Revert "[amdgpu] Add llvm.amdgcn.init.whole.wave intrinsic" (#108054)

Breaks bots, see #105822.

Reverts llvm/llvm-project#105822

11 files changed, 5 insertions, 72 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index 380dc7d3312f..0daaf6b65760 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -2738,11 +2738,6 @@ void AMDGPUDAGToDAGISel::SelectINTRINSIC_W_CHAIN(SDNode *N) {
   case Intrinsic::amdgcn_ds_bvh_stack_rtn:
     SelectDSBvhStackIntrinsic(N);
     return;
-  case Intrinsic::amdgcn_init_whole_wave:
-    CurDAG->getMachineFunction()
-        .getInfo<SIMachineFunctionInfo>()
-        ->setInitWholeWave();
-    break;
   }
 
   SelectCode(N);
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 53085d423cef..4dfd3f087c1a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -1772,14 +1772,6 @@ bool AMDGPUInstructionSelector::selectDSAppendConsume(MachineInstr &MI,
   return constrainSelectedInstRegOperands(*MIB, TII, TRI, RBI);
 }
 
-bool AMDGPUInstructionSelector::selectInitWholeWave(MachineInstr &MI) const {
-  MachineFunction *MF = MI.getParent()->getParent();
-  SIMachineFunctionInfo *MFInfo = MF->getInfo<SIMachineFunctionInfo>();
-
-  MFInfo->setInitWholeWave();
-  return selectImpl(MI, *CoverageInfo);
-}
-
 bool AMDGPUInstructionSelector::selectSBarrier(MachineInstr &MI) const {
   if (TM.getOptLevel() > CodeGenOptLevel::None) {
     unsigned WGSize = STI.getFlatWorkGroupSizes(MF->getFunction()).second;
@@ -2107,8 +2099,6 @@ bool AMDGPUInstructionSelector::selectG_INTRINSIC_W_SIDE_EFFECTS(
     return selectDSAppendConsume(I, true);
   case Intrinsic::amdgcn_ds_consume:
     return selectDSAppendConsume(I, false);
-  case Intrinsic::amdgcn_init_whole_wave:
-    return selectInitWholeWave(I);
   case Intrinsic::amdgcn_s_barrier:
     return selectSBarrier(I);
   case Intrinsic::amdgcn_raw_buffer_load_lds:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
index df39ecbd61bc..068db5c1c144 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
@@ -120,7 +120,6 @@ private:
   bool selectDSOrderedIntrinsic(MachineInstr &MI, Intrinsic::ID IID) const;
   bool selectDSGWSIntrinsic(MachineInstr &MI, Intrinsic::ID IID) const;
   bool selectDSAppendConsume(MachineInstr &MI, bool IsAppend) const;
-  bool selectInitWholeWave(MachineInstr &MI) const;
   bool selectSBarrier(MachineInstr &MI) const;
   bool selectDSBvhStackIntrinsic(MachineInstr &MI) const;
 
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.h b/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.h
index b1022e48b8d3..7efb7f825348 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.h
@@ -67,8 +67,6 @@ protected:
   // Kernel may need limited waves per EU for better performance.
   bool WaveLimiter = false;
 
-  bool HasInitWholeWave = false;
-
 public:
   AMDGPUMachineFunction(const Function &F, const AMDGPUSubtarget &ST);
 
@@ -111,9 +109,6 @@ public:
     return WaveLimiter;
   }
 
-  bool hasInitWholeWave() const { return HasInitWholeWave; }
-  void setInitWholeWave() { HasInitWholeWave = true; }
-
   unsigned allocateLDSGlobal(const DataLayout &DL, const GlobalVariable &GV) {
     return allocateLDSGlobal(DL, GV, DynLDSAlign);
   }
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index f2c9619cb827..46d98cad963b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -4997,7 +4997,6 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
       OpdsMapping[3] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, WaveSize);
       break;
     }
-    case Intrinsic::amdgcn_init_whole_wave:
     case Intrinsic::amdgcn_live_mask: {
       OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::VCCRegBankID, 1);
       break;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td b/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
index 2cd5fb2b9428..95c4859674ec 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSearchableTables.td
@@ -329,7 +329,6 @@ def : SourceOfDivergence<int_amdgcn_mov_dpp>;
 def : SourceOfDivergence<int_amdgcn_mov_dpp8>;
 def : SourceOfDivergence<int_amdgcn_update_dpp>;
 def : SourceOfDivergence<int_amdgcn_writelane>;
-def : SourceOfDivergence<int_amdgcn_init_whole_wave>;
 
 foreach intr = AMDGPUMFMAIntrinsics908 in
 def : SourceOfDivergence<intr>;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 7f659578a6d2..9c9c50513937 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -1739,9 +1739,6 @@ bool GCNTargetMachine::parseMachineFunctionInfo(
                                            ? DenormalMode::IEEE
                                            : DenormalMode::PreserveSign;
 
-  if (YamlMFI.HasInitWholeWave)
-    MFI->setInitWholeWave();
-
   return false;
 }
 
diff --git a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
index dfdc7ad32b00..8c951105101d 100644
--- a/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFrameLowering.cpp
@@ -1343,14 +1343,10 @@ void SIFrameLowering::processFunctionBeforeFrameFinalized(
 
