Disable same buffer for in and out on ARM A-32 (GH #683)

aes-armv4.S

@@ -32,6 +32,9 @@
 @ Profiler-assisted and platform-specific optimization resulted in 16%
 @ improvement on Cortex A8 core and ~21.5 cycles per byte.
 
+@ JW, JUL 2018: Begin defines from taken from arm_arch.h
+@               The defines were included through the header.
+
 # if !defined(__ARM_ARCH__)
 #  if defined(__CC_ARM)
 #   define __ARM_ARCH__ __TARGET_ARCH_ARM
@@ -86,6 +89,8 @@
 #  endif
 # endif
 
+@ JW, JUL 2018: End defines from taken from arm_arch.h
+@               Back to original Cryptogams code
 
 .text
 #if defined(__thumb2__) && !defined(__APPLE__)

rijndael.cpp

@@ -253,6 +253,10 @@ ANONYMOUS_NAMESPACE_END
 unsigned int Rijndael::Base::OptimalDataAlignment() const
 {
 	// CFB mode performs an extra memcpy if buffer is not aligned.
+#if (CRYPTOPP_AESNI_AVAILABLE)
+	if (HasAESNI())
+		return 1;
+#endif
 #if (CRYPTOPP_ARM_AES_AVAILABLE)
 	if (HasAES())
 		return 1;

strciphr.cpp

@@ -195,6 +195,8 @@ void CFB_CipherTemplate<BASE>::ProcessData(byte *outString, const byte *inString
 		if (!length) {return;}
 	}
 
+	// TODO: Figure out what is happening on ARM A-32. x86, Aarch64 and PowerPC are OK.
+#if !defined(__arm__)
 	if (policy.CanIterate() && length >= bytesPerIteration && IsAlignedOn(outString, alignment))
 	{
 		const CipherDir cipherDir = GetCipherDir(*this);
@@ -202,7 +204,9 @@ void CFB_CipherTemplate<BASE>::ProcessData(byte *outString, const byte *inString
 			policy.Iterate(outString, inString, cipherDir, length / bytesPerIteration);
 		else
 		{
-			// GCC and Clang does not like this on ARM.
+			// GCC and Clang does not like this on ARM. If we create
+			// an aligned temp input buffer, memcpy inString to it,
+			// and then use the temp input then things are [mostly] OK.
 			memcpy(outString, inString, length);
 			policy.Iterate(outString, outString, cipherDir, length / bytesPerIteration);
 		}
@@ -210,6 +214,7 @@ void CFB_CipherTemplate<BASE>::ProcessData(byte *outString, const byte *inString
 		outString = PtrAdd(outString, length - length % bytesPerIteration);
 		length %= bytesPerIteration;
 	}
+#endif
 
 	while (length >= bytesPerIteration)
 	{