test/CodeGen/ARM/bf16-shuffle.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=arm-eabi -mattr=+v8.6a,+neon -float-abi=hard < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16
; RUN: llc -mtriple=arm-eabi -mattr=+v8.6a,+neon,+bf16 -float-abi=hard < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOFP16
; RUN: llc -mtriple=arm-eabi -mattr=+v8.6a,+neon,+fullfp16,+bf16 -float-abi=hard < %s | FileCheck %s --check-prefixes=CHECK,CHECK-FP16

%struct.float16x4x2_t = type { [2 x <4 x bfloat>] }
%struct.float16x8x2_t = type { [2 x <8 x bfloat>] }

define dso_local <4 x bfloat> @test_vbsl_bf16(<4 x i16> %a, <4 x bfloat> %b, <4 x bfloat> %c) {
; CHECK-LABEL: test_vbsl_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vbsl d0, d1, d2
; CHECK-NEXT:    bx lr
entry:
  %0 = bitcast <4 x i16> %a to <8 x i8>
  %1 = bitcast <4 x bfloat> %b to <8 x i8>
  %2 = bitcast <4 x bfloat> %c to <8 x i8>
  %vbsl_v.i = tail call <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8> %0, <8 x i8> %1, <8 x i8> %2)
  %3 = bitcast <8 x i8> %vbsl_v.i to <4 x bfloat>
  ret <4 x bfloat> %3
}

define dso_local <8 x bfloat> @test_vbslq_bf16(<8 x i16> %a, <8 x bfloat> %b, <8 x bfloat> %c) {
; CHECK-LABEL: test_vbslq_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vbsl q0, q1, q2
; CHECK-NEXT:    bx lr
entry:
  %0 = bitcast <8 x i16> %a to <16 x i8>
  %1 = bitcast <8 x bfloat> %b to <16 x i8>
  %2 = bitcast <8 x bfloat> %c to <16 x i8>
  %vbslq_v.i = tail call <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8> %0, <16 x i8> %1, <16 x i8> %2)
  %3 = bitcast <16 x i8> %vbslq_v.i to <8 x bfloat>
  ret <8 x bfloat> %3
}

define dso_local %struct.float16x4x2_t @test_vzip_bf16(<4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vzip_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vzip.16 d0, d1
; CHECK-NEXT:    bx lr
entry:
  %vzip.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %vzip1.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %.fca.0.0.insert = insertvalue %struct.float16x4x2_t undef, <4 x bfloat> %vzip.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x4x2_t %.fca.0.0.insert, <4 x bfloat> %vzip1.i, 0, 1
  ret %struct.float16x4x2_t %.fca.0.1.insert
}

define dso_local %struct.float16x8x2_t @test_vzipq_bf16(<8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vzipq_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vzip.16 q0, q1
; CHECK-NEXT:    bx lr
entry:
  %vzip.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
  %vzip1.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>
  %.fca.0.0.insert = insertvalue %struct.float16x8x2_t undef, <8 x bfloat> %vzip.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x8x2_t %.fca.0.0.insert, <8 x bfloat> %vzip1.i, 0, 1
  ret %struct.float16x8x2_t %.fca.0.1.insert
}

define dso_local %struct.float16x4x2_t @test_vuzp_bf16(<4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vuzp_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vuzp.16 d0, d1
; CHECK-NEXT:    bx lr
entry:
  %vuzp.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %vuzp1.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %.fca.0.0.insert = insertvalue %struct.float16x4x2_t undef, <4 x bfloat> %vuzp.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x4x2_t %.fca.0.0.insert, <4 x bfloat> %vuzp1.i, 0, 1
  ret %struct.float16x4x2_t %.fca.0.1.insert
}

define dso_local %struct.float16x8x2_t @test_vuzpq_bf16(<8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vuzpq_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vuzp.16 q0, q1
; CHECK-NEXT:    bx lr
entry:
  %vuzp.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
  %vuzp1.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
  %.fca.0.0.insert = insertvalue %struct.float16x8x2_t undef, <8 x bfloat> %vuzp.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x8x2_t %.fca.0.0.insert, <8 x bfloat> %vuzp1.i, 0, 1
  ret %struct.float16x8x2_t %.fca.0.1.insert
}

