Handle ARM "ret void" and function alignment with proper padding.

src/IceCfg.cpp

 //===- subzero/src/IceCfg.cpp - Control flow graph implementation ---------===//
 //
 //                        The Subzero Code Generator
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements the Cfg class, including constant pool
@@ skipping 446 matching lines @@
 void Cfg::emitTextHeader(const IceString &MangledName, GlobalContext *Ctx,
                          const Assembler *Asm) {
   if (!ALLOW_DUMP)
     return;
   Ostream &Str = Ctx->getStrEmit();
   Str << "\t.text\n";
   if (Ctx->getFlags().getFunctionSections())
     Str << "\t.section\t.text." << MangledName << ",\"ax\",@progbits\n";
   if (!Asm->getInternal() || Ctx->getFlags().getDisableInternal()) {
     Str << "\t.globl\t" << MangledName << "\n";
-    Str << "\t.type\t" << MangledName << ",@function\n";
+    Str << "\t.type\t" << MangledName << ",%function\n";
   }
-  Str << "\t.p2align " << Asm->getBundleAlignLog2Bytes() << ",0x";
+  Str << "\t" << Asm->getNonExecPadDirective() << " "
+      << Asm->getBundleAlignLog2Bytes() << ",0x";
   for (uint8_t I : Asm->getNonExecBundlePadding())
     Str.write_hex(I);
   Str << "\n";
   Str << MangledName << ":\n";
 }
 
 void Cfg::emit() {
   if (!ALLOW_DUMP)
     return;
   TimerMarker T(TimerStack::TT_emit, this);
@@ skipping 61 matching lines @@
     }
   }
   // Print each basic block
   for (CfgNode *Node : Nodes)
     Node->dump(this);
   if (isVerbose(IceV_Instructions))
     Str << "}\n";
 }
 
 } // end of namespace Ice