[CodeGen][MachineLICM] Use RegUnits in HoistRegionPostRA

llvm/lib/CodeGen/MachineLICM.cpp

@@ -223,8 +223,8 @@ namespace {
     void HoistPostRA(MachineInstr *MI, unsigned Def, MachineLoop *CurLoop,
                      MachineBasicBlock *CurPreheader);
 
-    void ProcessMI(MachineInstr *MI, BitVector &PhysRegDefs,
-                   BitVector &PhysRegClobbers, SmallSet<int, 32> &StoredFIs,
+    void ProcessMI(MachineInstr *MI, BitVector &RUDefs, BitVector &RUClobbers,
+                   SmallSet<int, 32> &StoredFIs,
                    SmallVectorImpl<CandidateInfo> &Candidates,
                    MachineLoop *CurLoop);
 
@@ -423,10 +423,25 @@ static bool InstructionStoresToFI(const MachineInstr *MI, int FI) {
   return false;
 }
 
+static void applyBitsNotInRegMaskToRegUnitsMask(const TargetRegisterInfo *TRI,
+                                                BitVector &RUs,
+                                                const uint32_t *Mask) {
+  // FIXME: Use RUs better here
+  BitVector MaskedRegs(TRI->getNumRegs());
+  MaskedRegs.setBitsNotInMask(Mask);
+  for (const auto &Set : MaskedRegs.set_bits()) {
+    if (!Set)
+      continue;
+
+    for (MCRegUnitIterator RUI(Set, TRI); RUI.isValid(); ++RUI)
+      RUs.set(*RUI);
+  }
+}
+
 /// Examine the instruction for potentai LICM candidate. Also
 /// gather register def and frame object update information.
-void MachineLICMBase::ProcessMI(MachineInstr *MI, BitVector &PhysRegDefs,
-                                BitVector &PhysRegClobbers,
+void MachineLICMBase::ProcessMI(MachineInstr *MI, BitVector &RUDefs,
+                                BitVector &RUClobbers,
                                 SmallSet<int, 32> &StoredFIs,
                                 SmallVectorImpl<CandidateInfo> &Candidates,
                                 MachineLoop *CurLoop) {
@@ -448,7 +463,7 @@ void MachineLICMBase::ProcessMI(MachineInstr *MI, BitVector &PhysRegDefs,
     // We can't hoist an instruction defining a physreg that is clobbered in
     // the loop.
     if (MO.isRegMask()) {
-      PhysRegClobbers.setBitsNotInMask(MO.getRegMask());
+      applyBitsNotInRegMaskToRegUnitsMask(TRI, RUClobbers, MO.getRegMask());
       continue;
     }
 
@@ -460,16 +475,18 @@ void MachineLICMBase::ProcessMI(MachineInstr *MI, BitVector &PhysRegDefs,
     assert(Reg.isPhysical() && "Not expecting virtual register!");
 
     if (!MO.isDef()) {
-      if (Reg && (PhysRegDefs.test(Reg) || PhysRegClobbers.test(Reg)))
+      for (MCRegUnitIterator RUI(Reg, TRI); RUI.isValid(); ++RUI) {
         // If it's using a non-loop-invariant register, then it's obviously not
         // safe to hoist.
-        HasNonInvariantUse = true;
+        if (RUDefs.test(*RUI) || RUClobbers.test(*RUI))
+          HasNonInvariantUse = true;
+      }
       continue;
     }
 
