R600: Calculate store mask instead of using switch.

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

diff --git a/lib/CodeGen/MachineSink.cpp b/lib/CodeGen/MachineSink.cpp
index b117f8c3a206c9982e14fc79c41a638cea57fd4e..0a74f11b561ee491ff375a0900c6561acd2e02f9 100644 (file)
--- a/lib/CodeGen/MachineSink.cpp
+++ b/lib/CodeGen/MachineSink.cpp
@@ -18,18 +18,18 @@
 
 #define DEBUG_TYPE "machine-sink"
 #include "llvm/CodeGen/Passes.h"
-#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/CodeGen/MachineDominators.h"
-#include "llvm/CodeGen/MachineLoopInfo.h"
-#include "llvm/Analysis/AliasAnalysis.h"
-#include "llvm/Target/TargetRegisterInfo.h"
-#include "llvm/Target/TargetInstrInfo.h"
-#include "llvm/Target/TargetMachine.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetRegisterInfo.h"
 using namespace llvm;
 
 static cl::opt<bool>
@@ -60,9 +60,9 @@ namespace {
       initializeMachineSinkingPass(*PassRegistry::getPassRegistry());
     }
 
-    virtual bool runOnMachineFunction(MachineFunction &MF);
+    bool runOnMachineFunction(MachineFunction &MF) override;
 
-    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    void getAnalysisUsage(AnalysisUsage &AU) const override {
       AU.setPreservesCFG();
       MachineFunctionPass::getAnalysisUsage(AU);
       AU.addRequired<AliasAnalysis>();
@@ -72,7 +72,7 @@ namespace {
       AU.addPreserved<MachineLoopInfo>();
     }
 
-    virtual void releaseMemory() {
+    void releaseMemory() override {
       CEBCandidates.clear();
     }
 
@@ -98,16 +98,6 @@ namespace {
     bool PerformTrivialForwardCoalescing(MachineInstr *MI,
                                          MachineBasicBlock *MBB);
   };
-
-  // SuccessorSorter - Sort Successors according to their loop depth. 
-  struct SuccessorSorter {
-    SuccessorSorter(MachineLoopInfo *LoopInfo) : LI(LoopInfo) {}
-    bool operator()(const MachineBasicBlock *LHS,
-                    const MachineBasicBlock *RHS) const {
-      return LI->getLoopDepth(LHS) < LI->getLoopDepth(RHS);
-    }
-    MachineLoopInfo *LI;
-  };
 } // end anonymous namespace
 
 char MachineSinking::ID = 0;
@@ -308,12 +298,29 @@ bool MachineSinking::isWorthBreakingCriticalEdge(MachineInstr *MI,
   // to be sunk then it's probably worth it.
   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
     const MachineOperand &MO = MI->getOperand(i);
-    if (!MO.isReg()) continue;
+    if (!MO.isReg() || !MO.isUse())
+      continue;
     unsigned Reg = MO.getReg();
-    if (Reg == 0 || !TargetRegisterInfo::isPhysicalRegister(Reg))
+    if (Reg == 0)
       continue;
-    if (MRI->hasOneNonDBGUse(Reg))
-      return true;
+
+    // We don't move live definitions of physical registers,
+    // so sinking their uses won't enable any opportunities.
+    if (TargetRegisterInfo::isPhysicalRegister(Reg))
+      continue;
+
+    // If this instruction is the only user of a virtual register,
+    // check if breaking the edge will enable sinking
+    // both this instruction and the defining instruction.
+    if (MRI->hasOneNonDBGUse(Reg)) {
+      // If the definition resides in same MBB,
+      // claim it's likely we can sink these together.
+      // If definition resides elsewhere, we aren't
+      // blocking it from being sunk so don't break the edge.
+      MachineInstr *DefMI = MRI->getVRegDef(Reg);
+      if (DefMI->getParent() == MI->getParent())
+        return true;
+    }
   }
 
   return false;
@@ -394,7 +401,7 @@ static bool AvoidsSinking(MachineInstr *MI, MachineRegisterInfo *MRI) {
 /// collectDebgValues - Scan instructions following MI and collect any
 /// matching DBG_VALUEs.
 static void collectDebugValues(MachineInstr *MI,
-                               SmallVector<MachineInstr *, 2> & DbgValues) {
+                               SmallVectorImpl<MachineInstr *> &DbgValues) {
   DbgValues.clear();
   if (!MI->getOperand(0).isReg())
     return;
@@ -536,9 +543,14 @@ MachineBasicBlock *MachineSinking::FindSuccToSinkTo(MachineInstr *MI,
       // we should sink to.
       // We give successors with smaller loop depth higher priority.
       SmallVector<MachineBasicBlock*, 4> Succs(MBB->succ_begin(), MBB->succ_end());
-      std::stable_sort(Succs.begin(), Succs.end(), SuccessorSorter(LI));
-      for (SmallVector<MachineBasicBlock*, 4>::iterator SI = Succs.begin(),
-           E = Succs.end(); SI != E; ++SI) {
+      // Sort Successors according to their loop depth.
+      std::stable_sort(
+          Succs.begin(), Succs.end(),
+          [this](const MachineBasicBlock *LHS, const MachineBasicBlock *RHS) {
+            return LI->getLoopDepth(LHS) < LI->getLoopDepth(RHS);
+          });
+      for (SmallVectorImpl<MachineBasicBlock *>::iterator SI = Succs.begin(),
+             E = Succs.end(); SI != E; ++SI) {
         MachineBasicBlock *SuccBlock = *SI;
         bool LocalUse = false;
         if (AllUsesDominatedByBlock(Reg, SuccBlock, MBB,
@@ -615,9 +627,8 @@ bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
 
   DEBUG(dbgs() << "Sink instr " << *MI << "\tinto block " << *SuccToSinkTo);
 
-  // If the block has multiple predecessors, this would introduce computation on
-  // a path that it doesn't already exist.  We could split the critical edge,
-  // but for now we just punt.
+  // If the block has multiple predecessors, this is a critical edge.
+  // Decide if we can sink along it or need to break the edge.
   if (SuccToSinkTo->pred_size() > 1) {
     // We cannot sink a load across a critical edge - there may be stores in
     // other code paths.
@@ -697,7 +708,7 @@ bool MachineSinking::SinkInstruction(MachineInstr *MI, bool &SawStore) {
                        ++MachineBasicBlock::iterator(MI));
 
   // Move debug values.
-  for (SmallVector<MachineInstr *, 2>::iterator DBI = DbgValuesToSink.begin(),
+  for (SmallVectorImpl<MachineInstr *>::iterator DBI = DbgValuesToSink.begin(),
          DBE = DbgValuesToSink.end(); DBI != DBE; ++DBI) {
     MachineInstr *DbgMI = *DBI;
     SuccToSinkTo->splice(InsertPos, ParentBlock,  DbgMI,