1; RUN: llc -march=amdgcn -mcpu=gfx901 -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
2; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=CIVI %s
3; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=CIVI %s
4
5; GCN-LABEL: {{^}}s_shl_v2i16:
6; GFX9: s_load_dword [[LHS:s[0-9]+]]
7; GFX9: s_load_dword [[RHS:s[0-9]+]]
8; GFX9: v_mov_b32_e32 [[VLHS:v[0-9]+]], [[LHS]]
9; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[VLHS]]
10
11; CIVI: v_lshlrev_b32_e32
12; CIVI: v_and_b32_e32 v{{[0-9]+}}, 0xffff, v{{[0-9]+}}
13; CIVI: v_lshlrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
14; CIVI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
15; CIVI: v_or_b32_e32
16define void @s_shl_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> %lhs, <2 x i16> %rhs) #0 {
17  %result = shl <2 x i16> %lhs, %rhs
18  store <2 x i16> %result, <2 x i16> addrspace(1)* %out
19  ret void
20}
21
22; GCN-LABEL: {{^}}v_shl_v2i16:
23; GCN: {{buffer|flat}}_load_dword [[LHS:v[0-9]+]]
24; GCN: {{buffer|flat}}_load_dword [[RHS:v[0-9]+]]
25; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
26
27; VI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
28; VI: v_lshlrev_b16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
29; VI: v_lshlrev_b16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
30; VI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
31; VI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
32
33; CI: s_mov_b32 [[MASK:s[0-9]+]], 0xffff{{$}}
34; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[LHS]]
35; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
36; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
37; CI: v_lshl_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
38; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
39; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
40; CI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
41define void @v_shl_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
42  %tid = call i32 @llvm.amdgcn.workitem.id.x()
43  %tid.ext = sext i32 %tid to i64
44  %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
45  %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
46  %b_ptr = getelementptr <2 x i16>, <2 x i16> addrspace(1)* %in.gep, i32 1
47  %a = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
48  %b = load <2 x i16>, <2 x i16> addrspace(1)* %b_ptr
49  %result = shl <2 x i16> %a, %b
50  store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
51  ret void
52}
53
54; GCN-LABEL: {{^}}shl_v_s_v2i16:
55; GFX9: s_load_dword [[RHS:s[0-9]+]]
56; GFX9: {{buffer|flat}}_load_dword [[LHS:v[0-9]+]]
57; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
58define void @shl_v_s_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, <2 x i16> %sgpr) #0 {
59  %tid = call i32 @llvm.amdgcn.workitem.id.x()
60  %tid.ext = sext i32 %tid to i64
61  %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
62  %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
63  %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
64  %result = shl <2 x i16> %vgpr, %sgpr
65  store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
66  ret void
67}
68
69; GCN-LABEL: {{^}}shl_s_v_v2i16:
70; GFX9: s_load_dword [[LHS:s[0-9]+]]
71; GFX9: {{buffer|flat}}_load_dword [[RHS:v[0-9]+]]
72; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
73define void @shl_s_v_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, <2 x i16> %sgpr) #0 {
74  %tid = call i32 @llvm.amdgcn.workitem.id.x()
75  %tid.ext = sext i32 %tid to i64
76  %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
77  %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
78  %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
79  %result = shl <2 x i16> %sgpr, %vgpr
80  store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
81  ret void
82}
83
84; GCN-LABEL: {{^}}shl_imm_v_v2i16:
85; GCN: {{buffer|flat}}_load_dword [[RHS:v[0-9]+]]
86; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], [[RHS]], 8
87define void @shl_imm_v_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
88  %tid = call i32 @llvm.amdgcn.workitem.id.x()
89  %tid.ext = sext i32 %tid to i64
90  %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
91  %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
92  %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
93  %result = shl <2 x i16> <i16 8, i16 8>, %vgpr
94  store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
95  ret void
96}
97
98; GCN-LABEL: {{^}}shl_v_imm_v2i16:
99; GCN: {{buffer|flat}}_load_dword [[LHS:v[0-9]+]]
100; GFX9: v_pk_lshlrev_b16 [[RESULT:v[0-9]+]], 8, [[LHS]]
101define void @shl_v_imm_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
102  %tid = call i32 @llvm.amdgcn.workitem.id.x()
103  %tid.ext = sext i32 %tid to i64
104  %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
105  %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
106  %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
107  %result = shl <2 x i16> %vgpr, <i16 8, i16 8>
108  store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
109  ret void
110}
111
112; GCN-LABEL: {{^}}v_shl_v4i16:
113; GCN: {{buffer|flat}}_load_dwordx2
114; GCN: {{buffer|flat}}_load_dwordx2
115; GFX9: v_pk_lshlrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
116; GFX9: v_pk_lshlrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
117; GCN: {{buffer|flat}}_store_dwordx2
118define void @v_shl_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
119  %tid = call i32 @llvm.amdgcn.workitem.id.x()
120  %tid.ext = sext i32 %tid to i64
121  %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
122  %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
123  %b_ptr = getelementptr <4 x i16>, <4 x i16> addrspace(1)* %in.gep, i32 1
124  %a = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
125  %b = load <4 x i16>, <4 x i16> addrspace(1)* %b_ptr
126  %result = shl <4 x i16> %a, %b
127  store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
128  ret void
129}
130
131; GCN-LABEL: {{^}}shl_v_imm_v4i16:
132; GCN: {{buffer|flat}}_load_dwordx2
133; GFX9: v_pk_lshlrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
134; GFX9: v_pk_lshlrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
135; GCN: {{buffer|flat}}_store_dwordx2
136define void @shl_v_imm_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
137  %tid = call i32 @llvm.amdgcn.workitem.id.x()
138  %tid.ext = sext i32 %tid to i64
139  %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
140  %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
141  %vgpr = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
142  %result = shl <4 x i16> %vgpr, <i16 8, i16 8, i16 8, i16 8>
143  store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
144  ret void
145}
146
147declare i32 @llvm.amdgcn.workitem.id.x() #1
148
149attributes #0 = { nounwind }
150attributes #1 = { nounwind readnone }
151