//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "loop-unroll"
-#include "llvm/IntrinsicInst.h"
#include "llvm/Transforms/Scalar.h"
-#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/CodeMetrics.h"
+#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/IntrinsicInst.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
cl::desc("Allows loops to be partially unrolled until "
"-unroll-threshold loop size is reached."));
-// Temporary flag to be made default shortly.
static cl::opt<bool>
-UnrollWithSCEV("unroll-scev", cl::init(false), cl::Hidden,
- cl::desc("Use ScalarEvolution to analyze loop trip counts for unrolling"));
+UnrollRuntime("unroll-runtime", cl::ZeroOrMore, cl::init(false), cl::Hidden,
+ cl::desc("Unroll loops with run-time trip counts"));
namespace {
class LoopUnroll : public LoopPass {
// explicit -unroll-threshold).
static const unsigned OptSizeUnrollThreshold = 50;
+ // Default unroll count for loops with run-time trip count if
+ // -unroll-count is not set
+ static const unsigned UnrollRuntimeCount = 8;
+
unsigned CurrentCount;
unsigned CurrentThreshold;
bool CurrentAllowPartial;
INITIALIZE_PASS_DEPENDENCY(LoopInfo)
INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
INITIALIZE_PASS_DEPENDENCY(LCSSA)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
Pass *llvm::createLoopUnrollPass(int Threshold, int Count, int AllowPartial) {
}
/// ApproximateLoopSize - Approximate the size of the loop.
-static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls) {
+static unsigned ApproximateLoopSize(const Loop *L, unsigned &NumCalls,
+ bool &NotDuplicatable, const DataLayout *TD) {
CodeMetrics Metrics;
for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
I != E; ++I)
- Metrics.analyzeBasicBlock(*I);
+ Metrics.analyzeBasicBlock(*I, TD);
NumCalls = Metrics.NumInlineCandidates;
+ NotDuplicatable = Metrics.notDuplicatable;
unsigned LoopSize = Metrics.NumInsts;
// not user specified.
unsigned Threshold = CurrentThreshold;
if (!UserThreshold &&
- Header->getParent()->hasFnAttr(Attribute::OptimizeForSize))
+ Header->getParent()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex,
+ Attribute::OptimizeForSize))
Threshold = OptSizeUnrollThreshold;
// Find trip count and trip multiple if count is not available
unsigned TripCount = 0;
unsigned TripMultiple = 1;
- if (UnrollWithSCEV) {
- // Find "latch trip count". UnrollLoop assumes that control cannot exit
- // via the loop latch on any iteration prior to TripCount. The loop may exit
- // early via an earlier branch.
- BasicBlock *LatchBlock = L->getLoopLatch();
- if (LatchBlock) {
- TripCount = SE->getSmallConstantTripCount(L, LatchBlock);
- TripMultiple = SE->getSmallConstantTripMultiple(L, LatchBlock);
- }
- }
- else {
- TripCount = L->getSmallConstantTripCount();
- if (TripCount == 0)
- TripMultiple = L->getSmallConstantTripMultiple();
+ // Find "latch trip count". UnrollLoop assumes that control cannot exit
+ // via the loop latch on any iteration prior to TripCount. The loop may exit
+ // early via an earlier branch.
+ BasicBlock *LatchBlock = L->getLoopLatch();
+ if (LatchBlock) {
+ TripCount = SE->getSmallConstantTripCount(L, LatchBlock);
+ TripMultiple = SE->getSmallConstantTripMultiple(L, LatchBlock);
}
- // Automatically select an unroll count.
+ // Use a default unroll-count if the user doesn't specify a value
+ // and the trip count is a run-time value. The default is different
+ // for run-time or compile-time trip count loops.
unsigned Count = CurrentCount;
+ if (UnrollRuntime && CurrentCount == 0 && TripCount == 0)
+ Count = UnrollRuntimeCount;
+
if (Count == 0) {
// Conservative heuristic: if we know the trip count, see if we can
// completely unroll (subject to the threshold, checked below); otherwise
// Enforce the threshold.
if (Threshold != NoThreshold) {
+ const DataLayout *TD = getAnalysisIfAvailable<DataLayout>();
unsigned NumInlineCandidates;
- unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates);
+ bool notDuplicatable;
+ unsigned LoopSize = ApproximateLoopSize(L, NumInlineCandidates,
+ notDuplicatable, TD);
DEBUG(dbgs() << " Loop Size = " << LoopSize << "\n");
+ if (notDuplicatable) {
+ DEBUG(dbgs() << " Not unrolling loop which contains non duplicatable"
+ << " instructions.\n");
+ return false;
+ }
if (NumInlineCandidates != 0) {
DEBUG(dbgs() << " Not unrolling loop with inlinable calls.\n");
return false;
if (TripCount != 1 && Size > Threshold) {
DEBUG(dbgs() << " Too large to fully unroll with count: " << Count
<< " because size: " << Size << ">" << Threshold << "\n");
- if (!CurrentAllowPartial) {
+ if (!CurrentAllowPartial && !(UnrollRuntime && TripCount == 0)) {
DEBUG(dbgs() << " will not try to unroll partially because "
<< "-unroll-allow-partial not given\n");
return false;
}
- // Reduce unroll count to be modulo of TripCount for partial unrolling
- Count = Threshold / LoopSize;
- while (Count != 0 && TripCount%Count != 0) {
- Count--;
+ if (TripCount) {
+ // Reduce unroll count to be modulo of TripCount for partial unrolling
+ Count = Threshold / LoopSize;
+ while (Count != 0 && TripCount%Count != 0)
+ Count--;
+ }
+ else if (UnrollRuntime) {
+ // Reduce unroll count to be a lower power-of-two value
+ while (Count != 0 && Size > Threshold) {
+ Count >>= 1;
+ Size = LoopSize*Count;
+ }
}
if (Count < 2) {
DEBUG(dbgs() << " could not unroll partially\n");
}
// Unroll the loop.
- if (!UnrollLoop(L, Count, TripCount, TripMultiple, LI, &LPM))
+ if (!UnrollLoop(L, Count, TripCount, UnrollRuntime, TripMultiple, LI, &LPM))
return false;
return true;
projects / oota-llvm.git / blobdiff