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

Add CRYPTOPP_POWER5_AVAILABLE 

Power4 lacks 'vector long long'
Rename datatypes such as 'uint8x16_p8' to 'uint8x16_p'. Originally the p8 suffix indicated use with Power8 in-core crypto. We are now using Altivec/Power4 for general vector operations.
pull/548/head
Jeffrey Walton 2017-12-12 08:09:31 -05:00
parent b7e636ac51
commit 15d637771f
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
2 changed files with 45 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
config.h
View File

@ -629,7 +629,6 @@ NAMESPACE_END
#endif #endif
// An old Apple G5 with GCC 4.01 has AltiVec, but its only Power4 or so. // An old Apple G5 with GCC 4.01 has AltiVec, but its only Power4 or so.
// We need Power7 or above, so the makefile defines CRYPTOPP_DISABLE_ALTIVEC.
#if !defined(CRYPTOPP_ALTIVEC_AVAILABLE) && !defined(CRYPTOPP_DISABLE_ALTIVEC) #if !defined(CRYPTOPP_ALTIVEC_AVAILABLE) && !defined(CRYPTOPP_DISABLE_ALTIVEC)
# if defined(_ARCH_PWR4) || defined(__ALTIVEC__) || \ # if defined(_ARCH_PWR4) || defined(__ALTIVEC__) || \
(CRYPTOPP_XLC_VERSION >= 100000) || (CRYPTOPP_GCC_VERSION >= 40001) (CRYPTOPP_XLC_VERSION >= 100000) || (CRYPTOPP_GCC_VERSION >= 40001)
@ -637,12 +636,21 @@ NAMESPACE_END
# endif # endif
#endif #endif
// We need Power5 for 'vector unsigned long long'
#if !defined(CRYPTOPP_POWER5_AVAILABLE) && !defined(CRYPTOPP_DISABLE_POWER5) && defined(CRYPTOPP_ALTIVEC_AVAILABLE)
# if defined(_ARCH_PWR5) || (CRYPTOPP_XLC_VERSION >= 100000) || (CRYPTOPP_GCC_VERSION >= 40100)
# define CRYPTOPP_POWER5_AVAILABLE 1
# endif
#endif
// We need Power7 for unaligned loads and stores
#if !defined(CRYPTOPP_POWER7_AVAILABLE) && !defined(CRYPTOPP_DISABLE_POWER7) && defined(CRYPTOPP_ALTIVEC_AVAILABLE) #if !defined(CRYPTOPP_POWER7_AVAILABLE) && !defined(CRYPTOPP_DISABLE_POWER7) && defined(CRYPTOPP_ALTIVEC_AVAILABLE)
# if defined(_ARCH_PWR7) || (CRYPTOPP_XLC_VERSION >= 100000) || (CRYPTOPP_GCC_VERSION >= 40100) # if defined(_ARCH_PWR7) || (CRYPTOPP_XLC_VERSION >= 100000) || (CRYPTOPP_GCC_VERSION >= 40100)
# define CRYPTOPP_POWER7_AVAILABLE 1 # define CRYPTOPP_POWER7_AVAILABLE 1
# endif # endif
#endif #endif
// We need Power8 for in-core crypto
#if !defined(CRYPTOPP_POWER8_AVAILABLE) && !defined(CRYPTOPP_DISABLE_POWER8) && defined(CRYPTOPP_POWER7_AVAILABLE) #if !defined(CRYPTOPP_POWER8_AVAILABLE) && !defined(CRYPTOPP_DISABLE_POWER8) && defined(CRYPTOPP_POWER7_AVAILABLE)
# if defined(_ARCH_PWR8) || (CRYPTOPP_XLC_VERSION >= 130000) || (CRYPTOPP_GCC_VERSION >= 40800) # if defined(_ARCH_PWR8) || (CRYPTOPP_XLC_VERSION >= 130000) || (CRYPTOPP_GCC_VERSION >= 40800)
# define CRYPTOPP_POWER8_AVAILABLE 1 # define CRYPTOPP_POWER8_AVAILABLE 1

74
ppc-simd.h
View File

@ -29,27 +29,25 @@ NAMESPACE_BEGIN(CryptoPP)
#if defined(CRYPTOPP_ALTIVEC_AVAILABLE) || defined(CRYPTOPP_DOXYGEN_PROCESSING) #if defined(CRYPTOPP_ALTIVEC_AVAILABLE) || defined(CRYPTOPP_DOXYGEN_PROCESSING)
typedef __vector unsigned char uint8x16_p8; typedef __vector unsigned char uint8x16_p;
typedef __vector unsigned int uint32x4_p8; typedef __vector unsigned int uint32x4_p;
#if defined(CRYPTOPP_POWER7_AVAILABLE) #if defined(CRYPTOPP_POWER5_AVAILABLE)
typedef __vector unsigned long long uint64x2_p8; typedef __vector unsigned long long uint64x2_p;
#else
typedef __vector unsigned int uint64x2_p8;
#endif #endif
