[llvm-commits] CVS: llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
Chris Lattner
lattner at cs.uiuc.edu
Thu Jun 24 01:50:01 PDT 2004
Changes in directory llvm/lib/Transforms/Scalar:
IndVarSimplify.cpp updated: 1.66 -> 1.67
---
Log message:
Two fixes. First, stop using the ugly shouldSubstituteIndVar method.
Second, disable substitution of quadratic addrec expressions to avoid putting
multiplies in loops!
---
Diffs of the changes: (+7 -1)
Index: llvm/lib/Transforms/Scalar/IndVarSimplify.cpp
diff -u llvm/lib/Transforms/Scalar/IndVarSimplify.cpp:1.66 llvm/lib/Transforms/Scalar/IndVarSimplify.cpp:1.67
--- llvm/lib/Transforms/Scalar/IndVarSimplify.cpp:1.66 Sun Jun 20 00:04:01 2004
+++ llvm/lib/Transforms/Scalar/IndVarSimplify.cpp Thu Jun 24 01:49:18 2004
@@ -602,7 +602,13 @@
if (PN->getType()->isInteger()) { // FIXME: when we have fast-math, enable!
SCEVHandle SCEV = SE->getSCEV(PN);
if (SCEV->hasComputableLoopEvolution(L))
- if (SE->shouldSubstituteIndVar(SCEV)) // HACK!
+ // FIXME: Without a strength reduction pass, it is an extremely bad idea
+ // to indvar substitute anything more complex than a linear induction
+ // variable. Doing so will put expensive multiply instructions inside
+ // of the loop. For now just disable indvar subst on anything more
+ // complex than a linear addrec.
+ if (!isa<SCEVAddRecExpr>(SCEV) ||
+ cast<SCEVAddRecExpr>(SCEV)->getNumOperands() < 3)
IndVars.push_back(std::make_pair(PN, SCEV));
}
More information about the llvm-commits
mailing list