[LoopUnrollRuntime] Avoid high-cost trip count computation.

Summary:
Runtime unrolling of loops needs to emit an expression to compute the
loop's runtime trip-count.  Avoid runtime unrolling if this computation
will be expensive.

Depends on D8993.

Reviewers: atrick

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8994

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@234846 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Sanjoy Das
2015-04-14 03:20:38 +00:00
parent 17e08f50b9
commit 888e8e3a66
8 changed files with 63 additions and 14 deletions

View File

@@ -278,7 +278,8 @@ static void CloneLoopBlocks(Loop *L, Value *NewIter, const bool UnrollProlog,
/// ...
/// End:
///
bool llvm::UnrollRuntimeLoopProlog(Loop *L, unsigned Count, LoopInfo *LI,
bool llvm::UnrollRuntimeLoopProlog(Loop *L, unsigned Count,
bool AllowExpensiveTripCount, LoopInfo *LI,
LPPassManager *LPM) {
// for now, only unroll loops that contain a single exit
if (!L->getExitingBlock())
@@ -312,6 +313,12 @@ bool llvm::UnrollRuntimeLoopProlog(Loop *L, unsigned Count, LoopInfo *LI,
if (isa<SCEVCouldNotCompute>(TripCountSC))
return false;
BasicBlock *Header = L->getHeader();
const DataLayout &DL = Header->getModule()->getDataLayout();
SCEVExpander Expander(*SE, DL, "loop-unroll");
if (!AllowExpensiveTripCount && Expander.isHighCostExpansion(TripCountSC, L))
return false;
// We only handle cases when the unroll factor is a power of 2.
// Count is the loop unroll factor, the number of extra copies added + 1.
if (!isPowerOf2_32(Count))
@@ -332,18 +339,15 @@ bool llvm::UnrollRuntimeLoopProlog(Loop *L, unsigned Count, LoopInfo *LI,
auto *DT = DTWP ? &DTWP->getDomTree() : nullptr;
BasicBlock *PH = L->getLoopPreheader();
BasicBlock *Header = L->getHeader();
BasicBlock *Latch = L->getLoopLatch();
// It helps to splits the original preheader twice, one for the end of the
// prolog code and one for a new loop preheader
BasicBlock *PEnd = SplitEdge(PH, Header, DT, LI);
BasicBlock *NewPH = SplitBlock(PEnd, PEnd->getTerminator(), DT, LI);
BranchInst *PreHeaderBR = cast<BranchInst>(PH->getTerminator());
const DataLayout &DL = Header->getModule()->getDataLayout();
// Compute the number of extra iterations required, which is:
// extra iterations = run-time trip count % (loop unroll factor + 1)
SCEVExpander Expander(*SE, DL, "loop-unroll");
Value *TripCount = Expander.expandCodeFor(TripCountSC, TripCountSC->getType(),
PreHeaderBR);
Value *BECount = Expander.expandCodeFor(BECountSC, BECountSC->getType(),