xref: /llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.pops.exiting.wave.id.ll (revision 520e04540501f676580b9b73419265e1951a9c52)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
2; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GFX9,SDAG,GFX9-SDAG
3; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GFX9,GFX9-GISEL
4; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck %s -check-prefixes=GFX10,SDAG,GFX10-SDAG
5; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck %s -check-prefixes=GFX10,GFX10-GISEL
6
7declare void @foo(i32)
8
9define amdgpu_ps void @test(ptr addrspace(1) inreg %ptr) {
10; SDAG-LABEL: test:
11; SDAG:       ; %bb.0:
12; SDAG-NEXT:    s_mov_b32 s2, src_pops_exiting_wave_id
13; SDAG-NEXT:    v_mov_b32_e32 v0, 0
14; SDAG-NEXT:    v_mov_b32_e32 v1, s2
15; SDAG-NEXT:    global_store_dword v0, v1, s[0:1]
16; SDAG-NEXT:    s_endpgm
17;
18; GFX9-GISEL-LABEL: test:
19; GFX9-GISEL:       ; %bb.0:
20; GFX9-GISEL-NEXT:    s_mov_b32 s2, src_pops_exiting_wave_id
21; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s2
22; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
23; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
24; GFX9-GISEL-NEXT:    s_endpgm
25;
26; GFX10-GISEL-LABEL: test:
27; GFX10-GISEL:       ; %bb.0:
28; GFX10-GISEL-NEXT:    s_mov_b32 s2, src_pops_exiting_wave_id
29; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
30; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
31; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
32; GFX10-GISEL-NEXT:    s_endpgm
33  %id = call i32 @llvm.amdgcn.pops.exiting.wave.id()
34  store i32 %id, ptr addrspace(1) %ptr
35  ret void
36}
37
38define amdgpu_ps void @test_loop() {
39; GFX9-LABEL: test_loop:
40; GFX9:       ; %bb.0:
41; GFX9-NEXT:  .LBB1_1: ; %loop
42; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
43; GFX9-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
44; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
45; GFX9-NEXT:    s_cbranch_scc1 .LBB1_1
46; GFX9-NEXT:  ; %bb.2: ; %exit
47; GFX9-NEXT:    s_endpgm
48;
49; GFX10-LABEL: test_loop:
50; GFX10:       ; %bb.0:
51; GFX10-NEXT:  .LBB1_1: ; %loop
52; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
53; GFX10-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
54; GFX10-NEXT:    s_cmp_eq_u32 s0, 0
55; GFX10-NEXT:    s_cbranch_scc1 .LBB1_1
56; GFX10-NEXT:  ; %bb.2: ; %exit
57; GFX10-NEXT:    s_endpgm
58  br label %loop
59loop:
60  %id = call i32 @llvm.amdgcn.pops.exiting.wave.id()
61  %cond = icmp eq i32 %id, 0
62  br i1 %cond, label %loop, label %exit
63exit:
64  ret void
65}
66
67define amdgpu_ps i32 @test_if(i1 inreg %cond) {
68; SDAG-LABEL: test_if:
69; SDAG:       ; %bb.0: ; %entry
70; SDAG-NEXT:    s_bitcmp0_b32 s0, 0
71; SDAG-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
72; SDAG-NEXT:    s_cbranch_scc1 .LBB2_2
73; SDAG-NEXT:  ; %bb.1: ; %body
74; SDAG-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
75; SDAG-NEXT:  .LBB2_2: ; %exit
76; SDAG-NEXT:    ; return to shader part epilog
77;
78; GFX9-GISEL-LABEL: test_if:
79; GFX9-GISEL:       ; %bb.0: ; %entry
80; GFX9-GISEL-NEXT:    s_mov_b32 s1, s0
81; GFX9-GISEL-NEXT:    s_xor_b32 s1, s1, 1
82; GFX9-GISEL-NEXT:    s_and_b32 s1, s1, 1
83; GFX9-GISEL-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
84; GFX9-GISEL-NEXT:    s_cmp_lg_u32 s1, 0
85; GFX9-GISEL-NEXT:    s_cbranch_scc1 .LBB2_2
86; GFX9-GISEL-NEXT:  ; %bb.1: ; %body
87; GFX9-GISEL-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
88; GFX9-GISEL-NEXT:  .LBB2_2: ; %exit
89; GFX9-GISEL-NEXT:    ; return to shader part epilog
90;
91; GFX10-GISEL-LABEL: test_if:
92; GFX10-GISEL:       ; %bb.0: ; %entry
93; GFX10-GISEL-NEXT:    s_xor_b32 s0, s0, 1
94; GFX10-GISEL-NEXT:    s_and_b32 s1, s0, 1
95; GFX10-GISEL-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
96; GFX10-GISEL-NEXT:    s_cmp_lg_u32 s1, 0
