source: icGREP/icgrep-devel/llvm-3.8.0.src/test/CodeGen/AArch64/aarch64-minmaxv.ll @ 5027

Last change on this file since 5027 was 5027, checked in by cameron, 3 years ago

Upgrade to llvm 3.8

File size: 40.5 KB
Line 
1; RUN: llc -march=aarch64 -aarch64-neon-syntax=generic < %s | FileCheck %s
2
3target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
4target triple = "aarch64-linu--gnu"
5
6; CHECK-LABEL: smax_B
7; CHECK: smaxv {{b[0-9]+}}, {{v[0-9]+}}.16b
8define i8 @smax_B(<16 x i8>* nocapture readonly %arr)  {
9  %arr.load = load <16 x i8>, <16 x i8>* %arr
10  %rdx.shuf = shufflevector <16 x i8> %arr.load, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
11  %rdx.minmax.cmp22 = icmp sgt <16 x i8> %arr.load, %rdx.shuf
12  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i8> %arr.load, <16 x i8> %rdx.shuf
13  %rdx.shuf24 = shufflevector <16 x i8> %rdx.minmax.select23, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
14  %rdx.minmax.cmp25 = icmp sgt <16 x i8> %rdx.minmax.select23, %rdx.shuf24
15  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i8> %rdx.minmax.select23, <16 x i8> %rdx.shuf24
16  %rdx.shuf27 = shufflevector <16 x i8> %rdx.minmax.select26, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
17  %rdx.minmax.cmp28 = icmp sgt <16 x i8> %rdx.minmax.select26, %rdx.shuf27
18  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i8> %rdx.minmax.select26, <16 x i8> %rdx.shuf27
19  %rdx.shuf30 = shufflevector <16 x i8> %rdx.minmax.select29, <16 x i8> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
20  %rdx.minmax.cmp31 = icmp sgt <16 x i8> %rdx.minmax.select29, %rdx.shuf30
21  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
22  %rdx.minmax.select29.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 0
23  %rdx.shuf30.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 1
24  %r = select i1 %rdx.minmax.cmp31.elt, i8 %rdx.minmax.select29.elt, i8 %rdx.shuf30.elt
25  ret i8 %r
26}
27
28; CHECK-LABEL: smax_H
29; CHECK: smaxv {{h[0-9]+}}, {{v[0-9]+}}.8h
30define i16 @smax_H(<8 x i16>* nocapture readonly %arr) {
31  %rdx.minmax.select = load <8 x i16>, <8 x i16>* %arr
32  %rdx.shuf = shufflevector <8 x i16> %rdx.minmax.select, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
33  %rdx.minmax.cmp23 = icmp sgt <8 x i16> %rdx.minmax.select, %rdx.shuf
34  %rdx.minmax.select24 = select <8 x i1> %rdx.minmax.cmp23, <8 x i16> %rdx.minmax.select, <8 x i16> %rdx.shuf
35  %rdx.shuf25 = shufflevector <8 x i16> %rdx.minmax.select24, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
36  %rdx.minmax.cmp26 = icmp sgt <8 x i16> %rdx.minmax.select24, %rdx.shuf25
37  %rdx.minmax.select27 = select <8 x i1> %rdx.minmax.cmp26, <8 x i16> %rdx.minmax.select24, <8 x i16> %rdx.shuf25
38  %rdx.shuf28 = shufflevector <8 x i16> %rdx.minmax.select27, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
39  %rdx.minmax.cmp29 = icmp sgt <8 x i16> %rdx.minmax.select27, %rdx.shuf28
40  %rdx.minmax.cmp29.elt = extractelement <8 x i1> %rdx.minmax.cmp29, i32 0
41  %rdx.minmax.select27.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 0
42  %rdx.shuf28.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 1
43  %r = select i1 %rdx.minmax.cmp29.elt, i16 %rdx.minmax.select27.elt, i16 %rdx.shuf28.elt
44  ret i16 %r
45}
46
47; CHECK-LABEL: smax_S
48; CHECK: smaxv {{s[0-9]+}}, {{v[0-9]+}}.4s
49define i32 @smax_S(<4 x i32> * nocapture readonly %arr)  {
50  %rdx.minmax.select = load <4 x i32>, <4 x i32>* %arr
51  %rdx.shuf = shufflevector <4 x i32> %rdx.minmax.select, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
52  %rdx.minmax.cmp18 = icmp sgt <4 x i32> %rdx.minmax.select, %rdx.shuf
53  %rdx.minmax.select19 = select <4 x i1> %rdx.minmax.cmp18, <4 x i32> %rdx.minmax.select, <4 x i32> %rdx.shuf
54  %rdx.shuf20 = shufflevector <4 x i32> %rdx.minmax.select19, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
55  %rdx.minmax.cmp21 = icmp sgt <4 x i32> %rdx.minmax.select19, %rdx.shuf20
56  %rdx.minmax.cmp21.elt = extractelement <4 x i1> %rdx.minmax.cmp21, i32 0
57  %rdx.minmax.select19.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 0
58  %rdx.shuf20.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 1
59  %r = select i1 %rdx.minmax.cmp21.elt, i32 %rdx.minmax.select19.elt, i32 %rdx.shuf20.elt
60  ret i32 %r
61}
62
63; CHECK-LABEL: smax_D
64; CHECK-NOT: smaxv
65define i64 @smax_D(<2 x i64>* nocapture readonly %arr) {
