Mercurial > hg > CbC > CbC_llvm

annotate test/CodeGen/AMDGPU/ctpop64.ll @ 147:c2174574ed3a

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 10
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Wed, 14 Aug 2019 16:55:33 +0900
parents 3a76565eade5
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
1 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
2 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
3
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
4 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6 declare i64 @llvm.ctpop.i64(i64) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 declare <4 x i64> @llvm.ctpop.v4i64(<4 x i64>) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 declare <8 x i64> @llvm.ctpop.v8i64(<8 x i64>) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 declare <16 x i64> @llvm.ctpop.v16i64(<16 x i64>) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
11
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
12 declare i65 @llvm.ctpop.i65(i65) nounwind readnone
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
13 declare i128 @llvm.ctpop.i128(i128) nounwind readnone
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
14
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
15 ; FUNC-LABEL: {{^}}s_ctpop_i64:
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 134
diff changeset
16 ; SI: s_load_dwordx2 [[SVAL:s\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, 0x13
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 134
diff changeset
17 ; VI: s_load_dwordx2 [[SVAL:s\[[0-9]+:[0-9]+\]]], s{{\[[0-9]+:[0-9]+\]}}, 0x4c
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
18 ; GCN: s_bcnt1_i32_b64 [[SRESULT:s[0-9]+]], [[SVAL]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
19 ; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
20 ; GCN: buffer_store_dword [[VRESULT]],
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
21 ; GCN: s_endpgm
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 134
diff changeset
22 define amdgpu_kernel void @s_ctpop_i64(i32 addrspace(1)* noalias %out, [8 x i32], i64 %val) nounwind {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 %truncctpop = trunc i64 %ctpop to i32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 store i32 %truncctpop, i32 addrspace(1)* %out, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29 ; FUNC-LABEL: {{^}}v_ctpop_i64:
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
30 ; GCN: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}},
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
31 ; GCN: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT:v[0-9]+]], v[[LOVAL]], 0
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
32 ; SI-NEXT: v_bcnt_u32_b32_e32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
33 ; VI-NEXT: v_bcnt_u32_b32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
34 ; GCN: buffer_store_dword [[RESULT]],
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
36 define amdgpu_kernel void @v_ctpop_i64(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
37 %tid = call i32 @llvm.r600.read.tidig.x()
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
38 %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
39 %val = load i64, i64 addrspace(1)* %in.gep, align 8
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40 %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41 %truncctpop = trunc i64 %ctpop to i32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 store i32 %truncctpop, i32 addrspace(1)* %out, align 4
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
43 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
44 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
45
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 ; FUNC-LABEL: {{^}}v_ctpop_i64_user:
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
47 ; GCN: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}},
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
48 ; GCN: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT:v[0-9]+]], v[[LOVAL]], 0
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
49 ; SI-NEXT: v_bcnt_u32_b32_e32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
50 ; VI-NEXT: v_bcnt_u32_b32 [[RESULT:v[0-9]+]], v[[HIVAL]], [[MIDRESULT]]
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
51 ; GCN-DAG: v_or_b32_e32 v[[RESULT_LO:[0-9]+]], s{{[0-9]+}}, [[RESULT]]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
52 ; GCN-DAG: v_mov_b32_e32 v[[RESULT_HI:[0-9]+]], s{{[0-9]+}}
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
53 ; GCN: buffer_store_dwordx2 v{{\[}}[[RESULT_LO]]:[[RESULT_HI]]{{\]}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
54 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
55 define amdgpu_kernel void @v_ctpop_i64_user(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in, i64 %s.val) nounwind {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
56 %tid = call i32 @llvm.r600.read.tidig.x()
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
57 %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
58 %val = load i64, i64 addrspace(1)* %in.gep, align 8
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
59 %ctpop = call i64 @llvm.ctpop.i64(i64 %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
60 %or = or i64 %ctpop, %s.val
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
61 store i64 %or, i64 addrspace(1)* %out
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
62 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
63 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 ; FUNC-LABEL: {{^}}s_ctpop_v2i64:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
67 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
68 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
69 define amdgpu_kernel void @s_ctpop_v2i64(<2 x i32> addrspace(1)* noalias %out, <2 x i64> %val) nounwind {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
