MaskCpuFlags return cpuinfo so InitCpuFlags can call it

Reduce number of atomic references to cpu_info by making
InitCpuFlags call MaskCpuFlags and return the same value.

BUG=libyuv:641
TEST=libyuv_unittests pass

Change-Id: I5dfff8f7a10671bc8ef3ec0ed6f302791e752faa
Reviewed-on: https://chromium-review.googlesource.com/514145
Commit-Queue: Frank Barchard <fbarchard@google.com>
Reviewed-by: Cheng Wang <wangcheng@google.com>

include/libyuv/cpu_id.h

@@ -46,17 +46,14 @@ static const int kCpuHasMIPS = 0x10000;
 static const int kCpuHasDSPR2 = 0x20000;
 static const int kCpuHasMSA = 0x40000;
 
-// Internal function used to auto-init.
+// Optional init function. TestCpuFlag does an auto-init.
+// Returns cpu_info flags.
 LIBYUV_API
 int InitCpuFlags(void);
 
-// Internal function for parsing /proc/cpuinfo.
-LIBYUV_API
-int ArmCpuCaps(const char* cpuinfo_name);
-
 // Detect CPU has SSE2 etc.
 // Test_flag parameter should be one of kCpuHas constants above.
-// returns non-zero if instruction set is detected
+// Returns non-zero if instruction set is detected
 static __inline int TestCpuFlag(int test_flag) {
   LIBYUV_API extern int cpu_info_;
 #ifdef __ATOMIC_RELAXED
@@ -67,12 +64,18 @@ static __inline int TestCpuFlag(int test_flag) {
   return (!cpu_info ? InitCpuFlags() : cpu_info) & test_flag;
 }
 
+// Internal function for parsing /proc/cpuinfo.
+LIBYUV_API
+int ArmCpuCaps(const char* cpuinfo_name);
+
 // For testing, allow CPU flags to be disabled.
 // ie MaskCpuFlags(~kCpuHasSSSE3) to disable SSSE3.
 // MaskCpuFlags(-1) to enable all cpu specific optimizations.
 // MaskCpuFlags(1) to disable all cpu specific optimizations.
+// MaskCpuFlags(0) to reset state so next call will auto init.
+// Returns cpu_info flags.
 LIBYUV_API
-void MaskCpuFlags(int enable_flags);
+int MaskCpuFlags(int enable_flags);
 
 // Low level cpuid for X86. Returns zeros on other CPUs.
 // eax is the info type that you want.

source/cpu_id.cc

@@ -46,6 +46,7 @@ extern "C" {
 // cpu_info_ variable for SIMD instruction sets detected.
 LIBYUV_API int cpu_info_ = 0;
 
+// TODO(fbarchard): Consider using int for cpuid so casting is not needed.
 // Low level cpuid for X86.
 #if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
      defined(__x86_64__)) &&                                     \
@@ -55,7 +56,7 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
 #if defined(_MSC_VER)
 // Visual C version uses intrinsic or inline x86 assembly.
 #if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
-  __cpuidex((int*)(cpu_info), info_eax, info_ecx);
+  __cpuidex((int*)(cpu_info), info_eax, info_ecx);  // NOLINT
 #elif defined(_M_IX86)
   __asm {
     mov        eax, info_eax
@@ -69,7 +70,7 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
   }
 #else  // Visual C but not x86
   if (info_ecx == 0) {
-    __cpuid((int*)(cpu_info), info_eax);
+    __cpuid((int*)(cpu_info), info_eax);  // NOLINT
   } else {
     cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0u;
   }
@@ -122,8 +123,9 @@ void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
 // X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
 int GetXCR0() {
   uint32 xcr0 = 0u;
+// VS2010 SP1 required
 #if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
-  xcr0 = (uint32)(_xgetbv(0));  // VS2010 SP1 required.
+  xcr0 = (uint32)(_xgetbv(0));  // NOLINT
 #elif defined(__i386__) || defined(__x86_64__)
   asm(".byte 0x0f, 0x01, 0xd0" : "=a"(xcr0) : "c"(0) : "%edx");
 #endif  // defined(__i386__) || defined(__x86_64__)
@@ -170,7 +172,7 @@ LIBYUV_API SAFEBUFFERS int ArmCpuCaps(const char* cpuinfo_name) {
 LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name,
                                        const char ase[]) {
   char cpuinfo_line[512];
-  int len = (int)strlen(ase);
+  int len = (int)strlen(ase);  // NOLINT
   FILE* f = fopen(cpuinfo_name, "r");
   if (!f) {
     // ase enabled if /proc/cpuinfo is unavailable.
@@ -325,22 +327,19 @@ static SAFEBUFFERS int GetCpuFlags(void) {
 
 // Note that use of this function is not thread safe.
 LIBYUV_API
-void MaskCpuFlags(int enable_flags) {
+int MaskCpuFlags(int enable_flags) {
+  int cpu_info = GetCpuFlags() & enable_flags;
 #ifdef __ATOMIC_RELAXED
-  __atomic_store_n(&cpu_info_, GetCpuFlags() & enable_flags, __ATOMIC_RELAXED);
+  __atomic_store_n(&cpu_info_, cpu_info, __ATOMIC_RELAXED);
 #else
-  cpu_info_ = GetCpuFlags() & enable_flags;
+  cpu_info_ = cpu_info;
 #endif
+  return cpu_info;
 }
 
 LIBYUV_API
 int InitCpuFlags(void) {
-  MaskCpuFlags(-1);
-#ifdef __ATOMIC_RELAXED
-  return __atomic_load_n(&cpu_info_, __ATOMIC_RELAXED);
-#else
-  return cpu_info_;
-#endif
+  return MaskCpuFlags(-1);
 }
 
 #ifdef __cplusplus

unit_test/cpu_thread_test.cc

@@ -12,8 +12,11 @@
 
 #include "libyuv/cpu_id.h"
 
-// TODO(fbarchard): Port to other platforms
-#if defined(__linux__)
+#if defined(__clang__)
+#if __has_include(<pthread.h>)
+#define LIBYUV_HAVE_PTHREAD 1
+#endif
+#elif defined(__linux__)
 #define LIBYUV_HAVE_PTHREAD 1
 #endif