Move function nan_value to POSIX platform file

Strictly speaking NAN from math.h is not in a POSIX standard but in C99. However it seems that all our POSIX platforms have it.
Review URL: http://codereview.chromium.org/464015

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3405 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/platform-freebsd.cc

@@ -89,11 +89,6 @@ uint64_t OS::CpuFeaturesImpliedByPlatform() {
 }
 
 
-double OS::nan_value() {
-  return NAN;
-}
-
-
 int OS::ActivationFrameAlignment() {
   // 16 byte alignment on FreeBSD
   return 16;

src/platform-linux.cc

@@ -95,11 +95,6 @@ uint64_t OS::CpuFeaturesImpliedByPlatform() {
 }
 
 
-double OS::nan_value() {
-  return NAN;
-}
-
-
 #ifdef __arm__
 bool OS::ArmCpuHasFeature(CpuFeature feature) {
   const char* search_string = NULL;

src/platform-macos.cc

@@ -252,11 +252,6 @@ uint64_t OS::CpuFeaturesImpliedByPlatform() {
 }
 
 
-double OS::nan_value() {
-  return NAN;
-}
-
-
 int OS::ActivationFrameAlignment() {
   // OS X activation frames must be 16 byte-aligned; see "Mac OS X ABI
   // Function Call Guide".

src/platform-openbsd.cc

@@ -88,11 +88,6 @@ uint64_t OS::CpuFeaturesImpliedByPlatform() {
 }
 
 
-double OS::nan_value() {
-  return NAN;
-}
-
-
 int OS::ActivationFrameAlignment() {
   // 16 byte alignment on OpenBSD
   return 16;

src/platform-posix.cc

@@ -61,6 +61,13 @@ double modulo(double x, double y) {
   return fmod(x, y);
 }
 
+
+double OS::nan_value() {
+  // NAN from math.h is defined in C99 and not in POSIX.
+  return NAN;
+}
+
+
 // ----------------------------------------------------------------------------
 // POSIX date/time support.
 //