Mercurial > hg > Members > tobaru > cbc > CbC_llvm

comparison test/CodeGen/AMDGPU/max.ll @ 95:afa8332a0e37

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 3.8
author Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
date Tue, 13 Oct 2015 17:48:58 +0900
parents
children 7d135dc70f03
comparison
equal deleted inserted replaced
84:f3e34b893a5f 95:afa8332a0e37
1 ; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2
3 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
4
5 ; FUNC-LABEL: @v_test_imax_sge_i32
6 ; SI: v_max_i32_e32
7 define void @v_test_imax_sge_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
8 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
9 %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
10 %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
11 %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
12 %a = load i32, i32 addrspace(1)* %gep0, align 4
13 %b = load i32, i32 addrspace(1)* %gep1, align 4
14 %cmp = icmp sge i32 %a, %b
15 %val = select i1 %cmp, i32 %a, i32 %b
16 store i32 %val, i32 addrspace(1)* %outgep, align 4
17 ret void
18 }
19
20 ; FUNC-LABEL: @s_test_imax_sge_i32
21 ; SI: s_max_i32
22 define void @s_test_imax_sge_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
23 %cmp = icmp sge i32 %a, %b
24 %val = select i1 %cmp, i32 %a, i32 %b
25 store i32 %val, i32 addrspace(1)* %out, align 4
26 ret void
27 }
28
29 ; FUNC-LABEL: {{^}}s_test_imax_sge_imm_i32:
30 ; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
31 define void @s_test_imax_sge_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
32 %cmp = icmp sge i32 %a, 9
33 %val = select i1 %cmp, i32 %a, i32 9
34 store i32 %val, i32 addrspace(1)* %out, align 4
35 ret void
36 }
37
38 ; FUNC-LABEL: {{^}}s_test_imax_sgt_imm_i32:
39 ; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
40 define void @s_test_imax_sgt_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
41 %cmp = icmp sgt i32 %a, 9
42 %val = select i1 %cmp, i32 %a, i32 9
43 store i32 %val, i32 addrspace(1)* %out, align 4
44 ret void
45 }
46
47 ; FUNC-LABEL: @v_test_imax_sgt_i32
48 ; SI: v_max_i32_e32
49 define void @v_test_imax_sgt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
50 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
51 %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
52 %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
53 %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
54 %a = load i32, i32 addrspace(1)* %gep0, align 4
55 %b = load i32, i32 addrspace(1)* %gep1, align 4
56 %cmp = icmp sgt i32 %a, %b
57 %val = select i1 %cmp, i32 %a, i32 %b
58 store i32 %val, i32 addrspace(1)* %outgep, align 4
59 ret void
60 }
61
62 ; FUNC-LABEL: @s_test_imax_sgt_i32
63 ; SI: s_max_i32
64 define void @s_test_imax_sgt_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
65 %cmp = icmp sgt i32 %a, %b
66 %val = select i1 %cmp, i32 %a, i32 %b
67 store i32 %val, i32 addrspace(1)* %out, align 4
68 ret void
69 }
70
71 ; FUNC-LABEL: @v_test_umax_uge_i32
72 ; SI: v_max_u32_e32
73 define void @v_test_umax_uge_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
74 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
75 %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
76 %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
77 %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
78 %a = load i32, i32 addrspace(1)* %gep0, align 4
79 %b = load i32, i32 addrspace(1)* %gep1, align 4
80 %cmp = icmp uge i32 %a, %b
81 %val = select i1 %cmp, i32 %a, i32 %b
82 store i32 %val, i32 addrspace(1)* %outgep, align 4
83 ret void
84 }
85
86 ; FUNC-LABEL: @s_test_umax_uge_i32
87 ; SI: s_max_u32
88 define void @s_test_umax_uge_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
89 %cmp = icmp uge i32 %a, %b
90 %val = select i1 %cmp, i32 %a, i32 %b
91 store i32 %val, i32 addrspace(1)* %out, align 4
92 ret void
93 }
94
95 ; FUNC-LABEL: @v_test_umax_ugt_i32
96 ; SI: v_max_u32_e32
97 define void @v_test_umax_ugt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
98 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
99 %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
100 %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
101 %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
102 %a = load i32, i32 addrspace(1)* %gep0, align 4
103 %b = load i32, i32 addrspace(1)* %gep1, align 4
104 %cmp = icmp ugt i32 %a, %b
105 %val = select i1 %cmp, i32 %a, i32 %b
106 store i32 %val, i32 addrspace(1)* %outgep, align 4
107 ret void
108 }
109
110 ; FUNC-LABEL: @s_test_umax_ugt_i32
111 ; SI: s_max_u32
112 define void @s_test_umax_ugt_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
113 %cmp = icmp ugt i32 %a, %b
114 %val = select i1 %cmp, i32 %a, i32 %b
115 store i32 %val, i32 addrspace(1)* %out, align 4
116 ret void
117 }
118
119 ; Make sure redundant and removed
120 ; FUNC-LABEL: {{^}}simplify_demanded_bits_test_umax_ugt_i16:
121 ; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
122 ; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
123 ; SI: s_max_u32 [[MIN:s[0-9]+]], [[A]], [[B]]
124 ; SI-NEXT: v_mov_b32_e32 [[VMIN:v[0-9]+]], [[MIN]]
125 ; SI-NEXT: buffer_store_dword [[VMIN]]
126 define void @simplify_demanded_bits_test_umax_ugt_i16(i32 addrspace(1)* %out, i16 zeroext %a, i16 zeroext %b) nounwind {
127 %a.ext = zext i16 %a to i32
128 %b.ext = zext i16 %b to i32
129 %cmp = icmp ugt i32 %a.ext, %b.ext
130 %val = select i1 %cmp, i32 %a.ext, i32 %b.ext
131 %mask = and i32 %val, 65535
132 store i32 %mask, i32 addrspace(1)* %out
133 ret void
134 }
135
136 ; Make sure redundant sign_extend_inreg removed.
137
138 ; FUNC-LABEL: {{^}}simplify_demanded_bits_test_min_slt_i16:
139 ; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
140 ; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
141 ; SI: s_max_i32 [[MIN:s[0-9]+]], [[A]], [[B]]
142 ; SI-NEXT: v_mov_b32_e32 [[VMIN:v[0-9]+]], [[MIN]]
143 ; SI-NEXT: buffer_store_dword [[VMIN]]
144 define void @simplify_demanded_bits_test_min_slt_i16(i32 addrspace(1)* %out, i16 signext %a, i16 signext %b) nounwind {
145 %a.ext = sext i16 %a to i32
146 %b.ext = sext i16 %b to i32
147 %cmp = icmp sgt i32 %a.ext, %b.ext
148 %val = select i1 %cmp, i32 %a.ext, i32 %b.ext
149 %shl = shl i32 %val, 16
150 %sextinreg = ashr i32 %shl, 16
151 store i32 %sextinreg, i32 addrspace(1)* %out
152 ret void
153 }
154
155 ; FIXME: Should get match min/max through extends inserted by
156 ; legalization.
157
158 ; FUNC-LABEL: {{^}}s_test_imin_sge_i16:
159 ; SI: s_sext_i32_i16
160 ; SI: s_sext_i32_i16
161 ; SI: v_cmp_ge_i32_e32
162 ; SI: v_cndmask_b32
163 define void @s_test_imin_sge_i16(i16 addrspace(1)* %out, i16 %a, i16 %b) nounwind {
164 %cmp = icmp sge i16 %a, %b
165 %val = select i1 %cmp, i16 %a, i16 %b
166 store i16 %val, i16 addrspace(1)* %out
167 ret void
168 }