121
|
1 ; RUN: llc -march=amdgcn -mcpu=gfx901 -verify-machineinstrs -enable-packed-inlinable-literals < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
|
|
2 ; RUN: llc -march=amdgcn -mcpu=tonga -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_lshr_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_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[VLHS]]
|
|
10
|
|
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
|
|
12 ; CI: v_lshrrev_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
13 ; CI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}}
|
|
14 ; CIVI-DAG: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 16
|
|
15 ; CIVI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
16 define amdgpu_kernel void @s_lshr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> %lhs, <2 x i16> %rhs) #0 {
|
|
17 %result = lshr <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_lshr_v2i16:
|
|
23 ; GCN: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
|
|
24 ; GCN: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
|
|
25 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
|
|
26
|
|
27 ; VI: v_lshrrev_b16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
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
|
|
29 ; VI: v_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
30
|
|
31 ; CI: s_mov_b32 [[MASK:s[0-9]+]], 0xffff{{$}}
|
|
32 ; CI-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[LHS]]
|
|
33 ; CI-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 16, [[RHS]]
|
|
34 ; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
|
|
35 ; CI: v_and_b32_e32 v{{[0-9]+}}, [[MASK]], v{{[0-9]+}}
|
|
36 ; CI: v_bfe_u32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, 16
|
|
37 ; CI: v_lshrrev_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_or_b32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
40 define amdgpu_kernel void @v_lshr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
|
|
41 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
42 %tid.ext = sext i32 %tid to i64
|
|
43 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
44 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
45 %b_ptr = getelementptr <2 x i16>, <2 x i16> addrspace(1)* %in.gep, i32 1
|
|
46 %a = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
|
|
47 %b = load <2 x i16>, <2 x i16> addrspace(1)* %b_ptr
|
|
48 %result = lshr <2 x i16> %a, %b
|
|
49 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
|
|
50 ret void
|
|
51 }
|
|
52
|
|
53 ; GCN-LABEL: {{^}}lshr_v_s_v2i16:
|
|
54 ; GFX9: s_load_dword [[RHS:s[0-9]+]]
|
|
55 ; GFX9: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
|
|
56 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
|
|
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 {
|
|
58 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
59 %tid.ext = sext i32 %tid to i64
|
|
60 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
61 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
62 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
|
|
63 %result = lshr <2 x i16> %vgpr, %sgpr
|
|
64 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
|
|
65 ret void
|
|
66 }
|
|
67
|
|
68 ; GCN-LABEL: {{^}}lshr_s_v_v2i16:
|
|
69 ; GFX9: s_load_dword [[LHS:s[0-9]+]]
|
|
70 ; GFX9: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
|
|
71 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], [[LHS]]
|
|
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 {
|
|
73 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
74 %tid.ext = sext i32 %tid to i64
|
|
75 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
76 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
77 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
|
|
78 %result = lshr <2 x i16> %sgpr, %vgpr
|
|
79 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
|
|
80 ret void
|
|
81 }
|
|
82
|
|
83 ; GCN-LABEL: {{^}}lshr_imm_v_v2i16:
|
|
84 ; GCN: {{buffer|flat|global}}_load_dword [[RHS:v[0-9]+]]
|
|
85 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], [[RHS]], 8
|
|
86 define amdgpu_kernel void @lshr_imm_v_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
|
|
87 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
88 %tid.ext = sext i32 %tid to i64
|
|
89 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
90 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
91 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
|
|
92 %result = lshr <2 x i16> <i16 8, i16 8>, %vgpr
|
|
93 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
|
|
94 ret void
|
|
95 }
|
|
96
|
|
97 ; GCN-LABEL: {{^}}lshr_v_imm_v2i16:
|
|
98 ; GCN: {{buffer|flat|global}}_load_dword [[LHS:v[0-9]+]]
|
|
99 ; GFX9: v_pk_lshrrev_b16 [[RESULT:v[0-9]+]], 8, [[LHS]]
|
|
100 define amdgpu_kernel void @lshr_v_imm_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in) #0 {
|
|
101 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
102 %tid.ext = sext i32 %tid to i64
|
|
103 %in.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
104 %out.gep = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
105 %vgpr = load <2 x i16>, <2 x i16> addrspace(1)* %in.gep
|
|
106 %result = lshr <2 x i16> %vgpr, <i16 8, i16 8>
|
|
107 store <2 x i16> %result, <2 x i16> addrspace(1)* %out.gep
|
|
108 ret void
|
|
109 }
|
|
110
|
|
111 ; GCN-LABEL: {{^}}v_lshr_v4i16:
|
|
112 ; GCN: {{buffer|flat|global}}_load_dwordx2
|
|
113 ; GCN: {{buffer|flat|global}}_load_dwordx2
|
|
114 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
115 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
|
|
116 ; GCN: {{buffer|flat|global}}_store_dwordx2
|
|
117 define amdgpu_kernel void @v_lshr_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
|
|
118 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
119 %tid.ext = sext i32 %tid to i64
|
|
120 %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
121 %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
122 %b_ptr = getelementptr <4 x i16>, <4 x i16> addrspace(1)* %in.gep, i32 1
|
|
123 %a = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
|
|
124 %b = load <4 x i16>, <4 x i16> addrspace(1)* %b_ptr
|
|
125 %result = lshr <4 x i16> %a, %b
|
|
126 store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
|
|
127 ret void
|
|
128 }
|
|
129
|
|
130 ; GCN-LABEL: {{^}}lshr_v_imm_v4i16:
|
|
131 ; GCN: {{buffer|flat|global}}_load_dwordx2
|
|
132 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
|
|
133 ; GFX9: v_pk_lshrrev_b16 v{{[0-9]+}}, 8, v{{[0-9]+}}
|
|
134 ; GCN: {{buffer|flat|global}}_store_dwordx2
|
|
135 define amdgpu_kernel void @lshr_v_imm_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> addrspace(1)* %in) #0 {
|
|
136 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
137 %tid.ext = sext i32 %tid to i64
|
|
138 %in.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %in, i64 %tid.ext
|
|
139 %out.gep = getelementptr inbounds <4 x i16>, <4 x i16> addrspace(1)* %out, i64 %tid.ext
|
|
140 %vgpr = load <4 x i16>, <4 x i16> addrspace(1)* %in.gep
|
|
141 %result = lshr <4 x i16> %vgpr, <i16 8, i16 8, i16 8, i16 8>
|
|
142 store <4 x i16> %result, <4 x i16> addrspace(1)* %out.gep
|
|
143 ret void
|
|
144 }
|
|
145
|
|
146 declare i32 @llvm.amdgcn.workitem.id.x() #1
|
|
147
|
|
148 attributes #0 = { nounwind }
|
|
149 attributes #1 = { nounwind readnone }
|