// Use 8x16 for documentation because it is used frequently // Use 8x16 for documentation because it is used frequently
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
typedef uint8x16_p8 VectorType; typedef uint8x16_p VectorType;
#elif defined(CRYPTOPP_GCC_VERSION) #elif defined(CRYPTOPP_GCC_VERSION)
typedef uint64x2_p8 VectorType; typedef uint64x2_p VectorType;
#endif #endif
#if defined(CRYPTOPP_DOXYGEN_PROCESSING) #if defined(CRYPTOPP_DOXYGEN_PROCESSING)
/// \brief Default vector typedef /// \brief Default vector typedef
/// \details IBM XL C/C++ provides equally good support for all vector types, /// \details IBM XL C/C++ provides equally good support for all vector types,
/// including <tt>uint8x16_p8</tt>. GCC provides good support for /// including <tt>uint8x16_p</tt>. GCC provides good support for
/// <tt>uint64x2_p8</tt>. <tt>VectorType</tt> is typedef'd accordingly to /// <tt>uint64x2_p</tt>. <tt>VectorType</tt> is typedef'd accordingly to
/// minimize casting to and from buit-in function calls. /// minimize casting to and from buit-in function calls.
# define VectorType ... # define VectorType ...
#endif #endif
@ -68,8 +66,8 @@ inline void ReverseByteArrayLE(byte src[16])
#if defined(CRYPTOPP_XLC_VERSION) && defined(CRYPTOPP_LITTLE_ENDIAN) #if defined(CRYPTOPP_XLC_VERSION) && defined(CRYPTOPP_LITTLE_ENDIAN)
vec_st(vec_reve(vec_ld(0, src)), 0, src); vec_st(vec_reve(vec_ld(0, src)), 0, src);
#elif defined(CRYPTOPP_LITTLE_ENDIAN) #elif defined(CRYPTOPP_LITTLE_ENDIAN)
const uint8x16_p8 mask = {15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0}; const uint8x16_p mask = {15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0};
const uint8x16_p8 zero = {0}; const uint8x16_p zero = {0};
vec_vsx_st(vec_perm(vec_vsx_ld(0, src), zero, mask), 0, src); vec_vsx_st(vec_perm(vec_vsx_ld(0, src), zero, mask), 0, src);
#endif #endif
} }
@ -83,8 +81,8 @@ inline void ReverseByteArrayLE(byte src[16])
template <class T> template <class T>
inline T Reverse(const T& src) inline T Reverse(const T& src)
{ {
const uint8x16_p8 mask = {15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0}; const uint8x16_p mask = {15,14,13,12, 11,10,9,8, 7,6,5,4, 3,2,1,0};
const uint8x16_p8 zero = {0}; const uint8x16_p zero = {0};
return vec_perm(src, zero, mask); return vec_perm(src, zero, mask);
} }
@ -216,12 +214,12 @@ template <class T>
inline void VectorStoreBE(const T& src, uint8_t dest[16]) inline void VectorStoreBE(const T& src, uint8_t dest[16])
{ {
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
vec_xst_be((uint8x16_p8)src, 0, (uint8_t*)dest); vec_xst_be((uint8x16_p)src, 0, (uint8_t*)dest);
#else #else
# if defined(CRYPTOPP_LITTLE_ENDIAN) # if defined(CRYPTOPP_LITTLE_ENDIAN)
vec_vsx_st(Reverse((uint8x16_p8)src), 0, (uint8_t*)dest); vec_vsx_st(Reverse((uint8x16_p)src), 0, (uint8_t*)dest);
# else # else
vec_vsx_st((uint8x16_p8)src, 0, (uint8_t*)dest); vec_vsx_st((uint8x16_p)src, 0, (uint8_t*)dest);
# endif # endif
#endif #endif
} }
@ -239,12 +237,12 @@ template <class T>
inline void VectorStoreBE(const T& src, int off, uint8_t dest[16]) inline void VectorStoreBE(const T& src, int off, uint8_t dest[16])
{ {
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
vec_xst_be((uint8x16_p8)src, off, (uint8_t*)dest); vec_xst_be((uint8x16_p)src, off, (uint8_t*)dest);
#else #else