define dso_local %struct.float16x4x2_t @test_vtrn_bf16(<4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vtrn_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vtrn.16 d0, d1
; CHECK-NEXT:    bx lr
entry:
  %vtrn.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %vtrn1.i = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %.fca.0.0.insert = insertvalue %struct.float16x4x2_t undef, <4 x bfloat> %vtrn.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x4x2_t %.fca.0.0.insert, <4 x bfloat> %vtrn1.i, 0, 1
  ret %struct.float16x4x2_t %.fca.0.1.insert
}

define dso_local %struct.float16x8x2_t @test_vtrnq_bf16(<8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vtrnq_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vtrn.16 q0, q1
; CHECK-NEXT:    bx lr
entry:
  %vtrn.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %vtrn1.i = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %.fca.0.0.insert = insertvalue %struct.float16x8x2_t undef, <8 x bfloat> %vtrn.i, 0, 0
  %.fca.0.1.insert = insertvalue %struct.float16x8x2_t %.fca.0.0.insert, <8 x bfloat> %vtrn1.i, 0, 1
  ret %struct.float16x8x2_t %.fca.0.1.insert
}

define dso_local <4 x bfloat> @test_vmov_n_bf16(float %a.coerce) {
; CHECK-NOFP16-LABEL: test_vmov_n_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    .pad #4
; CHECK-NOFP16-NEXT:    sub sp, sp, #4
; CHECK-NOFP16-NEXT:    vmov r0, s0
; CHECK-NOFP16-NEXT:    strh r0, [sp, #2]
; CHECK-NOFP16-NEXT:    add r0, sp, #2
; CHECK-NOFP16-NEXT:    vld1.16 {d0[]}, [r0:16]
; CHECK-NOFP16-NEXT:    add sp, sp, #4
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: test_vmov_n_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    @ kill: def $s0 killed $s0 def $d0
; CHECK-FP16-NEXT:    vdup.16 d0, d0[0]
; CHECK-FP16-NEXT:    bx lr
entry:
  %0 = bitcast float %a.coerce to i32
  %tmp.0.extract.trunc = trunc i32 %0 to i16
  %1 = bitcast i16 %tmp.0.extract.trunc to bfloat
  %vecinit = insertelement <4 x bfloat> undef, bfloat %1, i32 0
  %vecinit4 = shufflevector <4 x bfloat> %vecinit, <4 x bfloat> undef, <4 x i32> zeroinitializer
  ret <4 x bfloat> %vecinit4
}

define dso_local <8 x bfloat> @test_vmovq_n_bf16(float %a.coerce) {
; CHECK-NOFP16-LABEL: test_vmovq_n_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    .pad #4
; CHECK-NOFP16-NEXT:    sub sp, sp, #4
; CHECK-NOFP16-NEXT:    vmov r0, s0
; CHECK-NOFP16-NEXT:    strh r0, [sp, #2]
; CHECK-NOFP16-NEXT:    add r0, sp, #2
; CHECK-NOFP16-NEXT:    vld1.16 {d0[], d1[]}, [r0:16]
; CHECK-NOFP16-NEXT:    add sp, sp, #4
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: test_vmovq_n_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    @ kill: def $s0 killed $s0 def $d0
; CHECK-FP16-NEXT:    vdup.16 q0, d0[0]
; CHECK-FP16-NEXT:    bx lr
entry:
  %0 = bitcast float %a.coerce to i32
  %tmp.0.extract.trunc = trunc i32 %0 to i16
  %1 = bitcast i16 %tmp.0.extract.trunc to bfloat
  %vecinit = insertelement <8 x bfloat> undef, bfloat %1, i32 0
  %vecinit8 = shufflevector <8 x bfloat> %vecinit, <8 x bfloat> undef, <8 x i32> zeroinitializer
  ret <8 x bfloat> %vecinit8
}

