1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 < %s | FileCheck --check-prefix=GFX12 %s
3 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 -mattr=+cumode < %s | FileCheck --check-prefix=GFX12 %s
5 define amdgpu_kernel void @global_last_use_load_0(ptr addrspace(1) %in, ptr addrspace(1) %out) {
6 ; GFX12-LABEL: global_last_use_load_0:
7 ; GFX12: ; %bb.0: ; %entry
8 ; GFX12-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
9 ; GFX12-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
10 ; GFX12-NEXT: v_mov_b32_e32 v0, 0
11 ; GFX12-NEXT: s_wait_kmcnt 0x0
12 ; GFX12-NEXT: s_load_b32 s2, s[2:3], 0x0
13 ; GFX12-NEXT: s_wait_kmcnt 0x0
14 ; GFX12-NEXT: v_mov_b32_e32 v1, s2
15 ; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
16 ; GFX12-NEXT: s_endpgm
18 %val = load i32, ptr addrspace(1) %in, align 4, !amdgpu.last.use !{}
19 store i32 %val, ptr addrspace(1) %out
23 define amdgpu_kernel void @global_last_use_load_1(ptr addrspace(1) %in, ptr addrspace(1) %out) {
24 ; GFX12-LABEL: global_last_use_load_1:
25 ; GFX12: ; %bb.0: ; %entry
26 ; GFX12-NEXT: v_mov_b32_e32 v1, v0
27 ; GFX12-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
28 ; GFX12-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
29 ; GFX12-NEXT: v_mov_b32_e32 v0, 0
30 ; GFX12-NEXT: s_mov_b32 s4, 0x3ff
31 ; GFX12-NEXT: s_wait_alu 0xfffe
32 ; GFX12-NEXT: v_and_b32_e64 v1, v1, s4
33 ; GFX12-NEXT: s_mov_b32 s4, 2
34 ; GFX12-NEXT: s_wait_alu 0xfffe
35 ; GFX12-NEXT: v_lshlrev_b32_e64 v1, s4, v1
36 ; GFX12-NEXT: s_wait_kmcnt 0x0
37 ; GFX12-NEXT: global_load_b32 v1, v1, s[2:3] th:TH_LOAD_LU
38 ; GFX12-NEXT: s_wait_loadcnt 0x0
39 ; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
40 ; GFX12-NEXT: s_endpgm
42 %tid = call i32 @llvm.amdgcn.workitem.id.x()
43 %val.gep = getelementptr inbounds i32, ptr addrspace(1) %in, i32 %tid
44 %val = load i32, ptr addrspace(1) %val.gep, align 4, !amdgpu.last.use !{}
45 store i32 %val, ptr addrspace(1) %out
49 define amdgpu_kernel void @global_last_use_and_volatile_load(ptr addrspace(1) %in, ptr addrspace(1) %out) {
50 ; GFX12-LABEL: global_last_use_and_volatile_load:
51 ; GFX12: ; %bb.0: ; %entry
52 ; GFX12-NEXT: v_mov_b32_e32 v0, 0
53 ; GFX12-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
54 ; GFX12-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
55 ; GFX12-NEXT: s_wait_kmcnt 0x0
56 ; GFX12-NEXT: global_load_b32 v1, v0, s[2:3] th:TH_LOAD_BYPASS scope:SCOPE_SYS
57 ; GFX12-NEXT: s_wait_bvhcnt 0x0
58 ; GFX12-NEXT: s_wait_samplecnt 0x0
59 ; GFX12-NEXT: s_wait_loadcnt 0x0
60 ; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
61 ; GFX12-NEXT: s_endpgm
63 %val = load volatile i32, ptr addrspace(1) %in, align 4, !amdgpu.last.use !{}
64 store i32 %val, ptr addrspace(1) %out
68 define amdgpu_kernel void @global_last_use_and_nontemporal_load(ptr addrspace(1) %in, ptr addrspace(1) %out) {
69 ; GFX12-LABEL: global_last_use_and_nontemporal_load:
70 ; GFX12: ; %bb.0: ; %entry
71 ; GFX12-NEXT: v_mov_b32_e32 v1, v0
72 ; GFX12-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
73 ; GFX12-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
74 ; GFX12-NEXT: v_mov_b32_e32 v0, 0
75 ; GFX12-NEXT: s_mov_b32 s4, 0x3ff
76 ; GFX12-NEXT: s_wait_alu 0xfffe
77 ; GFX12-NEXT: v_and_b32_e64 v1, v1, s4
78 ; GFX12-NEXT: s_mov_b32 s4, 2
79 ; GFX12-NEXT: s_wait_alu 0xfffe
80 ; GFX12-NEXT: v_lshlrev_b32_e64 v1, s4, v1
81 ; GFX12-NEXT: s_wait_kmcnt 0x0
82 ; GFX12-NEXT: global_load_b32 v1, v1, s[2:3] th:TH_LOAD_LU
83 ; GFX12-NEXT: s_wait_loadcnt 0x0
84 ; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
85 ; GFX12-NEXT: s_endpgm
87 %tid = call i32 @llvm.amdgcn.workitem.id.x()
88 %val.gep = getelementptr inbounds i32, ptr addrspace(1) %in, i32 %tid
89 %val = load i32, ptr addrspace(1) %val.gep, align 4, !amdgpu.last.use !{}, !nontemporal !0
90 store i32 %val, ptr addrspace(1) %out
94 declare i32 @llvm.amdgcn.workitem.id.x()