DAG: Fix missing legalization for any_extend_vector_inreg operands

[RKSimon]

Please don't forgot to subscribe cfe/llvm-commits on your tickets.

lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

1681

We don't normally do this so I think you need to comment this properly - explaining that *_EXTEND_VECTOR_INREG only reference the lower half of the input and that splitting the result has the same effect as splitting the input operand.

[arsenm]

Add comment

[RKSimon]

Thanks, LGTM

[arsenm]

r299389

lib/CodeGen/SelectionDAG/LegalizeTypes.h

private:
   bool SplitVectorOperand(SDNode *N, unsigned OpNo);
   SDValue SplitVecOp_VSELECT(SDNode *N, unsigned OpNo);
   SDValue SplitVecOp_UnaryOp(SDNode *N);
   SDValue SplitVecOp_TruncateHelper(SDNode *N);
 
   SDValue SplitVecOp_BITCAST(SDNode *N);
   SDValue SplitVecOp_EXTRACT_SUBVECTOR(SDNode *N);
   SDValue SplitVecOp_EXTRACT_VECTOR_ELT(SDNode *N);
+  SDValue SplitVecOp_ExtVecInRegOp(SDNode *N);
   SDValue SplitVecOp_STORE(StoreSDNode *N, unsigned OpNo);
   SDValue SplitVecOp_MSTORE(MaskedStoreSDNode *N, unsigned OpNo);
   SDValue SplitVecOp_MSCATTER(MaskedScatterSDNode *N, unsigned OpNo);
   SDValue SplitVecOp_MGATHER(MaskedGatherSDNode *N, unsigned OpNo);
   SDValue SplitVecOp_CONCAT_VECTORS(SDNode *N);
   SDValue SplitVecOp_VSETCC(SDNode *N);
   SDValue SplitVecOp_FP_ROUND(SDNode *N);
   SDValue SplitVecOp_FCOPYSIGN(SDNode *N);

lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

#endif
     case ISD::FP_EXTEND:
     case ISD::SIGN_EXTEND:
     case ISD::ZERO_EXTEND:
     case ISD::ANY_EXTEND:
     case ISD::FTRUNC:
     case ISD::FCANONICALIZE:
       Res = SplitVecOp_UnaryOp(N);
       break;
+
+    case ISD::ANY_EXTEND_VECTOR_INREG:
+    case ISD::SIGN_EXTEND_VECTOR_INREG:
+    case ISD::ZERO_EXTEND_VECTOR_INREG:
+      Res = SplitVecOp_ExtVecInRegOp(N);
+      break;
     }
   }
 
   // If the result is null, the sub-method took care of registering results etc.
   if (!Res.getNode()) return false;
 
   // If the result is N, the sub-method updated N in place.  Tell the legalizer
   // core about this.
SDValue Store =
       DAG.getStore(DAG.getEntryNode(), dl, Vec, StackPtr, MachinePointerInfo());
 
   // Load back the required element.
   StackPtr = TLI.getVectorElementPointer(DAG, StackPtr, VecVT, Idx);
   return DAG.getExtLoad(ISD::EXTLOAD, dl, N->getValueType(0), Store, StackPtr,
                         MachinePointerInfo(), EltVT);
 }
 
+SDValue DAGTypeLegalizer::SplitVecOp_ExtVecInRegOp(SDNode *N) {
+  SDValue Lo, Hi;
+
+  // *_EXTEND_VECTOR_INREG only reference the lower half of the input, so
+  // splitting the result has the same effect as splitting the input operand.
+  SplitVecRes_ExtVecInRegOp(N, Lo, Hi);
+
+  return DAG.getNode(ISD::CONCAT_VECTORS, SDLoc(N), N->getValueType(0), Lo, Hi);
+}
+
 SDValue DAGTypeLegalizer::SplitVecOp_MGATHER(MaskedGatherSDNode *MGT,
                                              unsigned OpNo) {
   EVT LoVT, HiVT;
   SDLoc dl(MGT);
   std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(MGT->getValueType(0));
 
   SDValue Ch = MGT->getChain();
   SDValue Ptr = MGT->getBasePtr();

test/CodeGen/AMDGPU/any_extend_vector_inreg.ll

+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+; GCN-LABEL: {{^}}any_extend_vector_inreg_v16i8_to_v4i32:
+; GCN: {{buffer|flat}}_load_dwordx4
+; GCN-DAG: {{buffer|flat}}_load_dwordx4
+; GCN-DAG: {{buffer|flat}}_load_dword
+
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+; GCN: {{buffer|flat}}_store_byte
+define amdgpu_kernel void @any_extend_vector_inreg_v16i8_to_v4i32(<8 x i8> addrspace(1)* nocapture readonly %arg, <16 x i8> addrspace(1)* %arg1) local_unnamed_addr #0 {
+bb:
+  %tmp = bitcast <8 x i8> addrspace(1)* %arg to <16 x i8> addrspace(1)*
+  %tmp2 = load <16 x i8>, <16 x i8> addrspace(1)* %tmp, align 16
+  %tmp3 = extractelement <16 x i8> %tmp2, i64 4
+  %tmp6 = extractelement <16 x i8> %tmp2, i64 11
+  %tmp10 = getelementptr inbounds <8 x i8>, <8 x i8> addrspace(1)* %arg, i64 2
+  %tmp11 = bitcast <8 x i8> addrspace(1)* %tmp10 to <16 x i8> addrspace(1)*
+  %tmp12 = load <16 x i8>, <16 x i8> addrspace(1)* %tmp11, align 16
+  %tmp13 = extractelement <16 x i8> %tmp12, i64 7
+  %tmp17 = extractelement <16 x i8> %tmp12, i64 12
+  %tmp21 = getelementptr inbounds <8 x i8>, <8 x i8> addrspace(1)* %arg, i64 4
+  %tmp22 = bitcast <8 x i8> addrspace(1)* %tmp21 to <16 x i8> addrspace(1)*
+  %tmp23 = load <16 x i8>, <16 x i8> addrspace(1)* %tmp22, align 16
+  %tmp24 = extractelement <16 x i8> %tmp23, i64 3
+  %tmp1 = insertelement <16 x i8> undef, i8 %tmp3, i32 2
+  %tmp4 = insertelement <16 x i8> %tmp1, i8 0, i32 3
+  %tmp5 = insertelement <16 x i8> %tmp4, i8 0, i32 4
+  %tmp7 = insertelement <16 x i8> %tmp5, i8 %tmp6, i32 5
+  %tmp8 = insertelement <16 x i8> %tmp7, i8 0, i32 6
+  %tmp9 = insertelement <16 x i8> %tmp8, i8 %tmp13, i32 7
+  %tmp14 = insertelement <16 x i8> %tmp9, i8 0, i32 8
+  %tmp15 = insertelement <16 x i8> %tmp14, i8 %tmp17, i32 9
+  %tmp16 = insertelement <16 x i8> %tmp15, i8 0, i32 10
+  %tmp18 = insertelement <16 x i8> %tmp16, i8 0, i32 11
+  %tmp19 = insertelement <16 x i8> %tmp18, i8 %tmp24, i32 12
+  store <16 x i8> %tmp19, <16 x i8> addrspace(1)* %arg1, align 1
+  ret void
+}
+
+attributes #0 = { nounwind }