define dso_local <4 x bfloat> @test_vdup_n_bf16(float %a.coerce) {
; CHECK-NOFP16-LABEL: test_vdup_n_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    .pad #4
; CHECK-NOFP16-NEXT:    sub sp, sp, #4
; CHECK-NOFP16-NEXT:    vmov r0, s0
; CHECK-NOFP16-NEXT:    strh r0, [sp, #2]
; CHECK-NOFP16-NEXT:    add r0, sp, #2
; CHECK-NOFP16-NEXT:    vld1.16 {d0[]}, [r0:16]
; CHECK-NOFP16-NEXT:    add sp, sp, #4
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: test_vdup_n_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    @ kill: def $s0 killed $s0 def $d0
; CHECK-FP16-NEXT:    vdup.16 d0, d0[0]
; CHECK-FP16-NEXT:    bx lr
entry:
  %0 = bitcast float %a.coerce to i32
  %tmp.0.extract.trunc = trunc i32 %0 to i16
  %1 = bitcast i16 %tmp.0.extract.trunc to bfloat
  %vecinit = insertelement <4 x bfloat> undef, bfloat %1, i32 0
  %vecinit4 = shufflevector <4 x bfloat> %vecinit, <4 x bfloat> undef, <4 x i32> zeroinitializer
  ret <4 x bfloat> %vecinit4
}

define dso_local <8 x bfloat> @test_vdupq_n_bf16(float %a.coerce) {
; CHECK-NOFP16-LABEL: test_vdupq_n_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    .pad #4
; CHECK-NOFP16-NEXT:    sub sp, sp, #4
; CHECK-NOFP16-NEXT:    vmov r0, s0
; CHECK-NOFP16-NEXT:    strh r0, [sp, #2]
; CHECK-NOFP16-NEXT:    add r0, sp, #2
; CHECK-NOFP16-NEXT:    vld1.16 {d0[], d1[]}, [r0:16]
; CHECK-NOFP16-NEXT:    add sp, sp, #4
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: test_vdupq_n_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    @ kill: def $s0 killed $s0 def $d0
; CHECK-FP16-NEXT:    vdup.16 q0, d0[0]
; CHECK-FP16-NEXT:    bx lr
entry:
  %0 = bitcast float %a.coerce to i32
  %tmp.0.extract.trunc = trunc i32 %0 to i16
  %1 = bitcast i16 %tmp.0.extract.trunc to bfloat
  %vecinit = insertelement <8 x bfloat> undef, bfloat %1, i32 0
  %vecinit8 = shufflevector <8 x bfloat> %vecinit, <8 x bfloat> undef, <8 x i32> zeroinitializer
  ret <8 x bfloat> %vecinit8
}

define dso_local <4 x bfloat> @test_vdup_lane_bf16(<4 x bfloat> %a) {
; CHECK-LABEL: test_vdup_lane_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vdup.16 d0, d0[3]
; CHECK-NEXT:    bx lr
entry:
  %shuffle = shufflevector <4 x bfloat> %a, <4 x bfloat> undef, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
  ret <4 x bfloat> %shuffle
}

define dso_local <8 x bfloat> @test_vdupq_lane_bf16(<4 x bfloat> %a) {
; CHECK-LABEL: test_vdupq_lane_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    @ kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    vdup.16 q0, d0[3]
; CHECK-NEXT:    bx lr
entry:
  %shuffle = shufflevector <4 x bfloat> %a, <4 x bfloat> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
  ret <8 x bfloat> %shuffle
}

define dso_local <4 x bfloat> @test_vext_bf16(<4 x bfloat> %a, <4 x bfloat> %b) {
; CHECK-LABEL: test_vext_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vext.16 d0, d0, d1, #2
; CHECK-NEXT:    bx lr
entry:
  %vext = shufflevector <4 x bfloat> %a, <4 x bfloat> %b, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
  ret <4 x bfloat> %vext
}

define dso_local <8 x bfloat> @test_vextq_bf16(<8 x bfloat> %a, <8 x bfloat> %b) {
; CHECK-LABEL: test_vextq_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vext.16 q0, q0, q1, #5
; CHECK-NEXT:    bx lr
entry:
  %vext = shufflevector <8 x bfloat> %a, <8 x bfloat> %b, <8 x i32> <i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12>
  ret <8 x bfloat> %vext
}

define dso_local <4 x bfloat> @test_vext_aligned_bf16(<8 x bfloat> %a) {
; CHECK-LABEL: test_vext_aligned_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vmov.f64 d0, d1
; CHECK-NEXT:    bx lr
entry:
  %vext = shufflevector <8 x bfloat> %a, <8 x bfloat> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  ret <4 x bfloat> %vext
}

