unsigned SequenceNum;
};
- /// This is a helper function for MergeStoresOfConstantsOrVecElts. Returns a
- /// constant build_vector of the stored constant values in Stores.
- SDValue getMergedConstantVectorStore(SelectionDAG &DAG,
- SDLoc SL,
- ArrayRef<MemOpLink> Stores,
- EVT Ty) const;
-
/// This is a helper function for MergeConsecutiveStores. When the source
/// elements of the consecutive stores are all constants or all extracted
/// vector elements, try to merge them into one larger store.
};
} // namespace
-SDValue DAGCombiner::getMergedConstantVectorStore(SelectionDAG &DAG,
- SDLoc SL,
- ArrayRef<MemOpLink> Stores,
- EVT Ty) const {
- SmallVector<SDValue, 8> BuildVector;
-
- for (const MemOpLink &Store : Stores)
- BuildVector.push_back(cast<StoreSDNode>(Store.MemNode)->getValue());
- return DAG.getNode(ISD::BUILD_VECTOR, SL, Ty, BuildVector);
-}
-
bool DAGCombiner::MergeStoresOfConstantsOrVecElts(
SmallVectorImpl<MemOpLink> &StoreNodes, EVT MemVT,
unsigned NumElem, bool IsConstantSrc, bool UseVector) {
EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT, NumElem);
assert(TLI.isTypeLegal(Ty) && "Illegal vector store");
if (IsConstantSrc) {
- StoredVal = getMergedConstantVectorStore(DAG, DL, StoreNodes, Ty);
+ // A vector store with a constant source implies that the constant is
+ // zero; we only handle merging stores of constant zeros because the zero
+ // can be materialized without a load.
+ // It may be beneficial to loosen this restriction to allow non-zero
+ // store merging.
+ StoredVal = DAG.getConstant(0, DL, Ty);
} else {
SmallVector<SDValue, 8> Ops;
for (unsigned i = 0; i < NumElem ; ++i) {
}
; GCN-LABEL: {{^}}merge_global_store_2_constants_f32_i32:
-; SI-DAG: s_mov_b32 [[SLO:s[0-9]+]], 4.0
-; SI-DAG: s_movk_i32 [[SHI:s[0-9]+]], 0x7b{{$}}
-; SI-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], [[SLO]]
-; SI-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], [[SHI]]
-; GCN: buffer_store_dwordx2 v{{\[}}[[VLO]]:[[VHI]]{{\]}}
+; GCN: buffer_store_dwordx2
define void @merge_global_store_2_constants_f32_i32(float addrspace(1)* %out) #0 {
%out.gep.1 = getelementptr float, float addrspace(1)* %out, i32 1
%out.gep.1.bc = bitcast float addrspace(1)* %out.gep.1 to i32 addrspace(1)*
}
; GCN-LABEL: {{^}}merge_global_store_4_constants_i32:
-; GCN-DAG: v_mov_b32_e32 v[[HI:[0-9]+]], 0x14d{{$}}
-; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, 0x1c8{{$}}
-; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, 0x7b{{$}}
-; GCN-DAG: v_mov_b32_e32 v[[LO:[0-9]+]], 0x4d2{{$}}
-; GCN: buffer_store_dwordx4 v{{\[}}[[LO]]:[[HI]]{{\]}}
+; GCN: buffer_store_dwordx4
define void @merge_global_store_4_constants_i32(i32 addrspace(1)* %out) #0 {
%out.gep.1 = getelementptr i32, i32 addrspace(1)* %out, i32 1
%out.gep.2 = getelementptr i32, i32 addrspace(1)* %out, i32 2