[X86] Make use of vpmovwb when possible in LowerMULH

If we have BWI, we can truncate in a much simpler way by using vpmovwb. This even works without VLX by using the wider zmm->ymm truncate with a subvector extract.

Differential Revision: https://reviews.llvm.org/D38375

llvm-svn: 314457

1 files changed, 8 insertions, 15 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d762c985e90..980db1012be 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -21631,17 +21631,7 @@ static SDValue LowerMULH(SDValue Op, const X86Subtarget &Subtarget,
         SDValue Mul = DAG.getNode(ISD::MUL, dl, MVT::v32i16, ExA, ExB);
         Mul = DAG.getNode(ISD::SRL, dl, MVT::v32i16, Mul,
                           DAG.getConstant(8, dl, MVT::v32i16));
-        // The ymm variant of PACKUS treats the 128-bit lanes separately, so
-        // before using PACKUS we need to permute the inputs to the correct
-        // lo/hi xmm lane.
-        const int Mask[] = { 0,  1,  2,  3,  4,  5,  6,  7,
-                            16, 17, 18, 19, 20, 21, 22, 23,
-                             8,  9, 10, 11, 12, 13, 14, 15,
-                            24, 25, 26, 27, 28, 29, 30, 31};
-        Mul = DAG.getVectorShuffle(MVT::v32i16, dl, Mul, Mul, Mask);
-        Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v16i16, Mul, Lo);
-        Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v16i16, Mul, Hi);
-        return DAG.getNode(X86ISD::PACKUS, dl, VT, Lo, Hi);
+        return DAG.getNode(ISD::TRUNCATE, dl, VT, Mul);
       }
       SDValue ALo = extract128BitVector(A, 0, DAG, dl);
       SDValue BLo = extract128BitVector(B, 0, DAG, dl);
@@ -21671,10 +21661,13 @@ static SDValue LowerMULH(SDValue Op, const X86Subtarget &Subtarget,
     SDValue ExA = DAG.getNode(ExAVX, dl, MVT::v16i16, A);
     SDValue ExB = DAG.getNode(ExAVX, dl, MVT::v16i16, B);
     SDValue Mul = DAG.getNode(ISD::MUL, dl, MVT::v16i16, ExA, ExB);
-    SDValue MulH = DAG.getNode(ISD::SRL, dl, MVT::v16i16, Mul,
-                               DAG.getConstant(8, dl, MVT::v16i16));
-    Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v8i16, MulH, Lo);
-    Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v8i16, MulH, Hi);
+    Mul = DAG.getNode(ISD::SRL, dl, MVT::v16i16, Mul,
+                      DAG.getConstant(8, dl, MVT::v16i16));
+    // If we have BWI we can use truncate instruction.
+    if (Subtarget.hasBWI())
+      return DAG.getNode(ISD::TRUNCATE, dl, VT, Mul);
+    Lo = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v8i16, Mul, Lo);
+    Hi = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v8i16, Mul, Hi);
     return DAG.getNode(X86ISD::PACKUS, dl, VT, Lo, Hi);
   }