; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
; RUN: llc -mtriple=aarch64 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-NOFP16
; RUN: llc -mtriple=aarch64 -mattr=+fullfp16 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD,CHECK-SD-FP16
; RUN: llc -mtriple=aarch64 -global-isel -global-isel-abort=1 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-NOFP16
; RUN: llc -mtriple=aarch64 -mattr=+fullfp16 -global-isel -global-isel-abort=1 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-GI,CHECK-GI-FP16

define fp128 @stofp_i128_f128(i128 %a) {
; CHECK-SD-LABEL: stofp_i128_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i128_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floattitf
entry:
  %c = sitofp i128 %a to fp128
  ret fp128 %c
}

define fp128 @utofp_i128_f128(i128 %a) {
; CHECK-SD-LABEL: utofp_i128_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i128_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatuntitf
entry:
  %c = uitofp i128 %a to fp128
  ret fp128 %c
}

define fp128 @stofp_i64_f128(i64 %a) {
; CHECK-SD-LABEL: stofp_i64_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i64_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatditf
entry:
  %c = sitofp i64 %a to fp128
  ret fp128 %c
}

define fp128 @utofp_i64_f128(i64 %a) {
; CHECK-SD-LABEL: utofp_i64_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i64_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatunditf
entry:
  %c = uitofp i64 %a to fp128
  ret fp128 %c
}

define fp128 @stofp_i32_f128(i32 %a) {
; CHECK-SD-LABEL: stofp_i32_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i32_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatsitf
entry:
  %c = sitofp i32 %a to fp128
  ret fp128 %c
}

define fp128 @utofp_i32_f128(i32 %a) {
; CHECK-SD-LABEL: utofp_i32_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i32_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatunsitf
entry:
  %c = uitofp i32 %a to fp128
  ret fp128 %c
}

define fp128 @stofp_i16_f128(i16 %a) {
; CHECK-SD-LABEL: stofp_i16_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    sxth w0, w0
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i16_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sxth w0, w0
; CHECK-GI-NEXT:    b __floatsitf
entry:
  %c = sitofp i16 %a to fp128
  ret fp128 %c
}

define fp128 @utofp_i16_f128(i16 %a) {
; CHECK-SD-LABEL: utofp_i16_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    and w0, w0, #0xffff
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i16_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    and w0, w0, #0xffff
; CHECK-GI-NEXT:    b __floatunsitf
entry:
  %c = uitofp i16 %a to fp128
  ret fp128 %c
}

define fp128 @stofp_i8_f128(i8 %a) {
; CHECK-SD-LABEL: stofp_i8_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    sxtb w0, w0
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i8_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sxtb w0, w0
; CHECK-GI-NEXT:    b __floatsitf
entry:
  %c = sitofp i8 %a to fp128
  ret fp128 %c
}

define fp128 @utofp_i8_f128(i8 %a) {
; CHECK-SD-LABEL: utofp_i8_f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    and w0, w0, #0xff
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i8_f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    and w0, w0, #0xff
; CHECK-GI-NEXT:    b __floatunsitf
entry:
  %c = uitofp i8 %a to fp128
  ret fp128 %c
}

define double @stofp_i128_f64(i128 %a) {
; CHECK-SD-LABEL: stofp_i128_f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i128_f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floattidf
entry:
  %c = sitofp i128 %a to double
  ret double %c
}

define double @utofp_i128_f64(i128 %a) {
; CHECK-SD-LABEL: utofp_i128_f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i128_f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatuntidf
entry:
  %c = uitofp i128 %a to double
  ret double %c
}

define double @stofp_i64_f64(i64 %a) {
; CHECK-LABEL: stofp_i64_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf d0, x0
; CHECK-NEXT:    ret
entry:
  %c = sitofp i64 %a to double
  ret double %c
}

define double @utofp_i64_f64(i64 %a) {
; CHECK-LABEL: utofp_i64_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf d0, x0
; CHECK-NEXT:    ret
entry:
  %c = uitofp i64 %a to double
  ret double %c
}

define double @stofp_i32_f64(i32 %a) {
; CHECK-LABEL: stofp_i32_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf d0, w0
; CHECK-NEXT:    ret
entry:
  %c = sitofp i32 %a to double
  ret double %c
}

define double @utofp_i32_f64(i32 %a) {
; CHECK-LABEL: utofp_i32_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf d0, w0
; CHECK-NEXT:    ret
entry:
  %c = uitofp i32 %a to double
  ret double %c
}

define double @stofp_i16_f64(i16 %a) {
; CHECK-LABEL: stofp_i16_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sxth w8, w0
; CHECK-NEXT:    scvtf d0, w8
; CHECK-NEXT:    ret
entry:
  %c = sitofp i16 %a to double
  ret double %c
}

define double @utofp_i16_f64(i16 %a) {
; CHECK-LABEL: utofp_i16_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    and w8, w0, #0xffff
; CHECK-NEXT:    ucvtf d0, w8
; CHECK-NEXT:    ret
entry:
  %c = uitofp i16 %a to double
  ret double %c
}

define double @stofp_i8_f64(i8 %a) {
; CHECK-LABEL: stofp_i8_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sxtb w8, w0
; CHECK-NEXT:    scvtf d0, w8
; CHECK-NEXT:    ret
entry:
  %c = sitofp i8 %a to double
  ret double %c
}

define double @utofp_i8_f64(i8 %a) {
; CHECK-LABEL: utofp_i8_f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    and w8, w0, #0xff
; CHECK-NEXT:    ucvtf d0, w8
; CHECK-NEXT:    ret
entry:
  %c = uitofp i8 %a to double
  ret double %c
}

define float @stofp_i128_f32(i128 %a) {
; CHECK-SD-LABEL: stofp_i128_f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_i128_f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floattisf
entry:
  %c = sitofp i128 %a to float
  ret float %c
}

define float @utofp_i128_f32(i128 %a) {
; CHECK-SD-LABEL: utofp_i128_f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_i128_f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    b __floatuntisf
entry:
  %c = uitofp i128 %a to float
  ret float %c
}

define float @stofp_i64_f32(i64 %a) {
; CHECK-LABEL: stofp_i64_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf s0, x0
; CHECK-NEXT:    ret
entry:
  %c = sitofp i64 %a to float
  ret float %c
}

define float @utofp_i64_f32(i64 %a) {
; CHECK-LABEL: utofp_i64_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf s0, x0
; CHECK-NEXT:    ret
entry:
  %c = uitofp i64 %a to float
  ret float %c
}

define float @stofp_i32_f32(i32 %a) {
; CHECK-LABEL: stofp_i32_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf s0, w0
; CHECK-NEXT:    ret
entry:
  %c = sitofp i32 %a to float
  ret float %c
}

define float @utofp_i32_f32(i32 %a) {
; CHECK-LABEL: utofp_i32_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf s0, w0
; CHECK-NEXT:    ret
entry:
  %c = uitofp i32 %a to float
  ret float %c
}

define float @stofp_i16_f32(i16 %a) {
; CHECK-LABEL: stofp_i16_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sxth w8, w0
; CHECK-NEXT:    scvtf s0, w8
; CHECK-NEXT:    ret
entry:
  %c = sitofp i16 %a to float
  ret float %c
}

define float @utofp_i16_f32(i16 %a) {
; CHECK-LABEL: utofp_i16_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    and w8, w0, #0xffff
; CHECK-NEXT:    ucvtf s0, w8
; CHECK-NEXT:    ret
entry:
  %c = uitofp i16 %a to float
  ret float %c
}

define float @stofp_i8_f32(i8 %a) {
; CHECK-LABEL: stofp_i8_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sxtb w8, w0
; CHECK-NEXT:    scvtf s0, w8
; CHECK-NEXT:    ret
entry:
  %c = sitofp i8 %a to float
  ret float %c
}

define float @utofp_i8_f32(i8 %a) {
; CHECK-LABEL: utofp_i8_f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    and w8, w0, #0xff
; CHECK-NEXT:    ucvtf s0, w8
; CHECK-NEXT:    ret
entry:
  %c = uitofp i8 %a to float
  ret float %c
}

define half @stofp_i128_f16(i128 %a) {
; CHECK-SD-NOFP16-LABEL: stofp_i128_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_i128_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -16
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_i128_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_i128_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    b __floattihf
entry:
  %c = sitofp i128 %a to half
  ret half %c
}

define half @utofp_i128_f16(i128 %a) {
; CHECK-SD-NOFP16-LABEL: utofp_i128_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_i128_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -16
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_i128_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #-16]! // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp], #16 // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_i128_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    b __floatuntihf
entry:
  %c = uitofp i128 %a to half
  ret half %c
}

define half @stofp_i64_f16(i64 %a) {
; CHECK-SD-NOFP16-LABEL: stofp_i64_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    scvtf s0, x0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_i64_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf h0, x0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_i64_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    scvtf s0, x0
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_i64_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf h0, x0
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp i64 %a to half
  ret half %c
}

define half @utofp_i64_f16(i64 %a) {
; CHECK-SD-NOFP16-LABEL: utofp_i64_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, x0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_i64_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf h0, x0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_i64_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ucvtf s0, x0
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_i64_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf h0, x0
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp i64 %a to half
  ret half %c
}

define half @stofp_i32_f16(i32 %a) {
; CHECK-SD-NOFP16-LABEL: stofp_i32_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    scvtf s0, w0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_i32_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf h0, w0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_i32_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    scvtf s0, w0
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_i32_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf h0, w0
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp i32 %a to half
  ret half %c
}

define half @utofp_i32_f16(i32 %a) {
; CHECK-SD-NOFP16-LABEL: utofp_i32_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, w0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_i32_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf h0, w0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_i32_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ucvtf s0, w0
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_i32_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf h0, w0
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp i32 %a to half
  ret half %c
}

define half @stofp_i16_f16(i16 %a) {
; CHECK-SD-NOFP16-LABEL: stofp_i16_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sxth w8, w0
; CHECK-SD-NOFP16-NEXT:    scvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_i16_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sxth w8, w0
; CHECK-SD-FP16-NEXT:    scvtf h0, w8
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_i16_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sxth w8, w0
; CHECK-GI-NOFP16-NEXT:    scvtf s0, w8
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_i16_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sxth w8, w0
; CHECK-GI-FP16-NEXT:    scvtf h0, w8
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp i16 %a to half
  ret half %c
}

define half @utofp_i16_f16(i16 %a) {
; CHECK-SD-NOFP16-LABEL: utofp_i16_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    and w8, w0, #0xffff
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_i16_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    and w8, w0, #0xffff
; CHECK-SD-FP16-NEXT:    ucvtf h0, w8
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_i16_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    and w8, w0, #0xffff
; CHECK-GI-NOFP16-NEXT:    ucvtf s0, w8
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_i16_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    and w8, w0, #0xffff
; CHECK-GI-FP16-NEXT:    ucvtf h0, w8
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp i16 %a to half
  ret half %c
}

define half @stofp_i8_f16(i8 %a) {
; CHECK-SD-NOFP16-LABEL: stofp_i8_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sxtb w8, w0
; CHECK-SD-NOFP16-NEXT:    scvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_i8_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sxtb w8, w0
; CHECK-SD-FP16-NEXT:    scvtf h0, w8
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_i8_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sxtb w8, w0
; CHECK-GI-NOFP16-NEXT:    scvtf s0, w8
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_i8_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sxtb w8, w0
; CHECK-GI-FP16-NEXT:    scvtf h0, w8
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp i8 %a to half
  ret half %c
}

define half @utofp_i8_f16(i8 %a) {
; CHECK-SD-NOFP16-LABEL: utofp_i8_f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    and w8, w0, #0xff
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_i8_f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    and w8, w0, #0xff
; CHECK-SD-FP16-NEXT:    ucvtf h0, w8
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_i8_f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    and w8, w0, #0xff
; CHECK-GI-NOFP16-NEXT:    ucvtf s0, w8
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_i8_f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    and w8, w0, #0xff
; CHECK-GI-FP16-NEXT:    ucvtf h0, w8
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp i8 %a to half
  ret half %c
}

define <2 x fp128> @stofp_v2i128_v2f128(<2 x i128> %a) {
; CHECK-SD-LABEL: stofp_v2i128_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x3
; CHECK-SD-NEXT:    mov x20, x2
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i128_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floattitf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattitf
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i128> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <2 x fp128> @utofp_v2i128_v2f128(<2 x i128> %a) {
; CHECK-SD-LABEL: utofp_v2i128_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x3
; CHECK-SD-NEXT:    mov x20, x2
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i128_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floatuntitf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntitf
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i128> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <3 x fp128> @stofp_v3i128_v3f128(<3 x i128> %a) {
; CHECK-SD-LABEL: stofp_v3i128_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #80
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    mov x19, x5
; CHECK-SD-NEXT:    mov x20, x4
; CHECK-SD-NEXT:    mov x21, x3
; CHECK-SD-NEXT:    mov x22, x2
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattitf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #80
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i128_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #80
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floattitf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattitf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #80
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i128> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <3 x fp128> @utofp_v3i128_v3f128(<3 x i128> %a) {
; CHECK-SD-LABEL: utofp_v3i128_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #80
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    mov x19, x5
; CHECK-SD-NEXT:    mov x20, x4
; CHECK-SD-NEXT:    mov x21, x3
; CHECK-SD-NEXT:    mov x22, x2
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntitf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #80
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i128_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #80
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floatuntitf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntitf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #80
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i128> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <2 x fp128> @stofp_v2i64_v2f128(<2 x i64> %a) {
; CHECK-SD-LABEL: stofp_v2i64_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov x0, v0.d[1]
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i64_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #32
; CHECK-GI-NEXT:    str d8, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 32
; CHECK-GI-NEXT:    .cfi_offset w30, -8
; CHECK-GI-NEXT:    .cfi_offset b8, -16
; CHECK-GI-NEXT:    fmov x0, d0
; CHECK-GI-NEXT:    mov d8, v0.d[1]
; CHECK-GI-NEXT:    bl __floatditf
; CHECK-GI-NEXT:    fmov x0, d8
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatditf
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldr d8, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #32
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i64> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <2 x fp128> @utofp_v2i64_v2f128(<2 x i64> %a) {
; CHECK-SD-LABEL: utofp_v2i64_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov x0, v0.d[1]
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i64_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #32
; CHECK-GI-NEXT:    str d8, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 32
; CHECK-GI-NEXT:    .cfi_offset w30, -8
; CHECK-GI-NEXT:    .cfi_offset b8, -16
; CHECK-GI-NEXT:    fmov x0, d0
; CHECK-GI-NEXT:    mov d8, v0.d[1]
; CHECK-GI-NEXT:    bl __floatunditf
; CHECK-GI-NEXT:    fmov x0, d8
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunditf
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldr d8, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #32
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i64> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <3 x fp128> @stofp_v3i64_v3f128(<3 x i64> %a) {
; CHECK-SD-LABEL: stofp_v3i64_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    str q2, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    str q1, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    bl __floatditf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i64_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NEXT:    .cfi_offset b8, -24
; CHECK-GI-NEXT:    .cfi_offset b9, -32
; CHECK-GI-NEXT:    fmov x0, d0
; CHECK-GI-NEXT:    fmov d8, d1
; CHECK-GI-NEXT:    fmov d9, d2
; CHECK-GI-NEXT:    bl __floatditf
; CHECK-GI-NEXT:    fmov x0, d8
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatditf
; CHECK-GI-NEXT:    fmov x0, d9
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatditf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i64> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <3 x fp128> @utofp_v3i64_v2f128(<3 x i64> %a) {
; CHECK-SD-LABEL: utofp_v3i64_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    str q2, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    str q1, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov x0, d0
; CHECK-SD-NEXT:    bl __floatunditf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i64_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NEXT:    .cfi_offset b8, -24
; CHECK-GI-NEXT:    .cfi_offset b9, -32
; CHECK-GI-NEXT:    fmov x0, d0
; CHECK-GI-NEXT:    fmov d8, d1
; CHECK-GI-NEXT:    fmov d9, d2
; CHECK-GI-NEXT:    bl __floatunditf
; CHECK-GI-NEXT:    fmov x0, d8
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunditf
; CHECK-GI-NEXT:    fmov x0, d9
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunditf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i64> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <2 x double> @stofp_v2i128_v2f64(<2 x i128> %a) {
; CHECK-SD-LABEL: stofp_v2i128_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x1
; CHECK-SD-NEXT:    mov x20, x0
; CHECK-SD-NEXT:    mov x0, x2
; CHECK-SD-NEXT:    mov x1, x3
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i128_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floattidf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattidf
; CHECK-GI-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.d[1], v0.d[0]
; CHECK-GI-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i128> %a to <2 x double>
  ret <2 x double> %c
}

define <2 x double> @utofp_v2i128_v2f64(<2 x i128> %a) {
; CHECK-SD-LABEL: utofp_v2i128_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x1
; CHECK-SD-NEXT:    mov x20, x0
; CHECK-SD-NEXT:    mov x0, x2
; CHECK-SD-NEXT:    mov x1, x3
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i128_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floatuntidf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntidf
; CHECK-GI-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.d[1], v0.d[0]
; CHECK-GI-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i128> %a to <2 x double>
  ret <2 x double> %c
}

define <3 x double> @stofp_v3i128_v3f64(<3 x i128> %a) {
; CHECK-SD-LABEL: stofp_v3i128_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    stp d9, d8, [sp, #-64]! // 16-byte Folded Spill
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    .cfi_offset b8, -56
; CHECK-SD-NEXT:    .cfi_offset b9, -64
; CHECK-SD-NEXT:    mov x19, x5
; CHECK-SD-NEXT:    mov x20, x4
; CHECK-SD-NEXT:    mov x21, x3
; CHECK-SD-NEXT:    mov x22, x2
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    fmov d8, d0
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    fmov d9, d0
; CHECK-SD-NEXT:    bl __floattidf
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov d2, d0
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov d0, d8
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    fmov d1, d9
; CHECK-SD-NEXT:    ldp d9, d8, [sp], #64 // 16-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i128_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    stp d9, d8, [sp, #-64]! // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    .cfi_offset b8, -56
; CHECK-GI-NEXT:    .cfi_offset b9, -64
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floattidf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    fmov d8, d0
; CHECK-GI-NEXT:    bl __floattidf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    fmov d9, d0
; CHECK-GI-NEXT:    bl __floattidf
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    fmov d2, d0
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    fmov d0, d8
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    fmov d1, d9
; CHECK-GI-NEXT:    ldp d9, d8, [sp], #64 // 16-byte Folded Reload
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i128> %a to <3 x double>
  ret <3 x double> %c
}

define <3 x double> @utofp_v3i128_v3f64(<3 x i128> %a) {
; CHECK-SD-LABEL: utofp_v3i128_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    stp d9, d8, [sp, #-64]! // 16-byte Folded Spill
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    .cfi_offset b8, -56
; CHECK-SD-NEXT:    .cfi_offset b9, -64
; CHECK-SD-NEXT:    mov x19, x5
; CHECK-SD-NEXT:    mov x20, x4
; CHECK-SD-NEXT:    mov x21, x3
; CHECK-SD-NEXT:    mov x22, x2
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    fmov d8, d0
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    fmov d9, d0
; CHECK-SD-NEXT:    bl __floatuntidf
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov d2, d0
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov d0, d8
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    fmov d1, d9
; CHECK-SD-NEXT:    ldp d9, d8, [sp], #64 // 16-byte Folded Reload
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i128_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    stp d9, d8, [sp, #-64]! // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    .cfi_offset b8, -56
; CHECK-GI-NEXT:    .cfi_offset b9, -64
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floatuntidf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    fmov d8, d0
; CHECK-GI-NEXT:    bl __floatuntidf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    fmov d9, d0
; CHECK-GI-NEXT:    bl __floatuntidf
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    fmov d2, d0
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    fmov d0, d8
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    fmov d1, d9
; CHECK-GI-NEXT:    ldp d9, d8, [sp], #64 // 16-byte Folded Reload
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i128> %a to <3 x double>
  ret <3 x double> %c
}

define <2 x double> @stofp_v2i64_v2f64(<2 x i64> %a) {
; CHECK-LABEL: stofp_v2i64_v2f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i64> %a to <2 x double>
  ret <2 x double> %c
}

define <2 x double> @utofp_v2i64_v2f64(<2 x i64> %a) {
; CHECK-LABEL: utofp_v2i64_v2f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i64> %a to <2 x double>
  ret <2 x double> %c
}

define <3 x double> @stofp_v3i64_v3f64(<3 x i64> %a) {
; CHECK-SD-LABEL: stofp_v3i64_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i64_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i64> %a to <3 x double>
  ret <3 x double> %c
}

define <3 x double> @utofp_v3i64_v3f64(<3 x i64> %a) {
; CHECK-SD-LABEL: utofp_v3i64_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i64_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i64> %a to <3 x double>
  ret <3 x double> %c
}

define <4 x double> @stofp_v4i64_v4f64(<4 x i64> %a) {
; CHECK-LABEL: stofp_v4i64_v4f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    scvtf v1.2d, v1.2d
; CHECK-NEXT:    ret
entry:
  %c = sitofp <4 x i64> %a to <4 x double>
  ret <4 x double> %c
}

define <4 x double> @utofp_v4i64_v4f64(<4 x i64> %a) {
; CHECK-LABEL: utofp_v4i64_v4f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-NEXT:    ret
entry:
  %c = uitofp <4 x i64> %a to <4 x double>
  ret <4 x double> %c
}

define <8 x double> @stofp_v8i64_v8f64(<8 x i64> %a) {
; CHECK-LABEL: stofp_v8i64_v8f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    scvtf v1.2d, v1.2d
; CHECK-NEXT:    scvtf v2.2d, v2.2d
; CHECK-NEXT:    scvtf v3.2d, v3.2d
; CHECK-NEXT:    ret
entry:
  %c = sitofp <8 x i64> %a to <8 x double>
  ret <8 x double> %c
}

define <8 x double> @utofp_v8i64_v8f64(<8 x i64> %a) {
; CHECK-LABEL: utofp_v8i64_v8f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-NEXT:    ret
entry:
  %c = uitofp <8 x i64> %a to <8 x double>
  ret <8 x double> %c
}

define <16 x double> @stofp_v16i64_v16f64(<16 x i64> %a) {
; CHECK-LABEL: stofp_v16i64_v16f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    scvtf v1.2d, v1.2d
; CHECK-NEXT:    scvtf v2.2d, v2.2d
; CHECK-NEXT:    scvtf v3.2d, v3.2d
; CHECK-NEXT:    scvtf v4.2d, v4.2d
; CHECK-NEXT:    scvtf v5.2d, v5.2d
; CHECK-NEXT:    scvtf v6.2d, v6.2d
; CHECK-NEXT:    scvtf v7.2d, v7.2d
; CHECK-NEXT:    ret
entry:
  %c = sitofp <16 x i64> %a to <16 x double>
  ret <16 x double> %c
}

define <16 x double> @utofp_v16i64_v16f64(<16 x i64> %a) {
; CHECK-LABEL: utofp_v16i64_v16f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-NEXT:    ret
entry:
  %c = uitofp <16 x i64> %a to <16 x double>
  ret <16 x double> %c
}

define <32 x double> @stofp_v32i64_v32f64(<32 x i64> %a) {
; CHECK-SD-LABEL: stofp_v32i64_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ldp q17, q16, [sp, #96]
; CHECK-SD-NEXT:    scvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ldp q19, q18, [sp, #64]
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ldp q21, q20, [sp, #32]
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v18.2d, v18.2d
; CHECK-SD-NEXT:    scvtf v19.2d, v19.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    scvtf v21.2d, v21.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q5, q6, [x8, #80]
; CHECK-SD-NEXT:    str q16, [x8, #240]
; CHECK-SD-NEXT:    ldp q22, q16, [sp]
; CHECK-SD-NEXT:    stp q3, q4, [x8, #48]
; CHECK-SD-NEXT:    stp q20, q19, [x8, #176]
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    stp q1, q2, [x8, #16]
; CHECK-SD-NEXT:    stp q18, q17, [x8, #208]
; CHECK-SD-NEXT:    scvtf v17.2d, v22.2d
; CHECK-SD-NEXT:    str q0, [x8]
; CHECK-SD-NEXT:    stp q16, q21, [x8, #144]
; CHECK-SD-NEXT:    stp q7, q17, [x8, #112]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i64_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q0, q1, [x8]
; CHECK-GI-NEXT:    scvtf v0.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v6.2d
; CHECK-GI-NEXT:    str q2, [x8, #32]
; CHECK-GI-NEXT:    ldp q2, q5, [sp]
; CHECK-GI-NEXT:    scvtf v6.2d, v7.2d
; CHECK-GI-NEXT:    stp q3, q4, [x8, #48]
; CHECK-GI-NEXT:    ldp q3, q4, [sp, #32]
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    stp q0, q1, [x8, #80]
; CHECK-GI-NEXT:    scvtf v0.2d, v3.2d
; CHECK-GI-NEXT:    ldp q1, q3, [sp, #64]
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q6, q2, [x8, #112]
; CHECK-GI-NEXT:    ldp q2, q6, [sp, #96]
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q5, q0, [x8, #144]
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v6.2d
; CHECK-GI-NEXT:    stp q4, q1, [x8, #176]
; CHECK-GI-NEXT:    stp q3, q2, [x8, #208]
; CHECK-GI-NEXT:    str q0, [x8, #240]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i64> %a to <32 x double>
  ret <32 x double> %c
}

define <32 x double> @utofp_v32i64_v32f64(<32 x i64> %a) {
; CHECK-SD-LABEL: utofp_v32i64_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ldp q17, q16, [sp, #96]
; CHECK-SD-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ldp q19, q18, [sp, #64]
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ldp q21, q20, [sp, #32]
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v18.2d, v18.2d
; CHECK-SD-NEXT:    ucvtf v19.2d, v19.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    ucvtf v21.2d, v21.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q5, q6, [x8, #80]
; CHECK-SD-NEXT:    str q16, [x8, #240]
; CHECK-SD-NEXT:    ldp q22, q16, [sp]
; CHECK-SD-NEXT:    stp q3, q4, [x8, #48]
; CHECK-SD-NEXT:    stp q20, q19, [x8, #176]
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    stp q1, q2, [x8, #16]
; CHECK-SD-NEXT:    stp q18, q17, [x8, #208]
; CHECK-SD-NEXT:    ucvtf v17.2d, v22.2d
; CHECK-SD-NEXT:    str q0, [x8]
; CHECK-SD-NEXT:    stp q16, q21, [x8, #144]
; CHECK-SD-NEXT:    stp q7, q17, [x8, #112]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i64_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q0, q1, [x8]
; CHECK-GI-NEXT:    ucvtf v0.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v6.2d
; CHECK-GI-NEXT:    str q2, [x8, #32]
; CHECK-GI-NEXT:    ldp q2, q5, [sp]
; CHECK-GI-NEXT:    ucvtf v6.2d, v7.2d
; CHECK-GI-NEXT:    stp q3, q4, [x8, #48]
; CHECK-GI-NEXT:    ldp q3, q4, [sp, #32]
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    stp q0, q1, [x8, #80]
; CHECK-GI-NEXT:    ucvtf v0.2d, v3.2d
; CHECK-GI-NEXT:    ldp q1, q3, [sp, #64]
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q6, q2, [x8, #112]
; CHECK-GI-NEXT:    ldp q2, q6, [sp, #96]
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q5, q0, [x8, #144]
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v6.2d
; CHECK-GI-NEXT:    stp q4, q1, [x8, #176]
; CHECK-GI-NEXT:    stp q3, q2, [x8, #208]
; CHECK-GI-NEXT:    str q0, [x8, #240]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i64> %a to <32 x double>
  ret <32 x double> %c
}

