; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme2 -mattr=+sve -aarch64-disable-multivector-spill-fill -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=NOPAIR
; RUN: llc -mtriple=aarch64-linux-gnu -aarch64-streaming-hazard-size=0 -mattr=+sme2  -mattr=+sve -verify-machineinstrs < %s | FileCheck %s --check-prefixes=PAIR

declare void @my_func()
declare void @my_func2(<vscale x 16 x i8> %v)

define void @fbyte(<vscale x 16 x i8> %v) #0{
; NOPAIR-LABEL: fbyte:
; NOPAIR:       // %bb.0:
; NOPAIR-NEXT:    str x29, [sp, #-32]! // 8-byte Folded Spill
; NOPAIR-NEXT:    stp x30, x19, [sp, #16] // 16-byte Folded Spill
; NOPAIR-NEXT:    addvl sp, sp, #-18
; NOPAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str z23, [sp, #2, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z22, [sp, #3, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z21, [sp, #4, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z20, [sp, #5, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z19, [sp, #6, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z18, [sp, #7, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z17, [sp, #8, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z16, [sp, #9, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z15, [sp, #10, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z14, [sp, #11, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z13, [sp, #12, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z12, [sp, #13, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z11, [sp, #14, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z10, [sp, #15, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    addvl sp, sp, #-1
; NOPAIR-NEXT:    str z0, [sp] // 16-byte Folded Spill
; NOPAIR-NEXT:    mrs x19, SVCR
; NOPAIR-NEXT:    tbz w19, #0, .LBB0_2
; NOPAIR-NEXT:  // %bb.1:
; NOPAIR-NEXT:    smstop sm
; NOPAIR-NEXT:  .LBB0_2:
; NOPAIR-NEXT:    rdvl x8, #1
; NOPAIR-NEXT:    addsvl x8, x8, #-1
; NOPAIR-NEXT:    cbz x8, .LBB0_4
; NOPAIR-NEXT:  // %bb.3:
; NOPAIR-NEXT:    brk #0x1
; NOPAIR-NEXT:  .LBB0_4:
; NOPAIR-NEXT:    ldr z0, [sp] // 16-byte Folded Reload
; NOPAIR-NEXT:    bl my_func2
; NOPAIR-NEXT:    tbz w19, #0, .LBB0_6
; NOPAIR-NEXT:  // %bb.5:
; NOPAIR-NEXT:    smstart sm
; NOPAIR-NEXT:  .LBB0_6:
; NOPAIR-NEXT:    addvl sp, sp, #1
; NOPAIR-NEXT:    ldr z23, [sp, #2, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z22, [sp, #3, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z21, [sp, #4, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z20, [sp, #5, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z19, [sp, #6, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z18, [sp, #7, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z17, [sp, #8, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z16, [sp, #9, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z15, [sp, #10, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z14, [sp, #11, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z13, [sp, #12, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z12, [sp, #13, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z11, [sp, #14, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z10, [sp, #15, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    addvl sp, sp, #18
; NOPAIR-NEXT:    ldp x30, x19, [sp, #16] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr x29, [sp], #32 // 8-byte Folded Reload
; NOPAIR-NEXT:    ret
;
; PAIR-LABEL: fbyte:
; PAIR:       // %bb.0:
; PAIR-NEXT:    str x29, [sp, #-32]! // 8-byte Folded Spill
; PAIR-NEXT:    stp x30, x19, [sp, #16] // 16-byte Folded Spill
; PAIR-NEXT:    addvl sp, sp, #-18
; PAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; PAIR-NEXT:    str z23, [sp, #2, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z22, [sp, #3, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z21, [sp, #4, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z20, [sp, #5, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z19, [sp, #6, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z18, [sp, #7, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z17, [sp, #8, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z16, [sp, #9, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z15, [sp, #10, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z14, [sp, #11, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z13, [sp, #12, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z12, [sp, #13, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z11, [sp, #14, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z10, [sp, #15, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    addvl sp, sp, #-1
; PAIR-NEXT:    str z0, [sp] // 16-byte Folded Spill
; PAIR-NEXT:    mrs x19, SVCR
; PAIR-NEXT:    tbz w19, #0, .LBB0_2
; PAIR-NEXT:  // %bb.1:
; PAIR-NEXT:    smstop sm
; PAIR-NEXT:  .LBB0_2:
; PAIR-NEXT:    rdvl x8, #1
; PAIR-NEXT:    addsvl x8, x8, #-1
; PAIR-NEXT:    cbz x8, .LBB0_4
; PAIR-NEXT:  // %bb.3:
; PAIR-NEXT:    brk #0x1
; PAIR-NEXT:  .LBB0_4:
; PAIR-NEXT:    ldr z0, [sp] // 16-byte Folded Reload
; PAIR-NEXT:    bl my_func2
; PAIR-NEXT:    tbz w19, #0, .LBB0_6
; PAIR-NEXT:  // %bb.5:
; PAIR-NEXT:    smstart sm
; PAIR-NEXT:  .LBB0_6:
; PAIR-NEXT:    addvl sp, sp, #1
; PAIR-NEXT:    ldr z23, [sp, #2, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z22, [sp, #3, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z21, [sp, #4, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z20, [sp, #5, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z19, [sp, #6, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z18, [sp, #7, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z17, [sp, #8, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z16, [sp, #9, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z15, [sp, #10, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z14, [sp, #11, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z13, [sp, #12, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z12, [sp, #13, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z11, [sp, #14, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z10, [sp, #15, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; PAIR-NEXT:    addvl sp, sp, #18
; PAIR-NEXT:    ldp x30, x19, [sp, #16] // 16-byte Folded Reload
; PAIR-NEXT:    ldr x29, [sp], #32 // 8-byte Folded Reload
; PAIR-NEXT:    ret
  call void @my_func2(<vscale x 16 x i8> %v)
  ret void
}

