arne
/
cryptopp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Port SPECK64 to Altivec 

SPECK64 runs about 4x faster than C++ for POWER4 and friends. If POWER7 is available it goes back to full speed due to efficient unaligned loads
pull/739/head
Jeffrey Walton 2018-11-12 20:56:15 -05:00
parent b372f7d387
commit a0608a6b80
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
5 changed files with 162 additions and 83 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
GNUmakefile
View File

@ -666,6 +666,12 @@ ifeq ($(DETECT_FEATURES),1)
endif endif
endif endif
# Drop SIMON64 and SPECK64 to Power4 if Power7 not available
ifeq ($(SIMON64_FLAG)$(SPECK64_FLAG),)
SPECK64_FLAG = $(ALTIVEC_FLAG)
SIMON64_FLAG = $(ALTIVEC_FLAG)
endif
# IBM XL C/C++ # IBM XL C/C++
TPROG = TestPrograms/test_ppc_power8.cxx TPROG = TestPrograms/test_ppc_power8.cxx
TOPT = -qarch=pwr8 -qaltivec TOPT = -qarch=pwr8 -qaltivec
@ -743,6 +749,12 @@ ifeq ($(DETECT_FEATURES),1)
endif endif
endif endif
# Drop SIMON64 and SPECK64 to Power4 if Power7 not available
ifeq ($(SIMON64_FLAG)$(SPECK64_FLAG),)
SPECK64_FLAG = $(ALTIVEC_FLAG)
SIMON64_FLAG = $(ALTIVEC_FLAG)
endif
ifeq ($(ALTIVEC_FLAG),) ifeq ($(ALTIVEC_FLAG),)
CXXFLAGS += -DCRYPTOPP_DISABLE_ALTIVEC CXXFLAGS += -DCRYPTOPP_DISABLE_ALTIVEC
else ifeq ($(POWER7_FLAG),) else ifeq ($(POWER7_FLAG),)

32
adv_simd.h
View File

@ -74,6 +74,14 @@
# include "ppc_simd.h" # include "ppc_simd.h"
#endif #endif
#ifndef CRYPTOPP_INLINE
# if defined(CRYPTOPP_DEBUG)
# define CRYPTOPP_INLINE static
# else
# define CRYPTOPP_INLINE inline
# endif
#endif
// ************************ All block ciphers *********************** // // ************************ All block ciphers *********************** //
ANONYMOUS_NAMESPACE_BEGIN ANONYMOUS_NAMESPACE_BEGIN
@ -103,7 +111,7 @@ NAMESPACE_BEGIN(CryptoPP)
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the /// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type. /// same word type.
template <typename F2, typename F6, typename W> template <typename F2, typename F6, typename W>
inline size_t AdvancedProcessBlocks64_6x2_NEON(F2 func2, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks64_6x2_NEON(F2 func2, F6 func6,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -348,7 +356,7 @@ inline size_t AdvancedProcessBlocks64_6x2_NEON(F2 func2, F6 func6,
/// \details The subkey type is usually word32 or word64. F1 and F6 must use the /// \details The subkey type is usually word32 or word64. F1 and F6 must use the
/// same word type. /// same word type.
template <typename F1, typename F6, typename W> template <typename F1, typename F6, typename W>
inline size_t AdvancedProcessBlocks128_6x1_NEON(F1 func1, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_6x1_NEON(F1 func1, F6 func6,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -510,7 +518,7 @@ inline size_t AdvancedProcessBlocks128_6x1_NEON(F1 func1, F6 func6,
/// vector type. The V parameter is used to avoid template argument /// vector type. The V parameter is used to avoid template argument
/// deduction/substitution failures. /// deduction/substitution failures.
template <typename F1, typename F4, typename W, typename V> template <typename F1, typename F4, typename W, typename V>
inline size_t AdvancedProcessBlocks128_4x1_NEON(F1 func1, F4 func4, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_4x1_NEON(F1 func1, F4 func4,
const V& unused, const W *subKeys, size_t rounds, const byte *inBlocks, const V& unused, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -651,7 +659,7 @@ inline size_t AdvancedProcessBlocks128_4x1_NEON(F1 func1, F4 func4,
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the /// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type. /// same word type.
template <typename F2, typename F6, typename W> template <typename F2, typename F6, typename W>
inline size_t AdvancedProcessBlocks128_6x2_NEON(F2 func2, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_6x2_NEON(F2 func2, F6 func6,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -882,7 +890,7 @@ NAMESPACE_BEGIN(CryptoPP)
/// \details The subkey type is usually word32 or word64. F1 and F2 must use the /// \details The subkey type is usually word32 or word64. F1 and F2 must use the
/// same word type. /// same word type.
template <typename F1, typename F2, typename W> template <typename F1, typename F2, typename W>
inline size_t AdvancedProcessBlocks64_2x1_SSE(F1 func1, F2 func2, CRYPTOPP_INLINE size_t AdvancedProcessBlocks64_2x1_SSE(F1 func1, F2 func2,
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks, MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -1037,7 +1045,7 @@ inline size_t AdvancedProcessBlocks64_2x1_SSE(F1 func1, F2 func2,
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the /// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type. /// same word type.
template <typename F2, typename F6, typename W> template <typename F2, typename F6, typename W>
inline size_t AdvancedProcessBlocks64_6x2_SSE(F2 func2, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks64_6x2_SSE(F2 func2, F6 func6,
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks, MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -1288,7 +1296,7 @@ inline size_t AdvancedProcessBlocks64_6x2_SSE(F2 func2, F6 func6,
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the /// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type. /// same word type.
template <typename F2, typename F6, typename W> template <typename F2, typename F6, typename W>
inline size_t AdvancedProcessBlocks128_6x2_SSE(F2 func2, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_6x2_SSE(F2 func2, F6 func6,
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks, MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -1483,7 +1491,7 @@ inline size_t AdvancedProcessBlocks128_6x2_SSE(F2 func2, F6 func6,
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the /// \details The subkey type is usually word32 or word64. F1 and F4 must use the
/// same word type. /// same word type.
