From: Erik de Castro Lopo <erikd@mega-nerd.com>
Date: Sat, 14 Sep 2013 23:46:17 +0000 (+1000)
Subject: Add CPU features (sse3, ssse3) detection code for x86-64.
X-Git-Tag: 1.3.1~237
X-Git-Url: http://git.xiph.org/?p=flac.git;a=commitdiff_plain;h=84c3e3d52c8d12ac694d70c14b69ee04d21d0aa9;ds=sidebyside

Add CPU features (sse3, ssse3) detection code for x86-64.

Patch-from: lvqcl <lvqcl.mail@gmail.com>
---

diff --git a/src/libFLAC/cpu.c b/src/libFLAC/cpu.c
index b7a7a3a2..5a453488 100644
--- a/src/libFLAC/cpu.c
+++ b/src/libFLAC/cpu.c
@@ -175,7 +175,7 @@ void FLAC__cpu_info(FLAC__CPUInfo *info)
 	info->data.ia32.ext3dnow = false;
 	info->data.ia32.extmmx = false;
 	if(info->data.ia32.cpuid) {
-		/* http://www.sandpile.org/ia32/cpuid.htm */
+		/* http://www.sandpile.org/x86/cpuid.htm */
 		FLAC__uint32 flags_edx, flags_ecx;
 		FLAC__cpu_info_asm_ia32(&flags_edx, &flags_ecx);
 		info->data.ia32.cmov  = (flags_edx & FLAC__CPUINFO_IA32_CPUID_CMOV )? true : false;
@@ -283,12 +283,12 @@ void FLAC__cpu_info(FLAC__CPUInfo *info)
 				info->data.ia32.fxsr = info->data.ia32.sse = info->data.ia32.sse2 = info->data.ia32.sse3 = info->data.ia32.ssse3 = false;
 #elif defined(_MSC_VER)
 # ifdef USE_TRY_CATCH_FLAVOR
-			_try {
+			__try {
 				__asm {
 					xorps xmm0,xmm0
 				}
 			}
-			_except(EXCEPTION_EXECUTE_HANDLER) {
+			__except(EXCEPTION_EXECUTE_HANDLER) {
 				if (_exception_code() == STATUS_ILLEGAL_INSTRUCTION)
 					info->data.ia32.fxsr = info->data.ia32.sse = info->data.ia32.sse2 = info->data.ia32.sse3 = info->data.ia32.ssse3 = false;
 			}
@@ -331,6 +331,32 @@ void FLAC__cpu_info(FLAC__CPUInfo *info)
 #endif
 
 /*
+ * x86-64-specific
+ */
+#elif defined FLAC__CPU_X86_64
+	info->type = FLAC__CPUINFO_TYPE_X86_64;
+#if !defined FLAC__NO_ASM && defined FLAC__HAS_X86INTRIN
+	info->use_asm = true;
+	info->data.x86_64.sse3 = false;
+	info->data.x86_64.ssse3 = false;
+	{
+		/* http://www.sandpile.org/x86/cpuid.htm */
+		FLAC__uint32 flags_edx, flags_ecx;
+		FLAC__cpu_info_x86(&flags_edx, &flags_ecx);
+		info->data.x86_64.sse3  = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSE3 )? true : false;
+		info->data.x86_64.ssse3 = (flags_ecx & FLAC__CPUINFO_IA32_CPUID_SSSE3)? true : false;
+	}
+#ifdef DEBUG
+	fprintf(stderr, "CPU info (x86-64):\n");
+	fprintf(stderr, "  SSE3 ....... %c\n", info->data.x86_64.sse3    ? 'Y' : 'n');
+	fprintf(stderr, "  SSSE3 ...... %c\n", info->data.x86_64.ssse3   ? 'Y' : 'n');
+#endif
+
+#else
+	info->use_asm = false;
+#endif
+
+/*
  * PPC-specific
  */
 #elif defined FLAC__CPU_PPC
@@ -396,10 +422,40 @@ void FLAC__cpu_info(FLAC__CPUInfo *info)
 # endif
 
