Index: lib/Transform/ScopInliner.cpp
===================================================================
--- lib/Transform/ScopInliner.cpp
+++ lib/Transform/ScopInliner.cpp
@@ -59,8 +59,13 @@
     // If the function is a nullptr, or the function is a declaration.
     if (!F)
       return false;
+
+    // HACK: remove this before upstreaming
+    if (F->getName().count("__radiation_rg_MOD_coe") == 0 &&
+            F->getName().count("__radiation_rg_MOD_inv") == 0) return false;
+
     if (F->isDeclaration()) {
-      DEBUG(dbgs() << "Skipping " << F->getName()
+      DEBUG(dbgs() << "SKIPPING " << F->getName()
                    << "because it is a declaration.\n");
       return false;
     }
@@ -73,16 +78,36 @@
     RegionInfo &RI = FAM.getResult<RegionInfoAnalysis>(*F);
     ScopDetection &SD = FAM.getResult<ScopAnalysis>(*F);
 
-    const bool HasScopAsTopLevelRegion =
-        SD.ValidRegions.count(RI.getTopLevelRegion()) > 0;
+    const auto TopLevelRegion = RI.getTopLevelRegion();
+
+    // Whether the entire function can be modeled as a Scop.
+    const bool IsFullyModeledAsScop =
+        SD.ValidRegions.count(TopLevelRegion) > 0;
+
+    // Whether the function has a Scop that is a *unique* child of the top-level
+    // region.
+    const bool IsModeledByTopLevelChild = [&] {
+        // If toplevel has more than 1 child, bail out.
+        if (std::distance(TopLevelRegion->begin(), TopLevelRegion->end()) > 1)
+            return false;
 
-    if (HasScopAsTopLevelRegion) {
-      DEBUG(dbgs() << "Skipping " << F->getName()
-                   << " has scop as top level region");
+        for (auto ScopRegion : SD.ValidRegions) {
+            if (ScopRegion->getParent() == TopLevelRegion) {
+                return true;
+            }
+        }
+        return false;
+    }();
+
+    if (IsFullyModeledAsScop || IsModeledByTopLevelChild) {
+      FAM.clear(*F);
+
+      DEBUG(dbgs() << "@ " << F->getName() << " DOES have scop as top level region.\n");
       F->addFnAttr(llvm::Attribute::AlwaysInline);
 
       ModuleAnalysisManager MAM;
       PB.registerModuleAnalyses(MAM);
+
       ModulePassManager MPM;
       MPM.addPass(AlwaysInlinerPass());
       Module *M = F->getParent();
@@ -90,7 +115,7 @@
       MPM.run(*M, MAM);
     } else {
       DEBUG(dbgs() << F->getName()
-                   << " does NOT have scop as top level region\n");
+                   << " does NOT have scop as top level region.\n");
     }
 
     return false;
Index: test/ScopInliner/scop-from-entry-successor.ll
===================================================================
--- /dev/null
+++ test/ScopInliner/scop-from-entry-successor.ll
@@ -0,0 +1,66 @@
+; RUN: opt %loadPolly -polly-detect-full-functions -polly-scop-inliner \
+; RUN: -polly-scops -analyze < %s | FileCheck %s
+
+; Check that we get the 2 nested loops by inlining `to_be_inlined` into
+; `inline_site`, when the `to_be_inlined` has a `Scop` from the *successor* of
+; the entry block to the exit node.
+
+; CHECK:    Max Loop Depth:  2
+
+; static const int N = 1000;
+;
+; void to_be_inlined(int A[]) {
+;     int B; // we do not allow dead code elimination on purpose to have an alloca.
+;     for(int i = 0; i < N; i++)
+;         A[i] *= 10;
+; }
+;
+; void inline_site(int A[]) {
+;     for(int i = 0; i < N; i++)
+;         to_be_inlined(A);
+; }
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.12.0"
+
+
+define void @to_be_inlined(i32* %A) {
+entry:
+  %alloc_to_block_modelling = alloca i32
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  br label %for.body
+
+for.body:                                         ; preds = %entry.split, %for.body
+  %indvars.iv1 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv1
+  %tmp = load i32, i32* %arrayidx, align 4
+  %mul = mul nsw i32 %tmp, 10
+  store i32 %mul, i32* %arrayidx, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv1, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+define void @inline_site(i32* %A) {
+entry:
+  br label %entry.split
+
+entry.split:                                      ; preds = %entry
+  br label %for.body
+
+for.body:                                         ; preds = %entry.split, %for.body
+  %i.01 = phi i32 [ 0, %entry.split ], [ %inc, %for.body ]
+  tail call void @to_be_inlined(i32* %A)
+  %inc = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %inc, 1000
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+