aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/X86/X86ISelDAGToDAG.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/X86/X86ISelDAGToDAG.cpp')
-rw-r--r--lib/Target/X86/X86ISelDAGToDAG.cpp126
1 files changed, 116 insertions, 10 deletions
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 8e2b1d6..5186482 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -187,6 +187,7 @@ namespace {
private:
SDNode *Select(SDNode *N);
+ SDNode *SelectGather(SDNode *N, unsigned Opc);
SDNode *SelectAtomic64(SDNode *Node, unsigned Opc);
SDNode *SelectAtomicLoadAdd(SDNode *Node, EVT NVT);
SDNode *SelectAtomicLoadArith(SDNode *Node, EVT NVT);
@@ -1905,6 +1906,20 @@ static bool isLoadIncOrDecStore(StoreSDNode *StoreNode, unsigned Opc,
ChainCheck = true;
continue;
}
+
+ // Make sure using Op as part of the chain would not cause a cycle here.
+ // In theory, we could check whether the chain node is a predecessor of
+ // the load. But that can be very expensive. Instead visit the uses and
+ // make sure they all have smaller node id than the load.
+ int LoadId = LoadNode->getNodeId();
+ for (SDNode::use_iterator UI = Op.getNode()->use_begin(),
+ UE = UI->use_end(); UI != UE; ++UI) {
+ if (UI.getUse().getResNo() != 0)
+ continue;
+ if (UI->getNodeId() > LoadId)
+ return false;
+ }
+
ChainOps.push_back(Op);
}
@@ -1938,6 +1953,38 @@ static unsigned getFusedLdStOpcode(EVT &LdVT, unsigned Opc) {
llvm_unreachable("unrecognized size for LdVT");
}
+/// SelectGather - Customized ISel for GATHER operations.
+///
+SDNode *X86DAGToDAGISel::SelectGather(SDNode *Node, unsigned Opc) {
+ // Operands of Gather: VSrc, Base, VIdx, VMask, Scale
+ SDValue Chain = Node->getOperand(0);
+ SDValue VSrc = Node->getOperand(2);
+ SDValue Base = Node->getOperand(3);
+ SDValue VIdx = Node->getOperand(4);
+ SDValue VMask = Node->getOperand(5);
+ ConstantSDNode *Scale = dyn_cast<ConstantSDNode>(Node->getOperand(6));
+ if (!Scale)
+ return 0;
+
+ SDVTList VTs = CurDAG->getVTList(VSrc.getValueType(), VSrc.getValueType(),
+ MVT::Other);
+
+ // Memory Operands: Base, Scale, Index, Disp, Segment
+ SDValue Disp = CurDAG->getTargetConstant(0, MVT::i32);
+ SDValue Segment = CurDAG->getRegister(0, MVT::i32);
+ const SDValue Ops[] = { VSrc, Base, getI8Imm(Scale->getSExtValue()), VIdx,
+ Disp, Segment, VMask, Chain};
+ SDNode *ResNode = CurDAG->getMachineNode(Opc, Node->getDebugLoc(),
+ VTs, Ops, array_lengthof(Ops));
+ // Node has 2 outputs: VDst and MVT::Other.
+ // ResNode has 3 outputs: VDst, VMask_wb, and MVT::Other.
+ // We replace VDst of Node with VDst of ResNode, and Other of Node with Other
+ // of ResNode.
+ ReplaceUses(SDValue(Node, 0), SDValue(ResNode, 0));
+ ReplaceUses(SDValue(Node, 1), SDValue(ResNode, 2));
+ return ResNode;
+}
+
SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
EVT NVT = Node->getValueType(0);
unsigned Opc, MOpc;
@@ -1953,23 +2000,82 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
switch (Opcode) {
default: break;
+ case ISD::INTRINSIC_W_CHAIN: {
+ unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue();
+ switch (IntNo) {
+ default: break;
+ case Intrinsic::x86_avx2_gather_d_pd:
+ case Intrinsic::x86_avx2_gather_d_pd_256:
+ case Intrinsic::x86_avx2_gather_q_pd:
+ case Intrinsic::x86_avx2_gather_q_pd_256:
+ case Intrinsic::x86_avx2_gather_d_ps:
+ case Intrinsic::x86_avx2_gather_d_ps_256:
+ case Intrinsic::x86_avx2_gather_q_ps:
+ case Intrinsic::x86_avx2_gather_q_ps_256:
+ case Intrinsic::x86_avx2_gather_d_q:
+ case Intrinsic::x86_avx2_gather_d_q_256:
+ case Intrinsic::x86_avx2_gather_q_q:
+ case Intrinsic::x86_avx2_gather_q_q_256:
+ case Intrinsic::x86_avx2_gather_d_d:
