; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 --verify-machineinstrs < %s | FileCheck -check-prefix=GFX11-PAL-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 --verify-machineinstrs < %s | FileCheck -check-prefix=GFX11-PAL-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250-PAL %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdpal -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250-PAL %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 --verify-machineinstrs < %s | FileCheck -check-prefix=GFX11-HSA %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 --verify-machineinstrs < %s | FileCheck -check-prefix=GFX11-HSA %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250-HSA %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 < %s | FileCheck -check-prefix=GFX1250-HSA %s

define amdgpu_kernel void @caller_internal() {
; GFX11-PAL-SDAG-LABEL: caller_internal:
; GFX11-PAL-SDAG:       ; %bb.0:
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s1, internal_func@abs32@hi
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s0, internal_func@abs32@lo
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s32, 0
; GFX11-PAL-SDAG-NEXT:    s_swappc_b64 s[30:31], s[0:1]
; GFX11-PAL-SDAG-NEXT:    s_endpgm
;
; GFX11-PAL-GISEL-LABEL: caller_internal:
; GFX11-PAL-GISEL:       ; %bb.0:
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s0, internal_func@abs32@lo
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s1, internal_func@abs32@hi
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s32, 0
; GFX11-PAL-GISEL-NEXT:    s_swappc_b64 s[30:31], s[0:1]
; GFX11-PAL-GISEL-NEXT:    s_endpgm
;
; GFX1250-PAL-LABEL: caller_internal:
; GFX1250-PAL:       ; %bb.0:
; GFX1250-PAL-NEXT:    s_mov_b64 s[0:1], internal_func@abs64
; GFX1250-PAL-NEXT:    s_mov_b32 s32, 0
; GFX1250-PAL-NEXT:    s_swap_pc_i64 s[30:31], s[0:1]
; GFX1250-PAL-NEXT:    s_endpgm
;
; GFX11-HSA-LABEL: caller_internal:
; GFX11-HSA:       ; %bb.0:
; GFX11-HSA-NEXT:    s_getpc_b64 s[0:1]
; GFX11-HSA-NEXT:    s_add_u32 s0, s0, internal_func@gotpcrel32@lo+4
; GFX11-HSA-NEXT:    s_addc_u32 s1, s1, internal_func@gotpcrel32@hi+12
; GFX11-HSA-NEXT:    s_mov_b32 s32, 0
; GFX11-HSA-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
; GFX11-HSA-NEXT:    s_waitcnt lgkmcnt(0)
; GFX11-HSA-NEXT:    s_swappc_b64 s[30:31], s[0:1]
; GFX11-HSA-NEXT:    s_endpgm
;
; GFX1250-HSA-LABEL: caller_internal:
; GFX1250-HSA:       ; %bb.0:
; GFX1250-HSA-NEXT:    s_get_pc_i64 s[0:1]
; GFX1250-HSA-NEXT:    s_add_nc_u64 s[0:1], s[0:1], internal_func@gotpcrel+4
; GFX1250-HSA-NEXT:    s_mov_b32 s32, 0
; GFX1250-HSA-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
; GFX1250-HSA-NEXT:    s_wait_kmcnt 0x0
; GFX1250-HSA-NEXT:    s_swap_pc_i64 s[30:31], s[0:1]
; GFX1250-HSA-NEXT:    s_endpgm
  call amdgpu_gfx void @internal_func()
  ret void
}

