| blob | 89c64d64b9c8447d2d06d5467fb2eac3d0a46e83 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=slp-vectorizer,dce -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
4 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
5 target triple = "x86_64-apple-macosx10.8.0"
7 define i32 @rollable(ptr noalias nocapture %in, ptr noalias nocapture %out, i64 %n) {
8 ; CHECK-LABEL: @rollable(
9 ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[N:%.*]], 0
10 ; CHECK-NEXT: br i1 [[TMP1]], label [[DOT_CRIT_EDGE:%.*]], label [[DOTLR_PH:%.*]]
11 ; CHECK: .lr.ph:
12 ; CHECK-NEXT: [[I_019:%.*]] = phi i64 [ [[TMP8:%.*]], [[DOTLR_PH]] ], [ 0, [[TMP0:%.*]] ]
13 ; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[I_019]], 2
14 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[IN:%.*]], i64 [[TMP2]]
15 ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[OUT:%.*]], i64 [[TMP2]]
16 ; CHECK-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4
17 ; CHECK-NEXT: [[TMP6:%.*]] = mul <4 x i32> [[TMP5]], <i32 7, i32 7, i32 7, i32 7>
18 ; CHECK-NEXT: [[TMP7:%.*]] = add <4 x i32> [[TMP6]], <i32 7, i32 14, i32 21, i32 28>
19 ; CHECK-NEXT: store <4 x i32> [[TMP7]], ptr [[TMP4]], align 4
20 ; CHECK-NEXT: [[TMP8]] = add i64 [[I_019]], 1
21 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[TMP8]], [[N]]
22 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]]
23 ; CHECK: ._crit_edge:
24 ; CHECK-NEXT: ret i32 undef
25 ;
26 %1 = icmp eq i64 %n, 0
27 br i1 %1, label %._crit_edge, label %.lr.ph
29 .lr.ph: ; preds = %0, %.lr.ph
30 %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ]
31 %2 = shl i64 %i.019, 2
32 %3 = getelementptr inbounds i32, ptr %in, i64 %2
33 %4 = load i32, ptr %3, align 4
34 %5 = or disjoint i64 %2, 1
35 %6 = getelementptr inbounds i32, ptr %in, i64 %5
36 %7 = load i32, ptr %6, align 4
37 %8 = or disjoint i64 %2, 2
38 %9 = getelementptr inbounds i32, ptr %in, i64 %8
39 %10 = load i32, ptr %9, align 4
40 %11 = or disjoint i64 %2, 3
41 %12 = getelementptr inbounds i32, ptr %in, i64 %11
42 %13 = load i32, ptr %12, align 4
43 %14 = mul i32 %4, 7
44 %15 = add i32 %14, 7
45 %16 = mul i32 %7, 7
46 %17 = add i32 %16, 14
47 %18 = mul i32 %10, 7
48 %19 = add i32 %18, 21
49 %20 = mul i32 %13, 7
50 %21 = add i32 %20, 28
51 %22 = getelementptr inbounds i32, ptr %out, i64 %2
52 store i32 %15, ptr %22, align 4
53 %23 = getelementptr inbounds i32, ptr %out, i64 %5
54 store i32 %17, ptr %23, align 4
55 %24 = getelementptr inbounds i32, ptr %out, i64 %8
56 store i32 %19, ptr %24, align 4
57 %25 = getelementptr inbounds i32, ptr %out, i64 %11
58 store i32 %21, ptr %25, align 4
59 %26 = add i64 %i.019, 1
60 %exitcond = icmp eq i64 %26, %n
61 br i1 %exitcond, label %._crit_edge, label %.lr.ph
63 ._crit_edge: ; preds = %.lr.ph, %0
64 ret i32 undef
65 }
67 define i32 @unrollable(ptr %in, ptr %out, i64 %n) nounwind ssp uwtable {
68 ; CHECK-LABEL: @unrollable(
69 ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[N:%.*]], 0
