; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefix=GCN %s
; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1250 < %s | FileCheck --check-prefix=GCN %s

declare void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 %col)

define amdgpu_ps void @global_prefetch(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 0)
  ret void
}

define amdgpu_ps void @global_prefetch_sgpr(ptr addrspace(1) inreg %ptr) {
; GCN-LABEL: global_prefetch_sgpr:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    v_mov_b32_e32 v0, 0
; GCN-NEXT:    global_prefetch_b8 v0, s[0:1]
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 0)
  ret void
}

define amdgpu_ps void @global_prefetch_offset(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch_offset:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off offset:512
; GCN-NEXT:    s_endpgm
entry:
  %gep = getelementptr inbounds i32, ptr addrspace(1) %ptr, i32 128
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %gep, i32 0)
  ret void
}

define amdgpu_ps void @global_prefetch_sgpr_voffset(ptr addrspace(1) inreg %ptr, i32 %offset) {
; GCN-LABEL: global_prefetch_sgpr_voffset:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v0, s[0:1]
; GCN-NEXT:    s_endpgm
entry:
  %gep = getelementptr i8, ptr addrspace(1) %ptr, i32 %offset
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %gep, i32 0)
  ret void
}

define amdgpu_ps void @global_prefetch_sgpr_voffset_offset(ptr addrspace(1) inreg %ptr, i32 %offset) {
; GCN-LABEL: global_prefetch_sgpr_voffset_offset:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v0, s[0:1] offset:128
; GCN-NEXT:    s_endpgm
entry:
  %gep1 = getelementptr i8, ptr addrspace(1) %ptr, i32 %offset
  %gep2 = getelementptr i8, ptr addrspace(1) %gep1, i32 128
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %gep2, i32 0)
  ret void
}

define amdgpu_ps void @global_prefetch_se(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch_se:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off scope:SCOPE_SE
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 8)
  ret void
}

define amdgpu_ps void @global_prefetch_se_nt(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch_se_nt:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off th:TH_LOAD_NT scope:SCOPE_SE
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 9)
  ret void
}

define amdgpu_ps void @global_prefetch_dev_ht(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch_dev_ht:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off th:TH_LOAD_HT scope:SCOPE_DEV
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 18)
  ret void
}

define amdgpu_ps void @global_prefetch_sys_lu(ptr addrspace(1) %ptr) {
; GCN-LABEL: global_prefetch_sys_lu:
; GCN:       ; %bb.0: ; %entry
; GCN-NEXT:    global_prefetch_b8 v[0:1], off th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GCN-NEXT:    s_endpgm
entry:
  tail call void @llvm.amdgcn.global.prefetch(ptr addrspace(1) %ptr, i32 27)
  ret void
}