define dso_local <4 x bfloat> @test_vext_unaligned_bf16(<8 x bfloat> %a) {
; CHECK-LABEL: test_vext_unaligned_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vext.16 d0, d0, d1, #3
; CHECK-NEXT:    bx lr
entry:
  %vext = shufflevector <8 x bfloat> %a, <8 x bfloat> undef, <4 x i32> <i32 3, i32 4, i32 5, i32 6>
  ret <4 x bfloat> %vext
}

define arm_aapcs_vfpcc <8 x bfloat> @shuffle3step0_bf16(<32 x bfloat> %src) {
; CHECK-NOFP16-LABEL: shuffle3step0_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    vmov r1, s0
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d0[3]
; CHECK-NOFP16-NEXT:    vrev32.16 d16, d3
; CHECK-NOFP16-NEXT:    vext.16 d17, d4, d5, #2
; CHECK-NOFP16-NEXT:    vext.16 d16, d16, d3, #1
; CHECK-NOFP16-NEXT:    vext.16 d16, d17, d16, #2
; CHECK-NOFP16-NEXT:    vext.16 d16, d16, d17, #1
; CHECK-NOFP16-NEXT:    vext.16 d17, d16, d16, #1
; CHECK-NOFP16-NEXT:    vmov.16 d16[0], r1
; CHECK-NOFP16-NEXT:    vmov.16 d16[1], r0
; CHECK-NOFP16-NEXT:    vmov r0, s3
; CHECK-NOFP16-NEXT:    vmov.16 d16[2], r0
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d2[1]
; CHECK-NOFP16-NEXT:    vmov.16 d16[3], r0
; CHECK-NOFP16-NEXT:    vorr q0, q8, q8
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: shuffle3step0_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    vmov r1, s0
; CHECK-FP16-NEXT:    vext.16 d17, d4, d5, #2
; CHECK-FP16-NEXT:    vmovx.f16 s8, s1
; CHECK-FP16-NEXT:    vrev32.16 d16, d3
; CHECK-FP16-NEXT:    vmov r0, s8
; CHECK-FP16-NEXT:    vext.16 d16, d16, d3, #1
; CHECK-FP16-NEXT:    vext.16 d16, d17, d16, #2
; CHECK-FP16-NEXT:    vext.16 d16, d16, d17, #1
; CHECK-FP16-NEXT:    vext.16 d17, d16, d16, #1
; CHECK-FP16-NEXT:    vmov.16 d16[0], r1
; CHECK-FP16-NEXT:    vmov.16 d16[1], r0
; CHECK-FP16-NEXT:    vmov r0, s3
; CHECK-FP16-NEXT:    vmovx.f16 s0, s4
; CHECK-FP16-NEXT:    vmov.16 d16[2], r0
; CHECK-FP16-NEXT:    vmov r0, s0
; CHECK-FP16-NEXT:    vmov.16 d16[3], r0
; CHECK-FP16-NEXT:    vorr q0, q8, q8
; CHECK-FP16-NEXT:    bx lr
entry:
  %s1 = shufflevector <32 x bfloat> %src, <32 x bfloat> undef, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>
  ret <8 x bfloat> %s1
}

