vdup_lane was missing

<rdar://problem/8278732>

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@110420 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Basic/arm_neon.td b/include/clang/Basic/arm_neon.td
index b42755ca72ae5b4efb9554f6c48df7d9f5661cc2..fa6ebb756ca3dccdf37edc6bf5342f3fb1082ed9 100644 (file)
--- a/include/clang/Basic/arm_neon.td
+++ b/include/clang/Basic/arm_neon.td
@@ -64,6 +64,7 @@ class WInst<string p, string t> : Inst<p, t, OP_NONE> {}
 // u: unsigned integer (int/float args)
 // f: float (int args)
 // d: default
+// g: default, ignore 'Q' size modifier.
 // w: double width elements, same num elts
 // n: double width elements, half num elts
 // h: half width elements, double num elts
@@ -239,6 +240,7 @@ def VCREATE: Inst<"dl", "csihfUcUsUiUlPcPsl", OP_CAST>;
 // E.3.19 Set all lanes to same value
 def VDUP_N : Inst<"ds", "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl", OP_DUP>;
 def VMOV_N : Inst<"ds", "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl", OP_DUP>;
+def VDUP_LANE : WInst<"dgi", "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
 
 ////////////////////////////////////////////////////////////////////////////////
 // E.3.20 Combining vectors

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 666be7421ffe28e4fbf7cbe7513b67fb4551fda6..d59131d1dedc8dfe8a1595f345e1e835e7cbdcf8 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -1012,8 +1012,10 @@ const llvm::VectorType *GetNeonType(LLVMContext &C, unsigned type, bool q) {
   return 0;
 }
 
-Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) {
+Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C, bool widen) {
   unsigned nElts = cast<llvm::VectorType>(V->getType())->getNumElements();
+  if (widen)
+    nElts <<= 1;
   SmallVector<Constant*, 16> Indices(nElts, C);
   Value* SV = llvm::ConstantVector::get(Indices.begin(), Indices.size());
   return Builder.CreateShuffleVector(V, V, SV, "lane");
@@ -1201,6 +1203,12 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
     Function *F = CGM.getIntrinsic(Int, Tys, 2);
     return EmitNeonCall(F, Ops, "vcvt_n");
   }
+  case ARM::BI__builtin_neon_vdup_lane_v: 
+    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
+    return EmitNeonSplat(Ops[0], cast<Constant>(Ops[1]));
+  case ARM::BI__builtin_neon_vdupq_lane_v:
+    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
+    return EmitNeonSplat(Ops[0], cast<Constant>(Ops[1]), true);
   case ARM::BI__builtin_neon_vext_v:
   case ARM::BI__builtin_neon_vextq_v: {
     ConstantInt *C = dyn_cast<ConstantInt>(Ops[2]);

diff --git a/lib/CodeGen/CodeGenFunction.h b/lib/CodeGen/CodeGenFunction.h
index a72cc0fd2ba66584ab741afb0e299f1db568468c..5934da0ef913fdfcc017be7f036dcdb42aeaee91 100644 (file)
--- a/lib/CodeGen/CodeGenFunction.h
+++ b/lib/CodeGen/CodeGenFunction.h
@@ -1437,7 +1437,8 @@ public:
                             llvm::SmallVectorImpl<llvm::Value*> &O,
                             const char *name, bool splat = false,
                             unsigned shift = 0, bool rightshift = false);
-  llvm::Value *EmitNeonSplat(llvm::Value *V, llvm::Constant *Idx);
+  llvm::Value *EmitNeonSplat(llvm::Value *V, llvm::Constant *Idx,
+                             bool widen = false);
   llvm::Value *EmitNeonShiftVector(llvm::Value *V, const llvm::Type *Ty,
                                    bool negateForRightShift);