Remove SSE2 from cpu.cpp, add sse-simd.cpp

We need to ensure SSE2 does not cross pollinate into other CPU functions since SSE2 is greater than the minimum arch. The minimum arch is i586/i686, and both lack SSE2 instructions

Filelist.txt

@@ -283,6 +283,7 @@ sosemanuk.h
 square.cpp
 square.h
 squaretb.cpp
+sse-simd.cpp
 stdcpp.h
 strciphr.cpp
 strciphr.h

GNUmakefile

@@ -226,9 +226,10 @@ endif  # -DCRYPTOPP_DISABLE_SSSE3
 endif  # -DCRYPTOPP_DISABLE_ASM
 endif  # CXXFLAGS
 
+# SSE2 is a core feature of x86_64
 ifeq ($(findstring -DCRYPTOPP_DISABLE_ASM,$(CXXFLAGS)),)
   ifeq ($(IS_X86),1)
-    CPU_FLAG = -msse2
+    SSE_FLAG = -msse2
   endif
 endif
 ifeq ($(findstring -DCRYPTOPP_DISABLE_SSSE3,$(CXXFLAGS)),)
@@ -996,13 +997,13 @@ endif
 aria-simd.o : aria-simd.cpp
 	$(CXX) $(strip $(CXXFLAGS) $(ARIA_FLAG) -c) $<
 
-# SSE4.2 or ARMv8a available
+# SSE4.1 or ARMv8a available
 blake2-simd.o : blake2-simd.cpp
 	$(CXX) $(strip $(CXXFLAGS) $(BLAKE2_FLAG) -c) $<
 
 # SSE2 on i586
-cpu.o : cpu.cpp
+sse-simd.o : sse-simd.cpp
-	$(CXX) $(strip $(CXXFLAGS) $(CPU_FLAG) -c) $<
+	$(CXX) $(strip $(CXXFLAGS) $(SSE_FLAG) -c) $<
 
 # SSE4.2 or ARMv8a available
 crc-simd.o : crc-simd.cpp

cpu.cpp

@@ -48,11 +48,6 @@ unsigned long int getauxval(unsigned long int) { return 0; }
 # include <setjmp.h>
 #endif
 
-// Needed by SunCC and MSVC
-#if (CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X32 || CRYPTOPP_BOOL_X64)
-# include <emmintrin.h>
-#endif
-
 NAMESPACE_BEGIN(CryptoPP)
 
 #ifndef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
@@ -63,7 +58,9 @@ extern "C" {
 
 // *************************** IA-32 CPUs ***************************
 
-#ifdef CRYPTOPP_CPUID_AVAILABLE
+#if (CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X32 || CRYPTOPP_BOOL_X64)
+
+extern bool CPU_ProbeSSE2();
 
 #if _MSC_VER >= 1500
 
@@ -94,12 +91,6 @@ extern "C"
 	{
 		longjmp(s_jmpNoCPUID, 1);
 	}
-
-	static jmp_buf s_jmpNoSSE2;
-	static void SigIllHandlerSSE2(int)
-	{
-		longjmp(s_jmpNoSSE2, 1);
-	}
 }
 #endif
 
@@ -186,64 +177,6 @@ bool CpuId(word32 func, word32 subfunc, word32 output[4])
 
 #endif
 
-static bool CPU_ProbeSSE2()
-{
-#if CRYPTOPP_BOOL_X64
-	return true;
-#elif defined(CRYPTOPP_NO_CPU_FEATURE_PROBES)
-	return false;
-#elif defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
-    __try
-	{
-# if CRYPTOPP_SSE2_ASM_AVAILABLE
-		AS2(por xmm0, xmm0)        // executing SSE2 instruction
-# elif CRYPTOPP_SSE2_INTRIN_AVAILABLE
-		__m128i x = _mm_setzero_si128();
-		return _mm_cvtsi128_si32(x) == 0;
-# endif
-	}
-	// GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION
-	__except (EXCEPTION_EXECUTE_HANDLER)
-	{
-		return false;
-	}
-	return true;
-#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, SigIllHandlerSSE2);
-	if (oldHandler == SIG_ERR)
-		return false;
-
-# ifndef __MINGW32__
-	volatile sigset_t oldMask;
-	if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
-		return false;
-# endif
-
-	if (setjmp(s_jmpNoSSE2))
-		result = false;
-	else
-	{
-# if CRYPTOPP_SSE2_ASM_AVAILABLE
-		__asm __volatile ("por %xmm0, %xmm0");
-# elif CRYPTOPP_SSE2_INTRIN_AVAILABLE
-		__m128i x = _mm_setzero_si128();
-		result = _mm_cvtsi128_si32(x) == 0;
-# endif
-	}
-
-# ifndef __MINGW32__
-	sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
-# endif
-
-	signal(SIGILL, oldHandler);
-	return result;
-#endif
-}
-
 bool CRYPTOPP_SECTION_INIT g_x86DetectionDone = false;
 bool CRYPTOPP_SECTION_INIT CRYPTOPP_SECTION_INIT g_hasSSE2 = false, CRYPTOPP_SECTION_INIT g_hasSSSE3 = false;
 bool CRYPTOPP_SECTION_INIT g_hasSSE41 = false, CRYPTOPP_SECTION_INIT g_hasSSE42 = false;

