs. Taken from + PearPC and Linux sources, mostly. + */ + const u16bit PVR_G4_7400 = 0x000C; + const u16bit PVR_G5_970 = 0x0039; + const u16bit PVR_G5_970FX = 0x003C; + const u16bit PVR_G5_970MP = 0x0044; + const u16bit PVR_G5_970GX = 0x0045; + const u16bit PVR_POWER6 = 0x003E; + const u16bit PVR_CELL_PPU = 0x0070; + + // Motorola produced G4s with PVR 0x800[0123C] (at least) + const u16bit PVR_G4_74xx_24 = 0x800; + + u32bit pvr = 0; + /* + On PowerPC, MSR 287 is PVR, the Processor Version Number + + Normally it is only accessible to ring 0, but Linux and NetBSD + (at least) will trap and emulate it for us. This is roughly 20x + saner than every other approach I've seen for this (all of which + are entirely OS specific, to boot). + + Apparently OS X doesn't support this, but then again OS X + doesn't really support PPC anymore, so I'm not worrying about it. + + For OSes that aren't (known to) support the emulation, leave pvr + as 0 which will cause all subsequent model number checks to fail. + */ +#if defined(BOTAN_TARGET_OS_IS_LINUX) || defined(BOTAN_TARGET_OS_IS_NETBSD) + asm volatile("mfspr %0, 287" : "=r" (pvr)); +#endif + // Top 16 bit suffice to identify model + pvr >>= 16; + + altivec_capable |= (pvr == PVR_G4_7400); + altivec_capable |= ((pvr >> 8) == PVR_G4_74xx_24); + altivec_capable |= (pvr == PVR_G5_970); + altivec_capable |= (pvr == PVR_G5_970FX); + altivec_capable |= (pvr == PVR_G5_970MP); + altivec_capable |= (pvr == PVR_G5_970GX); + altivec_capable |= (pvr == PVR_CELL_PPU); +#endif + + first_time = false; + } + + return altivec_capable; + } + } diff --git a/src/utils/cpuid.h b/src/utils/cpuid.h index 0b210768a..8b8021754 100644 --- a/src/utils/cpuid.h +++ b/src/utils/cpuid.h @@ -65,6 +65,7 @@ class CPUID static bool has_intel_aes() { return ((x86_processor_flags() >> CPUID_INTEL_AES_BIT) & 1); } + static bool has_altivec(); private: static u64bit x86_processor_flags(); }; diff --git a/src/utils/simd_32/info.txt b/src/utils/simd_32/info.txt new file mode 100644 index 000000000..64707c1e4 --- /dev/null +++ b/src/utils/simd_32/info.txt @@ -0,0 +1,16 @@ +define SIMD_32 + +load_on always + +<arch> +pentium-m +pentium4 +prescott +amd64 +</arch> + +<cc> +gcc +icc +msvc +</cc> diff --git a/src/utils/simd_32/simd_32.h b/src/utils/simd_32/simd_32.h new file mode 100644 index 000000000..be426efd6 --- /dev/null +++ b/src/utils/simd_32/simd_32.h @@ -0,0 +1,32 @@ +/** +* Lightweight wrappers for SIMD operations +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef BOTAN_SIMD_32_H__ +#define BOTAN_SIMD_32_H__ + +#include <botan/types.h> + +//#define BOTAN_TARGET_CPU_HAS_SSE2 + +#if defined(BOTAN_TARGET_CPU_HAS_SSE2) + + #include <botan/simd_sse.h> + namespace Botan { typedef SIMD_SSE2 SIMD_32; } + +#elif defined(BOTAN_TARGET_CPU_HAS_ALTIVEC) + + #include <botan/simd_altivec.h> + namespace Botan { typedef