template <typename F1, typename F4, typename W> template <typename F1, typename F4, typename W>
inline size_t AdvancedProcessBlocks128_4x1_SSE(F1 func1, F4 func4, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_4x1_SSE(F1 func1, F4 func4,
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks, MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -1614,7 +1622,7 @@ inline size_t AdvancedProcessBlocks128_4x1_SSE(F1 func1, F4 func4,
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the /// \details The subkey type is usually word32 or word64. F1 and F4 must use the
/// same word type. /// same word type.
template <typename F1, typename F4, typename W> template <typename F1, typename F4, typename W>
inline size_t AdvancedProcessBlocks64_4x1_SSE(F1 func1, F4 func4, CRYPTOPP_INLINE size_t AdvancedProcessBlocks64_4x1_SSE(F1 func1, F4 func4,
MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks, MAYBE_CONST W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -1797,7 +1805,7 @@ NAMESPACE_BEGIN(CryptoPP)
/// \details The subkey type is usually word32 or word64. F2 and F6 must use the /// \details The subkey type is usually word32 or word64. F2 and F6 must use the
/// same word type. /// same word type.
template <typename F2, typename F6, typename W> template <typename F2, typename F6, typename W>
inline size_t AdvancedProcessBlocks64_6x2_ALTIVEC(F2 func2, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks64_6x2_ALTIVEC(F2 func2, F6 func6,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -2068,7 +2076,7 @@ inline size_t AdvancedProcessBlocks64_6x2_ALTIVEC(F2 func2, F6 func6,
/// \details The subkey type is usually word32 or word64. F1 and F4 must use the /// \details The subkey type is usually word32 or word64. F1 and F4 must use the
/// same word type. /// same word type.
template <typename F1, typename F4, typename W> template <typename F1, typename F4, typename W>
inline size_t AdvancedProcessBlocks128_4x1_ALTIVEC(F1 func1, F4 func4, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_4x1_ALTIVEC(F1 func1, F4 func4,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
@ -2213,7 +2221,7 @@ inline size_t AdvancedProcessBlocks128_4x1_ALTIVEC(F1 func1, F4 func4,
/// \details The subkey type is usually word32 or word64. F1 and F6 must use the /// \details The subkey type is usually word32 or word64. F1 and F6 must use the
/// same word type. /// same word type.
template <typename F1, typename F6, typename W> template <typename F1, typename F6, typename W>
inline size_t AdvancedProcessBlocks128_6x1_ALTIVEC(F1 func1, F6 func6, CRYPTOPP_INLINE size_t AdvancedProcessBlocks128_6x1_ALTIVEC(F1 func1, F6 func6,
const W *subKeys, size_t rounds, const byte *inBlocks, const W *subKeys, size_t rounds, const byte *inBlocks,
const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {

52
speck.cpp
View File

@ -7,6 +7,14 @@
#include "misc.h" #include "misc.h"
#include "cpu.h" #include "cpu.h"
#ifndef CRYPTOPP_INLINE
# if defined(CRYPTOPP_DEBUG)
# define CRYPTOPP_INLINE static
# else
# define CRYPTOPP_INLINE inline
# endif
#endif
// Uncomment for benchmarking C++ against SSE or NEON. // Uncomment for benchmarking C++ against SSE or NEON.
// Do so in both speck.cpp and speck-simd.cpp. // Do so in both speck.cpp and speck-simd.cpp.
// #undef CRYPTOPP_SSSE3_AVAILABLE // #undef CRYPTOPP_SSSE3_AVAILABLE
@ -28,7 +36,7 @@ using CryptoPP::rotrConstant;
/// additional template parameters also made calling SPECK_Encrypt and SPECK_Decrypt /// additional template parameters also made calling SPECK_Encrypt and SPECK_Decrypt
/// kind of messy. /// kind of messy.
template <class W> template <class W>
inline void TF83(W& x, W& y, const W k) CRYPTOPP_INLINE void TF83(W& x, W& y, const W k)
{ {
x = rotrConstant<8>(x); x = rotrConstant<8>(x);
x += y; x ^= k; x += y; x ^= k;
@ -44,7 +52,7 @@ inline void TF83(W& x, W& y, const W k)
/// additional template parameters also made calling SPECK_Encrypt and SPECK_Decrypt /// additional template parameters also made calling SPECK_Encrypt and SPECK_Decrypt
/// kind of messy. /// kind of messy.
