[BranchFolding] Set correct mem refs

[oota-llvm.git] / lib / CodeGen / BranchFolding.cpp

diff --git a/lib/CodeGen/BranchFolding.cpp b/lib/CodeGen/BranchFolding.cpp
index 0b2495cc996e00b409c8a1be88a8d70b8f6391e2..df5cac5a9f7abb17a7fddd93cb64b39daf9145b4 100644 (file)
--- a/lib/CodeGen/BranchFolding.cpp
+++ b/lib/CodeGen/BranchFolding.cpp
@@ -371,7 +371,7 @@ static unsigned ComputeCommonTailLength(MachineBasicBlock *MBB1,
   }
   // Back past possible debugging pseudos at beginning of block.  This matters
   // when one block differs from the other only by whether debugging pseudos
-  // are present at the beginning.  (This way, the various checks later for
+  // are present at the beginning. (This way, the various checks later for
   // I1==MBB1->begin() work as expected.)
   if (I1 == MBB1->begin() && I2 != MBB2->begin()) {
     --I2;
@@ -744,18 +744,6 @@ bool BranchFolder::CreateCommonTailOnlyBlock(MachineBasicBlock *&PredBB,
   return true;
 }
 
-static bool hasIdenticalMMOs(const MachineInstr *MI1, const MachineInstr *MI2) {
-  auto I1 = MI1->memoperands_begin(), E1 = MI1->memoperands_end();
-  auto I2 = MI2->memoperands_begin(), E2 = MI2->memoperands_end();
-  if ((E1 - I1) != (E2 - I2))
-    return false;
-  for (; I1 != E1; ++I1, ++I2) {
-    if (**I1 != **I2)
-      return false;
-  }
-  return true;
-}
-
 static void
 removeMMOsFromMemoryOperations(MachineBasicBlock::iterator MBBIStartPos,
                                MachineBasicBlock &MBBCommon) {
@@ -792,8 +780,7 @@ removeMMOsFromMemoryOperations(MachineBasicBlock::iterator MBBIStartPos,
     assert(MBBICommon->isIdenticalTo(&*MBBI) && "Expected matching MIIs!");
 
     if (MBBICommon->mayLoad() || MBBICommon->mayStore())
-      if (!hasIdenticalMMOs(&*MBBI, &*MBBICommon))
-        MBBICommon->clearMemRefs();
+      MBBICommon->setMemRefs(MBBICommon->mergeMemRefsWith(*MBBI));
 
     ++MBBI;
     ++MBBICommon;
@@ -1099,13 +1086,19 @@ void BranchFolder::setCommonTailEdgeWeights(MachineBasicBlock &TailMBB) {
   if (TailMBB.succ_size() <= 1)
     return;
 
-  auto MaxEdgeFreq = *std::max_element(EdgeFreqLs.begin(), EdgeFreqLs.end());
-  uint64_t Scale = MaxEdgeFreq.getFrequency() / UINT32_MAX + 1;
+  auto SumEdgeFreq =
+      std::accumulate(EdgeFreqLs.begin(), EdgeFreqLs.end(), BlockFrequency(0))
+          .getFrequency();
   auto EdgeFreq = EdgeFreqLs.begin();
 
-  for (auto SuccI = TailMBB.succ_begin(), SuccE = TailMBB.succ_end();
-       SuccI != SuccE; ++SuccI, ++EdgeFreq)
-    TailMBB.setSuccWeight(SuccI, EdgeFreq->getFrequency() / Scale);
+  if (SumEdgeFreq > 0) {
+    for (auto SuccI = TailMBB.succ_begin(), SuccE = TailMBB.succ_end();
+         SuccI != SuccE; ++SuccI, ++EdgeFreq) {
+      auto Prob = BranchProbability::getBranchProbability(
+          EdgeFreq->getFrequency(), SumEdgeFreq);
+      TailMBB.setSuccProbability(SuccI, Prob);
+    }
+  }
 }
 
 //===----------------------------------------------------------------------===//
@@ -1558,6 +1551,14 @@ ReoptimizeBlock:
       // removed, move this block to the end of the function.
       MachineBasicBlock *PrevTBB = nullptr, *PrevFBB = nullptr;
       SmallVector<MachineOperand, 4> PrevCond;
+      // We're looking for cases where PrevBB could possibly fall through to
+      // FallThrough, but if FallThrough is an EH pad that wouldn't be useful
+      // so here we skip over any EH pads so we might have a chance to find
+      // a branch target from PrevBB.
+      while (FallThrough != MF.end() && FallThrough->isEHPad())
+        ++FallThrough;
+      // Now check to see if the current block is sitting between PrevBB and
+      // a block to which it could fall through.
       if (FallThrough != MF.end() &&
           !TII->AnalyzeBranch(PrevBB, PrevTBB, PrevFBB, PrevCond, true) &&
           PrevBB.isSuccessor(&*FallThrough)) {