From 27e01e675c1acd9f648847d96f296b611f22628f Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Fri, 26 Apr 2019 18:02:14 +0000
Subject: [PATCH] [X86][AVX] Fold extract_subvector(broadcast(x)) ->
 broadcast(x) iff x has one use

llvm-svn: 359332
---
 llvm/lib/Target/X86/X86ISelLowering.cpp                  |  7 +++++++
 llvm/test/CodeGen/X86/avx512-hadd-hsub.ll                |  6 ++----
 llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll |  2 +-
 llvm/test/CodeGen/X86/vector-fshl-rot-256.ll             | 14 +++++++-------
 llvm/test/CodeGen/X86/vector-fshl-rot-512.ll             | 16 ++++++++--------
 llvm/test/CodeGen/X86/vector-fshr-rot-256.ll             | 14 +++++++-------
 llvm/test/CodeGen/X86/vector-fshr-rot-512.ll             | 16 ++++++++--------
 llvm/test/CodeGen/X86/vector-rotate-256.ll               | 12 ++++++------
 llvm/test/CodeGen/X86/vector-rotate-512.ll               |  8 ++++----
 9 files changed, 50 insertions(+), 45 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 7aaf71a..23eec37 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -42590,6 +42590,13 @@ static SDValue combineExtractSubvector(SDNode *N, SelectionDAG &DAG,
         VT, SDLoc(N),
         InVec.getNode()->ops().slice(IdxVal, VT.getVectorNumElements()));
 
+  // If we're extracting from a broadcast then we're better off just
+  // broadcasting to the smaller type directly, assuming this is the only use.
+  // As its a broadcast we don't care about the extraction index.
+  if (InVec.getOpcode() == X86ISD::VBROADCAST && InVec.hasOneUse() &&
+      InVec.getOperand(0).getValueSizeInBits() <= VT.getSizeInBits())
+    return DAG.getNode(X86ISD::VBROADCAST, SDLoc(N), VT, InVec.getOperand(0));
+
   // If we're extracting the lowest subvector and we're the only user,
   // we may be able to perform this with a smaller vector width.
   if (IdxVal == 0 && InVec.hasOneUse()) {
diff --git a/llvm/test/CodeGen/X86/avx512-hadd-hsub.ll b/llvm/test/CodeGen/X86/avx512-hadd-hsub.ll
index f16d3bb..283f950 100644
--- a/llvm/test/CodeGen/X86/avx512-hadd-hsub.ll
+++ b/llvm/test/CodeGen/X86/avx512-hadd-hsub.ll
@@ -249,8 +249,7 @@ define double @fsub_noundef_ee (<8 x double> %x225, <8 x double> %x227) {
 ; KNL-LABEL: fsub_noundef_ee:
 ; KNL:       # %bb.0:
 ; KNL-NEXT:    vextractf32x4 $2, %zmm1, %xmm0
-; KNL-NEXT:    vbroadcastsd %xmm0, %zmm1
-; KNL-NEXT:    vextractf32x4 $2, %zmm1, %xmm1
+; KNL-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
 ; KNL-NEXT:    vsubpd %xmm0, %xmm1, %xmm0
 ; KNL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
 ; KNL-NEXT:    retq
@@ -258,8 +257,7 @@ define double @fsub_noundef_ee (<8 x double> %x225, <8 x double> %x227) {
 ; SKX-LABEL: fsub_noundef_ee:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vextractf32x4 $2, %zmm1, %xmm0
-; SKX-NEXT:    vbroadcastsd %xmm0, %zmm1
-; SKX-NEXT:    vextractf32x4 $2, %zmm1, %xmm1
+; SKX-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
 ; SKX-NEXT:    vsubpd %xmm0, %xmm1, %xmm0
 ; SKX-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
 ; SKX-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll b/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
index eca9506..abbeff1 100644
--- a/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
+++ b/llvm/test/CodeGen/X86/avx512-shuffles/partial_permute.ll
@@ -1819,7 +1819,7 @@ define <4 x i32> @test_16xi32_to_4xi32_perm_mask9(<16 x i32> %vec) {
 ; CHECK-NEXT:    vextractf64x4 $1, %zmm0, %ymm2
 ; CHECK-NEXT:    vpermps %ymm2, %ymm1, %ymm1
 ; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0
+; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0
 ; CHECK-NEXT:    vblendps {{.*#+}} xmm0 = xmm1[0,1],xmm0[2],xmm1[3]
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
index 75fb60f..b56b555 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-256.ll
@@ -542,7 +542,7 @@ define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX2-NEXT:    vpsllq %xmm3, %ymm0, %ymm3
@@ -625,7 +625,7 @@ define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
@@ -710,7 +710,7 @@ define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounw
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
@@ -723,7 +723,7 @@ define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounw
 ;
 ; AVX512-LABEL: splatvar_funnnel_v16i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
@@ -789,7 +789,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
@@ -810,7 +810,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
@@ -831,7 +831,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
diff --git a/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll b/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