66  %rdx.minmax.select = load <2 x i64>, <2 x i64>* %arr
67  %rdx.shuf = shufflevector <2 x i64> %rdx.minmax.select, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
68  %rdx.minmax.cmp18 = icmp sgt <2 x i64> %rdx.minmax.select, %rdx.shuf
69  %rdx.minmax.cmp18.elt = extractelement <2 x i1> %rdx.minmax.cmp18, i32 0
70  %rdx.minmax.select.elt = extractelement <2 x i64> %rdx.minmax.select, i32 0
71  %rdx.shuf.elt = extractelement <2 x i64> %rdx.minmax.select, i32 1
72  %r = select i1 %rdx.minmax.cmp18.elt, i64 %rdx.minmax.select.elt, i64 %rdx.shuf.elt
73  ret i64 %r
74}
75
76
77; CHECK-LABEL: umax_B
78; CHECK: umaxv {{b[0-9]+}}, {{v[0-9]+}}.16b
79define i8 @umax_B(<16 x i8>* nocapture readonly %arr)  {
80  %rdx.minmax.select = load <16 x i8>, <16 x i8>* %arr
81  %rdx.shuf = shufflevector <16 x i8> %rdx.minmax.select, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
82  %rdx.minmax.cmp22 = icmp ugt <16 x i8> %rdx.minmax.select, %rdx.shuf
83  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i8> %rdx.minmax.select, <16 x i8> %rdx.shuf
84  %rdx.shuf24 = shufflevector <16 x i8> %rdx.minmax.select23, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
85  %rdx.minmax.cmp25 = icmp ugt <16 x i8> %rdx.minmax.select23, %rdx.shuf24
86  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i8> %rdx.minmax.select23, <16 x i8> %rdx.shuf24
87  %rdx.shuf27 = shufflevector <16 x i8> %rdx.minmax.select26, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
88  %rdx.minmax.cmp28 = icmp ugt <16 x i8> %rdx.minmax.select26, %rdx.shuf27
89  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i8> %rdx.minmax.select26, <16 x i8> %rdx.shuf27
90  %rdx.shuf30 = shufflevector <16 x i8> %rdx.minmax.select29, <16 x i8> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
91  %rdx.minmax.cmp31 = icmp ugt <16 x i8> %rdx.minmax.select29, %rdx.shuf30
92  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
93  %rdx.minmax.select29.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 0
94  %rdx.shuf30.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 1
95  %r = select i1 %rdx.minmax.cmp31.elt, i8 %rdx.minmax.select29.elt, i8 %rdx.shuf30.elt
96  ret i8 %r
97}
98
99; CHECK-LABEL: umax_H
100; CHECK: umaxv {{h[0-9]+}}, {{v[0-9]+}}.8h
101define i16 @umax_H(<8 x i16>* nocapture readonly %arr)  {
102  %rdx.minmax.select = load <8 x i16>, <8 x i16>* %arr
103  %rdx.shuf = shufflevector <8 x i16> %rdx.minmax.select, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
104  %rdx.minmax.cmp23 = icmp ugt <8 x i16> %rdx.minmax.select, %rdx.shuf
105  %rdx.minmax.select24 = select <8 x i1> %rdx.minmax.cmp23, <8 x i16> %rdx.minmax.select, <8 x i16> %rdx.shuf
106  %rdx.shuf25 = shufflevector <8 x i16> %rdx.minmax.select24, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
107  %rdx.minmax.cmp26 = icmp ugt <8 x i16> %rdx.minmax.select24, %rdx.shuf25
108  %rdx.minmax.select27 = select <8 x i1> %rdx.minmax.cmp26, <8 x i16> %rdx.minmax.select24, <8 x i16> %rdx.shuf25
109  %rdx.shuf28 = shufflevector <8 x i16> %rdx.minmax.select27, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
110  %rdx.minmax.cmp29 = icmp ugt <8 x i16> %rdx.minmax.select27, %rdx.shuf28
111  %rdx.minmax.cmp29.elt = extractelement <8 x i1> %rdx.minmax.cmp29, i32 0
112  %rdx.minmax.select27.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 0
113  %rdx.shuf28.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 1
114  %r = select i1 %rdx.minmax.cmp29.elt, i16 %rdx.minmax.select27.elt, i16 %rdx.shuf28.elt
115  ret i16 %r
116}
117
118; CHECK-LABEL: umax_S
119; CHECK: umaxv {{s[0-9]+}}, {{v[0-9]+}}.4s
120define i32 @umax_S(<4 x i32>* nocapture readonly %arr) {
121  %rdx.minmax.select  = load <4 x i32>, <4 x i32>* %arr
122  %rdx.shuf = shufflevector <4 x i32> %rdx.minmax.select, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
123  %rdx.minmax.cmp18 = icmp ugt <4 x i32> %rdx.minmax.select, %rdx.shuf
124  %rdx.minmax.select19 = select <4 x i1> %rdx.minmax.cmp18, <4 x i32> %rdx.minmax.select, <4 x i32> %rdx.shuf
125  %rdx.shuf20 = shufflevector <4 x i32> %rdx.minmax.select19, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
126  %rdx.minmax.cmp21 = icmp ugt <4 x i32> %rdx.minmax.select19, %rdx.shuf20
127  %rdx.minmax.cmp21.elt = extractelement <4 x i1> %rdx.minmax.cmp21, i32 0
128  %rdx.minmax.select19.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 0
