1; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2
3declare i32 @llvm.r600.read.tidig.x() nounwind readnone
4
5; FUNC-LABEL: {{^}}v_test_imax_sge_i32:
6; SI: v_max_i32_e32
7define 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: {{^}}v_test_imax_sge_v4i32:
21; SI: v_max_i32_e32
22; SI: v_max_i32_e32
23; SI: v_max_i32_e32
24; SI: v_max_i32_e32
25define void @v_test_imax_sge_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %aptr, <4 x i32> addrspace(1)* %bptr) nounwind {
26  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
27  %gep0 = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %aptr, i32 %tid
28  %gep1 = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %bptr, i32 %tid
29  %outgep = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %out, i32 %tid
30  %a = load <4 x i32>, <4 x i32> addrspace(1)* %gep0, align 4
31  %b = load <4 x i32>, <4 x i32> addrspace(1)* %gep1, align 4
32  %cmp = icmp sge <4 x i32> %a, %b
33  %val = select <4 x i1> %cmp, <4 x i32> %a, <4 x i32> %b
34  store <4 x i32> %val, <4 x i32> addrspace(1)* %outgep, align 4
35  ret void
36}
37
38; FUNC-LABEL: @s_test_imax_sge_i32
39; SI: s_max_i32
40define void @s_test_imax_sge_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
41  %cmp = icmp sge i32 %a, %b
42  %val = select i1 %cmp, i32 %a, i32 %b
43  store i32 %val, i32 addrspace(1)* %out, align 4
44  ret void
45}
46
47; FUNC-LABEL: {{^}}s_test_imax_sge_imm_i32:
48; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
49define void @s_test_imax_sge_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
50  %cmp = icmp sge i32 %a, 9
51  %val = select i1 %cmp, i32 %a, i32 9
52  store i32 %val, i32 addrspace(1)* %out, align 4
53  ret void
54}
55
56; FUNC-LABEL: {{^}}v_test_imax_sge_i8:
57; SI: buffer_load_sbyte
58; SI: buffer_load_sbyte
59; SI: v_max_i32_e32
60define void @v_test_imax_sge_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %aptr, i8 addrspace(1)* %bptr) nounwind {
61  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
62  %gep0 = getelementptr i8, i8 addrspace(1)* %aptr, i32 %tid
63  %gep1 = getelementptr i8, i8 addrspace(1)* %bptr, i32 %tid
64  %outgep = getelementptr i8, i8 addrspace(1)* %out, i32 %tid
65  %a = load i8, i8 addrspace(1)* %gep0, align 1
66  %b = load i8, i8 addrspace(1)* %gep1, align 1
67  %cmp = icmp sge i8 %a, %b
68  %val = select i1 %cmp, i8 %a, i8 %b
69  store i8 %val, i8 addrspace(1)* %outgep, align 1
70  ret void
71}
72
73; FUNC-LABEL: {{^}}s_test_imax_sgt_imm_i32:
74; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
75define void @s_test_imax_sgt_imm_i32(i32 addrspace(1)* %out, i32 %a) nounwind {
76  %cmp = icmp sgt i32 %a, 9
77  %val = select i1 %cmp, i32 %a, i32 9
78  store i32 %val, i32 addrspace(1)* %out, align 4
79  ret void
80}
81
82; FUNC-LABEL: {{^}}s_test_imax_sgt_imm_v2i32:
83; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
84; SI: s_max_i32 {{s[0-9]+}}, {{s[0-9]+}}, 9
85define void @s_test_imax_sgt_imm_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a) nounwind {
86  %cmp = icmp sgt <2 x i32> %a, <i32 9, i32 9>
87  %val = select <2 x i1> %cmp, <2 x i32> %a, <2 x i32> <i32 9, i32 9>
88  store <2 x i32> %val, <2 x i32> addrspace(1)* %out, align 4
89  ret void
90}
