Cryptography: Optimize Whirlpool implementation by using public domain assembly code developed by Wei Dai

2025-11-11 11:08:02 -06:00 · 2015-12-30 06:57:34 +01:00
parent d462806ea9
commit 0c2c0a846d
17 changed files with 1547 additions and 1170 deletions
--- a/src/Crypto/cpu.c
+++ b/src/Crypto/cpu.c
@@ -0,0 +1,198 @@
+/* cpu.c - written and placed in the public domain by Wei Dai */
+
+#include "cpu.h"
+#include "misc.h"
+
+#ifndef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
+#include <signal.h>
+#include <setjmp.h>
+#endif
+
+#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+#include <emmintrin.h>
+#endif
+
+#ifdef CRYPTOPP_CPUID_AVAILABLE
+
+#if _MSC_VER >= 1400 && CRYPTOPP_BOOL_X64
+
+int CpuId(uint32 input, uint32 *output)
+{
+	__cpuid((int *)output, input);
+	return 1;
+}
+
+#else
+
+#ifndef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+typedef void (*SigHandler)(int);
+
+static jmp_buf s_jmpNoCPUID;
+static void SigIllHandlerCPUID(int p)
+{
+	longjmp(s_jmpNoCPUID, 1);
+}
+
+#if CRYPTOPP_BOOL_X64 == 0
+static jmp_buf s_jmpNoSSE2;
+static void SigIllHandlerSSE2(int p)
+{
+	longjmp(s_jmpNoSSE2, 1);
+}
+#endif
+#if defined(__cplusplus)
+}
+#endif
+#endif
+
+int CpuId(uint32 input, uint32 *output)
+{
+#ifdef CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY
+    __try
+	{
+		__asm
+		{
+			mov eax, input
+			cpuid
+			mov edi, output
+			mov [edi], eax
+			mov [edi+4], ebx
+			mov [edi+8], ecx
+			mov [edi+12], edx
+		}
+	}
+    __except (1)
+	{
+		return 0;
+    }
+	return 1;
+#else
+	SigHandler oldHandler = signal(SIGILL, SigIllHandlerCPUID);
+	if (oldHandler == SIG_ERR)
+		return 0;
+
+	int result = 1;
+	if (setjmp(s_jmpNoCPUID))
+		result = 0;
+	else
+	{
+		asm
+		(
+			// save ebx in case -fPIC is being used
+#if CRYPTOPP_BOOL_X86
+			"push %%ebx; cpuid; mov %%ebx, %%edi; pop %%ebx"
+#else
+			"pushq %%rbx; cpuid; mov %%ebx, %%edi; popq %%rbx"
+#endif
+			: "=a" (output[0]), "=D" (output[1]), "=c" (output[2]), "=d" (output[3])
+			: "a" (input)
+		);
+	}
+
+	signal(SIGILL, oldHandler);
+	return result;
+#endif
+}
+
+#endif
+
+static int TrySSE2()
+{
+#if CRYPTOPP_BOOL_X64
+	return 1;
+#elif defined(CRYPTOPP_MS_STYLE_INLINE_ASSEMBLY)
+    __try
+	{
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+        AS2(por xmm0, xmm0)        // executing SSE2 instruction
+#elif CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+		__m128i x = _mm_setzero_si128();
+		return _mm_cvtsi128_si32(x) == 0;
+#endif
+	}
+    __except (1)
+	{
+		return 0;
+    }
+	return 1;
+#else
+	SigHandler oldHandler = signal(SIGILL, SigIllHandlerSSE2);
+	if (oldHandler == SIG_ERR)
+		return 0;
+
+	int result = 1;
+	if (setjmp(s_jmpNoSSE2))
+		result = 0;
+	else
+	{
+#if CRYPTOPP_BOOL_SSE2_ASM_AVAILABLE
+		__asm __volatile ("por %xmm0, %xmm0");
+#elif CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE
+		__m128i x = _mm_setzero_si128();
+		result = _mm_cvtsi128_si32(x) == 0;
+#endif
+	}
+
+	signal(SIGILL, oldHandler);
+	return result;
+#endif
+}
+
+int g_x86DetectionDone = 0;
+int g_hasISSE = 0, g_hasSSE2 = 0, g_hasSSSE3 = 0, g_hasMMX = 0, g_hasAESNI = 0, g_hasCLMUL = 0, g_isP4 = 0;
+uint32 g_cacheLineSize = CRYPTOPP_L1_CACHE_LINE_SIZE;
+
+void DetectX86Features()
+{
+	uint32 cpuid[4], cpuid1[4], tmp;
+	if (!CpuId(0, cpuid))
+		return;
+	if (!CpuId(1, cpuid1))
+		return;
+
+	g_hasMMX = (cpuid1[3] & (1 << 23)) != 0;
+	if ((cpuid1[3] & (1 << 26)) != 0)
+		g_hasSSE2 = TrySSE2();
+	g_hasSSSE3 = g_hasSSE2 && (cpuid1[2] & (1<<9));
+	g_hasAESNI = g_hasSSE2 && (cpuid1[2] & (1<<25));
+	g_hasCLMUL = g_hasSSE2 && (cpuid1[2] & (1<<1));
+
+	if ((cpuid1[3] & (1 << 25)) != 0)
+		g_hasISSE = 1;
+	else
+	{
+		uint32 cpuid2[4];
+		CpuId(0x080000000, cpuid2);
+		if (cpuid2[0] >= 0x080000001)
+		{
+			CpuId(0x080000001, cpuid2);
+			g_hasISSE = (cpuid2[3] & (1 << 22)) != 0;
+		}
+	}
+
+	tmp = cpuid[2];
+	cpuid[2] = cpuid[3];
+	cpuid[3] = tmp;
+	if (memcmp(cpuid+1, "GenuineIntel", 12) == 0)
+	{
+		g_isP4 = ((cpuid1[0] >> 8) & 0xf) == 0xf;
+		g_cacheLineSize = 8 * GETBYTE(cpuid1[1], 1);
+	}
+	else if (memcmp(cpuid+1, "AuthenticAMD", 12) == 0)
+	{
+		CpuId(0x80000005, cpuid);
+		g_cacheLineSize = GETBYTE(cpuid[2], 0);
+	}
+
+	if (!g_cacheLineSize)
+		g_cacheLineSize = CRYPTOPP_L1_CACHE_LINE_SIZE;
+
+	g_x86DetectionDone = 1;
+}
+
+#endif