1; RUN: llc -march=amdgcn -mcpu=verde -show-mc-encoding -verify-machineinstrs < %s | FileCheck %s
2; RUN: llc -march=amdgcn -mcpu=tonga -show-mc-encoding -verify-machineinstrs < %s | FileCheck %s
3
4; Example of a simple geometry shader loading vertex attributes from the
5; ESGS ring buffer
6
7; FIXME: Out of bounds immediate offset crashes
8
9; CHECK-LABEL: {{^}}main:
10; CHECK: buffer_load_dword {{v[0-9]+}}, off, {{s\[[0-9]+:[0-9]+\]}}, 0 glc slc
11; CHECK: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, 0 offen glc slc
12; CHECK: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, 0 idxen glc slc
13; CHECK: buffer_load_dword {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 idxen offen glc slc
14; CHECK: s_movk_i32 [[K:s[0-9]+]], 0x4d2 ; encoding
15; CHECK: buffer_load_dword {{v[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, [[K]] idxen offen offset:65535 glc slc
16
17define amdgpu_vs void @main([17 x <4 x i32>] addrspace(4)* byval %arg, [32 x <4 x i32>] addrspace(4)* byval %arg1, [16 x <32 x i8>] addrspace(4)* byval %arg2, [2 x <4 x i32>] addrspace(4)* byval %arg3, [17 x <4 x i32>] addrspace(4)* inreg %arg4, [17 x <4 x i32>] addrspace(4)* inreg %arg5, i32 %arg6, i32 %arg7, i32 %arg8, i32 %arg9) {
18main_body:
19  %tmp = getelementptr [2 x <4 x i32>], [2 x <4 x i32>] addrspace(4)* %arg3, i64 0, i32 1
20  %tmp10 = load <4 x i32>, <4 x i32> addrspace(4)* %tmp, !tbaa !0
21  %tmp11 = shl i32 %arg6, 2
22  %tmp12 = call i32 @llvm.SI.buffer.load.dword.i32.i32(<4 x i32> %tmp10, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 0)
23  %tmp13 = bitcast i32 %tmp12 to float
24  %tmp14 = call i32 @llvm.SI.buffer.load.dword.i32.i32(<4 x i32> %tmp10, i32 %tmp11, i32 0, i32 0, i32 1, i32 0, i32 1, i32 1, i32 0)
25  %tmp15 = bitcast i32 %tmp14 to float
26  %tmp16 = call i32 @llvm.SI.buffer.load.dword.i32.i32(<4 x i32> %tmp10, i32 %tmp11, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 0)
27  %tmp17 = bitcast i32 %tmp16 to float
28  %tmp18 = call i32 @llvm.SI.buffer.load.dword.i32.v2i32(<4 x i32> %tmp10, <2 x i32> zeroinitializer, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 0)
29  %tmp19 = bitcast i32 %tmp18 to float
30
31  %tmp20 = call i32 @llvm.SI.buffer.load.dword.i32.v2i32(<4 x i32> %tmp10, <2 x i32> zeroinitializer, i32 0, i32 123, i32 1, i32 1, i32 1, i32 1, i32 0)
32  %tmp21 = bitcast i32 %tmp20 to float
33
34  %tmp22 = call i32 @llvm.SI.buffer.load.dword.i32.v2i32(<4 x i32> %tmp10, <2 x i32> zeroinitializer, i32 1234, i32 65535, i32 1, i32 1, i32 1, i32 1, i32 0)
35  %tmp23 = bitcast i32 %tmp22 to float
36
37  call void @llvm.amdgcn.exp.f32(i32 15, i32 12, float %tmp13, float %tmp15, float %tmp17, float %tmp19, i1 false, i1 false)
38  call void @llvm.amdgcn.exp.f32(i32 15, i32 12, float %tmp21, float %tmp23, float %tmp23, float %tmp23, i1 true, i1 false)
39  ret void
40}
41
42; Function Attrs: nounwind readonly
43declare i32 @llvm.SI.buffer.load.dword.i32.i32(<4 x i32>, i32, i32, i32, i32, i32, i32, i32, i32) #0
44
45; Function Attrs: nounwind readonly
46declare i32 @llvm.SI.buffer.load.dword.i32.v2i32(<4 x i32>, <2 x i32>, i32, i32, i32, i32, i32, i32, i32) #0
47
48declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #1
49
50attributes #0 = { nounwind readonly }
51attributes #1 = { nounwind inaccessiblememonly }
52
53!0 = !{!"const", !1, i32 1}
54!1 = !{!"tbaa root"}
55