; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s --check-prefixes=CHECK,CHECK-SD
; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI

; CHECK-GI:       warning: Instruction selection used fallback path for zext_v32i1
; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for zext_v64i1

; Test efficient codegen of vector extends up from legal type to 128 bit
; and 256 bit vector types.

;-----
; Vectors of i16.
;-----

define <8 x i16> @func1(<8 x i8> %v0) nounwind {
; CHECK-LABEL: func1:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ushll.8h v0, v0, #0
; CHECK-NEXT:    ret
  %r = zext <8 x i8> %v0 to <8 x i16>
  ret <8 x i16> %r
}

define <8 x i16> @func2(<8 x i8> %v0) nounwind {
; CHECK-LABEL: func2:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sshll.8h v0, v0, #0
; CHECK-NEXT:    ret
  %r = sext <8 x i8> %v0 to <8 x i16>
  ret <8 x i16> %r
}

define <16 x i16> @func3(<16 x i8> %v0) nounwind {
; CHECK-SD-LABEL: func3:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll2.8h v1, v0, #0
; CHECK-SD-NEXT:    ushll.8h v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: func3:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.8h v2, v0, #0
; CHECK-GI-NEXT:    ushll2.8h v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = zext <16 x i8> %v0 to <16 x i16>
  ret <16 x i16> %r
}

define <16 x i16> @func4(<16 x i8> %v0) nounwind {
; CHECK-SD-LABEL: func4:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll2.8h v1, v0, #0
; CHECK-SD-NEXT:    sshll.8h v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: func4:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.8h v2, v0, #0
; CHECK-GI-NEXT:    sshll2.8h v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = sext <16 x i8> %v0 to <16 x i16>
  ret <16 x i16> %r
}

;-----
; Vectors of i32.
;-----

define <4 x i32> @afunc1(<4 x i16> %v0) nounwind {
; CHECK-LABEL: afunc1:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ushll.4s v0, v0, #0
; CHECK-NEXT:    ret
  %r = zext <4 x i16> %v0 to <4 x i32>
  ret <4 x i32> %r
}

define <4 x i32> @afunc2(<4 x i16> %v0) nounwind {
; CHECK-LABEL: afunc2:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sshll.4s v0, v0, #0
; CHECK-NEXT:    ret
  %r = sext <4 x i16> %v0 to <4 x i32>
  ret <4 x i32> %r
}

define <8 x i32> @afunc3(<8 x i16> %v0) nounwind {
; CHECK-SD-LABEL: afunc3:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll2.4s v1, v0, #0
; CHECK-SD-NEXT:    ushll.4s v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: afunc3:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.4s v2, v0, #0
; CHECK-GI-NEXT:    ushll2.4s v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = zext <8 x i16> %v0 to <8 x i32>
  ret <8 x i32> %r
}

define <8 x i32> @afunc4(<8 x i16> %v0) nounwind {
; CHECK-SD-LABEL: afunc4:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll2.4s v1, v0, #0
; CHECK-SD-NEXT:    sshll.4s v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: afunc4:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.4s v2, v0, #0
; CHECK-GI-NEXT:    sshll2.4s v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = sext <8 x i16> %v0 to <8 x i32>
  ret <8 x i32> %r
}

define <8 x i32> @bfunc1(<8 x i8> %v0) nounwind {
; CHECK-SD-LABEL: bfunc1:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll.8h v0, v0, #0
; CHECK-SD-NEXT:    ushll2.4s v1, v0, #0
; CHECK-SD-NEXT:    ushll.4s v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: bfunc1:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.8h v1, v0, #0
; CHECK-GI-NEXT:    ushll.4s v0, v1, #0
; CHECK-GI-NEXT:    ushll2.4s v1, v1, #0
; CHECK-GI-NEXT:    ret
  %r = zext <8 x i8> %v0 to <8 x i32>
  ret <8 x i32> %r
}

define <8 x i32> @bfunc2(<8 x i8> %v0) nounwind {
; CHECK-SD-LABEL: bfunc2:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll.8h v0, v0, #0
; CHECK-SD-NEXT:    sshll2.4s v1, v0, #0
; CHECK-SD-NEXT:    sshll.4s v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: bfunc2:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.8h v1, v0, #0
; CHECK-GI-NEXT:    sshll.4s v0, v1, #0
; CHECK-GI-NEXT:    sshll2.4s v1, v1, #0
; CHECK-GI-NEXT:    ret
  %r = sext <8 x i8> %v0 to <8 x i32>
  ret <8 x i32> %r
}

