Add polynomial multiply instruction detection

cpu.cpp

@@ -319,8 +319,8 @@ void DetectX86Features()
 // http://community.arm.com/groups/android-community/blog/2014/10/10/runtime-detection-of-cpu-features-on-an-armv8-a-cpu
 //
 bool CRYPTOPP_SECTION_INIT g_ArmDetectionDone = false;
-bool CRYPTOPP_SECTION_INIT g_hasNEON = false, CRYPTOPP_SECTION_INIT g_hasCRC32 = false, CRYPTOPP_SECTION_INIT g_hasAES = false, CRYPTOPP_SECTION_INIT g_hasSHA1 = false;
+bool CRYPTOPP_SECTION_INIT g_hasNEON = false, CRYPTOPP_SECTION_INIT g_hasPMULL = false, CRYPTOPP_SECTION_INIT g_hasCRC32 = false;
-bool CRYPTOPP_SECTION_INIT g_hasSHA2 = false;
+bool CRYPTOPP_SECTION_INIT g_hasAES = false, CRYPTOPP_SECTION_INIT g_hasSHA1 = false, CRYPTOPP_SECTION_INIT g_hasSHA2 = false;
 word32 CRYPTOPP_SECTION_INIT g_cacheLineSize = CRYPTOPP_L1_CACHE_LINE_SIZE;
 
 #ifndef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
@@ -332,6 +332,12 @@ extern "C"
 		longjmp(s_jmpNoNEON, 1);
 	}
 
+	static jmp_buf s_jmpNoPMULL;
+	static void SigIllHandlerPMULL(int)
+	{
+		longjmp(s_jmpNoPMULL, 1);
+	}
+
 	static jmp_buf s_jmpNoCRC32;
 	static void SigIllHandlerCRC32(int)
 	{
@@ -426,6 +432,59 @@ static bool TryNEON()
 #endif  // CRYPTOPP_BOOL_NEON_INTRINSICS_AVAILABLE
 }
 
+static bool TryPMULL()
+{
+#if (CRYPTOPP_BOOL_ARM_CRYPTO_INTRINSICS_AVAILABLE)
+# if defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
+	volatile bool result = true;
+	__try
+	{
+		const poly64_t a1={1}, b1={2};
+		const poly64x2_t a2={1}, b2={2};
+		const poly128_t r1 = vmull_p64(a1, b1);
+		const poly128_t r2 = vmull_high_p64(a2, b2);
+
+		result = (r1 != r2);
+	}
+	__except (EXCEPTION_EXECUTE_HANDLER)
+	{
+		return false;
+	}
+	return result;
+# else
+	// longjmp and clobber warnings. Volatile is required.
+	// http://github.com/weidai11/cryptopp/issues/24 and http://stackoverflow.com/q/7721854
+	volatile bool result = true;
+
+	volatile SigHandler oldHandler = signal(SIGILL, SigIllHandlerPMULL);
+	if (oldHandler == SIG_ERR)
+		return false;
+
+	volatile sigset_t oldMask;
+	if (sigprocmask(0, NULL, (sigset_t*)&oldMask))
+		return false;
+
+	if (setjmp(s_jmpNoPMULL))
+		result = false;
+	else
+	{
+		const poly64_t a1={1}, b1={2};
+		const poly64x2_t a2={1}, b2={2};
+		const poly128_t r1 = vmull_p64(a1, b1);
+		const poly128_t r2 = vmull_high_p64(a2, b2);
+
+		result = (r1 != r2);
+	}
+
+	sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULL);
+	signal(SIGILL, oldHandler);
+	return result;
+# endif
+#else
+	return false;
+#endif  // CRYPTOPP_BOOL_CRYPTO_INTRINSICS_AVAILABLE
+}
+
 static bool TryCRC32()
 {
 #if (CRYPTOPP_BOOL_ARM_CRC32_INTRINSICS_AVAILABLE)
@@ -660,6 +719,7 @@ void DetectArmFeatures()
 #endif
 {
 	g_hasNEON = TryNEON();
+	g_hasPMULL = TryPMULL();
 	g_hasCRC32 = TryCRC32();
 	g_hasAES = TryAES();
 	g_hasSHA1 = TrySHA1();

cpu.h

@@ -364,7 +364,7 @@ inline int GetCacheLineSize()
 #elif (CRYPTOPP_BOOL_ARM32 || CRYPTOPP_BOOL_ARM64)
 
 extern bool g_ArmDetectionDone;
-extern bool g_hasNEON, g_hasCRC32, g_hasAES, g_hasSHA1, g_hasSHA2;
+extern bool g_hasNEON, g_hasPMULL, g_hasCRC32, g_hasAES, g_hasSHA1, g_hasSHA2;
 void CRYPTOPP_API DetectArmFeatures();
 
 //! \brief Determine if an ARM processor has Advanced SIMD available
@@ -380,6 +380,19 @@ inline bool HasNEON()
 	return g_hasNEON;
 }
 
