diff options
Diffstat (limited to 'src/Crypto')
| -rw-r--r-- | src/Crypto/Aes_hw_armv8.c | 316 | ||||
| -rw-r--r-- | src/Crypto/Aes_hw_cpu.h | 2 | ||||
| -rw-r--r-- | src/Crypto/Crypto.vcxproj | 4 | ||||
| -rw-r--r-- | src/Crypto/Crypto.vcxproj.filters | 15 | ||||
| -rw-r--r-- | src/Crypto/config.h | 39 | ||||
| -rw-r--r-- | src/Crypto/cpu.c | 38 | ||||
| -rw-r--r-- | src/Crypto/cpu.h | 18 |
7 files changed, 422 insertions, 10 deletions
diff --git a/src/Crypto/Aes_hw_armv8.c b/src/Crypto/Aes_hw_armv8.c new file mode 100644 index 00000000..b67ed1a5 --- /dev/null +++ b/src/Crypto/Aes_hw_armv8.c @@ -0,0 +1,316 @@ +/* +* AES using ARMv8 +* Contributed by Jeffrey Walton +* +* Further changes +* (C) 2017,2018 Jack Lloyd +* +* Botan is released under the Simplified BSD License (see license.txt) +*/ + +/* Modified and adapted for VeraCrypt */ + +#include "Common/Tcdefs.h" +#include "Aes_hw_cpu.h" +#if !defined(_UEFI) +#include <memory.h> +#include <stdlib.h> +#endif +#include "cpu.h" +#include "misc.h" + +#if CRYPTOPP_ARM_AES_AVAILABLE + +#include <arm_neon.h> + +// Single block encryption operations +VC_INLINE void aes_enc_block(uint8x16_t* B, uint8x16_t K) +{ + *B = vaesmcq_u8(vaeseq_u8(*B, K)); +} + +VC_INLINE void aes_enc_block_last(uint8x16_t* B, uint8x16_t K, uint8x16_t K2) +{ + *B = veorq_u8(vaeseq_u8(*B, K), K2); +} + +// 4-block parallel encryption operations +VC_INLINE void aes_enc_4_blocks(uint8x16_t* B0, uint8x16_t* B1, + uint8x16_t* B2, uint8x16_t* B3, uint8x16_t K) +{ + *B0 = vaesmcq_u8(vaeseq_u8(*B0, K)); + *B1 = vaesmcq_u8(vaeseq_u8(*B1, K)); + *B2 = vaesmcq_u8(vaeseq_u8(*B2, K)); + *B3 = vaesmcq_u8(vaeseq_u8(*B3, K)); +} + +VC_INLINE void aes_enc_4_blocks_last(uint8x16_t* B0, uint8x16_t* B1, + uint8x16_t* B2, uint8x16_t* B3, + uint8x16_t K, uint8x16_t K2) +{ + *B0 = veorq_u8(vaeseq_u8(*B0, K), K2); + *B1 = veorq_u8(vaeseq_u8(*B1, K), K2); + *B2 = veorq_u8(vaeseq_u8(*B2, K), K2); + *B3 = veorq_u8(vaeseq_u8(*B3, K), K2); +} + +// Single block decryption operations +VC_INLINE void aes_dec_block(uint8x16_t* B, uint8x16_t K) +{ + *B = vaesimcq_u8(vaesdq_u8(*B, K)); +} + +VC_INLINE void aes_dec_block_last(uint8x16_t* B, uint8x16_t K, uint8x16_t K2) +{ + *B = veorq_u8(vaesdq_u8(*B, K), K2); +} + +// 4-block parallel decryption operations +VC_INLINE void aes_dec_4_blocks(uint8x16_t* B0, uint8x16_t* B1, + uint8x16_t* B2, uint8x16_t* B3, uint8x16_t K) +{ + *B0 = vaesimcq_u8(vaesdq_u8(*B0, K)); + *B1 = vaesimcq_u8(vaesdq_u8(*B1, K)); + *B2 = vaesimcq_u8(vaesdq_u8(*B2, K)); + *B3 = vaesimcq_u8(vaesdq_u8(*B3, K)); +} + +VC_INLINE void aes_dec_4_blocks_last(uint8x16_t* B0, uint8x16_t* B1, + uint8x16_t* B2, uint8x16_t* B3, + uint8x16_t K, uint8x16_t K2) +{ + *B0 = veorq_u8(vaesdq_u8(*B0, K), K2); + *B1 = veorq_u8(vaesdq_u8(*B1, K), K2); + *B2 = veorq_u8(vaesdq_u8(*B2, K), K2); + *B3 = veorq_u8(vaesdq_u8(*B3, K), K2); +} + +VC_INLINE void aes256_hw_encrypt_blocks(uint8 buffer[], size_t blocks, const uint8* ks) +{ + const uint8x16_t K0 = vld1q_u8(ks + 0 * 16); + const uint8x16_t K1 = vld1q_u8(ks + 1 * 16); + const uint8x16_t K2 = vld1q_u8(ks + 2 * 16); + const uint8x16_t K3 = vld1q_u8(ks + 3 * 16); + const uint8x16_t K4 = vld1q_u8(ks + 4 * 16); + const uint8x16_t K5 = vld1q_u8(ks + 5 * 16); + const uint8x16_t K6 = vld1q_u8(ks + 6 * 16); + const uint8x16_t K7 = vld1q_u8(ks + 7 * 16); + const uint8x16_t K8 = vld1q_u8(ks + 8 * 16); + const uint8x16_t K9 = vld1q_u8(ks + 9 * 16); + const uint8x16_t K10 = vld1q_u8(ks + 10 * 16); + const uint8x16_t K11 = vld1q_u8(ks + 11 * 16); + const uint8x16_t K12 = vld1q_u8(ks + 12 * 16); + const uint8x16_t K13 = vld1q_u8(ks + 13 * 16); + const uint8x16_t K14 = vld1q_u8(ks + 14 * 16); + + while(blocks >= 4) { + uint8x16_t B0 = vld1q_u8(buffer); + uint8x16_t B1 = vld1q_u8(buffer + 16); + uint8x16_t B2 = vld1q_u8(buffer + 32); + uint8x16_t B3 = vld1q_u8(buffer + 48); + + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K0); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K1); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K2); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K3); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K4); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K5); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K6); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K7); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K8); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K9); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K10); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K11); + aes_enc_4_blocks(&B0, &B1, &B2, &B3, K12); + aes_enc_4_blocks_last(&B0, &B1, &B2, &B3, K13, K14); + + vst1q_u8(buffer, B0); + vst1q_u8(buffer + 16, B1); + vst1q_u8(buffer + 32, B2); + vst1q_u8(buffer + 48, B3); + + buffer += 16 * 4; + blocks -= 4; + } + + for(size_t i = 0; i != blocks; ++i) { + uint8x16_t B = vld1q_u8(buffer + 16 * i); + aes_enc_block(&B, K0); + aes_enc_block(&B, K1); + aes_enc_block(&B, K2); + aes_enc_block(&B, K3); + aes_enc_block(&B, K4); + aes_enc_block(&B, K5); + aes_enc_block(&B, K6); + aes_enc_block(&B, K7); + aes_enc_block(&B, K8); + aes_enc_block(&B, K9); + aes_enc_block(&B, K10); + aes_enc_block(&B, K11); + aes_enc_block(&B, K12); + aes_enc_block_last(&B, K13, K14); + vst1q_u8(buffer + 16 * i, B); + } +} + +VC_INLINE void aes256_hw_encrypt_block(uint8 buffer[], const uint8* ks) +{ + const uint8x16_t K0 = vld1q_u8(ks + 0 * 16); + const uint8x16_t K1 = vld1q_u8(ks + 1 * 16); + const uint8x16_t K2 = vld1q_u8(ks + 2 * 16); + const uint8x16_t K3 = vld1q_u8(ks + 3 * 16); + const uint8x16_t K4 = vld1q_u8(ks + 4 * 16); + const uint8x16_t K5 = vld1q_u8(ks + 5 * 16); + const uint8x16_t K6 = vld1q_u8(ks + 6 * 16); + const uint8x16_t K7 = vld1q_u8(ks + 7 * 16); + const uint8x16_t K8 = vld1q_u8(ks + 8 * 16); + const uint8x16_t K9 = vld1q_u8(ks + 9 * 16); + const uint8x16_t K10 = vld1q_u8(ks + 10 * 16); + const uint8x16_t K11 = vld1q_u8(ks + 11 * 16); + const uint8x16_t K12 = vld1q_u8(ks + 12 * 16); + const uint8x16_t K13 = vld1q_u8(ks + 13 * 16); + const uint8x16_t K14 = vld1q_u8(ks + 14 * 16); + + uint8x16_t B = vld1q_u8(buffer); + aes_enc_block(&B, K0); + aes_enc_block(&B, K1); + aes_enc_block(&B, K2); + aes_enc_block(&B, K3); + aes_enc_block(&B, K4); + aes_enc_block(&B, K5); + aes_enc_block(&B, K6); + aes_enc_block(&B, K7); + aes_enc_block(&B, K8); + aes_enc_block(&B, K9); + aes_enc_block(&B, K10); + aes_enc_block(&B, K11); + aes_enc_block(&B, K12); + aes_enc_block_last(&B, K13, K14); + vst1q_u8(buffer, B); +} + +VC_INLINE void aes256_hw_decrypt_blocks(uint8 buffer[], size_t blocks, const uint8* ks) +{ + const uint8x16_t K0 = vld1q_u8(ks + 0 * 16); + const uint8x16_t K1 = vld1q_u8(ks + 1 * 16); + const uint8x16_t K2 = vld1q_u8(ks + 2 * 16); + const uint8x16_t K3 = vld1q_u8(ks + 3 * 16); + const uint8x16_t K4 = vld1q_u8(ks + 4 * 16); + const uint8x16_t K5 = vld1q_u8(ks + 5 * 16); + const uint8x16_t K6 = vld1q_u8(ks + 6 * 16); + const uint8x16_t K7 = vld1q_u8(ks + 7 * 16); + const uint8x16_t K8 = vld1q_u8(ks + 8 * 16); + const uint8x16_t K9 = vld1q_u8(ks + 9 * 16); + const uint8x16_t K10 = vld1q_u8(ks + 10 * 16); + const uint8x16_t K11 = vld1q_u8(ks + 11 * 16); + const uint8x16_t K12 = vld1q_u8(ks + 12 * 16); + const uint8x16_t K13 = vld1q_u8(ks + 13 * 16); + const uint8x16_t K14 = vld1q_u8(ks + 14 * 16); + + while(blocks >= 4) { + uint8x16_t B0 = vld1q_u8(buffer); + uint8x16_t B1 = vld1q_u8(buffer + 16); + uint8x16_t B2 = vld1q_u8(buffer + 32); + uint8x16_t B3 = vld1q_u8(buffer + 48); + + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K0); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K1); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K2); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K3); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K4); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K5); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K6); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K7); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K8); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K9); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K10); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K11); + aes_dec_4_blocks(&B0, &B1, &B2, &B3, K12); + aes_dec_4_blocks_last(&B0, &B1, &B2, &B3, K13, K14); + + vst1q_u8(buffer, B0); + vst1q_u8(buffer + 16, B1); + vst1q_u8(buffer + 32, B2); + vst1q_u8(buffer + 48, B3); + + buffer += 16 * 4; + blocks -= 4; + } + + for(size_t i = 0; i != blocks; ++i) { + uint8x16_t B = vld1q_u8(buffer + 16 * i); + aes_dec_block(&B, K0); + aes_dec_block(&B, K1); + aes_dec_block(&B, K2); + aes_dec_block(&B, K3); + aes_dec_block(&B, K4); + aes_dec_block(&B, K5); + aes_dec_block(&B, K6); + aes_dec_block(&B, K7); + aes_dec_block(&B, K8); + aes_dec_block(&B, K9); + aes_dec_block(&B, K10); + aes_dec_block(&B, K11); + aes_dec_block(&B, K12); + aes_dec_block_last(&B, K13, K14); + vst1q_u8(buffer + 16 * i, B); + } +} + +VC_INLINE void aes256_hw_decrypt_block(uint8 buffer[], const uint8* ks) +{ + const uint8x16_t K0 = vld1q_u8(ks + 0 * 16); + const uint8x16_t K1 = vld1q_u8(ks + 1 * 16); + const uint8x16_t K2 = vld1q_u8(ks + 2 * 16); + const uint8x16_t K3 = vld1q_u8(ks + 3 * 16); + const uint8x16_t K4 = vld1q_u8(ks + 4 * 16); + const uint8x16_t K5 = vld1q_u8(ks + 5 * 16); + const uint8x16_t K6 = vld1q_u8(ks + 6 * 16); + const uint8x16_t K7 = vld1q_u8(ks + 7 * 16); + const uint8x16_t K8 = vld1q_u8(ks + 8 * 16); + const uint8x16_t K9 = vld1q_u8(ks + 9 * 16); + const uint8x16_t K10 = vld1q_u8(ks + 10 * 16); + const uint8x16_t K11 = vld1q_u8(ks + 11 * 16); + const uint8x16_t K12 = vld1q_u8(ks + 12 * 16); + const uint8x16_t K13 = vld1q_u8(ks + 13 * 16); + const uint8x16_t K14 = vld1q_u8(ks + 14 * 16); + + uint8x16_t B = vld1q_u8(buffer); + aes_dec_block(&B, K0); + aes_dec_block(&B, K1); + aes_dec_block(&B, K2); + aes_dec_block(&B, K3); + aes_dec_block(&B, K4); + aes_dec_block(&B, K5); + aes_dec_block(&B, K6); + aes_dec_block(&B, K7); + aes_dec_block(&B, K8); + aes_dec_block(&B, K9); + aes_dec_block(&B, K10); + aes_dec_block(&B, K11); + aes_dec_block(&B, K12); + aes_dec_block_last(&B, K13, K14); + vst1q_u8(buffer, B); +} + +void aes_hw_cpu_decrypt (const uint8 *ks, uint8 *data) +{ + aes256_hw_decrypt_block(data, ks); +} + +void aes_hw_cpu_decrypt_32_blocks (const uint8 *ks, uint8 *data) +{ + aes256_hw_decrypt_blocks(data, 32, ks); +} + +void aes_hw_cpu_encrypt (const uint8 *ks, uint8 *data) +{ + aes256_hw_encrypt_block(data, ks); +} + +void aes_hw_cpu_encrypt_32_blocks (const uint8 *ks, uint8 *data) +{ + aes256_hw_encrypt_blocks(data, 32, ks); +} + +#endif diff --git a/src/Crypto/Aes_hw_cpu.h b/src/Crypto/Aes_hw_cpu.h index 685a1dff..d9dda1af 100644 --- a/src/Crypto/Aes_hw_cpu.h +++ b/src/Crypto/Aes_hw_cpu.h @@ -22,8 +22,8 @@ extern "C" #if defined (TC_WINDOWS_BOOT) uint8 is_aes_hw_cpu_supported (); -#endif void aes_hw_cpu_enable_sse (); +#endif void aes_hw_cpu_decrypt (const uint8 *ks, uint8 *data); void VC_CDECL aes_hw_cpu_decrypt_32_blocks (const uint8 *ks, uint8 *data); void aes_hw_cpu_encrypt (const uint8 *ks, uint8 *data); diff --git