;-----
; Vectors of i64.
;-----

define <4 x i64> @zfunc1(<4 x i32> %v0) nounwind {
; CHECK-SD-LABEL: zfunc1:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll2.2d v1, v0, #0
; CHECK-SD-NEXT:    ushll.2d v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: zfunc1:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.2d v2, v0, #0
; CHECK-GI-NEXT:    ushll2.2d v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = zext <4 x i32> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <4 x i64> @zfunc2(<4 x i32> %v0) nounwind {
; CHECK-SD-LABEL: zfunc2:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll2.2d v1, v0, #0
; CHECK-SD-NEXT:    sshll.2d v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: zfunc2:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.2d v2, v0, #0
; CHECK-GI-NEXT:    sshll2.2d v1, v0, #0
; CHECK-GI-NEXT:    mov.16b v0, v2
; CHECK-GI-NEXT:    ret
  %r = sext <4 x i32> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <4 x i64> @bfunc3(<4 x i16> %v0) nounwind {
; CHECK-SD-LABEL: bfunc3:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll.4s v0, v0, #0
; CHECK-SD-NEXT:    ushll2.2d v1, v0, #0
; CHECK-SD-NEXT:    ushll.2d v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: bfunc3:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.4s v1, v0, #0
; CHECK-GI-NEXT:    ushll.2d v0, v1, #0
; CHECK-GI-NEXT:    ushll2.2d v1, v1, #0
; CHECK-GI-NEXT:    ret
  %r = zext <4 x i16> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <4 x i64> @cfunc4(<4 x i16> %v0) nounwind {
; CHECK-SD-LABEL: cfunc4:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll.4s v0, v0, #0
; CHECK-SD-NEXT:    sshll2.2d v1, v0, #0
; CHECK-SD-NEXT:    sshll.2d v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: cfunc4:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.4s v1, v0, #0
; CHECK-GI-NEXT:    sshll.2d v0, v1, #0
; CHECK-GI-NEXT:    sshll2.2d v1, v1, #0
; CHECK-GI-NEXT:    ret
  %r = sext <4 x i16> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <4 x i64> @zext_v4i8_to_v4i64(<4 x i8> %v0) nounwind {
; CHECK-SD-LABEL: zext_v4i8_to_v4i64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    bic.4h v0, #255, lsl #8
; CHECK-SD-NEXT:    ushll.4s v0, v0, #0
; CHECK-SD-NEXT:    ushll2.2d v1, v0, #0
; CHECK-SD-NEXT:    ushll.2d v0, v0, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: zext_v4i8_to_v4i64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.4s v0, v0, #0
; CHECK-GI-NEXT:    movi.2d v1, #0x000000000000ff
; CHECK-GI-NEXT:    ushll.2d v2, v0, #0
; CHECK-GI-NEXT:    ushll2.2d v3, v0, #0
; CHECK-GI-NEXT:    and.16b v0, v2, v1
; CHECK-GI-NEXT:    and.16b v1, v3, v1
; CHECK-GI-NEXT:    ret
  %r = zext <4 x i8> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <4 x i64> @sext_v4i8_to_v4i64(<4 x i8> %v0) nounwind {
; CHECK-SD-LABEL: sext_v4i8_to_v4i64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll.4s v0, v0, #0
; CHECK-SD-NEXT:    ushll.2d v1, v0, #0
; CHECK-SD-NEXT:    ushll2.2d v0, v0, #0
; CHECK-SD-NEXT:    shl.2d v0, v0, #56
; CHECK-SD-NEXT:    shl.2d v2, v1, #56
; CHECK-SD-NEXT:    sshr.2d v1, v0, #56
; CHECK-SD-NEXT:    sshr.2d v0, v2, #56
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sext_v4i8_to_v4i64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.4s v0, v0, #0
; CHECK-GI-NEXT:    ushll.2d v1, v0, #0
; CHECK-GI-NEXT:    ushll2.2d v0, v0, #0
; CHECK-GI-NEXT:    shl.2d v1, v1, #56
; CHECK-GI-NEXT:    shl.2d v2, v0, #56
; CHECK-GI-NEXT:    sshr.2d v0, v1, #56
; CHECK-GI-NEXT:    sshr.2d v1, v2, #56
; CHECK-GI-NEXT:    ret
  %r = sext <4 x i8> %v0 to <4 x i64>
  ret <4 x i64> %r
}

define <8 x i64> @zext_v8i8_to_v8i64(<8 x i8> %v0) nounwind {
; CHECK-SD-LABEL: zext_v8i8_to_v8i64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ushll.8h v0, v0, #0
; CHECK-SD-NEXT:    ushll.4s v1, v0, #0
; CHECK-SD-NEXT:    ushll2.4s v2, v0, #0
; CHECK-SD-NEXT:    ushll.2d v0, v1, #0
; CHECK-SD-NEXT:    ushll2.2d v3, v2, #0
; CHECK-SD-NEXT:    ushll2.2d v1, v1, #0
; CHECK-SD-NEXT:    ushll.2d v2, v2, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: zext_v8i8_to_v8i64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ushll.8h v0, v0, #0
; CHECK-GI-NEXT:    ushll.4s v1, v0, #0
; CHECK-GI-NEXT:    ushll2.4s v3, v0, #0
; CHECK-GI-NEXT:    ushll.2d v0, v1, #0
; CHECK-GI-NEXT:    ushll2.2d v1, v1, #0
; CHECK-GI-NEXT:    ushll.2d v2, v3, #0
; CHECK-GI-NEXT:    ushll2.2d v3, v3, #0
; CHECK-GI-NEXT:    ret
  %r = zext <8 x i8> %v0 to <8 x i64>
  ret <8 x i64> %r
}

define <8 x i64> @sext_v8i8_to_v8i64(<8 x i8> %v0) nounwind {
; CHECK-SD-LABEL: sext_v8i8_to_v8i64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sshll.8h v0, v0, #0
; CHECK-SD-NEXT:    sshll.4s v1, v0, #0
; CHECK-SD-NEXT:    sshll2.4s v2, v0, #0
; CHECK-SD-NEXT:    sshll.2d v0, v1, #0
; CHECK-SD-NEXT:    sshll2.2d v3, v2, #0
; CHECK-SD-NEXT:    sshll2.2d v1, v1, #0
; CHECK-SD-NEXT:    sshll.2d v2, v2, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sext_v8i8_to_v8i64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sshll.8h v0, v0, #0
; CHECK-GI-NEXT:    sshll.4s v1, v0, #0
; CHECK-GI-NEXT:    sshll2.4s v3, v0, #0
; CHECK-GI-NEXT:    sshll.2d v0, v1, #0
; CHECK-GI-NEXT:    sshll2.2d v1, v1, #0
; CHECK-GI-NEXT:    sshll.2d v2, v3, #0
; CHECK-GI-NEXT:    sshll2.2d v3, v3, #0
; CHECK-GI-NEXT:    ret
  %r = sext <8 x i8> %v0 to <8 x i64>
  ret <8 x i64> %r
}

