Merge from mainline.

Tevert part of the x86 subtarget logic changes: when -march=x86-64
is given, override the subtarget settings and enable 64-bit support.
This restores the earlier behavior, and fixes regressions on
Non-64-bit-capable x86-32 hosts.

This isn't necessarily the best approach, but the most obvious
alternative is to require -mcpu=x86-64 or -mattr=+64bit to be used
with -march=x86-64 when the host doesn't have 64-bit support. This
makes things little more consistent, but it's less convenient, and
it has the practical drawback of requiring lots of test changes, so
I opted for the above approach for now.

llvm-svn: 63792

Author: tlattner

llvm/lib/Target/X86/X86Subtarget.cpp

@@ -327,15 +327,16 @@ X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
   } else {
     // Otherwise, use CPUID to auto-detect feature set.
     AutoDetectSubtargetFeatures();
-    // If requesting codegen for X86-64, make sure that 64-bit features
-    // are enabled.
-    if (Is64Bit)
-      HasX86_64 = true;
     // Make sure SSE2 is enabled; it is available on all X86-64 CPUs.
     if (Is64Bit && X86SSELevel < SSE2)
       X86SSELevel = SSE2;
   }
 
+  // If requesting codegen for X86-64, make sure that 64-bit features
+  // are enabled.
+  if (Is64Bit)
+    HasX86_64 = true;
+
   DOUT << "Subtarget features: SSELevel " << X86SSELevel
        << ", 3DNowLevel " << X863DNowLevel
        << ", 64bit " << HasX86_64 << "\n";