define <2 x fp128> @stofp_v2i32_v2f128(<2 x i32> %a) {
; CHECK-SD-LABEL: stofp_v2i32_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #32
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 32
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov w0, s0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    mov w0, v1.s[1]
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #32
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i32_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #32
; CHECK-GI-NEXT:    str d8, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 32
; CHECK-GI-NEXT:    .cfi_offset w30, -8
; CHECK-GI-NEXT:    .cfi_offset b8, -16
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    fmov w0, s0
; CHECK-GI-NEXT:    mov s8, v0.s[1]
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    fmov w0, s8
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldr d8, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #32
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i32> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <2 x fp128> @utofp_v2i32_v2f128(<2 x i32> %a) {
; CHECK-SD-LABEL: utofp_v2i32_v2f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #32
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 32
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov w0, s0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    mov w0, v1.s[1]
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #32
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i32_v2f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #32
; CHECK-GI-NEXT:    str d8, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #24] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 32
; CHECK-GI-NEXT:    .cfi_offset w30, -8
; CHECK-GI-NEXT:    .cfi_offset b8, -16
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    fmov w0, s0
; CHECK-GI-NEXT:    mov s8, v0.s[1]
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    fmov w0, s8
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    mov v1.16b, v0.16b
; CHECK-GI-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #24] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldr d8, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #32
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i32> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <3 x fp128> @stofp_v3i32_v3f128(<3 x i32> %a) {
; CHECK-SD-LABEL: stofp_v3i32_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    str q0, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    fmov w0, s1
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov w0, s0
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov w0, v0.s[1]
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldp q0, q2, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i32_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NEXT:    .cfi_offset b8, -24
; CHECK-GI-NEXT:    .cfi_offset b9, -32
; CHECK-GI-NEXT:    fmov w0, s0
; CHECK-GI-NEXT:    mov s8, v0.s[1]
; CHECK-GI-NEXT:    mov s9, v0.s[2]
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    fmov w0, s8
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    fmov w0, s9
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i32> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <3 x fp128> @utofp_v3i32_v3f128(<3 x i32> %a) {
; CHECK-SD-LABEL: utofp_v3i32_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w30, -16
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    str q0, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    fmov w0, s1
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    fmov w0, s0
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    ldr q0, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov w0, v0.s[1]
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    mov v1.16b, v0.16b
; CHECK-SD-NEXT:    ldp q0, q2, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i32_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    str x30, [sp, #48] // 8-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w30, -16
; CHECK-GI-NEXT:    .cfi_offset b8, -24
; CHECK-GI-NEXT:    .cfi_offset b9, -32
; CHECK-GI-NEXT:    fmov w0, s0
; CHECK-GI-NEXT:    mov s8, v0.s[1]
; CHECK-GI-NEXT:    mov s9, v0.s[2]
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    fmov w0, s8
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    fmov w0, s9
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #48] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i32> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <2 x double> @stofp_v2i32_v2f64(<2 x i32> %a) {
; CHECK-LABEL: stofp_v2i32_v2f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i32> %a to <2 x double>
  ret <2 x double> %c
}

define <2 x double> @utofp_v2i32_v2f64(<2 x i32> %a) {
; CHECK-LABEL: utofp_v2i32_v2f64:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i32> %a to <2 x double>
  ret <2 x double> %c
}

define <3 x double> @stofp_v3i32_v3f64(<3 x i32> %a) {
; CHECK-SD-LABEL: stofp_v3i32_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v1.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll2 v0.2d, v0.4s, #0
; CHECK-SD-NEXT:    scvtf v3.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v0.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    fmov d0, d3
; CHECK-SD-NEXT:    ext v1.16b, v3.16b, v3.16b, #8
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i32_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v3.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v0.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    mov d1, v3.d[1]
; CHECK-GI-NEXT:    fmov d0, d3
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i32> %a to <3 x double>
  ret <3 x double> %c
}

define <3 x double> @utofp_v3i32_v3f64(<3 x i32> %a) {
; CHECK-SD-LABEL: utofp_v3i32_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v1.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-SD-NEXT:    ucvtf v3.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v0.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    fmov d0, d3
; CHECK-SD-NEXT:    ext v1.16b, v3.16b, v3.16b, #8
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i32_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v3.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v0.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    mov d1, v3.d[1]
; CHECK-GI-NEXT:    fmov d0, d3
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i32> %a to <3 x double>
  ret <3 x double> %c
}

define <4 x double> @stofp_v4i32_v4f64(<4 x i32> %a) {
; CHECK-SD-LABEL: stofp_v4i32_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v4i32_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <4 x i32> %a to <4 x double>
  ret <4 x double> %c
}

define <4 x double> @utofp_v4i32_v4f64(<4 x i32> %a) {
; CHECK-SD-LABEL: utofp_v4i32_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v4i32_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <4 x i32> %a to <4 x double>
  ret <4 x double> %c
}

define <8 x double> @stofp_v8i32_v8f64(<8 x i32> %a) {
; CHECK-SD-LABEL: stofp_v8i32_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v2.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll2 v3.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll2 v4.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll v5.2d, v1.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v5.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i32_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v2.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v3.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v4.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v5.2d, v1.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v5.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i32> %a to <8 x double>
  ret <8 x double> %c
}

define <8 x double> @utofp_v8i32_v8f64(<8 x i32> %a) {
; CHECK-SD-LABEL: utofp_v8i32_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v2.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll2 v3.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll2 v4.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll v5.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v5.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i32_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v2.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v3.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v4.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v5.2d, v1.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v5.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i32> %a to <8 x double>
  ret <8 x double> %c
}

define <16 x double> @stofp_v16i32_v16f64(<16 x i32> %a) {
; CHECK-SD-LABEL: stofp_v16i32_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v4.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll2 v5.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll v6.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshll v7.2d, v2.2s, #0
; CHECK-SD-NEXT:    sshll2 v16.2d, v2.4s, #0
; CHECK-SD-NEXT:    sshll2 v17.2d, v3.4s, #0
; CHECK-SD-NEXT:    sshll v18.2d, v3.2s, #0
; CHECK-SD-NEXT:    scvtf v1.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v7.2d, v17.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v18.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i32_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v4.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v5.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v6.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v7.2d, v1.4s, #0
; CHECK-GI-NEXT:    sshll v16.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v17.2d, v2.4s, #0
; CHECK-GI-NEXT:    sshll v18.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v19.2d, v3.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v18.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v19.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i32> %a to <16 x double>
  ret <16 x double> %c
}

define <16 x double> @utofp_v16i32_v16f64(<16 x i32> %a) {
; CHECK-SD-LABEL: utofp_v16i32_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v4.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll2 v5.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll v6.2d, v1.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v2.2s, #0
; CHECK-SD-NEXT:    ushll2 v16.2d, v2.4s, #0
; CHECK-SD-NEXT:    ushll2 v17.2d, v3.4s, #0
; CHECK-SD-NEXT:    ushll v18.2d, v3.2s, #0
; CHECK-SD-NEXT:    ucvtf v1.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v16.2d
; CHECK-SD-NEXT:    ucvtf v7.2d, v17.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v18.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i32_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v4.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v5.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v6.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v7.2d, v1.4s, #0
; CHECK-GI-NEXT:    ushll v16.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v17.2d, v2.4s, #0
; CHECK-GI-NEXT:    ushll v18.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v19.2d, v3.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v18.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v19.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i32> %a to <16 x double>
  ret <16 x double> %c
}

define <32 x double> @stofp_v32i32_v32f64(<32 x i32> %a) {
; CHECK-SD-LABEL: stofp_v32i32_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v17.2d, v6.4s, #0
; CHECK-SD-NEXT:    sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    sshll2 v16.2d, v7.4s, #0
; CHECK-SD-NEXT:    sshll v19.2d, v3.2s, #0
; CHECK-SD-NEXT:    sshll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    sshll2 v3.2d, v3.4s, #0
; CHECK-SD-NEXT:    sshll2 v20.2d, v4.4s, #0
; CHECK-SD-NEXT:    sshll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    sshll2 v18.2d, v5.4s, #0
; CHECK-SD-NEXT:    scvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    sshll2 v21.2d, v2.4s, #0
; CHECK-SD-NEXT:    scvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    sshll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    scvtf v18.2d, v18.2d
; CHECK-SD-NEXT:    stp q6, q17, [x8, #192]
; CHECK-SD-NEXT:    scvtf v17.2d, v19.2d
; CHECK-SD-NEXT:    stp q7, q16, [x8, #224]
; CHECK-SD-NEXT:    sshll2 v7.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    stp q4, q20, [x8, #128]
; CHECK-SD-NEXT:    scvtf v16.2d, v21.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    stp q17, q3, [x8, #96]
; CHECK-SD-NEXT:    sshll2 v3.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    scvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q5, q18, [x8, #160]
; CHECK-SD-NEXT:    stp q2, q16, [x8, #64]
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    stp q0, q4, [x8]
; CHECK-SD-NEXT:    stp q1, q3, [x8, #32]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i32_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v16.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v17.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    sshll v18.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    sshll v19.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    scvtf v16.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v18.2d, v18.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q16, q0, [x8]
; CHECK-GI-NEXT:    sshll v16.2d, v4.2s, #0
; CHECK-GI-NEXT:    sshll2 v0.2d, v4.4s, #0
; CHECK-GI-NEXT:    scvtf v4.2d, v19.2d
; CHECK-GI-NEXT:    stp q17, q1, [x8, #32]
; CHECK-GI-NEXT:    sshll v17.2d, v5.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v5.4s, #0
; CHECK-GI-NEXT:    stp q18, q2, [x8, #64]
; CHECK-GI-NEXT:    sshll2 v2.2d, v6.4s, #0
; CHECK-GI-NEXT:    scvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    sshll v16.2d, v6.2s, #0
; CHECK-GI-NEXT:    scvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    stp q4, q3, [x8, #96]
; CHECK-GI-NEXT:    sshll v4.2d, v7.2s, #0
; CHECK-GI-NEXT:    sshll2 v3.2d, v7.4s, #0
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    stp q5, q0, [x8, #128]
; CHECK-GI-NEXT:    scvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q6, q1, [x8, #160]
; CHECK-GI-NEXT:    stp q7, q2, [x8, #192]
; CHECK-GI-NEXT:    stp q0, q3, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i32> %a to <32 x double>
  ret <32 x double> %c
}

define <32 x double> @utofp_v32i32_v32f64(<32 x i32> %a) {
; CHECK-SD-LABEL: utofp_v32i32_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v17.2d, v6.4s, #0
; CHECK-SD-NEXT:    ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    ushll2 v16.2d, v7.4s, #0
; CHECK-SD-NEXT:    ushll v19.2d, v3.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    ushll2 v3.2d, v3.4s, #0
; CHECK-SD-NEXT:    ushll2 v20.2d, v4.4s, #0
; CHECK-SD-NEXT:    ushll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    ushll2 v18.2d, v5.4s, #0
; CHECK-SD-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ushll2 v21.2d, v2.4s, #0
; CHECK-SD-NEXT:    ucvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ushll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ucvtf v18.2d, v18.2d
; CHECK-SD-NEXT:    stp q6, q17, [x8, #192]
; CHECK-SD-NEXT:    ucvtf v17.2d, v19.2d
; CHECK-SD-NEXT:    stp q7, q16, [x8, #224]
; CHECK-SD-NEXT:    ushll2 v7.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    stp q4, q20, [x8, #128]
; CHECK-SD-NEXT:    ucvtf v16.2d, v21.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    stp q17, q3, [x8, #96]
; CHECK-SD-NEXT:    ushll2 v3.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q5, q18, [x8, #160]
; CHECK-SD-NEXT:    stp q2, q16, [x8, #64]
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    stp q0, q4, [x8]
; CHECK-SD-NEXT:    stp q1, q3, [x8, #32]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i32_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v16.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v17.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ushll v18.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    ushll v19.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v18.2d, v18.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q16, q0, [x8]
; CHECK-GI-NEXT:    ushll v16.2d, v4.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v4.4s, #0
; CHECK-GI-NEXT:    ucvtf v4.2d, v19.2d
; CHECK-GI-NEXT:    stp q17, q1, [x8, #32]
; CHECK-GI-NEXT:    ushll v17.2d, v5.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v5.4s, #0
; CHECK-GI-NEXT:    stp q18, q2, [x8, #64]
; CHECK-GI-NEXT:    ushll2 v2.2d, v6.4s, #0
; CHECK-GI-NEXT:    ucvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ushll v16.2d, v6.2s, #0
; CHECK-GI-NEXT:    ucvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    stp q4, q3, [x8, #96]
; CHECK-GI-NEXT:    ushll v4.2d, v7.2s, #0
; CHECK-GI-NEXT:    ushll2 v3.2d, v7.4s, #0
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    stp q5, q0, [x8, #128]
; CHECK-GI-NEXT:    ucvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    stp q6, q1, [x8, #160]
; CHECK-GI-NEXT:    stp q7, q2, [x8, #192]
; CHECK-GI-NEXT:    stp q0, q3, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i32> %a to <32 x double>
  ret <32 x double> %c
}

define <2 x fp128> @stofp_v2i16_v2f128(<2 x i16> %a) {
; CHECK-LABEL: stofp_v2i16_v2f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #32
; CHECK-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    fmov w8, s0
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    sxth w0, w8
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    mov w8, v1.s[1]
; CHECK-NEXT:    sxth w0, w8
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    mov v1.16b, v0.16b
; CHECK-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #32
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i16> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <2 x fp128> @utofp_v2i16_v2f128(<2 x i16> %a) {
; CHECK-LABEL: utofp_v2i16_v2f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #32
; CHECK-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    fmov w8, s0
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    and w0, w8, #0xffff
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    mov w8, v1.s[1]
; CHECK-NEXT:    and w0, w8, #0xffff
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    mov v1.16b, v0.16b
; CHECK-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #32
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i16> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <3 x fp128> @stofp_v3i16_v3f128(<3 x i16> %a) {
; CHECK-LABEL: stofp_v3i16_v3f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #48
; CHECK-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 48
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    smov w0, v0.h[0]
; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    ldr q1, [sp, #16] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    smov w0, v1.h[1]
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    ldr q1, [sp, #16] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-NEXT:    smov w0, v1.h[2]
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    mov v2.16b, v0.16b
; CHECK-NEXT:    ldp q0, q1, [sp] // 32-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #48
; CHECK-NEXT:    ret
entry:
  %c = sitofp <3 x i16> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <3 x fp128> @utofp_v3i16_v3f128(<3 x i16> %a) {
; CHECK-LABEL: utofp_v3i16_v3f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #48
; CHECK-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 48
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    umov w0, v0.h[0]
; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    ldr q1, [sp, #16] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    umov w0, v1.h[1]
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    ldr q1, [sp, #16] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-NEXT:    umov w0, v1.h[2]
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    mov v2.16b, v0.16b
; CHECK-NEXT:    ldp q0, q1, [sp] // 32-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #48
; CHECK-NEXT:    ret
entry:
  %c = uitofp <3 x i16> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <2 x double> @stofp_v2i16_v2f64(<2 x i16> %a) {
; CHECK-SD-LABEL: stofp_v2i16_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i16_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    shl v0.2d, v0.2d, #48
; CHECK-GI-NEXT:    sshr v0.2d, v0.2d, #48
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i16> %a to <2 x double>
  ret <2 x double> %c
}

define <2 x double> @utofp_v2i16_v2f64(<2 x i16> %a) {
; CHECK-SD-LABEL: utofp_v2i16_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    movi d1, #0x00ffff0000ffff
; CHECK-SD-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i16_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    movi v1.2d, #0x0000000000ffff
; CHECK-GI-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i16> %a to <2 x double>
  ret <2 x double> %c
}

define <3 x double> @stofp_v3i16_v3f64(<3 x i16> %a) {
; CHECK-SD-LABEL: stofp_v3i16_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll v0.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v1.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i16_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v1.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i16> %a to <3 x double>
  ret <3 x double> %c
}

define <3 x double> @utofp_v3i16_v3f64(<3 x i16> %a) {
; CHECK-SD-LABEL: utofp_v3i16_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll v0.2d, v1.2s, #0
; CHECK-SD-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v1.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i16_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll v0.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v1.2d
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i16> %a to <3 x double>
  ret <3 x double> %c
}

define <4 x double> @stofp_v4i16_v4f64(<4 x i16> %a) {
; CHECK-SD-LABEL: stofp_v4i16_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v4i16_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <4 x i16> %a to <4 x double>
  ret <4 x double> %c
}

define <4 x double> @utofp_v4i16_v4f64(<4 x i16> %a) {
; CHECK-SD-LABEL: utofp_v4i16_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v1.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v4i16_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <4 x i16> %a to <4 x double>
  ret <4 x double> %c
}

define <8 x double> @stofp_v8i16_v8f64(<8 x i16> %a) {
; CHECK-SD-LABEL: stofp_v8i16_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll v2.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshll2 v3.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll v4.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v4.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i16_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v2.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    sshll v3.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v4.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v4.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i16> %a to <8 x double>
  ret <8 x double> %c
}

define <8 x double> @utofp_v8i16_v8f64(<8 x i16> %a) {
; CHECK-SD-LABEL: utofp_v8i16_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll v2.2d, v1.2s, #0
; CHECK-SD-NEXT:    ushll2 v3.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll v4.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v4.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i16_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v2.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ushll v3.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v4.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i16> %a to <8 x double>
  ret <8 x double> %c
}

define <16 x double> @stofp_v16i16_v16f64(<16 x i16> %a) {
; CHECK-SD-LABEL: stofp_v16i16_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll v3.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-SD-NEXT:    sshll2 v4.2d, v2.4s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    sshll2 v5.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v6.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll v7.2d, v3.2s, #0
; CHECK-SD-NEXT:    sshll2 v16.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll2 v17.2d, v3.4s, #0
; CHECK-SD-NEXT:    sshll v18.2d, v1.2s, #0
; CHECK-SD-NEXT:    scvtf v1.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v7.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v17.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v18.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i16_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v3.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v4.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    sshll v5.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v6.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v7.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v16.2d, v3.4s, #0
; CHECK-GI-NEXT:    sshll v17.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v18.2d, v1.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v18.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i16> %a to <16 x double>
  ret <16 x double> %c
}

define <16 x double> @utofp_v16i16_v16f64(<16 x i16> %a) {
; CHECK-SD-LABEL: utofp_v16i16_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll v3.4s, v1.4h, #0
; CHECK-SD-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-SD-NEXT:    ushll2 v4.2d, v2.4s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ushll2 v5.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v6.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v3.2s, #0
; CHECK-SD-NEXT:    ushll2 v16.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll2 v17.2d, v3.4s, #0
; CHECK-SD-NEXT:    ushll v18.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v1.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v7.2d, v16.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v17.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v18.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i16_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v3.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v4.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    ushll v5.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v6.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v7.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v16.2d, v3.4s, #0
; CHECK-GI-NEXT:    ushll v17.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v18.2d, v1.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v18.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i16> %a to <16 x double>
  ret <16 x double> %c
}

define <32 x double> @stofp_v32i16_v32f64(<32 x i16> %a) {
; CHECK-SD-LABEL: stofp_v32i16_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v4.4s, v3.8h, #0
; CHECK-SD-NEXT:    sshll2 v5.4s, v2.8h, #0
; CHECK-SD-NEXT:    sshll2 v16.4s, v1.8h, #0
; CHECK-SD-NEXT:    sshll2 v17.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll v3.4s, v3.4h, #0
; CHECK-SD-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v6.2d, v4.4s, #0
; CHECK-SD-NEXT:    sshll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    sshll2 v7.2d, v5.4s, #0
; CHECK-SD-NEXT:    sshll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    sshll2 v18.2d, v16.4s, #0
; CHECK-SD-NEXT:    sshll v16.2d, v16.2s, #0
; CHECK-SD-NEXT:    sshll2 v19.2d, v17.4s, #0
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    stp q4, q6, [x8, #224]
; CHECK-SD-NEXT:    sshll v6.2d, v17.2s, #0
; CHECK-SD-NEXT:    scvtf v17.2d, v18.2d
; CHECK-SD-NEXT:    stp q5, q7, [x8, #160]
; CHECK-SD-NEXT:    sshll2 v7.2d, v3.4s, #0
; CHECK-SD-NEXT:    sshll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    scvtf v4.2d, v19.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v6.2d
; CHECK-SD-NEXT:    sshll2 v6.2d, v2.4s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    scvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    stp q16, q17, [x8, #96]
; CHECK-SD-NEXT:    sshll2 v16.2d, v1.4s, #0
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    stp q5, q4, [x8, #32]
; CHECK-SD-NEXT:    sshll2 v4.2d, v0.4s, #0
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v5.2d, v16.2d
; CHECK-SD-NEXT:    stp q3, q7, [x8, #192]
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q2, q6, [x8, #128]
; CHECK-SD-NEXT:    stp q1, q5, [x8, #64]
; CHECK-SD-NEXT:    stp q0, q3, [x8]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i16_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v4.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v6.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v16.4s, v2.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    sshll v18.4s, v3.4h, #0
; CHECK-GI-NEXT:    sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT:    sshll v5.2d, v4.2s, #0
; CHECK-GI-NEXT:    sshll2 v4.2d, v4.4s, #0
; CHECK-GI-NEXT:    sshll v7.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v17.2d, v6.2s, #0
; CHECK-GI-NEXT:    sshll2 v6.2d, v6.4s, #0
; CHECK-GI-NEXT:    sshll v19.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q5, q4, [x8]
; CHECK-GI-NEXT:    scvtf v5.2d, v17.2d
; CHECK-GI-NEXT:    sshll v4.2d, v16.2s, #0
; CHECK-GI-NEXT:    stp q7, q0, [x8, #32]
; CHECK-GI-NEXT:    sshll2 v7.2d, v16.4s, #0
; CHECK-GI-NEXT:    scvtf v16.2d, v19.2d
; CHECK-GI-NEXT:    sshll v0.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q5, q6, [x8, #64]
; CHECK-GI-NEXT:    scvtf v6.2d, v7.2d
; CHECK-GI-NEXT:    sshll v5.2d, v18.2s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    stp q16, q1, [x8, #96]
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    sshll2 v7.2d, v18.4s, #0
; CHECK-GI-NEXT:    sshll v1.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    stp q4, q6, [x8, #128]
; CHECK-GI-NEXT:    scvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q0, q2, [x8, #160]
; CHECK-GI-NEXT:    scvtf v0.2d, v3.2d
; CHECK-GI-NEXT:    stp q5, q4, [x8, #192]
; CHECK-GI-NEXT:    stp q1, q0, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i16> %a to <32 x double>
  ret <32 x double> %c
}

define <32 x double> @utofp_v32i16_v32f64(<32 x i16> %a) {
; CHECK-SD-LABEL: utofp_v32i16_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v4.4s, v3.8h, #0
; CHECK-SD-NEXT:    ushll2 v5.4s, v2.8h, #0
; CHECK-SD-NEXT:    ushll2 v16.4s, v1.8h, #0
; CHECK-SD-NEXT:    ushll2 v17.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll v3.4s, v3.4h, #0
; CHECK-SD-NEXT:    ushll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    ushll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v6.2d, v4.4s, #0
; CHECK-SD-NEXT:    ushll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    ushll2 v7.2d, v5.4s, #0
; CHECK-SD-NEXT:    ushll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    ushll2 v18.2d, v16.4s, #0
; CHECK-SD-NEXT:    ushll v16.2d, v16.2s, #0
; CHECK-SD-NEXT:    ushll2 v19.2d, v17.4s, #0
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    stp q4, q6, [x8, #224]
; CHECK-SD-NEXT:    ushll v6.2d, v17.2s, #0
; CHECK-SD-NEXT:    ucvtf v17.2d, v18.2d
; CHECK-SD-NEXT:    stp q5, q7, [x8, #160]
; CHECK-SD-NEXT:    ushll2 v7.2d, v3.4s, #0
; CHECK-SD-NEXT:    ushll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    ucvtf v4.2d, v19.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v6.2d
; CHECK-SD-NEXT:    ushll2 v6.2d, v2.4s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    stp q16, q17, [x8, #96]
; CHECK-SD-NEXT:    ushll2 v16.2d, v1.4s, #0
; CHECK-SD-NEXT:    ushll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    stp q5, q4, [x8, #32]
; CHECK-SD-NEXT:    ushll2 v4.2d, v0.4s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v5.2d, v16.2d
; CHECK-SD-NEXT:    stp q3, q7, [x8, #192]
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q2, q6, [x8, #128]
; CHECK-SD-NEXT:    stp q1, q5, [x8, #64]
; CHECK-SD-NEXT:    stp q0, q3, [x8]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i16_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v4.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v6.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v16.4s, v2.4h, #0
; CHECK-GI-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    ushll v18.4s, v3.4h, #0
; CHECK-GI-NEXT:    ushll2 v3.4s, v3.8h, #0
; CHECK-GI-NEXT:    ushll v5.2d, v4.2s, #0
; CHECK-GI-NEXT:    ushll2 v4.2d, v4.4s, #0
; CHECK-GI-NEXT:    ushll v7.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v17.2d, v6.2s, #0
; CHECK-GI-NEXT:    ushll2 v6.2d, v6.4s, #0
; CHECK-GI-NEXT:    ushll v19.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q5, q4, [x8]
; CHECK-GI-NEXT:    ucvtf v5.2d, v17.2d
; CHECK-GI-NEXT:    ushll v4.2d, v16.2s, #0
; CHECK-GI-NEXT:    stp q7, q0, [x8, #32]
; CHECK-GI-NEXT:    ushll2 v7.2d, v16.4s, #0
; CHECK-GI-NEXT:    ucvtf v16.2d, v19.2d
; CHECK-GI-NEXT:    ushll v0.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    stp q5, q6, [x8, #64]
; CHECK-GI-NEXT:    ucvtf v6.2d, v7.2d
; CHECK-GI-NEXT:    ushll v5.2d, v18.2s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    stp q16, q1, [x8, #96]
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ushll2 v7.2d, v18.4s, #0
; CHECK-GI-NEXT:    ushll v1.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    stp q4, q6, [x8, #128]
; CHECK-GI-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q0, q2, [x8, #160]
; CHECK-GI-NEXT:    ucvtf v0.2d, v3.2d
; CHECK-GI-NEXT:    stp q5, q4, [x8, #192]
; CHECK-GI-NEXT:    stp q1, q0, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i16> %a to <32 x double>
  ret <32 x double> %c
}

