AMDGPU: Use set for tracked registers

The majority of the time spent in the pass checking
for the register reads. Rather than searching all of
the defined registers for uses in each instruction,
use a set of defined registers and check the operands
of the instruction.

This process still is algorithmically not great,
but with the additional trick of skipping the analysis
for addresses with one use, this brings one slow
testcase into a reasonable range.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312206 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
index 1b2e5e6d0f76e8025f631c2e0e50738f544b4898..8b3fdd874385b90b5b57de04f7330d8fb48648d0 100644 (file)
--- a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
+++ b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
@@ -141,36 +141,35 @@ static void moveInstsAfter(MachineBasicBlock::iterator I,
   }
 }
 
-static void addDefsToList(const MachineInstr &MI,
-                          SmallVectorImpl<const MachineOperand *> &Defs) {
-  for (const MachineOperand &Def : MI.defs()) {
-    Defs.push_back(&Def);
-  }
+static void addDefsToList(const MachineInstr &MI, DenseSet<unsigned> &Defs) {
+  // XXX: Should this be looking for implicit defs?
+  for (const MachineOperand &Def : MI.defs())
+    Defs.insert(Def.getReg());
 }
 
 static bool memAccessesCanBeReordered(MachineBasicBlock::iterator A,
                                       MachineBasicBlock::iterator B,
                                       const SIInstrInfo *TII,
                                       AliasAnalysis * AA) {
-  return (TII->areMemAccessesTriviallyDisjoint(*A, *B, AA) ||
-    // RAW or WAR - cannot reorder
-    // WAW - cannot reorder
-    // RAR - safe to reorder
-    !(A->mayStore() || B->mayStore()));
+  // RAW or WAR - cannot reorder
+  // WAW - cannot reorder
+  // RAR - safe to reorder
+  return !(A->mayStore() || B->mayStore()) ||
+    TII->areMemAccessesTriviallyDisjoint(*A, *B, AA);
 }
 
 // Add MI and its defs to the lists if MI reads one of the defs that are
 // already in the list. Returns true in that case.
 static bool
 addToListsIfDependent(MachineInstr &MI,
-                      SmallVectorImpl<const MachineOperand *> &Defs,
+                      DenseSet<unsigned> &Defs,
                       SmallVectorImpl<MachineInstr*> &Insts) {
-  for (const MachineOperand *Def : Defs) {
-    bool ReadDef = MI.readsVirtualRegister(Def->getReg());
-    // If ReadDef is true, then there is a use of Def between I
-    // and the instruction that I will potentially be merged with. We
-    // will need to move this instruction after the merged instructions.
-    if (ReadDef) {
+  for (MachineOperand &Use : MI.operands()) {
+    // If one of the defs is read, then there is a use of Def between I and the
+    // instruction that I will potentially be merged with. We will need to move
+    // this instruction after the merged instructions.
+
+    if (Use.isReg() && Use.readsReg() && Defs.count(Use.getReg())) {
       Insts.push_back(&MI);
       addDefsToList(MI, Defs);
       return true;
@@ -249,7 +248,8 @@ bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI) {
 }
 
 bool SILoadStoreOptimizer::findMatchingDSInst(CombineInfo &CI) {
-  MachineBasicBlock::iterator E = CI.I->getParent()->end();
+  MachineBasicBlock *MBB = CI.I->getParent();
+  MachineBasicBlock::iterator E = MBB->end();
   MachineBasicBlock::iterator MBBI = CI.I;
 
   int AddrIdx = AMDGPU::getNamedOperandIdx(CI.I->getOpcode(),
@@ -258,12 +258,13 @@ bool SILoadStoreOptimizer::findMatchingDSInst(CombineInfo &CI) {
 
   // We only ever merge operations with the same base address register, so don't
   // bother scanning forward if there are no other uses.
-  if (MRI->hasOneNonDBGUse(AddrReg0.getReg()))
+  if (TargetRegisterInfo::isPhysicalRegister(AddrReg0.getReg()) ||
+      MRI->hasOneNonDBGUse(AddrReg0.getReg()))
     return false;
 
   ++MBBI;
 
-  SmallVector<const MachineOperand *, 8> DefsToMove;
+  DenseSet<unsigned> DefsToMove;
   addDefsToList(*CI.I, DefsToMove);
 
   for ( ; MBBI != E; ++MBBI) {
@@ -547,6 +548,8 @@ bool SILoadStoreOptimizer::runOnMachineFunction(MachineFunction &MF) {
   MRI = &MF.getRegInfo();
   AA = &getAnalysis<AAResultsWrapperPass>().getAAResults();
 
+  assert(MRI->isSSA() && "Must be run on SSA");
+
   DEBUG(dbgs() << "Running SILoadStoreOptimizer\n");
 
   bool Modified = false;