Enable Loop Sink pass for functions that has profile.

Summary: For functions with profile data, we are confident that loop sink will be optimal in sinking code.

Reviewers: davidxl, hfinkel

Subscribers: mehdi_amini, mzolotukhin, llvm-commits

Differential Revision: https://reviews.llvm.org/D26155

llvm-svn: 286325

lib/Transforms/IPO/PassManagerBuilder.cpp

@@ -626,10 +626,7 @@ void PassManagerBuilder::populateModulePassManager(
     // outer loop. LICM pass can help to promote the runtime check out if the
     // checked value is loop invariant.
     MPM.add(createLICMPass());
-
-    // Get rid of LCSSA nodes.
-    MPM.add(createInstructionSimplifierPass());
-  }
+ }
 
   // After vectorization and unrolling, assume intrinsics may tell us more
   // about pointer alignments.
@@ -650,6 +647,9 @@ void PassManagerBuilder::populateModulePassManager(
   if (MergeFunctions)
     MPM.add(createMergeFunctionsPass());
 
+  MPM.add(createLoopSinkPass());
+  // Get rid of LCSSA nodes.
+  MPM.add(createInstructionSimplifierPass());
   addExtensionsToPM(EP_OptimizerLast, MPM);
 }
 

lib/Transforms/Scalar/LoopSink.cpp

@@ -243,6 +243,11 @@ static bool sinkLoopInvariantInstructions(Loop &L, AAResults &AA, LoopInfo &LI,
   if (!Preheader)
     return false;
 
+  // Enable LoopSink only when runtime profile is available.
+  // With static profile, the sinking decision may be sub-optimal.
+  if (!Preheader->getParent()->getEntryCount())
+    return false;
+
   const BlockFrequency PreheaderFreq = BFI.getBlockFreq(Preheader);
   // If there are no basic blocks with lower frequency than the preheader then
   // we can avoid the detailed analysis as we will never find profitable sinking

test/Other/pass-pipelines.ll

@@ -82,6 +82,9 @@
 ; the runtime unrolling though.
 ; CHECK-O2: Loop Pass Manager
 ; CHECK-O2-NEXT: Loop Invariant Code Motion
+; CHECK-O2: FunctionPass Manager
+; CHECK-O2: Loop Pass Manager
+; CHECK-O2-NEXT: Loop Sink
 ; CHECK-O2-NOT: Manager
 ;
 ; FIXME: There really shouldn't be another pass manager, especially one that

test/Transforms/LICM/loopsink.ll

@@ -21,7 +21,7 @@
 ; CHECK: load i32, i32* @g
 ; CHECK: .b3:
 ; CHECK-NOT:  load i32, i32* @g
-define i32 @t1(i32, i32) #0 {
+define i32 @t1(i32, i32) #0 !prof !0 {
   %3 = icmp eq i32 %1, 0
   br i1 %3, label %.exit, label %.preheader
 
@@ -88,7 +88,7 @@ define i32 @t1(i32, i32) #0 {
 ; CHECK: .b6:
 ; CHECK: load i32, i32* @g
 ; CHECK: .b7:
-define i32 @t2(i32, i32) #0 {
+define i32 @t2(i32, i32) #0 !prof !0 {
   %3 = icmp eq i32 %1, 0
   br i1 %3, label %.exit, label %.preheader
 
@@ -150,7 +150,7 @@ define i32 @t2(i32, i32) #0 {
 ; CHECK: load i32, i32* @g
 ; CHECK: .b1:
 ; CHECK-NOT: load i32, i32* @g
-define i32 @t3(i32, i32) #0 {
+define i32 @t3(i32, i32) #0 !prof !0 {
   %3 = icmp eq i32 %1, 0
   br i1 %3, label %.exit, label %.preheader
 
@@ -201,7 +201,7 @@ define i32 @t3(i32, i32) #0 {
 ; CHECK: .b1:
 ; CHECK: load i32, i32* @g
 ; CHECK: .exit:
-define i32 @t4(i32, i32) #0 {
+define i32 @t4(i32, i32) #0 !prof !0 {
 .preheader:
   %invariant = load i32, i32* @g
   br label %.b1
@@ -235,7 +235,7 @@ define i32 @t4(i32, i32) #0 {
 ; CHECK: load i32, i32* @g
 ; CHECK: .b1:
 ; CHECK-NOT: load i32, i32* @g
-define i32 @t5(i32, i32*) #0 {
+define i32 @t5(i32, i32*) #0 !prof !0 {
   %3 = icmp eq i32 %0, 0
   br i1 %3, label %.exit, label %.preheader
 
@@ -281,6 +281,7 @@ define i32 @t5(i32, i32*) #0 {
 
 declare i32 @foo()
 
+!0 = !{!"function_entry_count", i64 1}
 !1 = !{!"branch_weights", i32 1, i32 2000}
 !2 = !{!"branch_weights", i32 2000, i32 1}
 !3 = !{!"branch_weights", i32 100, i32 1}

test/Transforms/LICM/sink.ll

@@ -15,7 +15,7 @@
 
 @g = global i32 0, align 4
 
-define i32 @foo(i32, i32) #0 {
+define i32 @foo(i32, i32) #0 !prof !2 {
   %3 = icmp eq i32 %1, 0
   br i1 %3, label %._crit_edge, label %.lr.ph.preheader
 
@@ -58,3 +58,4 @@ define i32 @foo(i32, i32) #0 {
 }
 
 !1 = !{!"branch_weights", i32 1, i32 2000}
+!2 = !{!"function_entry_count", i64 1}