define void @fhalf(<vscale x 8 x half> %v) #1{
; NOPAIR-LABEL: fhalf:
; NOPAIR:       // %bb.0:
; NOPAIR-NEXT:    stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
; NOPAIR-NEXT:    addvl sp, sp, #-18
; NOPAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str z23, [sp, #2, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z22, [sp, #3, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z21, [sp, #4, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z20, [sp, #5, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z19, [sp, #6, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z18, [sp, #7, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z17, [sp, #8, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z16, [sp, #9, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z15, [sp, #10, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z14, [sp, #11, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z13, [sp, #12, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z12, [sp, #13, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z11, [sp, #14, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z10, [sp, #15, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    smstop sm
; NOPAIR-NEXT:    bl my_func
; NOPAIR-NEXT:    smstart sm
; NOPAIR-NEXT:    ldr z23, [sp, #2, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z22, [sp, #3, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z21, [sp, #4, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z20, [sp, #5, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z19, [sp, #6, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z18, [sp, #7, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z17, [sp, #8, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z16, [sp, #9, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z15, [sp, #10, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z14, [sp, #11, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z13, [sp, #12, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z12, [sp, #13, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z11, [sp, #14, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z10, [sp, #15, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    addvl sp, sp, #18
; NOPAIR-NEXT:    ldp x29, x30, [sp], #16 // 16-byte Folded Reload
; NOPAIR-NEXT:    ret
;
; PAIR-LABEL: fhalf:
; PAIR:       // %bb.0:
; PAIR-NEXT:    stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
; PAIR-NEXT:    addvl sp, sp, #-18
; PAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; PAIR-NEXT:    ptrue pn8.b
; PAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; PAIR-NEXT:    st1b { z22.b, z23.b }, pn8, [sp, #2, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    st1b { z20.b, z21.b }, pn8, [sp, #4, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; PAIR-NEXT:    st1b { z18.b, z19.b }, pn8, [sp, #6, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    st1b { z16.b, z17.b }, pn8, [sp, #8, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; PAIR-NEXT:    st1b { z14.b, z15.b }, pn8, [sp, #10, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    st1b { z12.b, z13.b }, pn8, [sp, #12, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; PAIR-NEXT:    st1b { z10.b, z11.b }, pn8, [sp, #14, mul vl] // 32-byte Folded Spill
; PAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; PAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    smstop sm
; PAIR-NEXT:    bl my_func
; PAIR-NEXT:    smstart sm
; PAIR-NEXT:    ptrue pn8.b
; PAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ld1b { z22.b, z23.b }, pn8/z, [sp, #2, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z20.b, z21.b }, pn8/z, [sp, #4, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z18.b, z19.b }, pn8/z, [sp, #6, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z16.b, z17.b }, pn8/z, [sp, #8, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z14.b, z15.b }, pn8/z, [sp, #10, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z12.b, z13.b }, pn8/z, [sp, #12, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ld1b { z10.b, z11.b }, pn8/z, [sp, #14, mul vl] // 32-byte Folded Reload
; PAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; PAIR-NEXT:    addvl sp, sp, #18
; PAIR-NEXT:    ldp x29, x30, [sp], #16 // 16-byte Folded Reload
; PAIR-NEXT:    ret
  call void @my_func()
  ret void
}