template <class W> template <class W>
inline void TR83(W& x, W& y, const W k) CRYPTOPP_INLINE void TR83(W& x, W& y, const W k)
{ {
y ^= x; y ^= x;
y = rotrConstant<3>(y); y = rotrConstant<3>(y);
@ -59,7 +67,7 @@ inline void TR83(W& x, W& y, const W k)
/// \param p input array /// \param p input array
/// \param k subkey array /// \param k subkey array
template <class W, unsigned int R> template <class W, unsigned int R>
inline void SPECK_Encrypt(W c[2], const W p[2], const W k[R]) CRYPTOPP_INLINE void SPECK_Encrypt(W c[2], const W p[2], const W k[R])
{ {
c[0]=p[0]; c[1]=p[1]; c[0]=p[0]; c[1]=p[1];
@ -75,7 +83,7 @@ inline void SPECK_Encrypt(W c[2], const W p[2], const W k[R])
/// \param c input array /// \param c input array
/// \param k subkey array /// \param k subkey array
template <class W, unsigned int R> template <class W, unsigned int R>
inline void SPECK_Decrypt(W p[2], const W c[2], const W k[R]) CRYPTOPP_INLINE void SPECK_Decrypt(W p[2], const W c[2], const W k[R])
{ {
p[0]=c[0]; p[1]=c[1]; p[0]=c[0]; p[1]=c[1];
@ -91,7 +99,7 @@ inline void SPECK_Decrypt(W p[2], const W c[2], const W k[R])
/// \param key empty subkey array /// \param key empty subkey array
/// \param k user key array /// \param k user key array
template <class W, unsigned int R> template <class W, unsigned int R>
inline void SPECK_ExpandKey_2W(W key[R], const W k[2]) CRYPTOPP_INLINE void SPECK_ExpandKey_2W(W key[R], const W k[2])
{ {
CRYPTOPP_ASSERT(R==32); CRYPTOPP_ASSERT(R==32);
W i=0, B=k[0], A=k[1]; W i=0, B=k[0], A=k[1];
@ -111,7 +119,7 @@ inline void SPECK_ExpandKey_2W(W key[R], const W k[2])
/// \param key empty subkey array /// \param key empty subkey array
/// \param k user key array /// \param k user key array
template <class W, unsigned int R> template <class W, unsigned int R>
inline void SPECK_ExpandKey_3W(W key[R], const W k[3]) CRYPTOPP_INLINE void SPECK_ExpandKey_3W(W key[R], const W k[3])
{ {
CRYPTOPP_ASSERT(R==33 || R==26); CRYPTOPP_ASSERT(R==33 || R==26);
W i=0, C=k[0], B=k[1], A=k[2]; W i=0, C=k[0], B=k[1], A=k[2];
@ -138,7 +146,7 @@ inline void SPECK_ExpandKey_3W(W key[R], const W k[3])
/// \param key empty subkey array /// \param key empty subkey array
/// \param k user key array /// \param k user key array
template <class W, unsigned int R> template <class W, unsigned int R>
inline void SPECK_ExpandKey_4W(W key[R], const W k[4]) CRYPTOPP_INLINE void SPECK_ExpandKey_4W(W key[R], const W k[4])
{ {
CRYPTOPP_ASSERT(R==34 || R==27); CRYPTOPP_ASSERT(R==34 || R==27);
W i=0, D=k[0], C=k[1], B=k[2], A=k[3]; W i=0, D=k[0], C=k[1], B=k[2], A=k[3];
@ -200,11 +208,11 @@ extern size_t SPECK128_Dec_AdvancedProcessBlocks_SSSE3(const word64* subKeys, si
const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags); const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags);
#endif #endif
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
extern size_t SPECK64_Enc_AdvancedProcessBlocks_POWER7(const word32* subKeys, size_t rounds, extern size_t SPECK64_Enc_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags); const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags);
extern size_t SPECK64_Dec_AdvancedProcessBlocks_POWER7(const word32* subKeys, size_t rounds, extern size_t SPECK64_Dec_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags); const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags);
#endif #endif
@ -227,9 +235,9 @@ std::string SPECK64::Base::AlgorithmProvider() const
if (HasNEON()) if (HasNEON())
return "NEON"; return "NEON";
# endif # endif
# if (CRYPTOPP_POWER7_AVAILABLE) # if (CRYPTOPP_ALTIVEC_AVAILABLE)
if (HasPower7()) if (HasAltivec())
return "Power7"; return "Altivec";
# endif # endif
#endif #endif
return "C++"; return "C++";
@ -264,6 +272,12 @@ void SPECK64::Base::UncheckedSetKey(const byte *userKey, unsigned int keyLength,
default: default:
CRYPTOPP_ASSERT(0);; CRYPTOPP_ASSERT(0);;
} }
// Altivec loads the current subkey as a 16-byte vector
// The extra elements ensure memory backs the last subkey.
#if CRYPTOPP_ALTIVEC_AVAILABLE
m_rkeys.Grow(m_rkeys.size()+4);
#endif
} }
void SPECK64::Enc::ProcessAndXorBlock(const byte *inBlock, const byte *xorBlock, byte *outBlock) const void SPECK64::Enc::ProcessAndXorBlock(const byte *inBlock, const byte *xorBlock, byte *outBlock) const
@ -435,9 +449,9 @@ size_t SPECK64::Enc::AdvancedProcessBlocks(const byte *inBlocks, const byte *xor
return SPECK64_Enc_AdvancedProcessBlocks_NEON(m_rkeys, (size_t)m_rounds, return SPECK64_Enc_AdvancedProcessBlocks_NEON(m_rkeys, (size_t)m_rounds,
inBlocks, xorBlocks, outBlocks, length, flags); inBlocks, xorBlocks, outBlocks, length, flags);
#endif #endif
#if (CRYPTOPP_POWER7_AVAILABLE) #if (CRYPTOPP_ALTIVEC_AVAILABLE)
if (HasPower7()) if (HasAltivec())
return SPECK64_Enc_AdvancedProcessBlocks_POWER7(m_rkeys, (size_t)m_rounds, return SPECK64_Enc_AdvancedProcessBlocks_ALTIVEC(m_rkeys, (size_t)m_rounds,
inBlocks, xorBlocks, outBlocks, length, flags); inBlocks, xorBlocks, outBlocks, length, flags);
#endif #endif
return BlockTransformation::AdvancedProcessBlocks(inBlocks, xorBlocks, outBlocks, length, flags); return BlockTransformation::AdvancedProcessBlocks(inBlocks, xorBlocks, outBlocks, length, flags);
@ -456,9 +470,9 @@ size_t SPECK64::Dec::AdvancedProcessBlocks(const byte *inBlocks, const byte *xor
return SPECK64_Dec_AdvancedProcessBlocks_NEON(m_rkeys, (size_t)m_rounds, return SPECK64_Dec_AdvancedProcessBlocks_NEON(m_rkeys, (size_t)m_rounds,
inBlocks, xorBlocks, outBlocks, length, flags); inBlocks, xorBlocks, outBlocks, length, flags);
#endif #endif
#if (CRYPTOPP_POWER7_AVAILABLE) #if (CRYPTOPP_ALTIVEC_AVAILABLE)
if (HasPower7()) if (HasAltivec())
return SPECK64_Dec_AdvancedProcessBlocks_POWER7(m_rkeys, (size_t)m_rounds, return SPECK64_Dec_AdvancedProcessBlocks_ALTIVEC(m_rkeys, (size_t)m_rounds,
inBlocks, xorBlocks, outBlocks, length, flags); inBlocks, xorBlocks, outBlocks, length, flags);
#endif #endif
return BlockTransformation::AdvancedProcessBlocks(inBlocks, xorBlocks, outBlocks, length, flags); return BlockTransformation::AdvancedProcessBlocks(inBlocks, xorBlocks, outBlocks, length, flags);

52
speck128_simd.cpp
View File

@ -1,4 +1,4 @@
// speck-simd.cpp - written and placed in the public domain by Jeffrey Walton // speck128_simd.cpp - written and placed in the public domain by Jeffrey Walton
// //
// This source file uses intrinsics and built-ins to gain access to // This source file uses intrinsics and built-ins to gain access to
// SSSE3, ARM NEON and ARMv8a, and Power7 Altivec instructions. A separate // SSSE3, ARM NEON and ARMv8a, and Power7 Altivec instructions. A separate
@ -12,6 +12,14 @@
#include "misc.h" #include "misc.h"
#include "adv_simd.h" #include "adv_simd.h"
#ifndef CRYPTOPP_INLINE
# if defined(CRYPTOPP_DEBUG)
# define CRYPTOPP_INLINE static
# else
# define CRYPTOPP_INLINE inline
# endif
#endif
// Uncomment for benchmarking C++ against SSE or NEON. // Uncomment for benchmarking C++ against SSE or NEON.