cryptdll.vcxproj

@@ -237,6 +237,7 @@
     <ClCompile Include="sha-simd.cpp" />
     <ClCompile Include="simple.cpp" />
     <ClCompile Include="skipjack.cpp" />
+    <ClCompile Include="sse-simd.cpp" />
     <ClCompile Include="strciphr.cpp" />
     <ClCompile Include="trdlocal.cpp" />
   </ItemGroup>

cryptlib.vcxproj

@@ -294,6 +294,7 @@
     <ClCompile Include="sosemanuk.cpp" />
     <ClCompile Include="square.cpp" />
     <ClCompile Include="squaretb.cpp" />
+    <ClCompile Include="sse-simd.cpp" />
     <ClCompile Include="strciphr.cpp" />
     <ClCompile Include="tea.cpp" />
     <ClCompile Include="tftables.cpp" />

ppc-simd.cpp

@@ -1,4 +1,4 @@
-// crc-simd.cpp - written and placed in the public domain by
+// ppc-simd.cpp - written and placed in the public domain by
 //                Jeffrey Walton, Uri Blumenthal and Marcel Raad.
 //
 //    This source file uses intrinsics to gain access to AltiVec,

sse-simd.cpp

@@ -0,0 +1,101 @@
+// sse-simd.cpp - written and placed in the public domain by
+//                Jeffrey Walton, Uri Blumenthal and Marcel Raad.
+//
+//    This source file uses intrinsics to gain access to SSE for CPU
+//    feature testing. A separate source file is needed because additional
+//    CXXFLAGS are required to enable the appropriate instructions set in
+//    some build configurations.
+
+#include "pch.h"
+#include "config.h"
+
+#ifdef CRYPTOPP_GNU_STYLE_INLINE_ASSEMBLY
+# include <signal.h>
+# include <setjmp.h>
+#endif
+
+#ifndef EXCEPTION_EXECUTE_HANDLER
+# define EXCEPTION_EXECUTE_HANDLER 1
+#endif
+
+// Needed by SunCC and MSVC
+#if (CRYPTOPP_BOOL_X86 || CRYPTOPP_BOOL_X32 || CRYPTOPP_BOOL_X64)
+# include <emmintrin.h>
+#endif
+
+NAMESPACE_BEGIN(CryptoPP)
+
+#ifndef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
+extern "C" {
+    typedef void (*SigHandler)(int);
+};
+
+extern "C"
+{
+	static jmp_buf s_jmpNoSSE2;
+	static void SigIllHandlerSSE2(int)
+	{
+		longjmp(s_jmpNoSSE2, 1);
+	}
+}
+#endif  // Not CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
+
+bool CPU_ProbeSSE2()
+{
+#if CRYPTOPP_BOOL_X64
+	return true;
+#elif defined(CRYPTOPP_NO_CPU_FEATURE_PROBES)
+	return false;
+#elif defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
+    __try
+	{
+# if CRYPTOPP_SSE2_ASM_AVAILABLE
+		AS2(por xmm0, xmm0)        // executing SSE2 instruction
+# elif CRYPTOPP_SSE2_INTRIN_AVAILABLE
+		__m128i x = _mm_setzero_si128();
+		return _mm_cvtsi128_si32(x) == 0;
+# endif
+	}
+	// GetExceptionCode() == EXCEPTION_ILLEGAL_INSTRUCTION
+	__except (EXCEPTION_EXECUTE_HANDLER)
+	{
+		return false;
+	}
+	return true;
+#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, SigIllHandlerSSE2);
+	if (oldHandler == SIG_ERR)
+		return false;
+
+# ifndef __MINGW32__
+	volatile sigset_t oldMask;
+	if (sigprocmask(0, NULLPTR, (sigset_t*)&oldMask))
+		return false;
+# endif
+
+	if (setjmp(s_jmpNoSSE2))
+		result = false;
+	else
+	{
+# if CRYPTOPP_SSE2_ASM_AVAILABLE
+		__asm __volatile ("por %xmm0, %xmm0");
+# elif CRYPTOPP_SSE2_INTRIN_AVAILABLE
+		__m128i x = _mm_setzero_si128();
+		result = _mm_cvtsi128_si32(x) == 0;
+# endif
+	}
+
+# ifndef __MINGW32__
+	sigprocmask(SIG_SETMASK, (sigset_t*)&oldMask, NULLPTR);
+# endif
+
+	signal(SIGILL, oldHandler);
+	return result;
+#endif
+}
+
+NAMESPACE_END