| blob | db02d64ba9382a1d5f035e6f016a60d2936028c6 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -slp-vectorizer -S < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skylake -slp-threshold=-7 | FileCheck %s --check-prefixes=ALL,CHECK
3 ; RUN: opt -slp-vectorizer -S < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=skylake -slp-threshold=-8 -slp-min-tree-size=6 | FileCheck %s --check-prefixes=ALL,FORCE_REDUCTION
5 define void @Test(i32) {
6 ; CHECK-LABEL: @Test(
7 ; CHECK-NEXT: entry:
8 ; CHECK-NEXT: br label [[LOOP:%.*]]
9 ; CHECK: loop:
10 ; CHECK-NEXT: [[TMP1:%.*]] = phi <2 x i32> [ [[TMP15:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
11 ; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <8 x i32> <i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
12 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <8 x i32> [[SHUFFLE]], i32 1
13 ; CHECK-NEXT: [[TMP3:%.*]] = add <8 x i32> [[SHUFFLE]], <i32 0, i32 55, i32 285, i32 1240, i32 1496, i32 8555, i32 12529, i32 13685>
14 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <8 x i32> [[TMP3]], <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
15 ; CHECK-NEXT: [[BIN_RDX:%.*]] = and <8 x i32> [[TMP3]], [[RDX_SHUF]]
16 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <8 x i32> [[BIN_RDX]], <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
17 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = and <8 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
18 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <8 x i32> [[BIN_RDX2]], <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
19 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = and <8 x i32> [[BIN_RDX2]], [[RDX_SHUF3]]
20 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <8 x i32> [[BIN_RDX4]], i32 0
21 ; CHECK-NEXT: [[OP_EXTRA:%.*]] = and i32 [[TMP4]], [[TMP0:%.*]]
22 ; CHECK-NEXT: [[OP_EXTRA5:%.*]] = and i32 [[OP_EXTRA]], [[TMP0]]
23 ; CHECK-NEXT: [[OP_EXTRA6:%.*]] = and i32 [[OP_EXTRA5]], [[TMP0]]
24 ; CHECK-NEXT: [[OP_EXTRA7:%.*]] = and i32 [[OP_EXTRA6]], [[TMP0]]
25 ; CHECK-NEXT: [[OP_EXTRA8:%.*]] = and i32 [[OP_EXTRA7]], [[TMP0]]
26 ; CHECK-NEXT: [[OP_EXTRA9:%.*]] = and i32 [[OP_EXTRA8]], [[TMP0]]
27 ; CHECK-NEXT: [[OP_EXTRA10:%.*]] = and i32 [[OP_EXTRA9]], [[TMP0]]
28 ; CHECK-NEXT: [[OP_EXTRA11:%.*]] = and i32 [[OP_EXTRA10]], [[TMP0]]
29 ; CHECK-NEXT: [[OP_EXTRA12:%.*]] = and i32 [[OP_EXTRA11]], [[TMP0]]
30 ; CHECK-NEXT: [[OP_EXTRA13:%.*]] = and i32 [[OP_EXTRA12]], [[TMP0]]
31 ; CHECK-NEXT: [[OP_EXTRA14:%.*]] = and i32 [[OP_EXTRA13]], [[TMP0]]
32 ; CHECK-NEXT: [[OP_EXTRA15:%.*]] = and i32 [[OP_EXTRA14]], [[TMP0]]
33 ; CHECK-NEXT: [[OP_EXTRA16:%.*]] = and i32 [[OP_EXTRA15]], [[TMP0]]
34 ; CHECK-NEXT: [[OP_EXTRA17:%.*]] = and i32 [[OP_EXTRA16]], [[TMP0]]
35 ; CHECK-NEXT: [[OP_EXTRA18:%.*]] = and i32 [[OP_EXTRA17]], [[TMP0]]
36 ; CHECK-NEXT: [[OP_EXTRA19:%.*]] = and i32 [[OP_EXTRA18]], [[TMP0]]