define <2 x fp128> @stofp_v2i8_v2f128(<2 x i8> %a) {
; CHECK-LABEL: stofp_v2i8_v2f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #32
; CHECK-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    fmov w8, s0
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    sxtb w0, w8
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    mov w8, v1.s[1]
; CHECK-NEXT:    sxtb w0, w8
; CHECK-NEXT:    bl __floatsitf
; CHECK-NEXT:    mov v1.16b, v0.16b
; CHECK-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #32
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i8> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <2 x fp128> @utofp_v2i8_v2f128(<2 x i8> %a) {
; CHECK-LABEL: utofp_v2i8_v2f128:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sub sp, sp, #32
; CHECK-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-NEXT:    .cfi_def_cfa_offset 32
; CHECK-NEXT:    .cfi_offset w30, -16
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    fmov w8, s0
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    and w0, w8, #0xff
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-NEXT:    mov w8, v1.s[1]
; CHECK-NEXT:    and w0, w8, #0xff
; CHECK-NEXT:    bl __floatunsitf
; CHECK-NEXT:    mov v1.16b, v0.16b
; CHECK-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-NEXT:    add sp, sp, #32
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i8> %a to <2 x fp128>
  ret <2 x fp128> %c
}

define <3 x fp128> @stofp_v2i8_v3f128(<3 x i8> %a) {
; CHECK-SD-LABEL: stofp_v2i8_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    sxtb w0, w0
; CHECK-SD-NEXT:    mov w19, w2
; CHECK-SD-NEXT:    mov w20, w1
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    sxtb w0, w20
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    sxtb w0, w19
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatsitf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i8_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    sxtb w0, w0
; CHECK-GI-NEXT:    mov w19, w1
; CHECK-GI-NEXT:    mov w20, w2
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    sxtb w0, w19
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    sxtb w0, w20
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatsitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i8> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <3 x fp128> @utofp_v2i8_v3f128(<3 x i8> %a) {
; CHECK-SD-LABEL: utofp_v2i8_v3f128:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #64
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 64
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    and w0, w0, #0xff
; CHECK-SD-NEXT:    mov w19, w2
; CHECK-SD-NEXT:    mov w20, w1
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    and w0, w20, #0xff
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    and w0, w19, #0xff
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatunsitf
; CHECK-SD-NEXT:    mov v2.16b, v0.16b
; CHECK-SD-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    add sp, sp, #64
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i8_v3f128:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #64
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 64
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    and w0, w0, #0xff
; CHECK-GI-NEXT:    mov w19, w1
; CHECK-GI-NEXT:    mov w20, w2
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    and w0, w19, #0xff
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    and w0, w20, #0xff
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatunsitf
; CHECK-GI-NEXT:    mov v2.16b, v0.16b
; CHECK-GI-NEXT:    ldp q1, q0, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    add sp, sp, #64
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i8> %a to <3 x fp128>
  ret <3 x fp128> %c
}

define <2 x double> @stofp_v2i8_v2f64(<2 x i8> %a) {
; CHECK-SD-LABEL: stofp_v2i8_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i8_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    shl v0.2d, v0.2d, #56
; CHECK-GI-NEXT:    sshr v0.2d, v0.2d, #56
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i8> %a to <2 x double>
  ret <2 x double> %c
}

define <2 x double> @utofp_v2i8_v2f64(<2 x i8> %a) {
; CHECK-SD-LABEL: utofp_v2i8_v2f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-SD-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i8_v2f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    movi v1.2d, #0x000000000000ff
; CHECK-GI-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-GI-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i8> %a to <2 x double>
  ret <2 x double> %c
}

define <3 x double> @stofp_v3i8_v3f64(<3 x i8> %a) {
; CHECK-SD-LABEL: stofp_v3i8_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov s0, w0
; CHECK-SD-NEXT:    fmov s1, w2
; CHECK-SD-NEXT:    mov v0.s[1], w1
; CHECK-SD-NEXT:    shl v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    scvtf v2.2d, v1.2d
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i8_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    mov v0.h[1], w1
; CHECK-GI-NEXT:    mov v0.h[2], w2
; CHECK-GI-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-GI-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-GI-NEXT:    smov x8, v0.h[0]
; CHECK-GI-NEXT:    smov x9, v0.h[1]
; CHECK-GI-NEXT:    fmov d1, x8
; CHECK-GI-NEXT:    smov x8, v0.h[2]
; CHECK-GI-NEXT:    mov v1.d[1], x9
; CHECK-GI-NEXT:    smov x9, v0.h[3]
; CHECK-GI-NEXT:    fmov d2, x8
; CHECK-GI-NEXT:    scvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    mov v2.d[1], x9
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i8> %a to <3 x double>
  ret <3 x double> %c
}

define <3 x double> @utofp_v3i8_v3f64(<3 x i8> %a) {
; CHECK-SD-LABEL: utofp_v3i8_v3f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov s0, w0
; CHECK-SD-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-SD-NEXT:    fmov s2, w2
; CHECK-SD-NEXT:    mov v0.s[1], w1
; CHECK-SD-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-SD-NEXT:    and v1.8b, v2.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v1.2d
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 killed $q1
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i8_v3f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    movi d1, #0xff00ff00ff00ff
; CHECK-GI-NEXT:    mov v0.h[1], w1
; CHECK-GI-NEXT:    mov v0.h[2], w2
; CHECK-GI-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-NEXT:    umov w8, v0.h[0]
; CHECK-GI-NEXT:    umov w9, v0.h[1]
; CHECK-GI-NEXT:    fmov d1, x8
; CHECK-GI-NEXT:    umov w8, v0.h[2]
; CHECK-GI-NEXT:    mov v1.d[1], x9
; CHECK-GI-NEXT:    umov w9, v0.h[3]
; CHECK-GI-NEXT:    fmov d2, x8
; CHECK-GI-NEXT:    ucvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    mov v2.d[1], x9
; CHECK-GI-NEXT:    mov d1, v0.d[1]
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NEXT:    // kill: def $d2 killed $d2 killed $q2
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i8> %a to <3 x double>
  ret <3 x double> %c
}

define <4 x double> @stofp_v4i8_v4f64(<4 x i8> %a) {
; CHECK-SD-LABEL: stofp_v4i8_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    shl v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v4i8_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll v1.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    shl v1.2d, v1.2d, #56
; CHECK-GI-NEXT:    shl v0.2d, v0.2d, #56
; CHECK-GI-NEXT:    sshr v1.2d, v1.2d, #56
; CHECK-GI-NEXT:    sshr v2.2d, v0.2d, #56
; CHECK-GI-NEXT:    scvtf v0.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <4 x i8> %a to <4 x double>
  ret <4 x double> %c
}

define <4 x double> @utofp_v4i8_v4f64(<4 x i8> %a) {
; CHECK-SD-LABEL: utofp_v4i8_v4f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-SD-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-SD-NEXT:    and v1.8b, v2.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v4i8_v4f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    movi v1.2d, #0x000000000000ff
; CHECK-GI-NEXT:    ushll v2.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    and v2.16b, v2.16b, v1.16b
; CHECK-GI-NEXT:    and v1.16b, v0.16b, v1.16b
; CHECK-GI-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <4 x i8> %a to <4 x double>
  ret <4 x double> %c
}

define <8 x double> @stofp_v8i8_v8f64(<8 x i8> %a) {
; CHECK-SD-LABEL: stofp_v8i8_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    mov b1, v0.b[0]
; CHECK-SD-NEXT:    mov b2, v0.b[2]
; CHECK-SD-NEXT:    mov b3, v0.b[4]
; CHECK-SD-NEXT:    mov b4, v0.b[6]
; CHECK-SD-NEXT:    mov v1.b[4], v0.b[1]
; CHECK-SD-NEXT:    mov v2.b[4], v0.b[3]
; CHECK-SD-NEXT:    mov v3.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v4.b[4], v0.b[7]
; CHECK-SD-NEXT:    shl v0.2s, v1.2s, #24
; CHECK-SD-NEXT:    shl v1.2s, v2.2s, #24
; CHECK-SD-NEXT:    shl v2.2s, v3.2s, #24
; CHECK-SD-NEXT:    shl v3.2s, v4.2s, #24
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    sshr v2.2s, v2.2s, #24
; CHECK-SD-NEXT:    sshr v3.2s, v3.2s, #24
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    sshll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i8_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v2.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    sshll v3.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v4.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v4.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i8> %a to <8 x double>
  ret <8 x double> %c
}

define <8 x double> @utofp_v8i8_v8f64(<8 x i8> %a) {
; CHECK-SD-LABEL: utofp_v8i8_v8f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    mov b2, v0.b[0]
; CHECK-SD-NEXT:    mov b3, v0.b[2]
; CHECK-SD-NEXT:    mov b4, v0.b[4]
; CHECK-SD-NEXT:    mov b5, v0.b[6]
; CHECK-SD-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-SD-NEXT:    mov v2.b[4], v0.b[1]
; CHECK-SD-NEXT:    mov v3.b[4], v0.b[3]
; CHECK-SD-NEXT:    mov v4.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v5.b[4], v0.b[7]
; CHECK-SD-NEXT:    and v0.8b, v2.8b, v1.8b
; CHECK-SD-NEXT:    and v2.8b, v3.8b, v1.8b
; CHECK-SD-NEXT:    and v3.8b, v4.8b, v1.8b
; CHECK-SD-NEXT:    and v1.8b, v5.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ushll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    ushll v4.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i8_v8f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v2.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ushll v3.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v4.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i8> %a to <8 x double>
  ret <8 x double> %c
}

define <16 x double> @stofp_v16i8_v16f64(<16 x i8> %a) {
; CHECK-SD-LABEL: stofp_v16i8_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    mov b2, v0.b[0]
; CHECK-SD-NEXT:    mov b3, v0.b[2]
; CHECK-SD-NEXT:    mov b4, v0.b[4]
; CHECK-SD-NEXT:    mov b5, v0.b[6]
; CHECK-SD-NEXT:    mov b6, v1.b[0]
; CHECK-SD-NEXT:    mov b7, v1.b[2]
; CHECK-SD-NEXT:    mov b16, v1.b[4]
; CHECK-SD-NEXT:    mov b17, v1.b[6]
; CHECK-SD-NEXT:    mov v2.b[4], v0.b[1]
; CHECK-SD-NEXT:    mov v3.b[4], v0.b[3]
; CHECK-SD-NEXT:    mov v4.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v5.b[4], v0.b[7]
; CHECK-SD-NEXT:    mov v6.b[4], v1.b[1]
; CHECK-SD-NEXT:    mov v7.b[4], v1.b[3]
; CHECK-SD-NEXT:    mov v16.b[4], v1.b[5]
; CHECK-SD-NEXT:    mov v17.b[4], v1.b[7]
; CHECK-SD-NEXT:    shl v0.2s, v2.2s, #24
; CHECK-SD-NEXT:    shl v1.2s, v3.2s, #24
; CHECK-SD-NEXT:    shl v2.2s, v4.2s, #24
; CHECK-SD-NEXT:    shl v3.2s, v5.2s, #24
; CHECK-SD-NEXT:    shl v4.2s, v6.2s, #24
; CHECK-SD-NEXT:    shl v5.2s, v7.2s, #24
; CHECK-SD-NEXT:    shl v6.2s, v16.2s, #24
; CHECK-SD-NEXT:    shl v7.2s, v17.2s, #24
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    sshr v2.2s, v2.2s, #24
; CHECK-SD-NEXT:    sshr v3.2s, v3.2s, #24
; CHECK-SD-NEXT:    sshr v4.2s, v4.2s, #24
; CHECK-SD-NEXT:    sshr v5.2s, v5.2s, #24
; CHECK-SD-NEXT:    sshr v6.2s, v6.2s, #24
; CHECK-SD-NEXT:    sshr v7.2s, v7.2s, #24
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    sshll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    sshll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    sshll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    sshll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v7.2d, v7.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i8_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    sshll v2.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v3.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v4.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    sshll v5.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v6.2d, v1.4s, #0
; CHECK-GI-NEXT:    sshll v7.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v16.2d, v3.4s, #0
; CHECK-GI-NEXT:    sshll v17.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v18.2d, v0.4s, #0
; CHECK-GI-NEXT:    scvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v18.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i8> %a to <16 x double>
  ret <16 x double> %c
}

define <16 x double> @utofp_v16i8_v16f64(<16 x i8> %a) {
; CHECK-SD-LABEL: utofp_v16i8_v16f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    mov b3, v0.b[0]
; CHECK-SD-NEXT:    mov b4, v0.b[2]
; CHECK-SD-NEXT:    mov b5, v0.b[4]
; CHECK-SD-NEXT:    mov b6, v0.b[6]
; CHECK-SD-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-SD-NEXT:    mov b7, v2.b[0]
; CHECK-SD-NEXT:    mov b16, v2.b[2]
; CHECK-SD-NEXT:    mov b17, v2.b[4]
; CHECK-SD-NEXT:    mov b18, v2.b[6]
; CHECK-SD-NEXT:    mov v3.b[4], v0.b[1]
; CHECK-SD-NEXT:    mov v4.b[4], v0.b[3]
; CHECK-SD-NEXT:    mov v5.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v6.b[4], v0.b[7]
; CHECK-SD-NEXT:    mov v7.b[4], v2.b[1]
; CHECK-SD-NEXT:    mov v16.b[4], v2.b[3]
; CHECK-SD-NEXT:    mov v17.b[4], v2.b[5]
; CHECK-SD-NEXT:    mov v18.b[4], v2.b[7]
; CHECK-SD-NEXT:    and v0.8b, v3.8b, v1.8b
; CHECK-SD-NEXT:    and v2.8b, v4.8b, v1.8b
; CHECK-SD-NEXT:    and v3.8b, v5.8b, v1.8b
; CHECK-SD-NEXT:    and v4.8b, v6.8b, v1.8b
; CHECK-SD-NEXT:    and v5.8b, v7.8b, v1.8b
; CHECK-SD-NEXT:    and v6.8b, v16.8b, v1.8b
; CHECK-SD-NEXT:    and v7.8b, v17.8b, v1.8b
; CHECK-SD-NEXT:    and v1.8b, v18.8b, v1.8b
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ushll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    ushll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    ushll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    ushll v16.2d, v1.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v7.2d
; CHECK-SD-NEXT:    ucvtf v7.2d, v16.2d
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i8_v16f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    ushll v2.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v3.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v4.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    ushll v5.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v6.2d, v1.4s, #0
; CHECK-GI-NEXT:    ushll v7.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v16.2d, v3.4s, #0
; CHECK-GI-NEXT:    ushll v17.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v18.2d, v0.4s, #0
; CHECK-GI-NEXT:    ucvtf v0.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v16.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v18.2d
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i8> %a to <16 x double>
  ret <16 x double> %c
}

define <32 x double> @stofp_v32i8_v32f64(<32 x i8> %a) {
; CHECK-SD-LABEL: stofp_v32i8_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    mov b5, v1.b[6]
; CHECK-SD-NEXT:    mov b17, v1.b[4]
; CHECK-SD-NEXT:    mov b20, v1.b[2]
; CHECK-SD-NEXT:    mov b21, v1.b[0]
; CHECK-SD-NEXT:    mov b18, v0.b[0]
; CHECK-SD-NEXT:    mov b19, v0.b[6]
; CHECK-SD-NEXT:    mov b22, v0.b[4]
; CHECK-SD-NEXT:    ext v16.16b, v1.16b, v1.16b, #8
; CHECK-SD-NEXT:    mov b2, v3.b[0]
; CHECK-SD-NEXT:    mov b4, v3.b[2]
; CHECK-SD-NEXT:    mov b6, v3.b[4]
; CHECK-SD-NEXT:    mov b7, v3.b[6]
; CHECK-SD-NEXT:    mov v5.b[4], v1.b[7]
; CHECK-SD-NEXT:    mov v17.b[4], v1.b[5]
; CHECK-SD-NEXT:    mov v20.b[4], v1.b[3]
; CHECK-SD-NEXT:    mov v21.b[4], v1.b[1]
; CHECK-SD-NEXT:    mov v19.b[4], v0.b[7]
; CHECK-SD-NEXT:    mov v22.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v18.b[4], v0.b[1]
; CHECK-SD-NEXT:    mov b23, v16.b[0]
; CHECK-SD-NEXT:    mov v2.b[4], v3.b[1]
; CHECK-SD-NEXT:    mov v4.b[4], v3.b[3]
; CHECK-SD-NEXT:    mov v6.b[4], v3.b[5]
; CHECK-SD-NEXT:    mov v7.b[4], v3.b[7]
; CHECK-SD-NEXT:    mov b3, v0.b[2]
; CHECK-SD-NEXT:    shl v5.2s, v5.2s, #24
; CHECK-SD-NEXT:    shl v17.2s, v17.2s, #24
; CHECK-SD-NEXT:    shl v20.2s, v20.2s, #24
; CHECK-SD-NEXT:    mov b24, v16.b[4]
; CHECK-SD-NEXT:    mov v23.b[4], v16.b[1]
; CHECK-SD-NEXT:    shl v18.2s, v18.2s, #24
; CHECK-SD-NEXT:    shl v19.2s, v19.2s, #24
; CHECK-SD-NEXT:    sshr v5.2s, v5.2s, #24
; CHECK-SD-NEXT:    shl v1.2s, v2.2s, #24
; CHECK-SD-NEXT:    shl v2.2s, v4.2s, #24
; CHECK-SD-NEXT:    sshr v17.2s, v17.2s, #24
; CHECK-SD-NEXT:    mov v3.b[4], v0.b[3]
; CHECK-SD-NEXT:    shl v0.2s, v21.2s, #24
; CHECK-SD-NEXT:    shl v4.2s, v6.2s, #24
; CHECK-SD-NEXT:    shl v6.2s, v7.2s, #24
; CHECK-SD-NEXT:    mov b7, v16.b[2]
; CHECK-SD-NEXT:    sshll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    sshr v20.2s, v20.2s, #24
; CHECK-SD-NEXT:    mov b21, v16.b[6]
; CHECK-SD-NEXT:    sshll v17.2d, v17.2s, #0
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-SD-NEXT:    shl v22.2s, v22.2s, #24
; CHECK-SD-NEXT:    shl v3.2s, v3.2s, #24
; CHECK-SD-NEXT:    mov v24.b[4], v16.b[5]
; CHECK-SD-NEXT:    sshr v19.2s, v19.2s, #24
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    mov v7.b[4], v16.b[3]
; CHECK-SD-NEXT:    sshll v20.2d, v20.2s, #0
; CHECK-SD-NEXT:    scvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    mov v21.b[4], v16.b[7]
; CHECK-SD-NEXT:    sshr v3.2s, v3.2s, #24
; CHECK-SD-NEXT:    sshr v16.2s, v22.2s, #24
; CHECK-SD-NEXT:    sshll v19.2d, v19.2s, #0
; CHECK-SD-NEXT:    scvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    shl v22.2s, v24.2s, #24
; CHECK-SD-NEXT:    sshr v6.2s, v6.2s, #24
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    shl v7.2s, v7.2s, #24
; CHECK-SD-NEXT:    sshr v4.2s, v4.2s, #24
; CHECK-SD-NEXT:    stp q17, q5, [x8, #160]
; CHECK-SD-NEXT:    sshr v5.2s, v18.2s, #24
; CHECK-SD-NEXT:    shl v17.2s, v23.2s, #24
; CHECK-SD-NEXT:    sshll v3.2d, v3.2s, #0
; CHECK-SD-NEXT:    sshll v16.2d, v16.2s, #0
; CHECK-SD-NEXT:    shl v21.2s, v21.2s, #24
; CHECK-SD-NEXT:    scvtf v19.2d, v19.2d
; CHECK-SD-NEXT:    sshr v2.2s, v2.2s, #24
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #24
; CHECK-SD-NEXT:    stp q0, q20, [x8, #128]
; CHECK-SD-NEXT:    sshll v5.2d, v5.2s, #0
; CHECK-SD-NEXT:    sshr v0.2s, v7.2s, #24
; CHECK-SD-NEXT:    sshr v7.2s, v17.2s, #24
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    sshr v18.2s, v21.2s, #24
; CHECK-SD-NEXT:    sshr v20.2s, v22.2s, #24
; CHECK-SD-NEXT:    sshll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    sshll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    sshll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    sshll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    sshll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    stp q16, q19, [x8, #32]
; CHECK-SD-NEXT:    sshll v17.2d, v18.2s, #0
; CHECK-SD-NEXT:    sshll v16.2d, v20.2s, #0
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    stp q5, q3, [x8]
; CHECK-SD-NEXT:    scvtf v3.2d, v7.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    stp q4, q5, [x8, #96]
; CHECK-SD-NEXT:    stp q3, q0, [x8, #192]
; CHECK-SD-NEXT:    scvtf v0.2d, v1.2d
; CHECK-SD-NEXT:    stp q16, q17, [x8, #224]
; CHECK-SD-NEXT:    stp q0, q2, [x8, #64]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i8_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    sshll v4.8h, v1.8b, #0
; CHECK-GI-NEXT:    sshll2 v1.8h, v1.16b, #0
; CHECK-GI-NEXT:    sshll v3.4s, v2.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    sshll v6.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v16.4s, v4.4h, #0
; CHECK-GI-NEXT:    sshll2 v4.4s, v4.8h, #0
; CHECK-GI-NEXT:    sshll v18.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v5.2d, v3.2s, #0
; CHECK-GI-NEXT:    sshll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    sshll v7.2d, v2.2s, #0
; CHECK-GI-NEXT:    sshll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    sshll v17.2d, v6.2s, #0
; CHECK-GI-NEXT:    sshll2 v6.2d, v6.4s, #0
; CHECK-GI-NEXT:    sshll v19.2d, v0.2s, #0
; CHECK-GI-NEXT:    sshll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    sshll v20.2d, v16.2s, #0
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    stp q5, q3, [x8]
; CHECK-GI-NEXT:    sshll2 v5.2d, v16.4s, #0
; CHECK-GI-NEXT:    scvtf v16.2d, v19.2d
; CHECK-GI-NEXT:    stp q7, q2, [x8, #32]
; CHECK-GI-NEXT:    scvtf v2.2d, v20.2d
; CHECK-GI-NEXT:    sshll v3.2d, v4.2s, #0
; CHECK-GI-NEXT:    stp q17, q6, [x8, #64]
; CHECK-GI-NEXT:    sshll2 v4.2d, v4.4s, #0
; CHECK-GI-NEXT:    sshll v6.2d, v18.2s, #0
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    sshll2 v7.2d, v18.4s, #0
; CHECK-GI-NEXT:    stp q16, q0, [x8, #96]
; CHECK-GI-NEXT:    sshll v0.2d, v1.2s, #0
; CHECK-GI-NEXT:    sshll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    stp q2, q5, [x8, #128]
; CHECK-GI-NEXT:    scvtf v2.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q3, q4, [x8, #160]
; CHECK-GI-NEXT:    stp q6, q2, [x8, #192]
; CHECK-GI-NEXT:    stp q0, q1, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i8> %a to <32 x double>
  ret <32 x double> %c
}