; Extends of vectors of i1.

define <32 x i8> @zext_v32i1(<32 x i1> %arg) {
; CHECK-LABEL: zext_v32i1:
; CHECK:       // %bb.0:
; CHECK-NEXT:    ldr w8, [sp, #64]
; CHECK-NEXT:    fmov s0, w0
; CHECK-NEXT:    ldr w9, [sp, #72]
; CHECK-NEXT:    movi.16b v2, #1
; CHECK-NEXT:    fmov s1, w8
; CHECK-NEXT:    ldr w8, [sp, #80]
; CHECK-NEXT:    mov.b v0[1], w1
; CHECK-NEXT:    mov.b v1[1], w9
; CHECK-NEXT:    ldr w9, [sp]
; CHECK-NEXT:    mov.b v0[2], w2
; CHECK-NEXT:    mov.b v1[2], w8
; CHECK-NEXT:    ldr w8, [sp, #88]
; CHECK-NEXT:    mov.b v0[3], w3
; CHECK-NEXT:    mov.b v1[3], w8
; CHECK-NEXT:    ldr w8, [sp, #96]
; CHECK-NEXT:    mov.b v0[4], w4
; CHECK-NEXT:    mov.b v1[4], w8
; CHECK-NEXT:    ldr w8, [sp, #104]
; CHECK-NEXT:    mov.b v0[5], w5
; CHECK-NEXT:    mov.b v1[5], w8
; CHECK-NEXT:    ldr w8, [sp, #112]
; CHECK-NEXT:    mov.b v0[6], w6
; CHECK-NEXT:    mov.b v1[6], w8
; CHECK-NEXT:    ldr w8, [sp, #120]
; CHECK-NEXT:    mov.b v0[7], w7
; CHECK-NEXT:    mov.b v1[7], w8
; CHECK-NEXT:    ldr w8, [sp, #128]
; CHECK-NEXT:    mov.b v0[8], w9
; CHECK-NEXT:    ldr w9, [sp, #8]
; CHECK-NEXT:    mov.b v1[8], w8
; CHECK-NEXT:    ldr w8, [sp, #136]
; CHECK-NEXT:    mov.b v0[9], w9
; CHECK-NEXT:    ldr w9, [sp, #16]
; CHECK-NEXT:    mov.b v1[9], w8
; CHECK-NEXT:    ldr w8, [sp, #144]
; CHECK-NEXT:    mov.b v0[10], w9
; CHECK-NEXT:    ldr w9, [sp, #24]
; CHECK-NEXT:    mov.b v1[10], w8
; CHECK-NEXT:    ldr w8, [sp, #152]
; CHECK-NEXT:    mov.b v0[11], w9
; CHECK-NEXT:    ldr w9, [sp, #32]
; CHECK-NEXT:    mov.b v1[11], w8
; CHECK-NEXT:    ldr w8, [sp, #160]
; CHECK-NEXT:    mov.b v0[12], w9
; CHECK-NEXT:    ldr w9, [sp, #40]
; CHECK-NEXT:    mov.b v1[12], w8
; CHECK-NEXT:    ldr w8, [sp, #168]
; CHECK-NEXT:    mov.b v0[13], w9
; CHECK-NEXT:    ldr w9, [sp, #48]
; CHECK-NEXT:    mov.b v1[13], w8
; CHECK-NEXT:    ldr w8, [sp, #176]
; CHECK-NEXT:    mov.b v0[14], w9
; CHECK-NEXT:    ldr w9, [sp, #56]
; CHECK-NEXT:    mov.b v1[14], w8
; CHECK-NEXT:    ldr w8, [sp, #184]
; CHECK-NEXT:    mov.b v0[15], w9
; CHECK-NEXT:    mov.b v1[15], w8
; CHECK-NEXT:    and.16b v0, v0, v2
; CHECK-NEXT:    and.16b v1, v1, v2
; CHECK-NEXT:    ret
  %res = zext <32 x i1> %arg to <32 x i8>
  ret <32 x i8> %res
}