define arm_aapcs_vfpcc <8 x bfloat> @shuffle3step1_bf16(<32 x bfloat> %src) {
; CHECK-NOFP16-LABEL: shuffle3step1_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    vorr q3, q0, q0
; CHECK-NOFP16-NEXT:    vmov.u16 r1, d6[1]
; CHECK-NOFP16-NEXT:    vmov r0, s14
; CHECK-NOFP16-NEXT:    vmov.16 d0[0], r1
; CHECK-NOFP16-NEXT:    vmov.16 d0[1], r0
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d7[3]
; CHECK-NOFP16-NEXT:    vmov.16 d0[2], r0
; CHECK-NOFP16-NEXT:    vmov r0, s5
; CHECK-NOFP16-NEXT:    vdup.16 q1, d3[1]
; CHECK-NOFP16-NEXT:    vmov r1, s4
; CHECK-NOFP16-NEXT:    vmov.16 d0[3], r0
; CHECK-NOFP16-NEXT:    vmov r0, s8
; CHECK-NOFP16-NEXT:    vmov.16 d1[0], r1
; CHECK-NOFP16-NEXT:    vmov.16 d1[1], r0
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d4[3]
; CHECK-NOFP16-NEXT:    vmov.16 d1[2], r0
; CHECK-NOFP16-NEXT:    vmov r0, s11
; CHECK-NOFP16-NEXT:    vmov.16 d1[3], r0
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: shuffle3step1_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    vorr q3, q0, q0
; CHECK-FP16-NEXT:    vmovx.f16 s0, s12
; CHECK-FP16-NEXT:    vmovx.f16 s12, s15
; CHECK-FP16-NEXT:    vmov r1, s0
; CHECK-FP16-NEXT:    vmov r0, s14
; CHECK-FP16-NEXT:    vmov.16 d0[0], r1
; CHECK-FP16-NEXT:    vmov.16 d0[1], r0
; CHECK-FP16-NEXT:    vmov r0, s12
; CHECK-FP16-NEXT:    vmov.16 d0[2], r0
; CHECK-FP16-NEXT:    vmov r0, s5
; CHECK-FP16-NEXT:    vdup.16 q1, d3[1]
; CHECK-FP16-NEXT:    vmov r1, s4
; CHECK-FP16-NEXT:    vmovx.f16 s4, s9
; CHECK-FP16-NEXT:    vmov.16 d0[3], r0
; CHECK-FP16-NEXT:    vmov r0, s8
; CHECK-FP16-NEXT:    vmov.16 d1[0], r1
; CHECK-FP16-NEXT:    vmov.16 d1[1], r0
; CHECK-FP16-NEXT:    vmov r0, s4
; CHECK-FP16-NEXT:    vmov.16 d1[2], r0
; CHECK-FP16-NEXT:    vmov r0, s11
; CHECK-FP16-NEXT:    vmov.16 d1[3], r0
; CHECK-FP16-NEXT:    bx lr
entry:
  %s1 = shufflevector <32 x bfloat> %src, <32 x bfloat> undef, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>
  ret <8 x bfloat> %s1
}

define arm_aapcs_vfpcc <8 x bfloat> @shuffle3step2_bf16(<32 x bfloat> %src) {
; CHECK-NOFP16-LABEL: shuffle3step2_bf16:
; CHECK-NOFP16:       @ %bb.0: @ %entry
; CHECK-NOFP16-NEXT:    vext.16 d16, d0, d1, #2
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d4[1]
; CHECK-NOFP16-NEXT:    vext.16 d17, d16, d2, #3
; CHECK-NOFP16-NEXT:    vext.16 d16, d2, d16, #1
; CHECK-NOFP16-NEXT:    vdup.16 q1, d3[2]
; CHECK-NOFP16-NEXT:    vext.16 d16, d16, d17, #2
; CHECK-NOFP16-NEXT:    vmov r1, s4
; CHECK-NOFP16-NEXT:    vext.16 d0, d16, d16, #1
; CHECK-NOFP16-NEXT:    vmov.16 d1[0], r1
; CHECK-NOFP16-NEXT:    vmov.16 d1[1], r0
; CHECK-NOFP16-NEXT:    vmov r0, s10
; CHECK-NOFP16-NEXT:    vmov.16 d1[2], r0
; CHECK-NOFP16-NEXT:    vmov.u16 r0, d5[3]
; CHECK-NOFP16-NEXT:    vmov.16 d1[3], r0
; CHECK-NOFP16-NEXT:    bx lr
;
; CHECK-FP16-LABEL: shuffle3step2_bf16:
; CHECK-FP16:       @ %bb.0: @ %entry
; CHECK-FP16-NEXT:    vext.16 d16, d0, d1, #2
; CHECK-FP16-NEXT:    vmovx.f16 s12, s8
; CHECK-FP16-NEXT:    vmov r0, s12
; CHECK-FP16-NEXT:    vext.16 d17, d16, d2, #3
; CHECK-FP16-NEXT:    vext.16 d16, d2, d16, #1
; CHECK-FP16-NEXT:    vdup.16 q1, d3[2]
; CHECK-FP16-NEXT:    vext.16 d16, d16, d17, #2
; CHECK-FP16-NEXT:    vmov r1, s4
; CHECK-FP16-NEXT:    vmovx.f16 s4, s11
; CHECK-FP16-NEXT:    vext.16 d0, d16, d16, #1
; CHECK-FP16-NEXT:    vmov.16 d1[0], r1
; CHECK-FP16-NEXT:    vmov.16 d1[1], r0
; CHECK-FP16-NEXT:    vmov r0, s10
; CHECK-FP16-NEXT:    vmov.16 d1[2], r0
; CHECK-FP16-NEXT:    vmov r0, s4
; CHECK-FP16-NEXT:    vmov.16 d1[3], r0
; CHECK-FP16-NEXT:    bx lr
entry:
  %s1 = shufflevector <32 x bfloat> %src, <32 x bfloat> undef, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>
  ret <8 x bfloat> %s1
}


