diff options
| -rw-r--r-- | polly/lib/Analysis/ScopDetection.cpp | 15 | ||||
| -rw-r--r-- | polly/test/ScopInfo/multidim_parameter_addrec_product.ll | 64 |
2 files changed, 74 insertions, 5 deletions
diff --git a/polly/lib/Analysis/ScopDetection.cpp b/polly/lib/Analysis/ScopDetection.cpp index 254f0af8b52..e69d0b98eff 100644 --- a/polly/lib/Analysis/ScopDetection.cpp +++ b/polly/lib/Analysis/ScopDetection.cpp @@ -495,9 +495,6 @@ bool ScopDetection::hasAffineMemoryAccesses(DetectionContext &Context) const { // First step: collect parametric terms in all array references. SmallVector<const SCEV *, 4> Terms; for (const auto &Pair : Context.Accesses[BasePointer]) { - if (auto *AF = dyn_cast<SCEVAddRecExpr>(Pair.second)) - SE->collectParametricTerms(AF, Terms); - // In case the outermost expression is a plain add, we check if any of its // terms has the form 4 * %inst * %param * %param ..., aka a term that // contains a product between a parameter and an instruction that is @@ -530,6 +527,8 @@ bool ScopDetection::hasAffineMemoryAccesses(DetectionContext &Context) const { } } } + if (Terms.empty()) + SE->collectParametricTerms(Pair.second, Terms); } // Second step: find array shape. @@ -539,13 +538,19 @@ bool ScopDetection::hasAffineMemoryAccesses(DetectionContext &Context) const { if (!AllowNonAffine) for (const SCEV *DelinearizedSize : Shape->DelinearizedSizes) { if (auto *Unknown = dyn_cast<SCEVUnknown>(DelinearizedSize)) { - auto *value = dyn_cast<Value>(Unknown->getValue()); - if (isa<UndefValue>(value)) { + auto *V = dyn_cast<Value>(Unknown->getValue()); + if (isa<UndefValue>(V)) { invalid<ReportDifferentArrayElementSize>( Context, /*Assert=*/true, Context.Accesses[BasePointer].front().first, BaseValue); return false; } + if (auto *Load = dyn_cast<LoadInst>(V)) { + if (Context.CurRegion.contains(Load) && + isHoistableLoad(Load, CurRegion, *LI, *SE)) + Context.RequiredILS.insert(Load); + continue; + } } if (hasScalarDepsInsideRegion(DelinearizedSize, &CurRegion)) invalid<ReportNonAffineAccess>( diff --git a/polly/test/ScopInfo/multidim_parameter_addrec_product.ll b/polly/test/ScopInfo/multidim_parameter_addrec_product.ll new file mode 100644 index 00000000000..54aa361a72a --- /dev/null +++ b/polly/test/ScopInfo/multidim_parameter_addrec_product.ll @@ -0,0 +1,64 @@ +; RUN: opt %loadPolly -polly-scops -analyze < %s | FileCheck %s +; +; void foo(float *A, long *p) { +; for (long i = 0; i < 100; i++) +; for (long j = 0; j < 100; j++) +; A[i * (*p) + j] += i + j; +; } + +; CHECK: Invariant Accesses: { +; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK-NEXT: [p] -> { Stmt_bb5[i0, i1] -> MemRef_p[0] }; +; CHECK-NEXT: Execution Context: [p] -> { : } +; CHECK-NEXT: } + +; CHECK: ReadAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK: [p] -> { Stmt_bb5[i0, i1] -> MemRef_A[i0, i1] }; +; CHECK: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0] +; CHECK: [p] -> { Stmt_bb5[i0, i1] -> MemRef_A[i0, i1] }; + +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" + +define void @foo(float* %A, i64* %p) { +bb: + br label %bb2 + +bb2: ; preds = %bb16, %bb + %i.0 = phi i64 [ 0, %bb ], [ %tmp17, %bb16 ] + %exitcond1 = icmp ne i64 %i.0, 100 + br i1 %exitcond1, label %bb3, label %bb18 + +bb3: ; preds = %bb2 + br label %bb4 + +bb4: ; preds = %bb13, %bb3 + %j.0 = phi i64 [ 0, %bb3 ], [ %tmp14, %bb13 ] + %exitcond = icmp ne i64 %j.0, 100 + br i1 %exitcond, label %bb5, label %bb15 + +bb5: ; preds = %bb4 + %tmp = add nuw nsw i64 %i.0, %j.0 + %tmp6 = sitofp i64 %tmp to float + %pval = load i64, i64* %p, align 8 + %tmp8 = mul nsw i64 %i.0, %pval + %tmp9 = add nsw i64 %tmp8, %j.0 + %tmp10 = getelementptr inbounds float, float* %A, i64 %tmp9 + %tmp11 = load float, float* %tmp10, align 4 + %tmp12 = fadd float %tmp11, %tmp6 + store float %tmp12, float* %tmp10, align 4 + br label %bb13 + +bb13: ; preds = %bb5 + %tmp14 = add nuw nsw i64 %j.0, 1 + br label %bb4 + +bb15: ; preds = %bb4 + br label %bb16 + +bb16: ; preds = %bb15 + %tmp17 = add nuw nsw i64 %i.0, 1 + br label %bb2 + +bb18: ; preds = %bb2 + ret void +} |