define <32 x double> @utofp_v32i8_v32f64(<32 x i8> %a) {
; CHECK-SD-LABEL: utofp_v32i8_v32f64:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov b6, v1.b[6]
; CHECK-SD-NEXT:    mov b7, v1.b[4]
; CHECK-SD-NEXT:    ext v3.16b, v1.16b, v1.16b, #8
; CHECK-SD-NEXT:    mov b16, v1.b[2]
; CHECK-SD-NEXT:    mov b17, v1.b[0]
; CHECK-SD-NEXT:    mov b19, v0.b[6]
; CHECK-SD-NEXT:    mov b20, v0.b[4]
; CHECK-SD-NEXT:    movi d5, #0x0000ff000000ff
; CHECK-SD-NEXT:    mov b24, v0.b[2]
; CHECK-SD-NEXT:    mov b25, v0.b[0]
; CHECK-SD-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    mov v6.b[4], v1.b[7]
; CHECK-SD-NEXT:    mov v7.b[4], v1.b[5]
; CHECK-SD-NEXT:    mov b18, v3.b[0]
; CHECK-SD-NEXT:    mov b21, v3.b[2]
; CHECK-SD-NEXT:    mov b23, v3.b[4]
; CHECK-SD-NEXT:    mov v16.b[4], v1.b[3]
; CHECK-SD-NEXT:    mov v17.b[4], v1.b[1]
; CHECK-SD-NEXT:    mov b1, v3.b[6]
; CHECK-SD-NEXT:    mov v19.b[4], v0.b[7]
; CHECK-SD-NEXT:    mov v20.b[4], v0.b[5]
; CHECK-SD-NEXT:    mov v24.b[4], v0.b[3]
; CHECK-SD-NEXT:    mov v25.b[4], v0.b[1]
; CHECK-SD-NEXT:    and v6.8b, v6.8b, v5.8b
; CHECK-SD-NEXT:    and v7.8b, v7.8b, v5.8b
; CHECK-SD-NEXT:    mov v18.b[4], v3.b[1]
; CHECK-SD-NEXT:    mov v21.b[4], v3.b[3]
; CHECK-SD-NEXT:    mov v23.b[4], v3.b[5]
; CHECK-SD-NEXT:    and v16.8b, v16.8b, v5.8b
; CHECK-SD-NEXT:    and v17.8b, v17.8b, v5.8b
; CHECK-SD-NEXT:    mov v1.b[4], v3.b[7]
; CHECK-SD-NEXT:    and v3.8b, v19.8b, v5.8b
; CHECK-SD-NEXT:    ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v7.2s, #0
; CHECK-SD-NEXT:    and v20.8b, v20.8b, v5.8b
; CHECK-SD-NEXT:    ushll v16.2d, v16.2s, #0
; CHECK-SD-NEXT:    mov b4, v2.b[0]
; CHECK-SD-NEXT:    mov b22, v2.b[2]
; CHECK-SD-NEXT:    ushll v17.2d, v17.2s, #0
; CHECK-SD-NEXT:    ushll v0.2d, v3.2s, #0
; CHECK-SD-NEXT:    mov b19, v2.b[4]
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v7.2d
; CHECK-SD-NEXT:    ushll v20.2d, v20.2s, #0
; CHECK-SD-NEXT:    mov b7, v2.b[6]
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    and v24.8b, v24.8b, v5.8b
; CHECK-SD-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    mov v4.b[4], v2.b[1]
; CHECK-SD-NEXT:    ucvtf v20.2d, v20.2d
; CHECK-SD-NEXT:    mov v22.b[4], v2.b[3]
; CHECK-SD-NEXT:    mov v19.b[4], v2.b[5]
; CHECK-SD-NEXT:    stp q3, q6, [x8, #160]
; CHECK-SD-NEXT:    and v6.8b, v25.8b, v5.8b
; CHECK-SD-NEXT:    and v1.8b, v1.8b, v5.8b
; CHECK-SD-NEXT:    mov v7.b[4], v2.b[7]
; CHECK-SD-NEXT:    and v2.8b, v23.8b, v5.8b
; CHECK-SD-NEXT:    ushll v3.2d, v24.2s, #0
; CHECK-SD-NEXT:    stp q17, q16, [x8, #128]
; CHECK-SD-NEXT:    and v16.8b, v21.8b, v5.8b
; CHECK-SD-NEXT:    and v4.8b, v4.8b, v5.8b
; CHECK-SD-NEXT:    stp q20, q0, [x8, #32]
; CHECK-SD-NEXT:    ushll v6.2d, v6.2s, #0
; CHECK-SD-NEXT:    and v0.8b, v18.8b, v5.8b
; CHECK-SD-NEXT:    ushll v1.2d, v1.2s, #0
; CHECK-SD-NEXT:    ushll v2.2d, v2.2s, #0
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    ushll v16.2d, v16.2s, #0
; CHECK-SD-NEXT:    and v7.8b, v7.8b, v5.8b
; CHECK-SD-NEXT:    and v17.8b, v19.8b, v5.8b
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ushll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    and v18.8b, v22.8b, v5.8b
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    ushll v4.2d, v4.2s, #0
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    ushll v5.2d, v7.2s, #0
; CHECK-SD-NEXT:    ushll v7.2d, v17.2s, #0
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    stp q6, q3, [x8]
; CHECK-SD-NEXT:    ushll v3.2d, v18.2s, #0
; CHECK-SD-NEXT:    stp q2, q1, [x8, #224]
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v7.2d
; CHECK-SD-NEXT:    stp q0, q16, [x8, #192]
; CHECK-SD-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v4.2d
; CHECK-SD-NEXT:    stp q1, q5, [x8, #96]
; CHECK-SD-NEXT:    stp q0, q2, [x8, #64]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i8_v32f64:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    ushll v4.8h, v1.8b, #0
; CHECK-GI-NEXT:    ushll2 v1.8h, v1.16b, #0
; CHECK-GI-NEXT:    ushll v3.4s, v2.4h, #0
; CHECK-GI-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    ushll v6.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v16.4s, v4.4h, #0
; CHECK-GI-NEXT:    ushll2 v4.4s, v4.8h, #0
; CHECK-GI-NEXT:    ushll v18.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v5.2d, v3.2s, #0
; CHECK-GI-NEXT:    ushll2 v3.2d, v3.4s, #0
; CHECK-GI-NEXT:    ushll v7.2d, v2.2s, #0
; CHECK-GI-NEXT:    ushll2 v2.2d, v2.4s, #0
; CHECK-GI-NEXT:    ushll v17.2d, v6.2s, #0
; CHECK-GI-NEXT:    ushll2 v6.2d, v6.4s, #0
; CHECK-GI-NEXT:    ushll v19.2d, v0.2s, #0
; CHECK-GI-NEXT:    ushll2 v0.2d, v0.4s, #0
; CHECK-GI-NEXT:    ushll v20.2d, v16.2s, #0
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    stp q5, q3, [x8]
; CHECK-GI-NEXT:    ushll2 v5.2d, v16.4s, #0
; CHECK-GI-NEXT:    ucvtf v16.2d, v19.2d
; CHECK-GI-NEXT:    stp q7, q2, [x8, #32]
; CHECK-GI-NEXT:    ucvtf v2.2d, v20.2d
; CHECK-GI-NEXT:    ushll v3.2d, v4.2s, #0
; CHECK-GI-NEXT:    stp q17, q6, [x8, #64]
; CHECK-GI-NEXT:    ushll2 v4.2d, v4.4s, #0
; CHECK-GI-NEXT:    ushll v6.2d, v18.2s, #0
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    ushll2 v7.2d, v18.4s, #0
; CHECK-GI-NEXT:    stp q16, q0, [x8, #96]
; CHECK-GI-NEXT:    ushll v0.2d, v1.2s, #0
; CHECK-GI-NEXT:    ushll2 v1.2d, v1.4s, #0
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    stp q2, q5, [x8, #128]
; CHECK-GI-NEXT:    ucvtf v2.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    stp q3, q4, [x8, #160]
; CHECK-GI-NEXT:    stp q6, q2, [x8, #192]
; CHECK-GI-NEXT:    stp q0, q1, [x8, #224]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i8> %a to <32 x double>
  ret <32 x double> %c
}

define <2 x float> @stofp_v2i128_v2f32(<2 x i128> %a) {
; CHECK-SD-LABEL: stofp_v2i128_v2f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x1
; CHECK-SD-NEXT:    mov x20, x0
; CHECK-SD-NEXT:    mov x0, x2
; CHECK-SD-NEXT:    mov x1, x3
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i128_v2f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floattisf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattisf
; CHECK-GI-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[0]
; CHECK-GI-NEXT:    fmov d0, d1
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i128> %a to <2 x float>
  ret <2 x float> %c
}

define <2 x float> @utofp_v2i128_v2f32(<2 x i128> %a) {
; CHECK-SD-LABEL: utofp_v2i128_v2f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #48
; CHECK-SD-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NEXT:    mov x19, x1
; CHECK-SD-NEXT:    mov x20, x0
; CHECK-SD-NEXT:    mov x0, x2
; CHECK-SD-NEXT:    mov x1, x3
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NEXT:    add sp, sp, #48
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i128_v2f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #48
; CHECK-GI-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    bl __floatuntisf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntisf
; CHECK-GI-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[0]
; CHECK-GI-NEXT:    fmov d0, d1
; CHECK-GI-NEXT:    add sp, sp, #48
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i128> %a to <2 x float>
  ret <2 x float> %c
}

define <3 x float> @stofp_v3i128_v3f32(<3 x i128> %a) {
; CHECK-SD-LABEL: stofp_v3i128_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #80
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    mov x21, x1
; CHECK-SD-NEXT:    mov x22, x0
; CHECK-SD-NEXT:    mov x0, x4
; CHECK-SD-NEXT:    mov x1, x5
; CHECK-SD-NEXT:    mov x19, x3
; CHECK-SD-NEXT:    mov x20, x2
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floattisf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.s[1], v0.s[0]
; CHECK-SD-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.s[2], v0.s[0]
; CHECK-SD-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-NEXT:    add sp, sp, #80
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i128_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #80
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floattisf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattisf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floattisf
; CHECK-GI-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.s[1], v2.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v0.s[0]
; CHECK-GI-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NEXT:    add sp, sp, #80
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i128> %a to <3 x float>
  ret <3 x float> %c
}

define <3 x float> @utofp_v3i128_v3f32(<3 x i128> %a) {
; CHECK-SD-LABEL: utofp_v3i128_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sub sp, sp, #80
; CHECK-SD-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NEXT:    mov x21, x1
; CHECK-SD-NEXT:    mov x22, x0
; CHECK-SD-NEXT:    mov x0, x4
; CHECK-SD-NEXT:    mov x1, x5
; CHECK-SD-NEXT:    mov x19, x3
; CHECK-SD-NEXT:    mov x20, x2
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    mov x0, x22
; CHECK-SD-NEXT:    mov x1, x21
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    mov x0, x20
; CHECK-SD-NEXT:    mov x1, x19
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NEXT:    bl __floatuntisf
; CHECK-SD-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-SD-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.s[1], v0.s[0]
; CHECK-SD-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NEXT:    mov v1.s[2], v0.s[0]
; CHECK-SD-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-NEXT:    add sp, sp, #80
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i128_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sub sp, sp, #80
; CHECK-GI-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NEXT:    mov x19, x2
; CHECK-GI-NEXT:    mov x20, x3
; CHECK-GI-NEXT:    mov x21, x4
; CHECK-GI-NEXT:    mov x22, x5
; CHECK-GI-NEXT:    bl __floatuntisf
; CHECK-GI-NEXT:    mov x0, x19
; CHECK-GI-NEXT:    mov x1, x20
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntisf
; CHECK-GI-NEXT:    mov x0, x21
; CHECK-GI-NEXT:    mov x1, x22
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NEXT:    bl __floatuntisf
; CHECK-GI-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-NEXT:    // kill: def $s0 killed $s0 def $q0
; CHECK-GI-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v1.s[1], v2.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v0.s[0]
; CHECK-GI-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NEXT:    add sp, sp, #80
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i128> %a to <3 x float>
  ret <3 x float> %c
}

define <2 x float> @stofp_v2i64_v2f32(<2 x i64> %a) {
; CHECK-LABEL: stofp_v2i64_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    mov x8, v0.d[1]
; CHECK-NEXT:    fmov x9, d0
; CHECK-NEXT:    scvtf s0, x9
; CHECK-NEXT:    scvtf s1, x8
; CHECK-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i64> %a to <2 x float>
  ret <2 x float> %c
}

define <2 x float> @utofp_v2i64_v2f32(<2 x i64> %a) {
; CHECK-LABEL: utofp_v2i64_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    mov x8, v0.d[1]
; CHECK-NEXT:    fmov x9, d0
; CHECK-NEXT:    ucvtf s0, x9
; CHECK-NEXT:    ucvtf s1, x8
; CHECK-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i64> %a to <2 x float>
  ret <2 x float> %c
}

define <3 x float> @stofp_v3i64_v3f32(<3 x i64> %a) {
; CHECK-SD-LABEL: stofp_v3i64_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov x8, d0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
; CHECK-SD-NEXT:    scvtf s3, x8
; CHECK-SD-NEXT:    fmov x8, d1
; CHECK-SD-NEXT:    scvtf s1, x8
; CHECK-SD-NEXT:    fmov x8, d2
; CHECK-SD-NEXT:    mov v0.s[0], v3.s[0]
; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-SD-NEXT:    scvtf s1, x8
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i64_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov x8, d0
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    scvtf s0, x8
; CHECK-GI-NEXT:    scvtf s1, x9
; CHECK-GI-NEXT:    fmov x8, d2
; CHECK-GI-NEXT:    scvtf s2, x8
; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i64> %a to <3 x float>
  ret <3 x float> %c
}

define <3 x float> @utofp_v3i64_v3f32(<3 x i64> %a) {
; CHECK-SD-LABEL: utofp_v3i64_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    fmov x8, d0
; CHECK-SD-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-SD-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-SD-NEXT:    movi v0.2d, #0000000000000000
; CHECK-SD-NEXT:    ucvtf s3, x8
; CHECK-SD-NEXT:    fmov x8, d1
; CHECK-SD-NEXT:    ucvtf s1, x8
; CHECK-SD-NEXT:    fmov x8, d2
; CHECK-SD-NEXT:    mov v0.s[0], v3.s[0]
; CHECK-SD-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-SD-NEXT:    ucvtf s1, x8
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i64_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov x8, d0
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    ucvtf s0, x8
; CHECK-GI-NEXT:    ucvtf s1, x9
; CHECK-GI-NEXT:    fmov x8, d2
; CHECK-GI-NEXT:    ucvtf s2, x8
; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
; CHECK-GI-NEXT:    mov v0.s[2], v2.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i64> %a to <3 x float>
  ret <3 x float> %c
}

define <4 x float> @stofp_v4i64_v4f32(<4 x i64> %a) {
; CHECK-SD-LABEL: stofp_v4i64_v4f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    fmov x9, d0
; CHECK-SD-NEXT:    scvtf s0, x9
; CHECK-SD-NEXT:    mov x9, v1.d[1]
; CHECK-SD-NEXT:    scvtf s2, x8
; CHECK-SD-NEXT:    fmov x8, d1
; CHECK-SD-NEXT:    scvtf s1, x8
; CHECK-SD-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    scvtf s1, x9
; CHECK-SD-NEXT:    mov v0.s[3], v1.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v4i64_v4f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    fmov x9, d0
; CHECK-GI-NEXT:    scvtf s0, x9
; CHECK-GI-NEXT:    mov x9, v1.d[1]
; CHECK-GI-NEXT:    scvtf s2, x8
; CHECK-GI-NEXT:    fmov x8, d1
; CHECK-GI-NEXT:    scvtf s1, x8
; CHECK-GI-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-GI-NEXT:    scvtf s2, x9
; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v2.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <4 x i64> %a to <4 x float>
  ret <4 x float> %c
}

define <4 x float> @utofp_v4i64_v4f32(<4 x i64> %a) {
; CHECK-SD-LABEL: utofp_v4i64_v4f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    fmov x9, d0
; CHECK-SD-NEXT:    ucvtf s0, x9
; CHECK-SD-NEXT:    mov x9, v1.d[1]
; CHECK-SD-NEXT:    ucvtf s2, x8
; CHECK-SD-NEXT:    fmov x8, d1
; CHECK-SD-NEXT:    ucvtf s1, x8
; CHECK-SD-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    ucvtf s1, x9
; CHECK-SD-NEXT:    mov v0.s[3], v1.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v4i64_v4f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    fmov x9, d0
; CHECK-GI-NEXT:    ucvtf s0, x9
; CHECK-GI-NEXT:    mov x9, v1.d[1]
; CHECK-GI-NEXT:    ucvtf s2, x8
; CHECK-GI-NEXT:    fmov x8, d1
; CHECK-GI-NEXT:    ucvtf s1, x8
; CHECK-GI-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-GI-NEXT:    ucvtf s2, x9
; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v2.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <4 x i64> %a to <4 x float>
  ret <4 x float> %c
}

define <8 x float> @stofp_v8i64_v8f32(<8 x i64> %a) {
; CHECK-SD-LABEL: stofp_v8i64_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    mov x9, v2.d[1]
; CHECK-SD-NEXT:    fmov x10, d0
; CHECK-SD-NEXT:    fmov x11, d2
; CHECK-SD-NEXT:    scvtf s0, x10
; CHECK-SD-NEXT:    mov x10, v3.d[1]
; CHECK-SD-NEXT:    scvtf s4, x8
; CHECK-SD-NEXT:    scvtf s5, x9
; CHECK-SD-NEXT:    scvtf s2, x11
; CHECK-SD-NEXT:    fmov x9, d1
; CHECK-SD-NEXT:    fmov x11, d3
; CHECK-SD-NEXT:    mov x8, v1.d[1]
; CHECK-SD-NEXT:    scvtf s1, x9
; CHECK-SD-NEXT:    mov v0.s[1], v4.s[0]
; CHECK-SD-NEXT:    scvtf s3, x11
; CHECK-SD-NEXT:    mov v2.s[1], v5.s[0]
; CHECK-SD-NEXT:    scvtf s4, x8
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    scvtf s1, x10
; CHECK-SD-NEXT:    mov v2.s[2], v3.s[0]
; CHECK-SD-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-SD-NEXT:    mov v2.s[3], v1.s[0]
; CHECK-SD-NEXT:    mov v1.16b, v2.16b
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i64_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov x10, d0
; CHECK-GI-NEXT:    mov x9, v2.d[1]
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    scvtf s0, x10
; CHECK-GI-NEXT:    fmov x10, d2
; CHECK-GI-NEXT:    scvtf s5, x9
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    scvtf s4, x8
; CHECK-GI-NEXT:    mov x8, v1.d[1]
; CHECK-GI-NEXT:    scvtf s2, x10
; CHECK-GI-NEXT:    fmov x10, d3
; CHECK-GI-NEXT:    scvtf s1, x9
; CHECK-GI-NEXT:    mov x9, v3.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v4.s[0]
; CHECK-GI-NEXT:    scvtf s3, x10
; CHECK-GI-NEXT:    scvtf s4, x8
; CHECK-GI-NEXT:    mov v2.s[1], v5.s[0]
; CHECK-GI-NEXT:    scvtf s5, x9
; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v3.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v5.s[0]
; CHECK-GI-NEXT:    mov v1.16b, v2.16b
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i64> %a to <8 x float>
  ret <8 x float> %c
}

define <8 x float> @utofp_v8i64_v8f32(<8 x i64> %a) {
; CHECK-SD-LABEL: utofp_v8i64_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    mov x9, v2.d[1]
; CHECK-SD-NEXT:    fmov x10, d0
; CHECK-SD-NEXT:    fmov x11, d2
; CHECK-SD-NEXT:    ucvtf s0, x10
; CHECK-SD-NEXT:    mov x10, v3.d[1]
; CHECK-SD-NEXT:    ucvtf s4, x8
; CHECK-SD-NEXT:    ucvtf s5, x9
; CHECK-SD-NEXT:    ucvtf s2, x11
; CHECK-SD-NEXT:    fmov x9, d1
; CHECK-SD-NEXT:    fmov x11, d3
; CHECK-SD-NEXT:    mov x8, v1.d[1]
; CHECK-SD-NEXT:    ucvtf s1, x9
; CHECK-SD-NEXT:    mov v0.s[1], v4.s[0]
; CHECK-SD-NEXT:    ucvtf s3, x11
; CHECK-SD-NEXT:    mov v2.s[1], v5.s[0]
; CHECK-SD-NEXT:    ucvtf s4, x8
; CHECK-SD-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-SD-NEXT:    ucvtf s1, x10
; CHECK-SD-NEXT:    mov v2.s[2], v3.s[0]
; CHECK-SD-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-SD-NEXT:    mov v2.s[3], v1.s[0]
; CHECK-SD-NEXT:    mov v1.16b, v2.16b
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i64_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov x10, d0
; CHECK-GI-NEXT:    mov x9, v2.d[1]
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    ucvtf s0, x10
; CHECK-GI-NEXT:    fmov x10, d2
; CHECK-GI-NEXT:    ucvtf s5, x9
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    ucvtf s4, x8
; CHECK-GI-NEXT:    mov x8, v1.d[1]
; CHECK-GI-NEXT:    ucvtf s2, x10
; CHECK-GI-NEXT:    fmov x10, d3
; CHECK-GI-NEXT:    ucvtf s1, x9
; CHECK-GI-NEXT:    mov x9, v3.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v4.s[0]
; CHECK-GI-NEXT:    ucvtf s3, x10
; CHECK-GI-NEXT:    ucvtf s4, x8
; CHECK-GI-NEXT:    mov v2.s[1], v5.s[0]
; CHECK-GI-NEXT:    ucvtf s5, x9
; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v3.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v5.s[0]
; CHECK-GI-NEXT:    mov v1.16b, v2.16b
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i64> %a to <8 x float>
  ret <8 x float> %c
}

define <16 x float> @stofp_v16i64_v16f32(<16 x i64> %a) {
; CHECK-SD-LABEL: stofp_v16i64_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov x13, d2
; CHECK-SD-NEXT:    mov x9, v0.d[1]
; CHECK-SD-NEXT:    mov x10, v2.d[1]
; CHECK-SD-NEXT:    fmov x11, d0
; CHECK-SD-NEXT:    mov x12, v4.d[1]
; CHECK-SD-NEXT:    mov x8, v1.d[1]
; CHECK-SD-NEXT:    scvtf s16, x13
; CHECK-SD-NEXT:    fmov x13, d4
; CHECK-SD-NEXT:    scvtf s0, x11
; CHECK-SD-NEXT:    mov x11, v6.d[1]
; CHECK-SD-NEXT:    scvtf s17, x9
; CHECK-SD-NEXT:    scvtf s18, x10
; CHECK-SD-NEXT:    fmov x9, d1
; CHECK-SD-NEXT:    scvtf s1, x12
; CHECK-SD-NEXT:    fmov x12, d6
; CHECK-SD-NEXT:    scvtf s2, x13
; CHECK-SD-NEXT:    fmov x13, d3
; CHECK-SD-NEXT:    mov x10, v3.d[1]
; CHECK-SD-NEXT:    scvtf s4, x11
; CHECK-SD-NEXT:    mov v0.s[1], v17.s[0]
; CHECK-SD-NEXT:    scvtf s6, x9
; CHECK-SD-NEXT:    scvtf s3, x12
; CHECK-SD-NEXT:    mov v16.s[1], v18.s[0]
; CHECK-SD-NEXT:    mov x9, v5.d[1]
; CHECK-SD-NEXT:    fmov x11, d5
; CHECK-SD-NEXT:    scvtf s5, x13
; CHECK-SD-NEXT:    fmov x13, d7
; CHECK-SD-NEXT:    mov x12, v7.d[1]
; CHECK-SD-NEXT:    mov v2.s[1], v1.s[0]
; CHECK-SD-NEXT:    mov v0.s[2], v6.s[0]
; CHECK-SD-NEXT:    scvtf s6, x10
; CHECK-SD-NEXT:    scvtf s7, x11
; CHECK-SD-NEXT:    scvtf s1, x13
; CHECK-SD-NEXT:    mov v3.s[1], v4.s[0]
; CHECK-SD-NEXT:    mov v16.s[2], v5.s[0]
; CHECK-SD-NEXT:    scvtf s4, x8
; CHECK-SD-NEXT:    scvtf s5, x9
; CHECK-SD-NEXT:    mov v2.s[2], v7.s[0]
; CHECK-SD-NEXT:    mov v3.s[2], v1.s[0]
; CHECK-SD-NEXT:    scvtf s1, x12
; CHECK-SD-NEXT:    mov v16.s[3], v6.s[0]
; CHECK-SD-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-SD-NEXT:    mov v2.s[3], v5.s[0]
; CHECK-SD-NEXT:    mov v3.s[3], v1.s[0]
; CHECK-SD-NEXT:    mov v1.16b, v16.16b
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i64_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    mov x9, v0.d[1]
; CHECK-GI-NEXT:    fmov x13, d2
; CHECK-GI-NEXT:    fmov x11, d0
; CHECK-GI-NEXT:    mov x12, v4.d[1]
; CHECK-GI-NEXT:    mov x8, v1.d[1]
; CHECK-GI-NEXT:    mov x10, v2.d[1]
; CHECK-GI-NEXT:    scvtf s0, x11
; CHECK-GI-NEXT:    mov x11, v6.d[1]
; CHECK-GI-NEXT:    scvtf s16, x9
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    scvtf s1, x13
; CHECK-GI-NEXT:    fmov x13, d4
; CHECK-GI-NEXT:    scvtf s4, x12
; CHECK-GI-NEXT:    fmov x12, d6
; CHECK-GI-NEXT:    scvtf s17, x10
; CHECK-GI-NEXT:    mov x10, v3.d[1]
; CHECK-GI-NEXT:    scvtf s6, x11
; CHECK-GI-NEXT:    fmov x11, d5
; CHECK-GI-NEXT:    scvtf s18, x9
; CHECK-GI-NEXT:    scvtf s2, x13
; CHECK-GI-NEXT:    fmov x13, d3
; CHECK-GI-NEXT:    scvtf s3, x12
; CHECK-GI-NEXT:    mov x9, v5.d[1]
; CHECK-GI-NEXT:    mov x12, v7.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v16.s[0]
; CHECK-GI-NEXT:    scvtf s5, x11
; CHECK-GI-NEXT:    mov v1.s[1], v17.s[0]
; CHECK-GI-NEXT:    scvtf s19, x13
; CHECK-GI-NEXT:    fmov x13, d7
; CHECK-GI-NEXT:    mov v2.s[1], v4.s[0]
; CHECK-GI-NEXT:    mov v3.s[1], v6.s[0]
; CHECK-GI-NEXT:    scvtf s4, x8
; CHECK-GI-NEXT:    scvtf s6, x10
; CHECK-GI-NEXT:    scvtf s16, x9
; CHECK-GI-NEXT:    scvtf s17, x12
; CHECK-GI-NEXT:    scvtf s7, x13
; CHECK-GI-NEXT:    mov v0.s[2], v18.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v19.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v5.s[0]
; CHECK-GI-NEXT:    mov v3.s[2], v7.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-GI-NEXT:    mov v1.s[3], v6.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v16.s[0]
; CHECK-GI-NEXT:    mov v3.s[3], v17.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i64> %a to <16 x float>
  ret <16 x float> %c
}

