[BPI] Apply invoke heuristic before loop branch heuristic

Currently the loop branch heuristic is applied before the invoke heuristic which makes us overestimate the probability of the unwind destination of invokes inside loops. This in turn makes us grossly underestimate the frequencies of loops with invokes.

Reviewed By: skatkov, vsk

Differential Revision: https://reviews.llvm.org/D47371

llvm-svn: 334285

llvm/lib/Analysis/BranchProbabilityInfo.cpp

@@ -203,12 +203,10 @@ BranchProbabilityInfo::updatePostDominatedByColdCall(const BasicBlock *BB) {
 /// unreachable-terminated block as extremely unlikely.
 bool BranchProbabilityInfo::calcUnreachableHeuristics(const BasicBlock *BB) {
   const TerminatorInst *TI = BB->getTerminator();
+  (void) TI;
   assert(TI->getNumSuccessors() > 1 && "expected more than one successor!");
-
-  // Return false here so that edge weights for InvokeInst could be decided
-  // in calcInvokeHeuristics().
-  if (isa<InvokeInst>(TI))
-    return false;
+  assert(!isa<InvokeInst>(TI) &&
+         "Invokes should have already been handled by calcInvokeHeuristics");
 
   SmallVector<unsigned, 4> UnreachableEdges;
   SmallVector<unsigned, 4> ReachableEdges;
@@ -351,12 +349,10 @@ bool BranchProbabilityInfo::calcMetadataWeights(const BasicBlock *BB) {
 /// Return false, otherwise.
 bool BranchProbabilityInfo::calcColdCallHeuristics(const BasicBlock *BB) {
   const TerminatorInst *TI = BB->getTerminator();
+  (void) TI;
   assert(TI->getNumSuccessors() > 1 && "expected more than one successor!");
-
-  // Return false here so that edge weights for InvokeInst could be decided
-  // in calcInvokeHeuristics().
-  if (isa<InvokeInst>(TI))
-    return false;
+  assert(!isa<InvokeInst>(TI) &&
+         "Invokes should have already been handled by calcInvokeHeuristics");
 
   // Determine which successors are post-dominated by a cold block.
   SmallVector<unsigned, 4> ColdEdges;
@@ -975,6 +971,8 @@ void BranchProbabilityInfo::calculate(const Function &F, const LoopInfo &LI,
       continue;
     if (calcMetadataWeights(BB))
       continue;
+    if (calcInvokeHeuristics(BB))
+      continue;
     if (calcUnreachableHeuristics(BB))
       continue;
     if (calcColdCallHeuristics(BB))
@@ -987,7 +985,6 @@ void BranchProbabilityInfo::calculate(const Function &F, const LoopInfo &LI,
       continue;
     if (calcFloatingPointHeuristics(BB))
       continue;
-    calcInvokeHeuristics(BB);
   }
 
   PostDominatedByUnreachable.clear();

llvm/test/Analysis/BlockFrequencyInfo/loop_with_invoke.ll

@@ -0,0 +1,35 @@
+; RUN: opt < %s -analyze -block-freq | FileCheck %s
+; RUN: opt < %s -passes='print<block-freq>' -disable-output 2>&1 | FileCheck %s
+
+; CHECK-LABEL: Printing analysis {{.*}} for function 'loop_with_invoke':
+; CHECK-NEXT: block-frequency-info: loop_with_invoke
+define void @loop_with_invoke(i32 %n) personality i8 0 {
+; CHECK-NEXT: entry: float = 1.0, int = [[ENTRY:[0-9]+]]
+entry:
+  br label %loop
+
+; CHECK-NEXT: loop: float = 9905.6
+loop:
+  %i = phi i32 [ 0, %entry ], [ %i.next, %invoke.cont ]
+  invoke void @foo() to label %invoke.cont unwind label %lpad
+
+; CHECK-NEXT: invoke.cont: float = 9905.6
+invoke.cont:
+  %i.next = add i32 %i, 1
+  %cont = icmp ult i32 %i.next, %n
+  br i1 %cont, label %loop, label %exit, !prof !0
+
+; CHECK-NEXT: lpad: float = 0.0094467
+lpad:
+  %ll = landingpad { i8*, i32 }
+          cleanup
+  br label %exit
+
+; CHECK-NEXT: exit: float = 1.0, int = [[ENTRY]]
+exit:
+  ret void
+}
+
+declare void @foo()
+
+!0 = !{!"branch_weights", i32 9999, i32 1}

llvm/test/Analysis/BranchProbabilityInfo/loop.ll

@@ -489,3 +489,35 @@ for.inc:
   br label %for.cond
 ; CHECK: edge for.inc -> for.cond probability is 0x80000000 / 0x80000000 = 100.00% [HOT edge]
 }
+
+; The loop heuristic should not overwrite the invoke heuristic. The unwind destination
+; of an invoke should be considered VERY rare even in a loop.
+define void @test12(i32 %a) personality i8 0 {
+entry:
+  br label %loop
+; CHECK: edge entry -> loop probability is 0x80000000 / 0x80000000 = 100.00% [HOT edge]
+
+loop:
+  %i.0 = phi i32 [ 0, %entry ], [ %inc, %invoke.cont ]
+  invoke i32 @InvokeCall()
+          to label %invoke.cont unwind label %lpad
+; CHECK: edge loop -> invoke.cont probability is 0x7ffff800 / 0x80000000 = 100.00% [HOT edge]
+; CHECK: edge loop -> lpad probability is 0x00000800 / 0x80000000 = 0.00%
+
+invoke.cont:
+  %inc = add nsw i32 %i.0, 1
+  %cmp = icmp slt i32 %inc, %a
+  br i1 %cmp, label %loop, label %exit
+; CHECK: edge invoke.cont -> loop probability is 0x7c000000 / 0x80000000 = 96.88% [HOT edge]
+; CHECK: edge invoke.cont -> exit probability is 0x04000000 / 0x80000000 = 3.12%
+
+lpad:
+  %ll = landingpad { i8*, i32 }
+          cleanup
+  br label %exit
+
+exit:
+  ret void
+}
+
+declare i32 @InvokeCall()