# if defined(CRYPTOPP_LITTLE_ENDIAN) # if defined(CRYPTOPP_LITTLE_ENDIAN)
vec_vsx_st(Reverse((uint8x16_p8)src), off, (uint8_t*)dest); vec_vsx_st(Reverse((uint8x16_p)src), off, (uint8_t*)dest);
# else # else
vec_vsx_st((uint8x16_p8)src, off, (uint8_t*)dest); vec_vsx_st((uint8x16_p)src, off, (uint8_t*)dest);
# endif # endif
#endif #endif
} }
@ -262,12 +260,12 @@ inline void VectorStore(const T& src, byte dest[16])
{ {
// Do not call VectorStoreBE. It slows us down by about 0.5 cpb on LE. // Do not call VectorStoreBE. It slows us down by about 0.5 cpb on LE.
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
vec_xst_be((uint8x16_p8)src, 0, (uint8_t*)dest); vec_xst_be((uint8x16_p)src, 0, (uint8_t*)dest);
#else #else
# if defined(CRYPTOPP_LITTLE_ENDIAN) # if defined(CRYPTOPP_LITTLE_ENDIAN)
vec_vsx_st(Reverse((uint8x16_p8)src), 0, (uint8_t*)dest); vec_vsx_st(Reverse((uint8x16_p)src), 0, (uint8_t*)dest);
# else # else
vec_vsx_st((uint8x16_p8)src, 0, (uint8_t*)dest); vec_vsx_st((uint8x16_p)src, 0, (uint8_t*)dest);
# endif # endif
#endif #endif
} }
@ -286,12 +284,12 @@ inline void VectorStore(const T& src, int off, byte dest[16])
{ {
// Do not call VectorStoreBE. It slows us down by about 0.5 cpb on LE. // Do not call VectorStoreBE. It slows us down by about 0.5 cpb on LE.
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
vec_xst_be((uint8x16_p8)src, off, (uint8_t*)dest); vec_xst_be((uint8x16_p)src, off, (uint8_t*)dest);
#else #else
# if defined(CRYPTOPP_LITTLE_ENDIAN) # if defined(CRYPTOPP_LITTLE_ENDIAN)
vec_vsx_st(Reverse((uint8x16_p8)src), off, (uint8_t*)dest); vec_vsx_st(Reverse((uint8x16_p)src), off, (uint8_t*)dest);
# else # else
vec_vsx_st((uint8x16_p8)src, off, (uint8_t*)dest); vec_vsx_st((uint8x16_p)src, off, (uint8_t*)dest);
# endif # endif
#endif #endif
} }
@ -303,13 +301,13 @@ inline void VectorStore(const T& src, int off, byte dest[16])
/// \param vec2 the second vector /// \param vec2 the second vector
/// \param mask vector mask /// \param mask vector mask
/// \details VectorPermute returns a new vector from vec1 and vec2 /// \details VectorPermute returns a new vector from vec1 and vec2
/// based on mask. mask is an uint8x16_p8 type vector. The return /// based on mask. mask is an uint8x16_p type vector. The return
/// vector is the same type as vec1. /// vector is the same type as vec1.
/// \since Crypto++ 6.0 /// \since Crypto++ 6.0
template <class T1, class T2> template <class T1, class T2>
inline T1 VectorPermute(const T1& vec1, const T1& vec2, const T2& mask) inline T1 VectorPermute(const T1& vec1, const T1& vec2, const T2& mask)
{ {
return (T1)vec_perm(vec1, vec2, (uint8x16_p8)mask); return (T1)vec_perm(vec1, vec2, (uint8x16_p)mask);
} }
/// \brief XOR two vectors /// \brief XOR two vectors
@ -349,16 +347,16 @@ inline T1 VectorAdd(const T1& vec1, const T2& vec2)
/// \param vec2 the second vector /// \param vec2 the second vector
/// \details VectorShiftLeft() concatenates vec1 and vec2 and returns a /// \details VectorShiftLeft() concatenates vec1 and vec2 and returns a
/// new vector after shifting the concatenation by the specified number /// new vector after shifting the concatenation by the specified number
/// of bytes. Both vec1 and vec2 are cast to uint8x16_p8. The return /// of bytes. Both vec1 and vec2 are cast to uint8x16_p. The return
/// vector is the same type as vec1. /// vector is the same type as vec1.
