AMDGPU/SI: Handle hazard with sgpr lane selects for v_{read,write}lane

[nhaehnle]

LGTM

lib/Target/AMDGPU/GCNHazardRecognizer.h

class GCNHazardRecognizer final : public ScheduleHazardRecognizer {
   int checkSMRDHazards(MachineInstr *SMRD);
   int checkVMEMHazards(MachineInstr* VMEM);
   int checkDPPHazards(MachineInstr *DPP);
   int checkDivFMasHazards(MachineInstr *DivFMas);
   int checkGetRegHazards(MachineInstr *GetRegInstr);
   int checkSetRegHazards(MachineInstr *SetRegInstr);
   int createsVALUHazard(const MachineInstr &MI);
   int checkVALUHazards(MachineInstr *VALU);
+  int checkRWLaneHazards(MachineInstr *RWLane);
 public:
   GCNHazardRecognizer(const MachineFunction &MF);
   // We can only issue one instruction per cycle.
   bool atIssueLimit() const override { return true; }
   void EmitInstruction(SUnit *SU) override;
   void EmitInstruction(MachineInstr *MI) override;
   HazardType getHazardType(SUnit *SU, int Stalls) override;
   void EmitNoop() override;

lib/Target/AMDGPU/GCNHazardRecognizer.cpp

 static bool isSGetReg(unsigned Opcode) {
   return Opcode == AMDGPU::S_GETREG_B32;
 }
 
 static bool isSSetReg(unsigned Opcode) {
   return Opcode == AMDGPU::S_SETREG_B32 || Opcode == AMDGPU::S_SETREG_IMM32_B32;
 }
 
+static bool isRWLane(unsigned Opcode) {
+  return Opcode == AMDGPU::V_READLANE_B32 || Opcode == AMDGPU::V_WRITELANE_B32;
+}
+
 static bool getHWReg(const SIInstrInfo *TII, const MachineInstr &RegInstr) {
 
   const MachineOperand *RegOp = TII->getNamedOperand(RegInstr,
                                                      AMDGPU::OpName::simm16);
   return RegOp->getImm() & 0x3f;
 }
 
 ScheduleHazardRecognizer::HazardType
if (SIInstrInfo::isVALU(*MI) && checkVALUHazards(MI) > 0)
     return NoopHazard;
 
   if (SIInstrInfo::isDPP(*MI) && checkDPPHazards(MI) > 0)
     return NoopHazard;
 
   if (isDivFMas(MI->getOpcode()) && checkDivFMasHazards(MI) > 0)
     return NoopHazard;
 
+  if (isRWLane(MI->getOpcode()) && checkRWLaneHazards(MI) > 0)
+    return NoopHazard;
+
   if (isSGetReg(MI->getOpcode()) && checkGetRegHazards(MI) > 0)
     return NoopHazard;
 
   if (isSSetReg(MI->getOpcode()) && checkSetRegHazards(MI) > 0)
     return NoopHazard;
 
   return NoHazard;
 }
if (SIInstrInfo::isVMEM(*MI))
       WaitStates = std::max(WaitStates, checkVMEMHazards(MI));
 
     if (SIInstrInfo::isDPP(*MI))
       WaitStates = std::max(WaitStates, checkDPPHazards(MI));
 
     if (isDivFMas(MI->getOpcode()))
       WaitStates = std::max(WaitStates, checkDivFMasHazards(MI));
 
+    if (isRWLane(MI->getOpcode()))
+      WaitStates = std::max(WaitStates, checkRWLaneHazards(MI));
+
     return WaitStates;
   }
 
   if (isSGetReg(MI->getOpcode()))
     return std::max(0, checkGetRegHazards(MI));
 
