Merge pull request #1204 from yangfl/upstream

use builtin __AVX__ and __AES__ macros and reduce code duplication

Makefile.homebrew

@@ -34,7 +34,7 @@ endif
 # Seems like all recent Mac's have AES-NI, after firmware upgrade 2.2
 # Found no good way to detect it from command line. TODO: Might be some osx sysinfo magic
 ifeq ($(USE_AESNI),yes)
-	CXXFLAGS += -maes -DAESNI
+	CXXFLAGS += -maes
 endif
 ifeq ($(USE_AVX),1)
 	CXXFLAGS += -mavx

Makefile.linux

@@ -64,7 +64,7 @@ ifneq ($(shell $(GREP) -c aes /proc/cpuinfo),0)
 	ifeq ($(machine), aarch64)
 		CXXFLAGS += -DARM64AES
 	else
-		CPU_FLAGS += -maes -DAESNI
+		CPU_FLAGS += -maes
 	endif
 endif
 endif

Makefile.mingw

@@ -37,7 +37,7 @@ endif
 
 # don't change following line to ifeq ($(USE_AESNI),yes) !!!
 ifeq ($(USE_AESNI),1)
-	CPU_FLAGS += -maes -DAESNI
+	CPU_FLAGS += -maes
 else
 	CPU_FLAGS += -msse
 endif

Makefile.osx

@@ -21,7 +21,7 @@ ifeq ($(USE_UPNP),yes)
 endif
 
 ifeq ($(USE_AESNI),1)
-	CXXFLAGS += -maes -DAESNI
+	CXXFLAGS += -maes
 else
 	CXXFLAGS += -msse
 endif

build/CMakeLists.txt

@@ -234,7 +234,6 @@ endif ()
 
 if (WITH_AESNI)
   set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maes" )
-  add_definitions ( -DAESNI )
 endif()
 
 if (WITH_AVX)

libi2pd/CPU.cpp

@@ -21,23 +21,35 @@ namespace cpu
 
 	void Detect()
 	{
+#if defined(__AES__) || defined(__AVX__)
+
 #if defined(__x86_64__) || defined(__i386__)
 		int info[4];
 		__cpuid(0, info[0], info[1], info[2], info[3]);
 		if (info[0] >= 0x00000001) {
 			__cpuid(0x00000001, info[0], info[1], info[2], info[3]);
+#ifdef __AES__
 			aesni = info[2] & bit_AES;  // AESNI
+#endif  // __AES__
+#ifdef __AVX__
 			avx = info[2] & bit_AVX;  // AVX
+#endif  // __AVX__
 		}
-#endif
+#endif  // defined(__x86_64__) || defined(__i386__)
+
+#ifdef __AES__
 		if(aesni)
 		{
 			LogPrint(eLogInfo, "AESNI enabled");
 		}
+#endif  // __AES__
+#ifdef __AVX__
 		if(avx)
 		{
 			LogPrint(eLogInfo, "AVX enabled");
 		}
+#endif  // __AVX__
+#endif  // defined(__AES__) || defined(__AVX__)
 	}
 }
 }