SIMD_Altivec SIMD_32; } + +#else + + #include <botan/simd_scalar.h> + namespace Botan { typedef SIMD_Scalar SIMD_32; } + +#endif + +#endif diff --git a/src/utils/simd_32/simd_altivec.h b/src/utils/simd_32/simd_altivec.h new file mode 100644 index 000000000..e1aa62002 --- /dev/null +++ b/src/utils/simd_32/simd_altivec.h @@ -0,0 +1,202 @@ +/** +* Lightweight wrappers around AltiVec for 32-bit operations +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef BOTAN_SIMD_ALTIVEC_H__ +#define BOTAN_SIMD_ALTIVEC_H__ + +#include <botan/loadstor.h> +#include <botan/cpuid.h> + +#include <altivec.h> +#undef vector + +namespace Botan { + +class SIMD_Altivec + { + public: + bool enabled() const { return CPUID::has_altivec(); } + + SIMD_Altivec(const u32bit B[4]) + { + reg = (__vector unsigned int){B[0], B[1], B[2], B[3]}; + } + + SIMD_Altivec(u32bit B0, u32bit B1, u32bit B2, u32bit B3) + { + reg = (__vector unsigned int){B0, B1, B2, B3}; + } + + SIMD_Altivec(u32bit B) + { + reg = (__vector unsigned int){B, B, B, B}; + } + + static SIMD_Altivec load_le(const void* in) + { + const u32bit* in_32 = static_cast<const u32bit*>(in); + + __vector unsigned int R0 = vec_ld(0, in_32); + __vector unsigned int R1 = vec_ld(12, in_32); + + __vector unsigned char perm = vec_lvsl(0, in_32); + + perm = vec_xor(perm, vec_splat_u8(3)); + + R0 = vec_perm(R0, R1, perm); + + return SIMD_Altivec(R0); + } + + static SIMD_Altivec load_be(const void* in) + { + const u32bit* in_32 = static_cast<const u32bit*>(in); + + __vector unsigned int R0 = vec_ld(0, in_32); + __vector unsigned int R1 = vec_ld(12, in_32); + + __vector unsigned char perm = vec_lvsl(0, in_32); + + R0 = vec_perm(R0, R1, perm); + + return SIMD_Altivec(R0); + } + + void store_le(byte out[]) const + { + __vector unsigned char perm = vec_lvsl(0, (u32bit*)0); + + perm = vec_xor(perm, vec_splat_u8(3)); + + union { + __vector unsigned int V; + u32bit R[4]; + } vec; + + vec.V = vec_perm(reg, reg, perm); + + Botan::store_be(out, vec.R[0], vec.R[1], vec.R[2], vec.R[3]); + } + + void store_be(byte out[]) const + { + union { + __vector unsigned int V; + u32bit R[4]; + } vec; + + vec.V = reg; + + Botan::store_be(out, vec.R[0], vec.R[1], vec.R[2], vec.R[3]); + } + + void rotate_left(u32bit rot) + { + __vector unsigned int rot_vec = + (__vector unsigned int){rot, rot, rot, rot}; + + reg = vec_rl(reg, rot_vec); + } + + void rotate_right(u32bit rot) + { + rotate_left(32 - rot); + } + + void operator+=(const SIMD_Altivec& other) + { + reg = vec_add(reg, other.reg); + } + + SIMD_Altivec operator+(const SIMD_Altivec& other) const + { + return vec_add(reg, other.reg); + } + + void operator-=(const SIMD_Altivec& other) + { + reg = vec_sub(reg, other.reg); + } + + SIMD_Altivec operator-(const SIMD_Altivec& other) const + { + return vec_sub(reg, other.reg); + } + + void operator^=(const