define amdgpu_kernel void @caller_exterinal() {
; GFX11-PAL-SDAG-LABEL: caller_exterinal:
; GFX11-PAL-SDAG:       ; %bb.0:
; GFX11-PAL-SDAG-NEXT:    v_mov_b32_e32 v31, v0
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s12, s13
; GFX11-PAL-SDAG-NEXT:    s_mov_b64 s[10:11], s[6:7]
; GFX11-PAL-SDAG-NEXT:    s_mov_b64 s[8:9], s[4:5]
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s17, external_func@abs32@hi
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s16, external_func@abs32@lo
; GFX11-PAL-SDAG-NEXT:    s_mov_b64 s[4:5], s[0:1]
; GFX11-PAL-SDAG-NEXT:    s_mov_b64 s[6:7], s[2:3]
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s13, s14
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s14, s15
; GFX11-PAL-SDAG-NEXT:    s_mov_b32 s32, 0
; GFX11-PAL-SDAG-NEXT:    s_swappc_b64 s[30:31], s[16:17]
; GFX11-PAL-SDAG-NEXT:    s_endpgm
;
; GFX11-PAL-GISEL-LABEL: caller_exterinal:
; GFX11-PAL-GISEL:       ; %bb.0:
; GFX11-PAL-GISEL-NEXT:    v_mov_b32_e32 v31, v0
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s12, s13
; GFX11-PAL-GISEL-NEXT:    s_mov_b64 s[10:11], s[6:7]
; GFX11-PAL-GISEL-NEXT:    s_mov_b64 s[8:9], s[4:5]
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s16, external_func@abs32@lo
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s17, external_func@abs32@hi
; GFX11-PAL-GISEL-NEXT:    s_mov_b64 s[4:5], s[0:1]
; GFX11-PAL-GISEL-NEXT:    s_mov_b64 s[6:7], s[2:3]
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s13, s14
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s14, s15
; GFX11-PAL-GISEL-NEXT:    s_mov_b32 s32, 0
; GFX11-PAL-GISEL-NEXT:    s_swappc_b64 s[30:31], s[16:17]
; GFX11-PAL-GISEL-NEXT:    s_endpgm
;
; GFX1250-PAL-LABEL: caller_exterinal:
; GFX1250-PAL:       ; %bb.0:
; GFX1250-PAL-NEXT:    v_mov_b32_e32 v31, v0
; GFX1250-PAL-NEXT:    s_mov_b64 s[10:11], s[6:7]
; GFX1250-PAL-NEXT:    s_mov_b64 s[8:9], s[4:5]
; GFX1250-PAL-NEXT:    s_mov_b64 s[12:13], external_func@abs64
; GFX1250-PAL-NEXT:    s_mov_b64 s[4:5], s[0:1]
; GFX1250-PAL-NEXT:    s_mov_b64 s[6:7], s[2:3]
; GFX1250-PAL-NEXT:    s_mov_b32 s32, 0
; GFX1250-PAL-NEXT:    s_swap_pc_i64 s[30:31], s[12:13]
; GFX1250-PAL-NEXT:    s_endpgm
;
; GFX11-HSA-LABEL: caller_exterinal:
; GFX11-HSA:       ; %bb.0:
; GFX11-HSA-NEXT:    v_mov_b32_e32 v31, v0
; GFX11-HSA-NEXT:    s_mov_b32 s12, s13
; GFX11-HSA-NEXT:    s_mov_b64 s[10:11], s[6:7]
; GFX11-HSA-NEXT:    s_mov_b64 s[8:9], s[4:5]
; GFX11-HSA-NEXT:    s_getpc_b64 s[16:17]
; GFX11-HSA-NEXT:    s_add_u32 s16, s16, external_func@rel32@lo+4
; GFX11-HSA-NEXT:    s_addc_u32 s17, s17, external_func@rel32@hi+12
; GFX11-HSA-NEXT:    s_mov_b64 s[4:5], s[0:1]
; GFX11-HSA-NEXT:    s_mov_b64 s[6:7], s[2:3]
; GFX11-HSA-NEXT:    s_mov_b32 s13, s14
; GFX11-HSA-NEXT:    s_mov_b32 s14, s15
; GFX11-HSA-NEXT:    s_mov_b32 s32, 0
; GFX11-HSA-NEXT:    s_swappc_b64 s[30:31], s[16:17]
; GFX11-HSA-NEXT:    s_endpgm
;
; GFX1250-HSA-LABEL: caller_exterinal:
; GFX1250-HSA:       ; %bb.0:
; GFX1250-HSA-NEXT:    v_mov_b32_e32 v31, v0
; GFX1250-HSA-NEXT:    s_mov_b64 s[10:11], s[6:7]
; GFX1250-HSA-NEXT:    s_mov_b64 s[8:9], s[4:5]
; GFX1250-HSA-NEXT:    s_get_pc_i64 s[12:13]
; GFX1250-HSA-NEXT:    s_add_nc_u64 s[12:13], s[12:13], external_func@rel64+4
; GFX1250-HSA-NEXT:    s_mov_b64 s[4:5], s[0:1]
; GFX1250-HSA-NEXT:    s_mov_b64 s[6:7], s[2:3]
; GFX1250-HSA-NEXT:    s_mov_b32 s32, 0
; GFX1250-HSA-NEXT:    s_swap_pc_i64 s[30:31], s[12:13]
; GFX1250-HSA-NEXT:    s_endpgm
  call void @external_func()
  ret void
}

declare amdgpu_gfx void @internal_func()
declare hidden void @external_func()