libi2pd/Crypto.cpp

@@ -522,9 +522,9 @@ namespace crypto
 	{
 		uint64_t buf[256];
 		uint64_t hash[12]; // 96 bytes
+#ifdef __AVX__
 		if(i2p::cpu::avx)
 		{
-#ifdef AVX
 			__asm__
 				(
 					"vmovups %[key], %%ymm0 \n"
@@ -543,30 +543,9 @@ namespace crypto
 						[buf]"r"(buf), [hash]"r"(hash)
 					: "memory", "%xmm0"	// TODO: change to %ymm0 later
 					);
-#else
-			// ikeypad
-			buf[0] = key.GetLL ()[0] ^ IPAD;
-			buf[1] = key.GetLL ()[1] ^ IPAD;
-			buf[2] = key.GetLL ()[2] ^ IPAD;
-			buf[3] = key.GetLL ()[3] ^ IPAD;
-			buf[4] = IPAD;
-			buf[5] = IPAD;
-			buf[6] = IPAD;
-			buf[7] = IPAD;
-			// okeypad
-			hash[0] = key.GetLL ()[0] ^ OPAD;
-			hash[1] = key.GetLL ()[1] ^ OPAD;
-			hash[2] = key.GetLL ()[2] ^ OPAD;
-			hash[3] = key.GetLL ()[3] ^ OPAD;
-			hash[4] = OPAD;
-			hash[5] = OPAD;
-			hash[6] = OPAD;
-			hash[7] = OPAD;
-			// fill last 16 bytes with zeros (first hash size assumed 32 bytes in I2P)
-			memset (hash + 10, 0, 16);
-#endif
 		}
 		else
+#endif
 		{
 			// ikeypad
 			buf[0] = key.GetLL ()[0] ^ IPAD;
@@ -600,7 +579,7 @@ namespace crypto
 	}
 
 // AES
-#ifdef AESNI
+#ifdef __AES__
         #ifdef ARM64AES
                 void init_aesenc(void){
 			// TODO: Implementation
@@ -632,7 +611,7 @@ namespace crypto
 		"movaps	%%xmm3, "#round1"(%[sched]) \n"
 #endif
 
-#ifdef AESNI
+#ifdef __AES__
 	void ECBCryptoAESNI::ExpandKey (const AESKey& key)
 	{
 		__asm__
@@ -673,7 +652,7 @@ namespace crypto
 #endif
 
 
-#if AESNI
+#ifdef __AES__
 	#define EncryptAES256(sched) \
 		"pxor (%["#sched"]), %%xmm0 \n" \
 		"aesenc	16(%["#sched"]), %%xmm0 \n" \
@@ -694,9 +673,9 @@ namespace crypto
 
 	void ECBEncryption::Encrypt (const ChipherBlock * in, ChipherBlock * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[in]), %%xmm0 \n"
@@ -704,17 +683,15 @@ namespace crypto
 					"movups	%%xmm0, (%[out]) \n"
 					: : [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out) : "%xmm0", "memory"
 					);
-#else
-			AES_encrypt (in->buf, out->buf, &m_Key);
-#endif
 		}
 		else
+#endif
 		{
 			AES_encrypt (in->buf, out->buf, &m_Key);
 		}
 	}
 
-#ifdef AESNI
+#ifdef __AES__
 	#define DecryptAES256(sched) \
 		"pxor 224(%["#sched"]), %%xmm0 \n" \
 		"aesdec	208(%["#sched"]), %%xmm0 \n" \
@@ -735,9 +712,9 @@ namespace crypto
 
 	void ECBDecryption::Decrypt (const ChipherBlock * in, ChipherBlock * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[in]), %%xmm0 \n"
@@ -745,17 +722,15 @@ namespace crypto
 					"movups	%%xmm0, (%[out]) \n"
 					: : [sched]"r"(GetKeySchedule ()), [in]"r"(in), [out]"r"(out) : "%xmm0", "memory"
 					);
-#else
-			AES_decrypt (in->buf, out->buf, &m_Key);
-#endif
 		}
 		else
+#endif
 		{
 			AES_decrypt (in->buf, out->buf, &m_Key);
 		}
 	}
 
-#ifdef AESNI
+#ifdef __AES__
 	#define CallAESIMC(offset) \
 		"movaps "#offset"(%[shed]), %%xmm0 \n"	\
 		"aesimc %%xmm0, %%xmm0 \n" \
@@ -764,15 +739,13 @@ namespace crypto
 
 	void ECBEncryption::SetKey (const AESKey& key)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			ExpandKey (key);
-#else
-			AES_set_encrypt_key (key, 256, &m_Key);
-#endif
 		}
 		else
+#endif
 		{
 			AES_set_encrypt_key (key, 256, &m_Key);
 		}
@@ -780,9 +753,9 @@ namespace crypto
 