define void @ffloat(<vscale x 4 x i32> %v) #2 {
; NOPAIR-LABEL: ffloat:
; NOPAIR:       // %bb.0:
; NOPAIR-NEXT:    stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
; NOPAIR-NEXT:    addsvl sp, sp, #-18
; NOPAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; NOPAIR-NEXT:    str z23, [sp, #2, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z22, [sp, #3, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z21, [sp, #4, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z20, [sp, #5, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z19, [sp, #6, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z18, [sp, #7, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z17, [sp, #8, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z16, [sp, #9, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z15, [sp, #10, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z14, [sp, #11, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z13, [sp, #12, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z12, [sp, #13, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z11, [sp, #14, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z10, [sp, #15, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; NOPAIR-NEXT:    smstart sm
; NOPAIR-NEXT:    smstop sm
; NOPAIR-NEXT:    bl my_func
; NOPAIR-NEXT:    ldr z23, [sp, #2, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z22, [sp, #3, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z21, [sp, #4, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z20, [sp, #5, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z19, [sp, #6, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z18, [sp, #7, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z17, [sp, #8, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z16, [sp, #9, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z15, [sp, #10, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z14, [sp, #11, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z13, [sp, #12, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z12, [sp, #13, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z11, [sp, #14, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z10, [sp, #15, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; NOPAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; NOPAIR-NEXT:    addsvl sp, sp, #18
; NOPAIR-NEXT:    ldp x29, x30, [sp], #16 // 16-byte Folded Reload
; NOPAIR-NEXT:    ret
;
; PAIR-LABEL: ffloat:
; PAIR:       // %bb.0:
; PAIR-NEXT:    stp x29, x30, [sp, #-16]! // 16-byte Folded Spill
; PAIR-NEXT:    addsvl sp, sp, #-18
; PAIR-NEXT:    str p15, [sp, #4, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p14, [sp, #5, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p13, [sp, #6, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p12, [sp, #7, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p11, [sp, #8, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p10, [sp, #9, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p9, [sp, #10, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p8, [sp, #11, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p7, [sp, #12, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p6, [sp, #13, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p5, [sp, #14, mul vl] // 2-byte Spill
; PAIR-NEXT:    str p4, [sp, #15, mul vl] // 2-byte Spill
; PAIR-NEXT:    str z23, [sp, #2, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z22, [sp, #3, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z21, [sp, #4, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z20, [sp, #5, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z19, [sp, #6, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z18, [sp, #7, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z17, [sp, #8, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z16, [sp, #9, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z15, [sp, #10, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z14, [sp, #11, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z13, [sp, #12, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z12, [sp, #13, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z11, [sp, #14, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z10, [sp, #15, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z9, [sp, #16, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    str z8, [sp, #17, mul vl] // 16-byte Folded Spill
; PAIR-NEXT:    smstart sm
; PAIR-NEXT:    smstop sm
; PAIR-NEXT:    bl my_func
; PAIR-NEXT:    ldr z23, [sp, #2, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z22, [sp, #3, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z21, [sp, #4, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z20, [sp, #5, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z19, [sp, #6, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z18, [sp, #7, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z17, [sp, #8, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z16, [sp, #9, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z15, [sp, #10, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z14, [sp, #11, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z13, [sp, #12, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z12, [sp, #13, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z11, [sp, #14, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z10, [sp, #15, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z9, [sp, #16, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr z8, [sp, #17, mul vl] // 16-byte Folded Reload
; PAIR-NEXT:    ldr p15, [sp, #4, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p14, [sp, #5, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p13, [sp, #6, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p12, [sp, #7, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p11, [sp, #8, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p10, [sp, #9, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p9, [sp, #10, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p8, [sp, #11, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p7, [sp, #12, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p6, [sp, #13, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p5, [sp, #14, mul vl] // 2-byte Reload
; PAIR-NEXT:    ldr p4, [sp, #15, mul vl] // 2-byte Reload
; PAIR-NEXT:    addsvl sp, sp, #18
; PAIR-NEXT:    ldp x29, x30, [sp], #16 // 16-byte Folded Reload
; PAIR-NEXT:    ret
  call void @my_func()
  ret void
}



attributes #0 = { nounwind "aarch64_pstate_sm_compatible" }
attributes #1 = { nounwind "aarch64_pstate_sm_enabled" }
attributes #2 = { nounwind "aarch64_pstate_sm_body" }