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.
7 ; DEFAULT-LABEL: {{^}}main:
8 ; DEFAULT: s_load_dwordx8
9 ; DEFAULT: s_waitcnt lgkmcnt(0)
10 ; DEFAULT: buffer_load_format_xyzw
11 ; DEFAULT: buffer_load_format_xyzw
12 ; DEFAULT-DAG: s_waitcnt vmcnt(0)
15 ; DEFAULT-NEXT: s_endpgm
16 define amdgpu_vs void @main(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, ptr addrspace(4) inreg %arg3, ptr addrspace(4) inreg %arg4, i32 inreg %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9, ptr addrspace(4) inreg %constptr) #0 {
18 %tmp10 = load <16 x i8>, ptr addrspace(4) %arg3, !tbaa !0
19 %tmp10.cast.int = bitcast <16 x i8> %tmp10 to i128
20 %tmp10.cast = inttoptr i128 %tmp10.cast.int to ptr addrspace(8)
21 %tmp11 = call <4 x float> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f32(ptr addrspace(8) %tmp10.cast, i32 %arg6, i32 0, i32 0, i32 0)
22 %tmp12 = extractelement <4 x float> %tmp11, i32 0
23 %tmp13 = extractelement <4 x float> %tmp11, i32 1
24 call void @llvm.amdgcn.s.barrier() #1
25 %tmp14 = extractelement <4 x float> %tmp11, i32 2
26 %tmp15 = load float, ptr addrspace(4) %constptr, align 4
27 %tmp16 = getelementptr <16 x i8>, ptr addrspace(4) %arg3, i32 1
28 %tmp17 = load <16 x i8>, ptr addrspace(4) %tmp16, !tbaa !0
29 %tmp17.cast.int = bitcast <16 x i8> %tmp17 to i128
30 %tmp17.cast = inttoptr i128 %tmp17.cast.int to ptr addrspace(8)
31 %tmp18 = call <4 x float> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f32(ptr addrspace(8) %tmp17.cast, i32 %arg6, i32 0, i32 0, i32 0)
32 %tmp19 = extractelement <4 x float> %tmp18, i32 0
33 %tmp20 = extractelement <4 x float> %tmp18, i32 1
34 %tmp21 = extractelement <4 x float> %tmp18, i32 2
35 %tmp22 = extractelement <4 x float> %tmp18, i32 3
36 call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %tmp19, float %tmp20, float %tmp21, float %tmp22, i1 false, i1 false) #0
37 call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float %tmp12, float %tmp13, float %tmp14, float %tmp15, i1 true, i1 false) #0
41 ; ILPMAX-LABEL: {{^}}main2:
42 ; ILPMAX: s_load_dwordx8
43 ; ILPMAX: s_waitcnt lgkmcnt(0)
46 ; ILPMAX: s_waitcnt vmcnt(0)
48 ; ILPMAX-NEXT: exp param0
50 define amdgpu_vs void @main2(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, ptr addrspace(4) inreg %arg3, ptr addrspace(4) inreg %arg4, i32 inreg %arg5, i32 inreg %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10) #0 {
52 %tmp11 = load <16 x i8>, ptr addrspace(4) %arg4, align 16, !tbaa !0
53 %tmp12 = add i32 %arg5, %arg7
54 %tmp11.cast.int = bitcast <16 x i8> %tmp11 to i128
55 %tmp11.cast = inttoptr i128 %tmp11.cast.int to ptr addrspace(8)
56 %tmp13 = call <4 x float> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f32(ptr addrspace(8) %tmp11.cast, i32 %tmp12, i32 0, i32 0, i32 0)
57 %tmp14 = extractelement <4 x float> %tmp13, i32 0
58 %tmp15 = extractelement <4 x float> %tmp13, i32 1
59 %tmp16 = extractelement <4 x float> %tmp13, i32 2
60 %tmp17 = extractelement <4 x float> %tmp13, i32 3
61 %tmp18 = getelementptr [16 x <16 x i8>], ptr addrspace(4) %arg4, i64 0, i64 1
62 %tmp19 = load <16 x i8>, ptr addrspace(4) %tmp18, align 16, !tbaa !0
63 %tmp20 = add i32 %arg5, %arg7
64 %tmp19.cast.int = bitcast <16 x i8> %tmp19 to i128
65 %tmp19.cast = inttoptr i128 %tmp19.cast.int to ptr addrspace(8)
66 %tmp21 = call <4 x float> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f32(ptr addrspace(8) %tmp19.cast, i32 %tmp20, i32 0, i32 0, i32 0)
67 %tmp22 = extractelement <4 x float> %tmp21, i32 0
68 %tmp23 = extractelement <4 x float> %tmp21, i32 1
69 %tmp24 = extractelement <4 x float> %tmp21, i32 2
70 %tmp25 = extractelement <4 x float> %tmp21, i32 3
71 call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float %tmp14, float %tmp15, float %tmp16, float %tmp17, i1 false, i1 false) #0
72 call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %tmp22, float %tmp23, float %tmp24, float %tmp25, i1 true, i1 false) #0
76 declare void @llvm.amdgcn.s.barrier() #1
77 declare <4 x float> @llvm.amdgcn.struct.ptr.buffer.load.format.v4f32(ptr addrspace(8), i32, i32, i32, i32 immarg) #2
78 declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
80 attributes #0 = { nounwind }
81 attributes #1 = { convergent nounwind }
82 attributes #2 = { nounwind readonly }
84 !0 = !{!1, !1, i64 0, i32 1}