     if (MO.isImplicit()) {
-      for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
-        PhysRegClobbers.set(*AI);
+      for (MCRegUnitIterator RUI(Reg, TRI); RUI.isValid(); ++RUI)
+        RUClobbers.set(*RUI);
       if (!MO.isDead())
         // Non-dead implicit def? This cannot be hoisted.
         RuledOut = true;
@@ -488,19 +505,18 @@ void MachineLICMBase::ProcessMI(MachineInstr *MI, BitVector &PhysRegDefs,
     // If we have already seen another instruction that defines the same
     // register, then this is not safe.  Two defs is indicated by setting a
     // PhysRegClobbers bit.
-    for (MCRegAliasIterator AS(Reg, TRI, true); AS.isValid(); ++AS) {
-      if (PhysRegDefs.test(*AS))
-        PhysRegClobbers.set(*AS);
+    for (MCRegUnitIterator RUI(Reg, TRI); RUI.isValid(); ++RUI) {
+      if (RUDefs.test(*RUI)) {
+        RUClobbers.set(*RUI);
+        RuledOut = true;
+      } else if (RUClobbers.test(*RUI)) {
+        // MI defined register is seen defined by another instruction in
+        // the loop, it cannot be a LICM candidate.
+        RuledOut = true;
+      }
+
+      RUDefs.set(*RUI);
     }
-    // Need a second loop because MCRegAliasIterator can visit the same
-    // register twice.
-    for (MCRegAliasIterator AS(Reg, TRI, true); AS.isValid(); ++AS)
-      PhysRegDefs.set(*AS);
-
-    if (PhysRegClobbers.test(Reg))
-      // MI defined register is seen defined by another instruction in
-      // the loop, it cannot be a LICM candidate.
-      RuledOut = true;
   }
 
   // Only consider reloads for now and remats which do not have register
@@ -521,9 +537,9 @@ void MachineLICMBase::HoistRegionPostRA(MachineLoop *CurLoop,
   if (!Preheader)
     return;
 
-  unsigned NumRegs = TRI->getNumRegs();
-  BitVector PhysRegDefs(NumRegs); // Regs defined once in the loop.
-  BitVector PhysRegClobbers(NumRegs); // Regs defined more than once.
+  unsigned NumRegUnits = TRI->getNumRegUnits();
+  BitVector RUDefs(NumRegUnits);     // RUs defined once in the loop.
+  BitVector RUClobbers(NumRegUnits); // RUs defined more than once.
 
   SmallVector<CandidateInfo, 32> Candidates;
   SmallSet<int, 32> StoredFIs;
@@ -540,22 +556,22 @@ void MachineLICMBase::HoistRegionPostRA(MachineLoop *CurLoop,
     // FIXME: That means a reload that're reused in successor block(s) will not
     // be LICM'ed.
     for (const auto &LI : BB->liveins()) {
-      for (MCRegAliasIterator AI(LI.PhysReg, TRI, true); AI.isValid(); ++AI)
-        PhysRegDefs.set(*AI);
+      for (MCRegUnitIterator RUI(LI.PhysReg, TRI); RUI.isValid(); ++RUI)
+        RUDefs.set(*RUI);
     }
 
     // Funclet entry blocks will clobber all registers
-    if (const uint32_t *Mask = BB->getBeginClobberMask(TRI))
-      PhysRegClobbers.setBitsNotInMask(Mask);
+    if (const uint32_t *Mask = BB->getBeginClobberMask(TRI)) {
+      applyBitsNotInRegMaskToRegUnitsMask(TRI, RUClobbers, Mask);
+    }
 
     SpeculationState = SpeculateUnknown;
     for (MachineInstr &MI : *BB)
-      ProcessMI(&MI, PhysRegDefs, PhysRegClobbers, StoredFIs, Candidates,
-                CurLoop);
+      ProcessMI(&MI, RUDefs, RUClobbers, StoredFIs, Candidates, CurLoop);
   }
 
   // Gather the registers read / clobbered by the terminator.
-  BitVector TermRegs(NumRegs);
+  BitVector TermRUs(NumRegUnits);
   MachineBasicBlock::iterator TI = Preheader->getFirstTerminator();
   if (TI != Preheader->end()) {
     for (const MachineOperand &MO : TI->operands()) {
@@ -564,8 +580,8 @@ void MachineLICMBase::HoistRegionPostRA(MachineLoop *CurLoop,
       Register Reg = MO.getReg();
       if (!Reg)
         continue;
-      for (MCRegAliasIterator AI(Reg, TRI, true); AI.isValid(); ++AI)
-        TermRegs.set(*AI);
+      for (MCRegUnitIterator RUI(Reg, TRI); RUI.isValid(); ++RUI)
+        TermRUs.set(*RUI);
     }
   }
 
