[BPI] Look through bitcasts in calcZeroHeuristic

Constant hoisting may have hidden a constant behind a bitcast so that
it isn't folded into its users. However, this prevents BPI from
calculating some of its heuristics that are based upon constant
values. So, I've added a simple helper function to look through these
casts.

Differential Revision: https://reviews.llvm.org/D58166

llvm-svn: 354119

2 files changed, 110 insertions, 1 deletions

diff --git a/llvm/lib/Analysis/BranchProbabilityInfo.cpp b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
index e4de67b8b84..5eb95003f5d 100644
--- a/llvm/lib/Analysis/BranchProbabilityInfo.cpp
+++ b/llvm/lib/Analysis/BranchProbabilityInfo.cpp
@@ -660,8 +660,14 @@ bool BranchProbabilityInfo::calcZeroHeuristics(const BasicBlock *BB,
   if (!CI)
     return false;
 
+  auto GetConstantInt = [](Value *V) {
+    if (auto *I = dyn_cast<BitCastInst>(V))
+      return dyn_cast<ConstantInt>(I->getOperand(0));
+    return dyn_cast<ConstantInt>(V);
+  };
+
   Value *RHS = CI->getOperand(1);
-  ConstantInt *CV = dyn_cast<ConstantInt>(RHS);
+  ConstantInt *CV = GetConstantInt(RHS);
   if (!CV)
     return false;
 
diff --git a/llvm/test/Analysis/BranchProbabilityInfo/zero_heuristics.ll b/llvm/test/Analysis/BranchProbabilityInfo/zero_heuristics.ll
new file mode 100644
index 00000000000..c6e1cb8c265
--- /dev/null
+++ b/llvm/test/Analysis/BranchProbabilityInfo/zero_heuristics.ll
@@ -0,0 +1,103 @@
+; RUN: opt < %s -analyze -branch-prob | FileCheck %s
+
+@A = global i32 0, align 4
+@B = global i32 0, align 4
+
+; CHECK-LABEL: eq_opaque_minus_one
+define void @eq_opaque_minus_one(i32* %base) {
+entry:
+  %const = bitcast i32 -1 to i32
+  %tmp1 = load i32, i32* @B, align 4
+  br label %for.body
+
+; CHECK: edge for.body -> if.then probability is 0x30000000 / 0x80000000 = 37.50%
+; CHECK: edge for.body -> for.inc probability is 0x50000000 / 0x80000000 = 62.50%
+for.body:
+  %tmp4 = phi i32 [ %tmp1, %entry ], [ %tmp7, %for.inc ]
+  %inc.iv = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
+  %storemerge176.in = getelementptr inbounds i32, i32* %base, i32 %inc.iv
+  %storemerge176 = load i32, i32* %storemerge176.in, align 4
+  store i32 %storemerge176, i32* @A, align 4
+  %cmp20 = icmp eq i32 %storemerge176, %const
+  br i1 %cmp20, label %if.then, label %for.inc
+
+if.then:
+  %lnot.ext = zext i1 %cmp20 to i32
+  store i32 %lnot.ext, i32* @B, align 4
+  br label %for.inc
+
+for.inc:
+  %tmp7 = phi i32 [ %tmp4, %for.body ], [ %lnot.ext, %if.then ]
+  %inc = add nuw nsw i32 %inc.iv, 1
+  %cmp9 = icmp ult i32 %inc, 401
+  br i1 %cmp9, label %for.body, label %exit
+
+exit:
+  ret void
+}
+
+; CHECK-LABEL: ne_opaque_minus_one
+define void @ne_opaque_minus_one(i32* %base) {
+entry:
+  %const = bitcast i32 -1 to i32
+  %tmp1 = load i32, i32* @B, align 4
+  br label %for.body
+
+; CHECK: edge for.body -> if.then probability is 0x50000000 / 0x80000000 = 62.50%
+; CHECK: edge for.body -> for.inc probability is 0x30000000 / 0x80000000 = 37.50%
+for.body:
+  %tmp4 = phi i32 [ %tmp1, %entry ], [ %tmp7, %for.inc ]
+  %inc.iv = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
+  %storemerge176.in = getelementptr inbounds i32, i32* %base, i32 %inc.iv
+  %storemerge176 = load i32, i32* %storemerge176.in, align 4
+  store i32 %storemerge176, i32* @A, align 4
+  %cmp20 = icmp ne i32 %storemerge176, %const
+  br i1 %cmp20, label %if.then, label %for.inc
+
+if.then:
+  %lnot.ext = zext i1 %cmp20 to i32
+  store i32 %lnot.ext, i32* @B, align 4
+  br label %for.inc
+
+for.inc:
+  %tmp7 = phi i32 [ %tmp4, %for.body ], [ %lnot.ext, %if.then ]
+  %inc = add nuw nsw i32 %inc.iv, 1
+  %cmp9 = icmp ult i32 %inc, 401
+  br i1 %cmp9, label %for.body, label %exit
+
+exit:
+  ret void
+}
+
+; CHECK-LABEL: sgt_opaque_minus_one
+define void @sgt_opaque_minus_one(i32* %base) {
+entry:
+  %const = bitcast i32 -1 to i32
+  %tmp1 = load i32, i32* @B, align 4
+  br label %for.body
+
+; CHECK: edge for.body -> if.then probability is 0x50000000 / 0x80000000 = 62.50%
+; CHECK: edge for.body -> for.inc probability is 0x30000000 / 0x80000000 = 37.50%
+for.body:
+  %tmp4 = phi i32 [ %tmp1, %entry ], [ %tmp7, %for.inc ]
+  %inc.iv = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
+  %storemerge176.in = getelementptr inbounds i32, i32* %base, i32 %inc.iv
+  %storemerge176 = load i32, i32* %storemerge176.in, align 4
+  store i32 %storemerge176, i32* @A, align 4
+  %cmp20 = icmp sgt i32 %storemerge176, %const
+  br i1 %cmp20, label %if.then, label %for.inc
+
+if.then:
+  %lnot.ext = zext i1 %cmp20 to i32
+  store i32 %lnot.ext, i32* @B, align 4
+  br label %for.inc
+
+for.inc:
+  %tmp7 = phi i32 [ %tmp4, %for.body ], [ %lnot.ext, %if.then ]
+  %inc = add nuw nsw i32 %inc.iv, 1
+  %cmp9 = icmp ult i32 %inc, 401
+  br i1 %cmp9, label %for.body, label %exit
+
+exit:
+  ret void
+}