[ConstHoisting] choose to hoist when frequency is the same.

The patch is to adjust the strategy of frequency based consthoisting:
Previously when the candidate block has the same frequency with the existing
blocks containing a const, it will not hoist the const to the candidate block.
For that case, now we change the strategy to hoist the const if only existing
blocks have more than one block member. This is helpful for reducing code size.

Differential Revision: https://reviews.llvm.org/D35084

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@307328 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/Support/BlockFrequency.h b/include/llvm/Support/BlockFrequency.h
index 1b45cc52973f65890e2ad1288dede177144d23e8..2e75cbdd29c1613dd6616b81f652386d69955586 100644 (file)
--- a/include/llvm/Support/BlockFrequency.h
+++ b/include/llvm/Support/BlockFrequency.h
@@ -71,6 +71,10 @@ public:
   bool operator>=(BlockFrequency RHS) const {
     return Frequency >= RHS.Frequency;
   }
+
+  bool operator==(BlockFrequency RHS) const {
+    return Frequency == RHS.Frequency;
+  }
 };
 
 }

diff --git a/lib/Transforms/Scalar/ConstantHoisting.cpp b/lib/Transforms/Scalar/ConstantHoisting.cpp
index a49c9b68c97d0a80997ce6308843576ec67b0869..5f4a785b627bef483b2ab6f070a046013f27b75d 100644 (file)
--- a/lib/Transforms/Scalar/ConstantHoisting.cpp
+++ b/lib/Transforms/Scalar/ConstantHoisting.cpp
@@ -231,7 +231,8 @@ static void findBestInsertionSet(DominatorTree &DT, BlockFrequencyInfo &BFI,
     // Return the optimal insert points in BBs.
     if (Node == Entry) {
       BBs.clear();
-      if (InsertPtsFreq > BFI.getBlockFreq(Node))
+      if (InsertPtsFreq > BFI.getBlockFreq(Node) ||
+          (InsertPtsFreq == BFI.getBlockFreq(Node) && InsertPts.size() > 1))
         BBs.insert(Entry);
       else
         BBs.insert(InsertPts.begin(), InsertPts.end());
@@ -244,7 +245,15 @@ static void findBestInsertionSet(DominatorTree &DT, BlockFrequencyInfo &BFI,
     SmallPtrSet<BasicBlock *, 16> &ParentInsertPts = InsertPtsMap[Parent].first;
     BlockFrequency &ParentPtsFreq = InsertPtsMap[Parent].second;
     // Choose to insert in Node or in subtree of Node.
-    if (InsertPtsFreq > BFI.getBlockFreq(Node) || NodeInBBs) {
+    // Don't hoist to EHPad because we may not find a proper place to insert
+    // in EHPad.
+    // If the total frequency of InsertPts is the same as the frequency of the
+    // target Node, and InsertPts contains more than one nodes, choose hoisting
+    // to reduce code size.
+    if (NodeInBBs ||
+        (!Node->isEHPad() &&
+         (InsertPtsFreq > BFI.getBlockFreq(Node) ||
+          (InsertPtsFreq == BFI.getBlockFreq(Node) && InsertPts.size() > 1)))) {
       ParentInsertPts.insert(Node);
       ParentPtsFreq += BFI.getBlockFreq(Node);
     } else {

diff --git a/test/CodeGen/X86/constant-hoisting-bfi.ll b/test/CodeGen/X86/constant-hoisting-bfi.ll
index 83589b7706f756d330d10f04faaf51ac45fb9e6f..d73f7163fd87bc3448ac324cee134deeaab46cb9 100644 (file)
--- a/test/CodeGen/X86/constant-hoisting-bfi.ll
+++ b/test/CodeGen/X86/constant-hoisting-bfi.ll
@@ -4,13 +4,13 @@ target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 
 ; Check when BFI is enabled for constant hoisting, constant 214748364701
 ; will not be hoisted to the func entry.
-; CHECK-LABEL: @foo(
+; CHECK-LABEL: @test1(
 ; CHECK: entry:
 ; CHECK-NOT: bitcast i64 214748364701 to i64
 ; CHECK: if.then:
 
 ; Function Attrs: norecurse nounwind uwtable
-define i64 @foo(i64* nocapture %a) {
+define i64 @test1(i64* nocapture %a) {
 entry:
   %arrayidx = getelementptr inbounds i64, i64* %a, i64 9
   %t0 = load i64, i64* %arrayidx, align 8
@@ -52,7 +52,7 @@ return:                                           ; preds = %if.else5, %if.then,
 ; in while.body will be hoisted to while.body.preheader. 214748364701 in
 ; if.then16 and if.else10 will be merged and hoisted to the beginning of
 ; if.else10 because if.else10 dominates if.then16.
-; CHECK-LABEL: @goo(
+; CHECK-LABEL: @test2(
 ; CHECK: entry:
 ; CHECK-NOT: bitcast i64 214748364701 to i64
 ; CHECK: while.body.preheader:
@@ -61,7 +61,7 @@ return:                                           ; preds = %if.else5, %if.then,
 ; CHECK: if.else10:
 ; CHECK-NEXT: bitcast i64 214748364701 to i64
 ; CHECK-NOT: bitcast i64 214748364701 to i64
-define i64 @goo(i64* nocapture %a) {
+define i64 @test2(i64* nocapture %a) {
 entry:
   %arrayidx = getelementptr inbounds i64, i64* %a, i64 9
   %t0 = load i64, i64* %arrayidx, align 8
@@ -113,3 +113,47 @@ return:                                           ; preds = %while.cond.preheade
 }
 
 !0 = !{!"branch_weights", i32 1, i32 2000}
+
+; 214748364701 will be hoisted to entry block to reduce code size.
+; CHECK-LABEL: @test3(
+; CHECK: entry:
+; CHECK-NEXT: %const = bitcast i64 214748364701 to i64
+define i64 @test3(i64 %t0) {
+entry:
+  %cmp = icmp ult i64 %t0, 56
+  br i1 %cmp, label %if.then, label %if.else
+
+; CHECK: if.then:
+; CHECK-NOT: %const = bitcast i64 214748364701 to i64
+if.then:
+  %add1 = add i64 %t0, 214748364701
+  br label %return
+
+; CHECK: if.else:
+; CHECK-NOT: %const = bitcast i64 214748364701 to i64
+if.else:
+  %add2 = add i64 %t0, 214748364701
+  br label %return
+
+return:
+  %retval = phi i64 [ %add1, %if.then ], [ %add2, %if.else ]
+  ret i64 %retval
+}
+
+; 214748364701 will not be hoisted to entry block because it will only
+; increase its live range.
+; CHECK-LABEL: @test4(
+; CHECK: nextblock:
+; CHECK-NEXT: %add1 = add i64 %t0, 214748364701
+define i64 @test4(i64 %t0) {
+entry:
+  %cmp = icmp ult i64 %t0, 56
+  br label %nextblock
+
+nextblock:
+  %add1 = add i64 %t0, 214748364701
+  br label %return
+
+return:
+  ret i64 %add1
+}