[assembler] Don't define the reserved name _xgetbv

_xgetbv is reserved for the implementation and shouldn't be used by user
code. Newer GCCs and clang trunk define _xgetbv, leading to a name
collision if xsaveintrin.h gets included transitively.

This unbreaks building v8 with clang trunk and libstdc++ 4.9, which
happens to pull in xsaveintrin.h transitively through <algorithm>. Newer
versions of libstdc++ don't seem to do that anymore which is why this
issue never showed up before.

R=bmeurer@chromium.org

Change-Id: If94efaf4798e5420738064bcbf26880f904c76a9
Reviewed-on: https://chromium-review.googlesource.com/c/1414858Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Commit-Queue: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/master@{#58849}

src/x64/assembler-x64.cc

@@ -31,9 +31,10 @@ namespace internal {
 
 namespace {
 
-#if !V8_LIBC_MSVCRT
-
-V8_INLINE uint64_t _xgetbv(unsigned int xcr) {
+V8_INLINE uint64_t xgetbv(unsigned int xcr) {
+#if V8_LIBC_MSVCRT
+  return _xgetbv(xcr);
+#else
   unsigned eax, edx;
   // Check xgetbv; this uses a .byte sequence instead of the instruction
   // directly because older assemblers do not include support for xgetbv and
@@ -41,13 +42,9 @@ V8_INLINE uint64_t _xgetbv(unsigned int xcr) {
   // used.
   __asm__ volatile(".byte 0x0F, 0x01, 0xD0" : "=a"(eax), "=d"(edx) : "c"(xcr));
   return static_cast<uint64_t>(eax) | (static_cast<uint64_t>(edx) << 32);
+#endif
 }
 
-#define _XCR_XFEATURE_ENABLED_MASK 0
-
-#endif  // !V8_LIBC_MSVCRT
-
-
 bool OSHasAVXSupport() {
 #if V8_OS_MACOSX
   // Mac OS X up to 10.9 has a bug where AVX transitions were indeed being
@@ -67,7 +64,7 @@ bool OSHasAVXSupport() {
   if (kernel_version_major <= 13) return false;
 #endif  // V8_OS_MACOSX
   // Check whether OS claims to support AVX.
-  uint64_t feature_mask = _xgetbv(_XCR_XFEATURE_ENABLED_MASK);
+  uint64_t feature_mask = xgetbv(0);  // XCR_XFEATURE_ENABLED_MASK
   return (feature_mask & 0x6) == 0x6;
 }