/// \details On big endian machines VectorShiftLeft() is <tt>vec_sld(a, b, /// \details On big endian machines VectorShiftLeft() is <tt>vec_sld(a, b,
/// c)</tt>. On little endian machines VectorShiftLeft() is translated to /// c)</tt>. On little endian machines VectorShiftLeft() is translated to
/// <tt>vec_sld(b, a, 16-c)</tt>. You should always call the function as /// <tt>vec_sld(b, a, 16-c)</tt>. You should always call the function as
/// if on a big endian machine as shown below. /// if on a big endian machine as shown below.
/// <pre> /// <pre>
/// uint8x16_p8 r0 = {0}; /// uint8x16_p r0 = {0};
/// uint8x16_p8 r1 = VectorLoad(ptr); /// uint8x16_p r1 = VectorLoad(ptr);
/// uint8x16_p8 r5 = VectorShiftLeft<12>(r0, r1); /// uint8x16_p r5 = VectorShiftLeft<12>(r0, r1);
/// </pre> /// </pre>
/// \sa <A HREF="https://stackoverflow.com/q/46341923/608639">Is vec_sld /// \sa <A HREF="https://stackoverflow.com/q/46341923/608639">Is vec_sld
/// endian sensitive?</A> on Stack Overflow /// endian sensitive?</A> on Stack Overflow
@ -367,9 +365,9 @@ template <unsigned int C, class T1, class T2>
inline T1 VectorShiftLeft(const T1& vec1, const T2& vec2) inline T1 VectorShiftLeft(const T1& vec1, const T2& vec2)
{ {
#if defined(CRYPTOPP_LITTLE_ENDIAN) #if defined(CRYPTOPP_LITTLE_ENDIAN)
return (T1)vec_sld((uint8x16_p8)vec2, (uint8x16_p8)vec1, 16-C); return (T1)vec_sld((uint8x16_p)vec2, (uint8x16_p)vec1, 16-C);
#else #else
return (T1)vec_sld((uint8x16_p8)vec1, (uint8x16_p8)vec2, C); return (T1)vec_sld((uint8x16_p)vec1, (uint8x16_p)vec2, C);
#endif #endif
} }
@ -469,9 +467,9 @@ template <int func, int subfunc, class T>
inline T VectorSHA256(const T& vec) inline T VectorSHA256(const T& vec)
{ {
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
return (T)__vshasigmaw((uint32x4_p8)vec, func, subfunc); return (T)__vshasigmaw((uint32x4_p)vec, func, subfunc);
#elif defined(CRYPTOPP_GCC_VERSION) #elif defined(CRYPTOPP_GCC_VERSION)
return (T)__builtin_crypto_vshasigmaw((uint32x4_p8)vec, func, subfunc); return (T)__builtin_crypto_vshasigmaw((uint32x4_p)vec, func, subfunc);
#else #else
CRYPTOPP_ASSERT(0); CRYPTOPP_ASSERT(0);
#endif #endif
@ -489,9 +487,9 @@ template <int func, int subfunc, class T>
inline T VectorSHA512(const T& vec) inline T VectorSHA512(const T& vec)
{ {
#if defined(CRYPTOPP_XLC_VERSION) #if defined(CRYPTOPP_XLC_VERSION)
return (T)__vshasigmad((uint64x2_p8)vec, func, subfunc); return (T)__vshasigmad((uint64x2_p)vec, func, subfunc);
#elif defined(CRYPTOPP_GCC_VERSION) #elif defined(CRYPTOPP_GCC_VERSION)
return (T)__builtin_crypto_vshasigmad((uint64x2_p8)vec, func, subfunc); return (T)__builtin_crypto_vshasigmad((uint64x2_p)vec, func, subfunc);
#else #else
CRYPTOPP_ASSERT(0); CRYPTOPP_ASSERT(0);
#endif #endif