[ScopDetection] Do not allow required-invariant loads in non-affine region

These loads cannot be savely hoisted as the condition guarding the
non-affine region cannot be duplicated to also protect the hoisted load
later on. Today they are dropped in ScopInfo. By checking for this early, we
do not even try to model them and possibly can still optimize smaller regions
not containing this specific required-invariant load.

llvm-svn: 296744

2 files changed, 73 insertions, 1 deletions

diff --git a/polly/lib/Analysis/ScopDetection.cpp b/polly/lib/Analysis/ScopDetection.cpp
index 5b7fc34f932..a637afa5a41 100644
--- a/polly/lib/Analysis/ScopDetection.cpp
+++ b/polly/lib/Analysis/ScopDetection.cpp
@@ -347,10 +347,16 @@ bool ScopDetection::onlyValidRequiredInvariantLoads(
   if (!PollyInvariantLoadHoisting && !RequiredILS.empty())
     return false;
 
-  for (LoadInst *Load : RequiredILS)
+  for (LoadInst *Load : RequiredILS) {
     if (!isHoistableLoad(Load, CurRegion, *LI, *SE, *DT))
       return false;
 
+    for (auto NonAffineRegion : Context.NonAffineSubRegionSet)
+      if (NonAffineRegion->contains(Load) &&
+          Load->getParent() != NonAffineRegion->getEntry())
+        return false;
+  }
+
   Context.RequiredILS.insert(RequiredILS.begin(), RequiredILS.end());
 
   return true;
diff --git a/polly/test/ScopInfo/invariant_load_in_non_affine.ll b/polly/test/ScopInfo/invariant_load_in_non_affine.ll
new file mode 100644
index 00000000000..59cb265c290
--- /dev/null
+++ b/polly/test/ScopInfo/invariant_load_in_non_affine.ll
@@ -0,0 +1,66 @@
+; RUN: opt %loadPolly -polly-detect -analyze \
+; RUN:   -polly-invariant-load-hoisting=true < %s | FileCheck %s
+;
+; CHECK-NOT: Valid Region for Scop
+;
+;    void foo(float A[], float B[], long *p) {
+;      for (long i = 0; i < 1024; i++)
+;        for (long j = 0; j < 1024; j++)
+;          if (B[i])
+;            A[i * (*p) + j] += i * j;
+;    }
+;
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @foo(float* %A, float* %B, i64* %p) {
+bb:
+  br label %bb2
+
+bb2:                                              ; preds = %bb21, %bb
+  %i.0 = phi i64 [ 0, %bb ], [ %tmp22, %bb21 ]
+  %exitcond1 = icmp ne i64 %i.0, 1024
+  br i1 %exitcond1, label %bb3, label %bb23
+
+bb3:                                              ; preds = %bb2
+  br label %bb4
+
+bb4:                                              ; preds = %bb18, %bb3
+  %j.0 = phi i64 [ 0, %bb3 ], [ %tmp19, %bb18 ]
+  %exitcond = icmp ne i64 %j.0, 1024
+  br i1 %exitcond, label %bb5, label %bb20
+
+bb5:                                              ; preds = %bb4
+  %tmp = getelementptr inbounds float, float* %B, i64 %i.0
+  %tmp6 = load float, float* %tmp, align 4
+  %tmp7 = fcmp une float %tmp6, 0.000000e+00
+  br i1 %tmp7, label %bb8, label %bb17
+
+bb8:                                              ; preds = %bb5
+  %tmp9 = mul nuw nsw i64 %i.0, %j.0
+  %tmp10 = sitofp i64 %tmp9 to float
+  %tmp11 = load i64, i64* %p, align 8
+  %tmp12 = mul nsw i64 %i.0, %tmp11
+  %tmp13 = add nsw i64 %tmp12, %j.0
+  %tmp14 = getelementptr inbounds float, float* %A, i64 %tmp13
+  %tmp15 = load float, float* %tmp14, align 4
+  %tmp16 = fadd float %tmp15, %tmp10
+  store float %tmp16, float* %tmp14, align 4
+  br label %bb17
+
+bb17:                                             ; preds = %bb8, %bb5
+  br label %bb18
+
+bb18:                                             ; preds = %bb17
+  %tmp19 = add nuw nsw i64 %j.0, 1
+  br label %bb4
+
+bb20:                                             ; preds = %bb4
+  br label %bb21
+
+bb21:                                             ; preds = %bb20
+  %tmp22 = add nuw nsw i64 %i.0, 1
+  br label %bb2
+
+bb23:                                             ; preds = %bb2
+  ret void
+}