// Do so in both speck.cpp and speck-simd.cpp. // Do so in both speck.cpp and speck-simd.cpp.
// #undef CRYPTOPP_SSSE3_AVAILABLE // #undef CRYPTOPP_SSSE3_AVAILABLE
@ -42,7 +50,7 @@
# include <arm_acle.h> # include <arm_acle.h>
#endif #endif
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_POWER8_AVAILABLE)
# include "ppc_simd.h" # include "ppc_simd.h"
#endif #endif
@ -60,7 +68,7 @@ using CryptoPP::word64;
#if (CRYPTOPP_ARM_NEON_AVAILABLE) #if (CRYPTOPP_ARM_NEON_AVAILABLE)
template <class T> template <class T>
inline T UnpackHigh64(const T& a, const T& b) CRYPTOPP_INLINE T UnpackHigh64(const T& a, const T& b)
{ {
const uint64x1_t x(vget_high_u64((uint64x2_t)a)); const uint64x1_t x(vget_high_u64((uint64x2_t)a));
const uint64x1_t y(vget_high_u64((uint64x2_t)b)); const uint64x1_t y(vget_high_u64((uint64x2_t)b));
@ -68,7 +76,7 @@ inline T UnpackHigh64(const T& a, const T& b)
} }
template <class T> template <class T>
inline T UnpackLow64(const T& a, const T& b) CRYPTOPP_INLINE T UnpackLow64(const T& a, const T& b)
{ {
const uint64x1_t x(vget_low_u64((uint64x2_t)a)); const uint64x1_t x(vget_low_u64((uint64x2_t)a));
const uint64x1_t y(vget_low_u64((uint64x2_t)b)); const uint64x1_t y(vget_low_u64((uint64x2_t)b));
@ -76,7 +84,7 @@ inline T UnpackLow64(const T& a, const T& b)
} }
template <unsigned int R> template <unsigned int R>
inline uint64x2_t RotateLeft64(const uint64x2_t& val) CRYPTOPP_INLINE uint64x2_t RotateLeft64(const uint64x2_t& val)
{ {
const uint64x2_t a(vshlq_n_u64(val, R)); const uint64x2_t a(vshlq_n_u64(val, R));
const uint64x2_t b(vshrq_n_u64(val, 64 - R)); const uint64x2_t b(vshrq_n_u64(val, 64 - R));
@ -84,7 +92,7 @@ inline uint64x2_t RotateLeft64(const uint64x2_t& val)
} }
template <unsigned int R> template <unsigned int R>
inline uint64x2_t RotateRight64(const uint64x2_t& val) CRYPTOPP_INLINE uint64x2_t RotateRight64(const uint64x2_t& val)
{ {
const uint64x2_t a(vshlq_n_u64(val, 64 - R)); const uint64x2_t a(vshlq_n_u64(val, 64 - R));
const uint64x2_t b(vshrq_n_u64(val, R)); const uint64x2_t b(vshrq_n_u64(val, R));
@ -94,7 +102,7 @@ inline uint64x2_t RotateRight64(const uint64x2_t& val)
#if defined(__aarch32__) || defined(__aarch64__) #if defined(__aarch32__) || defined(__aarch64__)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline uint64x2_t RotateLeft64<8>(const uint64x2_t& val) CRYPTOPP_INLINE uint64x2_t RotateLeft64<8>(const uint64x2_t& val)
{ {
#if (CRYPTOPP_BIG_ENDIAN) #if (CRYPTOPP_BIG_ENDIAN)
const uint8_t maskb[16] = { 14,13,12,11, 10,9,8,15, 6,5,4,3, 2,1,0,7 }; const uint8_t maskb[16] = { 14,13,12,11, 10,9,8,15, 6,5,4,3, 2,1,0,7 };
@ -110,7 +118,7 @@ inline uint64x2_t RotateLeft64<8>(const uint64x2_t& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline uint64x2_t RotateRight64<8>(const uint64x2_t& val) CRYPTOPP_INLINE uint64x2_t RotateRight64<8>(const uint64x2_t& val)
{ {
#if (CRYPTOPP_BIG_ENDIAN) #if (CRYPTOPP_BIG_ENDIAN)
const uint8_t maskb[16] = { 8,15,14,13, 12,11,10,9, 0,7,6,5, 4,3,2,1 }; const uint8_t maskb[16] = { 8,15,14,13, 12,11,10,9, 0,7,6,5, 4,3,2,1 };
@ -125,7 +133,7 @@ inline uint64x2_t RotateRight64<8>(const uint64x2_t& val)
} }
#endif #endif
inline void SPECK128_Enc_Block(uint64x2_t &block0, uint64x2_t &block1, CRYPTOPP_INLINE void SPECK128_Enc_Block(uint64x2_t &block0, uint64x2_t &block1,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
// [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ... // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
@ -148,7 +156,7 @@ inline void SPECK128_Enc_Block(uint64x2_t &block0, uint64x2_t &block1,
block1 = UnpackHigh64(y1, x1); block1 = UnpackHigh64(y1, x1);
} }
inline void SPECK128_Enc_6_Blocks(uint64x2_t &block0, uint64x2_t &block1, CRYPTOPP_INLINE void SPECK128_Enc_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5, uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
@ -190,7 +198,7 @@ inline void SPECK128_Enc_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
block5 = UnpackHigh64(y3, x3); block5 = UnpackHigh64(y3, x3);
} }
inline void SPECK128_Dec_Block(uint64x2_t &block0, uint64x2_t &block1, CRYPTOPP_INLINE void SPECK128_Dec_Block(uint64x2_t &block0, uint64x2_t &block1,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
// [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ... // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
@ -213,7 +221,7 @@ inline void SPECK128_Dec_Block(uint64x2_t &block0, uint64x2_t &block1,
block1 = UnpackHigh64(y1, x1); block1 = UnpackHigh64(y1, x1);
} }
inline void SPECK128_Dec_6_Blocks(uint64x2_t &block0, uint64x2_t &block1, CRYPTOPP_INLINE void SPECK128_Dec_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5, uint64x2_t &block2, uint64x2_t &block3, uint64x2_t &block4, uint64x2_t &block5,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
