fix compile on GCC 4.4 and greater

Doxyfile

@@ -369,13 +369,13 @@ QUIET                  = NO
 # generated by doxygen. Possible values are YES and NO. If left blank 
 # NO is used.
 
-WARNINGS               = No
+WARNINGS               = NO
 
 # If WARN_IF_UNDOCUMENTED is set to YES, then doxygen will generate warnings 
 # for undocumented members. If EXTRACT_ALL is set to YES then this flag will 
 # automatically be disabled.
 
-WARN_IF_UNDOCUMENTED   = No
+WARN_IF_UNDOCUMENTED   = NO
 
 # If WARN_IF_DOC_ERROR is set to YES, doxygen will generate warnings for 
 # potential errors in the documentation, such as not documenting some 

GNUmakefile

@@ -34,7 +34,7 @@ ifneq ($(GCC42_OR_LATER),0)
 ifeq ($(UNAME),Darwin)
 CXXFLAGS += -arch x86_64 -arch i386
 else
-CXXFLAGS += -march=native -mtune=native
+CXXFLAGS += -march=native
 endif
 endif
 

cpu.h

@@ -12,10 +12,88 @@
 
 #include "config.h"
 
-#ifdef CRYPTOPP_MSVC6PP_OR_LATER
+#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
 #include <emmintrin.h>
 #endif
 
+#if CRYPTOPP_BOOL_AESNI_INTRINSICS_AVAILABLE
+#if !defined(__GNUC__) || defined(__SSSE3__)
+#include <tmmintrin.h>
+#else
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_shuffle_epi8 (__m128i a, __m128i b)
+{
+	asm ("pshufb %1, %0" : "+x"(a) : "xm"(b));
+  	return a;
+}
+#endif
+#if !defined(__GNUC__) || defined(__SSE4_1__)
+#include <smmintrin.h>
+#else
+__inline int __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_extract_epi32 (__m128i a, const int i)
+{
+	int r;
+	asm ("pextrd %2, %1, %0" : "=rm"(r) : "x"(a), "i"(i));
+  	return r;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_insert_epi32 (__m128i a, int b, const int i)
+{
+	asm ("pinsrd %2, %1, %0" : "+x"(a) : "rm"(b), "i"(i));
+  	return a;
+}
+#endif
+#if !defined(__GNUC__) || (defined(__AES__) && defined(__PCLMUL__))
+#include <wmmintrin.h>
+#else
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_clmulepi64_si128 (__m128i a, __m128i b, const int i)
+{
+	asm ("pclmulqdq %2, %1, %0" : "+x"(a) : "xm"(b), "i"(i));
+  	return a;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aeskeygenassist_si128 (__m128i a, const int i)
+{
+	__m128i r;
+	asm ("aeskeygenassist %2, %1, %0" : "=x"(r) : "xm"(a), "i"(i));
+  	return r;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aesimc_si128 (__m128i a)
+{
+	__m128i r;
+	asm ("aesimc %1, %0" : "=x"(r) : "xm"(a));
+  	return r;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aesenc_si128 (__m128i a, __m128i b)
+{
+	asm ("aesenc %1, %0" : "+x"(a) : "xm"(b));
+  	return a;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aesenclast_si128 (__m128i a, __m128i b)
+{
+	asm ("aesenclast %1, %0" : "+x"(a) : "xm"(b));
+  	return a;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aesdec_si128 (__m128i a, __m128i b)
+{
+	asm ("aesdec %1, %0" : "+x"(a) : "xm"(b));
+  	return a;
+}
+__inline __m128i __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_aesdeclast_si128 (__m128i a, __m128i b)
+{
+	asm ("aesdeclast %1, %0" : "+x"(a) : "xm"(b));
+  	return a;
+}
+#endif
+#endif
+
 NAMESPACE_BEGIN(CryptoPP)
 
 #if CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X64

gcm.cpp

@@ -10,15 +10,6 @@
 #include "gcm.h"
 #include "cpu.h"
 
-#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
-#include <emmintrin.h>
-#endif
-
-#if CRYPTOPP_BOOL_AESNI_INTRINSICS_AVAILABLE
-#include <tmmintrin.h>
-#include <wmmintrin.h>
-#endif
-
 NAMESPACE_BEGIN(CryptoPP)
 
 word16 GCM_Base::s_reductionTable[256];
@@ -88,9 +79,9 @@ inline static void Xor16(byte *a, const byte *b, const byte *c)
 
 #if CRYPTOPP_BOOL_AESNI_INTRINSICS_AVAILABLE
 static CRYPTOPP_ALIGN_DATA(16) const word64 s_clmulConstants64[] = {
-	0xe100000000000000, 0xc200000000000000,
+	W64LIT(0xe100000000000000), W64LIT(0xc200000000000000),
-	0x08090a0b0c0d0e0f, 0x0001020304050607,
+	W64LIT(0x08090a0b0c0d0e0f), W64LIT(0x0001020304050607),
-	0x0001020304050607, 0x08090a0b0c0d0e0f};
+	W64LIT(0x0001020304050607), W64LIT(0x08090a0b0c0d0e0f)};
 static const __m128i *s_clmulConstants = (const __m128i *)s_clmulConstants64;
 static const unsigned int s_clmulTableSizeInBlocks = 8;
 

rijndael.cpp

@@ -73,11 +73,6 @@ being unloaded from L1 cache, until that round is finished.
 #include "misc.h"
 #include "cpu.h"
 
-#if CRYPTOPP_BOOL_AESNI_INTRINSICS_AVAILABLE
-#include <smmintrin.h>
-#include <wmmintrin.h>
-#endif
-
 NAMESPACE_BEGIN(CryptoPP)
 
 #ifdef CRYPTOPP_ALLOW_UNALIGNED_DATA_ACCESS