1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -strict-whitespace %s --check-prefix=DEFAULT
2; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -strict-whitespace %s --check-prefix=DEFAULT
3; RUN: llc -march=amdgcn --misched=ilpmax -verify-machineinstrs < %s | FileCheck -strict-whitespace %s --check-prefix=ILPMAX
4; RUN: llc -march=amdgcn --misched=ilpmax -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -strict-whitespace %s --check-prefix=ILPMAX
5; The ilpmax scheduler is used for the second test to get the ordering we want for the test.
6
7; DEFAULT-LABEL: {{^}}main:
8; DEFAULT: s_load_dwordx4
9; DEFAULT: s_load_dwordx4
10; DEFAULT: s_waitcnt lgkmcnt(0)
11; DEFAULT: buffer_load_format_xyzw
12; DEFAULT: s_waitcnt vmcnt(0)
13; DEFAULT: buffer_load_format_xyzw
14; DEFAULT: s_waitcnt vmcnt(0)
15; DEFAULT: exp
16; DEFAULT-NEXT: exp
17; DEFAULT-NEXT: s_endpgm
18define amdgpu_vs void @main(<16 x i8> addrspace(4)* inreg %arg, <16 x i8> addrspace(4)* inreg %arg1, <32 x i8> addrspace(4)* inreg %arg2, <16 x i8> addrspace(4)* inreg %arg3, <16 x i8> addrspace(4)* inreg %arg4, i32 inreg %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, float addrspace(4)* inreg %constptr) #0 {
19main_body:
20  %tmp = getelementptr <16 x i8>, <16 x i8> addrspace(4)* %arg3, i32 0
21  %tmp10 = load <16 x i8>, <16 x i8> addrspace(4)* %tmp, !tbaa !0
22  %tmp10.cast = bitcast <16 x i8> %tmp10 to <4 x i32>
23  %tmp11 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %tmp10.cast, i32 %arg6, i32 0, i1 false, i1 false)
24  %tmp12 = extractelement <4 x float> %tmp11, i32 0
25  %tmp13 = extractelement <4 x float> %tmp11, i32 1
26  call void @llvm.amdgcn.s.barrier() #1
27  %tmp14 = extractelement <4 x float> %tmp11, i32 2
28  %tmp15 = load float, float addrspace(4)* %constptr, align 4
29  %tmp16 = getelementptr <16 x i8>, <16 x i8> addrspace(4)* %arg3, i32 1
30  %tmp17 = load <16 x i8>, <16 x i8> addrspace(4)* %tmp16, !tbaa !0
31  %tmp17.cast = bitcast <16 x i8> %tmp17 to <4 x i32>
32  %tmp18 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %tmp17.cast, i32 %arg6, i32 0, i1 false, i1 false)
33  %tmp19 = extractelement <4 x float> %tmp18, i32 0
34  %tmp20 = extractelement <4 x float> %tmp18, i32 1
35  %tmp21 = extractelement <4 x float> %tmp18, i32 2
36  %tmp22 = extractelement <4 x float> %tmp18, i32 3
37  call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %tmp19, float %tmp20, float %tmp21, float %tmp22, i1 false, i1 false) #0
38  call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float %tmp12, float %tmp13, float %tmp14, float %tmp15, i1 true, i1 false) #0
39  ret void
40}
41
42; ILPMAX-LABEL: {{^}}main2:
43; ILPMAX: s_load_dwordx4
44; ILPMAX: s_waitcnt lgkmcnt(0)
45; ILPMAX: buffer_load
46; ILPMAX: s_load_dwordx4
47; ILPMAX: s_waitcnt lgkmcnt(0)
48; ILPMAX: buffer_load
49; ILPMAX: s_waitcnt vmcnt(0)
50; ILPMAX: exp pos0
51; ILPMAX-NEXT: exp param0
52; ILPMAX: s_endpgm
53define amdgpu_vs void @main2([6 x <16 x i8>] addrspace(4)* inreg %arg, [17 x <16 x i8>] addrspace(4)* inreg %arg1, [17 x <4 x i32>] addrspace(4)* inreg %arg2, [34 x <8 x i32>] addrspace(4)* inreg %arg3, [16 x <16 x i8>] addrspace(4)* inreg %arg4, i32 inreg %arg5, i32 inreg %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10) #0 {
54main_body:
55  %tmp = getelementptr [16 x <16 x i8>], [16 x <16 x i8>] addrspace(4)* %arg4, i64 0, i64 0
56  %tmp11 = load <16 x i8>, <16 x i8> addrspace(4)* %tmp, align 16, !tbaa !0
57  %tmp12 = add i32 %arg5, %arg7
58  %tmp11.cast = bitcast <16 x i8> %tmp11 to <4 x i32>
59  %tmp13 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %tmp11.cast, i32 %tmp12, i32 0, i1 false, i1 false)
60  %tmp14 = extractelement <4 x float> %tmp13, i32 0
61  %tmp15 = extractelement <4 x float> %tmp13, i32 1
62  %tmp16 = extractelement <4 x float> %tmp13, i32 2
63  %tmp17 = extractelement <4 x float> %tmp13, i32 3
64  %tmp18 = getelementptr [16 x <16 x i8>], [16 x <16 x i8>] addrspace(4)* %arg4, i64 0, i64 1
65  %tmp19 = load <16 x i8>, <16 x i8> addrspace(4)* %tmp18, align 16, !tbaa !0
66  %tmp20 = add i32 %arg5, %arg7
67  %tmp19.cast = bitcast <16 x i8> %tmp19 to <4 x i32>
68  %tmp21 = call <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32> %tmp19.cast, i32 %tmp20, i32 0, i1 false, i1 false)
69  %tmp22 = extractelement <4 x float> %tmp21, i32 0
70  %tmp23 = extractelement <4 x float> %tmp21, i32 1
71  %tmp24 = extractelement <4 x float> %tmp21, i32 2
72  %tmp25 = extractelement <4 x float> %tmp21, i32 3
73  call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float %tmp14, float %tmp15, float %tmp16, float %tmp17, i1 false, i1 false) #0
74  call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %tmp22, float %tmp23, float %tmp24, float %tmp25, i1 true, i1 false) #0
75  ret void
76}
77
78declare void @llvm.amdgcn.s.barrier() #1
79declare <4 x float> @llvm.amdgcn.buffer.load.format.v4f32(<4 x i32>, i32, i32, i1, i1) #2
80declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
81
82attributes #0 = { nounwind }
83attributes #1 = { convergent nounwind }
84attributes #2 = { nounwind readonly }
85
86!0 = !{!1, !1, i64 0, i32 1}
87!1 = !{!"const", !2}
88!2 = !{!"tbaa root"}
89