define <16 x float> @utofp_v16i64_v16f32(<16 x i64> %a) {
; CHECK-SD-LABEL: utofp_v16i64_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov x13, d2
; CHECK-SD-NEXT:    mov x9, v0.d[1]
; CHECK-SD-NEXT:    mov x10, v2.d[1]
; CHECK-SD-NEXT:    fmov x11, d0
; CHECK-SD-NEXT:    mov x12, v4.d[1]
; CHECK-SD-NEXT:    mov x8, v1.d[1]
; CHECK-SD-NEXT:    ucvtf s16, x13
; CHECK-SD-NEXT:    fmov x13, d4
; CHECK-SD-NEXT:    ucvtf s0, x11
; CHECK-SD-NEXT:    mov x11, v6.d[1]
; CHECK-SD-NEXT:    ucvtf s17, x9
; CHECK-SD-NEXT:    ucvtf s18, x10
; CHECK-SD-NEXT:    fmov x9, d1
; CHECK-SD-NEXT:    ucvtf s1, x12
; CHECK-SD-NEXT:    fmov x12, d6
; CHECK-SD-NEXT:    ucvtf s2, x13
; CHECK-SD-NEXT:    fmov x13, d3
; CHECK-SD-NEXT:    mov x10, v3.d[1]
; CHECK-SD-NEXT:    ucvtf s4, x11
; CHECK-SD-NEXT:    mov v0.s[1], v17.s[0]
; CHECK-SD-NEXT:    ucvtf s6, x9
; CHECK-SD-NEXT:    ucvtf s3, x12
; CHECK-SD-NEXT:    mov v16.s[1], v18.s[0]
; CHECK-SD-NEXT:    mov x9, v5.d[1]
; CHECK-SD-NEXT:    fmov x11, d5
; CHECK-SD-NEXT:    ucvtf s5, x13
; CHECK-SD-NEXT:    fmov x13, d7
; CHECK-SD-NEXT:    mov x12, v7.d[1]
; CHECK-SD-NEXT:    mov v2.s[1], v1.s[0]
; CHECK-SD-NEXT:    mov v0.s[2], v6.s[0]
; CHECK-SD-NEXT:    ucvtf s6, x10
; CHECK-SD-NEXT:    ucvtf s7, x11
; CHECK-SD-NEXT:    ucvtf s1, x13
; CHECK-SD-NEXT:    mov v3.s[1], v4.s[0]
; CHECK-SD-NEXT:    mov v16.s[2], v5.s[0]
; CHECK-SD-NEXT:    ucvtf s4, x8
; CHECK-SD-NEXT:    ucvtf s5, x9
; CHECK-SD-NEXT:    mov v2.s[2], v7.s[0]
; CHECK-SD-NEXT:    mov v3.s[2], v1.s[0]
; CHECK-SD-NEXT:    ucvtf s1, x12
; CHECK-SD-NEXT:    mov v16.s[3], v6.s[0]
; CHECK-SD-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-SD-NEXT:    mov v2.s[3], v5.s[0]
; CHECK-SD-NEXT:    mov v3.s[3], v1.s[0]
; CHECK-SD-NEXT:    mov v1.16b, v16.16b
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i64_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    mov x9, v0.d[1]
; CHECK-GI-NEXT:    fmov x13, d2
; CHECK-GI-NEXT:    fmov x11, d0
; CHECK-GI-NEXT:    mov x12, v4.d[1]
; CHECK-GI-NEXT:    mov x8, v1.d[1]
; CHECK-GI-NEXT:    mov x10, v2.d[1]
; CHECK-GI-NEXT:    ucvtf s0, x11
; CHECK-GI-NEXT:    mov x11, v6.d[1]
; CHECK-GI-NEXT:    ucvtf s16, x9
; CHECK-GI-NEXT:    fmov x9, d1
; CHECK-GI-NEXT:    ucvtf s1, x13
; CHECK-GI-NEXT:    fmov x13, d4
; CHECK-GI-NEXT:    ucvtf s4, x12
; CHECK-GI-NEXT:    fmov x12, d6
; CHECK-GI-NEXT:    ucvtf s17, x10
; CHECK-GI-NEXT:    mov x10, v3.d[1]
; CHECK-GI-NEXT:    ucvtf s6, x11
; CHECK-GI-NEXT:    fmov x11, d5
; CHECK-GI-NEXT:    ucvtf s18, x9
; CHECK-GI-NEXT:    ucvtf s2, x13
; CHECK-GI-NEXT:    fmov x13, d3
; CHECK-GI-NEXT:    ucvtf s3, x12
; CHECK-GI-NEXT:    mov x9, v5.d[1]
; CHECK-GI-NEXT:    mov x12, v7.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v16.s[0]
; CHECK-GI-NEXT:    ucvtf s5, x11
; CHECK-GI-NEXT:    mov v1.s[1], v17.s[0]
; CHECK-GI-NEXT:    ucvtf s19, x13
; CHECK-GI-NEXT:    fmov x13, d7
; CHECK-GI-NEXT:    mov v2.s[1], v4.s[0]
; CHECK-GI-NEXT:    mov v3.s[1], v6.s[0]
; CHECK-GI-NEXT:    ucvtf s4, x8
; CHECK-GI-NEXT:    ucvtf s6, x10
; CHECK-GI-NEXT:    ucvtf s16, x9
; CHECK-GI-NEXT:    ucvtf s17, x12
; CHECK-GI-NEXT:    ucvtf s7, x13
; CHECK-GI-NEXT:    mov v0.s[2], v18.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v19.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v5.s[0]
; CHECK-GI-NEXT:    mov v3.s[2], v7.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v4.s[0]
; CHECK-GI-NEXT:    mov v1.s[3], v6.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v16.s[0]
; CHECK-GI-NEXT:    mov v3.s[3], v17.s[0]
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i64> %a to <16 x float>
  ret <16 x float> %c
}

define <32 x float> @stofp_v32i64_v32f32(<32 x i64> %a) {
; CHECK-SD-LABEL: stofp_v32i64_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    fmov x10, d0
; CHECK-SD-NEXT:    mov v16.16b, v1.16b
; CHECK-SD-NEXT:    fmov x11, d2
; CHECK-SD-NEXT:    ldp q24, q20, [sp]
; CHECK-SD-NEXT:    mov x9, v2.d[1]
; CHECK-SD-NEXT:    fmov x12, d3
; CHECK-SD-NEXT:    fmov x13, d4
; CHECK-SD-NEXT:    scvtf s0, x10
; CHECK-SD-NEXT:    ldp q21, q18, [sp, #32]
; CHECK-SD-NEXT:    scvtf s2, x8
; CHECK-SD-NEXT:    scvtf s1, x11
; CHECK-SD-NEXT:    mov x10, v4.d[1]
; CHECK-SD-NEXT:    fmov x11, d16
; CHECK-SD-NEXT:    ldp q19, q17, [sp, #96]
; CHECK-SD-NEXT:    scvtf s22, x9
; CHECK-SD-NEXT:    mov x8, v3.d[1]
; CHECK-SD-NEXT:    scvtf s4, x12
; CHECK-SD-NEXT:    mov x12, v24.d[1]
; CHECK-SD-NEXT:    mov x9, v16.d[1]
; CHECK-SD-NEXT:    scvtf s3, x11
; CHECK-SD-NEXT:    ldp q23, q16, [sp, #64]
; CHECK-SD-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-SD-NEXT:    scvtf s25, x10
; CHECK-SD-NEXT:    fmov x10, d6
; CHECK-SD-NEXT:    mov v1.s[1], v22.s[0]
; CHECK-SD-NEXT:    mov x11, v6.d[1]
; CHECK-SD-NEXT:    scvtf s2, x13
; CHECK-SD-NEXT:    mov x13, v21.d[1]
; CHECK-SD-NEXT:    fmov x14, d19
; CHECK-SD-NEXT:    scvtf s22, x9
; CHECK-SD-NEXT:    mov x9, v5.d[1]
; CHECK-SD-NEXT:    fmov x15, d17
; CHECK-SD-NEXT:    mov v0.s[2], v3.s[0]
; CHECK-SD-NEXT:    scvtf s3, x10
; CHECK-SD-NEXT:    fmov x10, d24
; CHECK-SD-NEXT:    mov v1.s[2], v4.s[0]
; CHECK-SD-NEXT:    scvtf s24, x12
; CHECK-SD-NEXT:    scvtf s6, x11
; CHECK-SD-NEXT:    fmov x11, d5
; CHECK-SD-NEXT:    fmov x12, d7
; CHECK-SD-NEXT:    mov v2.s[1], v25.s[0]
; CHECK-SD-NEXT:    scvtf s4, x10
; CHECK-SD-NEXT:    fmov x10, d21
; CHECK-SD-NEXT:    scvtf s21, x8
; CHECK-SD-NEXT:    mov x8, v23.d[1]
; CHECK-SD-NEXT:    scvtf s25, x13
; CHECK-SD-NEXT:    mov x13, v19.d[1]
; CHECK-SD-NEXT:    scvtf s26, x11
; CHECK-SD-NEXT:    mov x11, v20.d[1]
; CHECK-SD-NEXT:    mov v3.s[1], v6.s[0]
; CHECK-SD-NEXT:    scvtf s5, x10
; CHECK-SD-NEXT:    mov x10, v7.d[1]
; CHECK-SD-NEXT:    scvtf s7, x14
; CHECK-SD-NEXT:    mov v4.s[1], v24.s[0]
; CHECK-SD-NEXT:    scvtf s24, x12
; CHECK-SD-NEXT:    fmov x12, d20
; CHECK-SD-NEXT:    scvtf s20, x8
; CHECK-SD-NEXT:    fmov x8, d23
; CHECK-SD-NEXT:    scvtf s19, x13
; CHECK-SD-NEXT:    fmov x13, d18
; CHECK-SD-NEXT:    fmov x14, d16
; CHECK-SD-NEXT:    mov v2.s[2], v26.s[0]
; CHECK-SD-NEXT:    mov v5.s[1], v25.s[0]
; CHECK-SD-NEXT:    scvtf s23, x10
; CHECK-SD-NEXT:    mov v0.s[3], v22.s[0]
; CHECK-SD-NEXT:    scvtf s6, x8
; CHECK-SD-NEXT:    mov x8, v18.d[1]
; CHECK-SD-NEXT:    scvtf s18, x12
; CHECK-SD-NEXT:    mov x12, v16.d[1]
; CHECK-SD-NEXT:    scvtf s16, x13
; CHECK-SD-NEXT:    mov x13, v17.d[1]
; CHECK-SD-NEXT:    scvtf s17, x14
; CHECK-SD-NEXT:    mov v7.s[1], v19.s[0]
; CHECK-SD-NEXT:    scvtf s19, x9
; CHECK-SD-NEXT:    mov v3.s[2], v24.s[0]
; CHECK-SD-NEXT:    scvtf s24, x11
; CHECK-SD-NEXT:    mov v1.s[3], v21.s[0]
; CHECK-SD-NEXT:    mov v6.s[1], v20.s[0]
; CHECK-SD-NEXT:    scvtf s20, x15
; CHECK-SD-NEXT:    mov v4.s[2], v18.s[0]
; CHECK-SD-NEXT:    scvtf s18, x8
; CHECK-SD-NEXT:    mov v5.s[2], v16.s[0]
; CHECK-SD-NEXT:    scvtf s16, x12
; CHECK-SD-NEXT:    mov v2.s[3], v19.s[0]
; CHECK-SD-NEXT:    mov v3.s[3], v23.s[0]
; CHECK-SD-NEXT:    mov v6.s[2], v17.s[0]
; CHECK-SD-NEXT:    mov v7.s[2], v20.s[0]
; CHECK-SD-NEXT:    scvtf s17, x13
; CHECK-SD-NEXT:    mov v4.s[3], v24.s[0]
; CHECK-SD-NEXT:    mov v5.s[3], v18.s[0]
; CHECK-SD-NEXT:    mov v6.s[3], v16.s[0]
; CHECK-SD-NEXT:    mov v7.s[3], v17.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i64_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    str d12, [sp, #-48]! // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp d11, d10, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset b8, -8
; CHECK-GI-NEXT:    .cfi_offset b9, -16
; CHECK-GI-NEXT:    .cfi_offset b10, -24
; CHECK-GI-NEXT:    .cfi_offset b11, -32
; CHECK-GI-NEXT:    .cfi_offset b12, -48
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    mov x10, v1.d[1]
; CHECK-GI-NEXT:    mov x12, v4.d[1]
; CHECK-GI-NEXT:    mov x11, v3.d[1]
; CHECK-GI-NEXT:    ldp q30, q25, [sp, #112]
; CHECK-GI-NEXT:    fmov x9, d0
; CHECK-GI-NEXT:    ldp q21, q24, [sp, #48]
; CHECK-GI-NEXT:    ldp q11, q27, [sp, #80]
; CHECK-GI-NEXT:    scvtf s28, x8
; CHECK-GI-NEXT:    fmov x8, d1
; CHECK-GI-NEXT:    scvtf s16, x10
; CHECK-GI-NEXT:    mov x10, v6.d[1]
; CHECK-GI-NEXT:    scvtf s9, x12
; CHECK-GI-NEXT:    fmov x12, d6
; CHECK-GI-NEXT:    scvtf s0, x9
; CHECK-GI-NEXT:    mov x9, v2.d[1]
; CHECK-GI-NEXT:    scvtf s18, x11
; CHECK-GI-NEXT:    scvtf s17, x8
; CHECK-GI-NEXT:    fmov x8, d2
; CHECK-GI-NEXT:    fmov x11, d5
; CHECK-GI-NEXT:    ldp q31, q26, [sp, #144]
; CHECK-GI-NEXT:    mov x13, v21.d[1]
; CHECK-GI-NEXT:    scvtf s10, x10
; CHECK-GI-NEXT:    fmov x10, d21
; CHECK-GI-NEXT:    fmov x14, d30
; CHECK-GI-NEXT:    scvtf s1, x8
; CHECK-GI-NEXT:    fmov x8, d3
; CHECK-GI-NEXT:    scvtf s3, x12
; CHECK-GI-NEXT:    mov x12, v30.d[1]
; CHECK-GI-NEXT:    scvtf s22, x11
; CHECK-GI-NEXT:    fmov x11, d11
; CHECK-GI-NEXT:    scvtf s29, x9
; CHECK-GI-NEXT:    fmov x9, d4
; CHECK-GI-NEXT:    scvtf s4, x10
; CHECK-GI-NEXT:    scvtf s19, x8
; CHECK-GI-NEXT:    mov x8, v5.d[1]
; CHECK-GI-NEXT:    mov x10, v11.d[1]
; CHECK-GI-NEXT:    scvtf s5, x11
; CHECK-GI-NEXT:    mov x11, v31.d[1]
; CHECK-GI-NEXT:    scvtf s8, x13
; CHECK-GI-NEXT:    scvtf s11, x12
; CHECK-GI-NEXT:    fmov x12, d31
; CHECK-GI-NEXT:    scvtf s2, x9
; CHECK-GI-NEXT:    mov x9, v7.d[1]
; CHECK-GI-NEXT:    scvtf s6, x14
; CHECK-GI-NEXT:    fmov x13, d27
; CHECK-GI-NEXT:    scvtf s20, x8
; CHECK-GI-NEXT:    fmov x8, d7
; CHECK-GI-NEXT:    scvtf s12, x10
; CHECK-GI-NEXT:    scvtf s7, x12
; CHECK-GI-NEXT:    fmov x12, d25
; CHECK-GI-NEXT:    scvtf s30, x11
; CHECK-GI-NEXT:    mov x10, v25.d[1]
; CHECK-GI-NEXT:    mov x11, v26.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v28.s[0]
; CHECK-GI-NEXT:    scvtf s21, x9
; CHECK-GI-NEXT:    fmov x9, d24
; CHECK-GI-NEXT:    scvtf s23, x8
; CHECK-GI-NEXT:    scvtf s25, x12
; CHECK-GI-NEXT:    fmov x12, d26
; CHECK-GI-NEXT:    mov x8, v24.d[1]
; CHECK-GI-NEXT:    mov v1.s[1], v29.s[0]
; CHECK-GI-NEXT:    mov v2.s[1], v9.s[0]
; CHECK-GI-NEXT:    mov v3.s[1], v10.s[0]
; CHECK-GI-NEXT:    scvtf s24, x9
; CHECK-GI-NEXT:    mov x9, v27.d[1]
; CHECK-GI-NEXT:    scvtf s27, x13
; CHECK-GI-NEXT:    scvtf s26, x12
; CHECK-GI-NEXT:    mov v4.s[1], v8.s[0]
; CHECK-GI-NEXT:    mov v5.s[1], v12.s[0]
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v6.s[1], v11.s[0]
; CHECK-GI-NEXT:    ldp d11, d10, [sp, #16] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v7.s[1], v30.s[0]
; CHECK-GI-NEXT:    scvtf s28, x8
; CHECK-GI-NEXT:    scvtf s29, x9
; CHECK-GI-NEXT:    scvtf s30, x10
; CHECK-GI-NEXT:    scvtf s31, x11
; CHECK-GI-NEXT:    mov v0.s[2], v17.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v19.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v22.s[0]
; CHECK-GI-NEXT:    mov v3.s[2], v23.s[0]
; CHECK-GI-NEXT:    mov v4.s[2], v24.s[0]
; CHECK-GI-NEXT:    mov v5.s[2], v27.s[0]
; CHECK-GI-NEXT:    mov v6.s[2], v25.s[0]
; CHECK-GI-NEXT:    mov v7.s[2], v26.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v16.s[0]
; CHECK-GI-NEXT:    mov v1.s[3], v18.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v20.s[0]
; CHECK-GI-NEXT:    mov v3.s[3], v21.s[0]
; CHECK-GI-NEXT:    mov v4.s[3], v28.s[0]
; CHECK-GI-NEXT:    mov v5.s[3], v29.s[0]
; CHECK-GI-NEXT:    mov v6.s[3], v30.s[0]
; CHECK-GI-NEXT:    mov v7.s[3], v31.s[0]
; CHECK-GI-NEXT:    ldr d12, [sp], #48 // 8-byte Folded Reload
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i64> %a to <32 x float>
  ret <32 x float> %c
}

define <32 x float> @utofp_v32i64_v32f32(<32 x i64> %a) {
; CHECK-SD-LABEL: utofp_v32i64_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NEXT:    fmov x10, d0
; CHECK-SD-NEXT:    mov v16.16b, v1.16b
; CHECK-SD-NEXT:    fmov x11, d2
; CHECK-SD-NEXT:    ldp q24, q20, [sp]
; CHECK-SD-NEXT:    mov x9, v2.d[1]
; CHECK-SD-NEXT:    fmov x12, d3
; CHECK-SD-NEXT:    fmov x13, d4
; CHECK-SD-NEXT:    ucvtf s0, x10
; CHECK-SD-NEXT:    ldp q21, q18, [sp, #32]
; CHECK-SD-NEXT:    ucvtf s2, x8
; CHECK-SD-NEXT:    ucvtf s1, x11
; CHECK-SD-NEXT:    mov x10, v4.d[1]
; CHECK-SD-NEXT:    fmov x11, d16
; CHECK-SD-NEXT:    ldp q19, q17, [sp, #96]
; CHECK-SD-NEXT:    ucvtf s22, x9
; CHECK-SD-NEXT:    mov x8, v3.d[1]
; CHECK-SD-NEXT:    ucvtf s4, x12
; CHECK-SD-NEXT:    mov x12, v24.d[1]
; CHECK-SD-NEXT:    mov x9, v16.d[1]
; CHECK-SD-NEXT:    ucvtf s3, x11
; CHECK-SD-NEXT:    ldp q23, q16, [sp, #64]
; CHECK-SD-NEXT:    mov v0.s[1], v2.s[0]
; CHECK-SD-NEXT:    ucvtf s25, x10
; CHECK-SD-NEXT:    fmov x10, d6
; CHECK-SD-NEXT:    mov v1.s[1], v22.s[0]
; CHECK-SD-NEXT:    mov x11, v6.d[1]
; CHECK-SD-NEXT:    ucvtf s2, x13
; CHECK-SD-NEXT:    mov x13, v21.d[1]
; CHECK-SD-NEXT:    fmov x14, d19
; CHECK-SD-NEXT:    ucvtf s22, x9
; CHECK-SD-NEXT:    mov x9, v5.d[1]
; CHECK-SD-NEXT:    fmov x15, d17
; CHECK-SD-NEXT:    mov v0.s[2], v3.s[0]
; CHECK-SD-NEXT:    ucvtf s3, x10
; CHECK-SD-NEXT:    fmov x10, d24
; CHECK-SD-NEXT:    mov v1.s[2], v4.s[0]
; CHECK-SD-NEXT:    ucvtf s24, x12
; CHECK-SD-NEXT:    ucvtf s6, x11
; CHECK-SD-NEXT:    fmov x11, d5
; CHECK-SD-NEXT:    fmov x12, d7
; CHECK-SD-NEXT:    mov v2.s[1], v25.s[0]
; CHECK-SD-NEXT:    ucvtf s4, x10
; CHECK-SD-NEXT:    fmov x10, d21
; CHECK-SD-NEXT:    ucvtf s21, x8
; CHECK-SD-NEXT:    mov x8, v23.d[1]
; CHECK-SD-NEXT:    ucvtf s25, x13
; CHECK-SD-NEXT:    mov x13, v19.d[1]
; CHECK-SD-NEXT:    ucvtf s26, x11
; CHECK-SD-NEXT:    mov x11, v20.d[1]
; CHECK-SD-NEXT:    mov v3.s[1], v6.s[0]
; CHECK-SD-NEXT:    ucvtf s5, x10
; CHECK-SD-NEXT:    mov x10, v7.d[1]
; CHECK-SD-NEXT:    ucvtf s7, x14
; CHECK-SD-NEXT:    mov v4.s[1], v24.s[0]
; CHECK-SD-NEXT:    ucvtf s24, x12
; CHECK-SD-NEXT:    fmov x12, d20
; CHECK-SD-NEXT:    ucvtf s20, x8
; CHECK-SD-NEXT:    fmov x8, d23
; CHECK-SD-NEXT:    ucvtf s19, x13
; CHECK-SD-NEXT:    fmov x13, d18
; CHECK-SD-NEXT:    fmov x14, d16
; CHECK-SD-NEXT:    mov v2.s[2], v26.s[0]
; CHECK-SD-NEXT:    mov v5.s[1], v25.s[0]
; CHECK-SD-NEXT:    ucvtf s23, x10
; CHECK-SD-NEXT:    mov v0.s[3], v22.s[0]
; CHECK-SD-NEXT:    ucvtf s6, x8
; CHECK-SD-NEXT:    mov x8, v18.d[1]
; CHECK-SD-NEXT:    ucvtf s18, x12
; CHECK-SD-NEXT:    mov x12, v16.d[1]
; CHECK-SD-NEXT:    ucvtf s16, x13
; CHECK-SD-NEXT:    mov x13, v17.d[1]
; CHECK-SD-NEXT:    ucvtf s17, x14
; CHECK-SD-NEXT:    mov v7.s[1], v19.s[0]
; CHECK-SD-NEXT:    ucvtf s19, x9
; CHECK-SD-NEXT:    mov v3.s[2], v24.s[0]
; CHECK-SD-NEXT:    ucvtf s24, x11
; CHECK-SD-NEXT:    mov v1.s[3], v21.s[0]
; CHECK-SD-NEXT:    mov v6.s[1], v20.s[0]
; CHECK-SD-NEXT:    ucvtf s20, x15
; CHECK-SD-NEXT:    mov v4.s[2], v18.s[0]
; CHECK-SD-NEXT:    ucvtf s18, x8
; CHECK-SD-NEXT:    mov v5.s[2], v16.s[0]
; CHECK-SD-NEXT:    ucvtf s16, x12
; CHECK-SD-NEXT:    mov v2.s[3], v19.s[0]
; CHECK-SD-NEXT:    mov v3.s[3], v23.s[0]
; CHECK-SD-NEXT:    mov v6.s[2], v17.s[0]
; CHECK-SD-NEXT:    mov v7.s[2], v20.s[0]
; CHECK-SD-NEXT:    ucvtf s17, x13
; CHECK-SD-NEXT:    mov v4.s[3], v24.s[0]
; CHECK-SD-NEXT:    mov v5.s[3], v18.s[0]
; CHECK-SD-NEXT:    mov v6.s[3], v16.s[0]
; CHECK-SD-NEXT:    mov v7.s[3], v17.s[0]
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i64_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    str d12, [sp, #-48]! // 8-byte Folded Spill
; CHECK-GI-NEXT:    stp d11, d10, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NEXT:    .cfi_offset b8, -8
; CHECK-GI-NEXT:    .cfi_offset b9, -16
; CHECK-GI-NEXT:    .cfi_offset b10, -24
; CHECK-GI-NEXT:    .cfi_offset b11, -32
; CHECK-GI-NEXT:    .cfi_offset b12, -48
; CHECK-GI-NEXT:    mov x8, v0.d[1]
; CHECK-GI-NEXT:    mov x10, v1.d[1]
; CHECK-GI-NEXT:    mov x12, v4.d[1]
; CHECK-GI-NEXT:    mov x11, v3.d[1]
; CHECK-GI-NEXT:    ldp q30, q25, [sp, #112]
; CHECK-GI-NEXT:    fmov x9, d0
; CHECK-GI-NEXT:    ldp q21, q24, [sp, #48]
; CHECK-GI-NEXT:    ldp q11, q27, [sp, #80]
; CHECK-GI-NEXT:    ucvtf s28, x8
; CHECK-GI-NEXT:    fmov x8, d1
; CHECK-GI-NEXT:    ucvtf s16, x10
; CHECK-GI-NEXT:    mov x10, v6.d[1]
; CHECK-GI-NEXT:    ucvtf s9, x12
; CHECK-GI-NEXT:    fmov x12, d6
; CHECK-GI-NEXT:    ucvtf s0, x9
; CHECK-GI-NEXT:    mov x9, v2.d[1]
; CHECK-GI-NEXT:    ucvtf s18, x11
; CHECK-GI-NEXT:    ucvtf s17, x8
; CHECK-GI-NEXT:    fmov x8, d2
; CHECK-GI-NEXT:    fmov x11, d5
; CHECK-GI-NEXT:    ldp q31, q26, [sp, #144]
; CHECK-GI-NEXT:    mov x13, v21.d[1]
; CHECK-GI-NEXT:    ucvtf s10, x10
; CHECK-GI-NEXT:    fmov x10, d21
; CHECK-GI-NEXT:    fmov x14, d30
; CHECK-GI-NEXT:    ucvtf s1, x8
; CHECK-GI-NEXT:    fmov x8, d3
; CHECK-GI-NEXT:    ucvtf s3, x12
; CHECK-GI-NEXT:    mov x12, v30.d[1]
; CHECK-GI-NEXT:    ucvtf s22, x11
; CHECK-GI-NEXT:    fmov x11, d11
; CHECK-GI-NEXT:    ucvtf s29, x9
; CHECK-GI-NEXT:    fmov x9, d4
; CHECK-GI-NEXT:    ucvtf s4, x10
; CHECK-GI-NEXT:    ucvtf s19, x8
; CHECK-GI-NEXT:    mov x8, v5.d[1]
; CHECK-GI-NEXT:    mov x10, v11.d[1]
; CHECK-GI-NEXT:    ucvtf s5, x11
; CHECK-GI-NEXT:    mov x11, v31.d[1]
; CHECK-GI-NEXT:    ucvtf s8, x13
; CHECK-GI-NEXT:    ucvtf s11, x12
; CHECK-GI-NEXT:    fmov x12, d31
; CHECK-GI-NEXT:    ucvtf s2, x9
; CHECK-GI-NEXT:    mov x9, v7.d[1]
; CHECK-GI-NEXT:    ucvtf s6, x14
; CHECK-GI-NEXT:    fmov x13, d27
; CHECK-GI-NEXT:    ucvtf s20, x8
; CHECK-GI-NEXT:    fmov x8, d7
; CHECK-GI-NEXT:    ucvtf s12, x10
; CHECK-GI-NEXT:    ucvtf s7, x12
; CHECK-GI-NEXT:    fmov x12, d25
; CHECK-GI-NEXT:    ucvtf s30, x11
; CHECK-GI-NEXT:    mov x10, v25.d[1]
; CHECK-GI-NEXT:    mov x11, v26.d[1]
; CHECK-GI-NEXT:    mov v0.s[1], v28.s[0]
; CHECK-GI-NEXT:    ucvtf s21, x9
; CHECK-GI-NEXT:    fmov x9, d24
; CHECK-GI-NEXT:    ucvtf s23, x8
; CHECK-GI-NEXT:    ucvtf s25, x12
; CHECK-GI-NEXT:    fmov x12, d26
; CHECK-GI-NEXT:    mov x8, v24.d[1]
; CHECK-GI-NEXT:    mov v1.s[1], v29.s[0]
; CHECK-GI-NEXT:    mov v2.s[1], v9.s[0]
; CHECK-GI-NEXT:    mov v3.s[1], v10.s[0]
; CHECK-GI-NEXT:    ucvtf s24, x9
; CHECK-GI-NEXT:    mov x9, v27.d[1]
; CHECK-GI-NEXT:    ucvtf s27, x13
; CHECK-GI-NEXT:    ucvtf s26, x12
; CHECK-GI-NEXT:    mov v4.s[1], v8.s[0]
; CHECK-GI-NEXT:    mov v5.s[1], v12.s[0]
; CHECK-GI-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v6.s[1], v11.s[0]
; CHECK-GI-NEXT:    ldp d11, d10, [sp, #16] // 16-byte Folded Reload
; CHECK-GI-NEXT:    mov v7.s[1], v30.s[0]
; CHECK-GI-NEXT:    ucvtf s28, x8
; CHECK-GI-NEXT:    ucvtf s29, x9
; CHECK-GI-NEXT:    ucvtf s30, x10
; CHECK-GI-NEXT:    ucvtf s31, x11
; CHECK-GI-NEXT:    mov v0.s[2], v17.s[0]
; CHECK-GI-NEXT:    mov v1.s[2], v19.s[0]
; CHECK-GI-NEXT:    mov v2.s[2], v22.s[0]
; CHECK-GI-NEXT:    mov v3.s[2], v23.s[0]
; CHECK-GI-NEXT:    mov v4.s[2], v24.s[0]
; CHECK-GI-NEXT:    mov v5.s[2], v27.s[0]
; CHECK-GI-NEXT:    mov v6.s[2], v25.s[0]
; CHECK-GI-NEXT:    mov v7.s[2], v26.s[0]
; CHECK-GI-NEXT:    mov v0.s[3], v16.s[0]
; CHECK-GI-NEXT:    mov v1.s[3], v18.s[0]
; CHECK-GI-NEXT:    mov v2.s[3], v20.s[0]
; CHECK-GI-NEXT:    mov v3.s[3], v21.s[0]
; CHECK-GI-NEXT:    mov v4.s[3], v28.s[0]
; CHECK-GI-NEXT:    mov v5.s[3], v29.s[0]
; CHECK-GI-NEXT:    mov v6.s[3], v30.s[0]
; CHECK-GI-NEXT:    mov v7.s[3], v31.s[0]
; CHECK-GI-NEXT:    ldr d12, [sp], #48 // 8-byte Folded Reload
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i64> %a to <32 x float>
  ret <32 x float> %c
}