129  %rdx.shuf20.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 1
130  %r = select i1 %rdx.minmax.cmp21.elt, i32 %rdx.minmax.select19.elt, i32 %rdx.shuf20.elt
131  ret i32 %r
132}
133
134; CHECK-LABEL: umax_D
135; CHECK-NOT: umaxv
136define i64 @umax_D(<2 x i64>* nocapture readonly %arr)  {
137  %rdx.minmax.select = load <2 x i64>, <2 x i64>* %arr
138  %rdx.shuf = shufflevector <2 x i64> %rdx.minmax.select, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
139  %rdx.minmax.cmp18 = icmp ugt <2 x i64> %rdx.minmax.select, %rdx.shuf
140  %rdx.minmax.cmp18.elt = extractelement <2 x i1> %rdx.minmax.cmp18, i32 0
141  %rdx.minmax.select.elt = extractelement <2 x i64> %rdx.minmax.select, i32 0
142  %rdx.shuf.elt = extractelement <2 x i64> %rdx.minmax.select, i32 1
143  %r = select i1 %rdx.minmax.cmp18.elt, i64 %rdx.minmax.select.elt, i64 %rdx.shuf.elt
144  ret i64 %r
145}
146
147
148; CHECK-LABEL: smin_B
149; CHECK: sminv {{b[0-9]+}}, {{v[0-9]+}}.16b
150define i8 @smin_B(<16 x i8>* nocapture readonly %arr) {
151  %rdx.minmax.select = load <16 x i8>, <16 x i8>* %arr
152  %rdx.shuf = shufflevector <16 x i8> %rdx.minmax.select, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
153  %rdx.minmax.cmp22 = icmp slt <16 x i8> %rdx.minmax.select, %rdx.shuf
154  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i8> %rdx.minmax.select, <16 x i8> %rdx.shuf
155  %rdx.shuf24 = shufflevector <16 x i8> %rdx.minmax.select23, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
156  %rdx.minmax.cmp25 = icmp slt <16 x i8> %rdx.minmax.select23, %rdx.shuf24
157  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i8> %rdx.minmax.select23, <16 x i8> %rdx.shuf24
158  %rdx.shuf27 = shufflevector <16 x i8> %rdx.minmax.select26, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
159  %rdx.minmax.cmp28 = icmp slt <16 x i8> %rdx.minmax.select26, %rdx.shuf27
160  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i8> %rdx.minmax.select26, <16 x i8> %rdx.shuf27
161  %rdx.shuf30 = shufflevector <16 x i8> %rdx.minmax.select29, <16 x i8> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
162  %rdx.minmax.cmp31 = icmp slt <16 x i8> %rdx.minmax.select29, %rdx.shuf30
163  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
164  %rdx.minmax.select29.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 0
165  %rdx.shuf30.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 1
166  %r = select i1 %rdx.minmax.cmp31.elt, i8 %rdx.minmax.select29.elt, i8 %rdx.shuf30.elt
167  ret i8 %r
168}
169
170; CHECK-LABEL: smin_H
171; CHECK: sminv {{h[0-9]+}}, {{v[0-9]+}}.8h
172define i16 @smin_H(<8 x i16>* nocapture readonly %arr) {
173  %rdx.minmax.select = load <8 x i16>, <8 x i16>* %arr
174  %rdx.shuf = shufflevector <8 x i16> %rdx.minmax.select, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
175  %rdx.minmax.cmp23 = icmp slt <8 x i16> %rdx.minmax.select, %rdx.shuf
176  %rdx.minmax.select24 = select <8 x i1> %rdx.minmax.cmp23, <8 x i16> %rdx.minmax.select, <8 x i16> %rdx.shuf
177  %rdx.shuf25 = shufflevector <8 x i16> %rdx.minmax.select24, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
178  %rdx.minmax.cmp26 = icmp slt <8 x i16> %rdx.minmax.select24, %rdx.shuf25
179  %rdx.minmax.select27 = select <8 x i1> %rdx.minmax.cmp26, <8 x i16> %rdx.minmax.select24, <8 x i16> %rdx.shuf25
180  %rdx.shuf28 = shufflevector <8 x i16> %rdx.minmax.select27, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
181  %rdx.minmax.cmp29 = icmp slt <8 x i16> %rdx.minmax.select27, %rdx.shuf28
182  %rdx.minmax.cmp29.elt = extractelement <8 x i1> %rdx.minmax.cmp29, i32 0
183  %rdx.minmax.select27.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 0
184  %rdx.shuf28.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 1
185  %r = select i1 %rdx.minmax.cmp29.elt, i16 %rdx.minmax.select27.elt, i16 %rdx.shuf28.elt
186  ret i16 %r
187}
188
189; CHECK-LABEL: smin_S
190; CHECK: sminv {{s[0-9]+}}, {{v[0-9]+}}.4s
191define i32 @smin_S(<4 x i32>* nocapture readonly %arr) {
192  %rdx.minmax.select = load <4 x i32>, <4 x i32>* %arr
193  %rdx.shuf = shufflevector <4 x i32> %rdx.minmax.select, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
194  %rdx.minmax.cmp18 = icmp slt <4 x i32> %rdx.minmax.select, %rdx.shuf
195  %rdx.minmax.select19 = select <4 x i1> %rdx.minmax.cmp18, <4 x i32> %rdx.minmax.select, <4 x i32> %rdx.shuf
196  %rdx.shuf20 = shufflevector <4 x i32> %rdx.minmax.select19, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