define <32 x i8> @sext_v32i1(<32 x i1> %arg) {
; CHECK-SD-LABEL: sext_v32i1:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    ldr w8, [sp, #64]
; CHECK-SD-NEXT:    fmov s1, w0
; CHECK-SD-NEXT:    ldr w9, [sp, #72]
; CHECK-SD-NEXT:    fmov s0, w8
; CHECK-SD-NEXT:    ldr w8, [sp, #80]
; CHECK-SD-NEXT:    mov.b v1[1], w1
; CHECK-SD-NEXT:    mov.b v0[1], w9
; CHECK-SD-NEXT:    ldr w9, [sp]
; CHECK-SD-NEXT:    mov.b v1[2], w2
; CHECK-SD-NEXT:    mov.b v0[2], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #88]
; CHECK-SD-NEXT:    mov.b v1[3], w3
; CHECK-SD-NEXT:    mov.b v0[3], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #96]
; CHECK-SD-NEXT:    mov.b v1[4], w4
; CHECK-SD-NEXT:    mov.b v0[4], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #104]
; CHECK-SD-NEXT:    mov.b v1[5], w5
; CHECK-SD-NEXT:    mov.b v0[5], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #112]
; CHECK-SD-NEXT:    mov.b v1[6], w6
; CHECK-SD-NEXT:    mov.b v0[6], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #120]
; CHECK-SD-NEXT:    mov.b v1[7], w7
; CHECK-SD-NEXT:    mov.b v0[7], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #128]
; CHECK-SD-NEXT:    mov.b v1[8], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #8]
; CHECK-SD-NEXT:    mov.b v0[8], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #136]
; CHECK-SD-NEXT:    mov.b v1[9], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #16]
; CHECK-SD-NEXT:    mov.b v0[9], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #144]
; CHECK-SD-NEXT:    mov.b v1[10], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #24]
; CHECK-SD-NEXT:    mov.b v0[10], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #152]
; CHECK-SD-NEXT:    mov.b v1[11], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #32]
; CHECK-SD-NEXT:    mov.b v0[11], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #160]
; CHECK-SD-NEXT:    mov.b v1[12], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #40]
; CHECK-SD-NEXT:    mov.b v0[12], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #168]
; CHECK-SD-NEXT:    mov.b v1[13], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #48]
; CHECK-SD-NEXT:    mov.b v0[13], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #176]
; CHECK-SD-NEXT:    mov.b v1[14], w9
; CHECK-SD-NEXT:    ldr w9, [sp, #56]
; CHECK-SD-NEXT:    mov.b v0[14], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #184]
; CHECK-SD-NEXT:    mov.b v1[15], w9
; CHECK-SD-NEXT:    mov.b v0[15], w8
; CHECK-SD-NEXT:    shl.16b v1, v1, #7
; CHECK-SD-NEXT:    shl.16b v2, v0, #7
; CHECK-SD-NEXT:    cmlt.16b v0, v1, #0
; CHECK-SD-NEXT:    cmlt.16b v1, v2, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sext_v32i1:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    ldr w8, [sp, #64]
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    ldr w9, [sp, #72]
; CHECK-GI-NEXT:    fmov s1, w8
; CHECK-GI-NEXT:    ldr w8, [sp, #80]
; CHECK-GI-NEXT:    mov.b v0[1], w1
; CHECK-GI-NEXT:    mov.b v1[1], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #128]
; CHECK-GI-NEXT:    mov.b v0[2], w2
; CHECK-GI-NEXT:    mov.b v1[2], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #88]
; CHECK-GI-NEXT:    mov.b v0[3], w3
; CHECK-GI-NEXT:    mov.b v1[3], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #96]
; CHECK-GI-NEXT:    mov.b v0[4], w4
; CHECK-GI-NEXT:    mov.b v1[4], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #104]
; CHECK-GI-NEXT:    mov.b v0[5], w5
; CHECK-GI-NEXT:    mov.b v1[5], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #112]
; CHECK-GI-NEXT:    mov.b v0[6], w6
; CHECK-GI-NEXT:    mov.b v1[6], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #120]
; CHECK-GI-NEXT:    mov.b v0[7], w7
; CHECK-GI-NEXT:    mov.b v1[7], w8
; CHECK-GI-NEXT:    ldr w8, [sp]
; CHECK-GI-NEXT:    mov.b v0[8], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #8]
; CHECK-GI-NEXT:    mov.b v1[8], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #136]
; CHECK-GI-NEXT:    mov.b v0[9], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #16]
; CHECK-GI-NEXT:    mov.b v1[9], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #144]
; CHECK-GI-NEXT:    mov.b v0[10], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #24]
; CHECK-GI-NEXT:    mov.b v1[10], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #152]
; CHECK-GI-NEXT:    mov.b v0[11], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #32]
; CHECK-GI-NEXT:    mov.b v1[11], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #160]
; CHECK-GI-NEXT:    mov.b v0[12], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #40]
; CHECK-GI-NEXT:    mov.b v1[12], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #168]
; CHECK-GI-NEXT:    mov.b v0[13], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #48]
; CHECK-GI-NEXT:    mov.b v1[13], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #176]
; CHECK-GI-NEXT:    mov.b v0[14], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #56]
; CHECK-GI-NEXT:    mov.b v1[14], w9
; CHECK-GI-NEXT:    ldr w9, [sp, #184]
; CHECK-GI-NEXT:    mov.b v0[15], w8
; CHECK-GI-NEXT:    mov.b v1[15], w9
; CHECK-GI-NEXT:    shl.16b v0, v0, #7
; CHECK-GI-NEXT:    shl.16b v1, v1, #7
; CHECK-GI-NEXT:    cmlt.16b v0, v0, #0
; CHECK-GI-NEXT:    cmlt.16b v1, v1, #0
; CHECK-GI-NEXT:    ret
  %res = sext <32 x i1> %arg to <32 x i8>
  ret <32 x i8> %res
}