define dso_local <4 x bfloat> @test_vrev64_bf16(<4 x bfloat> %a) {
; CHECK-LABEL: test_vrev64_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vrev64.16 d0, d0
; CHECK-NEXT:    bx lr
entry:
  %shuffle.i = shufflevector <4 x bfloat> %a, <4 x bfloat> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
  ret <4 x bfloat> %shuffle.i
}

define dso_local <8 x bfloat> @test_vrev64q_bf16(<8 x bfloat> %a) {
; CHECK-LABEL: test_vrev64q_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vrev64.16 q0, q0
; CHECK-NEXT:    bx lr
entry:
  %shuffle.i = shufflevector <8 x bfloat> %a, <8 x bfloat> undef, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>
  ret <8 x bfloat> %shuffle.i
}

define dso_local <4 x bfloat> @test_vrev32_bf16(<4 x bfloat> %a) {
; CHECK-LABEL: test_vrev32_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vrev32.16 d0, d0
; CHECK-NEXT:    bx lr
entry:
  %shuffle.i = shufflevector <4 x bfloat> %a, <4 x bfloat> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
  ret <4 x bfloat> %shuffle.i
}

define dso_local <8 x bfloat> @test_vrev32q_bf16(<8 x bfloat> %a) {
; CHECK-LABEL: test_vrev32q_bf16:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vrev32.16 q0, q0
; CHECK-NEXT:    bx lr
entry:
  %shuffle.i = shufflevector <8 x bfloat> %a, <8 x bfloat> undef, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
  ret <8 x bfloat> %shuffle.i
}

define <4 x bfloat> @test_vld_dup1_4xbfloat(ptr %b) {
; CHECK-LABEL: test_vld_dup1_4xbfloat:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vld1.16 {d0[]}, [r0:16]
; CHECK-NEXT:    bx lr
entry:
  %b1 = load bfloat, ptr %b, align 2
  %vecinit = insertelement <4 x bfloat> undef, bfloat %b1, i32 0
  %vecinit2 = insertelement <4 x bfloat> %vecinit, bfloat %b1, i32 1
  %vecinit3 = insertelement <4 x bfloat> %vecinit2, bfloat %b1, i32 2
  %vecinit4 = insertelement <4 x bfloat> %vecinit3, bfloat %b1, i32 3
  ret <4 x bfloat> %vecinit4
}

define <8 x bfloat> @test_vld_dup1_8xbfloat(ptr %b) local_unnamed_addr {
; CHECK-LABEL: test_vld_dup1_8xbfloat:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    vld1.16 {d0[], d1[]}, [r0:16]
; CHECK-NEXT:    bx lr
entry:
  %b1 = load bfloat, ptr %b, align 2
  %vecinit = insertelement <8 x bfloat> undef, bfloat %b1, i32 0
  %vecinit8 = shufflevector <8 x bfloat> %vecinit, <8 x bfloat> undef, <8 x i32> zeroinitializer
  ret <8 x bfloat> %vecinit8
}

define <8 x bfloat> @test_shufflevector8xbfloat(<4 x bfloat> %a) {
; CHECK-LABEL: test_shufflevector8xbfloat:
; CHECK:       @ %bb.0: @ %entry
; CHECK-NEXT:    @ kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    vmov.f64 d1, d0
; CHECK-NEXT:    bx lr
entry:
  %r = shufflevector <4 x bfloat> %a, <4 x bfloat> %a, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x bfloat> %r
}

declare <8 x i8> @llvm.arm.neon.vbsl.v8i8(<8 x i8>, <8 x i8>, <8 x i8>)
declare <16 x i8> @llvm.arm.neon.vbsl.v16i8(<16 x i8>, <16 x i8>, <16 x i8>)