X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FX86%2FUtils%2FX86ShuffleDecode.cpp;h=04eb9c4e5037f0529347eec57de6161fe9c61e3c;hp=e66c0ff11ebb4807d1c06866315ab1ba9181b588;hb=f657b6395ac5c9fa57a3d4c871d7f32c565b11fe;hpb=fe1cc51fd12dfe817d2539af9e01f851d0757036

diff --git a/lib/Target/X86/Utils/X86ShuffleDecode.cpp b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
index e66c0ff11eb..04eb9c4e503 100644
--- a/lib/Target/X86/Utils/X86ShuffleDecode.cpp
+++ b/lib/Target/X86/Utils/X86ShuffleDecode.cpp
@@ -264,6 +264,26 @@ void DecodeUNPCKLMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
   }
 }
 
+/// \brief Decode a shuffle packed values at 128-bit granularity
+/// (SHUFF32x4/SHUFF64x2/SHUFI32x4/SHUFI64x2)
+/// immediate mask into a shuffle mask.
+void decodeVSHUF64x2FamilyMask(MVT VT, unsigned Imm,
+                        SmallVectorImpl<int> &ShuffleMask) {
+  unsigned NumLanes = VT.getSizeInBits() / 128;
+  unsigned NumElementsInLane = 128 / VT.getScalarSizeInBits();
+  unsigned ControlBitsMask = NumLanes - 1;
+  unsigned NumControlBits  = NumLanes / 2;
+
+  for (unsigned l = 0; l != NumLanes; ++l) {
+    unsigned LaneMask = (Imm >> (l * NumControlBits)) & ControlBitsMask;
+    // We actually need the other source.
+    if (l >= NumLanes / 2)
+      LaneMask += NumLanes;
+    for (unsigned i = 0; i != NumElementsInLane; ++i)
+      ShuffleMask.push_back(LaneMask * NumElementsInLane + i);
+  }
+}
+
 void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
                           SmallVectorImpl<int> &ShuffleMask) {
   unsigned HalfSize = VT.getVectorNumElements() / 2;