[llvm-commits] [llvm] r95282 - /llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
Evan Cheng
evan.cheng at apple.com
Wed Feb 3 18:45:02 PST 2010
Author: evancheng
Date: Wed Feb 3 20:45:02 2010
New Revision: 95282
URL: http://llvm.org/viewvc/llvm-project?rev=95282&view=rev
Log:
It's too risky to eliminate sext / zext of call results for tail call optimization even if the caller / callee attributes completely match. The callee may have been bitcast'ed (or otherwise lied about what it's doing).
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp?rev=95282&r1=95281&r2=95282&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp Wed Feb 3 20:45:02 2010
@@ -4213,12 +4213,6 @@
// causing miscompilation that has not been fully understood.
if (!Ret) return false;
- // Unless we are explicitly forcing tailcall optimization do not tailcall if
- // the called function is bitcast'ed. The analysis may not be entirely
- // accurate.
- if (!PerformTailCallOpt && isa<BitCastInst>(CS.getCalledValue()))
- return false;
-
// If I will have a chain, make sure no other instruction that will have a
// chain interposes between I and the return.
if (I->mayHaveSideEffects() || I->mayReadFromMemory() ||
@@ -4246,6 +4240,10 @@
if ((CalleeRetAttr ^ CallerRetAttr) & ~Attribute::NoAlias)
return false;
+ // It's not safe to eliminate thee sign / zero extension of the return value.
+ if ((CallerRetAttr & Attribute::ZExt) || (CallerRetAttr & Attribute::SExt))
+ return false;
+
// Otherwise, make sure the unmodified return value of I is the return value.
for (const Instruction *U = dyn_cast<Instruction>(Ret->getOperand(0)); ;
U = dyn_cast<Instruction>(U->getOperand(0))) {
More information about the llvm-commits
mailing list