37 ; CHECK-NEXT: [[OP_EXTRA20:%.*]] = and i32 [[OP_EXTRA19]], [[TMP0]]
38 ; CHECK-NEXT: [[OP_EXTRA21:%.*]] = and i32 [[OP_EXTRA20]], [[TMP0]]
39 ; CHECK-NEXT: [[OP_EXTRA22:%.*]] = and i32 [[OP_EXTRA21]], [[TMP0]]
40 ; CHECK-NEXT: [[OP_EXTRA23:%.*]] = and i32 [[OP_EXTRA22]], [[TMP0]]
41 ; CHECK-NEXT: [[OP_EXTRA24:%.*]] = and i32 [[OP_EXTRA23]], [[TMP0]]
42 ; CHECK-NEXT: [[OP_EXTRA25:%.*]] = and i32 [[OP_EXTRA24]], [[TMP0]]
43 ; CHECK-NEXT: [[OP_EXTRA26:%.*]] = and i32 [[OP_EXTRA25]], [[TMP0]]
44 ; CHECK-NEXT: [[OP_EXTRA27:%.*]] = and i32 [[OP_EXTRA26]], [[TMP0]]
45 ; CHECK-NEXT: [[OP_EXTRA28:%.*]] = and i32 [[OP_EXTRA27]], [[TMP0]]
46 ; CHECK-NEXT: [[OP_EXTRA29:%.*]] = and i32 [[OP_EXTRA28]], [[TMP0]]
47 ; CHECK-NEXT: [[OP_EXTRA30:%.*]] = and i32 [[OP_EXTRA29]], [[TMP0]]
48 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i32> undef, i32 [[OP_EXTRA30]], i32 0
49 ; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> [[TMP5]], i32 14910, i32 1
50 ; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i32> undef, i32 [[TMP2]], i32 0
51 ; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP2]], i32 1
52 ; CHECK-NEXT: [[TMP9:%.*]] = and <2 x i32> [[TMP6]], [[TMP8]]
53 ; CHECK-NEXT: [[TMP10:%.*]] = add <2 x i32> [[TMP6]], [[TMP8]]
54 ; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <2 x i32> [[TMP9]], <2 x i32> [[TMP10]], <2 x i32> <i32 0, i32 3>
55 ; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i32> [[TMP11]], i32 0
56 ; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x i32> undef, i32 [[TMP12]], i32 0
57 ; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i32> [[TMP11]], i32 1
58 ; CHECK-NEXT: [[TMP15]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP14]], i32 1
59 ; CHECK-NEXT: br label [[LOOP]]
60 ;
61 ; FORCE_REDUCTION-LABEL: @Test(
62 ; FORCE_REDUCTION-NEXT: entry:
63 ; FORCE_REDUCTION-NEXT: br label [[LOOP:%.*]]
64 ; FORCE_REDUCTION: loop:
65 ; FORCE_REDUCTION-NEXT: [[TMP1:%.*]] = phi <2 x i32> [ [[TMP13:%.*]], [[LOOP]] ], [ zeroinitializer, [[ENTRY:%.*]] ]
66 ; FORCE_REDUCTION-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> undef, <4 x i32> <i32 0, i32 1, i32 1, i32 1>
67 ; FORCE_REDUCTION-NEXT: [[TMP2:%.*]] = extractelement <4 x i32> [[SHUFFLE]], i32 1
68 ; FORCE_REDUCTION-NEXT: [[TMP3:%.*]] = add <4 x i32> [[SHUFFLE]], <i32 0, i32 55, i32 285, i32 1240>
69 ; FORCE_REDUCTION-NEXT: [[VAL_20:%.*]] = add i32 [[TMP2]], 1496
70 ; FORCE_REDUCTION-NEXT: [[VAL_34:%.*]] = add i32 [[TMP2]], 8555
71 ; FORCE_REDUCTION-NEXT: [[RDX_SHUF:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
72 ; FORCE_REDUCTION-NEXT: [[BIN_RDX:%.*]] = and <4 x i32> [[TMP3]], [[RDX_SHUF]]
73 ; FORCE_REDUCTION-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <4 x i32> [[BIN_RDX]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
74 ; FORCE_REDUCTION-NEXT: [[BIN_RDX2:%.*]] = and <4 x i32> [[BIN_RDX]], [[RDX_SHUF1]]