@ -278,7 +286,7 @@ inline void SPECK128_Dec_6_Blocks(uint64x2_t &block0, uint64x2_t &block1,
#endif #endif
template <unsigned int R> template <unsigned int R>
inline __m128i RotateLeft64(const __m128i& val) CRYPTOPP_INLINE __m128i RotateLeft64(const __m128i& val)
{ {
#if defined(CRYPTOPP_AVX512_ROTATE) #if defined(CRYPTOPP_AVX512_ROTATE)
return _mm_rol_epi64(val, R); return _mm_rol_epi64(val, R);
@ -291,7 +299,7 @@ inline __m128i RotateLeft64(const __m128i& val)
} }
template <unsigned int R> template <unsigned int R>
inline __m128i RotateRight64(const __m128i& val) CRYPTOPP_INLINE __m128i RotateRight64(const __m128i& val)
{ {
#if defined(CRYPTOPP_AVX512_ROTATE) #if defined(CRYPTOPP_AVX512_ROTATE)
return _mm_ror_epi64(val, R); return _mm_ror_epi64(val, R);
@ -305,7 +313,7 @@ inline __m128i RotateRight64(const __m128i& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline __m128i RotateLeft64<8>(const __m128i& val) CRYPTOPP_INLINE __m128i RotateLeft64<8>(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi64(val, 8); return _mm_roti_epi64(val, 8);
@ -317,7 +325,7 @@ inline __m128i RotateLeft64<8>(const __m128i& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline __m128i RotateRight64<8>(const __m128i& val) CRYPTOPP_INLINE __m128i RotateRight64<8>(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi64(val, 64-8); return _mm_roti_epi64(val, 64-8);
@ -327,7 +335,7 @@ inline __m128i RotateRight64<8>(const __m128i& val)
#endif #endif
} }
inline void SPECK128_Enc_Block(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK128_Enc_Block(__m128i &block0, __m128i &block1,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
// [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ... // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
@ -351,7 +359,7 @@ inline void SPECK128_Enc_Block(__m128i &block0, __m128i &block1,
block1 = _mm_unpackhi_epi64(y1, x1); block1 = _mm_unpackhi_epi64(y1, x1);
} }
inline void SPECK128_Enc_6_Blocks(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK128_Enc_6_Blocks(__m128i &block0, __m128i &block1,
__m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5, __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
@ -394,7 +402,7 @@ inline void SPECK128_Enc_6_Blocks(__m128i &block0, __m128i &block1,
block5 = _mm_unpackhi_epi64(y3, x3); block5 = _mm_unpackhi_epi64(y3, x3);
} }
inline void SPECK128_Dec_Block(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK128_Dec_Block(__m128i &block0, __m128i &block1,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
// [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ... // [A1 A2][B1 B2] ... => [A1 B1][A2 B2] ...
@ -418,7 +426,7 @@ inline void SPECK128_Dec_Block(__m128i &block0, __m128i &block1,
block1 = _mm_unpackhi_epi64(y1, x1); block1 = _mm_unpackhi_epi64(y1, x1);
} }
inline void SPECK128_Dec_6_Blocks(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK128_Dec_6_Blocks(__m128i &block0, __m128i &block1,
__m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5, __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
const word64 *subkeys, unsigned int rounds) const word64 *subkeys, unsigned int rounds)
{ {
@ -477,7 +485,7 @@ using CryptoPP::VectorXor;
// Rotate left by bit count // Rotate left by bit count
template<unsigned int C> template<unsigned int C>
inline uint64x2_p RotateLeft64(const uint64x2_p val) CRYPTOPP_INLINE uint64x2_p RotateLeft64(const uint64x2_p val)
{ {
const uint64x2_p m = {C, C}; const uint64x2_p m = {C, C};
return vec_rl(val, m); return vec_rl(val, m);
@ -485,7 +493,7 @@ inline uint64x2_p RotateLeft64(const uint64x2_p val)
// Rotate right by bit count // Rotate right by bit count
template<unsigned int C> template<unsigned int C>
inline uint64x2_p RotateRight64(const uint64x2_p val) CRYPTOPP_INLINE uint64x2_p RotateRight64(const uint64x2_p val)
{ {
const uint64x2_p m = {64-C, 64-C}; const uint64x2_p m = {64-C, 64-C};
return vec_rl(val, m); return vec_rl(val, m);

97
speck64_simd.cpp
View File

@ -1,7 +1,7 @@
// speck-simd.cpp - written and placed in the public domain by Jeffrey Walton // speck64_simd.cpp - written and placed in the public domain by Jeffrey Walton
// //
// This source file uses intrinsics and built-ins to gain access to // This source file uses intrinsics and built-ins to gain access to
// SSSE3, ARM NEON and ARMv8a, and Power7 Altivec instructions. A separate // SSSE3, ARM NEON and ARMv8a, and Altivec instructions. A separate
// source file is needed because additional CXXFLAGS are required to enable // source file is needed because additional CXXFLAGS are required to enable
// the appropriate instructions sets in some build configurations. // the appropriate instructions sets in some build configurations.
