When the inliner merges allocas, it must keep the larger alignment

For safety, the inliner cannot decrease the allignment on an alloca when
merging it with another.

I've included two variants of the test case for this: one with DataLayout
available, and one without. When DataLayout is not available, if only one of
the allocas uses the default alignment (getAlignment() == 0), then they cannot
be safely merged.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186425 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 203 insertions, 2 deletions

diff --git a/lib/Transforms/IPO/Inliner.cpp b/lib/Transforms/IPO/Inliner.cpp
index 663ddb7..f72121d 100644
--- a/lib/Transforms/IPO/Inliner.cpp
+++ b/lib/Transforms/IPO/Inliner.cpp
@@ -116,7 +116,8 @@ static void AdjustCallerSSPLevel(Function *Caller, Function *Callee) {
 /// any new allocas to the set if not possible.
 static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI,
                                  InlinedArrayAllocasTy &InlinedArrayAllocas,
-                                 int InlineHistory, bool InsertLifetime) {
+                                 int InlineHistory, bool InsertLifetime,
+                                 const DataLayout *TD) {
   Function *Callee = CS.getCalledFunction();
   Function *Caller = CS.getCaller();
 
@@ -189,6 +190,14 @@ static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI,
     bool MergedAwayAlloca = false;
     for (unsigned i = 0, e = AllocasForType.size(); i != e; ++i) {
       AllocaInst *AvailableAlloca = AllocasForType[i];
+
+      unsigned Align1 = AI->getAlignment(),
+               Align2 = AvailableAlloca->getAlignment();
+      // If we don't have data layout information, and only one alloca is using
+      // the target default, then we can't safely merge them because we can't
+      // pick the greater alignment.
+      if (!TD && (!Align1 || !Align2) && Align1 != Align2)
+        continue;
       
       // The available alloca has to be in the right function, not in some other
       // function in this SCC.
@@ -206,6 +215,11 @@ static bool InlineCallIfPossible(CallSite CS, InlineFunctionInfo &IFI,
                    << *AvailableAlloca << '\n');
       
       AI->replaceAllUsesWith(AvailableAlloca);
+
+      if (Align1 > Align2 || (!Align1 && TD &&
+          TD->getABITypeAlignment(AI->getAllocatedType()) > Align2))
+        AvailableAlloca->setAlignment(Align1);
+
       AI->eraseFromParent();
       MergedAwayAlloca = true;
       ++NumMergedAllocas;
@@ -482,7 +496,7 @@ bool Inliner::runOnSCC(CallGraphSCC &SCC) {
 
         // Attempt to inline the function.
         if (!InlineCallIfPossible(CS, InlineInfo, InlinedArrayAllocas,
-                                  InlineHistoryID, InsertLifetime))
+                                  InlineHistoryID, InsertLifetime, TD))
           continue;
         ++NumInlined;
         
