diff --git a/polly/lib/Analysis/ScopBuilder.cpp b/polly/lib/Analysis/ScopBuilder.cpp index c62cb2a85c83..4486164ce0e2 100644 --- a/polly/lib/Analysis/ScopBuilder.cpp +++ b/polly/lib/Analysis/ScopBuilder.cpp @@ -2536,18 +2536,39 @@ bool hasIntersectingAccesses(isl::set AllAccs, MemoryAccess *LoadMA, bool checkCandidatePairAccesses(MemoryAccess *LoadMA, MemoryAccess *StoreMA, isl::set Domain, SmallVector &MemAccs) { + // First check if the base value is the same. isl::map LoadAccs = LoadMA->getAccessRelation(); isl::map StoreAccs = StoreMA->getAccessRelation(); - - // Skip those with obviously unequal base addresses. bool Valid = LoadAccs.has_equal_space(StoreAccs); + LLVM_DEBUG(dbgs() << " == The accessed space below is " + << (Valid ? "" : "not ") << "equal!\n"); + LLVM_DEBUG(LoadMA->dump(); StoreMA->dump()); - // And check if the remaining for overlap with other memory accesses. if (Valid) { + // Then check if they actually access the same memory. + isl::map R = isl::manage(LoadAccs.copy()) + .intersect_domain(isl::manage(Domain.copy())); + isl::map W = isl::manage(StoreAccs.copy()) + .intersect_domain(isl::manage(Domain.copy())); + isl::set RS = R.range(); + isl::set WS = W.range(); + + isl::set InterAccs = + isl::manage(RS.copy()).intersect(isl::manage(WS.copy())); + Valid = !InterAccs.is_empty(); + LLVM_DEBUG(dbgs() << " == The accessed memory is " << (Valid ? "" : "not ") + << "overlapping!\n"); + } + + if (Valid) { + // Finally, check if they are no other instructions accessing this memory isl::map AllAccsRel = LoadAccs.unite(StoreAccs); AllAccsRel = AllAccsRel.intersect_domain(Domain); isl::set AllAccs = AllAccsRel.range(); Valid = !hasIntersectingAccesses(AllAccs, LoadMA, StoreMA, Domain, MemAccs); + + LLVM_DEBUG(dbgs() << " == The accessed memory is " << (Valid ? "not " : "") + << "accessed by other instructions!\n"); } return Valid; } diff --git a/polly/test/ScopInfo/reduction_different_index.ll b/polly/test/ScopInfo/reduction_different_index.ll new file mode 100644 index 000000000000..575e5a16d7b2 --- /dev/null +++ b/polly/test/ScopInfo/reduction_different_index.ll @@ -0,0 +1,38 @@ +; RUN: opt %loadPolly -polly-print-scops -disable-output < %s | FileCheck %s +; Verify if the following case is not detected as reduction. +; +; void f(int *A,int *sum) { +; for (int i = 0; i < 1024; i++) +; sum[0] = sum[1] + A[i]; +; } +; +; Verify that we don't detect the reduction on sum +; +; CHECK: ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_sum[1] }; +; CHECK-NEXT:ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_A[i0] }; +; CHECK-NEXT:MustWriteAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_sum[0] }; +; +target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" + +define dso_local void @f(ptr nocapture noundef readonly %A, ptr nocapture noundef %sum) local_unnamed_addr #0 { +entry: + br label %for.body + +for.cond.cleanup: ; preds = %for.body + ret void + +for.body: ; preds = %entry.split, %for.body + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %arrayidx = getelementptr inbounds i32, ptr %sum, i64 1 + %0 = load i32, ptr %arrayidx + %arrayidx1 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv + %1 = load i32, ptr %arrayidx1 + %add = add nsw i32 %1, %0 + store i32 %add, ptr %sum + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 + %exitcond.not = icmp eq i64 %indvars.iv.next, 1024 + br i1 %exitcond.not, label %for.cond.cleanup, label %for.body +} diff --git a/polly/test/ScopInfo/reduction_different_index1.ll b/polly/test/ScopInfo/reduction_different_index1.ll new file mode 100644 index 000000000000..39bd3c4b9abe --- /dev/null +++ b/polly/test/ScopInfo/reduction_different_index1.ll @@ -0,0 +1,45 @@ +; RUN: opt %loadPolly -polly-print-scops -disable-output < %s | FileCheck %s +; Verify if the following case is not detected as reduction. +; +; void f(int *A, int *sum, int i1, int i2) { +; for (int i = 0; i < 1024; i++) +; sum[i2] = sum[i1] + A[i]; +; } +; +; Verify that we don't detect the reduction on sum +; +; CHECK: ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_sum[i1] }; +; CHECK-NEXT:ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_A[i0] }; +; CHECK-NEXT:MustWriteAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_sum[i2] }; +; +target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" + +; Function Attrs: nofree norecurse nosync nounwind memory(argmem: readwrite) uwtable +define dso_local void @f(ptr nocapture noundef readonly %A, ptr nocapture noundef %sum, i32 noundef %i1, i32 noundef %i2) local_unnamed_addr #0 { +entry: + br label %entry.split + +entry.split: ; preds = %entry + %idxprom = sext i32 %i1 to i64 + %arrayidx = getelementptr inbounds i32, ptr %sum, i64 %idxprom + %idxprom3 = sext i32 %i2 to i64 + %arrayidx4 = getelementptr inbounds i32, ptr %sum, i64 %idxprom3 + br label %for.body + +for.cond.cleanup: ; preds = %for.body + ret void + +for.body: ; preds = %entry.split, %for.body + %indvars.iv = phi i64 [ 0, %entry.split ], [ %indvars.iv.next, %for.body ] + %0 = load i32, ptr %arrayidx, align 4 + %arrayidx2 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv + %1 = load i32, ptr %arrayidx2, align 4 + %add = add nsw i32 %1, %0 + store i32 %add, ptr %arrayidx4, align 4 + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 + %exitcond.not = icmp eq i64 %indvars.iv.next, 1024 + br i1 %exitcond.not, label %for.cond.cleanup, label %for.body +}