 /*
- * unknown CPI
+ * unknown CPU
  */
 #else
 	info->type = FLAC__CPUINFO_TYPE_UNKNOWN;
 	info->use_asm = false;
 #endif
 }
+
+#if defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64
+#ifdef FLAC__HAS_X86INTRIN
+
+#if defined _MSC_VER && (_MSC_VER >= 1400)
+#include <intrin.h> /* for __cpuid() */
+#endif
+
+void FLAC__cpu_info_x86(FLAC__uint32 *flags_edx, FLAC__uint32 *flags_ecx)
+{
+#if defined _MSC_VER && (_MSC_VER >= 1400)
+		int cpuinfo[4];
+		__cpuid(cpuinfo, 1);
+		*flags_ecx = cpuinfo[2];
+		*flags_edx = cpuinfo[3];
+#elif defined __GNUC__ && __GNUC__
+		FLAC__uint32 info = 1, flags_eax, flags_ebx;
+		__asm__ __volatile__ (
+			"xchg %%ebx, %%edi;"
+			"cpuid;"
+			"xchg %%edi, %%ebx;"
+			:"=a" (flags_eax), "=D" (flags_ebx), "=c" (*flags_ecx), "=d" (*flags_edx)
+			:"a" (info)
+		);
+#else
+		*flags_ecx = *flags_edx = 0;
+#endif
+}
+#endif /* FLAC__HAS_X86INTRIN */
+#endif /* FLAC__CPU_IA32 || FLAC__CPU_X86_64 */
diff --git a/src/libFLAC/include/private/cpu.h b/src/libFLAC/include/private/cpu.h
index d920a278..90bf9469 100644
--- a/src/libFLAC/include/private/cpu.h
+++ b/src/libFLAC/include/private/cpu.h
@@ -41,6 +41,7 @@
 
 typedef enum {
 	FLAC__CPUINFO_TYPE_IA32,
+	FLAC__CPUINFO_TYPE_X86_64,
 	FLAC__CPUINFO_TYPE_PPC,
 	FLAC__CPUINFO_TYPE_UNKNOWN
 } FLAC__CPUInfo_Type;
@@ -61,6 +62,11 @@ typedef struct {
 } FLAC__CPUInfo_IA32;
 
 typedef struct {
+	FLAC__bool sse3;
+	FLAC__bool ssse3;
+} FLAC__CPUInfo_x86_64;
+
+typedef struct {
 	FLAC__bool altivec;
 	FLAC__bool ppc64;
 } FLAC__CPUInfo_PPC;
@@ -70,6 +76,7 @@ typedef struct {
 	FLAC__CPUInfo_Type type;
 	union {
 		FLAC__CPUInfo_IA32 ia32;
+		FLAC__CPUInfo_x86_64 x86_64;
 		FLAC__CPUInfo_PPC ppc;
 	} data;
 } FLAC__CPUInfo;
@@ -77,6 +84,7 @@ typedef struct {
 void FLAC__cpu_info(FLAC__CPUInfo *info);
 
 #ifndef FLAC__NO_ASM
+
 #ifdef FLAC__CPU_IA32
 #ifdef FLAC__HAS_NASM
 FLAC__uint32 FLAC__cpu_have_cpuid_asm_ia32(void);
@@ -84,6 +92,13 @@ void         FLAC__cpu_info_asm_ia32(FLAC__uint32 *flags_edx, FLAC__uint32 *flag
 FLAC__uint32 FLAC__cpu_info_extended_amd_asm_ia32(void);
 #endif
 #endif
+
+#if defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64
+#ifdef FLAC__HAS_X86INTRIN
+void FLAC__cpu_info_x86(FLAC__uint32 *flags_edx, FLAC__uint32 *flags_ecx);
+#endif
+#endif
+
 #endif
 
 #endif