197  %rdx.minmax.cmp21 = icmp slt <4 x i32> %rdx.minmax.select19, %rdx.shuf20
198  %rdx.minmax.cmp21.elt = extractelement <4 x i1> %rdx.minmax.cmp21, i32 0
199  %rdx.minmax.select19.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 0
200  %rdx.shuf20.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 1
201  %r = select i1 %rdx.minmax.cmp21.elt, i32 %rdx.minmax.select19.elt, i32 %rdx.shuf20.elt
202  ret i32 %r
203}
204
205; CHECK-LABEL: smin_D
206; CHECK-NOT: sminv
207define i64 @smin_D(<2 x i64>* nocapture readonly %arr) {
208  %rdx.minmax.select = load <2 x i64>, <2 x i64>* %arr
209  %rdx.shuf = shufflevector <2 x i64> %rdx.minmax.select, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
210  %rdx.minmax.cmp18 = icmp slt <2 x i64> %rdx.minmax.select, %rdx.shuf
211  %rdx.minmax.cmp18.elt = extractelement <2 x i1> %rdx.minmax.cmp18, i32 0
212  %rdx.minmax.select.elt = extractelement <2 x i64> %rdx.minmax.select, i32 0
213  %rdx.shuf.elt = extractelement <2 x i64> %rdx.minmax.select, i32 1
214  %r = select i1 %rdx.minmax.cmp18.elt, i64 %rdx.minmax.select.elt, i64 %rdx.shuf.elt
215  ret i64 %r
216}
217
218
219; CHECK-LABEL: umin_B
220; CHECK: uminv {{b[0-9]+}}, {{v[0-9]+}}.16b
221define i8 @umin_B(<16 x i8>* nocapture readonly %arr)  {
222  %rdx.minmax.select = load <16 x i8>, <16 x i8>* %arr
223  %rdx.shuf = shufflevector <16 x i8> %rdx.minmax.select, <16 x i8> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
224  %rdx.minmax.cmp22 = icmp ult <16 x i8> %rdx.minmax.select, %rdx.shuf
225  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i8> %rdx.minmax.select, <16 x i8> %rdx.shuf
226  %rdx.shuf24 = shufflevector <16 x i8> %rdx.minmax.select23, <16 x i8> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
227  %rdx.minmax.cmp25 = icmp ult <16 x i8> %rdx.minmax.select23, %rdx.shuf24
228  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i8> %rdx.minmax.select23, <16 x i8> %rdx.shuf24
229  %rdx.shuf27 = shufflevector <16 x i8> %rdx.minmax.select26, <16 x i8> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
230  %rdx.minmax.cmp28 = icmp ult <16 x i8> %rdx.minmax.select26, %rdx.shuf27
231  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i8> %rdx.minmax.select26, <16 x i8> %rdx.shuf27
232  %rdx.shuf30 = shufflevector <16 x i8> %rdx.minmax.select29, <16 x i8> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
233  %rdx.minmax.cmp31 = icmp ult <16 x i8> %rdx.minmax.select29, %rdx.shuf30
234  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
235  %rdx.minmax.select29.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 0
236  %rdx.shuf30.elt = extractelement <16 x i8> %rdx.minmax.select29, i32 1
237  %r = select i1 %rdx.minmax.cmp31.elt, i8 %rdx.minmax.select29.elt, i8 %rdx.shuf30.elt
238  ret i8 %r
239}
240
241; CHECK-LABEL: umin_H
242; CHECK: uminv {{h[0-9]+}}, {{v[0-9]+}}.8h
243define i16 @umin_H(<8 x i16>* nocapture readonly %arr)  {
244  %rdx.minmax.select = load <8 x i16>, <8 x i16>* %arr
245  %rdx.shuf = shufflevector <8 x i16> %rdx.minmax.select, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
246  %rdx.minmax.cmp23 = icmp ult <8 x i16> %rdx.minmax.select, %rdx.shuf
247  %rdx.minmax.select24 = select <8 x i1> %rdx.minmax.cmp23, <8 x i16> %rdx.minmax.select, <8 x i16> %rdx.shuf
248  %rdx.shuf25 = shufflevector <8 x i16> %rdx.minmax.select24, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
249  %rdx.minmax.cmp26 = icmp ult <8 x i16> %rdx.minmax.select24, %rdx.shuf25
250  %rdx.minmax.select27 = select <8 x i1> %rdx.minmax.cmp26, <8 x i16> %rdx.minmax.select24, <8 x i16> %rdx.shuf25
251  %rdx.shuf28 = shufflevector <8 x i16> %rdx.minmax.select27, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
252  %rdx.minmax.cmp29 = icmp ult <8 x i16> %rdx.minmax.select27, %rdx.shuf28
253  %rdx.minmax.cmp29.elt = extractelement <8 x i1> %rdx.minmax.cmp29, i32 0
254  %rdx.minmax.select27.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 0
255  %rdx.shuf28.elt = extractelement <8 x i16> %rdx.minmax.select27, i32 1
256  %r = select i1 %rdx.minmax.cmp29.elt, i16 %rdx.minmax.select27.elt, i16 %rdx.shuf28.elt
257  ret i16 %r
258}
259
260; CHECK-LABEL: umin_S
261; CHECK: uminv {{s[0-9]+}}, {{v[0-9]+}}.4s
262define i32 @umin_S(<4 x i32>* nocapture readonly %arr) {
263  %rdx.minmax.select = load <4 x i32>, <4 x i32>* %arr