SIMD_Altivec& other) + { + reg = vec_xor(reg, other.reg); + } + + SIMD_Altivec operator^(const SIMD_Altivec& other) const + { + return vec_xor(reg, other.reg); + } + + void operator|=(const SIMD_Altivec& other) + { + reg = vec_or(reg, other.reg); + } + + void operator&=(const SIMD_Altivec& other) + { + reg = vec_and(reg, other.reg); + } + + SIMD_Altivec operator<<(u32bit shift) const + { + __vector unsigned int shift_vec = + (__vector unsigned int){shift, shift, shift, shift}; + + return vec_sl(reg, shift_vec); + } + + SIMD_Altivec operator>>(u32bit shift) const + { + __vector unsigned int shift_vec = + (__vector unsigned int){shift, shift, shift, shift}; + + return vec_sr(reg, shift_vec); + } + + SIMD_Altivec operator~() const + { + return vec_nor(reg, reg); + } + + SIMD_Altivec bswap() const + { + __vector unsigned char perm = vec_lvsl(0, (u32bit*)0); + + perm = vec_xor(perm, vec_splat_u8(3)); + + return SIMD_Altivec(vec_perm(reg, reg, perm)); + } + + static void transpose(SIMD_Altivec& B0, SIMD_Altivec& B1, + SIMD_Altivec& B2, SIMD_Altivec& B3) + { + __vector unsigned int T0 = vec_mergeh(B0.reg, B2.reg); + __vector unsigned int T1 = vec_mergel(B0.reg, B2.reg); + __vector unsigned int T2 = vec_mergeh(B1.reg, B3.reg); + __vector unsigned int T3 = vec_mergel(B1.reg, B3.reg); + + B0.reg = vec_mergeh(T0, T2); + B1.reg = vec_mergel(T0, T2); + B2.reg = vec_mergeh(T1, T3); + B3.reg = vec_mergel(T1, T3); + } + + private: + SIMD_Altivec(__vector unsigned int input) { reg = input; } + + __vector unsigned int reg; + }; + +} + +#endif diff --git a/src/utils/simd_32/simd_scalar.h b/src/utils/simd_32/simd_scalar.h new file mode 100644 index 000000000..5fc20b462 --- /dev/null +++ b/src/utils/simd_32/simd_scalar.h @@ -0,0 +1,202 @@ +/** +* Scalar emulation of SIMD 32-bit operations +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef BOTAN_SIMD_SCALAR_H__ +#define BOTAN_SIMD_SCALAR_H__ + +#include <botan/loadstor.h> +#include <botan/bswap.h> + +namespace Botan { + +class SIMD_Scalar + { + public: + bool enabled() const { return true; } + + SIMD_Scalar(const u32bit B[4]) + { + R0 = B[0]; + R1 = B[1]; + R2 = B[2]; + R3 = B[3]; + } + + SIMD_Scalar(u32bit B0, u32bit B1, u32bit B2, u32bit B3) + { + R0 = B0; + R1 = B1; + R2 = B2; + R3 = B3; + } + + SIMD_Scalar(u32bit B) + { + R0 = B; + R1 = B; + R2 = B; + R3 = B; + } + + static SIMD_Scalar load_le(const void* in) + { + const byte* in_b = static_cast<const byte*>(in); + return SIMD_Scalar(Botan::load_le<u32bit>(in_b, 0), + Botan::load_le<u32bit>(in_b, 1), + Botan::load_le<u32bit>(in_b, 2), + Botan::load_le<u32bit>(in_b, 3)); + } + + static SIMD_Scalar load_be(const void* in) + { + const byte* in_b = static_cast<const byte*>(in); + return SIMD_Scalar(Botan::load_be<u32bit>(in_b, 0), + Botan::load_be<u32bit>(in_b, 1), + Botan::load_be<u32bit>(in_b, 2), + Botan::load_be<u32bit>(in_b, 