Use a sign-extend instead of a zero-extend when promoting a

trip count value when the original loop iteration condition is
signed and the canonical induction variable won't undergo signed
overflow. This isn't required for correctness; it just preserves
more information about original loop iteration values.

Add a getTruncateOrSignExtend method to ScalarEvolution,
following getTruncateOrZeroExtend.

llvm-svn: 64918
This commit is contained in:
Dan Gohman 2009-02-18 17:22:41 +00:00
parent 5530918aff
commit 451474da4a
6 changed files with 96 additions and 19 deletions

View File

@ -264,6 +264,11 @@ namespace llvm {
/// extended, it is zero extended.
SCEVHandle getTruncateOrZeroExtend(const SCEVHandle &V, const Type *Ty);
/// getTruncateOrSignExtend - Return a SCEV corresponding to a conversion
/// of the input value to the specified type. If the type must be
/// extended, it is sign extended.
SCEVHandle getTruncateOrSignExtend(const SCEVHandle &V, const Type *Ty);
/// getIntegerSCEV - Given an integer or FP type, create a constant for the
/// specified signed integer value and return a SCEV for the constant.
SCEVHandle getIntegerSCEV(int Val, const Type *Ty);

View File

@ -755,6 +755,21 @@ SCEVHandle ScalarEvolution::getTruncateOrZeroExtend(const SCEVHandle &V,
return getZeroExtendExpr(V, Ty);
}
/// getTruncateOrSignExtend - Return a SCEV corresponding to a conversion
/// of the input value to the specified type. If the type must be
/// extended, it is sign extended.
SCEVHandle ScalarEvolution::getTruncateOrSignExtend(const SCEVHandle &V,
const Type *Ty) {
const Type *SrcTy = V->getType();
assert(SrcTy->isInteger() && Ty->isInteger() &&
"Cannot truncate or sign extend with non-integer arguments!");
if (SrcTy->getPrimitiveSizeInBits() == Ty->getPrimitiveSizeInBits())
return V; // No conversion
if (SrcTy->getPrimitiveSizeInBits() > Ty->getPrimitiveSizeInBits())
return getTruncateExpr(V, Ty);
return getSignExtendExpr(V, Ty);
}
// get - Get a canonical add expression, or something simpler if possible.
SCEVHandle ScalarEvolution::getAddExpr(std::vector<SCEVHandle> &Ops) {
assert(!Ops.empty() && "Cannot get empty add!");