 	void ECBDecryption::SetKey (const AESKey& key)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			ExpandKey (key); // expand encryption key first
 			// then  invert it using aesimc
 			__asm__
@@ -802,11 +775,9 @@ namespace crypto
 					CallAESIMC(208)
 					: : [shed]"r"(GetKeySchedule ()) : "%xmm0", "memory"
 					);
-#else
-			AES_set_decrypt_key (key, 256, &m_Key);
-#endif
 		}
 		else
+#endif
 		{
 			AES_set_decrypt_key (key, 256, &m_Key);
 		}
@@ -815,9 +786,9 @@ namespace crypto
 
 	void CBCEncryption::Encrypt (int numBlocks, const ChipherBlock * in, ChipherBlock * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[iv]), %%xmm1 \n"
@@ -837,16 +808,9 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
 					: "%xmm0", "%xmm1", "cc", "memory"
 					);
-#else
-			for (int i = 0; i < numBlocks; i++)
-			{
-				*m_LastBlock.GetChipherBlock () ^= in[i];
-				m_ECBEncryption.Encrypt (m_LastBlock.GetChipherBlock (), m_LastBlock.GetChipherBlock ());
-				out[i] = *m_LastBlock.GetChipherBlock ();
-			}
-#endif
 		}
 		else
+#endif
 		{
 			for (int i = 0; i < numBlocks; i++)
 			{
@@ -867,9 +831,9 @@ namespace crypto
 
 	void CBCEncryption::Encrypt (const uint8_t * in, uint8_t * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[iv]), %%xmm1 \n"
@@ -883,19 +847,17 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out)
 					: "%xmm0", "%xmm1", "memory"
 					);
-#else
-			Encrypt (1, (const ChipherBlock *)in, (ChipherBlock *)out);
-#endif
 		}
 		else
+#endif
 			Encrypt (1, (const ChipherBlock *)in, (ChipherBlock *)out);
 	}
 
 	void CBCDecryption::Decrypt (int numBlocks, const ChipherBlock * in, ChipherBlock * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[iv]), %%xmm1 \n"
@@ -916,17 +878,9 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out), [num]"r"(numBlocks)
 					: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
 					);
-#else
-			for (int i = 0; i < numBlocks; i++)
-			{
-				ChipherBlock tmp = in[i];
-				m_ECBDecryption.Decrypt (in + i, out + i);
-				out[i] ^= *m_IV.GetChipherBlock ();
-				*m_IV.GetChipherBlock () = tmp;
-			}
-#endif
 		}
 		else
+#endif
 		{
 			for (int i = 0; i < numBlocks; i++)
 			{
@@ -947,9 +901,9 @@ namespace crypto
 
 	void CBCDecryption::Decrypt (const uint8_t * in, uint8_t * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					"movups	(%[iv]), %%xmm1 \n"
@@ -963,19 +917,17 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out)
 					: "%xmm0", "%xmm1", "memory"
 					);
-#else
-			Decrypt (1, (const ChipherBlock *)in, (ChipherBlock *)out);
-#endif
 		}
 		else
+#endif
 			Decrypt (1, (const ChipherBlock *)in, (ChipherBlock *)out);
 	}
 
 	void TunnelEncryption::Encrypt (const uint8_t * in, uint8_t * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					// encrypt IV
@@ -1001,14 +953,9 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
 					: "%xmm0", "%xmm1", "cc", "memory"
 					);
-#else
-			m_IVEncryption.Encrypt ((const ChipherBlock *)in, (ChipherBlock *)out); // iv
-			m_LayerEncryption.SetIV (out);
-			m_LayerEncryption.Encrypt (in + 16, i2p::tunnel::TUNNEL_DATA_ENCRYPTED_SIZE, out + 16); // data
-			m_IVEncryption.Encrypt ((ChipherBlock *)out, (ChipherBlock *)out); // double iv
-#endif
 		}
 		else
+#endif
 		{
 			m_IVEncryption.Encrypt ((const ChipherBlock *)in, (ChipherBlock *)out); // iv
 			m_LayerEncryption.SetIV (out);
@@ -1019,9 +966,9 @@ namespace crypto
 