3)); + } + + void store_le(byte out[]) const + { + Botan::store_le(out, R0, R1, R2, R3); + } + + void store_be(byte out[]) const + { + Botan::store_be(out, R0, R1, R2, R3); + } + + void rotate_left(u32bit rot) + { + R0 = Botan::rotate_left(R0, rot); + R1 = Botan::rotate_left(R1, rot); + R2 = Botan::rotate_left(R2, rot); + R3 = Botan::rotate_left(R3, rot); + } + + void rotate_right(u32bit rot) + { + R0 = Botan::rotate_right(R0, rot); + R1 = Botan::rotate_right(R1, rot); + R2 = Botan::rotate_right(R2, rot); + R3 = Botan::rotate_right(R3, rot); + } + + void operator+=(const SIMD_Scalar& other) + { + R0 += other.R0; + R1 += other.R1; + R2 += other.R2; + R3 += other.R3; + } + + SIMD_Scalar operator+(const SIMD_Scalar& other) const + { + return SIMD_Scalar(R0 + other.R0, + R1 + other.R1, + R2 + other.R2, + R3 + other.R3); + } + + void operator-=(const SIMD_Scalar& other) + { + R0 -= other.R0; + R1 -= other.R1; + R2 -= other.R2; + R3 -= other.R3; + } + + SIMD_Scalar operator-(const SIMD_Scalar& other) const + { + return SIMD_Scalar(R0 - other.R0, + R1 - other.R1, + R2 - other.R2, + R3 - other.R3); + } + + void operator^=(const SIMD_Scalar& other) + { + R0 ^= other.R0; + R1 ^= other.R1; + R2 ^= other.R2; + R3 ^= other.R3; + } + + SIMD_Scalar operator^(const SIMD_Scalar& other) const + { + return SIMD_Scalar(R0 ^ other.R0, + R1 ^ other.R1, + R2 ^ other.R2, + R3 ^ other.R3); + } + + void operator|=(const SIMD_Scalar& other) + { + R0 |= other.R0; + R1 |= other.R1; + R2 |= other.R2; + R3 |= other.R3; + } + + void operator&=(const SIMD_Scalar& other) + { + R0 &= other.R0; + R1 &= other.R1; + R2 &= other.R2; + R3 &= other.R3; + } + + SIMD_Scalar operator<<(u32bit shift) const + { + return SIMD_Scalar(R0 << shift, + R1 << shift, + R2 << shift, + R3 << shift); + } + + SIMD_Scalar operator>>(u32bit shift) const + { + return SIMD_Scalar(R0 >> shift, + R1 >> shift, + R2 >> shift, + R3 >> shift); + } + + SIMD_Scalar operator~() const + { + return SIMD_Scalar(~R0, ~R1, ~R2, ~R3); + } + + SIMD_Scalar bswap() const + { + return SIMD_Scalar(reverse_bytes(R0), + reverse_bytes(R1), + reverse_bytes(R2), + reverse_bytes(R3)); + } + + static void transpose(SIMD_Scalar& B0, SIMD_Scalar& B1, + SIMD_Scalar& B2, SIMD_Scalar& B3) + { + SIMD_Scalar T0(B0.R0, B1.R0, B2.R0, B3.R0); + SIMD_Scalar T1(B0.R1, B1.R1, B2.R1, B3.R1); + SIMD_Scalar T2(B0.R2, B1.R2, B2.R2, B3.R2); + SIMD_Scalar T3(B0.R3, B1.R3, B2.R3, B3.R3); + + B0 = T0; + B1 = T1; + B2 = T2; + B3 = T3; + } + + private: + u32bit R0, R1, R2, R3; + }; + +} + +#endif diff --git a/src/utils/simd_32/simd_sse.h b/src/utils/simd_32/simd_sse.h new file mode 100644 index 000000000..c45d8032f --- /dev/null +++ b/src/utils/simd_32/simd_sse.h @@ -0,0 +1,156 @@ +/** +* Lightweight wrappers for SSE2 intrinsics for 32-bit operations +* (C) 2009 Jack Lloyd +* +* Distributed under the terms of the Botan license +*/ + +#ifndef BOTAN_SIMD_SSE_H__ +#define BOTAN_SIMD_SSE_H__ + +#include <botan/cpuid.h> + +#include <emmintrin.h> + +namespace Botan { + +class SIMD_SSE2 + { + public: + bool enabled() const { return CPUID::has_sse2(); } + + SIMD_SSE2(const u32bit B[4]) + { + reg = _mm_loadu_si128((const __m128i*)B); + } + + SIMD_SSE2(u32bit B0, u32bit B1, u32bit B2, u32bit B3) + { + reg = _mm_set_epi32(B0, B1, B2, B3); + } + + SIMD_SSE2(u32bit B) + { + reg = _mm_set1_epi32(B); + } + + static SIMD_SSE2 load_le(const void* in) + { + return _mm_loadu_si128((const __m128i*)in); + } + + static SIMD_SSE2 load_be(const void* in) + { + return load_le(in).bswap(); + } + + void store_le(byte out[]) const + { + _mm_storeu_si128((__m128i*)out, reg); + } + + void store_be(byte out[]) const + { + bswap().store_le(out); + } + + void rotate_left(u32bit rot) + { + reg = _mm_or_si128(_mm_slli_epi32(reg, rot), + _mm_srli_epi32(reg, 32-rot)); + } + + void rotate_right(u32bit rot) + { + rotate_left(32 - rot); + } + + void operator+=(const SIMD_SSE2& other) + { + reg = _mm_add_epi32(reg, other.reg); + } + + SIMD_SSE2 operator+(const SIMD_SSE2& other) const + { + return _mm_add_epi32(reg, other.reg); + } + + void operator-=(const SIMD_SSE2& other) + { + reg = _mm_sub_epi32(reg, other.reg); + } + + SIMD_SSE2 operator-(const SIMD_SSE2& other) const + { + return _mm_sub_epi32(reg, other.reg); + } + + void operator^=(const SIMD_SSE2& other) + { + reg = _mm_xor_si128(reg, other.reg); + } + + SIMD_SSE2 operator^(const SIMD_SSE2& other) const + { + return _mm_xor_si128(reg, other.reg); + } + + void operator|=(const SIMD_SSE2& other) + { + reg = _mm_or_si128(reg, other.reg); + } + + void operator&=(const SIMD_SSE2& other) + { + reg = _mm_and_si128(reg, other.reg); + } + + SIMD_SSE2 operator<<(u32bit shift) const + { + return _mm_slli_epi32(reg, shift); + } + + SIMD_SSE2 operator>>(u32bit shift) const + { + return _mm_srli_epi32(reg, shift); + } + + SIMD_SSE2 operator~() const + { + static const __m128i all_ones = _mm_set1_epi32(0xFFFFFFFF); + return _mm_xor_si128(reg, all_ones); + } + + SIMD_SSE2 bswap() const + { + __m128i T = reg; + + T = _mm_shufflehi_epi16(T, _MM_SHUFFLE(2, 3, 0, 1)); + T = _mm_shufflelo_epi16(T, _MM_SHUFFLE(2, 3, 0, 1)); + + return _mm_or_si128(_mm_srli_epi16(T, 8), + _mm_slli_epi16(T, 8)); + } + + static void transpose(SIMD_SSE2& B0, SIMD_SSE2& B1, + SIMD_SSE2& B2, SIMD_SSE2& B3) + { + __m128i T0 = _mm_unpacklo_epi32(B0.reg, B1.reg); + __m128i T1 = _mm_unpacklo_epi32(B2.reg, B3.reg); + __m128i T2 = _mm_unpackhi_epi32(B0.reg, B1.reg); + __m128i T3 = _mm_unpackhi_epi32(B2.reg, B3.reg); + B0.reg = _mm_unpacklo_epi64(T0, T1); + B1.reg = _mm_unpackhi_epi64(T0, T1); + B2.reg = _mm_unpacklo_epi64(T2, T3); + B3.reg = _mm_unpackhi_epi64(T2, T3); + } + + private: + SIMD_SSE2(__m128i in) { reg = in; } + + __m128i reg; + }; + +} + +#endif |