View File

@ -96,7 +96,8 @@ namespace {
Value *IndVar,
BasicBlock *ExitingBlock,
BranchInst *BI,
SCEVExpander &Rewriter);
SCEVExpander &Rewriter,
bool SignExtendTripCount);
void RewriteLoopExitValues(Loop *L, SCEV *IterationCount);
void DeleteTriviallyDeadInstructions(SmallPtrSet<Instruction*, 16> &Insts);
@ -235,7 +236,8 @@ void IndVarSimplify::LinearFunctionTestReplace(Loop *L,
Value *IndVar,
BasicBlock *ExitingBlock,
BranchInst *BI,
SCEVExpander &Rewriter) {
SCEVExpander &Rewriter,
bool SignExtendTripCount) {
// If the exiting block is not the same as the backedge block, we must compare
// against the preincremented value, otherwise we prefer to compare against
// the post-incremented value.
@ -253,11 +255,18 @@ void IndVarSimplify::LinearFunctionTestReplace(Loop *L,
if ((isa<SCEVConstant>(N) && !N->isZero()) ||
SE->isLoopGuardedByCond(L, ICmpInst::ICMP_NE, N, Zero)) {
// No overflow. Cast the sum.
IterationCount = SE->getTruncateOrZeroExtend(N, IndVar->getType());
if (SignExtendTripCount)
IterationCount = SE->getTruncateOrSignExtend(N, IndVar->getType());
else
IterationCount = SE->getTruncateOrZeroExtend(N, IndVar->getType());
} else {
// Potential overflow. Cast before doing the add.
IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
IndVar->getType());
if (SignExtendTripCount)
IterationCount = SE->getTruncateOrSignExtend(IterationCount,
IndVar->getType());
else
IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
IndVar->getType());
IterationCount =
SE->getAddExpr(IterationCount,
SE->getIntegerSCEV(1, IndVar->getType()));
@ -269,8 +278,12 @@ void IndVarSimplify::LinearFunctionTestReplace(Loop *L,
CmpIndVar = L->getCanonicalInductionVariableIncrement();
} else {
// We have to use the preincremented value...
IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
IndVar->getType());
if (SignExtendTripCount)
IterationCount = SE->getTruncateOrSignExtend(IterationCount,
IndVar->getType());
else
IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
IndVar->getType());
CmpIndVar = IndVar;
}
@ -464,10 +477,13 @@ static const Type *getEffectiveIndvarType(const PHINode *Phi) {
/// TestOrigIVForWrap - Analyze the original induction variable
/// that controls the loop's iteration to determine whether it
/// would ever undergo signed or unsigned overflow.
/// would ever undergo signed or unsigned overflow. Also, check
/// whether an induction variable in the same type that starts
/// at 0 would undergo signed overflow.
///
/// In addition to setting the NoSignedWrap and NoUnsignedWrap
/// variables, return the PHI for this induction variable.
/// In addition to setting the NoSignedWrap, NoUnsignedWrap, and
/// SignExtendTripCount variables, return the PHI for this induction
/// variable.
///
/// TODO: This duplicates a fair amount of ScalarEvolution logic.
/// Perhaps this can be merged with ScalarEvolution::getIterationCount
@ -477,7 +493,8 @@ static const PHINode *TestOrigIVForWrap(const Loop *L,
const BranchInst *BI,
const Instruction *OrigCond,
bool &NoSignedWrap,
bool &NoUnsignedWrap) {
bool &NoUnsignedWrap,
bool &SignExtendTripCount) {
// Verify that the loop is sane and find the exit condition.
const ICmpInst *Cmp = dyn_cast<ICmpInst>(OrigCond);
if (!Cmp) return 0;
@ -590,9 +607,13 @@ static const PHINode *TestOrigIVForWrap(const Loop *L,
// The original induction variable will start at some non-max value,
// it counts up by one, and the loop iterates only while it remans
// less than some value in the same type. As such, it will never wrap.
if (isSigned && !InitialVal->getValue().isMaxSignedValue())
if (isSigned && !InitialVal->getValue().isMaxSignedValue()) {
NoSignedWrap = true;
else if (!isSigned && !InitialVal->getValue().isMaxValue())
// If the original induction variable starts at zero or greater,
// the trip count can be considered signed.
if (InitialVal->getValue().isNonNegative())
SignExtendTripCount = true;
} else if (!isSigned && !InitialVal->getValue().isMaxValue())
NoUnsignedWrap = true;
return PN;
}
@ -678,6 +699,7 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
// using it. We can currently only handle loops with a single exit.
bool NoSignedWrap = false;
bool NoUnsignedWrap = false;
bool SignExtendTripCount = false;
const PHINode *OrigControllingPHI = 0;
if (!isa<SCEVCouldNotCompute>(IterationCount) && ExitingBlock)
// Can't rewrite non-branch yet.
@ -686,14 +708,16 @@ bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
// Determine if the OrigIV will ever undergo overflow.
OrigControllingPHI =
TestOrigIVForWrap(L, BI, OrigCond,
NoSignedWrap, NoUnsignedWrap);
NoSignedWrap, NoUnsignedWrap,
SignExtendTripCount);
// We'll be replacing the original condition, so it'll be dead.
DeadInsts.insert(OrigCond);
}
LinearFunctionTestReplace(L, IterationCount, IndVar,
ExitingBlock, BI, Rewriter);
ExitingBlock, BI, Rewriter,
SignExtendTripCount);
}
// Now that we have a canonical induction variable, we can rewrite any

View File

@ -1,5 +1,6 @@
; RUN: llvm-as < %s | opt -indvars | llvm-dis > %t
; RUN: grep sext %t | count 1
; RUN: grep sext %t | count 2
; RUN: grep { = sext i32 %n to i64} %t
; RUN: grep phi %t | count 1
; RUN: grep {phi i64} %t

View File

@ -1,6 +1,7 @@
; RUN: llvm-as < %s | opt -indvars | llvm-dis | not grep sext
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
; RUN: llvm-as < %s | opt -indvars | llvm-dis > %t
; RUN: grep sext %t | count 2
; RUN: grep { = sext i16 %N to i64} %t
; RUN: grep { = sext i32 %count to i64} %t
define i64 @test(i64* nocapture %first, i32 %count) nounwind readonly {
entry:

View File

@ -0,0 +1,31 @@
; RUN: llvm-as < %s | opt -indvars | llvm-dis > %t
; RUN: grep { = sext i32 %n} %t
; RUN: grep phi %t | count 1
; RUN: not grep zext %t
define void @foo(i64* nocapture %x, i32 %n) nounwind {
entry:
%tmp102 = icmp sgt i32 %n, 0 ; <i1> [#uses=1]
br i1 %tmp102, label %bb.nph, label %return
bb.nph: ; preds = %entry
br label %bb
bb: ; preds = %bb7, %bb.nph
%i.01 = phi i32 [ %tmp6, %bb7 ], [ 0, %bb.nph ] ; <i32> [#uses=3]
%tmp1 = sext i32 %i.01 to i64 ; <i64> [#uses=1]
%tmp4 = getelementptr i64* %x, i32 %i.01 ; <i64*> [#uses=1]
store i64 %tmp1, i64* %tmp4, align 8
%tmp6 = add i32 %i.01, 1 ; <i32> [#uses=2]
br label %bb7
bb7: ; preds = %bb
%tmp10 = icmp slt i32 %tmp6, %n ; <i1> [#uses=1]
br i1 %tmp10, label %bb, label %bb7.return_crit_edge
bb7.return_crit_edge: ; preds = %bb7
br label %return
return: ; preds = %bb7.return_crit_edge, %entry
ret void
}