+ case Intrinsic::x86_avx2_gather_d_d_256:
+ case Intrinsic::x86_avx2_gather_q_d:
+ case Intrinsic::x86_avx2_gather_q_d_256: {
+ unsigned Opc;
+ switch (IntNo) {
+ default: llvm_unreachable("Impossible intrinsic");
+ case Intrinsic::x86_avx2_gather_d_pd: Opc = X86::VGATHERDPDrm; break;
+ case Intrinsic::x86_avx2_gather_d_pd_256: Opc = X86::VGATHERDPDYrm; break;
+ case Intrinsic::x86_avx2_gather_q_pd: Opc = X86::VGATHERQPDrm; break;
+ case Intrinsic::x86_avx2_gather_q_pd_256: Opc = X86::VGATHERQPDYrm; break;
+ case Intrinsic::x86_avx2_gather_d_ps: Opc = X86::VGATHERDPSrm; break;
+ case Intrinsic::x86_avx2_gather_d_ps_256: Opc = X86::VGATHERDPSYrm; break;
+ case Intrinsic::x86_avx2_gather_q_ps: Opc = X86::VGATHERQPSrm; break;
+ case Intrinsic::x86_avx2_gather_q_ps_256: Opc = X86::VGATHERQPSYrm; break;
+ case Intrinsic::x86_avx2_gather_d_q: Opc = X86::VPGATHERDQrm; break;
+ case Intrinsic::x86_avx2_gather_d_q_256: Opc = X86::VPGATHERDQYrm; break;
+ case Intrinsic::x86_avx2_gather_q_q: Opc = X86::VPGATHERQQrm; break;
+ case Intrinsic::x86_avx2_gather_q_q_256: Opc = X86::VPGATHERQQYrm; break;
+ case Intrinsic::x86_avx2_gather_d_d: Opc = X86::VPGATHERDDrm; break;
+ case Intrinsic::x86_avx2_gather_d_d_256: Opc = X86::VPGATHERDDYrm; break;
+ case Intrinsic::x86_avx2_gather_q_d: Opc = X86::VPGATHERQDrm; break;
+ case Intrinsic::x86_avx2_gather_q_d_256: Opc = X86::VPGATHERQDYrm; break;
+ }
+ SDNode *RetVal = SelectGather(Node, Opc);
+ if (RetVal)
+ // We already called ReplaceUses inside SelectGather.
+ return NULL;
+ break;
+ }
+ }
+ break;
+ }
case X86ISD::GlobalBaseReg:
return getGlobalBaseReg();
+
case X86ISD::ATOMOR64_DAG:
- return SelectAtomic64(Node, X86::ATOMOR6432);
case X86ISD::ATOMXOR64_DAG:
- return SelectAtomic64(Node, X86::ATOMXOR6432);
case X86ISD::ATOMADD64_DAG:
- return SelectAtomic64(Node, X86::ATOMADD6432);
case X86ISD::ATOMSUB64_DAG:
- return SelectAtomic64(Node, X86::ATOMSUB6432);
case X86ISD::ATOMNAND64_DAG:
- return SelectAtomic64(Node, X86::ATOMNAND6432);
case X86ISD::ATOMAND64_DAG:
- return SelectAtomic64(Node, X86::ATOMAND6432);
- case X86ISD::ATOMSWAP64_DAG:
- return SelectAtomic64(Node, X86::ATOMSWAP6432);
+ case X86ISD::ATOMSWAP64_DAG: {
+ unsigned Opc;
+ switch (Opcode) {
+ default: llvm_unreachable("Impossible intrinsic");
+ case X86ISD::ATOMOR64_DAG: Opc = X86::ATOMOR6432; break;
+ case X86ISD::ATOMXOR64_DAG: Opc = X86::ATOMXOR6432; break;
+ case X86ISD::ATOMADD64_DAG: Opc = X86::ATOMADD6432; break;
+ case X86ISD::ATOMSUB64_DAG: Opc = X86::ATOMSUB6432; break;
+ case X86ISD::ATOMNAND64_DAG: Opc = X86::ATOMNAND6432; break;
+ case X86ISD::ATOMAND64_DAG: Opc = X86::ATOMAND6432; break;
+ case X86ISD::ATOMSWAP64_DAG: Opc = X86::ATOMSWAP6432; break;
+ }
+ SDNode *RetVal = SelectAtomic64(Node, Opc);
+ if (RetVal)
+ return RetVal;
+ break;
+ }
case ISD::ATOMIC_LOAD_ADD: {
SDNode *RetVal = SelectAtomicLoadAdd(Node, NVT);
@@ -2128,7 +2234,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
}
SDValue InFlag = CurDAG->getCopyToReg(CurDAG->getEntryNode(), dl, LoReg,
- N0, SDValue()).getValue(1);
+ N0, SDValue()).getValue(1);
if (foldedLoad) {
SDValue Ops[] = { Tmp0, Tmp1, Tmp2, Tmp3, Tmp4, N1.getOperand(0),
@@ -2168,7 +2274,7 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
// Copy the low half of the result, if it is needed.
if (!SDValue(Node, 0).use_empty()) {
SDValue Result = CurDAG->getCopyFromReg(CurDAG->getEntryNode(), dl,
- LoReg, NVT, InFlag);
+ LoReg, NVT, InFlag);
InFlag = Result.getValue(2);
ReplaceUses(SDValue(Node, 0), Result);
DEBUG(dbgs() << "=> "; Result.getNode()->dump(CurDAG); dbgs() << '\n');
generated by cgit v1.1 at 2025-08-15 16:07:32 +0000