; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,GISEL %s

define amdgpu_kernel void @test_bitcast_return_type_noinline() #0 {
; SDAG-LABEL: test_bitcast_return_type_noinline:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, ret_i32_noinline@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, ret_i32_noinline@rel32@hi+12
; SDAG-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; SDAG-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; SDAG-NEXT:    v_or_b32_e32 v0, v0, v1
; SDAG-NEXT:    v_or_b32_e32 v31, v0, v2
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_f32_e32 v0, 1.0, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_bitcast_return_type_noinline:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; GISEL-NEXT:    v_or_b32_e32 v0, v0, v1
; GISEL-NEXT:    v_or_b32_e32 v31, v0, v2
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, ret_i32_noinline@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, ret_i32_noinline@rel32@hi+12
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_f32_e32 v0, 1.0, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = call float @ret_i32_noinline()
  %op = fadd float %val, 1.0
  store volatile float %op, ptr addrspace(1) poison
  ret void
}

define amdgpu_kernel void @test_bitcast_return_type_alwaysinline() #0 {
; SDAG-LABEL: test_bitcast_return_type_alwaysinline:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, ret_i32_alwaysinline@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, ret_i32_alwaysinline@rel32@hi+12
; SDAG-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; SDAG-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; SDAG-NEXT:    v_or_b32_e32 v0, v0, v1
; SDAG-NEXT:    v_or_b32_e32 v31, v0, v2
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_f32_e32 v0, 1.0, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_bitcast_return_type_alwaysinline:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; GISEL-NEXT:    v_or_b32_e32 v0, v0, v1
; GISEL-NEXT:    v_or_b32_e32 v31, v0, v2
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, ret_i32_alwaysinline@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, ret_i32_alwaysinline@rel32@hi+12
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_f32_e32 v0, 1.0, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = call float @ret_i32_alwaysinline()
  %op = fadd float %val, 1.0
  store volatile float %op, ptr addrspace(1) poison
  ret void
}

define amdgpu_kernel void @test_bitcast_argument_type() #0 {
; SDAG-LABEL: test_bitcast_argument_type:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; SDAG-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; SDAG-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; SDAG-NEXT:    v_or_b32_e32 v0, v0, v1
; SDAG-NEXT:    v_or_b32_e32 v31, v0, v2
; SDAG-NEXT:    v_mov_b32_e32 v0, 2.0
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_i32_e32 v0, vcc, 1, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_bitcast_argument_type:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; GISEL-NEXT:    v_or_b32_e32 v0, v0, v1
; GISEL-NEXT:    v_or_b32_e32 v31, v0, v2
; GISEL-NEXT:    v_mov_b32_e32 v0, 2.0
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_i32_e32 v0, vcc, 1, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = call i32 @ident_i32(float 2.0)
  %op = add i32 %val, 1
  store volatile i32 %op, ptr addrspace(1) poison
  ret void
}

define amdgpu_kernel void @test_bitcast_argument_and_return_types() #0 {
; SDAG-LABEL: test_bitcast_argument_and_return_types:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; SDAG-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; SDAG-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; SDAG-NEXT:    v_or_b32_e32 v0, v0, v1
; SDAG-NEXT:    v_or_b32_e32 v31, v0, v2
; SDAG-NEXT:    v_mov_b32_e32 v0, 2.0
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_f32_e32 v0, 1.0, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_bitcast_argument_and_return_types:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; GISEL-NEXT:    v_or_b32_e32 v0, v0, v1
; GISEL-NEXT:    v_or_b32_e32 v31, v0, v2
; GISEL-NEXT:    v_mov_b32_e32 v0, 2.0
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_f32_e32 v0, 1.0, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = call float @ident_i32(float 2.0)
  %op = fadd float %val, 1.0
  store volatile float %op, ptr addrspace(1) poison
  ret void
}

define hidden i32 @use_workitem_id_x(i32 %arg0) #3 {
; GCN-LABEL: use_workitem_id_x:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_and_b32_e32 v1, 0x3ff, v31
; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
; GCN-NEXT:    s_setpc_b64 s[30:31]
  %id = call i32 @llvm.amdgcn.workitem.id.x()
  %op = add i32 %id, %arg0
  ret i32 %op
}

define amdgpu_kernel void @test_bitcast_use_workitem_id_x() #3 {
; SDAG-LABEL: test_bitcast_use_workitem_id_x:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    v_mov_b32_e32 v31, v0
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, use_workitem_id_x@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, use_workitem_id_x@rel32@hi+12
; SDAG-NEXT:    v_mov_b32_e32 v0, 9
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_f32_e32 v0, 1.0, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_bitcast_use_workitem_id_x:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    v_mov_b32_e32 v31, v0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, use_workitem_id_x@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, use_workitem_id_x@rel32@hi+12
; GISEL-NEXT:    v_mov_b32_e32 v0, 9
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_f32_e32 v0, 1.0, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = call float @use_workitem_id_x(i32 9)
  %op = fadd float %val, 1.0
  store volatile float %op, ptr addrspace(1) poison
  ret void
}