264  %rdx.shuf = shufflevector <4 x i32> %rdx.minmax.select, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
265  %rdx.minmax.cmp18 = icmp ult <4 x i32> %rdx.minmax.select, %rdx.shuf
266  %rdx.minmax.select19 = select <4 x i1> %rdx.minmax.cmp18, <4 x i32> %rdx.minmax.select, <4 x i32> %rdx.shuf
267  %rdx.shuf20 = shufflevector <4 x i32> %rdx.minmax.select19, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
268  %rdx.minmax.cmp21 = icmp ult <4 x i32> %rdx.minmax.select19, %rdx.shuf20
269  %rdx.minmax.cmp21.elt = extractelement <4 x i1> %rdx.minmax.cmp21, i32 0
270  %rdx.minmax.select19.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 0
271  %rdx.shuf20.elt = extractelement <4 x i32> %rdx.minmax.select19, i32 1
272  %r = select i1 %rdx.minmax.cmp21.elt, i32 %rdx.minmax.select19.elt, i32 %rdx.shuf20.elt
273  ret i32 %r
274}
275
276; CHECK-LABEL: umin_D
277; CHECK-NOT: uminv
278define i64 @umin_D(<2 x i64>* nocapture readonly %arr)  {
279  %rdx.minmax.select = load <2 x i64>, <2 x i64>* %arr
280  %rdx.shuf = shufflevector <2 x i64> %rdx.minmax.select, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
281  %rdx.minmax.cmp18 = icmp ult <2 x i64> %rdx.minmax.select, %rdx.shuf
282  %rdx.minmax.cmp18.elt = extractelement <2 x i1> %rdx.minmax.cmp18, i32 0
283  %rdx.minmax.select.elt = extractelement <2 x i64> %rdx.minmax.select, i32 0
284  %rdx.shuf.elt = extractelement <2 x i64> %rdx.minmax.select, i32 1
285  %r = select i1 %rdx.minmax.cmp18.elt, i64 %rdx.minmax.select.elt, i64 %rdx.shuf.elt
286  ret i64 %r
287}
288
289; CHECK-LABEL: fmaxnm_S
290; CHECK: fmaxnmv
291define float @fmaxnm_S(<4 x float>* nocapture readonly %arr) {
292  %rdx.minmax.select  = load <4 x float>, <4 x float>* %arr
293  %rdx.shuf = shufflevector <4 x float> %rdx.minmax.select, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
294  %rdx.minmax.cmp = fcmp fast oge <4 x float> %rdx.minmax.select, %rdx.shuf
295  %rdx.minmax.select1 = select <4 x i1> %rdx.minmax.cmp, <4 x float> %rdx.minmax.select, <4 x float> %rdx.shuf
296  %rdx.shuf1 = shufflevector <4 x float> %rdx.minmax.select1, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
297  %rdx.minmax.cmp1 = fcmp fast oge <4 x float> %rdx.minmax.select1, %rdx.shuf1
298  %rdx.minmax.cmp1.elt = extractelement <4 x i1> %rdx.minmax.cmp1, i32 0
299  %rdx.minmax.select1.elt = extractelement <4 x float> %rdx.minmax.select1, i32 0
300  %rdx.shuf1.elt = extractelement <4 x float> %rdx.minmax.select1, i32 1
301  %r = select i1 %rdx.minmax.cmp1.elt, float %rdx.minmax.select1.elt, float %rdx.shuf1.elt
302  ret float %r
303}
304
305; CHECK-LABEL: fminnm_S
306; CHECK: fminnmv
307define float @fminnm_S(<4 x float>* nocapture readonly %arr) {
308  %rdx.minmax.select  = load <4 x float>, <4 x float>* %arr
309  %rdx.shuf = shufflevector <4 x float> %rdx.minmax.select, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
310  %rdx.minmax.cmp = fcmp fast ole <4 x float> %rdx.minmax.select, %rdx.shuf
311  %rdx.minmax.select1 = select <4 x i1> %rdx.minmax.cmp, <4 x float> %rdx.minmax.select, <4 x float> %rdx.shuf
312  %rdx.shuf1 = shufflevector <4 x float> %rdx.minmax.select1, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
313  %rdx.minmax.cmp1 = fcmp fast ole <4 x float> %rdx.minmax.select1, %rdx.shuf1
314  %rdx.minmax.cmp1.elt = extractelement <4 x i1> %rdx.minmax.cmp1, i32 0
315  %rdx.minmax.select1.elt = extractelement <4 x float> %rdx.minmax.select1, i32 0
316  %rdx.shuf1.elt = extractelement <4 x float> %rdx.minmax.select1, i32 1
317  %r = select i1 %rdx.minmax.cmp1.elt, float %rdx.minmax.select1.elt, float %rdx.shuf1.elt
318  ret float %r
319}
320
321define i16 @oversized_umax_256(<16 x i16>* nocapture readonly %arr)  {
322; CHECK-LABEL: oversized_umax_256
323; CHECK: umax [[V0:v[0-9]+]].8h, {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
324; CHECK: umaxv {{h[0-9]+}}, [[V0]]
325  %rdx.minmax.select = load <16 x i16>, <16 x i16>* %arr
326  %rdx.shuf = shufflevector <16 x i16> %rdx.minmax.select, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
327  %rdx.minmax.cmp22 = icmp ugt <16 x i16> %rdx.minmax.select, %rdx.shuf
328  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i16> %rdx.minmax.select, <16 x i16> %rdx.shuf
329  %rdx.shuf24 = shufflevector <16 x i16> %rdx.minmax.select23, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
330  %rdx.minmax.cmp25 = icmp ugt <16 x i16> %rdx.minmax.select23, %rdx.shuf24
