# REQUIRES: asserts
# RUN: llc --mtriple=hexagon %s -run-pass=pipeliner -debug-only=pipeliner \
# RUN: -window-sched=force -filetype=null -verify-machineinstrs 2>&1 \
# RUN: | FileCheck %s

# CHECK: Best window offset is {{[0-9]+}} and Best II is {{[0-9]+}}.

--- |
  define void @sqrt_approx(i32 noundef %N, ptr noalias %x, ptr noalias %y) #0 {
  entry:
    %isZeroLength = icmp eq i32 %N, 0
    br i1 %isZeroLength, label %loop.exit, label %loop.preheader

  loop.preheader:                                   ; preds = %entry
    %half_splat = tail call <32 x i32> @llvm.hexagon.V6.lvsplatw.128B(i32 1056964608)
    %one_splat = tail call <32 x i32> @llvm.hexagon.V6.lvsplatw.128B(i32 1065353216)
    %two_splat = tail call <32 x i32> @llvm.hexagon.V6.lvsplatw.128B(i32 1073741824)
    br label %loop.body

  loop.exit:                                        ; preds = %loop.body, %entry
    ret void

  loop.body:                                        ; preds = %loop.body, %loop.preheader
    %lsr.iv1 = phi ptr [ %cgep3, %loop.body ], [ %x, %loop.preheader ]
    %lsr.iv = phi ptr [ %cgep, %loop.body ], [ %y, %loop.preheader ]
    %index = phi i32 [ 0, %loop.preheader ], [ %index.next, %loop.body ]
    %vec_x = load <32 x i32>, ptr %lsr.iv1, align 128
    %vec_sqrt_1 = tail call <32 x i32> @llvm.hexagon.V6.vaddw.128B(<32 x i32> %one_splat, <32 x i32> %vec_x)
    %vec_sqrt_2 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_1, <32 x i32> %half_splat)
    %vec_recip_1 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_2, <32 x i32> %half_splat)
    %vec_recip_2 = tail call <32 x i32> @llvm.hexagon.V6.vsubw.128B(<32 x i32> %two_splat, <32 x i32> %vec_recip_1)
    %vec_y1 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_2, <32 x i32> %vec_recip_2)
    %vec_recip_3 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_2, <32 x i32> %vec_y1)
    %vec_recop_4 = tail call <32 x i32> @llvm.hexagon.V6.vsubw.128B(<32 x i32> %two_splat, <32 x i32> %vec_recip_3)
    %vec_y2 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_y1, <32 x i32> %vec_recop_4)
    %vec_sqrt_3 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_x, <32 x i32> %vec_y2)
    %vec_sqrt_4 = tail call <32 x i32> @llvm.hexagon.V6.vaddw.128B(<32 x i32> %vec_y2, <32 x i32> %vec_sqrt_3)
    %vec_sqrt_5 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_4, <32 x i32> %half_splat)
    %vec_recip_5 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_5, <32 x i32> %half_splat)
    %vec_recip_6 = tail call <32 x i32> @llvm.hexagon.V6.vsubw.128B(<32 x i32> %two_splat, <32 x i32> %vec_recip_5)
    %vec_y3 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_5, <32 x i32> %vec_recip_6)
    %vec_recip_7 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_5, <32 x i32> %vec_y3)
    %vec_recop_8 = tail call <32 x i32> @llvm.hexagon.V6.vsubw.128B(<32 x i32> %two_splat, <32 x i32> %vec_recip_7)
    %vec_y4 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_y3, <32 x i32> %vec_recop_8)
    %vec_sqrt_7 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_x, <32 x i32> %vec_y4)
    %vec_sqrt_8 = tail call <32 x i32> @llvm.hexagon.V6.vaddw.128B(<32 x i32> %vec_y4, <32 x i32> %vec_sqrt_7)
    %vec_sqrt_9 = tail call <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32> %vec_sqrt_8, <32 x i32> %half_splat)
    store <32 x i32> %vec_sqrt_9, ptr %lsr.iv, align 128
    %index.next = add nuw i32 %index, 32
    %continue = icmp ult i32 %index.next, %N
    %cgep = getelementptr i8, ptr %lsr.iv, i32 128
    %cgep3 = getelementptr i8, ptr %lsr.iv1, i32 128
    br i1 %continue, label %loop.body, label %loop.exit
  }