@@ -583,24 +599,36 @@ void MachineLICMBase::HoistRegionPostRA(MachineLoop *CurLoop,
       continue;
 
     unsigned Def = Candidate.Def;
-    if (!PhysRegClobbers.test(Def) && !TermRegs.test(Def)) {
-      bool Safe = true;
-      MachineInstr *MI = Candidate.MI;
-      for (const MachineOperand &MO : MI->all_uses()) {
-        if (!MO.getReg())
-          continue;
-        Register Reg = MO.getReg();
-        if (PhysRegDefs.test(Reg) ||
-            PhysRegClobbers.test(Reg)) {
+    bool Safe = true;
+    for (MCRegUnitIterator RUI(Def, TRI); RUI.isValid(); ++RUI) {
+      if (RUClobbers.test(*RUI) || TermRUs.test(*RUI)) {
+        Safe = false;
+        break;
+      }
+    }
+
+    if (!Safe)
+      continue;
+
+    MachineInstr *MI = Candidate.MI;
+    for (const MachineOperand &MO : MI->all_uses()) {
+      if (!MO.getReg())
+        continue;
+      for (MCRegUnitIterator RUI(MO.getReg(), TRI); RUI.isValid(); ++RUI) {
+        if (RUDefs.test(*RUI) || RUClobbers.test(*RUI)) {
           // If it's using a non-loop-invariant register, then it's obviously
           // not safe to hoist.
           Safe = false;
           break;
         }
       }
-      if (Safe)
-        HoistPostRA(MI, Candidate.Def, CurLoop, CurPreheader);
+
+      if (!Safe)
+        break;
     }
+
+    if (Safe)
+      HoistPostRA(MI, Candidate.Def, CurLoop, CurPreheader);
   }
 }
 

llvm/test/CodeGen/AMDGPU/indirect-call.ll

@@ -886,12 +886,12 @@ define void @test_indirect_call_vgpr_ptr_inreg_arg(ptr %fptr) {
 ; GCN-NEXT:    v_writelane_b32 v40, s62, 30
 ; GCN-NEXT:    v_writelane_b32 v40, s63, 31
 ; GCN-NEXT:    s_mov_b64 s[6:7], exec
-; GCN-NEXT:    s_movk_i32 s4, 0x7b
 ; GCN-NEXT:  .LBB6_1: ; =>This Inner Loop Header: Depth=1
 ; GCN-NEXT:    v_readfirstlane_b32 s8, v0
 ; GCN-NEXT:    v_readfirstlane_b32 s9, v1
 ; GCN-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
 ; GCN-NEXT:    s_and_saveexec_b64 s[10:11], vcc
+; GCN-NEXT:    s_movk_i32 s4, 0x7b
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[8:9]
 ; GCN-NEXT:    ; implicit-def: $vgpr0_vgpr1
 ; GCN-NEXT:    s_xor_b64 exec, exec, s[10:11]
@@ -980,12 +980,12 @@ define void @test_indirect_call_vgpr_ptr_inreg_arg(ptr %fptr) {
 ; GISEL-NEXT:    v_writelane_b32 v40, s62, 30
 ; GISEL-NEXT:    v_writelane_b32 v40, s63, 31
 ; GISEL-NEXT:    s_mov_b64 s[6:7], exec
-; GISEL-NEXT:    s_movk_i32 s4, 0x7b
 ; GISEL-NEXT:  .LBB6_1: ; =>This Inner Loop Header: Depth=1
 ; GISEL-NEXT:    v_readfirstlane_b32 s8, v0
 ; GISEL-NEXT:    v_readfirstlane_b32 s9, v1
 ; GISEL-NEXT:    v_cmp_eq_u64_e32 vcc, s[8:9], v[0:1]
 ; GISEL-NEXT:    s_and_saveexec_b64 s[10:11], vcc
+; GISEL-NEXT:    s_movk_i32 s4, 0x7b
 ; GISEL-NEXT:    s_swappc_b64 s[30:31], s[8:9]
 ; GISEL-NEXT:    ; implicit-def: $vgpr0
 ; GISEL-NEXT:    s_xor_b64 exec, exec, s[10:11]