331  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i16> %rdx.minmax.select23, <16 x i16> %rdx.shuf24
332  %rdx.shuf27 = shufflevector <16 x i16> %rdx.minmax.select26, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
333  %rdx.minmax.cmp28 = icmp ugt <16 x i16> %rdx.minmax.select26, %rdx.shuf27
334  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i16> %rdx.minmax.select26, <16 x i16> %rdx.shuf27
335  %rdx.shuf30 = shufflevector <16 x i16> %rdx.minmax.select29, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
336  %rdx.minmax.cmp31 = icmp ugt <16 x i16> %rdx.minmax.select29, %rdx.shuf30
337  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
338  %rdx.minmax.select29.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 0
339  %rdx.shuf30.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 1
340  %r = select i1 %rdx.minmax.cmp31.elt, i16 %rdx.minmax.select29.elt, i16 %rdx.shuf30.elt
341  ret i16 %r
342}
343
344define i32 @oversized_umax_512(<16 x i32>* nocapture readonly %arr)  {
345; CHECK-LABEL: oversized_umax_512
346; CHECK: umax v
347; CHECK-NEXT: umax v
348; CHECK-NEXT: umax [[V0:v[0-9]+]].4s, {{v[0-9]+}}.4s, {{v[0-9]+}}.4s
349; CHECK-NEXT: umaxv {{s[0-9]+}}, [[V0]]
350  %arr.load = load <16 x i32>, <16 x i32>* %arr
351  %rdx.shuf = shufflevector <16 x i32> %arr.load, <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
352  %rdx.minmax.cmp22 = icmp ugt <16 x i32> %arr.load, %rdx.shuf
353  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i32> %arr.load, <16 x i32> %rdx.shuf
354  %rdx.shuf24 = shufflevector <16 x i32> %rdx.minmax.select23, <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
355  %rdx.minmax.cmp25 = icmp ugt <16 x i32> %rdx.minmax.select23, %rdx.shuf24
356  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i32> %rdx.minmax.select23, <16 x i32> %rdx.shuf24
357  %rdx.shuf27 = shufflevector <16 x i32> %rdx.minmax.select26, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
358  %rdx.minmax.cmp28 = icmp ugt <16 x i32> %rdx.minmax.select26, %rdx.shuf27
359  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i32> %rdx.minmax.select26, <16 x i32> %rdx.shuf27
360  %rdx.shuf30 = shufflevector <16 x i32> %rdx.minmax.select29, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
361  %rdx.minmax.cmp31 = icmp ugt <16 x i32> %rdx.minmax.select29, %rdx.shuf30
362  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
363  %rdx.minmax.select29.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 0
364  %rdx.shuf30.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 1
365  %r = select i1 %rdx.minmax.cmp31.elt, i32 %rdx.minmax.select29.elt, i32 %rdx.shuf30.elt
366  ret i32 %r
367}
368
369define i16 @oversized_umin_256(<16 x i16>* nocapture readonly %arr)  {
370; CHECK-LABEL: oversized_umin_256
371; CHECK: umin [[V0:v[0-9]+]].8h, {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
372; CHECK: uminv {{h[0-9]+}}, [[V0]]
373  %rdx.minmax.select = load <16 x i16>, <16 x i16>* %arr
374  %rdx.shuf = shufflevector <16 x i16> %rdx.minmax.select, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
375  %rdx.minmax.cmp22 = icmp ult <16 x i16> %rdx.minmax.select, %rdx.shuf
376  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i16> %rdx.minmax.select, <16 x i16> %rdx.shuf
377  %rdx.shuf24 = shufflevector <16 x i16> %rdx.minmax.select23, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
378  %rdx.minmax.cmp25 = icmp ult <16 x i16> %rdx.minmax.select23, %rdx.shuf24
379  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i16> %rdx.minmax.select23, <16 x i16> %rdx.shuf24
380  %rdx.shuf27 = shufflevector <16 x i16> %rdx.minmax.select26, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
381  %rdx.minmax.cmp28 = icmp ult <16 x i16> %rdx.minmax.select26, %rdx.shuf27
382  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i16> %rdx.minmax.select26, <16 x i16> %rdx.shuf27
383  %rdx.shuf30 = shufflevector <16 x i16> %rdx.minmax.select29, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
384  %rdx.minmax.cmp31 = icmp ult <16 x i16> %rdx.minmax.select29, %rdx.shuf30
385  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
386  %rdx.minmax.select29.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 0
387  %rdx.shuf30.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 1
388  %r = select i1 %rdx.minmax.cmp31.elt, i16 %rdx.minmax.select29.elt, i16 %rdx.shuf30.elt
389  ret i16 %r
390}
391