define <64 x i8> @zext_v64i1(<64 x i1> %arg) {
; CHECK-LABEL: zext_v64i1:
; CHECK:       // %bb.0:
; CHECK-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 16
; CHECK-NEXT:    .cfi_offset w29, -16
; CHECK-NEXT:    ldr w8, [sp, #336]
; CHECK-NEXT:    ldr w9, [sp, #208]
; CHECK-NEXT:    fmov s0, w0
; CHECK-NEXT:    ldr w10, [sp, #80]
; CHECK-NEXT:    ldr w11, [sp, #216]
; CHECK-NEXT:    movi.16b v4, #1
; CHECK-NEXT:    fmov s3, w8
; CHECK-NEXT:    fmov s2, w9
; CHECK-NEXT:    ldr w8, [sp, #344]
; CHECK-NEXT:    fmov s1, w10
; CHECK-NEXT:    ldr w12, [sp, #88]
; CHECK-NEXT:    mov.b v0[1], w1
; CHECK-NEXT:    ldr w9, [sp, #224]
; CHECK-NEXT:    ldr w10, [sp, #96]
; CHECK-NEXT:    mov.b v3[1], w8
; CHECK-NEXT:    mov.b v2[1], w11
; CHECK-NEXT:    ldr w8, [sp, #352]
; CHECK-NEXT:    mov.b v1[1], w12
; CHECK-NEXT:    ldr w11, [sp, #144]
; CHECK-NEXT:    mov.b v0[2], w2
; CHECK-NEXT:    mov.b v3[2], w8
; CHECK-NEXT:    mov.b v2[2], w9
; CHECK-NEXT:    ldr w8, [sp, #360]
; CHECK-NEXT:    mov.b v1[2], w10
; CHECK-NEXT:    ldr w9, [sp, #232]
; CHECK-NEXT:    ldr w10, [sp, #104]
; CHECK-NEXT:    mov.b v0[3], w3
; CHECK-NEXT:    mov.b v3[3], w8
; CHECK-NEXT:    mov.b v2[3], w9
; CHECK-NEXT:    ldr w8, [sp, #368]
; CHECK-NEXT:    mov.b v1[3], w10
; CHECK-NEXT:    ldr w9, [sp, #240]
; CHECK-NEXT:    ldr w10, [sp, #112]
; CHECK-NEXT:    mov.b v0[4], w4
; CHECK-NEXT:    mov.b v3[4], w8
; CHECK-NEXT:    mov.b v2[4], w9
; CHECK-NEXT:    ldr w8, [sp, #376]
; CHECK-NEXT:    mov.b v1[4], w10
; CHECK-NEXT:    ldr w9, [sp, #248]
; CHECK-NEXT:    ldr w10, [sp, #120]
; CHECK-NEXT:    mov.b v0[5], w5
; CHECK-NEXT:    mov.b v3[5], w8
; CHECK-NEXT:    mov.b v2[5], w9
; CHECK-NEXT:    ldr w8, [sp, #384]
; CHECK-NEXT:    mov.b v1[5], w10
; CHECK-NEXT:    ldr w9, [sp, #256]
; CHECK-NEXT:    ldr w10, [sp, #128]
; CHECK-NEXT:    mov.b v0[6], w6
; CHECK-NEXT:    mov.b v3[6], w8
; CHECK-NEXT:    mov.b v2[6], w9
; CHECK-NEXT:    ldr w8, [sp, #392]
; CHECK-NEXT:    mov.b v1[6], w10
; CHECK-NEXT:    ldr w9, [sp, #264]
; CHECK-NEXT:    ldr w10, [sp, #136]
; CHECK-NEXT:    mov.b v0[7], w7
; CHECK-NEXT:    mov.b v3[7], w8
; CHECK-NEXT:    mov.b v2[7], w9
; CHECK-NEXT:    ldr w8, [sp, #16]
; CHECK-NEXT:    mov.b v1[7], w10
; CHECK-NEXT:    ldr w9, [sp, #400]
; CHECK-NEXT:    ldr w10, [sp, #272]
; CHECK-NEXT:    mov.b v0[8], w8
; CHECK-NEXT:    ldr w8, [sp, #24]
; CHECK-NEXT:    mov.b v3[8], w9
; CHECK-NEXT:    mov.b v2[8], w10
; CHECK-NEXT:    ldr w9, [sp, #408]
; CHECK-NEXT:    mov.b v1[8], w11
; CHECK-NEXT:    ldr w10, [sp, #280]
; CHECK-NEXT:    ldr w11, [sp, #152]
; CHECK-NEXT:    mov.b v0[9], w8
; CHECK-NEXT:    ldr w8, [sp, #32]
; CHECK-NEXT:    mov.b v3[9], w9
; CHECK-NEXT:    mov.b v2[9], w10
; CHECK-NEXT:    ldr w9, [sp, #416]
; CHECK-NEXT:    mov.b v1[9], w11
; CHECK-NEXT:    ldr w10, [sp, #288]
; CHECK-NEXT:    ldr w11, [sp, #160]
; CHECK-NEXT:    mov.b v0[10], w8
; CHECK-NEXT:    ldr w8, [sp, #40]
; CHECK-NEXT:    mov.b v3[10], w9
; CHECK-NEXT:    mov.b v2[10], w10
; CHECK-NEXT:    ldr w9, [sp, #424]
; CHECK-NEXT:    mov.b v1[10], w11
; CHECK-NEXT:    ldr w10, [sp, #296]
; CHECK-NEXT:    ldr w11, [sp, #168]
; CHECK-NEXT:    mov.b v0[11], w8
; CHECK-NEXT:    ldr w8, [sp, #48]
; CHECK-NEXT:    mov.b v3[11], w9
; CHECK-NEXT:    mov.b v2[11], w10
; CHECK-NEXT:    ldr w9, [sp, #432]
; CHECK-NEXT:    mov.b v1[11], w11
; CHECK-NEXT:    ldr w10, [sp, #304]
; CHECK-NEXT:    ldr w11, [sp, #176]
; CHECK-NEXT:    mov.b v0[12], w8
; CHECK-NEXT:    ldr w8, [sp, #56]
; CHECK-NEXT:    mov.b v3[12], w9
; CHECK-NEXT:    mov.b v2[12], w10
; CHECK-NEXT:    ldr w9, [sp, #440]
; CHECK-NEXT:    mov.b v1[12], w11
; CHECK-NEXT:    ldr w10, [sp, #312]
; CHECK-NEXT:    ldr w11, [sp, #184]
; CHECK-NEXT:    mov.b v0[13], w8
; CHECK-NEXT:    ldr w8, [sp, #64]
; CHECK-NEXT:    mov.b v3[13], w9
; CHECK-NEXT:    mov.b v2[13], w10
; CHECK-NEXT:    ldr w9, [sp, #448]
; CHECK-NEXT:    mov.b v1[13], w11
; CHECK-NEXT:    ldr w10, [sp, #320]
; CHECK-NEXT:    ldr w11, [sp, #192]
; CHECK-NEXT:    mov.b v0[14], w8
; CHECK-NEXT:    ldr w8, [sp, #72]
; CHECK-NEXT:    mov.b v3[14], w9
; CHECK-NEXT:    mov.b v2[14], w10
; CHECK-NEXT:    ldr w9, [sp, #456]
; CHECK-NEXT:    mov.b v1[14], w11
; CHECK-NEXT:    ldr w10, [sp, #328]
; CHECK-NEXT:    ldr w11, [sp, #200]
; CHECK-NEXT:    mov.b v0[15], w8
; CHECK-NEXT:    mov.b v3[15], w9
; CHECK-NEXT:    mov.b v2[15], w10
; CHECK-NEXT:    mov.b v1[15], w11
; CHECK-NEXT:    and.16b v0, v0, v4
; CHECK-NEXT:    and.16b v2, v2, v4
; CHECK-NEXT:    and.16b v3, v3, v4
; CHECK-NEXT:    and.16b v1, v1, v4
; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-NEXT:    ret
  %res = zext <64 x i1> %arg to <64 x i8>
  ret <64 x i8> %res
}