   if (isSSetReg(MI->getOpcode()))
     return std::max(0, checkSetRegHazards(MI));
auto IsHazardFn = [this, Reg, TRI] (MachineInstr *MI) {
              TRI->regsOverlap(MI->getOperand(DataIdx).getReg(), Reg);
     };
     int WaitStatesNeededForDef =
         VALUWaitStates - getWaitStatesSince(IsHazardFn);
     WaitStatesNeeded = std::max(WaitStatesNeeded, WaitStatesNeededForDef);
   }
   return WaitStatesNeeded;
 }
+
+int GCNHazardRecognizer::checkRWLaneHazards(MachineInstr *RWLane) {
+  const SIInstrInfo *TII = ST.getInstrInfo();
+  const SIRegisterInfo *TRI = ST.getRegisterInfo();
+  const MachineRegisterInfo &MRI =
+      RWLane->getParent()->getParent()->getRegInfo();
+
+  const MachineOperand *LaneSelectOp =
+      TII->getNamedOperand(*RWLane, AMDGPU::OpName::src1);
+
+  if (!LaneSelectOp->isReg() || !TRI->isSGPRReg(MRI, LaneSelectOp->getReg()))
+    return 0;
+
+  unsigned LaneSelectReg = LaneSelectOp->getReg();
+  auto IsHazardFn = [TII] (MachineInstr *MI) {
+    return TII->isVALU(*MI);
+  };
+
+  const int RWLaneWaitStates = 4;
+  int WaitStatesSince = getWaitStatesSinceDef(LaneSelectReg, IsHazardFn);
+  return RWLaneWaitStates - WaitStatesSince;
+}

test/CodeGen/MIR/AMDGPU/inserted-wait-states.mir

 # RUN: llc -march=amdgcn -mcpu=tahiti -run-pass post-RA-hazard-rec  %s -o - | FileCheck %s -check-prefixes=GCN
 # RUN: llc -march=amdgcn -mcpu=hawaii -run-pass post-RA-hazard-rec  %s -o - | FileCheck %s -check-prefixes=GCN,CIVI
 # RUN: llc -march=amdgcn -mcpu=fiji -run-pass post-RA-hazard-rec  %s -o - | FileCheck %s -check-prefixes=GCN,CIVI,VI
 
 --- |
   define void @div_fmas() { ret void }
   define void @s_getreg() { ret void }
   define void @s_setreg() { ret void }
   define void @vmem_gt_8dw_store() { ret void }
+  define void @readwrite_lane() { ret void }
 ...
 ---
 # GCN-LABEL: name: div_fmas
 
 # GCN-LABEL: bb.0:
 # GCN: S_MOV_B64
 # GCN-NOT: S_NOP
 # GCN: V_DIV_FMAS
bb.1:
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
     FLAT_ATOMIC_CMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
     FLAT_ATOMIC_FCMPSWAP_X2 %vgpr0_vgpr1, %vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit %exec, implicit %flat_scr
     %vgpr3 = V_MOV_B32_e32 0, implicit %exec
     S_ENDPGM
 
 ...
+
+...
+---
+
+# GCN-LABEL: name: readwrite_lane
+
+# GCN-LABEL: bb.0:
+# GCN: V_ADD_I32
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: V_READLANE_B32
+
+# GCN-LABEL: bb.1:
+# GCN: V_ADD_I32
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: V_WRITELANE_B32
+
+# GCN-LABEL: bb.2:
+# GCN: V_ADD_I32
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: V_READLANE_B32
+
+# GCN-LABEL: bb.3:
+# GCN: V_ADD_I32
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: S_NOP
+# GCN: V_WRITELANE_B32
+
+name: readwrite_lane
+
+body: |
+  bb.0:
+    successors: %bb.1
+    %vgpr0,%sgpr0_sgpr1 = V_ADD_I32_e64 %vgpr1, %vgpr2, implicit %vcc, implicit %exec
+    %sgpr4 = V_READLANE_B32 %vgpr4, %sgpr0
+    S_BRANCH %bb.1
+
+  bb.1:
+    successors: %bb.2
+    %vgpr0,%sgpr0_sgpr1 = V_ADD_I32_e64 %vgpr1, %vgpr2, implicit %vcc, implicit %exec
+    %vgpr4 = V_WRITELANE_B32 %sgpr0, %sgpr0
+    S_BRANCH %bb.2
+
+  bb.2:
+    successors: %bb.3
+    %vgpr0,implicit %vcc = V_ADD_I32_e32 %vgpr1, %vgpr2, implicit %vcc, implicit %exec
+    %sgpr4 = V_READLANE_B32 %vgpr4, %vcc_lo
+    S_BRANCH %bb.3
+
+  bb.3:
+    %vgpr0,implicit %vcc = V_ADD_I32_e32 %vgpr1, %vgpr2, implicit %vcc, implicit %exec
+    %vgpr4 = V_WRITELANE_B32 %sgpr4, %vcc_lo
+    S_ENDPGM
+
+...