Drop ChaCha requirements to POWER7

This costs about 0.6 cpb (700 MB/s on GCC112), but it makes the faster algorithm available to more machines. In the future we may want to provide both POWER7 and POWER8

1 files changed, 29 insertions, 8 deletions

diff --git a/chacha_simd.cpp b/chacha_simd.cpp
index 6e619281..5190f161 100644
--- a/chacha_simd.cpp
+++ b/chacha_simd.cpp
@@ -54,7 +54,7 @@
 # include <arm_acle.h>

 #endif

 

-#if defined(CRYPTOPP_POWER8_AVAILABLE)

+#if defined(CRYPTOPP_POWER7_AVAILABLE)

 # include "ppc_simd.h"

 #endif

 

@@ -201,16 +201,25 @@ inline __m128i RotateLeft<16>(const __m128i val)
 

 #endif  // CRYPTOPP_SSE2_INTRIN_AVAILABLE || CRYPTOPP_SSE2_ASM_AVAILABLE

 

-// **************************** POWER8 **************************** //

+// **************************** POWER7 **************************** //

 

-#if (CRYPTOPP_POWER8_AVAILABLE)

+#if (CRYPTOPP_POWER7_AVAILABLE)

+

+// POWER8 is optional and runs about 0.6 cpb faster because

+// of the native 64-bit vector add. That's about 700 MB/s on

+// GCC112 from the compile farm. Use -mcpu=power8 to engage

+// POWER8. POWER7 lacks 64-bit element support, so code built

+// with -mcpu=power8 will SIGILL on POWER7 machines.

 

 using CryptoPP::uint8x16_p;

 using CryptoPP::uint32x4_p;

-using CryptoPP::uint64x2_p;

 using CryptoPP::VectorLoad;

 using CryptoPP::VectorStore;

 

+#if (_ARCH_PWR8 || _ARCH_PWR9)

+using CryptoPP::uint64x2_p;

+#endif

+

 // Permutes bytes in packed 32-bit words to little endian.

 // State is already in proper endian order. Input and

 // output must be permuted during load and save.

@@ -290,10 +299,22 @@ inline uint32x4_p Shuffle<3>(const uint32x4_p& val)
 // Helper to perform 64-bit addition across two elements of 32-bit vectors

 inline uint32x4_p VectorAdd64(const uint32x4_p& a, const uint32x4_p& b)

 {

+#if (_ARCH_PWR8 || _ARCH_PWR9)

     return (uint32x4_p)vec_add((uint64x2_p)a, (uint64x2_p)b);

+#else

+	// The carry mask selects carries from elements 1 and 3 and sets remaining

+	// elements to 0. The mask also shifts the carried values left by 4 bytes

+	// so the carries are added to elements 0 and 2.

+	const uint8x16_p cmask = {4,5,6,7, 16,16,16,16, 12,13,14,15, 16,16,16,16};

+	const uint32x4_p zero = {0, 0, 0, 0};

+

+	uint32x4_p cy = vec_addc(a, b);

+    cy = vec_perm(cy, zero, cmask);

+    return vec_add(vec_add(a, b), cy);

+#endif

 }

 

-#endif  // CRYPTOPP_POWER8_AVAILABLE

+#endif  // CRYPTOPP_POWER7_AVAILABLE

 

 ANONYMOUS_NAMESPACE_END

 

@@ -835,9 +856,9 @@ void ChaCha_OperateKeystream_SSE2(const word32 *state, const byte* input, byte *
 

 #endif  // CRYPTOPP_SSE2_INTRIN_AVAILABLE || CRYPTOPP_SSE2_ASM_AVAILABLE

 

-#if (CRYPTOPP_POWER8_AVAILABLE)

+#if (CRYPTOPP_POWER7_AVAILABLE)

 

-void ChaCha_OperateKeystream_POWER8(const word32 *state, const byte* input, byte *output, unsigned int rounds)

+void ChaCha_OperateKeystream_POWER7(const word32 *state, const byte* input, byte *output, unsigned int rounds)

 {

     const uint32x4_p state0 = VectorLoad(state + 0*4);

     const uint32x4_p state1 = VectorLoad(state + 1*4);

@@ -1099,6 +1120,6 @@ void ChaCha_OperateKeystream_POWER8(const word32 *state, const byte* input, byte
     VectorStore32LE(output + 15*16, r3_3);

 }

 

-#endif  // CRYPTOPP_POWER8_AVAILABLE

+#endif  // CRYPTOPP_POWER7_AVAILABLE

 

 NAMESPACE_END