 	void TunnelDecryption::Decrypt (const uint8_t * in, uint8_t * out)
 	{
+#ifdef __AES__
 		if(i2p::cpu::aesni)
 		{
-#ifdef AESNI
 			__asm__
 				(
 					// decrypt IV
@@ -1048,14 +995,9 @@ namespace crypto
 						[in]"r"(in), [out]"r"(out), [num]"r"(63) // 63 blocks = 1008 bytes
 					: "%xmm0", "%xmm1", "%xmm2", "cc", "memory"
 					);
-#else
-			m_IVDecryption.Decrypt ((const ChipherBlock *)in, (ChipherBlock *)out); // iv
-			m_LayerDecryption.SetIV (out);
-			m_LayerDecryption.Decrypt (in + 16, i2p::tunnel::TUNNEL_DATA_ENCRYPTED_SIZE, out + 16); // data
-			m_IVDecryption.Decrypt ((ChipherBlock *)out, (ChipherBlock *)out); // double iv
-#endif
 		}
 		else
+#endif
 		{
 			m_IVDecryption.Decrypt ((const ChipherBlock *)in, (ChipherBlock *)out); // iv
 			m_LayerDecryption.SetIV (out);

libi2pd/Crypto.h

@@ -69,9 +69,9 @@ namespace crypto
 
 		void operator^=(const ChipherBlock& other) // XOR
 		{
+#ifdef __AVX__
 			if (i2p::cpu::avx)
 			{
-#ifdef AVX
 				__asm__
 					(
 						"vmovups (%[buf]), %%xmm0 \n"
@@ -82,12 +82,9 @@ namespace crypto
 						: [buf]"r"(buf), [other]"r"(other.buf)
 						: "%xmm0", "%xmm1", "memory"
 						);
-#else
-				for (int i = 0; i < 16; i++)
-					buf[i] ^= other.buf[i];
-#endif
 			}
 			else
+#endif
 			{
 				// TODO: implement it better
 				for (int i = 0; i < 16; i++)
@@ -123,7 +120,7 @@ namespace crypto
 	};
 
 
-#ifdef AESNI
+#ifdef __AES__
 	#ifdef ARM64AES
 		void init_aesenc(void) __attribute__((constructor));
 	#endif
@@ -143,7 +140,7 @@ namespace crypto
 	};
 #endif
 
-#ifdef AESNI
+#ifdef __AES__
 	class ECBEncryption: public ECBCryptoAESNI
 #else
 	class ECBEncryption
@@ -159,7 +156,7 @@ namespace crypto
 		AES_KEY m_Key;
 	};
 
-#ifdef AESNI
+#ifdef __AES__
 	class ECBDecryption: public ECBCryptoAESNI
 #else
 	class ECBDecryption

libi2pd/Identity.cpp

@@ -719,7 +719,9 @@ namespace data
 	XORMetric operator^(const IdentHash& key1, const IdentHash& key2)
 	{
 		XORMetric m;
-#if defined(__AVX__) // for AVX
+#ifdef __AVX__
+		if(i2p::cpu::avx)
+		{
 			__asm__
 			(
 				"vmovups %1, %%ymm0 \n"
@@ -730,13 +732,16 @@ namespace data
 				: "m"(*key1), "m"(*key2)
 				: "memory", "%xmm0", "%xmm1" // should be replaced by %ymm0/1 once supported by compiler
 			);
-#else
+		}
+		else
+#endif
+		{
 			const uint64_t * hash1 = key1.GetLL (), * hash2 = key2.GetLL ();
 			m.metric_ll[0] = hash1[0] ^ hash2[0];
 			m.metric_ll[1] = hash1[1] ^ hash2[1];
 			m.metric_ll[2] = hash1[2] ^ hash2[2];
 			m.metric_ll[3] = hash1[3] ^ hash2[3];
-#endif
+		}
 
 		return m;
 	}