1; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-always-inline %s | FileCheck -check-prefixes=CALLS-ENABLED,ALL %s
2; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-stress-function-calls -amdgpu-always-inline %s | FileCheck -check-prefixes=STRESS-CALLS,ALL %s
3
4target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"
5
6@lds0 = addrspace(3) global i32 undef, align 4
7@lds1 = addrspace(3) global [512 x i32] undef, align 4
8@nested.lds.address = addrspace(1) global i32 addrspace(3)* @lds0, align 4
9@gds0 = addrspace(2) global i32 undef, align 4
10
11@alias.lds0 = alias i32, i32 addrspace(3)* @lds0
12@lds.cycle = addrspace(3) global i32 ptrtoint (i32 addrspace(3)* @lds.cycle to i32), align 4
13
14
15; ALL-LABEL: define i32 @load_lds_simple() #0 {
16define i32 @load_lds_simple() {
17  %load = load i32, i32 addrspace(3)* @lds0, align 4
18  ret i32 %load
19}
20
21; ALL-LABEL: define i32 @load_gds_simple() #0 {
22define i32 @load_gds_simple() {
23  %load = load i32, i32 addrspace(2)* @gds0, align 4
24  ret i32 %load
25}
26
27; ALL-LABEL: define i32 @load_lds_const_gep() #0 {
28define i32 @load_lds_const_gep() {
29  %load = load i32, i32 addrspace(3)* getelementptr inbounds ([512 x i32], [512 x i32] addrspace(3)* @lds1, i64 0, i64 4), align 4
30  ret i32 %load
31}
32
33; ALL-LABEL: define i32 @load_lds_var_gep(i32 %idx) #0 {
34define i32 @load_lds_var_gep(i32 %idx) {
35  %gep = getelementptr inbounds [512 x i32], [512 x i32] addrspace(3)* @lds1, i32 0, i32 %idx
36  %load = load i32, i32 addrspace(3)* %gep, align 4
37  ret i32 %load
38}
39
40; ALL-LABEL: define i32 addrspace(3)* @load_nested_address(i32 %idx) #0 {
41define i32 addrspace(3)* @load_nested_address(i32 %idx) {
42  %load = load i32 addrspace(3)*, i32 addrspace(3)* addrspace(1)* @nested.lds.address, align 4
43  ret i32 addrspace(3)* %load
44}
45
46; ALL-LABEL: define i32 @load_lds_alias() #0 {
47define i32 @load_lds_alias() {
48  %load = load i32, i32 addrspace(3)* @alias.lds0, align 4
49  ret i32 %load
50}
51
52; ALL-LABEL: define i32 @load_lds_cycle() #0 {
53define i32 @load_lds_cycle() {
54  %load = load i32, i32 addrspace(3)* @lds.cycle, align 4
55  ret i32 %load
56}
57
58; ALL-LABEL: define i1 @icmp_lds_address() #0 {
59define i1 @icmp_lds_address() {
60  ret i1 icmp eq (i32 addrspace(3)* @lds0, i32 addrspace(3)* null)
61}
62
63; ALL-LABEL: define i32 @transitive_call() #0 {
64define i32 @transitive_call() {
65  %call = call i32 @load_lds_simple()
66  ret i32 %call
67}
68
69; ALL-LABEL: define i32 @recursive_call_lds(i32 %arg0) #0 {
70define i32 @recursive_call_lds(i32 %arg0) {
71  %load = load i32, i32 addrspace(3)* @lds0, align 4
72  %add = add i32 %arg0, %load
73  %call = call i32 @recursive_call_lds(i32 %add)
74  ret i32 %call
75}
76
77; Test we don't break the IR and have both alwaysinline and noinline
78; FIXME: We should really not override noinline.
79
80; ALL-LABEL: define i32 @load_lds_simple_noinline() #0 {
81define i32 @load_lds_simple_noinline() noinline {
82  %load = load i32, i32 addrspace(3)* @lds0, align 4
83  ret i32 %load
84}
85
86; ALL-LABEL: define i32 @recursive_call_lds_noinline(i32 %arg0) #0 {
87define i32 @recursive_call_lds_noinline(i32 %arg0) noinline {
88  %load = load i32, i32 addrspace(3)* @lds0, align 4
89  %add = add i32 %arg0, %load
90  %call = call i32 @recursive_call_lds(i32 %add)
91  ret i32 %call
92}
93
94; ALL: attributes #0 = { alwaysinline }
95