a/src/Crypto/Crypto.vcxproj b/src/Crypto/Crypto.vcxproj index 4aebc084..c17bd607 100644 --- a/src/Crypto/Crypto.vcxproj +++ b/src/Crypto/Crypto.vcxproj @@ -226,6 +226,10 @@ </ClCompile> <ClCompile Include="Aeskey.c" /> <ClCompile Include="Aestab.c" /> + <ClCompile Include="Aes_hw_armv8.c"> + <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">true</ExcludedFromBuild> + <ExcludedFromBuild Condition="'$(Configuration)|$(Platform)'=='Release|x64'">true</ExcludedFromBuild> + </ClCompile> <ClCompile Include="blake2s.c" /> <ClCompile Include="blake2s_SSE2.c" /> <ClCompile Include="blake2s_SSE41.c" /> diff --git a/src/Crypto/Crypto.vcxproj.filters b/src/Crypto/Crypto.vcxproj.filters index 3d384f97..f2b1b54d 100644 --- a/src/Crypto/Crypto.vcxproj.filters +++ b/src/Crypto/Crypto.vcxproj.filters @@ -90,6 +90,12 @@ <ClCompile Include="Sha2Intel.c"> <Filter>Source Files</Filter> </ClCompile> + <ClCompile Include="Aescrypt.c"> + <Filter>Source Files</Filter> + </ClCompile> + <ClCompile Include="Aes_hw_armv8.c"> + <Filter>Source Files</Filter> + </ClCompile> </ItemGroup> <ItemGroup> <ClInclude Include="Aes.h"> @@ -167,15 +173,6 @@ <ClInclude Include="t1ha_selfcheck.h"> <Filter>Header Files</Filter> </ClInclude> - <ClInclude Include="blake2s-load-sse2.h"> - <Filter>Header Files</Filter> - </ClInclude> - <ClInclude Include="blake2s-load-sse41.h"> - <Filter>Header Files</Filter> - </ClInclude> - <ClInclude Include="blake2s-round.h"> - <Filter>Header Files</Filter> - </ClInclude> </ItemGroup> <ItemGroup> <CustomBuild Include="Aes_hw_cpu.asm"> diff --git a/src/Crypto/config.h b/src/Crypto/config.h index 1c2aff72..dd8e3f06 100644 --- a/src/Crypto/config.h +++ b/src/Crypto/config.h @@ -29,6 +29,11 @@ #define CRYPTOPP_CLANG_INTEGRATED_ASSEMBLER 1 #endif +#if defined(_MSC_VER) && !defined(__clang__) +# undef CRYPTOPP_LLVM_CLANG_VERSION +# define CRYPTOPP_MSC_VERSION (_MSC_VER) +#endif + // Clang due to "Inline assembly operands don't work with .intel_syntax", http://llvm.org/bugs/show_bug.cgi?id=24232 // TODO: supply the upper version when LLVM fixes it. We set it to 20.0 for compilation purposes. #if (defined(CRYPTOPP_LLVM_CLANG_VERSION) && CRYPTOPP_LLVM_CLANG_VERSION <= 200000) || (defined(CRYPTOPP_APPLE_CLANG_VERSION) && CRYPTOPP_APPLE_CLANG_VERSION <= 200000) || defined(CRYPTOPP_CLANG_INTEGRATED_ASSEMBLER) @@ -201,6 +206,40 @@ #define CRYPTOPP_BOOL_X64 0 #endif +#if defined(__arm64__) || defined(__aarch64__) || defined(_M_ARM64) + #define CRYPTOPP_BOOL_ARMV8 1 + #define CRYPTOPP_BOOL_ARM64 1 +#else + #define CRYPTOPP_BOOL_ARMV8 0 + #define CRYPTOPP_BOOL_ARM64 0 +#endif + +// ARMv8 and ASIMD. -march=armv8-a or above must be present +// Requires GCC 4.8, Clang 3.3 or Visual Studio 2017 +// Do not use APPLE_CLANG_VERSION; use __ARM_FEATURE_XXX instead. +#if !defined(CRYPTOPP_ARM_ASIMD_AVAILABLE) && !defined(CRYPTOPP_DISABLE_ARM_ASIMD) +# if defined(__aarch32__) || defined(__aarch64__) || defined(__arm64__) || defined(_M_ARM64) +# if defined(__ARM_NEON) || defined(__ARM_ASIMD) || defined(__ARM_FEATURE_NEON) || defined(__ARM_FEATURE_ASIMD) || \ + (CRYPTOPP_GCC_VERSION >= 40800) || (CRYPTOPP_LLVM_CLANG_VERSION >= 30300) || \ + (CRYPTOPP_APPLE_CLANG_VERSION >= 40000) || (CRYPTOPP_MSC_VERSION >= 1916) +# define CRYPTOPP_ARM_NEON_AVAILABLE 1 +# define CRYPTOPP_ARM_ASIMD_AVAILABLE 1 +# endif // Compilers +# endif // Platforms +#endif + +// ARMv8 and AES. -march=armv8-a+crypto or above must be present +// Requires GCC 4.8, Clang 3.3 or Visual Studio 2017 +#if !defined(CRYPTOPP_ARM_AES_AVAILABLE) && !defined(CRYPTOPP_DISABLE_ARM_AES) +# if defined(__aarch32__) || defined(__aarch64__) || defined(_M_ARM64) +# if defined(__ARM_FEATURE_CRYPTO) || (CRYPTOPP_GCC_VERSION >= 40800) || \ + (CRYPTOPP_LLVM_CLANG_VERSION >= 30300) || (CRYPTOPP_APPLE_CLANG_VERSION >= 40300) || \ + (CRYPTOPP_MSC_VERSION >= 1916) +# define CRYPTOPP_ARM_AES_AVAILABLE 1 +# endif // Compilers +# endif // Platforms +#endif + // Undo the ASM and Intrinsic related defines due to X32. #if CRYPTOPP_BOOL_X32 # undef CRYPTOPP_BOOL_X64 diff --git a/src/Crypto/cpu.c b/src/Crypto/cpu.c index e611e9bb..85278a92 100644 --- a/src/Crypto/cpu.c +++ b/src/Crypto/cpu.c @@ -469,3 +469,41 @@ void DisableCPUExtendedFeatures () #endif +#if CRYPTOPP_BOOL_ARMV8 + +volatile int g_hasAESARM = 0; + +#ifndef HWCAP_AES +# define HWCAP_AES (1 << 3) +#endif + +inline int CPU_QueryAES() +{ +#if defined(CRYPTOPP_ARM_AES_AVAILABLE) +#if defined(__linux__) && defined(__aarch64__) + if ((getauxval(AT_HWCAP) & HWCAP_AES) != 0) + return 1; +#elif defined(__APPLE__) && defined(__aarch64__) + // Apple Sillcon (M1) and later + return 1; +#elif defined(_WIN32) && defined(_M_ARM64) +#ifdef TC_WINDOWS_DRIVER + if (ExIsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) != 0) + return 1; +#else + if (IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) != 0) + return 1; +#endif +#endif + return 0; +#else + return 0; +#endif +} + +void DetectArmFeatures() +{ + g_hasAESARM = CPU_QueryAES(); +} + +#endif
\ No newline at end of file diff --git a/src/Crypto/cpu.h b/src/Crypto/cpu.h index b0df6462..761258f2 100644 --- a/src/Crypto/cpu.h +++ b/src/Crypto/cpu.h @@ -288,6 +288,24 @@ void DisableCPUExtendedFeatures (); } #endif +#elif CRYPTOPP_BOOL_ARMV8 +#if defined(__cplusplus) +extern "C" { +#endif + +#if !defined(CRYPTOPP_DISABLE_AESNI) && !defined(WOLFCRYPT_BACKEND) +#define TC_AES_HW_CPU +#endif + +extern volatile int g_hasAESARM; +void DetectArmFeatures(); + +#define HasAESNI() g_hasAESARM + +#if defined(__cplusplus) +} +#endif + #else #define HasSSE2() 0 |