define <64 x i8> @sext_v64i1(<64 x i1> %arg) {
; CHECK-SD-LABEL: sext_v64i1:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w29, -16
; CHECK-SD-NEXT:    ldr w8, [sp, #336]
; CHECK-SD-NEXT:    ldr w9, [sp, #208]
; CHECK-SD-NEXT:    fmov s2, w0
; CHECK-SD-NEXT:    ldr w10, [sp, #80]
; CHECK-SD-NEXT:    ldr w11, [sp, #216]
; CHECK-SD-NEXT:    ldr w12, [sp, #88]
; CHECK-SD-NEXT:    fmov s0, w8
; CHECK-SD-NEXT:    fmov s1, w9
; CHECK-SD-NEXT:    ldr w8, [sp, #344]
; CHECK-SD-NEXT:    fmov s3, w10
; CHECK-SD-NEXT:    mov.b v2[1], w1
; CHECK-SD-NEXT:    ldr w9, [sp, #224]
; CHECK-SD-NEXT:    ldr w10, [sp, #96]
; CHECK-SD-NEXT:    mov.b v0[1], w8
; CHECK-SD-NEXT:    mov.b v1[1], w11
; CHECK-SD-NEXT:    ldr w8, [sp, #352]
; CHECK-SD-NEXT:    mov.b v3[1], w12
; CHECK-SD-NEXT:    ldr w11, [sp, #144]
; CHECK-SD-NEXT:    mov.b v2[2], w2
; CHECK-SD-NEXT:    mov.b v0[2], w8
; CHECK-SD-NEXT:    mov.b v1[2], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #360]
; CHECK-SD-NEXT:    mov.b v3[2], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #232]
; CHECK-SD-NEXT:    ldr w10, [sp, #104]
; CHECK-SD-NEXT:    mov.b v2[3], w3
; CHECK-SD-NEXT:    mov.b v0[3], w8
; CHECK-SD-NEXT:    mov.b v1[3], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #368]
; CHECK-SD-NEXT:    mov.b v3[3], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #240]
; CHECK-SD-NEXT:    ldr w10, [sp, #112]
; CHECK-SD-NEXT:    mov.b v2[4], w4
; CHECK-SD-NEXT:    mov.b v0[4], w8
; CHECK-SD-NEXT:    mov.b v1[4], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #376]
; CHECK-SD-NEXT:    mov.b v3[4], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #248]
; CHECK-SD-NEXT:    ldr w10, [sp, #120]
; CHECK-SD-NEXT:    mov.b v2[5], w5
; CHECK-SD-NEXT:    mov.b v0[5], w8
; CHECK-SD-NEXT:    mov.b v1[5], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #384]
; CHECK-SD-NEXT:    mov.b v3[5], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #256]
; CHECK-SD-NEXT:    ldr w10, [sp, #128]
; CHECK-SD-NEXT:    mov.b v2[6], w6
; CHECK-SD-NEXT:    mov.b v0[6], w8
; CHECK-SD-NEXT:    mov.b v1[6], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #392]
; CHECK-SD-NEXT:    mov.b v3[6], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #264]
; CHECK-SD-NEXT:    ldr w10, [sp, #136]
; CHECK-SD-NEXT:    mov.b v2[7], w7
; CHECK-SD-NEXT:    mov.b v0[7], w8
; CHECK-SD-NEXT:    mov.b v1[7], w9
; CHECK-SD-NEXT:    ldr w8, [sp, #16]
; CHECK-SD-NEXT:    mov.b v3[7], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #400]
; CHECK-SD-NEXT:    ldr w10, [sp, #272]
; CHECK-SD-NEXT:    mov.b v2[8], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #24]
; CHECK-SD-NEXT:    mov.b v0[8], w9
; CHECK-SD-NEXT:    mov.b v1[8], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #408]
; CHECK-SD-NEXT:    mov.b v3[8], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #280]
; CHECK-SD-NEXT:    ldr w11, [sp, #152]
; CHECK-SD-NEXT:    mov.b v2[9], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #32]
; CHECK-SD-NEXT:    mov.b v0[9], w9
; CHECK-SD-NEXT:    mov.b v1[9], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #416]
; CHECK-SD-NEXT:    mov.b v3[9], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #288]
; CHECK-SD-NEXT:    ldr w11, [sp, #160]
; CHECK-SD-NEXT:    mov.b v2[10], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #40]
; CHECK-SD-NEXT:    mov.b v0[10], w9
; CHECK-SD-NEXT:    mov.b v1[10], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #424]
; CHECK-SD-NEXT:    mov.b v3[10], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #296]
; CHECK-SD-NEXT:    ldr w11, [sp, #168]
; CHECK-SD-NEXT:    mov.b v2[11], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #48]
; CHECK-SD-NEXT:    mov.b v0[11], w9
; CHECK-SD-NEXT:    mov.b v1[11], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #432]
