AMDGPU/GlobalISel: Implement select() for @llvm.amdgcn.exp

[tstellar]

Remove stray debug statement.

[arsenm]

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

155

I think these are supposed to be const references now

174

Ditto

test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.exp.mir

3–4

You should be able to drop the IR section

[tstellar]

Address review comments.

[tstellar]

Ping.

[arsenm]

LGTM

lib/Target/AMDGPU/AMDGPUInstructionSelector.h

struct GEPInfo {
     GEPInfo(const MachineInstr &GEP) : GEP(GEP), Imm(0) { }
   };
 
   MachineOperand getSubOperand64(MachineOperand &MO, unsigned SubIdx) const;
   bool selectCOPY(MachineInstr &I) const;
   bool selectG_CONSTANT(MachineInstr &I) const;
   bool selectG_ADD(MachineInstr &I) const;
   bool selectG_GEP(MachineInstr &I) const;
+  bool selectG_INTRINSIC_W_SIDE_EFFECTS(MachineInstr &I) const;
   bool hasVgprParts(ArrayRef<GEPInfo> AddrInfo) const;
   void getAddrModeInfo(const MachineInstr &Load, const MachineRegisterInfo &MRI,
                        SmallVectorImpl<GEPInfo> &AddrInfo) const;
   bool selectSMRD(MachineInstr &I, ArrayRef<GEPInfo> AddrInfo) const;
   bool selectG_LOAD(MachineInstr &I) const;
   bool selectG_STORE(MachineInstr &I) const;
   unsigned getSALUOpcode(const MachineInstr &I) const;
   bool selectSimpleSALU(MachineInstr &I) const;

lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

default:
     llvm_unreachable("do not know to split immediate with this sub index.");
   case AMDGPU::sub0:
     return MachineOperand::CreateImm(Imm.getLoBits(32).getSExtValue());
   case AMDGPU::sub1:
     return MachineOperand::CreateImm(Imm.getHiBits(32).getSExtValue());
   }
 }
 
+static int64_t getConstant(const MachineInstr *MI) {
+  return MI->getOperand(1).getCImm()->getSExtValue();
+}
+
 bool AMDGPUInstructionSelector::selectG_ADD(MachineInstr &I) const {
   MachineBasicBlock *BB = I.getParent();
   MachineFunction *MF = BB->getParent();
   MachineRegisterInfo &MRI = MF->getRegInfo();
   unsigned Size = RBI.getSizeInBits(I.getOperand(0).getReg(), MRI, TRI);
   unsigned DstLo = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
   unsigned DstHi = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
 
bool AMDGPUInstructionSelector::selectG_ADD(MachineInstr &I) const {
   I.eraseFromParent();
   return true;
 }
 
 bool AMDGPUInstructionSelector::selectG_GEP(MachineInstr &I) const {
   return selectG_ADD(I);
 }
 
+static MachineInstr *
+buildEXP(const TargetInstrInfo &TII, MachineInstr *Insert, unsigned Tgt,
+         unsigned Reg0, unsigned Reg1, unsigned Reg2, unsigned Reg3,
+         unsigned VM, bool Compr, unsigned Enabled, bool Done) {
+  DebugLoc DL = Insert->getDebugLoc();
+  MachineBasicBlock &BB = *Insert->getParent();
+  unsigned Opcode = Done ? AMDGPU::EXP_DONE : AMDGPU::EXP;
+  return BuildMI(BB, Insert, DL, TII.get(Opcode))
+          .addImm(Tgt)
+          .addReg(Reg0)
+          .addReg(Reg1)
+          .addReg(Reg2)
+          .addReg(Reg3)
+          .addImm(VM)
+          .addImm(Compr)
+          .addImm(Enabled);
+}
+
+bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
+                                                 MachineInstr &I) const {
+  MachineBasicBlock *BB = I.getParent();
+  MachineFunction *MF = BB->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  DebugLoc DL = I.getDebugLoc();
+
+  unsigned IntrinsicID = I.getOperand(0).getIntrinsicID();
+  switch (IntrinsicID) {
+  case Intrinsic::amdgcn_exp: {
+    int64_t Tgt = getConstant(MRI.getVRegDef(I.getOperand(1).getReg()));
+    int64_t Enabled = getConstant(MRI.getVRegDef(I.getOperand(2).getReg()));
+    int64_t Done = getConstant(MRI.getVRegDef(I.getOperand(7).getReg()));
+    int64_t VM = getConstant(MRI.getVRegDef(I.getOperand(8).getReg()));
+
+    MachineInstr *Exp = buildEXP(TII, &I, Tgt, I.getOperand(3).getReg(),
+                                 I.getOperand(4).getReg(),
+                                 I.getOperand(5).getReg(),
+                                 I.getOperand(6).getReg(),
+                                 VM, false, Enabled, Done);
+
+    // operands to use appropriate classes.
+    bool Ret = constrainSelectedInstRegOperands(*Exp, TII, TRI, RBI);
+
+    I.eraseFromParent();
+    return Ret;
+  }
+  case Intrinsic::amdgcn_exp_compr: {
+    int64_t Tgt = getConstant(MRI.getVRegDef(I.getOperand(1).getReg()));
+    int64_t Enabled = getConstant(MRI.getVRegDef(I.getOperand(2).getReg()));
+    unsigned Reg0 = I.getOperand(3).getReg();
+    unsigned Reg1 = I.getOperand(4).getReg();
+    unsigned Undef = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+    int64_t Done = getConstant(MRI.getVRegDef(I.getOperand(5).getReg()));
+    int64_t VM = getConstant(MRI.getVRegDef(I.getOperand(6).getReg()));
+
+    BuildMI(*BB, &I, DL, TII.get(AMDGPU::IMPLICIT_DEF), Undef);
+    MachineInstr *Exp = buildEXP(TII, &I, Tgt, Reg0, Reg1, Undef, Undef, VM,
+                                 true,  Enabled, Done);
+    bool Ret = constrainSelectedInstRegOperands(*Exp, TII, TRI, RBI);
+
+    I.eraseFromParent();
+    return Ret;
+  }
+  }
+  return false;
+}
+
 bool AMDGPUInstructionSelector::selectG_STORE(MachineInstr &I) const {
   MachineBasicBlock *BB = I.getParent();
   DebugLoc DL = I.getDebugLoc();
 