   // Allocate spill slots for WWM reserved VGPRs.
   // For chain functions, we only need to do this if we have calls to
-  // llvm.amdgcn.cs.chain (otherwise there's no one to save them for, since
-  // chain functions do not return) and the function did not contain a call to
-  // llvm.amdgcn.init.whole.wave (since in that case there are no inactive lanes
-  // when entering the function).
-  bool IsChainWithoutRestores =
-      FuncInfo->isChainFunction() &&
-      (!MF.getFrameInfo().hasTailCall() || FuncInfo->hasInitWholeWave());
-  if (!FuncInfo->isEntryFunction() && !IsChainWithoutRestores) {
+  // llvm.amdgcn.cs.chain.
+  bool IsChainWithoutCalls =
+      FuncInfo->isChainFunction() && !MF.getFrameInfo().hasTailCall();
+  if (!FuncInfo->isEntryFunction() && !IsChainWithoutCalls) {
     for (Register Reg : FuncInfo->getWWMReservedRegs()) {
       const TargetRegisterClass *RC = TRI->getPhysRegBaseClass(Reg);
       FuncInfo->allocateWWMSpill(MF, Reg, TRI->getSpillSize(*RC),
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index f3eee9c807c1..b7543238c130 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -583,16 +583,6 @@ def SI_INIT_EXEC_FROM_INPUT : SPseudoInstSI <
   let Defs = [EXEC];
 }
 
-// Sets EXEC to all lanes and returns the previous EXEC.
-def SI_INIT_WHOLE_WAVE : SPseudoInstSI <
-  (outs SReg_1:$dst), (ins),
-  [(set i1:$dst, (int_amdgcn_init_whole_wave))]> {
-  let Defs = [EXEC];
-  let Uses = [EXEC];
-
-  let isConvergent = 1;
-}
-
 // Return for returning shaders to a shader variant epilog.
 def SI_RETURN_TO_EPILOG : SPseudoInstSI <
   (outs), (ins variable_ops), [(AMDGPUreturn_to_epilog)]> {
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
index 7cebfa29fe7b..7af5e7388f84 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
@@ -289,8 +289,6 @@ struct SIMachineFunctionInfo final : public yaml::MachineFunctionInfo {
   StringValue SGPRForEXECCopy;
   StringValue LongBranchReservedReg;
 
-  bool HasInitWholeWave = false;
-
   SIMachineFunctionInfo() = default;
   SIMachineFunctionInfo(const llvm::SIMachineFunctionInfo &,
                         const TargetRegisterInfo &TRI,
@@ -338,7 +336,6 @@ template <> struct MappingTraits<SIMachineFunctionInfo> {
                        StringValue()); // Don't print out when it's empty.
     YamlIO.mapOptional("longBranchReservedReg", MFI.LongBranchReservedReg,
                        StringValue());
-    YamlIO.mapOptional("hasInitWholeWave", MFI.HasInitWholeWave, false);
   }
 };
 
diff --git a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
index 7a493d8d1d1d..f9d7ead4ff3e 100644
--- a/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
+++ b/llvm/lib/Target/AMDGPU/SIWholeQuadMode.cpp
@@ -594,8 +594,7 @@ char SIWholeQuadMode::scanInstructions(MachineFunction &MF,
         KillInstrs.push_back(&MI);
         BBI.NeedsLowering = true;
       } else if (Opcode == AMDGPU::SI_INIT_EXEC ||
-                 Opcode == AMDGPU::SI_INIT_EXEC_FROM_INPUT ||
-                 Opcode == AMDGPU::SI_INIT_WHOLE_WAVE) {
+                 Opcode == AMDGPU::SI_INIT_EXEC_FROM_INPUT) {
         InitExecInstrs.push_back(&MI);
       } else if (WQMOutputs) {
         // The function is in machine SSA form, which means that physical
@@ -1583,29 +1582,6 @@ void SIWholeQuadMode::lowerInitExec(MachineInstr &MI) {
   MachineBasicBlock *MBB = MI.getParent();
   bool IsWave32 = ST->isWave32();
 
-  if (MI.getOpcode() == AMDGPU::SI_INIT_WHOLE_WAVE) {
-    assert(MBB == &MBB->getParent()->front() &&
-           "init whole wave not in entry block");
-    Register EntryExec = MRI->createVirtualRegister(TRI->getBoolRC());
-    MachineInstr *SaveExec =
-        BuildMI(*MBB, MBB->begin(), MI.getDebugLoc(),
-                TII->get(IsWave32 ? AMDGPU::S_OR_SAVEEXEC_B32
-                                  : AMDGPU::S_OR_SAVEEXEC_B64),
-                EntryExec)
-            .addImm(-1);
-
-    // Replace all uses of MI's destination reg with EntryExec.
-    MRI->replaceRegWith(MI.getOperand(0).getReg(), EntryExec);
-    MI.eraseFromParent();
-
-    if (LIS) {
-      LIS->RemoveMachineInstrFromMaps(MI);
-      LIS->InsertMachineInstrInMaps(*SaveExec);
-      LIS->createAndComputeVirtRegInterval(EntryExec);
-    }
-    return;
-  }
-
   if (MI.getOpcode() == AMDGPU::SI_INIT_EXEC) {
     // This should be before all vector instructions.
     MachineInstr *InitMI =