@_ZTIi = external global ptr
declare i32 @__gxx_personality_v0(...)
define amdgpu_kernel void @test_invoke() #0 personality ptr @__gxx_personality_v0 {
; SDAG-LABEL: test_invoke:
; SDAG:       ; %bb.0:
; SDAG-NEXT:    s_mov_b32 s32, 0
; SDAG-NEXT:    s_mov_b32 flat_scratch_lo, s13
; SDAG-NEXT:    s_add_i32 s12, s12, s17
; SDAG-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; SDAG-NEXT:    s_add_u32 s0, s0, s17
; SDAG-NEXT:    s_addc_u32 s1, s1, 0
; SDAG-NEXT:    s_mov_b32 s13, s15
; SDAG-NEXT:    s_mov_b32 s12, s14
; SDAG-NEXT:    s_getpc_b64 s[18:19]
; SDAG-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; SDAG-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; SDAG-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; SDAG-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; SDAG-NEXT:    v_or_b32_e32 v0, v0, v1
; SDAG-NEXT:    v_or_b32_e32 v31, v0, v2
; SDAG-NEXT:    v_mov_b32_e32 v0, 2.0
; SDAG-NEXT:    s_mov_b32 s14, s16
; SDAG-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; SDAG-NEXT:    v_add_f32_e32 v0, 1.0, v0
; SDAG-NEXT:    flat_store_dword v[0:1], v0
; SDAG-NEXT:    s_waitcnt vmcnt(0)
; SDAG-NEXT:    s_endpgm
;
; GISEL-LABEL: test_invoke:
; GISEL:       ; %bb.0:
; GISEL-NEXT:    s_mov_b32 s32, 0
; GISEL-NEXT:    s_mov_b32 flat_scratch_lo, s13
; GISEL-NEXT:    s_add_i32 s12, s12, s17
; GISEL-NEXT:    s_lshr_b32 flat_scratch_hi, s12, 8
; GISEL-NEXT:    s_add_u32 s0, s0, s17
; GISEL-NEXT:    s_addc_u32 s1, s1, 0
; GISEL-NEXT:    s_mov_b32 s13, s15
; GISEL-NEXT:    s_mov_b32 s12, s14
; GISEL-NEXT:    v_lshlrev_b32_e32 v1, 10, v1
; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 20, v2
; GISEL-NEXT:    s_getpc_b64 s[18:19]
; GISEL-NEXT:    s_add_u32 s18, s18, ident_i32@rel32@lo+4
; GISEL-NEXT:    s_addc_u32 s19, s19, ident_i32@rel32@hi+12
; GISEL-NEXT:    v_or_b32_e32 v0, v0, v1
; GISEL-NEXT:    v_or_b32_e32 v31, v0, v2
; GISEL-NEXT:    v_mov_b32_e32 v0, 2.0
; GISEL-NEXT:    s_mov_b32 s14, s16
; GISEL-NEXT:    s_swappc_b64 s[30:31], s[18:19]
; GISEL-NEXT:    v_add_f32_e32 v0, 1.0, v0
; GISEL-NEXT:    flat_store_dword v[0:1], v0
; GISEL-NEXT:    s_waitcnt vmcnt(0)
; GISEL-NEXT:    s_endpgm
  %val = invoke float @ident_i32(float 2.0)
          to label %continue unwind label %broken

broken:
  landingpad { ptr, i32 } catch ptr @_ZTIi
  ret void

continue:
  %op = fadd float %val, 1.0
  store volatile float %op, ptr addrspace(1) poison
  ret void
}

; Callees appears last in source file to test that we still lower their
; arguments before we lower any calls to them.

define hidden i32 @ret_i32_noinline() #0 {
; GCN-LABEL: ret_i32_noinline:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, 4
; GCN-NEXT:    s_setpc_b64 s[30:31]
  ret i32 4
}

define hidden i32 @ret_i32_alwaysinline() #1 {
; GCN-LABEL: ret_i32_alwaysinline:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    v_mov_b32_e32 v0, 4
; GCN-NEXT:    s_setpc_b64 s[30:31]
  ret i32 4
}

define hidden i32 @ident_i32(i32 %i) #0 {
; GCN-LABEL: ident_i32:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT:    s_setpc_b64 s[30:31]
  ret i32 %i
}

declare i32 @llvm.amdgcn.workitem.id.x() #2

attributes #0 = { nounwind noinline }
attributes #1 = { alwaysinline nounwind }
attributes #2 = { nounwind readnone speculatable }
attributes #3 = { nounwind noinline "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" }