Avoid inserting noop's in the middle of a loop.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@60141 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/LoopAligner.cpp b/lib/CodeGen/LoopAligner.cpp
index b8d0059..b67f5c3 100644 (file)
--- a/lib/CodeGen/LoopAligner.cpp
+++ b/lib/CodeGen/LoopAligner.cpp
@@ -64,8 +64,14 @@ bool LoopAligner::runOnMachineFunction(MachineFunction &MF) {
 
   for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
     MachineBasicBlock *MBB = I;
-    if (MLI->isLoopHeader(MBB))
+    if (MLI->isLoopHeader(MBB)) {
+      MachineBasicBlock *PredBB = prior(I);
+      if (MLI->getLoopFor(MBB) == MLI->getLoopFor(PredBB))
+        // If previously BB is in the same loop, don't align this BB. We want
+        // to prevent adding noop's inside a loop.
+        continue;
       MBB->setAlignment(Align);
+    }
   }
 
   return true;

diff --git a/test/CodeGen/X86/avoid-loop-align.ll b/test/CodeGen/X86/avoid-loop-align.ll
new file mode 100644 (file)
index 0000000..dfc5818
--- /dev/null
+++ b/test/CodeGen/X86/avoid-loop-align.ll
@@ -0,0 +1,32 @@
+; RUN: llvm-as < %s | llc -mtriple=i386-apple-darwin | grep align | count 1
+
+@A = common global [100 x i32] zeroinitializer, align 32               ; <[100 x i32]*> [#uses=1]
+
+define i8* @test(i8* %Q, i32* %L) nounwind {
+entry:
+       %tmp = tail call i32 (...)* @foo() nounwind             ; <i32> [#uses=2]
+       %tmp1 = inttoptr i32 %tmp to i8*                ; <i8*> [#uses=1]
+       br label %bb1
+
+bb:            ; preds = %bb1, %bb1
+       %indvar.next = add i32 %P.0.rec, 1              ; <i32> [#uses=1]
+       br label %bb1
+
+bb1:           ; preds = %bb, %entry
+       %P.0.rec = phi i32 [ 0, %entry ], [ %indvar.next, %bb ]         ; <i32> [#uses=2]
+       %P.0 = getelementptr i8* %tmp1, i32 %P.0.rec            ; <i8*> [#uses=3]
+       %tmp2 = load i8* %P.0, align 1          ; <i8> [#uses=1]
+       switch i8 %tmp2, label %bb4 [
+               i8 12, label %bb
+               i8 42, label %bb
+       ]
+
+bb4:           ; preds = %bb1
+       %tmp3 = ptrtoint i8* %P.0 to i32                ; <i32> [#uses=1]
+       %tmp4 = sub i32 %tmp3, %tmp             ; <i32> [#uses=1]
+       %tmp5 = getelementptr [100 x i32]* @A, i32 0, i32 %tmp4         ; <i32*> [#uses=1]
+       store i32 4, i32* %tmp5, align 4
+       ret i8* %P.0
+}
+
+declare i32 @foo(...)