annotate test/CodeGen/AMDGPU/lshr.v2i16.ll @ 128:c347d3398279 default tip

fix
author mir3636
date Wed, 06 Dec 2017 14:37:17 +0900
parents 803732b1fca8
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
121
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
1 ; RUN: llc -march=amdgcn -mcpu=gfx901 -verify-machineinstrs -enable-packed-inlinable-literals < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=CIVI %s
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
3 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=CIVI %s
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
4
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
5 ; GCN-LABEL: {{^}}s_lshr_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
6 ; GFX9: s_load_dword [[LHS:s[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
7 ; GFX9: s_load_dword [[RHS:s[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
8 ; GFX9: v_mov_b32_e32 [[VLHS:v[0-9]+]], [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
9 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[VLHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
10
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
11 ; VI: v_lshrrev_b32_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
12 ; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
13 ; CI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
14 ; CIVI-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 16
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
15 ; CIVI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
16 define amdgpu_kernel void @s_lshr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> %lhs, <2 x i16> %rhs) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
17 %result = lshr <2 x i16> %lhs, %rhs
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
18 store <2 x i16> %result, <2 x i16> addrspace(1)* %out
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
19 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
20 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
21
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
22 ; GCN-LABEL: {{^}}v_lshr_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
23 ; GCN: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
24 ; GCN: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
25 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
26
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
27 ; VI: v_lshrrev_b16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
28 ; VI: v_lshrrev_b16_sdwa v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
29 ; VI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
30
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
31 ; CI: s_mov_b32 [[MASK:s[0-9]+]], 0xffff{{$}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
32 ; CI-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
33 ; CI-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[RHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
34 ; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
35 ; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
36 ; CI: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 16
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
37 ; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
38 ; CI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
39 ; CI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
40 define amdgpu_kernel void @v_lshr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
41 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
42 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
43 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
44 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
45 %b_ptr = getelementptr <2 x i16>, <2 x i16> addrspace(1)* %in.gep, i32 1
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
46 %a = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
47 %b = load <2 x i16>, <2 x i16> addrspace(1)* %b_ptr
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
48 %result = lshr <2 x i16> %a, %b
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
49 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
50 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
51 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
52
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
53 ; GCN-LABEL: {{^}}lshr_v_s_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
54 ; GFX9: s_load_dword [[RHS:s[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
55 ; GFX9: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
56 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
57 define amdgpu_kernel void @lshr_v_s_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, <2 x i16> %sgpr) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
58 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
59 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
60 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
61 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
62 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
63 %result = lshr <2 x i16> %vgpr, %sgpr
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
64 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
65 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
66 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
67
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
68 ; GCN-LABEL: {{^}}lshr_s_v_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
69 ; GFX9: s_load_dword [[LHS:s[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
70 ; GFX9: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
71 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
72 define amdgpu_kernel void @lshr_s_v_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in, <2 x i16> %sgpr) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
73 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
74 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
75 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
76 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
77 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
78 %result = lshr <2 x i16> %sgpr, %vgpr
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
79 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
80 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
81 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
82
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
83 ; GCN-LABEL: {{^}}lshr_imm_v_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
84 ; GCN: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
85 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], 8
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
86 define amdgpu_kernel void @lshr_imm_v_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
87 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
88 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
89 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
90 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
91 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
92 %result = lshr <2 x i16> <i16 8, i16 8>, %vgpr
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
93 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
94 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
95 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
96
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
97 ; GCN-LABEL: {{^}}lshr_v_imm_v2i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
98 ; GCN: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
99 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], 8, [[LHS]]
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
100 define amdgpu_kernel void @lshr_v_imm_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
101 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
102 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
103 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
104 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
105 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
106 %result = lshr <2 x i16> %vgpr, <i16 8, i16 8>
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
107 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
108 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
109 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
110
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
111 ; GCN-LABEL: {{^}}v_lshr_v4i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
112 ; GCN: {{buffer|flat|global}}_load_dwordx2
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
113 ; GCN: {{buffer|flat|global}}_load_dwordx2
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
114 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
115 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
116 ; GCN: {{buffer|flat|global}}_store_dwordx2
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
117 define amdgpu_kernel void @v_lshr_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
118 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
119 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
120 %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
121 %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
122 %b_ptr = getelementptr <4 x i16>, <4 x i16> addrspace(1)* %in.gep, i32 1
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
123 %a = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
124 %b = load <4 x i16>, <4 x i16> addrspace(1)* %b_ptr
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
125 %result = lshr <4 x i16> %a, %b
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
126 store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
127 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
128 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
129
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
130 ; GCN-LABEL: {{^}}lshr_v_imm_v4i16:
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
131 ; GCN: {{buffer|flat|global}}_load_dwordx2
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
132 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
133 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
134 ; GCN: {{buffer|flat|global}}_store_dwordx2
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
135 define amdgpu_kernel void @lshr_v_imm_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
136 %tid = call i32 @llvm.amdgcn.workitem.id.x()
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
137 %tid.ext = sext i32 %tid to i64
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
138 %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
139 %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
140 %vgpr = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
141 %result = lshr <4 x i16> %vgpr, <i16 8, i16 8, i16 8, i16 8>
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
142 store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
143 ret void
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
144 }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
145
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
146 declare i32 @llvm.amdgcn.workitem.id.x() #1
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
147
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
148 attributes #0 = { nounwind }
803732b1fca8 LLVM 5.0
kono
parents:
diff changeset
149 attributes #1 = { nounwind readnone }