   // FIXME: Select store instruction based on address space
   MachineInstr *Flat = BuildMI(*BB, &I, DL, TII.get(AMDGPU::FLAT_STORE_DWORD))
           .add(I.getOperand(1))
           .add(I.getOperand(0))
bool AMDGPUInstructionSelector::select(MachineInstr &I,
   case TargetOpcode::G_ADD:
     return selectG_ADD(I);
   case TargetOpcode::G_BITCAST:
     return selectCOPY(I);
   case TargetOpcode::G_CONSTANT:
     return selectG_CONSTANT(I);
   case TargetOpcode::G_GEP:
     return selectG_GEP(I);
+  case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
+    return selectG_INTRINSIC_W_SIDE_EFFECTS(I);
   case TargetOpcode::G_LOAD:
     return selectG_LOAD(I);
   case TargetOpcode::G_STORE:
     return selectG_STORE(I);
   }
   return false;
 }

test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgcn.exp.mir

+# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s
+
+--- |
+  define void @exp0() { ret void }
+...
+
+---
+name: exp0
+legalized:       true
+regBankSelected: true
+
+# CHECK: name: exp0
+body: |
+  bb.0:
+    liveins: $vgpr0
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:sgpr(s32) = G_CONSTANT i32 1
+    %2:sgpr(s32) = G_CONSTANT i32 15
+    %3:sgpr(s1)  = G_CONSTANT i1 0
+    %4:sgpr(s1)  = G_CONSTANT i1 1
+
+    ; CHECK: EXP 1, %0, %0, %0, %0, 0, 0, 15, implicit $exec
+    G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.exp), %1:sgpr(s32), %2:sgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %3:sgpr(s1), %3:sgpr(s1)
+
+    ; CHECK: EXP_DONE 1, %0, %0, %0, %0, 0, 0, 15, implicit $exec
+    G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.exp), %1:sgpr(s32), %2:sgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %0:vgpr(s32), %4:sgpr(s1), %3:sgpr(s1)
+
+    %5:vgpr(<2 x s16>) = G_BITCAST %0(s32)
+
+    ; CHECK: [[UNDEF0:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+    ; CHECK: EXP 1, %0, %0, [[UNDEF0]], [[UNDEF0]], 0, 1, 15, implicit $exec
+    G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.exp.compr), %1:sgpr(s32), %2:sgpr(s32), %5:vgpr(<2 x s16>), %5:vgpr(<2 x s16>), %3:sgpr(s1), %3:sgpr(s1)
+
+    ; CHECK: [[UNDEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+    ; CHECK: EXP_DONE 1, %0, %0, [[UNDEF1]], [[UNDEF1]], 0, 1, 15, implicit $exec
+    G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.exp.compr), %1:sgpr(s32), %2:sgpr(s32), %5:vgpr(<2 x s16>), %5:vgpr(<2 x s16>), %4:sgpr(s1), %3:sgpr(s1)
+