AMDGPU: Remove ATOMIC_PK_FADD

[arsenm]

Summary

The f32 and v2f16 cases should be handled the same way.

[rampitec]

LGTM, thanks!

[arsenm]

0ee1eba58114d4cbe9d2c976e39887b6df4508f7

llvm/lib/Target/AMDGPU/AMDGPUISelLowering.h

enum NodeType : unsigned {
   BUFFER_ATOMIC_AND,
   BUFFER_ATOMIC_OR,
   BUFFER_ATOMIC_XOR,
   BUFFER_ATOMIC_INC,
   BUFFER_ATOMIC_DEC,
   BUFFER_ATOMIC_CMPSWAP,
   BUFFER_ATOMIC_CSUB,
   BUFFER_ATOMIC_FADD,
-  ATOMIC_PK_FADD,
 
   LAST_AMDGPU_ISD_NUMBER
 };
 
 
 } // End namespace AMDGPUISD
 
 } // End namespace llvm
 
 #endif

llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp

const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   NODE_NAME_CASE(BUFFER_ATOMIC_AND)
   NODE_NAME_CASE(BUFFER_ATOMIC_OR)
   NODE_NAME_CASE(BUFFER_ATOMIC_XOR)
   NODE_NAME_CASE(BUFFER_ATOMIC_INC)
   NODE_NAME_CASE(BUFFER_ATOMIC_DEC)
   NODE_NAME_CASE(BUFFER_ATOMIC_CMPSWAP)
   NODE_NAME_CASE(BUFFER_ATOMIC_CSUB)
   NODE_NAME_CASE(BUFFER_ATOMIC_FADD)
-  NODE_NAME_CASE(ATOMIC_PK_FADD)
 
   case AMDGPUISD::LAST_AMDGPU_ISD_NUMBER: break;
   }
   return nullptr;
 }
 
 SDValue AMDGPUTargetLowering::getSqrtEstimate(SDValue Operand,
                                               SelectionDAG &DAG, int Enabled,

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

SDValue SITargetLowering::LowerINTRINSIC_VOID(SDValue Op,
   }
 
   case Intrinsic::amdgcn_global_atomic_fadd: {
     SDValue Ops[] = {
       Chain,
       Op.getOperand(2), // ptr
       Op.getOperand(3)  // vdata
     };
-    EVT VT = Op.getOperand(3).getValueType();
 
+    EVT VT = Op.getOperand(3).getValueType();
     auto *M = cast<MemSDNode>(Op);
-    if (VT.isVector()) {
-      return DAG.getMemIntrinsicNode(
-        AMDGPUISD::ATOMIC_PK_FADD, DL, Op->getVTList(), Ops, VT,
-        M->getMemOperand());
-    }
 
     return DAG.getAtomic(ISD::ATOMIC_LOAD_FADD, DL, VT,
                          DAG.getVTList(VT, MVT::Other), Ops,
                          M->getMemOperand()).getValue(1);
   }
   case Intrinsic::amdgcn_end_cf:
     return SDValue(DAG.getMachineNode(AMDGPU::SI_END_CF, DL, MVT::Other,
                                       Op->getOperand(2), Chain), 0);