[X86][AMX] Reduce the compiling time for non-amx code.

[LuoYuanke]

@nikic , do you know how to check the compiling time locally?

[nikic]

Compile-time looks good: http://llvm-compile-time-tracker.com/compare.php?from=a7c079aaa227d55ad9cb6b916500d99b4fdf26d2&to=3b586204feec6114692ead49d107e3b9d7c2f2c4&stat=instructions I think it mostly recovers the regression.

[LuoYuanke]

In D126280#3533729, @nikic wrote:

Compile-time looks good: http://llvm-compile-time-tracker.com/compare.php?from=a7c079aaa227d55ad9cb6b916500d99b4fdf26d2&to=3b586204feec6114692ead49d107e3b9d7c2f2c4&stat=instructions I think it mostly recovers the regression.

Thanks, I'll land this patch.

llvm/lib/Target/X86/X86FastPreTileConfig.cpp

bool X86FastPreTileConfig::runOnMachineFunction(MachineFunction &MFunc) {
   ST = &MFunc.getSubtarget<X86Subtarget>();
   TII = ST->getInstrInfo();
   X86FI = MFunc.getInfo<X86MachineFunctionInfo>();
   MFI = &MFunc.getFrameInfo();
   TRI = ST->getRegisterInfo();
   CfgSS = -1;
 
   unsigned NumVirtRegs = MRI->getNumVirtRegs();
+  // Abandon early if there is no tile register to config.
+  bool HasVirtTileReg = false;
+  for (unsigned I = 0, E = NumVirtRegs; I != E; ++I) {
+    Register VirtReg = Register::index2VirtReg(I);
+    if (MRI->getRegClass(VirtReg)->getID() == X86::TILERegClassID) {
+      HasVirtTileReg = true;
+      break;
+    }
+  }
+  if (!HasVirtTileReg)
+    return false;
+
   StackSlotForVirtReg.resize(NumVirtRegs);
   MayLiveAcrossBlocks.clear();
   // We will create register during config. *3 is to make sure
   // the virtual register number doesn't exceed the size of
   // the bit vector.
   MayLiveAcrossBlocks.resize(NumVirtRegs * 3);
   bool Change = false;
   assert(MRI->isSSA());