392define i32 @oversized_umin_512(<16 x i32>* nocapture readonly %arr)  {
393; CHECK-LABEL: oversized_umin_512
394; CHECK: umin v
395; CHECK-NEXT: umin v
396; CHECK-NEXT: umin [[V0:v[0-9]+]].4s, {{v[0-9]+}}.4s, {{v[0-9]+}}.4s
397; CHECK-NEXT: uminv {{s[0-9]+}}, [[V0]]
398  %arr.load = load <16 x i32>, <16 x i32>* %arr
399  %rdx.shuf = shufflevector <16 x i32> %arr.load, <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
400  %rdx.minmax.cmp22 = icmp ult <16 x i32> %arr.load, %rdx.shuf
401  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i32> %arr.load, <16 x i32> %rdx.shuf
402  %rdx.shuf24 = shufflevector <16 x i32> %rdx.minmax.select23, <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
403  %rdx.minmax.cmp25 = icmp ult <16 x i32> %rdx.minmax.select23, %rdx.shuf24
404  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i32> %rdx.minmax.select23, <16 x i32> %rdx.shuf24
405  %rdx.shuf27 = shufflevector <16 x i32> %rdx.minmax.select26, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
406  %rdx.minmax.cmp28 = icmp ult <16 x i32> %rdx.minmax.select26, %rdx.shuf27
407  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i32> %rdx.minmax.select26, <16 x i32> %rdx.shuf27
408  %rdx.shuf30 = shufflevector <16 x i32> %rdx.minmax.select29, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
409  %rdx.minmax.cmp31 = icmp ult <16 x i32> %rdx.minmax.select29, %rdx.shuf30
410  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
411  %rdx.minmax.select29.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 0
412  %rdx.shuf30.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 1
413  %r = select i1 %rdx.minmax.cmp31.elt, i32 %rdx.minmax.select29.elt, i32 %rdx.shuf30.elt
414  ret i32 %r
415}
416
417define i16 @oversized_smax_256(<16 x i16>* nocapture readonly %arr)  {
418; CHECK-LABEL: oversized_smax_256
419; CHECK: smax [[V0:v[0-9]+]].8h, {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
420; CHECK: smaxv {{h[0-9]+}}, [[V0]]
421  %arr.load = load <16 x i16>, <16 x i16>* %arr
422  %rdx.shuf = shufflevector <16 x i16> %arr.load, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
423  %rdx.minmax.cmp22 = icmp sgt <16 x i16> %arr.load, %rdx.shuf
424  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i16> %arr.load, <16 x i16> %rdx.shuf
425  %rdx.shuf24 = shufflevector <16 x i16> %rdx.minmax.select23, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
426  %rdx.minmax.cmp25 = icmp sgt <16 x i16> %rdx.minmax.select23, %rdx.shuf24
427  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i16> %rdx.minmax.select23, <16 x i16> %rdx.shuf24
428  %rdx.shuf27 = shufflevector <16 x i16> %rdx.minmax.select26, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
429  %rdx.minmax.cmp28 = icmp sgt <16 x i16> %rdx.minmax.select26, %rdx.shuf27
430  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i16> %rdx.minmax.select26, <16 x i16> %rdx.shuf27
431  %rdx.shuf30 = shufflevector <16 x i16> %rdx.minmax.select29, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
432  %rdx.minmax.cmp31 = icmp sgt <16 x i16> %rdx.minmax.select29, %rdx.shuf30
433  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
434  %rdx.minmax.select29.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 0
435  %rdx.shuf30.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 1
436  %r = select i1 %rdx.minmax.cmp31.elt, i16 %rdx.minmax.select29.elt, i16 %rdx.shuf30.elt
437  ret i16 %r
438}
439
440define i32 @oversized_smax_512(<16 x i32>* nocapture readonly %arr)  {
441; CHECK-LABEL: oversized_smax_512
442; CHECK: smax v
443; CHECK-NEXT: smax v
444; CHECK-NEXT: smax [[V0:v[0-9]+]].4s, {{v[0-9]+}}.4s, {{v[0-9]+}}.4s
445; CHECK-NEXT: smaxv {{s[0-9]+}}, [[V0]]
446  %arr.load = load <16 x i32>, <16 x i32>* %arr
447  %rdx.shuf = shufflevector <16 x i32> %arr.load, <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
448  %rdx.minmax.cmp22 = icmp sgt <16 x i32> %arr.load, %rdx.shuf
449  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i32> %arr.load, <16 x i32> %rdx.shuf
450  %rdx.shuf24 = shufflevector <16 x i32> %rdx.minmax.select23, <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
451  %rdx.minmax.cmp25 = icmp sgt <16 x i32> %rdx.minmax.select23, %rdx.shuf24
452  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i32> %rdx.minmax.select23, <16 x i32> %rdx.shuf24
453  %rdx.shuf27 = shufflevector <16 x i32> %rdx.minmax.select26, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
