From 87aeff8bbb47a4eafdd83e114220af4f2cd56a28 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Thu, 28 Feb 2019 10:53:58 +0000
Subject: [PATCH] [X86][AVX] Fold vf64 concat_vectors(movddup(x),movddup(x)) ->
 broadcast(x)

llvm-svn: 355078
---
 llvm/lib/Target/X86/X86ISelLowering.cpp    | 12 +++++++++++-
 llvm/test/CodeGen/X86/widened-broadcast.ll | 19 ++++---------------
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 460c95a..76f6578 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -41777,6 +41777,8 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
                                       ArrayRef<SDValue> Ops, SelectionDAG &DAG,
                                       TargetLowering::DAGCombinerInfo &DCI,
                                       const X86Subtarget &Subtarget) {
+  assert(Subtarget.hasAVX() && "AVX assumed for concat_vectors");
+
   if (llvm::all_of(Ops, [](SDValue Op) { return Op.isUndef(); }))
     return DAG.getUNDEF(VT);
 
@@ -41811,8 +41813,16 @@ static SDValue combineConcatVectorOps(const SDLoc &DL, MVT VT,
         Op0.getOpcode() == X86ISD::SUBV_BROADCAST)
       return DAG.getNode(Op0.getOpcode(), DL, VT, Op0.getOperand(0));
 
+    // concat_vectors(movddup(x),movddup(x)) -> broadcast(x)
+    if (Op0.getOpcode() == X86ISD::MOVDDUP && VT == MVT::v4f64 &&
+        (Subtarget.hasAVX2() || MayFoldLoad(Op0.getOperand(0))))
+      return DAG.getNode(X86ISD::VBROADCAST, DL, VT,
+                         DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, MVT::f64,
+                                     Op0.getOperand(0),
+                                     DAG.getIntPtrConstant(0, DL)));
+
     // concat_vectors(scalar_to_vector(x),scalar_to_vector(x)) -> broadcast(x)
-    if (Op0.getOpcode() == ISD::SCALAR_TO_VECTOR && Subtarget.hasAVX() &&
+    if (Op0.getOpcode() == ISD::SCALAR_TO_VECTOR &&
         (Subtarget.hasAVX2() ||
          (VT.getScalarSizeInBits() >= 32 && MayFoldLoad(Op0.getOperand(0)))) &&
         Op0.getOperand(0).getValueType() == VT.getScalarType())
diff --git a/llvm/test/CodeGen/X86/widened-broadcast.ll b/llvm/test/CodeGen/X86/widened-broadcast.ll
index ae4f901..b43c8a4 100644
--- a/llvm/test/CodeGen/X86/widened-broadcast.ll
+++ b/llvm/test/CodeGen/X86/widened-broadcast.ll
@@ -43,21 +43,10 @@ define <8 x float> @load_splat_8f32_4f32_01010101(<4 x float>* %ptr) nounwind uw
 ; SSE42-NEXT:    movapd %xmm0, %xmm1
 ; SSE42-NEXT:    retq
 ;
-; AVX1-LABEL: load_splat_8f32_4f32_01010101:
-; AVX1:       # %bb.0: # %entry
-; AVX1-NEXT:    vmovddup {{.*#+}} xmm0 = mem[0,0]
-; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: load_splat_8f32_4f32_01010101:
-; AVX2:       # %bb.0: # %entry
-; AVX2-NEXT:    vbroadcastsd (%rdi), %ymm0
-; AVX2-NEXT:    retq
-;
-; AVX512-LABEL: load_splat_8f32_4f32_01010101:
-; AVX512:       # %bb.0: # %entry
-; AVX512-NEXT:    vbroadcastsd (%rdi), %ymm0
-; AVX512-NEXT:    retq
+; AVX-LABEL: load_splat_8f32_4f32_01010101:
+; AVX:       # %bb.0: # %entry
+; AVX-NEXT:    vbroadcastsd (%rdi), %ymm0
+; AVX-NEXT:    retq
 entry:
   %ld = load <4 x float>, <4 x float>* %ptr
   %ret = shufflevector <4 x float> %ld, <4 x float> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
-- 
2.7.4