• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=CI %s
3; RUN: llc -global-isel -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GFX9 %s
4
5; TODO: Merge with DAG test
6
7define amdgpu_kernel void @is_local_vgpr(i8* addrspace(1)* %ptr.ptr) {
8; CI-LABEL: is_local_vgpr:
9; CI:       ; %bb.0:
10; CI-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
11; CI-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
12; CI-NEXT:    s_waitcnt lgkmcnt(0)
13; CI-NEXT:    v_mov_b32_e32 v0, s0
14; CI-NEXT:    v_mov_b32_e32 v1, s1
15; CI-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
16; CI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
17; CI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
18; CI-NEXT:    s_load_dword s0, s[4:5], 0x10
19; CI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
20; CI-NEXT:    v_cmp_eq_u32_e32 vcc, s0, v1
21; CI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
22; CI-NEXT:    flat_store_dword v[0:1], v0
23; CI-NEXT:    s_endpgm
24;
25; GFX9-LABEL: is_local_vgpr:
26; GFX9:       ; %bb.0:
27; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
28; GFX9-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
29; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
30; GFX9-NEXT:    global_load_dwordx2 v[0:1], v0, s[0:1]
31; GFX9-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_SH_MEM_BASES, 16, 16)
32; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
33; GFX9-NEXT:    s_waitcnt vmcnt(0)
34; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s0, v1
35; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
36; GFX9-NEXT:    global_store_dword v[0:1], v0, off
37; GFX9-NEXT:    s_endpgm
38  %id = call i32 @llvm.amdgcn.workitem.id.x()
39  %gep = getelementptr inbounds i8*, i8* addrspace(1)* %ptr.ptr, i32 %id
40  %ptr = load volatile i8*, i8* addrspace(1)* %gep
41  %val = call i1 @llvm.amdgcn.is.shared(i8* %ptr)
42  %ext = zext i1 %val to i32
43  store i32 %ext, i32 addrspace(1)* undef
44  ret void
45}
46
47define amdgpu_kernel void @is_local_sgpr(i8* %ptr) {
48; CI-LABEL: is_local_sgpr:
49; CI:       ; %bb.0:
50; CI-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
51; CI-NEXT:    s_waitcnt lgkmcnt(0)
52; CI-NEXT:    s_load_dword s0, s[4:5], 0x10
53; CI-NEXT:    s_waitcnt lgkmcnt(0)
54; CI-NEXT:    s_cmp_lg_u32 s1, s0
55; CI-NEXT:    s_cselect_b32 s0, 1, 0
56; CI-NEXT:    s_and_b32 s0, s0, 1
57; CI-NEXT:    s_cmp_lg_u32 s0, 0
58; CI-NEXT:    s_cbranch_scc1 BB1_2
59; CI-NEXT:  ; %bb.1: ; %bb0
60; CI-NEXT:    v_mov_b32_e32 v0, 0
61; CI-NEXT:    flat_store_dword v[0:1], v0
62; CI-NEXT:  BB1_2: ; %bb1
63; CI-NEXT:    s_endpgm
64;
65; GFX9-LABEL: is_local_sgpr:
66; GFX9:       ; %bb.0:
67; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
68; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
69; GFX9-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_SH_MEM_BASES, 16, 16)
70; GFX9-NEXT:    s_lshl_b32 s0, s0, 16
71; GFX9-NEXT:    s_cmp_lg_u32 s1, s0
72; GFX9-NEXT:    s_cselect_b32 s0, 1, 0
73; GFX9-NEXT:    s_and_b32 s0, s0, 1
74; GFX9-NEXT:    s_cmp_lg_u32 s0, 0
75; GFX9-NEXT:    s_cbranch_scc1 BB1_2
76; GFX9-NEXT:  ; %bb.1: ; %bb0
77; GFX9-NEXT:    v_mov_b32_e32 v0, 0
78; GFX9-NEXT:    global_store_dword v[0:1], v0, off
79; GFX9-NEXT:  BB1_2: ; %bb1
80; GFX9-NEXT:    s_endpgm
81  %val = call i1 @llvm.amdgcn.is.shared(i8* %ptr)
82  br i1 %val, label %bb0, label %bb1
83
84bb0:
85  store volatile i32 0, i32 addrspace(1)* undef
86  br label %bb1
87
88bb1:
89  ret void
90}
91
92declare i32 @llvm.amdgcn.workitem.id.x() #0
93declare i1 @llvm.amdgcn.is.shared(i8* nocapture) #0
94
95attributes #0 = { nounwind readnone speculatable }
96