; CHECK-SD-NEXT:    mov.b v3[11], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #304]
; CHECK-SD-NEXT:    ldr w11, [sp, #176]
; CHECK-SD-NEXT:    mov.b v2[12], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #56]
; CHECK-SD-NEXT:    mov.b v0[12], w9
; CHECK-SD-NEXT:    mov.b v1[12], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #440]
; CHECK-SD-NEXT:    mov.b v3[12], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #312]
; CHECK-SD-NEXT:    ldr w11, [sp, #184]
; CHECK-SD-NEXT:    mov.b v2[13], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #64]
; CHECK-SD-NEXT:    mov.b v0[13], w9
; CHECK-SD-NEXT:    mov.b v1[13], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #448]
; CHECK-SD-NEXT:    mov.b v3[13], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #320]
; CHECK-SD-NEXT:    ldr w11, [sp, #192]
; CHECK-SD-NEXT:    mov.b v2[14], w8
; CHECK-SD-NEXT:    ldr w8, [sp, #72]
; CHECK-SD-NEXT:    mov.b v0[14], w9
; CHECK-SD-NEXT:    mov.b v1[14], w10
; CHECK-SD-NEXT:    ldr w9, [sp, #456]
; CHECK-SD-NEXT:    mov.b v3[14], w11
; CHECK-SD-NEXT:    ldr w10, [sp, #328]
; CHECK-SD-NEXT:    ldr w11, [sp, #200]
; CHECK-SD-NEXT:    mov.b v2[15], w8
; CHECK-SD-NEXT:    mov.b v0[15], w9
; CHECK-SD-NEXT:    mov.b v1[15], w10
; CHECK-SD-NEXT:    mov.b v3[15], w11
; CHECK-SD-NEXT:    shl.16b v2, v2, #7
; CHECK-SD-NEXT:    shl.16b v4, v1, #7
; CHECK-SD-NEXT:    shl.16b v5, v0, #7
; CHECK-SD-NEXT:    shl.16b v3, v3, #7
; CHECK-SD-NEXT:    cmlt.16b v0, v2, #0
; CHECK-SD-NEXT:    cmlt.16b v2, v4, #0
; CHECK-SD-NEXT:    cmlt.16b v1, v3, #0
; CHECK-SD-NEXT:    cmlt.16b v3, v5, #0
; CHECK-SD-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sext_v64i1:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 16
; CHECK-GI-NEXT:    .cfi_offset w29, -16
; CHECK-GI-NEXT:    ldr w13, [sp, #80]
; CHECK-GI-NEXT:    ldr w11, [sp, #208]
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    ldr w9, [sp, #336]
; CHECK-GI-NEXT:    ldr w8, [sp, #88]
; CHECK-GI-NEXT:    ldr w10, [sp, #216]
; CHECK-GI-NEXT:    fmov s1, w13
; CHECK-GI-NEXT:    fmov s2, w11
; CHECK-GI-NEXT:    ldr w12, [sp, #344]
; CHECK-GI-NEXT:    fmov s3, w9
; CHECK-GI-NEXT:    mov.b v0[1], w1
; CHECK-GI-NEXT:    ldr w9, [sp, #224]
; CHECK-GI-NEXT:    ldr w11, [sp, #400]
; CHECK-GI-NEXT:    mov.b v1[1], w8
; CHECK-GI-NEXT:    mov.b v2[1], w10
; CHECK-GI-NEXT:    ldr w8, [sp, #96]
; CHECK-GI-NEXT:    mov.b v3[1], w12
; CHECK-GI-NEXT:    ldr w10, [sp, #352]
; CHECK-GI-NEXT:    mov.b v0[2], w2
; CHECK-GI-NEXT:    mov.b v1[2], w8
; CHECK-GI-NEXT:    mov.b v2[2], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #104]
; CHECK-GI-NEXT:    mov.b v3[2], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #232]
; CHECK-GI-NEXT:    ldr w10, [sp, #360]
; CHECK-GI-NEXT:    mov.b v0[3], w3
; CHECK-GI-NEXT:    mov.b v1[3], w8
; CHECK-GI-NEXT:    mov.b v2[3], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #112]
; CHECK-GI-NEXT:    mov.b v3[3], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #240]
; CHECK-GI-NEXT:    ldr w10, [sp, #368]
; CHECK-GI-NEXT:    mov.b v0[4], w4
; CHECK-GI-NEXT:    mov.b v1[4], w8
; CHECK-GI-NEXT:    mov.b v2[4], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #120]
; CHECK-GI-NEXT:    mov.b v3[4], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #248]
; CHECK-GI-NEXT:    ldr w10, [sp, #376]
; CHECK-GI-NEXT:    mov.b v0[5], w5
; CHECK-GI-NEXT:    mov.b v1[5], w8
; CHECK-GI-NEXT:    mov.b v2[5], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #128]
; CHECK-GI-NEXT:    mov.b v3[5], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #256]
; CHECK-GI-NEXT:    ldr w10, [sp, #384]
; CHECK-GI-NEXT:    mov.b v0[6], w6
; CHECK-GI-NEXT:    mov.b v1[6], w8
; CHECK-GI-NEXT:    mov.b v2[6], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #136]
; CHECK-GI-NEXT:    mov.b v3[6], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #264]