75 ; FORCE_REDUCTION-NEXT: [[TMP4:%.*]] = extractelement <4 x i32> [[BIN_RDX2]], i32 0
76 ; FORCE_REDUCTION-NEXT: [[TMP5:%.*]] = and i32 [[TMP4]], [[VAL_20]]
77 ; FORCE_REDUCTION-NEXT: [[TMP6:%.*]] = and i32 [[TMP5]], [[VAL_34]]
78 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA:%.*]] = and i32 [[TMP6]], [[TMP0:%.*]]
79 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA3:%.*]] = and i32 [[OP_EXTRA]], [[TMP0]]
80 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA4:%.*]] = and i32 [[OP_EXTRA3]], [[TMP0]]
81 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA5:%.*]] = and i32 [[OP_EXTRA4]], [[TMP0]]
82 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA6:%.*]] = and i32 [[OP_EXTRA5]], [[TMP0]]
83 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA7:%.*]] = and i32 [[OP_EXTRA6]], [[TMP0]]
84 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA8:%.*]] = and i32 [[OP_EXTRA7]], [[TMP0]]
85 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA9:%.*]] = and i32 [[OP_EXTRA8]], [[TMP0]]
86 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA10:%.*]] = and i32 [[OP_EXTRA9]], [[TMP0]]
87 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA11:%.*]] = and i32 [[OP_EXTRA10]], [[TMP0]]
88 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA12:%.*]] = and i32 [[OP_EXTRA11]], [[TMP0]]
89 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA13:%.*]] = and i32 [[OP_EXTRA12]], [[TMP0]]
90 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA14:%.*]] = and i32 [[OP_EXTRA13]], [[TMP0]]
91 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA15:%.*]] = and i32 [[OP_EXTRA14]], [[TMP0]]
92 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA16:%.*]] = and i32 [[OP_EXTRA15]], [[TMP0]]
93 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA17:%.*]] = and i32 [[OP_EXTRA16]], [[TMP0]]
94 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA18:%.*]] = and i32 [[OP_EXTRA17]], [[TMP0]]
95 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA19:%.*]] = and i32 [[OP_EXTRA18]], [[TMP0]]
96 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA20:%.*]] = and i32 [[OP_EXTRA19]], [[TMP0]]
97 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA21:%.*]] = and i32 [[OP_EXTRA20]], [[TMP0]]
98 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA22:%.*]] = and i32 [[OP_EXTRA21]], [[TMP0]]
99 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA23:%.*]] = and i32 [[OP_EXTRA22]], [[TMP0]]
100 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA24:%.*]] = and i32 [[OP_EXTRA23]], [[TMP0]]
101 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA25:%.*]] = and i32 [[OP_EXTRA24]], [[TMP0]]
102 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA26:%.*]] = and i32 [[OP_EXTRA25]], [[TMP0]]
103 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA27:%.*]] = and i32 [[OP_EXTRA26]], [[TMP0]]