define <2 x float> @stofp_v2i32_v2f32(<2 x i32> %a) {
; CHECK-LABEL: stofp_v2i32_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i32> %a to <2 x float>
  ret <2 x float> %c
}

define <2 x float> @utofp_v2i32_v2f32(<2 x i32> %a) {
; CHECK-LABEL: utofp_v2i32_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i32> %a to <2 x float>
  ret <2 x float> %c
}

define <3 x float> @stofp_v3i32_v3f32(<3 x i32> %a) {
; CHECK-LABEL: stofp_v3i32_v3f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <3 x i32> %a to <3 x float>
  ret <3 x float> %c
}

define <3 x float> @utofp_v3i32_v3f32(<3 x i32> %a) {
; CHECK-LABEL: utofp_v3i32_v3f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <3 x i32> %a to <3 x float>
  ret <3 x float> %c
}

define <4 x float> @stofp_v4i32_v4f32(<4 x i32> %a) {
; CHECK-LABEL: stofp_v4i32_v4f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <4 x i32> %a to <4 x float>
  ret <4 x float> %c
}

define <4 x float> @utofp_v4i32_v4f32(<4 x i32> %a) {
; CHECK-LABEL: utofp_v4i32_v4f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <4 x i32> %a to <4 x float>
  ret <4 x float> %c
}

define <8 x float> @stofp_v8i32_v8f32(<8 x i32> %a) {
; CHECK-LABEL: stofp_v8i32_v8f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    scvtf v1.4s, v1.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <8 x i32> %a to <8 x float>
  ret <8 x float> %c
}

define <8 x float> @utofp_v8i32_v8f32(<8 x i32> %a) {
; CHECK-LABEL: utofp_v8i32_v8f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <8 x i32> %a to <8 x float>
  ret <8 x float> %c
}

define <16 x float> @stofp_v16i32_v16f32(<16 x i32> %a) {
; CHECK-LABEL: stofp_v16i32_v16f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    scvtf v1.4s, v1.4s
; CHECK-NEXT:    scvtf v2.4s, v2.4s
; CHECK-NEXT:    scvtf v3.4s, v3.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <16 x i32> %a to <16 x float>
  ret <16 x float> %c
}

define <16 x float> @utofp_v16i32_v16f32(<16 x i32> %a) {
; CHECK-LABEL: utofp_v16i32_v16f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <16 x i32> %a to <16 x float>
  ret <16 x float> %c
}

define <32 x float> @stofp_v32i32_v32f32(<32 x i32> %a) {
; CHECK-LABEL: stofp_v32i32_v32f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    scvtf v1.4s, v1.4s
; CHECK-NEXT:    scvtf v2.4s, v2.4s
; CHECK-NEXT:    scvtf v3.4s, v3.4s
; CHECK-NEXT:    scvtf v4.4s, v4.4s
; CHECK-NEXT:    scvtf v5.4s, v5.4s
; CHECK-NEXT:    scvtf v6.4s, v6.4s
; CHECK-NEXT:    scvtf v7.4s, v7.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <32 x i32> %a to <32 x float>
  ret <32 x float> %c
}

define <32 x float> @utofp_v32i32_v32f32(<32 x i32> %a) {
; CHECK-LABEL: utofp_v32i32_v32f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <32 x i32> %a to <32 x float>
  ret <32 x float> %c
}

define <2 x float> @stofp_v2i16_v2f32(<2 x i16> %a) {
; CHECK-LABEL: stofp_v2i16_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-NEXT:    scvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i16> %a to <2 x float>
  ret <2 x float> %c
}

define <2 x float> @utofp_v2i16_v2f32(<2 x i16> %a) {
; CHECK-LABEL: utofp_v2i16_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    movi d1, #0x00ffff0000ffff
; CHECK-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i16> %a to <2 x float>
  ret <2 x float> %c
}

define <3 x float> @stofp_v3i16_v3f32(<3 x i16> %a) {
; CHECK-LABEL: stofp_v3i16_v3f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <3 x i16> %a to <3 x float>
  ret <3 x float> %c
}

define <3 x float> @utofp_v3i16_v3f32(<3 x i16> %a) {
; CHECK-LABEL: utofp_v3i16_v3f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <3 x i16> %a to <3 x float>
  ret <3 x float> %c
}

define <4 x float> @stofp_v4i16_v4f32(<4 x i16> %a) {
; CHECK-LABEL: stofp_v4i16_v4f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <4 x i16> %a to <4 x float>
  ret <4 x float> %c
}

define <4 x float> @utofp_v4i16_v4f32(<4 x i16> %a) {
; CHECK-LABEL: utofp_v4i16_v4f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <4 x i16> %a to <4 x float>
  ret <4 x float> %c
}

define <8 x float> @stofp_v8i16_v8f32(<8 x i16> %a) {
; CHECK-SD-LABEL: stofp_v8i16_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v1.4s, v1.4s
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i16_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v0.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v1.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i16> %a to <8 x float>
  ret <8 x float> %c
}

define <8 x float> @utofp_v8i16_v8f32(<8 x i16> %a) {
; CHECK-SD-LABEL: utofp_v8i16_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v1.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i16_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v2.4s, v0.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v1.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i16> %a to <8 x float>
  ret <8 x float> %c
}

define <16 x float> @stofp_v16i16_v16f32(<16 x i16> %a) {
; CHECK-SD-LABEL: stofp_v16i16_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v3.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll2 v4.4s, v1.8h, #0
; CHECK-SD-NEXT:    sshll v5.4s, v1.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v2.4s
; CHECK-SD-NEXT:    scvtf v1.4s, v3.4s
; CHECK-SD-NEXT:    scvtf v3.4s, v4.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v5.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i16_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v4.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v5.4s, v1.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v2.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v3.4s
; CHECK-GI-NEXT:    scvtf v2.4s, v4.4s
; CHECK-GI-NEXT:    scvtf v3.4s, v5.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i16> %a to <16 x float>
  ret <16 x float> %c
}

define <16 x float> @utofp_v16i16_v16f32(<16 x i16> %a) {
; CHECK-SD-LABEL: utofp_v16i16_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v3.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll2 v4.4s, v1.8h, #0
; CHECK-SD-NEXT:    ushll v5.4s, v1.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v2.4s
; CHECK-SD-NEXT:    ucvtf v1.4s, v3.4s
; CHECK-SD-NEXT:    ucvtf v3.4s, v4.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v5.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i16_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v3.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v4.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v5.4s, v1.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v2.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v3.4s
; CHECK-GI-NEXT:    ucvtf v2.4s, v4.4s
; CHECK-GI-NEXT:    ucvtf v3.4s, v5.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i16> %a to <16 x float>
  ret <16 x float> %c
}

define <32 x float> @stofp_v32i16_v32f32(<32 x i16> %a) {
; CHECK-SD-LABEL: stofp_v32i16_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    sshll2 v4.4s, v0.8h, #0
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    sshll2 v5.4s, v1.8h, #0
; CHECK-SD-NEXT:    sshll v6.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll v7.4s, v2.4h, #0
; CHECK-SD-NEXT:    sshll2 v16.4s, v2.8h, #0
; CHECK-SD-NEXT:    sshll2 v17.4s, v3.8h, #0
; CHECK-SD-NEXT:    sshll v18.4s, v3.4h, #0
; CHECK-SD-NEXT:    scvtf v1.4s, v4.4s
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    scvtf v3.4s, v5.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v6.4s
; CHECK-SD-NEXT:    scvtf v4.4s, v7.4s
; CHECK-SD-NEXT:    scvtf v5.4s, v16.4s
; CHECK-SD-NEXT:    scvtf v7.4s, v17.4s
; CHECK-SD-NEXT:    scvtf v6.4s, v18.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i16_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v4.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v5.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v6.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v7.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v16.4s, v2.4h, #0
; CHECK-GI-NEXT:    sshll2 v17.4s, v2.8h, #0
; CHECK-GI-NEXT:    sshll v18.4s, v3.4h, #0
; CHECK-GI-NEXT:    sshll2 v19.4s, v3.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v4.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v5.4s
; CHECK-GI-NEXT:    scvtf v2.4s, v6.4s
; CHECK-GI-NEXT:    scvtf v3.4s, v7.4s
; CHECK-GI-NEXT:    scvtf v4.4s, v16.4s
; CHECK-GI-NEXT:    scvtf v5.4s, v17.4s
; CHECK-GI-NEXT:    scvtf v6.4s, v18.4s
; CHECK-GI-NEXT:    scvtf v7.4s, v19.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i16> %a to <32 x float>
  ret <32 x float> %c
}

define <32 x float> @utofp_v32i16_v32f32(<32 x i16> %a) {
; CHECK-SD-LABEL: utofp_v32i16_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ushll2 v4.4s, v0.8h, #0
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ushll2 v5.4s, v1.8h, #0
; CHECK-SD-NEXT:    ushll v6.4s, v1.4h, #0
; CHECK-SD-NEXT:    ushll v7.4s, v2.4h, #0
; CHECK-SD-NEXT:    ushll2 v16.4s, v2.8h, #0
; CHECK-SD-NEXT:    ushll2 v17.4s, v3.8h, #0
; CHECK-SD-NEXT:    ushll v18.4s, v3.4h, #0
; CHECK-SD-NEXT:    ucvtf v1.4s, v4.4s
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ucvtf v3.4s, v5.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v6.4s
; CHECK-SD-NEXT:    ucvtf v4.4s, v7.4s
; CHECK-SD-NEXT:    ucvtf v5.4s, v16.4s
; CHECK-SD-NEXT:    ucvtf v7.4s, v17.4s
; CHECK-SD-NEXT:    ucvtf v6.4s, v18.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i16_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v4.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v5.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v6.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v7.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v16.4s, v2.4h, #0
; CHECK-GI-NEXT:    ushll2 v17.4s, v2.8h, #0
; CHECK-GI-NEXT:    ushll v18.4s, v3.4h, #0
; CHECK-GI-NEXT:    ushll2 v19.4s, v3.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v4.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v5.4s
; CHECK-GI-NEXT:    ucvtf v2.4s, v6.4s
; CHECK-GI-NEXT:    ucvtf v3.4s, v7.4s
; CHECK-GI-NEXT:    ucvtf v4.4s, v16.4s
; CHECK-GI-NEXT:    ucvtf v5.4s, v17.4s
; CHECK-GI-NEXT:    ucvtf v6.4s, v18.4s
; CHECK-GI-NEXT:    ucvtf v7.4s, v19.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i16> %a to <32 x float>
  ret <32 x float> %c
}

define <2 x float> @stofp_v2i8_v2f32(<2 x i8> %a) {
; CHECK-LABEL: stofp_v2i8_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    shl v0.2s, v0.2s, #24
; CHECK-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-NEXT:    scvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <2 x i8> %a to <2 x float>
  ret <2 x float> %c
}

define <2 x float> @utofp_v2i8_v2f32(<2 x i8> %a) {
; CHECK-LABEL: utofp_v2i8_v2f32:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <2 x i8> %a to <2 x float>
  ret <2 x float> %c
}

define <3 x float> @stofp_v3i8_v3f32(<3 x i8> %a) {
; CHECK-SD-LABEL: stofp_v3i8_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov s0, w0
; CHECK-SD-NEXT:    mov v0.h[1], w1
; CHECK-SD-NEXT:    mov v0.h[2], w2
; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v3i8_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    mov v0.h[1], w1
; CHECK-GI-NEXT:    mov v0.h[2], w2
; CHECK-GI-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-GI-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-GI-NEXT:    mov v1.h[0], v0.h[2]
; CHECK-GI-NEXT:    mov v1.h[1], v0.h[3]
; CHECK-GI-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <3 x i8> %a to <3 x float>
  ret <3 x float> %c
}

define <3 x float> @utofp_v3i8_v3f32(<3 x i8> %a) {
; CHECK-SD-LABEL: utofp_v3i8_v3f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    fmov s0, w0
; CHECK-SD-NEXT:    mov v0.h[1], w1
; CHECK-SD-NEXT:    mov v0.h[2], w2
; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v3i8_v3f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    fmov s0, w0
; CHECK-GI-NEXT:    movi d1, #0xff00ff00ff00ff
; CHECK-GI-NEXT:    mov v0.h[1], w1
; CHECK-GI-NEXT:    mov v0.h[2], w2
; CHECK-GI-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-NEXT:    mov v1.h[0], v0.h[2]
; CHECK-GI-NEXT:    mov v1.h[1], v0.h[3]
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll v1.4s, v1.4h, #0
; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <3 x i8> %a to <3 x float>
  ret <3 x float> %c
}

define <4 x float> @stofp_v4i8_v4f32(<4 x i8> %a) {
; CHECK-SD-LABEL: stofp_v4i8_v4f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v4i8_v4f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    shl v0.4s, v0.4s, #24
; CHECK-GI-NEXT:    sshr v0.4s, v0.4s, #24
; CHECK-GI-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <4 x i8> %a to <4 x float>
  ret <4 x float> %c
}

define <4 x float> @utofp_v4i8_v4f32(<4 x i8> %a) {
; CHECK-SD-LABEL: utofp_v4i8_v4f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v4i8_v4f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    movi v1.2d, #0x0000ff000000ff
; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-GI-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <4 x i8> %a to <4 x float>
  ret <4 x float> %c
}

define <8 x float> @stofp_v8i8_v8f32(<8 x i8> %a) {
; CHECK-SD-LABEL: stofp_v8i8_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    zip1 v1.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    shl v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshr v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v1.4s
; CHECK-SD-NEXT:    scvtf v1.4s, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i8_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v0.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v1.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i8> %a to <8 x float>
  ret <8 x float> %c
}

define <8 x float> @utofp_v8i8_v8f32(<8 x i8> %a) {
; CHECK-SD-LABEL: utofp_v8i8_v8f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    zip1 v1.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    bic v1.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v1.4s, v1.4h, #0
; CHECK-SD-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v1.4s
; CHECK-SD-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i8_v8f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v2.4s, v0.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v1.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i8> %a to <8 x float>
  ret <8 x float> %c
}

define <16 x float> @stofp_v16i8_v16f32(<16 x i8> %a) {
; CHECK-SD-LABEL: stofp_v16i8_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    zip1 v2.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v3.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v1.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    shl v2.4h, v2.4h, #8
; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshr v2.4h, v2.4h, #8
; CHECK-SD-NEXT:    shl v3.4h, v3.4h, #8
; CHECK-SD-NEXT:    shl v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    sshr v3.4h, v3.4h, #8
; CHECK-SD-NEXT:    sshr v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    sshll v4.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v2.4s
; CHECK-SD-NEXT:    sshll v3.4s, v3.4h, #0
; CHECK-SD-NEXT:    sshll v5.4s, v1.4h, #0
; CHECK-SD-NEXT:    scvtf v1.4s, v4.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v3.4s
; CHECK-SD-NEXT:    scvtf v3.4s, v5.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i8_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v1.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    sshll v2.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    sshll v3.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v4.4s, v0.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v2.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v1.4s
; CHECK-GI-NEXT:    scvtf v2.4s, v3.4s
; CHECK-GI-NEXT:    scvtf v3.4s, v4.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i8> %a to <16 x float>
  ret <16 x float> %c
}

define <16 x float> @utofp_v16i8_v16f32(<16 x i8> %a) {
; CHECK-SD-LABEL: utofp_v16i8_v16f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v1.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    zip1 v2.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v3.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v1.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    bic v2.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v2.4s, v2.4h, #0
; CHECK-SD-NEXT:    bic v3.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v1.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v4.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v2.4s
; CHECK-SD-NEXT:    ushll v3.4s, v3.4h, #0
; CHECK-SD-NEXT:    ushll v5.4s, v1.4h, #0
; CHECK-SD-NEXT:    ucvtf v1.4s, v4.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v3.4s
; CHECK-SD-NEXT:    ucvtf v3.4s, v5.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i8_v16f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v1.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    ushll v2.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    ushll v3.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v4.4s, v0.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v2.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-GI-NEXT:    ucvtf v2.4s, v3.4s
; CHECK-GI-NEXT:    ucvtf v3.4s, v4.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i8> %a to <16 x float>
  ret <16 x float> %c
}

define <32 x float> @stofp_v32i8_v32f32(<32 x i8> %a) {
; CHECK-SD-LABEL: stofp_v32i8_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    ext v3.16b, v1.16b, v1.16b, #8
; CHECK-SD-NEXT:    zip1 v4.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    shl v4.4h, v4.4h, #8
; CHECK-SD-NEXT:    zip1 v5.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v6.8b, v2.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v2.8b, v2.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v1.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v7.8b, v3.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v3.8b, v3.8b, v0.8b
; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshr v4.4h, v4.4h, #8
; CHECK-SD-NEXT:    shl v5.4h, v5.4h, #8
; CHECK-SD-NEXT:    shl v6.4h, v6.4h, #8
; CHECK-SD-NEXT:    shl v2.4h, v2.4h, #8
; CHECK-SD-NEXT:    shl v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    shl v7.4h, v7.4h, #8
; CHECK-SD-NEXT:    shl v3.4h, v3.4h, #8
; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NEXT:    sshll v4.4s, v4.4h, #0
; CHECK-SD-NEXT:    sshr v5.4h, v5.4h, #8
; CHECK-SD-NEXT:    sshr v6.4h, v6.4h, #8
; CHECK-SD-NEXT:    sshr v2.4h, v2.4h, #8
; CHECK-SD-NEXT:    sshr v1.4h, v1.4h, #8
; CHECK-SD-NEXT:    sshr v7.4h, v7.4h, #8
; CHECK-SD-NEXT:    sshr v3.4h, v3.4h, #8
; CHECK-SD-NEXT:    sshll v16.4s, v0.4h, #0
; CHECK-SD-NEXT:    scvtf v0.4s, v4.4s
; CHECK-SD-NEXT:    sshll v5.4s, v5.4h, #0
; CHECK-SD-NEXT:    sshll v6.4s, v6.4h, #0
; CHECK-SD-NEXT:    sshll v17.4s, v2.4h, #0
; CHECK-SD-NEXT:    sshll v18.4s, v1.4h, #0
; CHECK-SD-NEXT:    sshll v7.4s, v7.4h, #0
; CHECK-SD-NEXT:    sshll v19.4s, v3.4h, #0
; CHECK-SD-NEXT:    scvtf v1.4s, v16.4s
; CHECK-SD-NEXT:    scvtf v4.4s, v5.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v6.4s
; CHECK-SD-NEXT:    scvtf v3.4s, v17.4s
; CHECK-SD-NEXT:    scvtf v5.4s, v18.4s
; CHECK-SD-NEXT:    scvtf v6.4s, v7.4s
; CHECK-SD-NEXT:    scvtf v7.4s, v19.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i8_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-GI-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    sshll v3.8h, v1.8b, #0
; CHECK-GI-NEXT:    sshll2 v1.8h, v1.16b, #0
; CHECK-GI-NEXT:    sshll v4.4s, v2.4h, #0
; CHECK-GI-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    sshll v5.4s, v0.4h, #0
; CHECK-GI-NEXT:    sshll2 v6.4s, v0.8h, #0
; CHECK-GI-NEXT:    sshll v7.4s, v3.4h, #0
; CHECK-GI-NEXT:    sshll2 v16.4s, v3.8h, #0
; CHECK-GI-NEXT:    sshll v17.4s, v1.4h, #0
; CHECK-GI-NEXT:    sshll2 v18.4s, v1.8h, #0
; CHECK-GI-NEXT:    scvtf v0.4s, v4.4s
; CHECK-GI-NEXT:    scvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    scvtf v2.4s, v5.4s
; CHECK-GI-NEXT:    scvtf v3.4s, v6.4s
; CHECK-GI-NEXT:    scvtf v4.4s, v7.4s
; CHECK-GI-NEXT:    scvtf v5.4s, v16.4s
; CHECK-GI-NEXT:    scvtf v6.4s, v17.4s
; CHECK-GI-NEXT:    scvtf v7.4s, v18.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i8> %a to <32 x float>
  ret <32 x float> %c
}

define <32 x float> @utofp_v32i8_v32f32(<32 x i8> %a) {
; CHECK-SD-LABEL: utofp_v32i8_v32f32:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
; CHECK-SD-NEXT:    ext v3.16b, v1.16b, v1.16b, #8
; CHECK-SD-NEXT:    zip1 v4.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v0.8b, v0.8b, v0.8b
; CHECK-SD-NEXT:    bic v4.4h, #255, lsl #8
; CHECK-SD-NEXT:    zip1 v5.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v6.8b, v2.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v2.8b, v2.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v1.8b, v1.8b, v0.8b
; CHECK-SD-NEXT:    zip1 v7.8b, v3.8b, v0.8b
; CHECK-SD-NEXT:    zip2 v3.8b, v3.8b, v0.8b
; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v4.4s, v4.4h, #0
; CHECK-SD-NEXT:    bic v5.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v6.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v2.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v1.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v7.4h, #255, lsl #8
; CHECK-SD-NEXT:    bic v3.4h, #255, lsl #8
; CHECK-SD-NEXT:    ushll v16.4s, v0.4h, #0
; CHECK-SD-NEXT:    ucvtf v0.4s, v4.4s
; CHECK-SD-NEXT:    ushll v5.4s, v5.4h, #0
; CHECK-SD-NEXT:    ushll v6.4s, v6.4h, #0
; CHECK-SD-NEXT:    ushll v17.4s, v2.4h, #0
; CHECK-SD-NEXT:    ushll v18.4s, v1.4h, #0
; CHECK-SD-NEXT:    ushll v7.4s, v7.4h, #0
; CHECK-SD-NEXT:    ushll v19.4s, v3.4h, #0
; CHECK-SD-NEXT:    ucvtf v1.4s, v16.4s
; CHECK-SD-NEXT:    ucvtf v4.4s, v5.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v6.4s
; CHECK-SD-NEXT:    ucvtf v3.4s, v17.4s
; CHECK-SD-NEXT:    ucvtf v5.4s, v18.4s
; CHECK-SD-NEXT:    ucvtf v6.4s, v7.4s
; CHECK-SD-NEXT:    ucvtf v7.4s, v19.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i8_v32f32:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-GI-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NEXT:    ushll v3.8h, v1.8b, #0
; CHECK-GI-NEXT:    ushll2 v1.8h, v1.16b, #0
; CHECK-GI-NEXT:    ushll v4.4s, v2.4h, #0
; CHECK-GI-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NEXT:    ushll v5.4s, v0.4h, #0
; CHECK-GI-NEXT:    ushll2 v6.4s, v0.8h, #0
; CHECK-GI-NEXT:    ushll v7.4s, v3.4h, #0
; CHECK-GI-NEXT:    ushll2 v16.4s, v3.8h, #0
; CHECK-GI-NEXT:    ushll v17.4s, v1.4h, #0
; CHECK-GI-NEXT:    ushll2 v18.4s, v1.8h, #0
; CHECK-GI-NEXT:    ucvtf v0.4s, v4.4s
; CHECK-GI-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-GI-NEXT:    ucvtf v2.4s, v5.4s
; CHECK-GI-NEXT:    ucvtf v3.4s, v6.4s
; CHECK-GI-NEXT:    ucvtf v4.4s, v7.4s
; CHECK-GI-NEXT:    ucvtf v5.4s, v16.4s
; CHECK-GI-NEXT:    ucvtf v6.4s, v17.4s
; CHECK-GI-NEXT:    ucvtf v7.4s, v18.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i8> %a to <32 x float>
  ret <32 x float> %c
}