91; FUNC-LABEL: @v_test_imax_sgt_i32
92; SI: v_max_i32_e32
93define void @v_test_imax_sgt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
94  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
95  %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
96  %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
97  %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
98  %a = load i32, i32 addrspace(1)* %gep0, align 4
99  %b = load i32, i32 addrspace(1)* %gep1, align 4
100  %cmp = icmp sgt i32 %a, %b
101  %val = select i1 %cmp, i32 %a, i32 %b
102  store i32 %val, i32 addrspace(1)* %outgep, align 4
103  ret void
104}
105
106; FUNC-LABEL: @s_test_imax_sgt_i32
107; SI: s_max_i32
108define void @s_test_imax_sgt_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
109  %cmp = icmp sgt i32 %a, %b
110  %val = select i1 %cmp, i32 %a, i32 %b
111  store i32 %val, i32 addrspace(1)* %out, align 4
112  ret void
113}
114
115; FUNC-LABEL: @v_test_umax_uge_i32
116; SI: v_max_u32_e32
117define void @v_test_umax_uge_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
118  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
119  %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
120  %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
121  %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
122  %a = load i32, i32 addrspace(1)* %gep0, align 4
123  %b = load i32, i32 addrspace(1)* %gep1, align 4
124  %cmp = icmp uge i32 %a, %b
125  %val = select i1 %cmp, i32 %a, i32 %b
126  store i32 %val, i32 addrspace(1)* %outgep, align 4
127  ret void
128}
129
130; FUNC-LABEL: @s_test_umax_uge_i32
131; SI: s_max_u32
132define void @s_test_umax_uge_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
133  %cmp = icmp uge i32 %a, %b
134  %val = select i1 %cmp, i32 %a, i32 %b
135  store i32 %val, i32 addrspace(1)* %out, align 4
136  ret void
137}
138
139; FUNC-LABEL: {{^}}s_test_umax_uge_v3i32:
140; SI: s_max_u32
141; SI: s_max_u32
142; SI: s_max_u32
143; SI-NOT: s_max_u32
144; SI: s_endpgm
145define void @s_test_umax_uge_v3i32(<3 x i32> addrspace(1)* %out, <3 x i32> %a, <3 x i32> %b) nounwind {
146  %cmp = icmp uge <3 x i32> %a, %b
147  %val = select <3 x i1> %cmp, <3 x i32> %a, <3 x i32> %b
148  store <3 x i32> %val, <3 x i32> addrspace(1)* %out, align 4
149  ret void
150}
151
152; FUNC-LABEL: {{^}}v_test_umax_uge_i8:
153; SI: buffer_load_ubyte
154; SI: buffer_load_ubyte
155; SI: v_max_u32_e32
156define void @v_test_umax_uge_i8(i8 addrspace(1)* %out, i8 addrspace(1)* %aptr, i8 addrspace(1)* %bptr) nounwind {
157  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
158  %gep0 = getelementptr i8, i8 addrspace(1)* %aptr, i32 %tid
159  %gep1 = getelementptr i8, i8 addrspace(1)* %bptr, i32 %tid
160  %outgep = getelementptr i8, i8 addrspace(1)* %out, i32 %tid
161  %a = load i8, i8 addrspace(1)* %gep0, align 1
162  %b = load i8, i8 addrspace(1)* %gep1, align 1
163  %cmp = icmp uge i8 %a, %b
164  %val = select i1 %cmp, i8 %a, i8 %b
165  store i8 %val, i8 addrspace(1)* %outgep, align 1
166  ret void
167}
168
169; FUNC-LABEL: @v_test_umax_ugt_i32
170; SI: v_max_u32_e32
171define void @v_test_umax_ugt_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %aptr, i32 addrspace(1)* %bptr) nounwind {