70 ; CHECK-NEXT: br i1 [[TMP1]], label [[DOT_CRIT_EDGE:%.*]], label [[DOTLR_PH:%.*]]
71 ; CHECK: .lr.ph:
72 ; CHECK-NEXT: [[I_019:%.*]] = phi i64 [ [[TMP14:%.*]], [[DOTLR_PH]] ], [ 0, [[TMP0:%.*]] ]
73 ; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[I_019]], 2
74 ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[IN:%.*]], i64 [[TMP2]]
75 ; CHECK-NEXT: [[TMP4:%.*]] = or disjoint i64 [[TMP2]], 2
76 ; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[IN]], i64 [[TMP4]]
77 ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[OUT:%.*]], i64 [[TMP2]]
78 ; CHECK-NEXT: [[TMP7:%.*]] = load <2 x i32>, ptr [[TMP3]], align 4
79 ; CHECK-NEXT: [[TMP8:%.*]] = mul <2 x i32> [[TMP7]], <i32 7, i32 7>
80 ; CHECK-NEXT: [[TMP9:%.*]] = add <2 x i32> [[TMP8]], <i32 7, i32 14>
81 ; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[OUT]], i64 [[TMP4]]
82 ; CHECK-NEXT: [[TMP11:%.*]] = load <2 x i32>, ptr [[TMP5]], align 4
83 ; CHECK-NEXT: [[TMP12:%.*]] = mul <2 x i32> [[TMP11]], <i32 7, i32 7>
84 ; CHECK-NEXT: [[TMP13:%.*]] = add <2 x i32> [[TMP12]], <i32 21, i32 28>
85 ; CHECK-NEXT: store <2 x i32> [[TMP9]], ptr [[TMP6]], align 4
86 ; CHECK-NEXT: [[BARRIER:%.*]] = call i32 @goo(i32 0)
87 ; CHECK-NEXT: store <2 x i32> [[TMP13]], ptr [[TMP10]], align 4
88 ; CHECK-NEXT: [[TMP14]] = add i64 [[I_019]], 1
89 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[TMP14]], [[N]]
90 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE]], label [[DOTLR_PH]]
91 ; CHECK: ._crit_edge:
92 ; CHECK-NEXT: ret i32 undef
93 ;
94 %1 = icmp eq i64 %n, 0
95 br i1 %1, label %._crit_edge, label %.lr.ph
97 .lr.ph: ; preds = %0, %.lr.ph
98 %i.019 = phi i64 [ %26, %.lr.ph ], [ 0, %0 ]
99 %2 = shl i64 %i.019, 2
100 %3 = getelementptr inbounds i32, ptr %in, i64 %2
101 %4 = load i32, ptr %3, align 4
102 %5 = or disjoint i64 %2, 1
103 %6 = getelementptr inbounds i32, ptr %in, i64 %5
104 %7 = load i32, ptr %6, align 4
105 %8 = or disjoint i64 %2, 2
106 %9 = getelementptr inbounds i32, ptr %in, i64 %8
107 %10 = load i32, ptr %9, align 4
108 %11 = or disjoint i64 %2, 3
109 %12 = getelementptr inbounds i32, ptr %in, i64 %11
110 %13 = load i32, ptr %12, align 4
111 %14 = mul i32 %4, 7
112 %15 = add i32 %14, 7
113 %16 = mul i32 %7, 7
114 %17 = add i32 %16, 14
115 %18 = mul i32 %10, 7
116 %19 = add i32 %18, 21
117 %20 = mul i32 %13, 7
118 %21 = add i32 %20, 28
119 %22 = getelementptr inbounds i32, ptr %out, i64 %2
120 store i32 %15, ptr %22, align 4
121 %23 = getelementptr inbounds i32, ptr %out, i64 %5
122 store i32 %17, ptr %23, align 4
123 %barrier = call i32 @goo(i32 0) ; <---------------- memory barrier.
124 %24 = getelementptr inbounds i32, ptr %out, i64 %8
125 store i32 %19, ptr %24, align 4
126 %25 = getelementptr inbounds i32, ptr %out, i64 %11
127 store i32 %21, ptr %25, align 4
128 %26 = add i64 %i.019, 1
129 %exitcond = icmp eq i64 %26, %n
130 br i1 %exitcond, label %._crit_edge, label %.lr.ph
132 ._crit_edge: ; preds = %.lr.ph, %0
133 ret i32 undef
134 }
136 declare i32 @goo(i32)