define <2 x half> @stofp_v2i128_v2f16(<2 x i128> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v2i128_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sub sp, sp, #48
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NOFP16-NEXT:    mov x19, x1
; CHECK-SD-NOFP16-NEXT:    mov x20, x0
; CHECK-SD-NOFP16-NEXT:    mov x0, x2
; CHECK-SD-NOFP16-NEXT:    mov x1, x3
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x20
; CHECK-SD-NOFP16-NEXT:    mov x1, x19
; CHECK-SD-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    add sp, sp, #48
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v2i128_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sub sp, sp, #48
; CHECK-SD-FP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -32
; CHECK-SD-FP16-NEXT:    mov x19, x1
; CHECK-SD-FP16-NEXT:    mov x20, x0
; CHECK-SD-FP16-NEXT:    mov x0, x2
; CHECK-SD-FP16-NEXT:    mov x1, x3
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    mov x0, x20
; CHECK-SD-FP16-NEXT:    mov x1, x19
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    add sp, sp, #48
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v2i128_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sub sp, sp, #48
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NOFP16-NEXT:    mov x19, x2
; CHECK-GI-NOFP16-NEXT:    mov x20, x3
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x19
; CHECK-GI-NOFP16-NEXT:    mov x1, x20
; CHECK-GI-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    fcvt h1, s0
; CHECK-GI-NOFP16-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-GI-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NOFP16-NEXT:    add sp, sp, #48
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v2i128_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sub sp, sp, #48
; CHECK-GI-FP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-FP16-NEXT:    .cfi_offset w30, -32
; CHECK-GI-FP16-NEXT:    mov x19, x2
; CHECK-GI-FP16-NEXT:    mov x20, x3
; CHECK-GI-FP16-NEXT:    bl __floattihf
; CHECK-GI-FP16-NEXT:    mov x0, x19
; CHECK-GI-FP16-NEXT:    mov x1, x20
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floattihf
; CHECK-GI-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-FP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-GI-FP16-NEXT:    fmov d0, d1
; CHECK-GI-FP16-NEXT:    add sp, sp, #48
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <2 x i128> %a to <2 x half>
  ret <2 x half> %c
}

define <2 x half> @utofp_v2i128_v2f16(<2 x i128> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v2i128_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sub sp, sp, #48
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -32
; CHECK-SD-NOFP16-NEXT:    mov x19, x1
; CHECK-SD-NOFP16-NEXT:    mov x20, x0
; CHECK-SD-NOFP16-NEXT:    mov x0, x2
; CHECK-SD-NOFP16-NEXT:    mov x1, x3
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x20
; CHECK-SD-NOFP16-NEXT:    mov x1, x19
; CHECK-SD-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    add sp, sp, #48
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v2i128_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sub sp, sp, #48
; CHECK-SD-FP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-SD-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -32
; CHECK-SD-FP16-NEXT:    mov x19, x1
; CHECK-SD-FP16-NEXT:    mov x20, x0
; CHECK-SD-FP16-NEXT:    mov x0, x2
; CHECK-SD-FP16-NEXT:    mov x1, x3
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    mov x0, x20
; CHECK-SD-FP16-NEXT:    mov x1, x19
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    add sp, sp, #48
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v2i128_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sub sp, sp, #48
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -32
; CHECK-GI-NOFP16-NEXT:    mov x19, x2
; CHECK-GI-NOFP16-NEXT:    mov x20, x3
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x19
; CHECK-GI-NOFP16-NEXT:    mov x1, x20
; CHECK-GI-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    fcvt h1, s0
; CHECK-GI-NOFP16-NEXT:    ldr q0, [sp] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-GI-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NOFP16-NEXT:    add sp, sp, #48
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v2i128_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sub sp, sp, #48
; CHECK-GI-FP16-NEXT:    str x30, [sp, #16] // 8-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x20, x19, [sp, #32] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    .cfi_def_cfa_offset 48
; CHECK-GI-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-FP16-NEXT:    .cfi_offset w30, -32
; CHECK-GI-FP16-NEXT:    mov x19, x2
; CHECK-GI-FP16-NEXT:    mov x20, x3
; CHECK-GI-FP16-NEXT:    bl __floatuntihf
; CHECK-GI-FP16-NEXT:    mov x0, x19
; CHECK-GI-FP16-NEXT:    mov x1, x20
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floatuntihf
; CHECK-GI-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    ldp x20, x19, [sp, #32] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldr x30, [sp, #16] // 8-byte Folded Reload
; CHECK-GI-FP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-GI-FP16-NEXT:    fmov d0, d1
; CHECK-GI-FP16-NEXT:    add sp, sp, #48
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <2 x i128> %a to <2 x half>
  ret <2 x half> %c
}

define <3 x half> @stofp_v3i128_v3f16(<3 x i128> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v3i128_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sub sp, sp, #80
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NOFP16-NEXT:    mov x21, x1
; CHECK-SD-NOFP16-NEXT:    mov x22, x0
; CHECK-SD-NOFP16-NEXT:    mov x0, x4
; CHECK-SD-NOFP16-NEXT:    mov x1, x5
; CHECK-SD-NOFP16-NEXT:    mov x19, x3
; CHECK-SD-NOFP16-NEXT:    mov x20, x2
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x22
; CHECK-SD-NOFP16-NEXT:    mov x1, x21
; CHECK-SD-NOFP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x20
; CHECK-SD-NOFP16-NEXT:    mov x1, x19
; CHECK-SD-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floattisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-SD-NOFP16-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-SD-NOFP16-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    add sp, sp, #80
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v3i128_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sub sp, sp, #80
; CHECK-SD-FP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-FP16-NEXT:    .cfi_offset w21, -24
; CHECK-SD-FP16-NEXT:    .cfi_offset w22, -32
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -48
; CHECK-SD-FP16-NEXT:    mov x21, x1
; CHECK-SD-FP16-NEXT:    mov x22, x0
; CHECK-SD-FP16-NEXT:    mov x0, x4
; CHECK-SD-FP16-NEXT:    mov x1, x5
; CHECK-SD-FP16-NEXT:    mov x19, x3
; CHECK-SD-FP16-NEXT:    mov x20, x2
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    mov x0, x22
; CHECK-SD-FP16-NEXT:    mov x1, x21
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    mov x0, x20
; CHECK-SD-FP16-NEXT:    mov x1, x19
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floattihf
; CHECK-SD-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-SD-FP16-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-SD-FP16-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    add sp, sp, #80
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v3i128_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sub sp, sp, #80
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NOFP16-NEXT:    mov x19, x2
; CHECK-GI-NOFP16-NEXT:    mov x20, x3
; CHECK-GI-NOFP16-NEXT:    mov x21, x4
; CHECK-GI-NOFP16-NEXT:    mov x22, x5
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x19
; CHECK-GI-NOFP16-NEXT:    mov x1, x20
; CHECK-GI-NOFP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x21
; CHECK-GI-NOFP16-NEXT:    mov x1, x22
; CHECK-GI-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floattisf
; CHECK-GI-NOFP16-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NOFP16-NEXT:    add sp, sp, #80
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v3i128_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sub sp, sp, #80
; CHECK-GI-FP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-FP16-NEXT:    .cfi_offset w21, -24
; CHECK-GI-FP16-NEXT:    .cfi_offset w22, -32
; CHECK-GI-FP16-NEXT:    .cfi_offset w30, -48
; CHECK-GI-FP16-NEXT:    mov x19, x2
; CHECK-GI-FP16-NEXT:    mov x20, x3
; CHECK-GI-FP16-NEXT:    mov x21, x4
; CHECK-GI-FP16-NEXT:    mov x22, x5
; CHECK-GI-FP16-NEXT:    bl __floattihf
; CHECK-GI-FP16-NEXT:    mov x0, x19
; CHECK-GI-FP16-NEXT:    mov x1, x20
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floattihf
; CHECK-GI-FP16-NEXT:    mov x0, x21
; CHECK-GI-FP16-NEXT:    mov x1, x22
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floattihf
; CHECK-GI-FP16-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    mov v1.h[1], v2.h[0]
; CHECK-GI-FP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-GI-FP16-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-FP16-NEXT:    add sp, sp, #80
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <3 x i128> %a to <3 x half>
  ret <3 x half> %c
}

define <3 x half> @utofp_v3i128_v3f16(<3 x i128> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v3i128_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sub sp, sp, #80
; CHECK-SD-NOFP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w21, -24
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w22, -32
; CHECK-SD-NOFP16-NEXT:    .cfi_offset w30, -48
; CHECK-SD-NOFP16-NEXT:    mov x21, x1
; CHECK-SD-NOFP16-NEXT:    mov x22, x0
; CHECK-SD-NOFP16-NEXT:    mov x0, x4
; CHECK-SD-NOFP16-NEXT:    mov x1, x5
; CHECK-SD-NOFP16-NEXT:    mov x19, x3
; CHECK-SD-NOFP16-NEXT:    mov x20, x2
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x22
; CHECK-SD-NOFP16-NEXT:    mov x1, x21
; CHECK-SD-NOFP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    mov x0, x20
; CHECK-SD-NOFP16-NEXT:    mov x1, x19
; CHECK-SD-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-NOFP16-NEXT:    bl __floatuntisf
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s0
; CHECK-SD-NOFP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-SD-NOFP16-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-NOFP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-SD-NOFP16-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    add sp, sp, #80
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v3i128_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sub sp, sp, #80
; CHECK-SD-FP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-SD-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-SD-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-SD-FP16-NEXT:    .cfi_offset w21, -24
; CHECK-SD-FP16-NEXT:    .cfi_offset w22, -32
; CHECK-SD-FP16-NEXT:    .cfi_offset w30, -48
; CHECK-SD-FP16-NEXT:    mov x21, x1
; CHECK-SD-FP16-NEXT:    mov x22, x0
; CHECK-SD-FP16-NEXT:    mov x0, x4
; CHECK-SD-FP16-NEXT:    mov x1, x5
; CHECK-SD-FP16-NEXT:    mov x19, x3
; CHECK-SD-FP16-NEXT:    mov x20, x2
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    mov x0, x22
; CHECK-SD-FP16-NEXT:    mov x1, x21
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    mov x0, x20
; CHECK-SD-FP16-NEXT:    mov x1, x19
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-SD-FP16-NEXT:    bl __floatuntihf
; CHECK-SD-FP16-NEXT:    ldr q1, [sp] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-SD-FP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v1.h[1], v0.h[0]
; CHECK-SD-FP16-NEXT:    ldr q0, [sp, #16] // 16-byte Folded Reload
; CHECK-SD-FP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-SD-FP16-NEXT:    mov v0.16b, v1.16b
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    add sp, sp, #80
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v3i128_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sub sp, sp, #80
; CHECK-GI-NOFP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w21, -24
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w22, -32
; CHECK-GI-NOFP16-NEXT:    .cfi_offset w30, -48
; CHECK-GI-NOFP16-NEXT:    mov x19, x2
; CHECK-GI-NOFP16-NEXT:    mov x20, x3
; CHECK-GI-NOFP16-NEXT:    mov x21, x4
; CHECK-GI-NOFP16-NEXT:    mov x22, x5
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x19
; CHECK-GI-NOFP16-NEXT:    mov x1, x20
; CHECK-GI-NOFP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    mov x0, x21
; CHECK-GI-NOFP16-NEXT:    mov x1, x22
; CHECK-GI-NOFP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-NOFP16-NEXT:    bl __floatuntisf
; CHECK-GI-NOFP16-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    fcvt h0, s0
; CHECK-GI-NOFP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-NOFP16-NEXT:    mov v1.h[1], v2.h[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-GI-NOFP16-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-NOFP16-NEXT:    add sp, sp, #80
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v3i128_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sub sp, sp, #80
; CHECK-GI-FP16-NEXT:    str x30, [sp, #32] // 8-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x22, x21, [sp, #48] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    stp x20, x19, [sp, #64] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    .cfi_def_cfa_offset 80
; CHECK-GI-FP16-NEXT:    .cfi_offset w19, -8
; CHECK-GI-FP16-NEXT:    .cfi_offset w20, -16
; CHECK-GI-FP16-NEXT:    .cfi_offset w21, -24
; CHECK-GI-FP16-NEXT:    .cfi_offset w22, -32
; CHECK-GI-FP16-NEXT:    .cfi_offset w30, -48
; CHECK-GI-FP16-NEXT:    mov x19, x2
; CHECK-GI-FP16-NEXT:    mov x20, x3
; CHECK-GI-FP16-NEXT:    mov x21, x4
; CHECK-GI-FP16-NEXT:    mov x22, x5
; CHECK-GI-FP16-NEXT:    bl __floatuntihf
; CHECK-GI-FP16-NEXT:    mov x0, x19
; CHECK-GI-FP16-NEXT:    mov x1, x20
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp, #16] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floatuntihf
; CHECK-GI-FP16-NEXT:    mov x0, x21
; CHECK-GI-FP16-NEXT:    mov x1, x22
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    str q0, [sp] // 16-byte Folded Spill
; CHECK-GI-FP16-NEXT:    bl __floatuntihf
; CHECK-GI-FP16-NEXT:    ldp q2, q1, [sp] // 32-byte Folded Reload
; CHECK-GI-FP16-NEXT:    // kill: def $h0 killed $h0 def $q0
; CHECK-GI-FP16-NEXT:    ldr x30, [sp, #32] // 8-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldp x20, x19, [sp, #64] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    ldp x22, x21, [sp, #48] // 16-byte Folded Reload
; CHECK-GI-FP16-NEXT:    mov v1.h[1], v2.h[0]
; CHECK-GI-FP16-NEXT:    mov v1.h[2], v0.h[0]
; CHECK-GI-FP16-NEXT:    mov v0.16b, v1.16b
; CHECK-GI-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-GI-FP16-NEXT:    add sp, sp, #80
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <3 x i128> %a to <3 x half>
  ret <3 x half> %c
}

define <2 x half> @stofp_v2i64_v2f16(<2 x i64> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v2i64_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NOFP16-NEXT:    fmov x9, d0
; CHECK-SD-NOFP16-NEXT:    scvtf s1, x9
; CHECK-SD-NOFP16-NEXT:    scvtf s0, x8
; CHECK-SD-NOFP16-NEXT:    fcvt h2, s0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v2i64_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    mov x8, v0.d[1]
; CHECK-SD-FP16-NEXT:    fmov x9, d0
; CHECK-SD-FP16-NEXT:    scvtf h0, x9
; CHECK-SD-FP16-NEXT:    scvtf h1, x8
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i64_v2f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i64> %a to <2 x half>
  ret <2 x half> %c
}

define <2 x half> @utofp_v2i64_v2f16(<2 x i64> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v2i64_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    mov x8, v0.d[1]
; CHECK-SD-NOFP16-NEXT:    fmov x9, d0
; CHECK-SD-NOFP16-NEXT:    ucvtf s1, x9
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, x8
; CHECK-SD-NOFP16-NEXT:    fcvt h2, s0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v2i64_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    mov x8, v0.d[1]
; CHECK-SD-FP16-NEXT:    fmov x9, d0
; CHECK-SD-FP16-NEXT:    ucvtf h0, x9
; CHECK-SD-FP16-NEXT:    ucvtf h1, x8
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i64_v2f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i64> %a to <2 x half>
  ret <2 x half> %c
}

define <3 x half> @stofp_v3i64_v3f16(<3 x i64> %a) {
; CHECK-LABEL: stofp_v3i64_v3f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-NEXT:    scvtf v1.2d, v2.2d
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <3 x i64> %a to <3 x half>
  ret <3 x half> %c
}

define <3 x half> @utofp_v3i64_v3f16(<3 x i64> %a) {
; CHECK-LABEL: utofp_v3i64_v3f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-NEXT:    // kill: def $d2 killed $d2 def $q2
; CHECK-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-NEXT:    ucvtf v1.2d, v2.2d
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <3 x i64> %a to <3 x half>
  ret <3 x half> %c
}

define <4 x half> @stofp_v4i64_v4f16(<4 x i64> %a) {
; CHECK-LABEL: stofp_v4i64_v4f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.2d, v0.2d
; CHECK-NEXT:    scvtf v1.2d, v1.2d
; CHECK-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <4 x i64> %a to <4 x half>
  ret <4 x half> %c
}

define <4 x half> @utofp_v4i64_v4f16(<4 x i64> %a) {
; CHECK-LABEL: utofp_v4i64_v4f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <4 x i64> %a to <4 x half>
  ret <4 x half> %c
}

define <8 x half> @stofp_v8i64_v8f16(<8 x i64> %a) {
; CHECK-SD-LABEL: stofp_v8i64_v8f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-SD-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v8i64_v8f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <8 x i64> %a to <8 x half>
  ret <8 x half> %c
}

define <8 x half> @utofp_v8i64_v8f16(<8 x i64> %a) {
; CHECK-SD-LABEL: utofp_v8i64_v8f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-SD-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v8i64_v8f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <8 x i64> %a to <8 x half>
  ret <8 x half> %c
}

define <16 x half> @stofp_v16i64_v16f16(<16 x i64> %a) {
; CHECK-SD-LABEL: stofp_v16i64_v16f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-SD-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-SD-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-SD-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v7.2d
; CHECK-SD-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-SD-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn2 v6.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i64_v16f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-GI-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-GI-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn2 v6.4s, v7.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i64> %a to <16 x half>
  ret <16 x half> %c
}

define <16 x half> @utofp_v16i64_v16f16(<16 x i64> %a) {
; CHECK-SD-LABEL: utofp_v16i64_v16f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-SD-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-SD-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-SD-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v7.2d
; CHECK-SD-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-SD-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn2 v6.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i64_v16f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-GI-NEXT:    fcvtn v2.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-GI-NEXT:    fcvtn2 v2.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn2 v6.4s, v7.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i64> %a to <16 x half>
  ret <16 x half> %c
}

define <32 x half> @stofp_v32i64_v32f16(<32 x i64> %a) {
; CHECK-SD-LABEL: stofp_v32i64_v32f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ldp q17, q16, [sp]
; CHECK-SD-NEXT:    scvtf v18.2d, v0.2d
; CHECK-SD-NEXT:    ldp q19, q0, [sp, #64]
; CHECK-SD-NEXT:    scvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ldp q21, q20, [sp, #96]
; CHECK-SD-NEXT:    scvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    scvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    ldp q23, q22, [sp, #32]
; CHECK-SD-NEXT:    scvtf v19.2d, v19.2d
; CHECK-SD-NEXT:    fcvtn v18.2s, v18.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    scvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-SD-NEXT:    scvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    scvtf v23.2d, v23.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    scvtf v21.2d, v21.2d
; CHECK-SD-NEXT:    fcvtn v17.2s, v17.2d
; CHECK-SD-NEXT:    scvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    fcvtn v24.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn v19.2s, v19.2d
; CHECK-SD-NEXT:    scvtf v2.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn2 v18.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-SD-NEXT:    scvtf v1.2d, v7.2d
; CHECK-SD-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-SD-NEXT:    fcvtn v5.2s, v23.2d
; CHECK-SD-NEXT:    scvtf v3.2d, v22.2d
; CHECK-SD-NEXT:    fcvtn v7.2s, v21.2d
; CHECK-SD-NEXT:    fcvtn2 v17.4s, v16.2d
; CHECK-SD-NEXT:    scvtf v16.2d, v20.2d
; CHECK-SD-NEXT:    fcvtn2 v19.4s, v0.2d
; CHECK-SD-NEXT:    fcvtn2 v24.4s, v2.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v18.4s
; CHECK-SD-NEXT:    fcvtn2 v6.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v5.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v2.4h, v17.4s
; CHECK-SD-NEXT:    fcvtn2 v7.4s, v16.2d
; CHECK-SD-NEXT:    fcvtn v3.4h, v19.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v24.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-SD-NEXT:    fcvtn2 v2.8h, v5.4s
; CHECK-SD-NEXT:    fcvtn2 v3.8h, v7.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v32i64_v32f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ldp q16, q17, [sp]
; CHECK-GI-NEXT:    scvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ldp q20, q21, [sp, #64]
; CHECK-GI-NEXT:    scvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    ldp q18, q19, [sp, #32]
; CHECK-GI-NEXT:    scvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    scvtf v16.2d, v16.2d
; CHECK-GI-NEXT:    ldp q22, q23, [sp, #96]
; CHECK-GI-NEXT:    scvtf v20.2d, v20.2d
; CHECK-GI-NEXT:    scvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    scvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    scvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    scvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    scvtf v18.2d, v18.2d
; CHECK-GI-NEXT:    scvtf v21.2d, v21.2d
; CHECK-GI-NEXT:    scvtf v22.2d, v22.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-GI-NEXT:    fcvtn v16.2s, v16.2d
; CHECK-GI-NEXT:    scvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    fcvtn v20.2s, v20.2d
; CHECK-GI-NEXT:    scvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    scvtf v19.2d, v19.2d
; CHECK-GI-NEXT:    scvtf v23.2d, v23.2d
; CHECK-GI-NEXT:    fcvtn v24.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-GI-NEXT:    fcvtn v18.2s, v18.2d
; CHECK-GI-NEXT:    fcvtn v22.2s, v22.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-GI-NEXT:    fcvtn2 v16.4s, v17.2d
; CHECK-GI-NEXT:    fcvtn2 v20.4s, v21.2d
; CHECK-GI-NEXT:    fcvtn2 v24.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn2 v6.4s, v7.2d
; CHECK-GI-NEXT:    fcvtn2 v18.4s, v19.2d
; CHECK-GI-NEXT:    fcvtn2 v22.4s, v23.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-GI-NEXT:    fcvtn v2.4h, v16.4s
; CHECK-GI-NEXT:    fcvtn v3.4h, v20.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v24.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-GI-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NEXT:    fcvtn2 v3.8h, v22.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <32 x i64> %a to <32 x half>
  ret <32 x half> %c
}

define <32 x half> @utofp_v32i64_v32f16(<32 x i64> %a) {
; CHECK-SD-LABEL: utofp_v32i64_v32f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ldp q17, q16, [sp]
; CHECK-SD-NEXT:    ucvtf v18.2d, v0.2d
; CHECK-SD-NEXT:    ldp q19, q0, [sp, #64]
; CHECK-SD-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-SD-NEXT:    ldp q21, q20, [sp, #96]
; CHECK-SD-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-SD-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-SD-NEXT:    ldp q23, q22, [sp, #32]
; CHECK-SD-NEXT:    ucvtf v19.2d, v19.2d
; CHECK-SD-NEXT:    fcvtn v18.2s, v18.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-SD-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-SD-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-SD-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-SD-NEXT:    ucvtf v23.2d, v23.2d
; CHECK-SD-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-SD-NEXT:    ucvtf v21.2d, v21.2d
; CHECK-SD-NEXT:    fcvtn v17.2s, v17.2d
; CHECK-SD-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-SD-NEXT:    fcvtn v24.2s, v2.2d
; CHECK-SD-NEXT:    fcvtn v19.2s, v19.2d
; CHECK-SD-NEXT:    ucvtf v2.2d, v3.2d
; CHECK-SD-NEXT:    fcvtn2 v18.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-SD-NEXT:    ucvtf v1.2d, v7.2d
; CHECK-SD-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-SD-NEXT:    fcvtn v5.2s, v23.2d
; CHECK-SD-NEXT:    ucvtf v3.2d, v22.2d
; CHECK-SD-NEXT:    fcvtn v7.2s, v21.2d
; CHECK-SD-NEXT:    fcvtn2 v17.4s, v16.2d
; CHECK-SD-NEXT:    ucvtf v16.2d, v20.2d
; CHECK-SD-NEXT:    fcvtn2 v19.4s, v0.2d
; CHECK-SD-NEXT:    fcvtn2 v24.4s, v2.2d
; CHECK-SD-NEXT:    fcvtn v0.4h, v18.4s
; CHECK-SD-NEXT:    fcvtn2 v6.4s, v1.2d
; CHECK-SD-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v5.4s, v3.2d
; CHECK-SD-NEXT:    fcvtn v2.4h, v17.4s
; CHECK-SD-NEXT:    fcvtn2 v7.4s, v16.2d
; CHECK-SD-NEXT:    fcvtn v3.4h, v19.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v24.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-SD-NEXT:    fcvtn2 v2.8h, v5.4s
; CHECK-SD-NEXT:    fcvtn2 v3.8h, v7.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v32i64_v32f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ldp q16, q17, [sp]
; CHECK-GI-NEXT:    ucvtf v0.2d, v0.2d
; CHECK-GI-NEXT:    ldp q20, q21, [sp, #64]
; CHECK-GI-NEXT:    ucvtf v4.2d, v4.2d
; CHECK-GI-NEXT:    ldp q18, q19, [sp, #32]
; CHECK-GI-NEXT:    ucvtf v1.2d, v1.2d
; CHECK-GI-NEXT:    ucvtf v16.2d, v16.2d
; CHECK-GI-NEXT:    ldp q22, q23, [sp, #96]
; CHECK-GI-NEXT:    ucvtf v20.2d, v20.2d
; CHECK-GI-NEXT:    ucvtf v2.2d, v2.2d
; CHECK-GI-NEXT:    ucvtf v5.2d, v5.2d
; CHECK-GI-NEXT:    ucvtf v6.2d, v6.2d
; CHECK-GI-NEXT:    ucvtf v17.2d, v17.2d
; CHECK-GI-NEXT:    ucvtf v18.2d, v18.2d
; CHECK-GI-NEXT:    ucvtf v21.2d, v21.2d
; CHECK-GI-NEXT:    ucvtf v22.2d, v22.2d
; CHECK-GI-NEXT:    fcvtn v0.2s, v0.2d
; CHECK-GI-NEXT:    fcvtn v4.2s, v4.2d
; CHECK-GI-NEXT:    fcvtn v16.2s, v16.2d
; CHECK-GI-NEXT:    ucvtf v3.2d, v3.2d
; CHECK-GI-NEXT:    fcvtn v20.2s, v20.2d
; CHECK-GI-NEXT:    ucvtf v7.2d, v7.2d
; CHECK-GI-NEXT:    ucvtf v19.2d, v19.2d
; CHECK-GI-NEXT:    ucvtf v23.2d, v23.2d
; CHECK-GI-NEXT:    fcvtn v24.2s, v2.2d
; CHECK-GI-NEXT:    fcvtn v6.2s, v6.2d
; CHECK-GI-NEXT:    fcvtn v18.2s, v18.2d
; CHECK-GI-NEXT:    fcvtn v22.2s, v22.2d
; CHECK-GI-NEXT:    fcvtn2 v0.4s, v1.2d
; CHECK-GI-NEXT:    fcvtn2 v4.4s, v5.2d
; CHECK-GI-NEXT:    fcvtn2 v16.4s, v17.2d
; CHECK-GI-NEXT:    fcvtn2 v20.4s, v21.2d
; CHECK-GI-NEXT:    fcvtn2 v24.4s, v3.2d
; CHECK-GI-NEXT:    fcvtn2 v6.4s, v7.2d
; CHECK-GI-NEXT:    fcvtn2 v18.4s, v19.2d
; CHECK-GI-NEXT:    fcvtn2 v22.4s, v23.2d
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v4.4s
; CHECK-GI-NEXT:    fcvtn v2.4h, v16.4s
; CHECK-GI-NEXT:    fcvtn v3.4h, v20.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v24.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v6.4s
; CHECK-GI-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NEXT:    fcvtn2 v3.8h, v22.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <32 x i64> %a to <32 x half>
  ret <32 x half> %c
}

