AVX-512: MUL operation lowering for v8i64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193083 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4b11f2b61f3dc36dc9b54f64dcdc68e09ed4e472..dea4a4616f38ef6ee98363d3e17c435062f475fe 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -12402,8 +12402,8 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
     return DAG.getVectorShuffle(VT, dl, Evens, Odds, ShufMask);
   }
 
-  assert((VT == MVT::v2i64 || VT == MVT::v4i64) &&
-         "Only know how to lower V2I64/V4I64 multiply");
+  assert((VT == MVT::v2i64 || VT == MVT::v4i64 || VT == MVT::v8i64) &&
+         "Only know how to lower V2I64/V4I64/V8I64 multiply");
 
   //  Ahi = psrlqi(a, 32);
   //  Bhi = psrlqi(b, 32);
@@ -12422,7 +12422,8 @@ static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
   SDValue Bhi = DAG.getNode(X86ISD::VSRLI, dl, VT, B, ShAmt);
 
   // Bit cast to 32-bit vectors for MULUDQ
-  EVT MulVT = (VT == MVT::v2i64) ? MVT::v4i32 : MVT::v8i32;
+  EVT MulVT = (VT == MVT::v2i64) ? MVT::v4i32 :
+                                  (VT == MVT::v4i64) ? MVT::v8i32 : MVT::v16i32;
   A = DAG.getNode(ISD::BITCAST, dl, MulVT, A);
   B = DAG.getNode(ISD::BITCAST, dl, MulVT, B);
   Ahi = DAG.getNode(ISD::BITCAST, dl, MulVT, Ahi);

diff --git a/test/CodeGen/X86/avx512-arith.ll b/test/CodeGen/X86/avx512-arith.ll
index d5af76fdfa4db9e653522107dae0b0ad9f05a670..e27600ecd734175cad1a26eed12a83d6060536bf 100644 (file)
--- a/test/CodeGen/X86/avx512-arith.ll
+++ b/test/CodeGen/X86/avx512-arith.ll
@@ -74,6 +74,15 @@ entry:
   ret <16 x float> %sub.i
 }
 
+; CHECK-LABEL: imulq512
+; CHECK: vpmuludq
+; CHECK: vpmuludq
+; CHECK: ret
+define <8 x i64> @imulq512(<8 x i64> %y, <8 x i64> %x) {
+  %z = mul <8 x i64>%x, %y
+  ret <8 x i64>%z
+}
+
 ; CHECK-LABEL: mulpd512
 ; CHECK: vmulpd
 ; CHECK: ret
@@ -259,4 +268,4 @@ entry:
   %c = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer
   %d = and <8 x i64> %p1, %c
   ret <8 x i64>%d
-}
\ No newline at end of file
+}