Add support for additional in-reg vbroadcast patterns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@157127 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 439cd4f862524bf235979a6add2b57ea1f10bda3..203c8733aae6b1a5f75a69111f2ed0395be599ef 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -5026,12 +5026,18 @@ X86TargetLowering::LowerVectorBroadcast(SDValue &Op, SelectionDAG &DAG) const {
     }
   }
 
+  bool IsLoad = ISD::isNormalLoad(Ld.getNode());
+  unsigned ScalarSize = Ld.getValueType().getSizeInBits();
+
+  // Handle AVX2 in-register broadcasts.
+  if (!IsLoad && Subtarget->hasAVX2() &&
+      (ScalarSize == 32 || (Is256 && ScalarSize == 64)))
+    return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
+
   // The scalar source must be a normal load.
-  if (!ISD::isNormalLoad(Ld.getNode()))
+  if (!IsLoad)
     return SDValue();
 
-  unsigned ScalarSize = Ld.getValueType().getSizeInBits();
-
   if (ScalarSize == 32 || (Is256 && ScalarSize == 64))
     return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
 

diff --git a/test/CodeGen/X86/avx2-vbroadcast.ll b/test/CodeGen/X86/avx2-vbroadcast.ll
index 6eba694bd4a39bebad5b81bb455c806fb26dcf2b..46b41fa953b4ef9f694725311e47cc7d5fd64601 100644 (file)
--- a/test/CodeGen/X86/avx2-vbroadcast.ll
+++ b/test/CodeGen/X86/avx2-vbroadcast.ll
@@ -222,3 +222,40 @@ footer349VF:
 ret:
   ret void
 }
+
+; CHECK: _inreg0
+; CHECK: broadcastss
+; CHECK: ret
+define <8 x i32> @_inreg0(i32 %scalar) nounwind uwtable readnone ssp {
+  %in = insertelement <8 x i32> undef, i32 %scalar, i32 0
+  %wide = shufflevector <8 x i32> %in, <8 x i32> undef, <8 x i32> zeroinitializer
+  ret <8 x i32> %wide
+}
+
+; CHECK: _inreg1
+; CHECK: broadcastss
+; CHECK: ret
+define <8 x float> @_inreg1(float %scalar) nounwind uwtable readnone ssp {
+  %in = insertelement <8 x float> undef, float %scalar, i32 0
+  %wide = shufflevector <8 x float> %in, <8 x float> undef, <8 x i32> zeroinitializer
+  ret <8 x float> %wide
+}
+
+; CHECK: _inreg2
+; CHECK: broadcastss
+; CHECK: ret
+define <4 x float> @_inreg2(float %scalar) nounwind uwtable readnone ssp {
+  %in = insertelement <4 x float> undef, float %scalar, i32 0
+  %wide = shufflevector <4 x float> %in, <4 x float> undef, <4 x i32> zeroinitializer
+  ret <4 x float> %wide
+}
+
+; CHECK: _inreg3
+; CHECK: broadcastsd
+; CHECK: ret
+define <4 x double> @_inreg3(double %scalar) nounwind uwtable readnone ssp {
+  %in = insertelement <4 x double> undef, double %scalar, i32 0
+  %wide = shufflevector <4 x double> %in, <4 x double> undef, <4 x i32> zeroinitializer
+  ret <4 x double> %wide
+}
+