From 65ffec49f73d1f8856211b107712c58cc9636b78 Mon Sep 17 00:00:00 2001
From: Bob Wilson <bob.wilson@apple.com>
Date: Tue, 21 Sep 2010 17:56:22 +0000
Subject: [PATCH] Define the TargetLowering::getTgtMemIntrinsic hook for ARM so
 that NEON load and store intrinsics are represented with MemIntrinsicSDNodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114454 91177308-0d34-0410-b5e6-96231b3b80d8
---
 .../SelectionDAG/SelectionDAGBuilder.cpp      |  3 +-
 lib/Target/ARM/ARMISelLowering.cpp            | 61 +++++++++++++++++++
 lib/Target/ARM/ARMISelLowering.h              |  3 +
 3 files changed, 66 insertions(+), 1 deletion(-)

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 66323999cc9..7370a0bb27e 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -3031,7 +3031,8 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
   bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I, Intrinsic);
 
   // Add the intrinsic ID as an integer operand if it's not a target intrinsic.
-  if (!IsTgtIntrinsic)
+  if (!IsTgtIntrinsic || Info.opc == ISD::INTRINSIC_VOID ||
+      Info.opc == ISD::INTRINSIC_W_CHAIN)
     Ops.push_back(DAG.getConstant(Intrinsic, TLI.getPointerTy()));
 
   // Add all operands of the call to the operand list.
diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 9e824c691f2..db061c383b5 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -29,6 +29,7 @@
 #include "llvm/Function.h"
 #include "llvm/GlobalValue.h"
 #include "llvm/Instruction.h"
+#include "llvm/Instructions.h"
 #include "llvm/Intrinsics.h"
 #include "llvm/Type.h"
 #include "llvm/CodeGen/CallingConvLower.h"
@@ -5542,3 +5543,63 @@ bool ARMTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
     return ARM::getVFPf64Imm(Imm) != -1;
   return false;
 }
+
+/// getTgtMemIntrinsic - Represent NEON load and store intrinsics as 
+/// MemIntrinsicNodes.  The associated MachineMemOperands record the alignment
+/// specified in the intrinsic calls.
+bool ARMTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
+                                           const CallInst &I,
+                                           unsigned Intrinsic) const {
+  switch (Intrinsic) {
+  case Intrinsic::arm_neon_vld1:
+  case Intrinsic::arm_neon_vld2:
+  case Intrinsic::arm_neon_vld3:
+  case Intrinsic::arm_neon_vld4:
+  case Intrinsic::arm_neon_vld2lane:
+  case Intrinsic::arm_neon_vld3lane:
+  case Intrinsic::arm_neon_vld4lane: {
+    Info.opc = ISD::INTRINSIC_W_CHAIN;
+    // Conservatively set memVT to the entire set of vectors loaded.
+    uint64_t NumElts = getTargetData()->getTypeAllocSize(I.getType()) / 8;
+    Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
+    Info.ptrVal = I.getArgOperand(0);
+    Info.offset = 0;
+    Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
+    Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
+    Info.vol = false; // volatile loads with NEON intrinsics not supported
+    Info.readMem = true;
+    Info.writeMem = false;
+    return true;
+  }
+  case Intrinsic::arm_neon_vst1:
+  case Intrinsic::arm_neon_vst2:
+  case Intrinsic::arm_neon_vst3:
+  case Intrinsic::arm_neon_vst4:
+  case Intrinsic::arm_neon_vst2lane:
+  case Intrinsic::arm_neon_vst3lane:
+  case Intrinsic::arm_neon_vst4lane: {
+    Info.opc = ISD::INTRINSIC_VOID;
+    // Conservatively set memVT to the entire set of vectors stored.
+    unsigned NumElts = 0;
+    for (unsigned ArgI = 1, ArgE = I.getNumArgOperands(); ArgI < ArgE; ++ArgI) {
+      const Type *ArgTy = I.getArgOperand(ArgI)->getType();
+      if (!ArgTy->isVectorTy())
+        break;
+      NumElts += getTargetData()->getTypeAllocSize(ArgTy) / 8;
+    }
+    Info.memVT = EVT::getVectorVT(I.getType()->getContext(), MVT::i64, NumElts);
+    Info.ptrVal = I.getArgOperand(0);
+    Info.offset = 0;
+    Value *AlignArg = I.getArgOperand(I.getNumArgOperands() - 1);
+    Info.align = cast<ConstantInt>(AlignArg)->getZExtValue();
+    Info.vol = false; // volatile stores with NEON intrinsics not supported
+    Info.readMem = false;
+    Info.writeMem = true;
+    return true;
+  }
+  default:
+    break;
+  }
+
+  return false;
+}
diff --git a/lib/Target/ARM/ARMISelLowering.h b/lib/Target/ARM/ARMISelLowering.h
index 58b8b9eb44d..a00d2ab5426 100644
--- a/lib/Target/ARM/ARMISelLowering.h
+++ b/lib/Target/ARM/ARMISelLowering.h
@@ -290,6 +290,9 @@ namespace llvm {
     /// materialize the FP immediate as a load from a constant pool.
     virtual bool isFPImmLegal(const APFloat &Imm, EVT VT) const;
 
+    virtual bool getTgtMemIntrinsic(IntrinsicInfo &Info,
+                                    const CallInst &I,
+                                    unsigned Intrinsic) const;
   protected:
     std::pair<const TargetRegisterClass*, uint8_t>
     findRepresentativeClass(EVT VT) const;