+//! \brief Determine if an ARM processor provides Polynomial Multiplication (long)
+//! \returns true if the hardware is capable of polynomial multiplications at runtime, false otherwise.
+//! \details The multiplication instructions are available under Aarch64 (ARM-64) and Aarch32 (ARM-32).
+//! \details Runtime support requires compile time support. When compiling with GCC, you may
+//!   need to compile with <tt>-march=armv8-a+crypto</tt>; while Apple requires
+//!   <tt>-arch arm64</tt>. Also see ARM's <tt>__ARM_FEATURE_CRYPTO</tt> preprocessor macro.
+inline bool HasPMULL()
+{
+	if (!g_ArmDetectionDone)
+		DetectArmFeatures();
+	return g_hasPMULL;
+}
+
 //! \brief Determine if an ARM processor has CRC32 available
 //! \returns true if the hardware is capable of CRC32 at runtime, false otherwise.
 //! \details CRC32 instructions provide access to the processor's CRC32 and CRC32-C intructions.
@@ -485,20 +498,6 @@ inline int GetCacheLineSize()
 #else
 	#define CRYPTOPP_GNU_STYLE_INLINE_ASSEMBLY
 
-#if defined(CRYPTOPP_LLVM_CLANG_VERSION) || defined(CRYPTOPP_APPLE_CLANG_VERSION) || defined(CRYPTOPP_CLANG_INTEGRATED_ASSEMBLER)
-	#define NEW_LINE "\n"
-	#define INTEL_PREFIX ".intel_syntax;"
-	#define INTEL_NOPREFIX ".intel_syntax;"
-	#define ATT_PREFIX ".att_syntax;"
-	#define ATT_NOPREFIX ".att_syntax;"
-#else
-	#define NEW_LINE
-	#define INTEL_PREFIX ".intel_syntax prefix;"
-	#define INTEL_NOPREFIX ".intel_syntax noprefix;"
-	#define ATT_PREFIX ".att_syntax prefix;"
-	#define ATT_NOPREFIX ".att_syntax noprefix;"
-#endif
-
 	// define these in two steps to allow arguments to be expanded
 	#define GNU_AS1(x) #x ";" NEW_LINE
 	#define GNU_AS2(x, y) #x ", " #y ";" NEW_LINE
@@ -519,21 +518,6 @@ inline int GetCacheLineSize()
 #define IF0(y)
 #define IF1(y) y
 
-// Should be confined to GCC, but its used to help manage Clang 3.4 compiler error.
-//   Also see LLVM Bug 24232, http://llvm.org/bugs/show_bug.cgi?id=24232 .
-#ifndef INTEL_PREFIX
-	#define INTEL_PREFIX
-#endif
-#ifndef INTEL_NOPREFIX
-	#define INTEL_NOPREFIX
-#endif
-#ifndef ATT_PREFIX
-	#define ATT_PREFIX
-#endif
-#ifndef ATT_NOPREFIX
-	#define ATT_NOPREFIX
-#endif
-
 #ifdef CRYPTOPP_GENERATE_X64_MASM
 #define ASM_MOD(x, y) ((x) MOD (y))
 #define XMMWORD_PTR XMMWORD PTR
@@ -666,6 +650,21 @@ inline int GetCacheLineSize()
 
 #endif  //  X86/X32/X64
 
+// Applies to both X86/X32/X64 and ARM32/ARM64
+#if defined(CRYPTOPP_LLVM_CLANG_VERSION) || defined(CRYPTOPP_APPLE_CLANG_VERSION) || defined(CRYPTOPP_CLANG_INTEGRATED_ASSEMBLER)
+	#define NEW_LINE "\n"
+	#define INTEL_PREFIX ".intel_syntax;"
+	#define INTEL_NOPREFIX ".intel_syntax;"
+	#define ATT_PREFIX ".att_syntax;"
+	#define ATT_NOPREFIX ".att_syntax;"
+#else
+	#define NEW_LINE
+	#define INTEL_PREFIX ".intel_syntax prefix;"
+	#define INTEL_NOPREFIX ".intel_syntax noprefix;"
+	#define ATT_PREFIX ".att_syntax prefix;"
+	#define ATT_NOPREFIX ".att_syntax noprefix;"
+#endif
+
 NAMESPACE_END
 
 #endif  // CRYPTOPP_CPU_H

validat1.cpp

@@ -315,13 +315,14 @@ bool TestSettings()
 
 #elif (CRYPTOPP_BOOL_ARM32 || CRYPTOPP_BOOL_ARM64)
 	bool hasNEON = HasNEON();
+	bool hasPMULL = HasPMULL();
 	bool hasCRC32 = HasCRC32();
 	bool hasAES = HasAES();
 	bool hasSHA1 = HasSHA1();
 	bool hasSHA2 = HasSHA2();
 
 	cout << "passed:  ";
-	cout << "hasNEON == " << hasNEON << ", hasCRC32 == " << hasCRC32 << ", hasAES == " << hasAES << ", hasSHA1 == " << hasSHA1 << ", hasSHA2 == " << hasSHA2 << endl;
+	cout << "hasNEON == " << hasNEON << ", hasPMULL == " << hasPMULL << ", hasCRC32 == " << hasCRC32 << ", hasAES == " << hasAES << ", hasSHA1 == " << hasSHA1 << ", hasSHA2 == " << hasSHA2 << endl;
 #endif
 
 	if (!pass)