1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
2; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-annotate-kernel-features  %s | FileCheck -check-prefix=AKF_GCN %s
3; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-attributor %s | FileCheck -check-prefix=ATTRIBUTOR_GCN %s
4
5; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
6
7target datalayout = "A5"
8
9define internal void @indirect() {
10; AKF_GCN-LABEL: define {{[^@]+}}@indirect() {
11; AKF_GCN-NEXT:    ret void
12;
13; ATTRIBUTOR_GCN-LABEL: define {{[^@]+}}@indirect
14; ATTRIBUTOR_GCN-SAME: () #[[ATTR0:[0-9]+]] {
15; ATTRIBUTOR_GCN-NEXT:    ret void
16;
17; GFX9-LABEL: indirect:
18; GFX9:       ; %bb.0:
19; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
20; GFX9-NEXT:    s_setpc_b64 s[30:31]
21  ret void
22}
23
24define amdgpu_kernel void @test_simple_indirect_call() {
25; AKF_GCN-LABEL: define {{[^@]+}}@test_simple_indirect_call
26; AKF_GCN-SAME: () #[[ATTR0:[0-9]+]] {
27; AKF_GCN-NEXT:    [[FPTR:%.*]] = alloca void ()*, align 8, addrspace(5)
28; AKF_GCN-NEXT:    [[FPTR_CAST:%.*]] = addrspacecast void ()* addrspace(5)* [[FPTR]] to void ()**
29; AKF_GCN-NEXT:    store void ()* @indirect, void ()** [[FPTR_CAST]], align 8
30; AKF_GCN-NEXT:    [[FP:%.*]] = load void ()*, void ()** [[FPTR_CAST]], align 8
31; AKF_GCN-NEXT:    call void [[FP]]()
32; AKF_GCN-NEXT:    ret void
33;
34; ATTRIBUTOR_GCN-LABEL: define {{[^@]+}}@test_simple_indirect_call
35; ATTRIBUTOR_GCN-SAME: () #[[ATTR1:[0-9]+]] {
36; ATTRIBUTOR_GCN-NEXT:    [[FPTR:%.*]] = alloca void ()*, align 8, addrspace(5)
37; ATTRIBUTOR_GCN-NEXT:    [[FPTR_CAST:%.*]] = addrspacecast void ()* addrspace(5)* [[FPTR]] to void ()**
38; ATTRIBUTOR_GCN-NEXT:    store void ()* @indirect, void ()** [[FPTR_CAST]], align 8
39; ATTRIBUTOR_GCN-NEXT:    [[FP:%.*]] = load void ()*, void ()** [[FPTR_CAST]], align 8
40; ATTRIBUTOR_GCN-NEXT:    call void [[FP]]()
41; ATTRIBUTOR_GCN-NEXT:    ret void
42;
43; GFX9-LABEL: test_simple_indirect_call:
44; GFX9:       ; %bb.0:
45; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x4
46; GFX9-NEXT:    s_add_u32 flat_scratch_lo, s6, s9
47; GFX9-NEXT:    s_addc_u32 flat_scratch_hi, s7, 0
48; GFX9-NEXT:    s_add_u32 s0, s0, s9
49; GFX9-NEXT:    s_addc_u32 s1, s1, 0
50; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
51; GFX9-NEXT:    s_lshr_b32 s4, s4, 16
52; GFX9-NEXT:    s_mul_i32 s4, s4, s5
53; GFX9-NEXT:    v_mul_lo_u32 v0, s4, v0
54; GFX9-NEXT:    s_getpc_b64 s[6:7]
55; GFX9-NEXT:    s_add_u32 s6, s6, indirect@rel32@lo+4
56; GFX9-NEXT:    s_addc_u32 s7, s7, indirect@rel32@hi+12
57; GFX9-NEXT:    v_mov_b32_e32 v3, s6
58; GFX9-NEXT:    v_mov_b32_e32 v4, s7
59; GFX9-NEXT:    v_mad_u32_u24 v0, v1, s5, v0
60; GFX9-NEXT:    v_add_lshl_u32 v0, v0, v2, 3
61; GFX9-NEXT:    s_mov_b32 s32, 0
62; GFX9-NEXT:    ds_write_b64 v0, v[3:4]
63; GFX9-NEXT:    s_swappc_b64 s[30:31], s[6:7]
64; GFX9-NEXT:    s_endpgm
65  %fptr = alloca void()*, addrspace(5)
66  %fptr.cast = addrspacecast void()* addrspace(5)* %fptr to void()**
67  store void()* @indirect, void()** %fptr.cast
68  %fp = load void()*, void()** %fptr.cast
69  call void %fp()
70  ret void
71}
72
73;.
74; AKF_GCN: attributes #[[ATTR0]] = { "amdgpu-calls" "amdgpu-stack-objects" }
75;.
76; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "uniform-work-group-size"="false" }
77; ATTRIBUTOR_GCN: attributes #[[ATTR1]] = { "uniform-work-group-size"="false" }
78;.
79