index 70dddd45..c096a6c 100644
--- a/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshl-rot-512.ll
@@ -316,7 +316,7 @@ define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %amt) nounw
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -332,7 +332,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -348,7 +348,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm1, %zmm1
+; AVX512BW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -363,7 +363,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %zmm1
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -383,7 +383,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -409,7 +409,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -435,7 +435,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %zmm1
+; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -458,7 +458,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %zmm1
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
index f77423b..bc477c2e 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-256.ll
@@ -587,7 +587,7 @@ define <4 x i64> @splatvar_funnnel_v4i64(<4 x i64> %x, <4 x i64> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v4i64:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastq %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastq %xmm1, %xmm1
 ; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [63,63]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX2-NEXT:    vpsrlq %xmm3, %ymm0, %ymm3
@@ -676,7 +676,7 @@ define <8 x i32> @splatvar_funnnel_v8i32(<8 x i32> %x, <8 x i32> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
@@ -769,7 +769,7 @@ define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounw
 ;
 ; AVX2-LABEL: splatvar_funnnel_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
@@ -784,7 +784,7 @@ define <16 x i16> @splatvar_funnnel_v16i16(<16 x i16> %x, <16 x i16> %amt) nounw
 ;
 ; AVX512-LABEL: splatvar_funnnel_v16i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512-NEXT:    vpsubw %xmm1, %xmm2, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
@@ -857,7 +857,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX2-LABEL: splatvar_funnnel_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
@@ -880,7 +880,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX512F-LABEL: splatvar_funnnel_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512F-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
@@ -903,7 +903,7 @@ define <32 x i8> @splatvar_funnnel_v32i8(<32 x i8> %x, <32 x i8> %amt) nounwind
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpsubb %xmm1, %xmm2, %xmm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
diff --git a/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll b/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
index 7746e70..ee8e2a9 100644
--- a/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
+++ b/llvm/test/CodeGen/X86/vector-fshr-rot-512.ll
@@ -328,7 +328,7 @@ define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %amt) nounw
 define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
@@ -346,7 +346,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpsubw %xmm2, %xmm3, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
@@ -364,7 +364,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastw %xmm1, %zmm1
+; AVX512BW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -379,7 +379,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %zmm1
+; AVX512VLBW-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero
@@ -399,7 +399,7 @@ define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounw
 define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
 ; AVX512F-LABEL: splatvar_funnnel_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512F-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
@@ -427,7 +427,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512VL-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpxor %xmm3, %xmm3, %xmm3
 ; AVX512VL-NEXT:    vpsubb %xmm2, %xmm3, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
@@ -455,7 +455,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512BW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpbroadcastb %xmm1, %zmm1
+; AVX512BW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512BW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512BW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512BW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
@@ -478,7 +478,7 @@ define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind
 ;
 ; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
 ; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %zmm1
+; AVX512VLBW-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VLBW-NEXT:    vmovdqa {{.*#+}} xmm2 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
 ; AVX512VLBW-NEXT:    vpand %xmm2, %xmm1, %xmm3
 ; AVX512VLBW-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,zero,zero,zero,zero,xmm3[1],zero,zero,zero,zero,zero,zero,zero
diff --git a/llvm/test/CodeGen/X86/vector-rotate-256.ll b/llvm/test/CodeGen/X86/vector-rotate-256.ll
index ac50c36..d7ca744 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-256.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-256.ll
@@ -609,7 +609,7 @@ define <8 x i32> @splatvar_rotate_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
 ;
 ; AVX2-LABEL: splatvar_rotate_v8i32:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastd %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastd %xmm1, %xmm1
 ; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [31,31,31,31]
 ; AVX2-NEXT:    vpand %xmm2, %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxdq {{.*#+}} xmm2 = xmm1[0],zero,xmm1[1],zero
@@ -697,7 +697,7 @@ define <16 x i16> @splatvar_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; AVX2-LABEL: splatvar_rotate_v16i16:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
@@ -710,7 +710,7 @@ define <16 x i16> @splatvar_rotate_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind
 ;
 ; AVX512-LABEL: splatvar_rotate_v16i16:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpbroadcastw %xmm1, %ymm1
+; AVX512-NEXT:    vpbroadcastw %xmm1, %xmm1
 ; AVX512-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512-NEXT:    vpmovzxwq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero
 ; AVX512-NEXT:    vpsllw %xmm2, %ymm0, %ymm2
@@ -779,7 +779,7 @@ define <32 x i8> @splatvar_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; AVX2-LABEL: splatvar_rotate_v32i8:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX2-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX2-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
@@ -800,7 +800,7 @@ define <32 x i8> @splatvar_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; AVX512F-LABEL: splatvar_rotate_v32i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512F-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
@@ -821,7 +821,7 @@ define <32 x i8> @splatvar_rotate_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind {
 ;
 ; AVX512VL-LABEL: splatvar_rotate_v32i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm1, %ymm1
+; AVX512VL-NEXT:    vpbroadcastb %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm1, %xmm1
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm2 = xmm1[0],zero,zero,zero,zero,zero,zero,zero,xmm1[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm2, %ymm0, %ymm3
diff --git a/llvm/test/CodeGen/X86/vector-rotate-512.ll b/llvm/test/CodeGen/X86/vector-rotate-512.ll
index 01b4676..54725e3 100644
--- a/llvm/test/CodeGen/X86/vector-rotate-512.ll
+++ b/llvm/test/CodeGen/X86/vector-rotate-512.ll
@@ -311,7 +311,7 @@ define <16 x i32> @splatvar_rotate_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind
 define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind {
 ; AVX512F-LABEL: splatvar_rotate_v32i16:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -327,7 +327,7 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 ;
 ; AVX512VL-LABEL: splatvar_rotate_v32i16:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastw %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastw %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxwq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -373,7 +373,7 @@ define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind
 define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ; AVX512F-LABEL: splatvar_rotate_v64i8:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512F-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512F-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512F-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512F-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
@@ -399,7 +399,7 @@ define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind {
 ;
 ; AVX512VL-LABEL: splatvar_rotate_v64i8:
 ; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpbroadcastb %xmm2, %ymm2
+; AVX512VL-NEXT:    vpbroadcastb %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpand {{.*}}(%rip), %xmm2, %xmm2
 ; AVX512VL-NEXT:    vpmovzxbq {{.*#+}} xmm3 = xmm2[0],zero,zero,zero,zero,zero,zero,zero,xmm2[1],zero,zero,zero,zero,zero,zero,zero
 ; AVX512VL-NEXT:    vpsllw %xmm3, %ymm0, %ymm4
-- 
2.7.4