@ -46,10 +46,18 @@
# include <arm_acle.h> # include <arm_acle.h>
#endif #endif
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
# include "ppc_simd.h" # include "ppc_simd.h"
#endif #endif
#ifndef CRYPTOPP_INLINE
# if defined(CRYPTOPP_DEBUG)
# define CRYPTOPP_INLINE static
# else
# define CRYPTOPP_INLINE inline
# endif
#endif
// Squash MS LNK4221 and libtool warnings // Squash MS LNK4221 and libtool warnings
extern const char SPECK64_SIMD_FNAME[] = __FILE__; extern const char SPECK64_SIMD_FNAME[] = __FILE__;
@ -64,7 +72,7 @@ using CryptoPP::word64;
#if (CRYPTOPP_ARM_NEON_AVAILABLE) #if (CRYPTOPP_ARM_NEON_AVAILABLE)
template <class T> template <class T>
inline T UnpackHigh32(const T& a, const T& b) CRYPTOPP_INLINE T UnpackHigh32(const T& a, const T& b)
{ {
const uint32x2_t x(vget_high_u32((uint32x4_t)a)); const uint32x2_t x(vget_high_u32((uint32x4_t)a));
const uint32x2_t y(vget_high_u32((uint32x4_t)b)); const uint32x2_t y(vget_high_u32((uint32x4_t)b));
@ -73,7 +81,7 @@ inline T UnpackHigh32(const T& a, const T& b)
} }
template <class T> template <class T>
inline T UnpackLow32(const T& a, const T& b) CRYPTOPP_INLINE T UnpackLow32(const T& a, const T& b)
{ {
const uint32x2_t x(vget_low_u32((uint32x4_t)a)); const uint32x2_t x(vget_low_u32((uint32x4_t)a));
const uint32x2_t y(vget_low_u32((uint32x4_t)b)); const uint32x2_t y(vget_low_u32((uint32x4_t)b));
@ -82,7 +90,7 @@ inline T UnpackLow32(const T& a, const T& b)
} }
template <unsigned int R> template <unsigned int R>
inline uint32x4_t RotateLeft32(const uint32x4_t& val) CRYPTOPP_INLINE uint32x4_t RotateLeft32(const uint32x4_t& val)
{ {
const uint32x4_t a(vshlq_n_u32(val, R)); const uint32x4_t a(vshlq_n_u32(val, R));
const uint32x4_t b(vshrq_n_u32(val, 32 - R)); const uint32x4_t b(vshrq_n_u32(val, 32 - R));
@ -90,7 +98,7 @@ inline uint32x4_t RotateLeft32(const uint32x4_t& val)
} }
template <unsigned int R> template <unsigned int R>
inline uint32x4_t RotateRight32(const uint32x4_t& val) CRYPTOPP_INLINE uint32x4_t RotateRight32(const uint32x4_t& val)
{ {
const uint32x4_t a(vshlq_n_u32(val, 32 - R)); const uint32x4_t a(vshlq_n_u32(val, 32 - R));
const uint32x4_t b(vshrq_n_u32(val, R)); const uint32x4_t b(vshrq_n_u32(val, R));
@ -100,7 +108,7 @@ inline uint32x4_t RotateRight32(const uint32x4_t& val)
#if defined(__aarch32__) || defined(__aarch64__) #if defined(__aarch32__) || defined(__aarch64__)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline uint32x4_t RotateLeft32<8>(const uint32x4_t& val) CRYPTOPP_INLINE uint32x4_t RotateLeft32<8>(const uint32x4_t& val)
{ {
#if (CRYPTOPP_BIG_ENDIAN) #if (CRYPTOPP_BIG_ENDIAN)
const uint8_t maskb[16] = { 14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3 }; const uint8_t maskb[16] = { 14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3 };
@ -116,7 +124,7 @@ inline uint32x4_t RotateLeft32<8>(const uint32x4_t& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline uint32x4_t RotateRight32<8>(const uint32x4_t& val) CRYPTOPP_INLINE uint32x4_t RotateRight32<8>(const uint32x4_t& val)
{ {
#if (CRYPTOPP_BIG_ENDIAN) #if (CRYPTOPP_BIG_ENDIAN)
const uint8_t maskb[16] = { 12,15,14,13, 8,11,10,9, 4,7,6,5, 0,3,2,1 }; const uint8_t maskb[16] = { 12,15,14,13, 8,11,10,9, 4,7,6,5, 0,3,2,1 };
@ -131,7 +139,7 @@ inline uint32x4_t RotateRight32<8>(const uint32x4_t& val)
} }
#endif // Aarch32 or Aarch64 #endif // Aarch32 or Aarch64
inline void SPECK64_Enc_Block(uint32x4_t &block0, uint32x4_t &block1, CRYPTOPP_INLINE void SPECK64_Enc_Block(uint32x4_t &block0, uint32x4_t &block1,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
// [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ... // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
@ -154,7 +162,7 @@ inline void SPECK64_Enc_Block(uint32x4_t &block0, uint32x4_t &block1,
block1 = UnpackHigh32(y1, x1); block1 = UnpackHigh32(y1, x1);
} }
inline void SPECK64_Dec_Block(uint32x4_t &block0, uint32x4_t &block1, CRYPTOPP_INLINE void SPECK64_Dec_Block(uint32x4_t &block0, uint32x4_t &block1,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
// [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ... // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
@ -177,7 +185,7 @@ inline void SPECK64_Dec_Block(uint32x4_t &block0, uint32x4_t &block1,
block1 = UnpackHigh32(y1, x1); block1 = UnpackHigh32(y1, x1);
} }
inline void SPECK64_Enc_6_Blocks(uint32x4_t &block0, uint32x4_t &block1, CRYPTOPP_INLINE void SPECK64_Enc_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5, uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