  declare <32 x i32> @llvm.hexagon.V6.lvsplatw.128B(i32)
  declare <32 x i32> @llvm.hexagon.V6.vmpyowh.rnd.128B(<32 x i32>, <32 x i32>)
  declare <32 x i32> @llvm.hexagon.V6.vaddw.128B(<32 x i32>, <32 x i32>)
  declare <32 x i32> @llvm.hexagon.V6.vsubw.128B(<32 x i32>, <32 x i32>)

  attributes #0 = { "target-features"="+hvx-length128b,+hvxv69,+v66,-long-calls" }
...
---
name:            sqrt_approx
tracksRegLiveness: true
body:             |
  bb.0.entry:
    successors: %bb.2(0x30000000), %bb.1(0x50000000)
    liveins: $r0, $r1, $r2

    %0:intregs = COPY $r2
    %1:intregs = COPY $r1
    %2:intregs = COPY $r0
    %3:predregs = C2_cmpeqi %2, 0
    J2_jumpt killed %3, %bb.2, implicit-def dead $pc
    J2_jump %bb.1, implicit-def dead $pc

  bb.1.loop.preheader:
    successors: %bb.3(0x80000000)

    %4:intregs = A2_tfrsi 1056964608
    %5:hvxvr = V6_lvsplatw killed %4
    %6:intregs = A2_tfrsi 1065353216
    %7:hvxvr = V6_lvsplatw killed %6
    %8:intregs = A2_tfrsi 1073741824
    %9:hvxvr = V6_lvsplatw killed %8
    %10:intregs = A2_addi %2, 31
    %11:intregs = S2_lsr_i_r %10, 5
    %12:intregs = COPY %11
    J2_loop0r %bb.3, %12, implicit-def $lc0, implicit-def $sa0, implicit-def $usr
    J2_jump %bb.3, implicit-def dead $pc

  bb.2.loop.exit:
    PS_jmpret $r31, implicit-def dead $pc

  bb.3.loop.body (machine-block-address-taken):
    successors: %bb.3(0x7c000000), %bb.2(0x04000000)

    %13:intregs = PHI %1, %bb.1, %14, %bb.3
    %15:intregs = PHI %0, %bb.1, %16, %bb.3
    %17:hvxvr, %14:intregs = V6_vL32b_pi %13, 128 :: (load (s1024) from %ir.lsr.iv1)
    %18:hvxvr = V6_vaddw %7, %17
    %19:hvxvr = V6_vmpyowh_rnd killed %18, %5
    %20:hvxvr = V6_vmpyowh_rnd %19, %5
    %21:hvxvr = V6_vsubw %9, killed %20
    %22:hvxvr = V6_vmpyowh_rnd %19, killed %21
    %23:hvxvr = V6_vmpyowh_rnd %19, %22
    %24:hvxvr = V6_vsubw %9, killed %23
    %25:hvxvr = V6_vmpyowh_rnd %22, killed %24
    %26:hvxvr = V6_vmpyowh_rnd %17, %25
    %27:hvxvr = V6_vaddw %25, killed %26
    %28:hvxvr = V6_vmpyowh_rnd killed %27, %5
    %29:hvxvr = V6_vmpyowh_rnd %28, %5
    %30:hvxvr = V6_vsubw %9, killed %29
    %31:hvxvr = V6_vmpyowh_rnd %28, killed %30
    %32:hvxvr = V6_vmpyowh_rnd %28, %31
    %33:hvxvr = V6_vsubw %9, killed %32
    %34:hvxvr = V6_vmpyowh_rnd %31, killed %33
    %35:hvxvr = V6_vmpyowh_rnd %17, %34
    %36:hvxvr = V6_vaddw %34, killed %35
    %37:hvxvr = V6_vmpyowh_rnd killed %36, %5
    %16:intregs = V6_vS32b_pi %15, 128, killed %37 :: (store (s1024) into %ir.lsr.iv)
    ENDLOOP0 %bb.3, implicit-def $pc, implicit-def $lc0, implicit $sa0, implicit $lc0
    J2_jump %bb.2, implicit-def dead $pc
...