104 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA28:%.*]] = and i32 [[OP_EXTRA27]], [[TMP0]]
105 ; FORCE_REDUCTION-NEXT: [[OP_EXTRA29:%.*]] = and i32 [[OP_EXTRA28]], [[TMP2]]
106 ; FORCE_REDUCTION-NEXT: [[VAL_39:%.*]] = add i32 [[TMP2]], 12529
107 ; FORCE_REDUCTION-NEXT: [[VAL_40:%.*]] = and i32 [[OP_EXTRA29]], [[VAL_39]]
108 ; FORCE_REDUCTION-NEXT: [[VAL_41:%.*]] = add i32 [[TMP2]], 13685
109 ; FORCE_REDUCTION-NEXT: [[TMP7:%.*]] = insertelement <2 x i32> undef, i32 [[VAL_40]], i32 0
110 ; FORCE_REDUCTION-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP2]], i32 1
111 ; FORCE_REDUCTION-NEXT: [[TMP9:%.*]] = insertelement <2 x i32> undef, i32 [[VAL_41]], i32 0
112 ; FORCE_REDUCTION-NEXT: [[TMP10:%.*]] = insertelement <2 x i32> [[TMP9]], i32 14910, i32 1
113 ; FORCE_REDUCTION-NEXT: [[TMP11:%.*]] = and <2 x i32> [[TMP8]], [[TMP10]]
114 ; FORCE_REDUCTION-NEXT: [[TMP12:%.*]] = add <2 x i32> [[TMP8]], [[TMP10]]
115 ; FORCE_REDUCTION-NEXT: [[TMP13]] = shufflevector <2 x i32> [[TMP11]], <2 x i32> [[TMP12]], <2 x i32> <i32 0, i32 3>
116 ; FORCE_REDUCTION-NEXT: br label [[LOOP]]
117 ;
118 entry:
119 br label %loop
121 loop:
122 %local_4_39.us = phi i32 [ %val_42, %loop ], [ 0, %entry ]
123 %local_8_43.us = phi i32 [ %val_43, %loop ], [ 0, %entry ]
124 %val_0 = add i32 %local_4_39.us, 0
125 %val_1 = and i32 %local_8_43.us, %val_0
126 %val_2 = and i32 %val_1, %0
127 %val_3 = and i32 %val_2, %0
128 %val_4 = and i32 %val_3, %0
129 %val_5 = and i32 %val_4, %0
130 %val_6 = add i32 %local_8_43.us, 55
131 %val_7 = and i32 %val_5, %val_6
132 %val_8 = and i32 %val_7, %0
133 %val_9 = and i32 %val_8, %0
134 %val_10 = and i32 %val_9, %0
135 %val_11 = add i32 %local_8_43.us, 285
136 %val_12 = and i32 %val_10, %val_11
137 %val_13 = and i32 %val_12, %0
138 %val_14 = and i32 %val_13, %0
139 %val_15 = and i32 %val_14, %0
140 %val_16 = and i32 %val_15, %0
141 %val_17 = and i32 %val_16, %0
142 %val_18 = add i32 %local_8_43.us, 1240
143 %val_19 = and i32 %val_17, %val_18
144 %val_20 = add i32 %local_8_43.us, 1496
145 %val_21 = and i32 %val_19, %val_20
146 %val_22 = and i32 %val_21, %0
147 %val_23 = and i32 %val_22, %0
148 %val_24 = and i32 %val_23, %0
149 %val_25 = and i32 %val_24, %0
150 %val_26 = and i32 %val_25, %0
151 %val_27 = and i32 %val_26, %0
152 %val_28 = and i32 %val_27, %0
153 %val_29 = and i32 %val_28, %0
154 %val_30 = and i32 %val_29, %0
155 %val_31 = and i32 %val_30, %0
156 %val_32 = and i32 %val_31, %0
157 %val_33 = and i32 %val_32, %0
158 %val_34 = add i32 %local_8_43.us, 8555
159 %val_35 = and i32 %val_33, %val_34
160 %val_36 = and i32 %val_35, %0
161 %val_37 = and i32 %val_36, %0
162 %val_38 = and i32 %val_37, %0
163 %val_39 = add i32 %local_8_43.us, 12529
164 %val_40 = and i32 %val_38, %val_39
165 %val_41 = add i32 %local_8_43.us, 13685
166 %val_42 = and i32 %val_40, %val_41
167 %val_43 = add i32 %local_8_43.us, 14910
168 br label %loop
169 }