@ -219,7 +227,7 @@ inline void SPECK64_Enc_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
block5 = UnpackHigh32(y3, x3); block5 = UnpackHigh32(y3, x3);
} }
inline void SPECK64_Dec_6_Blocks(uint32x4_t &block0, uint32x4_t &block1, CRYPTOPP_INLINE void SPECK64_Dec_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5, uint32x4_t &block2, uint32x4_t &block3, uint32x4_t &block4, uint32x4_t &block5,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
@ -268,7 +276,7 @@ inline void SPECK64_Dec_6_Blocks(uint32x4_t &block0, uint32x4_t &block1,
#if defined(CRYPTOPP_SSE41_AVAILABLE) #if defined(CRYPTOPP_SSE41_AVAILABLE)
template <unsigned int R> template <unsigned int R>
inline __m128i RotateLeft32(const __m128i& val) CRYPTOPP_INLINE __m128i RotateLeft32(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi32(val, R); return _mm_roti_epi32(val, R);
@ -279,7 +287,7 @@ inline __m128i RotateLeft32(const __m128i& val)
} }
template <unsigned int R> template <unsigned int R>
inline __m128i RotateRight32(const __m128i& val) CRYPTOPP_INLINE __m128i RotateRight32(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi32(val, 32-R); return _mm_roti_epi32(val, 32-R);
@ -291,7 +299,7 @@ inline __m128i RotateRight32(const __m128i& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline __m128i RotateLeft32<8>(const __m128i& val) CRYPTOPP_INLINE __m128i RotateLeft32<8>(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi32(val, 8); return _mm_roti_epi32(val, 8);
@ -303,7 +311,7 @@ inline __m128i RotateLeft32<8>(const __m128i& val)
// Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks. // Faster than two Shifts and an Or. Thanks to Louis Wingers and Bryan Weeks.
template <> template <>
inline __m128i RotateRight32<8>(const __m128i& val) CRYPTOPP_INLINE __m128i RotateRight32<8>(const __m128i& val)
{ {
#if defined(__XOP__) #if defined(__XOP__)
return _mm_roti_epi32(val, 32-8); return _mm_roti_epi32(val, 32-8);
@ -313,7 +321,7 @@ inline __m128i RotateRight32<8>(const __m128i& val)
#endif #endif
} }
inline void SPECK64_Enc_Block(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK64_Enc_Block(__m128i &block0, __m128i &block1,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
// [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ... // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
@ -339,7 +347,7 @@ inline void SPECK64_Enc_Block(__m128i &block0, __m128i &block1,
block1 = _mm_unpackhi_epi32(y1, x1); block1 = _mm_unpackhi_epi32(y1, x1);
} }
inline void SPECK64_Dec_Block(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK64_Dec_Block(__m128i &block0, __m128i &block1,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
// [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ... // [A1 A2 A3 A4][B1 B2 B3 B4] ... => [A1 A3 B1 B3][A2 A4 B2 B4] ...
@ -365,7 +373,7 @@ inline void SPECK64_Dec_Block(__m128i &block0, __m128i &block1,
block1 = _mm_unpackhi_epi32(y1, x1); block1 = _mm_unpackhi_epi32(y1, x1);
} }
inline void SPECK64_Enc_6_Blocks(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK64_Enc_6_Blocks(__m128i &block0, __m128i &block1,
__m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5, __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
@ -416,7 +424,7 @@ inline void SPECK64_Enc_6_Blocks(__m128i &block0, __m128i &block1,
block5 = _mm_unpackhi_epi32(y3, x3); block5 = _mm_unpackhi_epi32(y3, x3);
} }
inline void SPECK64_Dec_6_Blocks(__m128i &block0, __m128i &block1, CRYPTOPP_INLINE void SPECK64_Dec_6_Blocks(__m128i &block0, __m128i &block1,
__m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5, __m128i &block2, __m128i &block3, __m128i &block4, __m128i &block5,
const word32 *subkeys, unsigned int rounds) const word32 *subkeys, unsigned int rounds)
{ {
@ -469,19 +477,20 @@ inline void SPECK64_Dec_6_Blocks(__m128i &block0, __m128i &block1,
#endif // CRYPTOPP_SSE41_AVAILABLE #endif // CRYPTOPP_SSE41_AVAILABLE
// ***************************** Power7 ***************************** // // ***************************** Altivec ***************************** //
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
using CryptoPP::uint8x16_p; using CryptoPP::uint8x16_p;
using CryptoPP::uint32x4_p; using CryptoPP::uint32x4_p;
using CryptoPP::VectorAdd; using CryptoPP::VectorAdd;
using CryptoPP::VectorSub; using CryptoPP::VectorSub;
using CryptoPP::VectorXor; using CryptoPP::VectorXor;
using CryptoPP::VectorLoad;
// Rotate left by bit count // Rotate left by bit count
template<unsigned int C> template<unsigned int C>
inline uint32x4_p RotateLeft32(const uint32x4_p val) CRYPTOPP_INLINE uint32x4_p RotateLeft32(const uint32x4_p val)
{ {
