1 ; RUN: llc -march=amdgcn -mcpu=hawaii -enable-amdgpu-aa=0 -verify-machineinstrs -mattr=-promote-alloca,-load-store-opt < %s | FileCheck -check-prefix=GCN %s
3 @sPrivateStorage = internal addrspace(3) global [256 x [8 x <4 x i64>]] undef
5 ; GCN-LABEL: {{^}}ds_reorder_vector_split:
7 ; Write zeroinitializer
8 ; GCN-DAG: ds_write_b64 [[PTR:v[0-9]+]], [[VAL:v\[[0-9]+:[0-9]+\]]] offset:24
9 ; GCN-DAG: ds_write_b64 [[PTR]], [[VAL]] offset:16
10 ; GCN-DAG: ds_write_b64 [[PTR]], [[VAL]] offset:8
11 ; GCN-DAG: ds_write_b64 [[PTR]], [[VAL]]{{$}}
13 ; GCN: s_waitcnt vmcnt
15 ; GCN-DAG: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:24
16 ; GCN-DAG: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:16
17 ; GCN-DAG: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:8
18 ; Appears to be dead store of vector component.
19 ; GCN-DAG: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]$}}
22 ; GCN-DAG: ds_read_b64 {{v\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:8
23 ; GCN-DAG: ds_read_b64 {{v\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:16
24 ; GCN-DAG: ds_read_b64 {{v\[[0-9]+:[0-9]+\]}}, v{{[0-9]+}} offset:24
26 ; GCN-DAG: buffer_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
27 ; GCN-DAG: buffer_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8
28 ; GCN-DAG: buffer_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:16
29 ; GCN-DAG: buffer_store_dwordx2 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:24
32 define amdgpu_kernel void @ds_reorder_vector_split(<4 x i64> addrspace(1)* nocapture readonly %srcValues, i32 addrspace(1)* nocapture readonly %offsets, <4 x i64> addrspace(1)* nocapture %destBuffer, i32 %alignmentOffset) #0 {
34 %tmp = tail call i32 @llvm.r600.read.local.size.y()
35 %tmp1 = tail call i32 @llvm.r600.read.local.size.z()
36 %tmp2 = tail call i32 @llvm.amdgcn.workitem.id.x()
37 %tmp3 = tail call i32 @llvm.amdgcn.workitem.id.y()
38 %tmp4 = tail call i32 @llvm.amdgcn.workitem.id.z()
39 %tmp6 = mul i32 %tmp2, %tmp
40 %tmp10 = add i32 %tmp3, %tmp6
41 %tmp11 = mul i32 %tmp10, %tmp1
42 %tmp9 = add i32 %tmp11, %tmp4
43 %x.i.i = tail call i32 @llvm.amdgcn.workgroup.id.x() #1
44 %x.i.12.i = tail call i32 @llvm.r600.read.local.size.x() #1
45 %mul.26.i = mul i32 %x.i.12.i, %x.i.i
46 %add.i = add i32 %tmp2, %mul.26.i
47 %arrayidx = getelementptr [256 x [8 x <4 x i64>]], [256 x [8 x <4 x i64>]] addrspace(3)* @sPrivateStorage, i32 0, i32 %tmp9, i32 %add.i
48 store <4 x i64> zeroinitializer, <4 x i64> addrspace(3)* %arrayidx
49 %tmp12 = sext i32 %add.i to i64
50 %arrayidx1 = getelementptr inbounds <4 x i64>, <4 x i64> addrspace(1)* %srcValues, i64 %tmp12
51 %tmp13 = load <4 x i64>, <4 x i64> addrspace(1)* %arrayidx1
52 %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %offsets, i64 %tmp12
53 %tmp14 = load i32, i32 addrspace(1)* %arrayidx2
54 %add.ptr = getelementptr [256 x [8 x <4 x i64>]], [256 x [8 x <4 x i64>]] addrspace(3)* @sPrivateStorage, i32 0, i32 %tmp9, i32 0, i32 %alignmentOffset
55 %mul.i = shl i32 %tmp14, 2
56 %arrayidx.i = getelementptr inbounds i64, i64 addrspace(3)* %add.ptr, i32 %mul.i
57 %tmp15 = bitcast i64 addrspace(3)* %arrayidx.i to <4 x i64> addrspace(3)*
58 store <4 x i64> %tmp13, <4 x i64> addrspace(3)* %tmp15
59 %add.ptr6 = getelementptr [256 x [8 x <4 x i64>]], [256 x [8 x <4 x i64>]] addrspace(3)* @sPrivateStorage, i32 0, i32 %tmp9, i32 %tmp14, i32 %alignmentOffset
60 %tmp16 = sext i32 %tmp14 to i64
61 %tmp17 = sext i32 %alignmentOffset to i64
62 %add.ptr9 = getelementptr inbounds <4 x i64>, <4 x i64> addrspace(1)* %destBuffer, i64 %tmp16, i64 %tmp17
63 %tmp18 = bitcast <4 x i64> %tmp13 to i256
64 %trunc = trunc i256 %tmp18 to i64
65 store i64 %trunc, i64 addrspace(1)* %add.ptr9
66 %arrayidx10.1 = getelementptr inbounds i64, i64 addrspace(3)* %add.ptr6, i32 1
67 %tmp19 = load i64, i64 addrspace(3)* %arrayidx10.1
68 %arrayidx11.1 = getelementptr inbounds i64, i64 addrspace(1)* %add.ptr9, i64 1
69 store i64 %tmp19, i64 addrspace(1)* %arrayidx11.1
70 %arrayidx10.2 = getelementptr inbounds i64, i64 addrspace(3)* %add.ptr6, i32 2
71 %tmp20 = load i64, i64 addrspace(3)* %arrayidx10.2
72 %arrayidx11.2 = getelementptr inbounds i64, i64 addrspace(1)* %add.ptr9, i64 2
73 store i64 %tmp20, i64 addrspace(1)* %arrayidx11.2
74 %arrayidx10.3 = getelementptr inbounds i64, i64 addrspace(3)* %add.ptr6, i32 3
75 %tmp21 = load i64, i64 addrspace(3)* %arrayidx10.3
76 %arrayidx11.3 = getelementptr inbounds i64, i64 addrspace(1)* %add.ptr9, i64 3
77 store i64 %tmp21, i64 addrspace(1)* %arrayidx11.3
81 ; Function Attrs: nounwind readnone
82 declare i32 @llvm.amdgcn.workgroup.id.x() #1
84 ; Function Attrs: nounwind readnone
85 declare i32 @llvm.r600.read.local.size.x() #1
87 ; Function Attrs: nounwind readnone
88 declare i32 @llvm.amdgcn.workitem.id.x() #1
90 ; Function Attrs: nounwind readnone
91 declare i32 @llvm.r600.read.local.size.y() #1
93 ; Function Attrs: nounwind readnone
94 declare i32 @llvm.r600.read.local.size.z() #1
96 ; Function Attrs: nounwind readnone
97 declare i32 @llvm.amdgcn.workitem.id.y() #1
99 ; Function Attrs: nounwind readnone
100 declare i32 @llvm.amdgcn.workitem.id.z() #1
102 attributes #0 = { norecurse nounwind }
103 attributes #1 = { nounwind readnone }