454  %rdx.minmax.cmp28 = icmp sgt <16 x i32> %rdx.minmax.select26, %rdx.shuf27
455  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i32> %rdx.minmax.select26, <16 x i32> %rdx.shuf27
456  %rdx.shuf30 = shufflevector <16 x i32> %rdx.minmax.select29, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
457  %rdx.minmax.cmp31 = icmp sgt <16 x i32> %rdx.minmax.select29, %rdx.shuf30
458  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
459  %rdx.minmax.select29.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 0
460  %rdx.shuf30.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 1
461  %r = select i1 %rdx.minmax.cmp31.elt, i32 %rdx.minmax.select29.elt, i32 %rdx.shuf30.elt
462  ret i32 %r
463}
464
465define i16 @oversized_smin_256(<16 x i16>* nocapture readonly %arr)  {
466; CHECK-LABEL: oversized_smin_256
467; CHECK: smin [[V0:v[0-9]+]].8h, {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
468; CHECK: sminv {{h[0-9]+}}, [[V0]]
469  %rdx.minmax.select = load <16 x i16>, <16 x i16>* %arr
470  %rdx.shuf = shufflevector <16 x i16> %rdx.minmax.select, <16 x i16> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
471  %rdx.minmax.cmp22 = icmp slt <16 x i16> %rdx.minmax.select, %rdx.shuf
472  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i16> %rdx.minmax.select, <16 x i16> %rdx.shuf
473  %rdx.shuf24 = shufflevector <16 x i16> %rdx.minmax.select23, <16 x i16> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
474  %rdx.minmax.cmp25 = icmp slt <16 x i16> %rdx.minmax.select23, %rdx.shuf24
475  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i16> %rdx.minmax.select23, <16 x i16> %rdx.shuf24
476  %rdx.shuf27 = shufflevector <16 x i16> %rdx.minmax.select26, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
477  %rdx.minmax.cmp28 = icmp slt <16 x i16> %rdx.minmax.select26, %rdx.shuf27
478  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i16> %rdx.minmax.select26, <16 x i16> %rdx.shuf27
479  %rdx.shuf30 = shufflevector <16 x i16> %rdx.minmax.select29, <16 x i16> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
480  %rdx.minmax.cmp31 = icmp slt <16 x i16> %rdx.minmax.select29, %rdx.shuf30
481  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
482  %rdx.minmax.select29.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 0
483  %rdx.shuf30.elt = extractelement <16 x i16> %rdx.minmax.select29, i32 1
484  %r = select i1 %rdx.minmax.cmp31.elt, i16 %rdx.minmax.select29.elt, i16 %rdx.shuf30.elt
485  ret i16 %r
486}
487
488define i32 @oversized_smin_512(<16 x i32>* nocapture readonly %arr)  {
489; CHECK-LABEL: oversized_smin_512
490; CHECK: smin v
491; CHECK-NEXT: smin v
492; CHECK-NEXT: smin [[V0:v[0-9]+]].4s, {{v[0-9]+}}.4s, {{v[0-9]+}}.4s
493; CHECK-NEXT: sminv {{s[0-9]+}}, [[V0]]
494  %arr.load = load <16 x i32>, <16 x i32>* %arr
495  %rdx.shuf = shufflevector <16 x i32> %arr.load, <16 x i32> undef, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
496  %rdx.minmax.cmp22 = icmp slt <16 x i32> %arr.load, %rdx.shuf
497  %rdx.minmax.select23 = select <16 x i1> %rdx.minmax.cmp22, <16 x i32> %arr.load, <16 x i32> %rdx.shuf
498  %rdx.shuf24 = shufflevector <16 x i32> %rdx.minmax.select23, <16 x i32> undef, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
499  %rdx.minmax.cmp25 = icmp slt <16 x i32> %rdx.minmax.select23, %rdx.shuf24
500  %rdx.minmax.select26 = select <16 x i1> %rdx.minmax.cmp25, <16 x i32> %rdx.minmax.select23, <16 x i32> %rdx.shuf24
501  %rdx.shuf27 = shufflevector <16 x i32> %rdx.minmax.select26, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
502  %rdx.minmax.cmp28 = icmp slt <16 x i32> %rdx.minmax.select26, %rdx.shuf27
503  %rdx.minmax.select29 = select <16 x i1> %rdx.minmax.cmp28, <16 x i32> %rdx.minmax.select26, <16 x i32> %rdx.shuf27
504  %rdx.shuf30 = shufflevector <16 x i32> %rdx.minmax.select29, <16 x i32> undef, <16 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
505  %rdx.minmax.cmp31 = icmp slt <16 x i32> %rdx.minmax.select29, %rdx.shuf30
506  %rdx.minmax.cmp31.elt = extractelement <16 x i1> %rdx.minmax.cmp31, i32 0
507  %rdx.minmax.select29.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 0
508  %rdx.shuf30.elt = extractelement <16 x i32> %rdx.minmax.select29, i32 1
509  %r = select i1 %rdx.minmax.cmp31.elt, i32 %rdx.minmax.select29.elt, i32 %rdx.shuf30.elt
510  ret i32 %r
511}
Note: See TracBrowser for help on using the repository browser.