; CHECK-GI-NEXT:    ldr w10, [sp, #392]
; CHECK-GI-NEXT:    mov.b v0[7], w7
; CHECK-GI-NEXT:    mov.b v1[7], w8
; CHECK-GI-NEXT:    mov.b v2[7], w9
; CHECK-GI-NEXT:    ldr w8, [sp, #16]
; CHECK-GI-NEXT:    mov.b v3[7], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #144]
; CHECK-GI-NEXT:    ldr w10, [sp, #272]
; CHECK-GI-NEXT:    mov.b v0[8], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #24]
; CHECK-GI-NEXT:    mov.b v1[8], w9
; CHECK-GI-NEXT:    mov.b v2[8], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #152]
; CHECK-GI-NEXT:    mov.b v3[8], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #280]
; CHECK-GI-NEXT:    ldr w11, [sp, #408]
; CHECK-GI-NEXT:    mov.b v0[9], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #32]
; CHECK-GI-NEXT:    mov.b v1[9], w9
; CHECK-GI-NEXT:    mov.b v2[9], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #160]
; CHECK-GI-NEXT:    mov.b v3[9], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #288]
; CHECK-GI-NEXT:    ldr w11, [sp, #416]
; CHECK-GI-NEXT:    mov.b v0[10], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #40]
; CHECK-GI-NEXT:    mov.b v1[10], w9
; CHECK-GI-NEXT:    mov.b v2[10], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #168]
; CHECK-GI-NEXT:    mov.b v3[10], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #296]
; CHECK-GI-NEXT:    ldr w11, [sp, #424]
; CHECK-GI-NEXT:    mov.b v0[11], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #48]
; CHECK-GI-NEXT:    mov.b v1[11], w9
; CHECK-GI-NEXT:    mov.b v2[11], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #176]
; CHECK-GI-NEXT:    mov.b v3[11], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #304]
; CHECK-GI-NEXT:    ldr w11, [sp, #432]
; CHECK-GI-NEXT:    mov.b v0[12], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #56]
; CHECK-GI-NEXT:    mov.b v1[12], w9
; CHECK-GI-NEXT:    mov.b v2[12], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #184]
; CHECK-GI-NEXT:    mov.b v3[12], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #312]
; CHECK-GI-NEXT:    ldr w11, [sp, #440]
; CHECK-GI-NEXT:    mov.b v0[13], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #64]
; CHECK-GI-NEXT:    mov.b v1[13], w9
; CHECK-GI-NEXT:    mov.b v2[13], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #192]
; CHECK-GI-NEXT:    mov.b v3[13], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #320]
; CHECK-GI-NEXT:    ldr w11, [sp, #448]
; CHECK-GI-NEXT:    mov.b v0[14], w8
; CHECK-GI-NEXT:    ldr w8, [sp, #72]
; CHECK-GI-NEXT:    mov.b v1[14], w9
; CHECK-GI-NEXT:    mov.b v2[14], w10
; CHECK-GI-NEXT:    ldr w9, [sp, #200]
; CHECK-GI-NEXT:    mov.b v3[14], w11
; CHECK-GI-NEXT:    ldr w10, [sp, #328]
; CHECK-GI-NEXT:    ldr w11, [sp, #456]
; CHECK-GI-NEXT:    mov.b v0[15], w8
; CHECK-GI-NEXT:    mov.b v1[15], w9
; CHECK-GI-NEXT:    mov.b v2[15], w10
; CHECK-GI-NEXT:    mov.b v3[15], w11
; CHECK-GI-NEXT:    shl.16b v0, v0, #7
; CHECK-GI-NEXT:    shl.16b v1, v1, #7
; CHECK-GI-NEXT:    shl.16b v2, v2, #7
; CHECK-GI-NEXT:    shl.16b v3, v3, #7
; CHECK-GI-NEXT:    cmlt.16b v0, v0, #0
; CHECK-GI-NEXT:    cmlt.16b v1, v1, #0
; CHECK-GI-NEXT:    cmlt.16b v2, v2, #0
; CHECK-GI-NEXT:    cmlt.16b v3, v3, #0
; CHECK-GI-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
; CHECK-GI-NEXT:    ret
  %res = sext <64 x i1> %arg to <64 x i8>
  ret <64 x i8> %res
}

; X0 & X1 are the real return registers, SDAG messes with v0 too for unknown reasons.
define <1 x i128> @sext_v1x64(<1 x i64> %arg) {
; CHECK-SD-LABEL: sext_v1x64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    asr x1, x0, #63
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: sext_v1x64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    fmov x8, d0
; CHECK-GI-NEXT:    fmov x0, d0
; CHECK-GI-NEXT:    asr x1, x8, #63
; CHECK-GI-NEXT:    ret
  %res = sext <1 x i64> %arg to <1 x i128>
  ret <1 x i128> %res
}