define <2 x half> @stofp_v2i32_v2f16(<2 x i32> %a) {
; CHECK-SD-LABEL: stofp_v2i32_v2f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v2i32_v2f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.2s, v0.2s
; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <2 x i32> %a to <2 x half>
  ret <2 x half> %c
}

define <2 x half> @utofp_v2i32_v2f16(<2 x i32> %a) {
; CHECK-SD-LABEL: utofp_v2i32_v2f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v2i32_v2f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <2 x i32> %a to <2 x half>
  ret <2 x half> %c
}

define <3 x half> @stofp_v3i32_v3f16(<3 x i32> %a) {
; CHECK-LABEL: stofp_v3i32_v3f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <3 x i32> %a to <3 x half>
  ret <3 x half> %c
}

define <3 x half> @utofp_v3i32_v3f16(<3 x i32> %a) {
; CHECK-LABEL: utofp_v3i32_v3f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <3 x i32> %a to <3 x half>
  ret <3 x half> %c
}

define <4 x half> @stofp_v4i32_v4f16(<4 x i32> %a) {
; CHECK-LABEL: stofp_v4i32_v4f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <4 x i32> %a to <4 x half>
  ret <4 x half> %c
}

define <4 x half> @utofp_v4i32_v4f16(<4 x i32> %a) {
; CHECK-LABEL: utofp_v4i32_v4f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <4 x i32> %a to <4 x half>
  ret <4 x half> %c
}

define <8 x half> @stofp_v8i32_v8f16(<8 x i32> %a) {
; CHECK-LABEL: stofp_v8i32_v8f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    scvtf v1.4s, v1.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <8 x i32> %a to <8 x half>
  ret <8 x half> %c
}

define <8 x half> @utofp_v8i32_v8f16(<8 x i32> %a) {
; CHECK-LABEL: utofp_v8i32_v8f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <8 x i32> %a to <8 x half>
  ret <8 x half> %c
}

define <16 x half> @stofp_v16i32_v16f16(<16 x i32> %a) {
; CHECK-SD-LABEL: stofp_v16i32_v16f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v2.4s
; CHECK-SD-NEXT:    scvtf v4.4s, v1.4s
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-SD-NEXT:    scvtf v2.4s, v3.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: stofp_v16i32_v16f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    scvtf v2.4s, v2.4s
; CHECK-GI-NEXT:    scvtf v4.4s, v1.4s
; CHECK-GI-NEXT:    scvtf v3.4s, v3.4s
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v3.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = sitofp <16 x i32> %a to <16 x half>
  ret <16 x half> %c
}

define <16 x half> @utofp_v16i32_v16f16(<16 x i32> %a) {
; CHECK-SD-LABEL: utofp_v16i32_v16f16:
; CHECK-SD:       // %bb.0: // %entry
; CHECK-SD-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-SD-NEXT:    ucvtf v4.4s, v1.4s
; CHECK-SD-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-SD-NEXT:    ucvtf v2.4s, v3.4s
; CHECK-SD-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-SD-NEXT:    fcvtn2 v1.8h, v2.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: utofp_v16i32_v16f16:
; CHECK-GI:       // %bb.0: // %entry
; CHECK-GI-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-GI-NEXT:    ucvtf v4.4s, v1.4s
; CHECK-GI-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-GI-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-GI-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NEXT:    fcvtn2 v1.8h, v3.4s
; CHECK-GI-NEXT:    ret
entry:
  %c = uitofp <16 x i32> %a to <16 x half>
  ret <16 x half> %c
}

define <32 x half> @stofp_v32i32_v32f16(<32 x i32> %a) {
; CHECK-LABEL: stofp_v32i32_v32f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    scvtf v0.4s, v0.4s
; CHECK-NEXT:    scvtf v2.4s, v2.4s
; CHECK-NEXT:    scvtf v4.4s, v4.4s
; CHECK-NEXT:    scvtf v6.4s, v6.4s
; CHECK-NEXT:    scvtf v16.4s, v1.4s
; CHECK-NEXT:    scvtf v17.4s, v3.4s
; CHECK-NEXT:    scvtf v5.4s, v5.4s
; CHECK-NEXT:    scvtf v7.4s, v7.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-NEXT:    fcvtn v2.4h, v4.4s
; CHECK-NEXT:    fcvtn v3.4h, v6.4s
; CHECK-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-NEXT:    fcvtn2 v2.8h, v5.4s
; CHECK-NEXT:    fcvtn2 v3.8h, v7.4s
; CHECK-NEXT:    ret
entry:
  %c = sitofp <32 x i32> %a to <32 x half>
  ret <32 x half> %c
}

define <32 x half> @utofp_v32i32_v32f16(<32 x i32> %a) {
; CHECK-LABEL: utofp_v32i32_v32f16:
; CHECK:       // %bb.0: // %entry
; CHECK-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-NEXT:    ucvtf v16.4s, v1.4s
; CHECK-NEXT:    ucvtf v17.4s, v3.4s
; CHECK-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-NEXT:    fcvtn v2.4h, v4.4s
; CHECK-NEXT:    fcvtn v3.4h, v6.4s
; CHECK-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-NEXT:    fcvtn2 v2.8h, v5.4s
; CHECK-NEXT:    fcvtn2 v3.8h, v7.4s
; CHECK-NEXT:    ret
entry:
  %c = uitofp <32 x i32> %a to <32 x half>
  ret <32 x half> %c
}

define <2 x half> @stofp_v2i16_v2f16(<2 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v2i16_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-SD-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v2i16_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v2i16_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-GI-NOFP16-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-GI-NOFP16-NEXT:    scvtf v0.2s, v0.2s
; CHECK-GI-NOFP16-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v2i16_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <2 x i16> %a to <2 x half>
  ret <2 x half> %c
}

define <2 x half> @utofp_v2i16_v2f16(<2 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v2i16_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-SD-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v2i16_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v2i16_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    movi d1, #0x00ffff0000ffff
; CHECK-GI-NOFP16-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-GI-NOFP16-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v2i16_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <2 x i16> %a to <2 x half>
  ret <2 x half> %c
}

define <3 x half> @stofp_v3i16_v3f16(<3 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v3i16_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v3i16_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v3i16_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v3i16_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <3 x i16> %a to <3 x half>
  ret <3 x half> %c
}

define <3 x half> @utofp_v3i16_v3f16(<3 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v3i16_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v3i16_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v3i16_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v3i16_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <3 x i16> %a to <3 x half>
  ret <3 x half> %c
}

define <4 x half> @stofp_v4i16_v4f16(<4 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v4i16_v4f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v4i16_v4f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v4i16_v4f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v4i16_v4f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <4 x i16> %a to <4 x half>
  ret <4 x half> %c
}

define <4 x half> @utofp_v4i16_v4f16(<4 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v4i16_v4f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v4i16_v4f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v4i16_v4f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v4i16_v4f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <4 x i16> %a to <4 x half>
  ret <4 x half> %c
}

define <8 x half> @stofp_v8i16_v8f16(<8 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v8i16_v8f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v2.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v1.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v1.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v8i16_v8f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v8i16_v8f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v8i16_v8f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <8 x i16> %a to <8 x half>
  ret <8 x half> %c
}

define <8 x half> @utofp_v8i16_v8f16(<8 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v8i16_v8f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v2.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v8i16_v8f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v8i16_v8f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v8i16_v8f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <8 x i16> %a to <8 x half>
  ret <8 x half> %c
}

define <16 x half> @stofp_v16i16_v16f16(<16 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v16i16_v16f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v3.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v4.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v5.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v2.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v3.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v2.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v3.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v16i16_v16f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    scvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v16i16_v16f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v2.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v3.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v16i16_v16f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    scvtf v1.8h, v1.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <16 x i16> %a to <16 x half>
  ret <16 x half> %c
}

define <16 x half> @utofp_v16i16_v16f16(<16 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v16i16_v16f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v3.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v4.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v5.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v2.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v3.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v16i16_v16f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ucvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v16i16_v16f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v2.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v3.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v4.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v5.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v16i16_v16f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ucvtf v1.8h, v1.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <16 x i16> %a to <16 x half>
  ret <16 x half> %c
}

define <32 x half> @stofp_v32i16_v32f16(<32 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v32i16_v32f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v4.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v5.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v6.4s, v2.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v7.4s, v3.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v3.4s, v3.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v4.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v5.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v6.4s, v6.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v7.4s, v7.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v16.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v17.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v18.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v19.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v32i16_v32f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    scvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    scvtf v2.8h, v2.8h
; CHECK-SD-FP16-NEXT:    scvtf v3.8h, v3.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v32i16_v32f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v4.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v5.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v6.4s, v2.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v7.4s, v3.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v4.4s, v4.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v5.4s, v5.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v6.4s, v6.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v7.4s, v7.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v16.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v17.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v18.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v19.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v32i16_v32f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    scvtf v1.8h, v1.8h
; CHECK-GI-FP16-NEXT:    scvtf v2.8h, v2.8h
; CHECK-GI-FP16-NEXT:    scvtf v3.8h, v3.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <32 x i16> %a to <32 x half>
  ret <32 x half> %c
}

define <32 x half> @utofp_v32i16_v32f16(<32 x i16> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v32i16_v32f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v4.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v5.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v6.4s, v2.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v7.4s, v3.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v3.4s, v3.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v16.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v17.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v18.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v19.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v32i16_v32f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ucvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    ucvtf v2.8h, v2.8h
; CHECK-SD-FP16-NEXT:    ucvtf v3.8h, v3.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v32i16_v32f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v4.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v5.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v6.4s, v2.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v7.4s, v3.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v3.4s, v3.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v16.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v17.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v18.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v19.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v32i16_v32f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ucvtf v1.8h, v1.8h
; CHECK-GI-FP16-NEXT:    ucvtf v2.8h, v2.8h
; CHECK-GI-FP16-NEXT:    ucvtf v3.8h, v3.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <32 x i16> %a to <32 x half>
  ret <32 x half> %c
}

define <2 x half> @stofp_v2i8_v2f16(<2 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v2i8_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT:    mov w8, v0.s[1]
; CHECK-SD-NOFP16-NEXT:    fmov w9, s0
; CHECK-SD-NOFP16-NEXT:    sxtb w9, w9
; CHECK-SD-NOFP16-NEXT:    sxtb w8, w8
; CHECK-SD-NOFP16-NEXT:    scvtf s1, w9
; CHECK-SD-NOFP16-NEXT:    scvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h2, s0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v2i8_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-FP16-NEXT:    mov w8, v0.s[1]
; CHECK-SD-FP16-NEXT:    fmov w9, s0
; CHECK-SD-FP16-NEXT:    sxtb w9, w9
; CHECK-SD-FP16-NEXT:    sxtb w8, w8
; CHECK-SD-FP16-NEXT:    scvtf h0, w9
; CHECK-SD-FP16-NEXT:    scvtf h1, w8
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v2i8_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    shl v0.2s, v0.2s, #24
; CHECK-GI-NOFP16-NEXT:    sshr v0.2s, v0.2s, #24
; CHECK-GI-NOFP16-NEXT:    scvtf v0.2s, v0.2s
; CHECK-GI-NOFP16-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v2i8_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-GI-FP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <2 x i8> %a to <2 x half>
  ret <2 x half> %c
}

define <2 x half> @utofp_v2i8_v2f16(<2 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v2i8_v2f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-NOFP16-NEXT:    mov w8, v0.s[1]
; CHECK-SD-NOFP16-NEXT:    fmov w9, s0
; CHECK-SD-NOFP16-NEXT:    and w9, w9, #0xff
; CHECK-SD-NOFP16-NEXT:    and w8, w8, #0xff
; CHECK-SD-NOFP16-NEXT:    ucvtf s1, w9
; CHECK-SD-NOFP16-NEXT:    ucvtf s0, w8
; CHECK-SD-NOFP16-NEXT:    fcvt h2, s0
; CHECK-SD-NOFP16-NEXT:    fcvt h0, s1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], v2.h[0]
; CHECK-SD-NOFP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v2i8_v2f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-SD-FP16-NEXT:    mov w8, v0.s[1]
; CHECK-SD-FP16-NEXT:    fmov w9, s0
; CHECK-SD-FP16-NEXT:    and w9, w9, #0xff
; CHECK-SD-FP16-NEXT:    and w8, w8, #0xff
; CHECK-SD-FP16-NEXT:    ucvtf h0, w9
; CHECK-SD-FP16-NEXT:    ucvtf h1, w8
; CHECK-SD-FP16-NEXT:    mov v0.h[1], v1.h[0]
; CHECK-SD-FP16-NEXT:    // kill: def $d0 killed $d0 killed $q0
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v2i8_v2f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-GI-NOFP16-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.2s, v0.2s
; CHECK-GI-NOFP16-NEXT:    mov v1.s[0], v0.s[0]
; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], v0.s[1]
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v2i8_v2f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    movi d1, #0x0000ff000000ff
; CHECK-GI-FP16-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-FP16-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <2 x i8> %a to <2 x half>
  ret <2 x half> %c
}

define <3 x half> @stofp_v3i8_v3f16(<3 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v3i8_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    fmov s0, w0
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], w1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[2], w2
; CHECK-SD-NOFP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NOFP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v3i8_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    fmov s0, w0
; CHECK-SD-FP16-NEXT:    mov v0.h[1], w1
; CHECK-SD-FP16-NEXT:    mov v0.h[2], w2
; CHECK-SD-FP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-FP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v3i8_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    fmov s0, w0
; CHECK-GI-NOFP16-NEXT:    mov v0.h[1], w1
; CHECK-GI-NOFP16-NEXT:    mov v0.h[2], w2
; CHECK-GI-NOFP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-GI-NOFP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-GI-NOFP16-NEXT:    mov v1.h[0], v0.h[2]
; CHECK-GI-NOFP16-NEXT:    mov v1.h[1], v0.h[3]
; CHECK-GI-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v1.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v3i8_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    fmov s0, w0
; CHECK-GI-FP16-NEXT:    mov v0.b[1], w1
; CHECK-GI-FP16-NEXT:    mov v0.b[2], w2
; CHECK-GI-FP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <3 x i8> %a to <3 x half>
  ret <3 x half> %c
}

define <3 x half> @utofp_v3i8_v3f16(<3 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v3i8_v3f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    fmov s0, w0
; CHECK-SD-NOFP16-NEXT:    mov v0.h[1], w1
; CHECK-SD-NOFP16-NEXT:    mov v0.h[2], w2
; CHECK-SD-NOFP16-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v3i8_v3f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    fmov s0, w0
; CHECK-SD-FP16-NEXT:    mov v0.h[1], w1
; CHECK-SD-FP16-NEXT:    mov v0.h[2], w2
; CHECK-SD-FP16-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v3i8_v3f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    fmov s0, w0
; CHECK-GI-NOFP16-NEXT:    movi d1, #0xff00ff00ff00ff
; CHECK-GI-NOFP16-NEXT:    mov v0.h[1], w1
; CHECK-GI-NOFP16-NEXT:    mov v0.h[2], w2
; CHECK-GI-NOFP16-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-NOFP16-NEXT:    mov v1.h[0], v0.h[2]
; CHECK-GI-NOFP16-NEXT:    mov v1.h[1], v0.h[3]
; CHECK-GI-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v1.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    mov v0.d[1], v1.d[0]
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v3i8_v3f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    fmov s0, w0
; CHECK-GI-FP16-NEXT:    mov v0.b[1], w1
; CHECK-GI-FP16-NEXT:    mov v0.b[2], w2
; CHECK-GI-FP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <3 x i8> %a to <3 x half>
  ret <3 x half> %c
}

define <4 x half> @stofp_v4i8_v4f16(<4 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v4i8_v4f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-NOFP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-NOFP16-NEXT:    sshll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v4i8_v4f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-SD-FP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-SD-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v4i8_v4f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    shl v0.4s, v0.4s, #24
; CHECK-GI-NOFP16-NEXT:    sshr v0.4s, v0.4s, #24
; CHECK-GI-NOFP16-NEXT:    scvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v4i8_v4f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    shl v0.4h, v0.4h, #8
; CHECK-GI-FP16-NEXT:    sshr v0.4h, v0.4h, #8
; CHECK-GI-FP16-NEXT:    scvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <4 x i8> %a to <4 x half>
  ret <4 x half> %c
}

define <4 x half> @utofp_v4i8_v4f16(<4 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v4i8_v4f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v4i8_v4f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    bic v0.4h, #255, lsl #8
; CHECK-SD-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v4i8_v4f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    movi v1.2d, #0x0000ff000000ff
; CHECK-GI-NOFP16-NEXT:    ushll v0.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    and v0.16b, v0.16b, v1.16b
; CHECK-GI-NOFP16-NEXT:    ucvtf v0.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v0.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v4i8_v4f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    movi d1, #0xff00ff00ff00ff
; CHECK-GI-FP16-NEXT:    and v0.8b, v0.8b, v1.8b
; CHECK-GI-FP16-NEXT:    ucvtf v0.4h, v0.4h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <4 x i8> %a to <4 x half>
  ret <4 x half> %c
}

define <8 x half> @stofp_v8i8_v8f16(<8 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v8i8_v8f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v2.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v1.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v1.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v8i8_v8f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v8i8_v8f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    sshll v1.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v8i8_v8f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <8 x i8> %a to <8 x half>
  ret <8 x half> %c
}

define <8 x half> @utofp_v8i8_v8f16(<8 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v8i8_v8f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v2.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v1.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v1.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v8i8_v8f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v8i8_v8f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    ushll v1.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v1.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v2.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v2.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v8i8_v8f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <8 x i8> %a to <8 x half>
  ret <8 x half> %c
}

define <16 x half> @stofp_v16i8_v16f16(<16 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v16i8_v16f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll2 v1.8h, v0.16b, #0
; CHECK-SD-NOFP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    sshll v2.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v3.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v4.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v5.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v2.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v3.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v2.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v3.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v3.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v16i8_v16f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sshll2 v1.8h, v0.16b, #0
; CHECK-SD-FP16-NEXT:    sshll v0.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    scvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v16i8_v16f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v1.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NOFP16-NEXT:    sshll v2.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v3.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v4.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v5.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v16i8_v16f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sshll v1.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    sshll2 v2.8h, v0.16b, #0
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v1.8h
; CHECK-GI-FP16-NEXT:    scvtf v1.8h, v2.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <16 x i8> %a to <16 x half>
  ret <16 x half> %c
}

define <16 x half> @utofp_v16i8_v16f16(<16 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v16i8_v16f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll2 v1.8h, v0.16b, #0
; CHECK-SD-NOFP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    ushll v2.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v3.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v4.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v5.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v2.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v3.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v2.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v3.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v16i8_v16f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ushll2 v1.8h, v0.16b, #0
; CHECK-SD-FP16-NEXT:    ushll v0.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    ucvtf v1.8h, v1.8h
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v0.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v16i8_v16f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v1.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NOFP16-NEXT:    ushll v2.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v3.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v2.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v3.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v4.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v5.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v2.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v3.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v5.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v16i8_v16f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ushll v1.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    ushll2 v2.8h, v0.16b, #0
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v1.8h
; CHECK-GI-FP16-NEXT:    ucvtf v1.8h, v2.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <16 x i8> %a to <16 x half>
  ret <16 x half> %c
}

define <32 x half> @stofp_v32i8_v32f16(<32 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: stofp_v32i8_v32f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v3.8h, v1.16b, #0
; CHECK-SD-NOFP16-NEXT:    sshll v1.8h, v1.8b, #0
; CHECK-SD-NOFP16-NEXT:    sshll v4.4s, v2.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v5.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v6.4s, v3.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll v7.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v3.4s, v3.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-SD-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    scvtf v4.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v5.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v6.4s, v6.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v7.4s, v7.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v16.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v17.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v18.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    scvtf v19.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v3.4h, v6.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v2.4h, v7.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v18.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v16.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v3.8h, v17.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v2.8h, v19.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: stofp_v32i8_v32f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    sshll2 v3.8h, v0.16b, #0
; CHECK-SD-FP16-NEXT:    sshll2 v4.8h, v1.16b, #0
; CHECK-SD-FP16-NEXT:    sshll v5.8h, v1.8b, #0
; CHECK-SD-FP16-NEXT:    scvtf v0.8h, v2.8h
; CHECK-SD-FP16-NEXT:    scvtf v1.8h, v3.8h
; CHECK-SD-FP16-NEXT:    scvtf v3.8h, v4.8h
; CHECK-SD-FP16-NEXT:    scvtf v2.8h, v5.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: stofp_v32i8_v32f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.8h, v0.16b, #0
; CHECK-GI-NOFP16-NEXT:    sshll v3.8h, v1.8b, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v1.8h, v1.16b, #0
; CHECK-GI-NOFP16-NEXT:    sshll v4.4s, v2.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v5.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v6.4s, v3.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll v7.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v2.4s, v2.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v3.4s, v3.8h, #0
; CHECK-GI-NOFP16-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    scvtf v4.4s, v4.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v5.4s, v5.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v6.4s, v6.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v7.4s, v7.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v16.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v17.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v18.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    scvtf v19.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: stofp_v32i8_v32f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    sshll v2.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    sshll2 v3.8h, v0.16b, #0
; CHECK-GI-FP16-NEXT:    sshll v4.8h, v1.8b, #0
; CHECK-GI-FP16-NEXT:    sshll2 v5.8h, v1.16b, #0
; CHECK-GI-FP16-NEXT:    scvtf v0.8h, v2.8h
; CHECK-GI-FP16-NEXT:    scvtf v1.8h, v3.8h
; CHECK-GI-FP16-NEXT:    scvtf v2.8h, v4.8h
; CHECK-GI-FP16-NEXT:    scvtf v3.8h, v5.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = sitofp <32 x i8> %a to <32 x half>
  ret <32 x half> %c
}

define <32 x half> @utofp_v32i8_v32f16(<32 x i8> %a) {
; CHECK-SD-NOFP16-LABEL: utofp_v32i8_v32f16:
; CHECK-SD-NOFP16:       // %bb.0: // %entry
; CHECK-SD-NOFP16-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v3.8h, v1.16b, #0
; CHECK-SD-NOFP16-NEXT:    ushll v1.8h, v1.8b, #0
; CHECK-SD-NOFP16-NEXT:    ushll v4.4s, v2.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v5.4s, v0.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v6.4s, v3.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll v7.4s, v1.4h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v3.4s, v3.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-SD-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-SD-NOFP16-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v16.4s, v0.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v17.4s, v3.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v18.4s, v2.4s
; CHECK-SD-NOFP16-NEXT:    ucvtf v19.4s, v1.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v3.4h, v6.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn v2.4h, v7.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v0.8h, v18.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v1.8h, v16.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v3.8h, v17.4s
; CHECK-SD-NOFP16-NEXT:    fcvtn2 v2.8h, v19.4s
; CHECK-SD-NOFP16-NEXT:    ret
;
; CHECK-SD-FP16-LABEL: utofp_v32i8_v32f16:
; CHECK-SD-FP16:       // %bb.0: // %entry
; CHECK-SD-FP16-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-SD-FP16-NEXT:    ushll2 v3.8h, v0.16b, #0
; CHECK-SD-FP16-NEXT:    ushll2 v4.8h, v1.16b, #0
; CHECK-SD-FP16-NEXT:    ushll v5.8h, v1.8b, #0
; CHECK-SD-FP16-NEXT:    ucvtf v0.8h, v2.8h
; CHECK-SD-FP16-NEXT:    ucvtf v1.8h, v3.8h
; CHECK-SD-FP16-NEXT:    ucvtf v3.8h, v4.8h
; CHECK-SD-FP16-NEXT:    ucvtf v2.8h, v5.8h
; CHECK-SD-FP16-NEXT:    ret
;
; CHECK-GI-NOFP16-LABEL: utofp_v32i8_v32f16:
; CHECK-GI-NOFP16:       // %bb.0: // %entry
; CHECK-GI-NOFP16-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.8h, v0.16b, #0
; CHECK-GI-NOFP16-NEXT:    ushll v3.8h, v1.8b, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v1.8h, v1.16b, #0
; CHECK-GI-NOFP16-NEXT:    ushll v4.4s, v2.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v5.4s, v0.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v6.4s, v3.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll v7.4s, v1.4h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v2.4s, v2.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v0.4s, v0.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v3.4s, v3.8h, #0
; CHECK-GI-NOFP16-NEXT:    ushll2 v1.4s, v1.8h, #0
; CHECK-GI-NOFP16-NEXT:    ucvtf v4.4s, v4.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v5.4s, v5.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v6.4s, v6.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v7.4s, v7.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v16.4s, v2.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v17.4s, v0.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v18.4s, v3.4s
; CHECK-GI-NOFP16-NEXT:    ucvtf v19.4s, v1.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v0.4h, v4.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v1.4h, v5.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v2.4h, v6.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn v3.4h, v7.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v0.8h, v16.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v1.8h, v17.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v2.8h, v18.4s
; CHECK-GI-NOFP16-NEXT:    fcvtn2 v3.8h, v19.4s
; CHECK-GI-NOFP16-NEXT:    ret
;
; CHECK-GI-FP16-LABEL: utofp_v32i8_v32f16:
; CHECK-GI-FP16:       // %bb.0: // %entry
; CHECK-GI-FP16-NEXT:    ushll v2.8h, v0.8b, #0
; CHECK-GI-FP16-NEXT:    ushll2 v3.8h, v0.16b, #0
; CHECK-GI-FP16-NEXT:    ushll v4.8h, v1.8b, #0
; CHECK-GI-FP16-NEXT:    ushll2 v5.8h, v1.16b, #0
; CHECK-GI-FP16-NEXT:    ucvtf v0.8h, v2.8h
; CHECK-GI-FP16-NEXT:    ucvtf v1.8h, v3.8h
; CHECK-GI-FP16-NEXT:    ucvtf v2.8h, v4.8h
; CHECK-GI-FP16-NEXT:    ucvtf v3.8h, v5.8h
; CHECK-GI-FP16-NEXT:    ret
entry:
  %c = uitofp <32 x i8> %a to <32 x half>
  ret <32 x half> %c
}