[NaryReassociate] enhances nsw by leveraging @llvm.assume

[jingyue]

Summary

nsw are flaky and can often be removed by optimizations. This patch enhances
nsw by leveraging @llvm.assume in the IR. Specifically, NaryReassociate now
understands that

assume(a + b >= 0) && assume(a >= 0) ==> a +nsw b

As a result, it can split more sext(a + b) into sext(a) + sext(b) for CSE.

[meheff]

LGTM. Just one nit comment/question.

test/Transforms/NaryReassociate/NVPTX/nary-gep.ll

39

Should all "CHECK"s be replaced with "check"? or none?

[jingyue]

test/Transforms/NaryReassociate/NVPTX/nary-gep.ll

39

Didn't mean this at all. I must have hit some magic button! :)

[jingyue]

fixed typos

[jingyue]

fixed another typo

lib/Transforms/Scalar/NaryReassociate.cpp

 // result, ab2 becomes dead and ab will be used only once in the second
 // iteration.
 //
 // Limitations and TODO items:
 //
 // 1) We only considers n-ary adds for now. This should be extended and
 // generalized.
 //
-// 2) Besides arithmetic operations, similar reassociation can be applied to
-// GEPs. For example, if
-//   X = &arr[a]
-// dominates
-//   Y = &arr[a + b]
-// we may rewrite Y into X + b.
-//
 //===----------------------------------------------------------------------===//
 
+#include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/PatternMatch.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 using namespace PatternMatch;
 
 #define DEBUG_TYPE "nary-reassociate"
 
 namespace {
bool doInitialization(Module &M) override {
     return false;
   }
   bool runOnFunction(Function &F) override;
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addPreserved<DominatorTreeWrapperPass>();
     AU.addPreserved<ScalarEvolution>();
     AU.addPreserved<TargetLibraryInfoWrapperPass>();
+    AU.addRequired<AssumptionCacheTracker>();
     AU.addRequired<DominatorTreeWrapperPass>();
     AU.addRequired<ScalarEvolution>();
     AU.addRequired<TargetLibraryInfoWrapperPass>();
     AU.addRequired<TargetTransformInfoWrapperPass>();
     AU.setPreservesCFG();
   }
 
 private:
private:
   // \c CandidateExpr. Returns null if not found.
   Instruction *findClosestMatchingDominator(const SCEV *CandidateExpr,
                                             Instruction *Dominatee);
   // GetElementPtrInst implicitly sign-extends an index if the index is shorter
   // than the pointer size. This function returns whether Index is shorter than
   // GEP's pointer size, i.e., whether Index needs to be sign-extended in order
   // to be an index of GEP.
   bool requiresSignExtension(Value *Index, GetElementPtrInst *GEP);
+  // Returns whether V is known to be non-negative at context \c Ctxt.
+  bool isKnownNonNegative(Value *V, Instruction *Ctxt);
+  // Returns whether AO may sign overflow at context \c Ctxt. It computes a
+  // conservative result -- it answers true when not sure.
+  bool maySignOverflow(AddOperator *AO, Instruction *Ctxt);
 
+  AssumptionCache *AC;
+  const DataLayout *DL;
   DominatorTree *DT;
   ScalarEvolution *SE;
   TargetLibraryInfo *TLI;
   TargetTransformInfo *TTI;
-  const DataLayout *DL;
   // A lookup table quickly telling which instructions compute the given SCEV.
   // Note that there can be multiple instructions at different locations
   // computing to the same SCEV, so we map a SCEV to an instruction list.  For
   // example,
   //
   //   if (p1)
   //     foo(a + b);
   //   if (p2)
   //     bar(a + b);
   DenseMap<const SCEV *, SmallVector<Instruction *, 2>> SeenExprs;
 };
 } // anonymous namespace
 
 char NaryReassociate::ID = 0;
 INITIALIZE_PASS_BEGIN(NaryReassociate, "nary-reassociate", "Nary reassociation",
                       false, false)
+INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 INITIALIZE_PASS_END(NaryReassociate, "nary-reassociate", "Nary reassociation",
                     false, false)
 
 FunctionPass *llvm::createNaryReassociatePass() {
   return new NaryReassociate();
 }
 
 bool NaryReassociate::runOnFunction(Function &F) {
   if (skipOptnoneFunction(F))
     return false;
 
+  AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
   DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
   SE = &getAnalysis<ScalarEvolution>();
   TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
   TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
 
   bool Changed = false, ChangedInThisIteration;
   do {
     ChangedInThisIteration = doOneIteration(F);

 
 bool NaryReassociate::requiresSignExtension(Value *Index,
                                             GetElementPtrInst *GEP) {
   unsigned PointerSizeInBits =
       DL->getPointerSizeInBits(GEP->getType()->getPointerAddressSpace());
   return cast<IntegerType>(Index->getType())->getBitWidth() < PointerSizeInBits;
 }
 
