Add support to use pextrw and pinsrw to extract and insert a word element

from a 128-bit vector.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27304 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 69 insertions, 4 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 8036066d0f..cfc6af84ca 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -255,9 +255,9 @@ X86TargetLowering::X86TargetLowering(TargetMachine &TM)
     setOperationAction(ISD::SUB , (MVT::ValueType)VT, Expand);
     setOperationAction(ISD::MUL , (MVT::ValueType)VT, Expand);
     setOperationAction(ISD::LOAD, (MVT::ValueType)VT, Expand);
-    setOperationAction(ISD::VECTOR_SHUFFLE, (MVT::ValueType)VT, Expand);
+    setOperationAction(ISD::VECTOR_SHUFFLE,     (MVT::ValueType)VT, Expand);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
-    setOperationAction(ISD::INSERT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
+    setOperationAction(ISD::INSERT_VECTOR_ELT,  (MVT::ValueType)VT, Expand);
   }
 
   if (Subtarget->hasMMX()) {
@@ -316,6 +316,8 @@ X86TargetLowering::X86TargetLowering(TargetMachine &TM)
     setOperationAction(ISD::VECTOR_SHUFFLE,   MVT::v8i16, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,   MVT::v4i32, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,   MVT::v2i64, Custom);
+    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v8i16, Custom);
+    setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v8i16, Custom);
   }
 
   computeRegisterProperties();
@@ -2657,6 +2659,37 @@ SDOperand X86TargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
 
     return SDOperand();
   }
+  case ISD::EXTRACT_VECTOR_ELT: {
+    // Transform it so it match pextrw which produces a 32-bit result.
+    MVT::ValueType VT = Op.getValueType();
+    if (MVT::getSizeInBits(VT) == 16) {
+      MVT::ValueType EVT = (MVT::ValueType)(VT+1);
+      SDOperand Extract = DAG.getNode(X86ISD::PEXTRW, EVT,
+                                      Op.getOperand(0), Op.getOperand(1));
+      SDOperand Assert  = DAG.getNode(ISD::AssertZext, EVT, Extract,
+                                      DAG.getValueType(VT));
+      return DAG.getNode(ISD::TRUNCATE, VT, Assert);
+    }
+
+    return SDOperand();
+  }
+  case ISD::INSERT_VECTOR_ELT: {
+    // Transform it so it match pinsrw which expects a 16-bit value in a R32
+    // as its second argument.
+    MVT::ValueType VT = Op.getValueType();
+    MVT::ValueType BaseVT = MVT::getVectorBaseType(VT);
+    if (MVT::getSizeInBits(BaseVT) == 16) {
+      SDOperand N1 = Op.getOperand(1);
+      SDOperand N2 = Op.getOperand(2);
+      if (N1.getValueType() != MVT::i32)
+        N1 = DAG.getNode(ISD::ANY_EXTEND, MVT::i32, N1);
+      if (N2.getValueType() != MVT::i32)
+        N2 = DAG.getConstant(cast<ConstantSDNode>(N2)->getValue(), MVT::i32);
+      return DAG.getNode(ISD::INSERT_VECTOR_ELT, VT, Op.getOperand(0), N1, N2);
+    }
+
+    return SDOperand();
+  }
   }
 }
 
@@ -2692,6 +2725,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::Wrapper:            return "X86ISD::Wrapper";
   case X86ISD::S2VEC:              return "X86ISD::S2VEC";
   case X86ISD::ZEXT_S2VEC:         return "X86ISD::ZEXT_S2VEC";
+  case X86ISD::PEXTRW:             return "X86ISD::PEXTRW";
   }
 }
 
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 3bd2b88c6c..3f8ec844f4 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -153,6 +153,10 @@ namespace llvm {
       /// ZEXT_S2VEC - SCALAR_TO_VECTOR with zero extension. The destination base
       /// does not have to match the operand type.
       ZEXT_S2VEC,
+
+      /// PEXTRW - Extract a 16-bit value from a vector and zero extend it to
+      /// i32, corresponds to X86::PINSRW.
+      PEXTRW,
     };
 
     // X86 specific condition code. These correspond to X86_*_COND in
diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 2245a2218f..38158a24f4 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -28,8 +28,8 @@ def X86s2vec   : SDNode<"X86ISD::S2VEC",
 def X86zexts2vec : SDNode<"X86ISD::ZEXT_S2VEC",
                           SDTypeProfile<1, 1, []>, []>;
 
-def SDTUnpckl : SDTypeProfile<1, 2,
-                              [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>]>;
+def X86pextrw  : SDNode<"X86ISD::PEXTRW",
+                        SDTypeProfile<1, 2, []>, []>;
 
 //===----------------------------------------------------------------------===//
 // SSE pattern fragments
@@ -1409,6 +1409,33 @@ def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
                                 UNPCKH_shuffle_mask)))]>;
 }
 
+// Extract / Insert
+def PEXTRWrr : PDIi8<0xC5, MRMSrcReg,
+                     (ops R32:$dst, VR128:$src1, i32i8imm:$src2),
+                     "pextrw {$src2, $src1, $dst|$dst, $src1, $src2}",
+                     [(set R32:$dst, (X86pextrw (v8i16 VR128:$src1),
+                                      (i32 imm:$src2)))]>;
+def PEXTRWrm : PDIi8<0xC5, MRMSrcMem,
+                     (ops R32:$dst, i128mem:$src1, i32i8imm:$src2),
+                     "pextrw {$src2, $src1, $dst|$dst, $src1, $src2}",
+                     [(set R32:$dst, (X86pextrw (loadv8i16 addr:$src1),
+                                      (i32 imm:$src2)))]>;
+
+let isTwoAddress = 1 in {
+def PINSRWrr : PDIi8<0xC4, MRMSrcReg,
+                     (ops VR128:$dst, VR128:$src1, R32:$src2, i32i8imm:$src3),
+                     "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
+                     [(set VR128:$dst, (v8i16 (vector_insert (v8i16 VR128:$src1),
+                               R32:$src2, (i32 imm:$src3))))]>;
+def PINSRWrm : PDIi8<0xC4, MRMSrcMem,
+                     (ops VR128:$dst, VR128:$src1, i16mem:$src2, i32i8imm:$src3),
+                     "pinsrw {$src3, $src2, $dst|$dst, $src2, $src3}",
+                     [(set VR128:$dst,
+                       (v8i16 (vector_insert (v8i16 VR128:$src1),
+                               (i32 (anyext (loadi16 addr:$src2))),
+                               (i32 imm:$src3))))]>;
+}
+
 //===----------------------------------------------------------------------===//
 // Miscellaneous Instructions
 //===----------------------------------------------------------------------===//