172  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
173  %gep0 = getelementptr i32, i32 addrspace(1)* %aptr, i32 %tid
174  %gep1 = getelementptr i32, i32 addrspace(1)* %bptr, i32 %tid
175  %outgep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
176  %a = load i32, i32 addrspace(1)* %gep0, align 4
177  %b = load i32, i32 addrspace(1)* %gep1, align 4
178  %cmp = icmp ugt i32 %a, %b
179  %val = select i1 %cmp, i32 %a, i32 %b
180  store i32 %val, i32 addrspace(1)* %outgep, align 4
181  ret void
182}
183
184; FUNC-LABEL: {{^}}s_test_umax_ugt_i32:
185; SI: s_max_u32
186define void @s_test_umax_ugt_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind {
187  %cmp = icmp ugt i32 %a, %b
188  %val = select i1 %cmp, i32 %a, i32 %b
189  store i32 %val, i32 addrspace(1)* %out, align 4
190  ret void
191}
192
193; FUNC-LABEL: {{^}}s_test_umax_ugt_imm_v2i32:
194; SI: s_max_u32 {{s[0-9]+}}, {{s[0-9]+}}, 15
195; SI: s_max_u32 {{s[0-9]+}}, {{s[0-9]+}}, 23
196define void @s_test_umax_ugt_imm_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a) nounwind {
197  %cmp = icmp ugt <2 x i32> %a, <i32 15, i32 23>
198  %val = select <2 x i1> %cmp, <2 x i32> %a, <2 x i32> <i32 15, i32 23>
199  store <2 x i32> %val, <2 x i32> addrspace(1)* %out, align 4
200  ret void
201}
202
203; Make sure redundant and removed
204; FUNC-LABEL: {{^}}simplify_demanded_bits_test_umax_ugt_i16:
205; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
206; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
207; SI: s_max_u32 [[MAX:s[0-9]+]], [[A]], [[B]]
208; SI-NEXT: v_mov_b32_e32 [[VMAX:v[0-9]+]], [[MAX]]
209; SI-NEXT: buffer_store_dword [[VMAX]]
210define void @simplify_demanded_bits_test_umax_ugt_i16(i32 addrspace(1)* %out, i16 zeroext %a, i16 zeroext %b) nounwind {
211  %a.ext = zext i16 %a to i32
212  %b.ext = zext i16 %b to i32
213  %cmp = icmp ugt i32 %a.ext, %b.ext
214  %val = select i1 %cmp, i32 %a.ext, i32 %b.ext
215  %mask = and i32 %val, 65535
216  store i32 %mask, i32 addrspace(1)* %out
217  ret void
218}
219
220; Make sure redundant sign_extend_inreg removed.
221
222; FUNC-LABEL: {{^}}simplify_demanded_bits_test_max_slt_i16:
223; SI-DAG: s_load_dword [[A:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
224; SI-DAG: s_load_dword [[B:s[0-9]+]], {{s\[[0-9]+:[0-9]+\]}}, 0xc
225; SI: s_max_i32 [[MAX:s[0-9]+]], [[A]], [[B]]
226; SI-NEXT: v_mov_b32_e32 [[VMAX:v[0-9]+]], [[MAX]]
227; SI-NEXT: buffer_store_dword [[VMAX]]
228define void @simplify_demanded_bits_test_max_slt_i16(i32 addrspace(1)* %out, i16 signext %a, i16 signext %b) nounwind {
229  %a.ext = sext i16 %a to i32
230  %b.ext = sext i16 %b to i32
231  %cmp = icmp sgt i32 %a.ext, %b.ext
232  %val = select i1 %cmp, i32 %a.ext, i32 %b.ext
233  %shl = shl i32 %val, 16
234  %sextinreg = ashr i32 %shl, 16
235  store i32 %sextinreg, i32 addrspace(1)* %out
236  ret void
237}
238
239; FUNC-LABEL: {{^}}s_test_imax_sge_i16:
240; SI: s_load_dword
241; SI: s_load_dword
242; SI: s_sext_i32_i16
243; SI: s_sext_i32_i16
244; SI: s_max_i32
245define void @s_test_imax_sge_i16(i16 addrspace(1)* %out, i16 %a, i16 %b) nounwind {
246  %cmp = icmp sge i16 %a, %b
247  %val = select i1 %cmp, i16 %a, i16 %b
248  store i16 %val, i16 addrspace(1)* %out
249  ret void
250}
251