[llvm-commits] CVS: reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp

Brian Gaeke gaeke at cs.uiuc.edu
Sun Jul 11 19:34:13 PDT 2004


Changes in directory reopt/lib/LightWtProfiling:

UnpackTraceFunction.cpp updated: 1.95 -> 1.96

---
Log message:

Use a throwaway MCFI for CreateCodeToLoadConst(), because you can no
longer pass in an Instruction* to MachineCodeForInstruction::get()
which is not attached to a Function*. Since this seems to work, it's
likely that CreateCodeToLoadConst() doesn't need to have an MCFI passed
in at all.  We'll have to look into this.

Fix the setting of fpIsTraceFP in rewriteEpilog().


---
Diffs of the changes:  (+4 -2)

Index: reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp
diff -u reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp:1.95 reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp:1.96
--- reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp:1.95	Sun Jul 11 05:02:55 2004
+++ reopt/lib/LightWtProfiling/UnpackTraceFunction.cpp	Sun Jul 11 19:33:26 2004
@@ -313,8 +313,8 @@
                                                   std::vector<MachineInstr *> &mvec) {
   const TargetInstrInfo &TII = *TM->getInstrInfo ();
   TmpInstruction *tmp = new TmpInstruction (C);
-  TII.CreateCodeToLoadConst (*TM, const_cast<Function *> (MF.getFunction ()), C, tmp, mvec,
-                             MachineCodeForInstruction::get (tmp));
+  MachineCodeForInstruction throwaway;
+  TII.CreateCodeToLoadConst (*TM, const_cast<Function *> (MF.getFunction ()), C, tmp, mvec, throwaway);
                              
   DEBUG (for (std::vector<MachineInstr *>::iterator i = mvec.begin (), e = mvec.end ();
               i != e; ++i)
@@ -463,6 +463,8 @@
   // (FIXME: may not work once we start doing optimizations!!! We will probably
   // have to use a separate MBB)
   MBB.clear ();
+  // This is just an educated guess...
+  fpIsTraceFP = true;
   
   // Insert stores from each live-out variable's reg. in the trace
   // to its stack slot in the trace function, from which it will be





More information about the llvm-commits mailing list