diff --git a/test/Transforms/Inline/alloca-merge-align-nodl.ll b/test/Transforms/Inline/alloca-merge-align-nodl.ll
new file mode 100644
index 0000000..203f52b
--- /dev/null
+++ b/test/Transforms/Inline/alloca-merge-align-nodl.ll
@@ -0,0 +1,93 @@
+; RUN: opt < %s -inline -S | FileCheck %s
+; This variant of the test has no data layout information.
+target triple = "powerpc64-unknown-linux-gnu"
+
+%struct.s = type { i32, i32 }
+
+define void @foo(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32], align 4
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 4, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+define void @foo0(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32]
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 4, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+declare void @bar(i32*) #1
+
+define void @goo(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32], align 32
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 32, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+; CHECK-LABEL: @main
+; CHECK: alloca [2 x i32], align 32
+; CHECK-NOT: alloca [2 x i32]
+; CHECK: ret i32 0
+
+define signext i32 @main() {
+entry:
+  %a = alloca i64, align 8
+  %tmpcast = bitcast i64* %a to %struct.s*
+  store i64 0, i64* %a, align 8
+  %a1 = bitcast i64* %a to i32*
+  store i32 1, i32* %a1, align 8, !tbaa !0
+  call void @foo(%struct.s* byval %tmpcast)
+  store i32 2, i32* %a1, align 8, !tbaa !0
+  call void @goo(%struct.s* byval %tmpcast)
+  ret i32 0
+}
+
+; CHECK-LABEL: @test0
+; CHECK: alloca [2 x i32], align 32
+; CHECK: alloca [2 x i32]
+; CHECK: ret i32 0
+
+define signext i32 @test0() {
+entry:
+  %a = alloca i64, align 8
+  %tmpcast = bitcast i64* %a to %struct.s*
+  store i64 0, i64* %a, align 8
+  %a1 = bitcast i64* %a to i32*
+  store i32 1, i32* %a1, align 8, !tbaa !0
+  call void @foo0(%struct.s* byval %tmpcast)
+  store i32 2, i32* %a1, align 8, !tbaa !0
+  call void @goo(%struct.s* byval %tmpcast)
+  ret i32 0
+}
+
+!0 = metadata !{metadata !"int", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}
+
diff --git a/test/Transforms/Inline/alloca-merge-align.ll b/test/Transforms/Inline/alloca-merge-align.ll
new file mode 100644
index 0000000..de4707d
--- /dev/null
+++ b/test/Transforms/Inline/alloca-merge-align.ll
@@ -0,0 +1,94 @@
+; RUN: opt < %s -inline -S | FileCheck %s
+
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+%struct.s = type { i32, i32 }
+
+define void @foo(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32], align 4
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 4, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+define void @foo0(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32]
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 4, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+declare void @bar(i32*) #1
+
+define void @goo(%struct.s* byval nocapture readonly %a) {
+entry:
+  %x = alloca [2 x i32], align 32
+  %a1 = getelementptr inbounds %struct.s* %a, i64 0, i32 0
+  %0 = load i32* %a1, align 4, !tbaa !0
+  %arrayidx = getelementptr inbounds [2 x i32]* %x, i64 0, i64 0
+  store i32 %0, i32* %arrayidx, align 32, !tbaa !0
+  %b = getelementptr inbounds %struct.s* %a, i64 0, i32 1
+  %1 = load i32* %b, align 4, !tbaa !0
+  %arrayidx2 = getelementptr inbounds [2 x i32]* %x, i64 0, i64 1
+  store i32 %1, i32* %arrayidx2, align 4, !tbaa !0
+  call void @bar(i32* %arrayidx) #2
+  ret void
+}
+
+; CHECK-LABEL: @main
+; CHECK: alloca [2 x i32], align 32
+; CHECK-NOT: alloca [2 x i32]
+; CHECK: ret i32 0
+
+define signext i32 @main() {
+entry:
+  %a = alloca i64, align 8
+  %tmpcast = bitcast i64* %a to %struct.s*
+  store i64 0, i64* %a, align 8
+  %a1 = bitcast i64* %a to i32*
+  store i32 1, i32* %a1, align 8, !tbaa !0
+  call void @foo(%struct.s* byval %tmpcast)
+  store i32 2, i32* %a1, align 8, !tbaa !0
+  call void @goo(%struct.s* byval %tmpcast)
+  ret i32 0
+}
+
+; CHECK-LABEL: @test0
+; CHECK: alloca [2 x i32], align 32
+; CHECK-NOT: alloca [2 x i32]
+; CHECK: ret i32 0
+
+define signext i32 @test0() {
+entry:
+  %a = alloca i64, align 8
+  %tmpcast = bitcast i64* %a to %struct.s*
+  store i64 0, i64* %a, align 8
+  %a1 = bitcast i64* %a to i32*
+  store i32 1, i32* %a1, align 8, !tbaa !0
+  call void @foo0(%struct.s* byval %tmpcast)
+  store i32 2, i32* %a1, align 8, !tbaa !0
+  call void @goo(%struct.s* byval %tmpcast)
+  ret i32 0
+}
+
+!0 = metadata !{metadata !"int", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA"}
+