70 %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71 %truncctpop = trunc <2 x i64> %ctpop to <2 x i32>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72 store <2 x i32> %truncctpop, <2 x i32> addrspace(1)* %out, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
73 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
74 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
75
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
76 ; FUNC-LABEL: {{^}}s_ctpop_v4i64:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
77 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
78 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
79 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
80 ; GCN: s_bcnt1_i32_b64
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
82 define amdgpu_kernel void @s_ctpop_v4i64(<4 x i32> addrspace(1)* noalias %out, <4 x i64> %val) nounwind {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83 %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 %truncctpop = trunc <4 x i64> %ctpop to <4 x i32>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 store <4 x i32> %truncctpop, <4 x i32> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
86 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
87 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89 ; FUNC-LABEL: {{^}}v_ctpop_v2i64:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
91 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
95 define amdgpu_kernel void @v_ctpop_v2i64(<2 x i32> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %in) nounwind {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
96 %tid = call i32 @llvm.r600.read.tidig.x()
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
97 %in.gep = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %in, i32 %tid
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
98 %val = load <2 x i64>, <2 x i64> addrspace(1)* %in.gep, align 16
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
99 %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
100 %truncctpop = trunc <2 x i64> %ctpop to <2 x i32>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
101 store <2 x i32> %truncctpop, <2 x i32> addrspace(1)* %out, align 8
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
102 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
103 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
104
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
105 ; FUNC-LABEL: {{^}}v_ctpop_v4i64:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
106 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
107 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
108 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
109 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
110 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
111 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
112 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
113 ; GCN: v_bcnt_u32_b32
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
114 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
115 define amdgpu_kernel void @v_ctpop_v4i64(<4 x i32> addrspace(1)* noalias %out, <4 x i64> addrspace(1)* noalias %in) nounwind {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
116 %tid = call i32 @llvm.r600.read.tidig.x()
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
117 %in.gep = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i32 %tid
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
118 %val = load <4 x i64>, <4 x i64> addrspace(1)* %in.gep, align 32
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
119 %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %val) nounwind readnone
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
120 %truncctpop = trunc <4 x i64> %ctpop to <4 x i32>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
121 store <4 x i32> %truncctpop, <4 x i32> addrspace(1)* %out, align 16
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
122 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
123 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
124
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
125 ; FUNC-LABEL: {{^}}ctpop_i64_in_br:
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
126 ; SI-DAG: s_load_dwordx2 s{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}}, s[{{[0-9]+:[0-9]+}}], 0xd
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
127 ; VI-DAG: s_load_dwordx2 s{{\[}}[[LOVAL:[0-9]+]]:[[HIVAL:[0-9]+]]{{\]}}, s[{{[0-9]+:[0-9]+}}], 0x34
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
128 ; GCN-DAG: s_bcnt1_i32_b64 [[RESULT:s[0-9]+]], {{s\[}}[[LOVAL]]:[[HIVAL]]{{\]}}
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
129 ; GCN-DAG: s_mov_b32 [[ZERO:s[0-9]+]], 0
100
7d135dc70f03 LLVM 3.9
Miyagi Mitsuki <e135756@ie.u-ryukyu.ac.jp>
parents: 95
diff changeset
130 ; GCN-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], [[RESULT]]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
131 ; GCN-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], [[ZERO]]
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
132 ; GCN: buffer_store_dwordx2 {{v\[}}[[VLO]]:[[VHI]]{{\]}}
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
133 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
134 define amdgpu_kernel void @ctpop_i64_in_br(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %ctpop_arg, i32 %cond) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
135 entry:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
136 %tmp0 = icmp eq i32 %cond, 0
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
137 br i1 %tmp0, label %if, label %else
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
138
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
139 if:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
140 %tmp2 = call i64 @llvm.ctpop.i64(i64 %ctpop_arg)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
