Cherry-pick some commits from Crypto++ to make OpenMP for scrypt work on Windows

vendor/README

@@ -5,3 +5,8 @@ spdlog: https://github.com/gabime/spdlog/tree/v0.16.3/include/spdlog
 cryptopp: https://github.com/weidai11/cryptopp/tree/CRYPTOPP_8_0_0
  - changed: added CMakeLists.txt and cryptopp-config.cmake from https://github.com/noloader/cryptopp-cmake/tree/CRYPTOPP_8_0_0
  - changed: commented out line including winapifamily.h in CMakeLists.txt
+ - cherry-picked commits to get OpenMP for scrypt on Windows:
+   - https://github.com/weidai11/cryptopp/commit/aa043b38a7930725c31a0cd7016986d1c581c573
+   - https://github.com/weidai11/cryptopp/commit/672f5c7f3dad8ae12b2d0ce0940ccb7c8e257bf8
+   - https://github.com/weidai11/cryptopp/commit/7e96a283a3192d29aac5b60e5b4ff19248f00d9a
+   - https://github.com/weidai11/cryptopp/commit/ca32b63038d5f7b13e2e00809cd9184a1efe8c24

vendor/cryptopp/vendor_cryptopp/cryptest.nmake

@@ -139,6 +139,7 @@ LDLIBS =
 # CXXFLAGS = $(CXXFLAGS) /DDEBUG /D_DEBUG /Oi /Oy- /Od /MTd
 # Release build. Add /OPT:REF to linker
 CXXFLAGS = $(CXXFLAGS) /DNDEBUG /D_NDEBUG /Oi /Oy /O2 /MT
+# Linker flags.
 LDFLAGS = $(LDFLAGS) /OPT:REF
 
 # Attempt to detect when <sdkddkver.h> and <winapifamily.h> are available

vendor/cryptopp/vendor_cryptopp/salsa.cpp

@@ -90,9 +90,14 @@ void Salsa20_Core(word32* data, unsigned int rounds)
 		x[15] ^= rotlConstant<18>(x[14]+x[13]);
 	}
 
+#ifdef _MSC_VER
+	for (size_t i = 0; i < 16; ++i)
+		data[i] += x[i];
+#else
 	#pragma omp simd
 	for (size_t i = 0; i < 16; ++i)
 		data[i] += x[i];
+#endif
 }
 
 std::string Salsa20_Policy::AlgorithmProvider() const

vendor/cryptopp/vendor_cryptopp/scrypt.cpp

@@ -14,6 +14,8 @@
 #include "sha.h"
 
 #include <sstream>
+#include <limits>
+
 #ifdef _OPENMP
 # include <omp.h>
 #endif
@@ -53,9 +55,14 @@ static inline void BlockCopy(byte* dest, byte* src, size_t len)
 
 static inline void BlockXOR(byte* dest, byte* src, size_t len)
 {
+#ifdef _MSC_VER
+    for (size_t i = 0; i < len; ++i)
+        dest[i] ^= src[i];
+#else
     #pragma omp simd
     for (size_t i = 0; i < len; ++i)
         dest[i] ^= src[i];
+#endif
 }
 
 static inline void PBKDF2_SHA256(byte* buf, size_t dkLen,
@@ -171,6 +178,16 @@ void Scrypt::ValidateParameters(size_t derivedLen, word64 cost, word64 blockSize
         }
     }
 
+    // https://github.com/weidai11/cryptopp/issues/787
+    CRYPTOPP_ASSERT(parallelization <= std::numeric_limits<int>::max());
+    if (parallelization > static_cast<word64>(std::numeric_limits<int>::max()))
+    {
+        std::ostringstream oss;
+        oss << " parallelization " << parallelization << " is larger than ";
+        oss << std::numeric_limits<int>::max();
+        throw InvalidArgument("Scrypt: " + oss.str());
+    }
+
     CRYPTOPP_ASSERT(IsPowerOf2(cost));
     if (IsPowerOf2(cost) == false)
         throw InvalidArgument("Scrypt: cost must be a power of 2");
@@ -245,10 +262,13 @@ size_t Scrypt::DeriveKey(byte*derived, size_t derivedLen, const byte*secret, siz
     // 1: (B_0 ... B_{p-1}) <-- PBKDF2(P, S, 1, p * MFLen)
     PBKDF2_SHA256(B, B.size(), secret, secretLen, salt, saltLen, 1);
 
+    // Visual Studio and OpenMP 2.0 fixup. We must use int, not size_t.
+    int maxParallel=0;
+    if (!SafeConvert(parallel, maxParallel))
+        maxParallel = std::numeric_limits<int>::max();
+
     #ifdef _OPENMP
-    int threads = STDMIN(omp_get_max_threads(),
+    int threads = STDMIN(omp_get_max_threads(), maxParallel);
-        static_cast<int>(STDMIN(static_cast<size_t>(parallel),
-        static_cast<size_t>(std::numeric_limits<int>::max()))));
     #endif
 
     // http://stackoverflow.com/q/49604260/608639
@@ -260,7 +280,7 @@ size_t Scrypt::DeriveKey(byte*derived, size_t derivedLen, const byte*secret, siz
 
         // 2: for i = 0 to p - 1 do
         #pragma omp for
-        for (size_t i = 0; i < static_cast<size_t>(parallel); ++i)
+        for (int i = 0; i < maxParallel; ++i)
         {
             // 3: B_i <-- MF(B_i, N)
             const ptrdiff_t offset = static_cast<ptrdiff_t>(blockSize*i*128);

vendor/cryptopp/vendor_cryptopp/scrypt.h

@@ -76,7 +76,9 @@ public:
     /// \details The parameter <tt>blockSize</tt> ("r" in the documents) specifies the block
     ///   size.
     /// \details The <tt>parallelization</tt> parameter ("p" in the documents) is a positive
-    ///   integer less than or equal to <tt>((2^32-1) * 32) / (128 * r)</tt>.
+    ///   integer less than or equal to <tt>((2^32-1) * 32) / (128 * r)</tt>. Due to Microsoft
+    ///   and its OpenMP 2.0 implementation <tt>parallelization</tt> is limited to
+    ///   <tt>std::numeric_limits<int>::max()</tt>.
     /// \details Scrypt always returns 1 because it only performs 1 iteration. Other
     ///   derivation functions, like PBKDF's, will return more interesting values.
     /// \details The Crypto++ implementation of Scrypt is limited by C++ datatypes. For