const uint32x4_p m = {C, C, C, C}; const uint32x4_p m = {C, C, C, C};
return vec_rl(val, m); return vec_rl(val, m);
@ -489,7 +498,7 @@ inline uint32x4_p RotateLeft32(const uint32x4_p val)
// Rotate right by bit count // Rotate right by bit count
template<unsigned int C> template<unsigned int C>
inline uint32x4_p RotateRight32(const uint32x4_p val) CRYPTOPP_INLINE uint32x4_p RotateRight32(const uint32x4_p val)
{ {
const uint32x4_p m = {32-C, 32-C, 32-C, 32-C}; const uint32x4_p m = {32-C, 32-C, 32-C, 32-C};
return vec_rl(val, m); return vec_rl(val, m);
@ -512,7 +521,14 @@ void SPECK64_Enc_Block(uint32x4_p &block0, uint32x4_p &block1,
for (int i=0; i < static_cast<int>(rounds); ++i) for (int i=0; i < static_cast<int>(rounds); ++i)
{ {
#if CRYPTOPP_POWER7_AVAILABLE
const uint32x4_p rk = vec_splats(subkeys[i]); const uint32x4_p rk = vec_splats(subkeys[i]);
#else
// subkeys has extra elements so memory backs the last subkey
const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
uint32x4_p rk = VectorLoad(0, subkeys+i);
rk = vec_perm(rk, rk, m);
#endif
x1 = RotateRight32<8>(x1); x1 = RotateRight32<8>(x1);
x1 = VectorAdd(x1, y1); x1 = VectorAdd(x1, y1);
@ -552,7 +568,14 @@ void SPECK64_Dec_Block(uint32x4_p &block0, uint32x4_p &block1,
for (int i = static_cast<int>(rounds-1); i >= 0; --i) for (int i = static_cast<int>(rounds-1); i >= 0; --i)
{ {
#if CRYPTOPP_POWER7_AVAILABLE
const uint32x4_p rk = vec_splats(subkeys[i]); const uint32x4_p rk = vec_splats(subkeys[i]);
#else
// subkeys has extra elements so memory backs the last subkey
const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
uint32x4_p rk = VectorLoad(0, subkeys+i);
rk = vec_perm(rk, rk, m);
#endif
y1 = VectorXor(y1, x1); y1 = VectorXor(y1, x1);
y1 = RotateRight32<3>(y1); y1 = RotateRight32<3>(y1);
@ -597,7 +620,14 @@ void SPECK64_Enc_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
for (int i=0; i < static_cast<int>(rounds); ++i) for (int i=0; i < static_cast<int>(rounds); ++i)
{ {
#if CRYPTOPP_POWER7_AVAILABLE
const uint32x4_p rk = vec_splats(subkeys[i]); const uint32x4_p rk = vec_splats(subkeys[i]);
#else
// subkeys has extra elements so memory backs the last subkey
const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
uint32x4_p rk = VectorLoad(0, subkeys+i);
rk = vec_perm(rk, rk, m);
#endif
x1 = RotateRight32<8>(x1); x1 = RotateRight32<8>(x1);
x2 = RotateRight32<8>(x2); x2 = RotateRight32<8>(x2);
@ -659,7 +689,14 @@ void SPECK64_Dec_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
for (int i = static_cast<int>(rounds-1); i >= 0; --i) for (int i = static_cast<int>(rounds-1); i >= 0; --i)
{ {
#if CRYPTOPP_POWER7_AVAILABLE
const uint32x4_p rk = vec_splats(subkeys[i]); const uint32x4_p rk = vec_splats(subkeys[i]);
#else
// subkeys has extra elements so memory backs the last subkey
const uint8x16_p m = {0,1,2,3, 0,1,2,3, 0,1,2,3, 0,1,2,3};
uint32x4_p rk = VectorLoad(0, subkeys+i);
rk = vec_perm(rk, rk, m);
#endif
y1 = VectorXor(y1, x1); y1 = VectorXor(y1, x1);
y2 = VectorXor(y2, x2); y2 = VectorXor(y2, x2);
@ -699,7 +736,7 @@ void SPECK64_Dec_6_Blocks(uint32x4_p &block0, uint32x4_p &block1,
block5 = (uint32x4_p)vec_perm(x3, y3, m4); block5 = (uint32x4_p)vec_perm(x3, y3, m4);
} }
#endif // CRYPTOPP_POWER7_AVAILABLE #endif // CRYPTOPP_ALTIVEC_AVAILABLE
ANONYMOUS_NAMESPACE_END ANONYMOUS_NAMESPACE_END
@ -743,17 +780,17 @@ size_t SPECK64_Dec_AdvancedProcessBlocks_SSE41(const word32* subKeys, size_t rou
} }
#endif #endif
// ***************************** Power7 ***************************** // // ***************************** Altivec ***************************** //
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_ALTIVEC_AVAILABLE)
size_t SPECK64_Enc_AdvancedProcessBlocks_POWER7(const word32* subKeys, size_t rounds, size_t SPECK64_Enc_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Enc_Block, SPECK64_Enc_6_Blocks, return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Enc_Block, SPECK64_Enc_6_Blocks,
subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags); subKeys, rounds, inBlocks, xorBlocks, outBlocks, length, flags);
} }
size_t SPECK64_Dec_AdvancedProcessBlocks_POWER7(const word32* subKeys, size_t rounds, size_t SPECK64_Dec_AdvancedProcessBlocks_ALTIVEC(const word32* subKeys, size_t rounds,
const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags) const byte *inBlocks, const byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)
{ {
return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Dec_Block, SPECK64_Dec_6_Blocks, return AdvancedProcessBlocks64_6x2_ALTIVEC(SPECK64_Dec_Block, SPECK64_Dec_6_Blocks,