Give backends an opportunity to finish layout post-relaxation

[colinl]

Summary

Overriding this function gives targets access to the layout post-relaxation and before writing fragments to output.

Hexagon is using this to more intelligently pad nops by using a larger encoding for the previous instruction if possible.

We aren't able to do this with the existing writeNopData since the previous instructions have already been encoded to the output stream.

I imagine other targets would be able to do similar optimizations and possibly more.

llvm/trunk/include/llvm/MC/MCAsmBackend.h

public:
   virtual unsigned getMinimumNopSize() const { return 1; }
 
   /// Write an (optimal) nop sequence of Count bytes to the given output. If the
   /// target cannot generate such a sequence, it should return an error.
   ///
   /// \return - True on success.
   virtual bool writeNopData(uint64_t Count, MCObjectWriter *OW) const = 0;
 
+  /// Give backend an opportunity to finish layout after relaxation
+  virtual void finishLayout(MCAssembler const &Asm,
+                            MCAsmLayout &Layout) const {}
+
   /// Handle any target-specific assembler flags. By default, do nothing.
   virtual void handleAssemblerFlag(MCAssemblerFlag Flag) {}
 
   /// \brief Generate the compact unwind encoding for the CFI instructions.
   virtual uint32_t
       generateCompactUnwindEncoding(ArrayRef<MCCFIInstruction>) const {
     return 0;
   }
 };
 
 } // End llvm namespace
 
 #endif

llvm/trunk/lib/MC/MCAssembler.cpp

bool MCAssembler::layoutOnce(MCAsmLayout &Layout) {
   return WasRelaxed;
 }
 
 void MCAssembler::finishLayout(MCAsmLayout &Layout) {
   // The layout is done. Mark every fragment as valid.
   for (unsigned int i = 0, n = Layout.getSectionOrder().size(); i != n; ++i) {
     Layout.getFragmentOffset(&*Layout.getSectionOrder()[i]->rbegin());
   }
+  getBackend().finishLayout(*this, Layout);
 }