Fix push/pop emit methods for ARM assembler.

src/IceInstARM32.cpp

 //===- subzero/src/IceInstARM32.cpp - ARM32 instruction implementation ----===//
 //
 //                        The Subzero Code Generator
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 ///
 /// \file
@@ skipping 1099 matching lines @@
         PrintComma = true;
       }
     }
     Str << "}";
     NeedNewline = true;
   }
 
   for (const Operand *Op : Dests) {
     if (isScalarIntegerType(Op->getType()))
       continue;
-    startNextInst(Func);
     if (NeedNewline) {
       Str << "\n";
+      startNextInst(Func);
       NeedNewline = false;
     }
     Str << "\t"
         << "vpop"
         << "\t{";
     Op->emit(Func);
     Str << "}";
     NeedNewline = true;
   }
   assert(NeedNewline); // caller will add the newline
@@ skipping 87 matching lines @@
     if (isScalarIntegerType(getSrc(i)->getType())) {
       ++IntegerCount;
     }
   }
   Ostream &Str = Func->getContext()->getStrEmit();
   bool NeedNewline = false;
   for (SizeT i = getSrcSize(); i > 0; --i) {
     Operand *Op = getSrc(i - 1);
     if (isScalarIntegerType(Op->getType()))
       continue;
+    if (NeedNewline) {
+      Str << "\n";
+      startNextInst(Func);
+      NeedNewline = false;
+    }
     Str << "\t"
         << "vpush"
         << "\t{";
     Op->emit(Func);
     Str << "}";
     NeedNewline = true;
   }
   if (IntegerCount != 0) {
-    startNextInst(Func);
     if (NeedNewline) {
       Str << "\n";
+      startNextInst(Func);
       NeedNewline = false;
     }
     Str << "\t"
         << "push"
         << "\t{";
     bool PrintComma = false;
     for (SizeT i = 0; i < getSrcSize(); ++i) {
       Operand *Op = getSrc(i);
       if (isScalarIntegerType(Op->getType())) {
         if (PrintComma)
@@ skipping 511 matching lines @@
 template class InstARM32UnaryopGPR<InstARM32::Uxt, true>;
 template class InstARM32UnaryopFP<InstARM32::Vsqrt>;
 
 template class InstARM32FourAddrGPR<InstARM32::Mla>;
 template class InstARM32FourAddrGPR<InstARM32::Mls>;
 
 template class InstARM32CmpLike<InstARM32::Cmp>;
 template class InstARM32CmpLike<InstARM32::Tst>;
 
 } // end of namespace Ice