xref: /llvm-project/llvm/test/CodeGen/AMDGPU/pal-simple-indirect-call.ll (revision d34a10a47d25103725174c62ed5e84e8ca380249)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
2; Check that no attributes are added to graphics functions
3; RUN: opt -S -mtriple=amdgcn-amd-amdpal -amdgpu-annotate-kernel-features  %s | FileCheck -check-prefixes=AKF_GCN %s
4; RUN: opt -S -mtriple=amdgcn-amd-amdpal -passes=amdgpu-attributor %s | FileCheck -check-prefixes=ATTRIBUTOR_GCN %s
5
6; Check that it doesn't crash
7; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9 %s
8; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10 %s
9; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1010 -global-isel < %s | FileCheck -check-prefixes=GFX10 %s
10
11target datalayout = "A5"
12
13
14define amdgpu_cs void @test_simple_indirect_call() {
15; AKF_GCN-LABEL: define {{[^@]+}}@test_simple_indirect_call() {
16; AKF_GCN-NEXT:    [[PC:%.*]] = call i64 @llvm.amdgcn.s.getpc()
17; AKF_GCN-NEXT:    [[FUN:%.*]] = inttoptr i64 [[PC]] to ptr
18; AKF_GCN-NEXT:    call amdgpu_gfx void [[FUN]]()
19; AKF_GCN-NEXT:    ret void
20;
21; ATTRIBUTOR_GCN-LABEL: define {{[^@]+}}@test_simple_indirect_call
22; ATTRIBUTOR_GCN-SAME: () #[[ATTR0:[0-9]+]] {
23; ATTRIBUTOR_GCN-NEXT:    [[PC:%.*]] = call i64 @llvm.amdgcn.s.getpc()
24; ATTRIBUTOR_GCN-NEXT:    [[FUN:%.*]] = inttoptr i64 [[PC]] to ptr
25; ATTRIBUTOR_GCN-NEXT:    call amdgpu_gfx void [[FUN]]()
26; ATTRIBUTOR_GCN-NEXT:    ret void
27;
28; Attributor adds work-group-size attribute. This should be ok.
29; GFX9-LABEL: test_simple_indirect_call:
30; GFX9:       ; %bb.0:
31; GFX9-NEXT:    s_getpc_b64 s[8:9]
32; GFX9-NEXT:    s_mov_b32 s8, s0
33; GFX9-NEXT:    s_load_dwordx4 s[8:11], s[8:9], 0x10
34; GFX9-NEXT:    s_getpc_b64 s[4:5]
35; GFX9-NEXT:    s_mov_b32 s32, 0
36; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
37; GFX9-NEXT:    s_add_u32 s8, s8, s0
38; GFX9-NEXT:    s_addc_u32 s9, s9, 0
39; GFX9-NEXT:    s_mov_b64 s[0:1], s[8:9]
40; GFX9-NEXT:    s_mov_b64 s[2:3], s[10:11]
41; GFX9-NEXT:    s_swappc_b64 s[30:31], s[4:5]
42; GFX9-NEXT:    s_endpgm
43; GFX10-LABEL: test_simple_indirect_call:
44; GFX10:       ; %bb.0:
45; GFX10-NEXT:    s_getpc_b64 s[8:9]
46; GFX10-NEXT:    s_mov_b32 s8, s0
47; GFX10-NEXT:    s_getpc_b64 s[4:5]
48; GFX10-NEXT:    s_load_dwordx4 s[8:11], s[8:9], 0x10
49; GFX10-NEXT:    s_mov_b32 s32, 0
50; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
51; GFX10-NEXT:    s_bitset0_b32 s11, 21
52; GFX10-NEXT:    s_add_u32 s8, s8, s0
53; GFX10-NEXT:    s_addc_u32 s9, s9, 0
54; GFX10-NEXT:    s_mov_b64 s[0:1], s[8:9]
55; GFX10-NEXT:    s_mov_b64 s[2:3], s[10:11]
56; GFX10-NEXT:    s_swappc_b64 s[30:31], s[4:5]
57; GFX10-NEXT:    s_endpgm
58
59
60  %pc = call i64 @llvm.amdgcn.s.getpc()
61  %fun = inttoptr i64 %pc to ptr
62  call amdgpu_gfx void %fun()
63  ret void
64}
65
66; Function Attrs: nounwind readnone speculatable willreturn
67declare i64 @llvm.amdgcn.s.getpc() #0
68
69attributes #0 = { nounwind readnone speculatable willreturn }
70;.
71; AKF_GCN: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
72;.
73; ATTRIBUTOR_GCN: attributes #[[ATTR0]] = { "uniform-work-group-size"="false" }
74; ATTRIBUTOR_GCN: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
75;.
76