141 br label %endif
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
142
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
143 else:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
144 %tmp3 = getelementptr i64, i64 addrspace(1)* %in, i32 1
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
145 %tmp4 = load i64, i64 addrspace(1)* %tmp3
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
146 br label %endif
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
147
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
148 endif:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
149 %tmp5 = phi i64 [%tmp2, %if], [%tmp4, %else]
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
150 store i64 %tmp5, i64 addrspace(1)* %out
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
151 ret void
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
152 }
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
153
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
154 ; FUNC-LABEL: {{^}}s_ctpop_i128:
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
155 ; GCN: s_bcnt1_i32_b64 [[SRESULT0:s[0-9]+]],
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
156 ; GCN: s_bcnt1_i32_b64 [[SRESULT1:s[0-9]+]],
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
157 ; GCN: s_add_i32 s{{[0-9]+}}, [[SRESULT1]], [[SRESULT0]]
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
158 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
159 define amdgpu_kernel void @s_ctpop_i128(i32 addrspace(1)* noalias %out, i128 %val) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
160 %ctpop = call i128 @llvm.ctpop.i128(i128 %val) nounwind readnone
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
161 %truncctpop = trunc i128 %ctpop to i32
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
162 store i32 %truncctpop, i32 addrspace(1)* %out, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
163 ret void
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
164 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
165
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
166 ; FUNC-LABEL: {{^}}s_ctpop_i65:
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
167 ; GCN: s_bcnt1_i32_b64 [[REG0:s[0-9]+]],
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
168 ; GCN: s_and_b32
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
169 ; GCN: s_bcnt1_i32_b64 [[REG1:s[0-9]+]],
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
170 ; GCN: s_add_i32 {{s[0-9]+}}, [[REG0]], [[REG1]]
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
171 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
172 define amdgpu_kernel void @s_ctpop_i65(i32 addrspace(1)* noalias %out, i65 %val) nounwind {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
173 %ctpop = call i65 @llvm.ctpop.i65(i65 %val) nounwind readnone
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
174 %truncctpop = trunc i65 %ctpop to i32
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
175 store i32 %truncctpop, i32 addrspace(1)* %out, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
176 ret void
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
177 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
178
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
179 ; FIXME: Should not have extra add
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
180
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
181 ; FUNC-LABEL: {{^}}v_ctpop_i128:
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
182 ; SI: buffer_load_dwordx4 v{{\[}}[[VAL0:[0-9]+]]:[[VAL3:[0-9]+]]{{\]}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
183 ; VI: flat_load_dwordx4 v{{\[}}[[VAL0:[0-9]+]]:[[VAL3:[0-9]+]]{{\]}}, v{{\[[0-9]+:[0-9]+\]}}
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
184
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
185 ; GCN-DAG: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT0:v[0-9]+]], v{{[0-9]+}}, 0
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
186 ; GCN-DAG: v_bcnt_u32_b32{{(_e32)*(_e64)*}} [[MIDRESULT1:v[0-9]+]], v[[VAL3]], [[MIDRESULT0]]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
187
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
188 ; GCN-DAG: v_bcnt_u32_b32{{(_e64)*}} [[MIDRESULT2:v[0-9]+]], v[[VAL0]], 0
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
189 ; GCN-DAG: v_bcnt_u32_b32{{(_e32)*(_e64)*}} [[MIDRESULT3:v[0-9]+]], v{{[0-9]+}}, [[MIDRESULT2]]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
190
147
c2174574ed3a LLVM 10
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 134
diff changeset
191 ; GCN: v_add_{{[iu]}}32_e32 [[RESULT:v[0-9]+]], vcc, [[MIDRESULT2]], [[MIDRESULT1]]
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
192
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
193 ; GCN: buffer_store_dword [[RESULT]],
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
194 ; GCN: s_endpgm
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
195 define amdgpu_kernel void @v_ctpop_i128(i32 addrspace(1)* noalias %out, i128 addrspace(1)* noalias %in) nounwind {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
196 %tid = call i32 @llvm.r600.read.tidig.x()
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
197 %in.gep = getelementptr i128, i128 addrspace(1)* %in, i32 %tid
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
198 %val = load i128, i128 addrspace(1)* %in.gep, align 8
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
199 %ctpop = call i128 @llvm.ctpop.i128(i128 %val) nounwind readnone
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
200 %truncctpop = trunc i128 %ctpop to i32
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
201 store i32 %truncctpop, i32 addrspace(1)* %out, align 4
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
202 ret void
1172e4bd9c6f update 4.0.0
mir3636
parents: 100
diff changeset
203 }