97; GFX10-GISEL-NEXT:    s_cbranch_scc1 .LBB2_2
98; GFX10-GISEL-NEXT:  ; %bb.1: ; %body
99; GFX10-GISEL-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
100; GFX10-GISEL-NEXT:  .LBB2_2: ; %exit
101; GFX10-GISEL-NEXT:    ; return to shader part epilog
102entry:
103  %id1 = call i32 @llvm.amdgcn.pops.exiting.wave.id()
104  br i1 %cond, label %body, label %exit
105body:
106  %id2 = call i32 @llvm.amdgcn.pops.exiting.wave.id()
107  br label %exit
108exit:
109  %id = phi i32 [ %id1, %entry ], [ %id2, %body ]
110  ret i32 %id
111}
112
113define amdgpu_ps void @test_call(ptr addrspace(1) inreg %ptr) {
114; GFX9-SDAG-LABEL: test_call:
115; GFX9-SDAG:       ; %bb.0:
116; GFX9-SDAG-NEXT:    s_mov_b32 s36, SCRATCH_RSRC_DWORD0
117; GFX9-SDAG-NEXT:    s_mov_b32 s37, SCRATCH_RSRC_DWORD1
118; GFX9-SDAG-NEXT:    s_mov_b32 s38, -1
119; GFX9-SDAG-NEXT:    s_mov_b32 s39, 0xe00000
120; GFX9-SDAG-NEXT:    s_add_u32 s36, s36, s2
121; GFX9-SDAG-NEXT:    s_addc_u32 s37, s37, 0
122; GFX9-SDAG-NEXT:    s_getpc_b64 s[0:1]
123; GFX9-SDAG-NEXT:    s_add_u32 s0, s0, foo@gotpcrel32@lo+4
124; GFX9-SDAG-NEXT:    s_addc_u32 s1, s1, foo@gotpcrel32@hi+12
125; GFX9-SDAG-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
126; GFX9-SDAG-NEXT:    s_mov_b32 s6, src_pops_exiting_wave_id
127; GFX9-SDAG-NEXT:    s_mov_b64 s[0:1], s[36:37]
128; GFX9-SDAG-NEXT:    s_mov_b64 s[8:9], 36
129; GFX9-SDAG-NEXT:    s_mov_b64 s[2:3], s[38:39]
130; GFX9-SDAG-NEXT:    v_mov_b32_e32 v0, s6
131; GFX9-SDAG-NEXT:    s_mov_b32 s32, 0
132; GFX9-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
133; GFX9-SDAG-NEXT:    s_swappc_b64 s[30:31], s[4:5]
134; GFX9-SDAG-NEXT:    s_endpgm
135;
136; GFX9-GISEL-LABEL: test_call:
137; GFX9-GISEL:       ; %bb.0:
138; GFX9-GISEL-NEXT:    s_mov_b32 s36, SCRATCH_RSRC_DWORD0
139; GFX9-GISEL-NEXT:    s_mov_b32 s37, SCRATCH_RSRC_DWORD1
140; GFX9-GISEL-NEXT:    s_mov_b32 s38, -1
141; GFX9-GISEL-NEXT:    s_mov_b32 s39, 0xe00000
142; GFX9-GISEL-NEXT:    s_add_u32 s36, s36, s2
143; GFX9-GISEL-NEXT:    s_addc_u32 s37, s37, 0
144; GFX9-GISEL-NEXT:    s_getpc_b64 s[0:1]
145; GFX9-GISEL-NEXT:    s_add_u32 s0, s0, foo@gotpcrel32@lo+4
146; GFX9-GISEL-NEXT:    s_addc_u32 s1, s1, foo@gotpcrel32@hi+12
147; GFX9-GISEL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
148; GFX9-GISEL-NEXT:    s_mov_b32 s2, src_pops_exiting_wave_id
149; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s2
150; GFX9-GISEL-NEXT:    s_mov_b64 s[0:1], s[36:37]
151; GFX9-GISEL-NEXT:    s_mov_b64 s[2:3], s[38:39]
152; GFX9-GISEL-NEXT:    s_mov_b64 s[8:9], 36
153; GFX9-GISEL-NEXT:    s_mov_b32 s32, 0
154; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
155; GFX9-GISEL-NEXT:    s_swappc_b64 s[30:31], s[4:5]
156; GFX9-GISEL-NEXT:    s_endpgm
157;
158; GFX10-LABEL: test_call:
159; GFX10:       ; %bb.0:
160; GFX10-NEXT:    s_mov_b32 s36, SCRATCH_RSRC_DWORD0
161; GFX10-NEXT:    s_mov_b32 s37, SCRATCH_RSRC_DWORD1
162; GFX10-NEXT:    s_mov_b32 s38, -1
163; GFX10-NEXT:    s_mov_b32 s39, 0x31c16000
164; GFX10-NEXT:    s_add_u32 s36, s36, s2
165; GFX10-NEXT:    s_addc_u32 s37, s37, 0
166; GFX10-NEXT:    s_getpc_b64 s[0:1]
167; GFX10-NEXT:    s_add_u32 s0, s0, foo@gotpcrel32@lo+4
168; GFX10-NEXT:    s_addc_u32 s1, s1, foo@gotpcrel32@hi+12
169; GFX10-NEXT:    s_mov_b64 s[8:9], 36
170; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x0
171; GFX10-NEXT:    s_mov_b32 s0, src_pops_exiting_wave_id
172; GFX10-NEXT:    s_mov_b32 s32, 0
173; GFX10-NEXT:    v_mov_b32_e32 v0, s0
174; GFX10-NEXT:    s_mov_b64 s[0:1], s[36:37]
175; GFX10-NEXT:    s_mov_b64 s[2:3], s[38:39]
176; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
177; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
178; GFX10-NEXT:    s_endpgm
179  %id = call i32 @llvm.amdgcn.pops.exiting.wave.id()
180  call void @foo(i32 %id)
181  ret void
182}
183
184;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
185; GFX10-SDAG: {{.*}}
186