+bool NaryReassociate::isKnownNonNegative(Value *V, Instruction *Ctxt) {
+  bool NonNegative, Negative;
+  // TODO: ComputeSignBits is expensive. Consider caching the results.
+  ComputeSignBit(V, NonNegative, Negative, *DL, 0, AC, Ctxt, DT);
+  return NonNegative;
+}
+
+bool NaryReassociate::maySignOverflow(AddOperator *AO, Instruction *Ctxt) {
+  if (AO->hasNoSignedWrap())
+    return false;
+
+  Value *LHS = AO->getOperand(0), *RHS = AO->getOperand(1);
+  // If LHS or RHS has the same sign as the sum, AO doesn't sign overflow.
+  // TODO: handle the negative case as well.
+  if (isKnownNonNegative(AO, Ctxt) &&
+      (isKnownNonNegative(LHS, Ctxt) || isKnownNonNegative(RHS, Ctxt)))
+    return false;
+
+  return true;
+}
+
 GetElementPtrInst *
 NaryReassociate::tryReassociateGEPAtIndex(GetElementPtrInst *GEP, unsigned I,
                                           Type *IndexedType) {
   Value *IndexToSplit = GEP->getOperand(I + 1);
-  if (SExtInst *SExt = dyn_cast<SExtInst>(IndexToSplit))
+  if (SExtInst *SExt = dyn_cast<SExtInst>(IndexToSplit)) {
     IndexToSplit = SExt->getOperand(0);
+  } else if (ZExtInst *ZExt = dyn_cast<ZExtInst>(IndexToSplit)) {
+    // zext can be treated as sext if the source is non-negative.
+    if (isKnownNonNegative(ZExt->getOperand(0), GEP))
+      IndexToSplit = ZExt->getOperand(0);
+  }
 
   if (AddOperator *AO = dyn_cast<AddOperator>(IndexToSplit)) {
     // If the I-th index needs sext and the underlying add is not equipped with
     // nsw, we cannot split the add because
     //   sext(LHS + RHS) != sext(LHS) + sext(RHS).
-    if (requiresSignExtension(IndexToSplit, GEP) && !AO->hasNoSignedWrap())
+    if (requiresSignExtension(IndexToSplit, GEP) && maySignOverflow(AO, GEP))
       return nullptr;
     Value *LHS = AO->getOperand(0), *RHS = AO->getOperand(1);
     // IndexToSplit = LHS + RHS.
     if (auto *NewGEP = tryReassociateGEPAtIndex(GEP, I, LHS, RHS, IndexedType))
       return NewGEP;
     // Symmetrically, try IndexToSplit = RHS + LHS.
     if (LHS != RHS) {
       if (auto *NewGEP =
               tryReassociateGEPAtIndex(GEP, I, RHS, LHS, IndexedType))
         return NewGEP;
     }
   }
   return nullptr;
 }
 
-GetElementPtrInst *
-NaryReassociate::tryReassociateGEPAtIndex(GetElementPtrInst *GEP, unsigned I,
-                                          Value *LHS, Value *RHS,
-                                          Type *IndexedType) {
+GetElementPtrInst *NaryReassociate::tryReassociateGEPAtIndex(
+    GetElementPtrInst *GEP, unsigned I, Value *LHS, Value *RHS,
   // Look for GEP's closest dominator that has the same SCEV as GEP except that
   // the I-th index is replaced with LHS.
   SmallVector<const SCEV *, 4> IndexExprs;
   for (auto Index = GEP->idx_begin(); Index != GEP->idx_end(); ++Index)
     IndexExprs.push_back(SE->getSCEV(*Index));
   // Replace the I-th index with LHS.
   IndexExprs[I] = SE->getSCEV(LHS);
+  if (isKnownNonNegative(LHS, GEP) &&
+      DL->getTypeSizeInBits(LHS->getType()) <
+          DL->getTypeSizeInBits(GEP->getOperand(I)->getType())) {
+    // Zero-extend LHS if it is non-negative. InstCombine canonicalizes sext to
+    // zext if the source operand is proved non-negative. We should do that
+    // consistently so that CandidateExpr more likely appears before. See
+    // @reassociate_gep_assume for an example of this canonicalization.
+    IndexExprs[I] =
+        SE->getZeroExtendExpr(IndexExprs[I], GEP->getOperand(I)->getType());
+  }
   const SCEV *CandidateExpr = SE->getGEPExpr(
       GEP->getSourceElementType(), SE->getSCEV(GEP->getPointerOperand()),
       IndexExprs, GEP->isInBounds());
 
   auto *Candidate = findClosestMatchingDominator(CandidateExpr, GEP);
   if (Candidate == nullptr)
     return nullptr;
 

test/Transforms/NaryReassociate/NVPTX/nary-gep.ll

-; RUN: opt < %s -nary-reassociate -early-cse -S | FileCheck %s
+; RUN: opt < %s -nary-reassociate -early-cse -S | FileCHECK %s
 
 target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
 target triple = "nvptx64-unknown-unknown"
 
 declare void @foo(float*)
 
