Use VLDM / VSTM to spill/reload 128-bit Neon registers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78468 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp
index ae6f916f42b2343c30764d59a5de1f0c641a7dd1..c179d8334dfd2855bee257f737ce1d350d6944ac 100644 (file)
--- a/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -655,11 +655,15 @@ storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
     AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::FSTD))
                    .addReg(SrcReg, getKillRegState(isKill))
                    .addFrameIndex(FI).addImm(0));
-  } else {
-    assert(RC == ARM::SPRRegisterClass && "Unknown regclass!");
+  } else if (RC == ARM::SPRRegisterClass) {
     AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::FSTS))
                    .addReg(SrcReg, getKillRegState(isKill))
                    .addFrameIndex(FI).addImm(0));
+  } else {
+    assert(RC == ARM::QPRRegisterClass && "Unknown regclass!");
+    // FIXME: Neon instructions should support predicates
+    BuildMI(MBB, I, DL, get(ARM::VSTRQ)).addReg(SrcReg, getKillRegState(isKill))
+      .addFrameIndex(FI).addImm(0);
   }
 }
 
@@ -676,10 +680,13 @@ loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
   } else if (RC == ARM::DPRRegisterClass || RC == ARM::DPR_VFP2RegisterClass) {
     AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::FLDD), DestReg)
                    .addFrameIndex(FI).addImm(0));
-  } else {
-    assert(RC == ARM::SPRRegisterClass && "Unknown regclass!");
+  } else if (RC == ARM::SPRRegisterClass) {
     AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::FLDS), DestReg)
                    .addFrameIndex(FI).addImm(0));
+  } else {
+    assert(RC == ARM::QPRRegisterClass && "Unknown regclass!");
+    // FIXME: Neon instructions should support predicates
+    BuildMI(MBB, I, DL, get(ARM::VLDRQ), DestReg).addFrameIndex(FI).addImm(0);
   }
 }
 
@@ -928,6 +935,8 @@ int llvm::rewriteARMFrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
       NumBits = 8;
       break;
     }
+    case ARMII::AddrMode4:
+     break;
     case ARMII::AddrMode5: {
       ImmIdx = FrameRegIdx+1;
       InstrOffs = ARM_AM::getAM5Offset(MI.getOperand(ImmIdx).getImm());

diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index 8e73a43b2d955e70c4e85aa1643f2489eec422d6..3304900c9f2e69c3d4b52755ecb6a2618a1465c7 100644 (file)
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -78,6 +78,8 @@ public:
                        SDValue &Offset, SDValue &Opc);
   bool SelectAddrMode3Offset(SDValue Op, SDValue N,
                              SDValue &Offset, SDValue &Opc);
+  bool SelectAddrMode4(SDValue Op, SDValue N, SDValue &Addr,
+                       SDValue &Mode);
   bool SelectAddrMode5(SDValue Op, SDValue N, SDValue &Base,
                        SDValue &Offset);
   bool SelectAddrMode6(SDValue Op, SDValue N, SDValue &Addr, SDValue &Update,
@@ -383,6 +385,12 @@ bool ARMDAGToDAGISel::SelectAddrMode3Offset(SDValue Op, SDValue N,
   return true;
 }
 
+bool ARMDAGToDAGISel::SelectAddrMode4(SDValue Op, SDValue N,
+                                      SDValue &Addr, SDValue &Mode) {
+  Addr = N;
+  Mode = CurDAG->getTargetConstant(0, MVT::i32);
+  return true;
+}
 
 bool ARMDAGToDAGISel::SelectAddrMode5(SDValue Op, SDValue N,
                                       SDValue &Base, SDValue &Offset) {

diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td
index fb84d523fa851614e201a71b26b107931f44cc90..5bc7212881c82b599da07d2909e55469028803a3 100644 (file)
--- a/lib/Target/ARM/ARMInstrFormats.td
+++ b/lib/Target/ARM/ARMInstrFormats.td
@@ -1199,6 +1199,10 @@ class NI<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern>
   : NeonI<oops, iops, AddrModeNone, IndexModeNone, itin, asm, "", pattern> {
 }
 
+class NI4<dag oops, dag iops, InstrItinClass itin, string asm, list<dag> pattern>
+  : NeonI<oops, iops, AddrMode4, IndexModeNone, itin, asm, "", pattern> {
+}
+
 class NLdSt<dag oops, dag iops, InstrItinClass itin,
             string asm, list<dag> pattern>
   : NeonI<oops, iops, AddrMode6, IndexModeNone, itin, asm, "", pattern> {

diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index 68dac9f49516e868f9a8a1f4264b71005ec52b09..599ea1a801f679e65062d64688e0cd6ade03dc7a 100644 (file)
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -298,7 +298,7 @@ def am3offset : Operand<i32>,
 // addrmode4 := reg, <mode|W>
 //
 def addrmode4 : Operand<i32>,
-                ComplexPattern<i32, 2, "", []> {
+                ComplexPattern<i32, 2, "SelectAddrMode4", []> {
   let PrintMethod = "printAddrMode4Operand";
   let MIOperandInfo = (ops GPR, i32imm);
 }

diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index afa42d57ec2bff51c69304ee104d4011152b96e4..36cd094ee837dbcecd3d47754f915c2c8eeedcdf 100644 (file)
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -138,10 +138,10 @@ def VLDMS : NI<(outs),
 */
 
 // Use vldmia to load a Q register as a D register pair.
-def VLDRQ : NI<(outs QPR:$dst), (ins GPR:$addr),
+def VLDRQ : NI4<(outs QPR:$dst), (ins addrmode4:$addr),
                NoItinerary,
                "vldmia $addr, ${dst:dregpair}",
-               [(set QPR:$dst, (v2f64 (load GPR:$addr)))]> {
+               [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]> {
   let Inst{27-25} = 0b110;
   let Inst{24}    = 0; // P bit
   let Inst{23}    = 1; // U bit
@@ -150,10 +150,10 @@ def VLDRQ : NI<(outs QPR:$dst), (ins GPR:$addr),
 }
 
 // Use vstmia to store a Q register as a D register pair.
-def VSTRQ : NI<(outs), (ins QPR:$src, GPR:$addr),
+def VSTRQ : NI4<(outs), (ins QPR:$src, addrmode4:$addr),
                NoItinerary,
                "vstmia $addr, ${src:dregpair}",
-               [(store (v2f64 QPR:$src), GPR:$addr)]> {
+               [(store (v2f64 QPR:$src), addrmode4:$addr)]> {
   let Inst{27-25} = 0b110;
   let Inst{24}    = 0; // P bit
   let Inst{23}    = 1; // U bit
@@ -161,7 +161,6 @@ def VSTRQ : NI<(outs), (ins QPR:$src, GPR:$addr),
   let Inst{11-9}  = 0b101;
 }
 
-
 //   VLD1     : Vector Load (multiple single elements)
 class VLD1D<string OpcodeStr, ValueType Ty, Intrinsic IntOp>
   : NLdSt<(outs DPR:$dst), (ins addrmode6:$addr),