[SLSR] handles off bounds GEPs

Summary:
The old requirement on GEP candidates being in bounds is unnecessary.
For off-bound GEPs, we still have

  &B[i * S] = B + (i * S) * e = B + (i * e) * S

Test Plan: slsr_offbound_gep in slsr-gep.ll

Reviewers: meheff

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D8809

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@233949 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp b/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
index e71031c79c2eaaeff5d7bd15e3b1c8cfcbe675d3..940ab58f1586b7b604f8d18eb7389accf07369f5 100644 (file)
--- a/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
+++ b/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp
@@ -333,7 +333,8 @@ void StraightLineStrengthReduce::allocateCandidateAndFindBasisForMul(
 void StraightLineStrengthReduce::allocateCandidateAndFindBasisForGEP(
     const SCEV *B, ConstantInt *Idx, Value *S, uint64_t ElementSize,
     Instruction *I) {
-  // I = B + sext(Idx *nsw S) *nsw ElementSize
+  // I = B + sext(Idx *nsw S) * ElementSize
+  //   = B + (sext(Idx) * sext(S)) * ElementSize
   //   = B + (sext(Idx) * ElementSize) * sext(S)
   // Casting to IntegerType is safe because we skipped vector GEPs.
   IntegerType *IntPtrTy = cast<IntegerType>(DL->getIntPtrType(I->getType()));
@@ -367,7 +368,7 @@ void StraightLineStrengthReduce::factorArrayIndex(Value *ArrayIdx,
   // sext'ed multiplication.
   if (match(ArrayIdx, m_NSWMul(m_Value(LHS), m_ConstantInt(RHS)))) {
     // SLSR is currently unsafe if i * S may overflow.
-    // GEP = Base + sext(LHS *nsw RHS) *nsw ElementSize
+    // GEP = Base + sext(LHS *nsw RHS) * ElementSize
     allocateCandidateAndFindBasisForGEP(Base, RHS, LHS, ElementSize, GEP);
   }
 }
@@ -472,19 +473,25 @@ void StraightLineStrengthReduce::rewriteCandidateWithBasis(
   case Candidate::GEP:
     {
       Type *IntPtrTy = DL->getIntPtrType(C.Ins->getType());
+      bool InBounds = cast<GetElementPtrInst>(C.Ins)->isInBounds();
       if (BumpWithUglyGEP) {
         // C = (char *)Basis + Bump
         unsigned AS = Basis.Ins->getType()->getPointerAddressSpace();
         Type *CharTy = Type::getInt8PtrTy(Basis.Ins->getContext(), AS);
         Reduced = Builder.CreateBitCast(Basis.Ins, CharTy);
-        // We only considered inbounds GEP as candidates.
-        Reduced = Builder.CreateInBoundsGEP(Reduced, Bump);
+        if (InBounds)
+          Reduced = Builder.CreateInBoundsGEP(Reduced, Bump);
+        else
+          Reduced = Builder.CreateGEP(Reduced, Bump);
         Reduced = Builder.CreateBitCast(Reduced, C.Ins->getType());
       } else {
         // C = gep Basis, Bump
         // Canonicalize bump to pointer size.
         Bump = Builder.CreateSExtOrTrunc(Bump, IntPtrTy);
-        Reduced = Builder.CreateInBoundsGEP(Basis.Ins, Bump);
+        if (InBounds)
+          Reduced = Builder.CreateInBoundsGEP(Basis.Ins, Bump);
+        else
+          Reduced = Builder.CreateGEP(Basis.Ins, Bump);
       }
     }
     break;

diff --git a/test/Transforms/StraightLineStrengthReduce/slsr-gep.ll b/test/Transforms/StraightLineStrengthReduce/slsr-gep.ll
index 47e66371ddc6a42daef1c2eb44f8bc8cdf752a2e..e22013b782452453df8c82301958e02b621d3e7e 100644 (file)
--- a/test/Transforms/StraightLineStrengthReduce/slsr-gep.ll
+++ b/test/Transforms/StraightLineStrengthReduce/slsr-gep.ll
@@ -107,3 +107,28 @@ define i64 @slsr_gep_uglygep([10 x [5 x %struct.S]]* %input, i64 %s, i64 %t) {
   %2 = add i64 %1, %v2
   ret i64 %2
 }
+
+define i32 @slsr_out_of_bounds_gep(i32* %input, i32 %s) {
+; CHECK-LABEL: @slsr_out_of_bounds_gep(
+  ; v0 = input[0];
+  %p0 = getelementptr i32, i32* %input, i64 0
+  %v0 = load i32, i32* %p0
+
+  ; v1 = input[(long)s];
+  %t = sext i32 %s to i64
+  %p1 = getelementptr i32, i32* %input, i64 %t
+; CHECK: %p1 = getelementptr i32, i32* %input, i64 %t
+  %v1 = load i32, i32* %p1
+
+  ; v2 = input[(long)(s * 2)];
+  %s2 = mul nsw i32 %s, 2
+  %t2 = sext i32 %s2 to i64
+  %p2 = getelementptr i32, i32* %input, i64 %t2
+; CHECK: %p2 = getelementptr i32, i32* %p1, i64 %t
+  %v2 = load i32, i32* %p2
+
+  ; return v0 + v1 + v2;
+  %1 = add i32 %v0, %v1
+  %2 = add i32 %1, %v2
+  ret i32 %2
+}