; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv7s-none-eabi %s -o - | FileCheck %s

declare <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %shuffle.i.i307, <8 x i8> %shuffle.i27.i308, <8 x i8> %vtbl2.i25.i)

; Check that we get the motivating example:
; The bitcasts force the values to go through the GPRs, whereas
; they are defined on VPRs and used on VPRs.
;
define void @motivatingExample(<2 x i64>* %addr, <8 x i8>* %addr2) {
; CHECK-LABEL: motivatingExample:
; CHECK:       @ %bb.0:
; CHECK-NEXT:    vld1.64 {d16, d17}, [r0]
; CHECK-NEXT:    vldr d18, [r1]
; CHECK-NEXT:    vtbl.8 d16, {d16, d17}, d18
; CHECK-NEXT:    vstr d16, [r1]
; CHECK-NEXT:    bx lr
  %shuffle.i.bc.i309 = load <2 x i64>, <2 x i64>* %addr
  %vtbl2.i25.i = load <8 x i8>, <8 x i8>* %addr2
  %shuffle.i.extract.i310 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 0
  %shuffle.i27.extract.i311 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 1
  %tmp45 = bitcast i64 %shuffle.i.extract.i310 to <8 x i8>
  %tmp46 = bitcast i64 %shuffle.i27.extract.i311 to <8 x i8>
  %vtbl2.i25.i313 = tail call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %tmp45, <8 x i8> %tmp46, <8 x i8> %vtbl2.i25.i)
  store <8 x i8> %vtbl2.i25.i313, <8 x i8>* %addr2
  ret void
}

; Check that we do not perform the transformation for dynamic index.
define void @dynamicIndex(<2 x i64>* %addr, <8 x i8>* %addr2, i32 %index) {
; CHECK-LABEL: dynamicIndex:
; CHECK:       @ %bb.0:
; CHECK-NEXT:    .save {r4, r6, r7, lr}
; CHECK-NEXT:    push {r4, r6, r7, lr}
; CHECK-NEXT:    .setfp r7, sp, #8
; CHECK-NEXT:    add r7, sp, #8
; CHECK-NEXT:    .pad #16
; CHECK-NEXT:    sub sp, #16
; CHECK-NEXT:    mov r4, sp
; CHECK-NEXT:    bfc r4, #0, #4
; CHECK-NEXT:    mov sp, r4
; CHECK-NEXT:    vld1.64 {d16, d17}, [r0]
; CHECK-NEXT:    adds r0, r2, r2
; CHECK-NEXT:    and r2, r0, #3
; CHECK-NEXT:    adds r0, #1
; CHECK-NEXT:    mov r12, sp
; CHECK-NEXT:    and r0, r0, #3
; CHECK-NEXT:    lsls r2, r2, #2
; CHECK-NEXT:    mov r3, r12
; CHECK-NEXT:    vst1.64 {d16, d17}, [r3:128], r2
; CHECK-NEXT:    orr.w r0, r12, r0, lsl #2
; CHECK-NEXT:    sub.w r4, r7, #8
; CHECK-NEXT:    ldr r2, [r3]
; CHECK-NEXT:    ldr r0, [r0]
; CHECK-NEXT:    vldr d18, [r1]
; CHECK-NEXT:    vmov d16, r2, r0
; CHECK-NEXT:    vtbl.8 d16, {d16, d17}, d18
; CHECK-NEXT:    vstr d16, [r1]
; CHECK-NEXT:    mov sp, r4
; CHECK-NEXT:    pop {r4, r6, r7, pc}
  %shuffle.i.bc.i309 = load <2 x i64>, <2 x i64>* %addr
  %vtbl2.i25.i = load <8 x i8>, <8 x i8>* %addr2
  %shuffle.i.extract.i310 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 %index
  %shuffle.i27.extract.i311 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 1
  %tmp45 = bitcast i64 %shuffle.i.extract.i310 to <8 x i8>
  %tmp46 = bitcast i64 %shuffle.i27.extract.i311 to <8 x i8>
  %vtbl2.i25.i313 = tail call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %tmp45, <8 x i8> %tmp46, <8 x i8> %vtbl2.i25.i)
  store <8 x i8> %vtbl2.i25.i313, <8 x i8>* %addr2
  ret void
}

; Check that we do not perform the transformation when there are several uses
; of the result of the bitcast.
define i64 @severalUses(<2 x i64>* %addr, <8 x i8>* %addr2) {
; CHECK-LABEL: severalUses:
; CHECK:       @ %bb.0:
; CHECK-NEXT:    vld1.64 {d16, d17}, [r0]
; CHECK-NEXT:    vmov r0, r2, d16
; CHECK-NEXT:    vldr d18, [r1]
; CHECK-NEXT:    vtbl.8 d16, {d16, d17}, d18
; CHECK-NEXT:    vstr d16, [r1]
; CHECK-NEXT:    mov r1, r2
; CHECK-NEXT:    bx lr
  %shuffle.i.bc.i309 = load <2 x i64>, <2 x i64>* %addr
  %vtbl2.i25.i = load <8 x i8>, <8 x i8>* %addr2
  %shuffle.i.extract.i310 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 0
  %shuffle.i27.extract.i311 = extractelement <2 x i64> %shuffle.i.bc.i309, i32 1
  %tmp45 = bitcast i64 %shuffle.i.extract.i310 to <8 x i8>
  %tmp46 = bitcast i64 %shuffle.i27.extract.i311 to <8 x i8>
  %vtbl2.i25.i313 = tail call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %tmp45, <8 x i8> %tmp46, <8 x i8> %vtbl2.i25.i)
  store <8 x i8> %vtbl2.i25.i313, <8 x i8>* %addr2
  ret i64 %shuffle.i.extract.i310
}