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

1 files changed, 20 insertions, 12 deletions

diff --git a/adv_simd.h b/adv_simd.h
index 76954e99..c79a9f25 100644
--- a/adv_simd.h
+++ b/adv_simd.h
@@ -74,6 +74,14 @@
 # include "ppc_simd.h"

 #endif

 

+#ifndef CRYPTOPP_INLINE

+# if defined(CRYPTOPP_DEBUG)

+#  define CRYPTOPP_INLINE static

+# else

+#  define CRYPTOPP_INLINE inline

+# endif

+#endif

+

 // ************************ All block ciphers *********************** //

 

 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

 ///   same word type.

 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 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

 ///   same word type.

 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 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

 ///   deduction/substitution failures.

 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 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

 ///   same word type.

 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 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

 ///   same word type.

 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,

         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

 ///   same word type.

 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,

         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

 ///   same word type.

 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,

         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

 ///   same word type.

 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,

         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

 ///   same word type.

 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,

     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

 ///   same word type.

 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 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

 ///   same word type.

 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 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

 ///   same word type.

 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 byte *xorBlocks, byte *outBlocks, size_t length, word32 flags)

 {