 ; foo(&a[i]);
 ; foo(&a[i + j]);

 ;   =>
 ; t1 = &a[sext(j)];
 ; foo(t1);
 ; t2 = t1 + sext(i);
 ; foo(t2);
 ; t3 = t2 + sext(i); // sext(i) should be GVN'ed.
 ; foo(t3);
 define void @reassociate_gep_nsw(float* %a, i32 %i, i32 %j) {
 ; CHECK-LABEL: @reassociate_gep_nsw(
   %idxprom.j = sext i32 %j to i64
   %1 = getelementptr float, float* %a, i64 %idxprom.j
 ; CHECK: [[t1:[^ ]+]] = getelementptr float, float* %a, i64 %idxprom.j
   call void @foo(float* %1)
 ; CHECK: call void @foo(float* [[t1]])
 
   %2 = add nsw i32 %i, %j
   %idxprom.2 = sext i32 %2 to i64
   %3 = getelementptr float, float* %a, i64 %idxprom.2
 ; CHECK: [[sexti:[^ ]+]] = sext i32 %i to i64
 ; CHECK: [[t2:[^ ]+]] = getelementptr float, float* [[t1]], i64 [[sexti]]
   call void @foo(float* %3)
 ; CHECK: call void @foo(float* [[t2]])
 
   %4 = add nsw i32 %2, %i
   %idxprom.4 = sext i32 %4 to i64
   %5 = getelementptr float, float* %a, i64 %idxprom.4
 ; CHECK: [[t3:[^ ]+]] = getelementptr float, float* [[t2]], i64 [[sexti]]
   call void @foo(float* %5)
 ; CHECK: call void @foo(float* [[t3]])
 
   ret void
 }
 
+; assume(j >= 0);
+; foo(&a[zext(j)]);
+; assume(i + j >= 0);
+; foo(&a[zext(i + j)]);
+;   =>
+; t1 = &a[zext(j)];
+; foo(t1);
+; t2 = t1 + sext(i);
+; foo(t2);
+define void @reassociate_gep_assume(float* %a, i32 %i, i32 %j) {
+; CHECK-LABEL: @reassociate_gep_assume(
+  ; assume(j >= 0)
+  %cmp = icmp sgt i32 %j, -1
+  call void @llvm.assume(i1 %cmp)
+  %1 = add i32 %i, %j
+  %cmp2 = icmp sgt i32 %1, -1
+  call void @llvm.assume(i1 %cmp2)
+
+  %idxprom.j = zext i32 %j to i64
+  %2 = getelementptr float, float* %a, i64 %idxprom.j
+; CHECK: [[t1:[^ ]+]] = getelementptr float, float* %a, i64 %idxprom.j
+  call void @foo(float* %2)
+; CHECK: call void @foo(float* [[t1]])
+
+  %idxprom.1 = zext i32 %1 to i64
+  %3 = getelementptr float, float* %a, i64 %idxprom.1
+; CHECK: [[sexti:[^ ]+]] = sext i32 %i to i64
+; CHECK: [[t2:[^ ]+]] = getelementptr float, float* [[t1]], i64 [[sexti]]
+  call void @foo(float* %3)
+; CHECK: call void @foo(float* [[t2]])
+
+  ret void
+}
+
 ; Do not split the second GEP because sext(i + j) != sext(i) + sext(j).
 define void @reassociate_gep_no_nsw(float* %a, i32 %i, i32 %j) {
 ; CHECK-LABEL: @reassociate_gep_no_nsw(
   %1 = add i32 %i, %j
   %2 = getelementptr float, float* %a, i32 %j
 ; CHECK: getelementptr float, float* %a, i32 %j
   call void @foo(float* %2)
   %3 = getelementptr float, float* %a, i32 %1

 ; CHECK: call void @foo(float* [[t1]])
   %3 = getelementptr float, float* %a, i128 %1
 ; CHECK: [[truncj:[^ ]+]] = trunc i128 %j to i64
 ; CHECK: [[t2:[^ ]+]] = getelementptr float, float* [[t1]], i64 [[truncj]]
   call void @foo(float* %3)
 ; CHECK: call void @foo(float* [[t2]])
   ret void
 }
+
+declare void @llvm.assume(i1)