diff options
| author | Jeffrey Walton <noloader@gmail.com> | 2017-08-17 12:33:43 -0400 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2017-08-17 12:33:43 -0400 |
| commit | e2c377effd9a4eca0291c217e9c3256b033511dc (patch) | |
| tree | a33fa7d52e14b923a034bafb9a05172a983a0e7c /shacal2-simd.cpp | |
| parent | 74d21068a16c2b3e879336ae7eed180d4466e8b5 (diff) | |
| download | cryptopp-git-e2c377effd9a4eca0291c217e9c3256b033511dc.tar.gz | |
Split source files to support Base Implementation + SIMD implementation (GH #461)
Split source files to support Base Implementation + SIMD implementation
Diffstat (limited to 'shacal2-simd.cpp')
| -rw-r--r-- | shacal2-simd.cpp | 111 |
1 files changed, 111 insertions, 0 deletions
diff --git a/shacal2-simd.cpp b/shacal2-simd.cpp new file mode 100644 index 00000000..9597cc90 --- /dev/null +++ b/shacal2-simd.cpp @@ -0,0 +1,111 @@ +// shacla2-simd.cpp - written and placed in the public domain by
+// Jeffrey Walton and Jack Lloyd
+//
+// Jack Lloyd and the Botan team allowed Crypto++ to use parts of
+// Botan's implementation under the same license as Crypto++
+// is released. The code for SHACAL2_Enc_ProcessAndXorBlock_SHANI
+// below is Botan's x86_encrypt_blocks with minor tweaks. Many thanks
+// to the Botan team. Also see http://github.com/randombit/botan/.
+//
+// This source file uses intrinsics to gain access to SHA-NI and
+// ARMv8a SHA instructions. A separate source file is needed because
+// additional CXXFLAGS are required to enable the appropriate instructions
+// sets in some build configurations.
+
+#include "pch.h"
+#include "config.h"
+#include "sha.h"
+#include "misc.h"
+
+// Clang and GCC hoops...
+#if !(defined(__ARM_FEATURE_CRYPTO) || defined(_MSC_VER))
+# undef CRYPTOPP_ARM_SHA_AVAILABLE
+#endif
+
+#if (CRYPTOPP_SHANI_AVAILABLE)
+# include "nmmintrin.h"
+# include "immintrin.h"
+#endif
+
+#if (CRYPTOPP_ARM_SHA_AVAILABLE)
+# include "arm_neon.h"
+#endif
+
+// Don't include <arm_acle.h> when using Apple Clang. Early Apple compilers
+// fail to compile with <arm_acle.h> included. Later Apple compilers compile
+// intrinsics without <arm_acle.h> included.
+#if (CRYPTOPP_ARM_SHA_AVAILABLE) && !defined(CRYPTOPP_APPLE_CLANG_VERSION)
+# include "arm_acle.h"
+#endif
+
+// Clang __m128i casts
+#define M128_CAST(x) ((__m128i *)(void *)(x))
+#define CONST_M128_CAST(x) ((const __m128i *)(const void *)(x))
+
+NAMESPACE_BEGIN(CryptoPP)
+
+#if CRYPTOPP_SHANI_AVAILABLE
+void SHACAL2_Enc_ProcessAndXorBlock_SHANI(const word32* subKeys, const byte *inBlock, const byte *xorBlock, byte *outBlock)
+{
+ CRYPTOPP_ASSERT(subKeys);
+ CRYPTOPP_ASSERT(inBlock);
+ CRYPTOPP_ASSERT(outBlock);
+
+ const __m128i MASK1 = _mm_set_epi8(8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7);
+ const __m128i MASK2 = _mm_set_epi8(0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15);
+
+ __m128i B0 = _mm_shuffle_epi8(_mm_loadu_si128(CONST_M128_CAST(inBlock + 0)), MASK1);
+ __m128i B1 = _mm_shuffle_epi8(_mm_loadu_si128(CONST_M128_CAST(inBlock + 16)), MASK2);
+
+ __m128i TMP = _mm_alignr_epi8(B0, B1, 8);
+ B1 = _mm_blend_epi16(B1, B0, 0xF0);
+ B0 = TMP;
+
+#if 0
+ // SSE2 + SSSE3, but 0.2 cpb slower on a Celeraon J3455
+ const __m128i MASK1 = _mm_set_epi8(8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7);
+ const __m128i MASK2 = _mm_set_epi8(0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15);
+
+ __m128i B0 = _mm_loadu_si128(CONST_M128_CAST(inBlock + 0));
+ __m128i B1 = _mm_loadu_si128(CONST_M128_CAST(inBlock + 16));
+
+ __m128i TMP = _mm_shuffle_epi8(_mm_unpacklo_epi64(B0, B1), MASK2);
+ B1 = _mm_shuffle_epi8(_mm_unpackhi_epi64(B0, B1), MASK2);
+ B0 = TMP;
+#endif
+
+ const byte* keys = reinterpret_cast<const byte*>(subKeys);
+ for (size_t i = 0; i != 8; ++i)
+ {
+ const __m128i RK0 = _mm_load_si128(CONST_M128_CAST(keys + 32*i));
+ const __m128i RK2 = _mm_load_si128(CONST_M128_CAST(keys + 32*i+16));
+ const __m128i RK1 = _mm_srli_si128(RK0, 8);
+ const __m128i RK3 = _mm_srli_si128(RK2, 8);
+
+ B1 = _mm_sha256rnds2_epu32(B1, B0, RK0);
+ B0 = _mm_sha256rnds2_epu32(B0, B1, RK1);
+ B1 = _mm_sha256rnds2_epu32(B1, B0, RK2);
+ B0 = _mm_sha256rnds2_epu32(B0, B1, RK3);
+ }
+
+ TMP = _mm_shuffle_epi8(_mm_unpackhi_epi64(B0, B1), MASK1);
+ B1 = _mm_shuffle_epi8(_mm_unpacklo_epi64(B0, B1), MASK1);
+ B0 = TMP;
+
+ if (xorBlock)
+ {
+ _mm_storeu_si128(M128_CAST(outBlock + 0),
+ _mm_xor_si128(B0, _mm_loadu_si128(CONST_M128_CAST(xorBlock + 0))));
+
+ _mm_storeu_si128(M128_CAST(outBlock + 16),
+ _mm_xor_si128(B1, _mm_loadu_si128(CONST_M128_CAST(xorBlock + 16))));
+ }
+ else
+ {
+ _mm_storeu_si128(M128_CAST(outBlock + 0), B0);
+ _mm_storeu_si128(M128_CAST(outBlock + 16), B1);
+ }
+}
+#endif
+
+NAMESPACE_END
|