1 ; RUN: llc -march=x86-64 < %s > /dev/null
3 ; ScalarEvolution misses an opportunity to fold ((trunc x) + (trunc -x) + y),
4 ; but LSR should tolerate this.
7 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"
8 target triple = "x86_64-apple-darwin11.0"
10 define fastcc void @formatValue(i64 %arg5) nounwind {
12 %t = trunc i64 %arg5 to i32 ; <i32> [#uses=1]
13 %t13 = sub i64 0, %arg5 ; <i64> [#uses=1]
14 %t14 = and i64 %t13, 4294967295 ; <i64> [#uses=1]
17 bb15: ; preds = %bb21, %bb12
18 %t16 = phi i64 [ 0, %bb12 ], [ %t23, %bb15 ] ; <i64> [#uses=2]
19 %t17 = mul i64 %t14, %t16 ; <i64> [#uses=1]
20 %t18 = add i64 undef, %t17 ; <i64> [#uses=1]
21 %t19 = trunc i64 %t18 to i32 ; <i32> [#uses=1]
22 %t22 = icmp eq i32 %t19, %t ; <i1> [#uses=1]
23 %t23 = add i64 %t16, 1 ; <i64> [#uses=1]
24 br i1 %t22, label %bb24, label %bb15
26 bb24: ; preds = %bb21, %bb11
30 ; ScalarEvolution should be able to correctly expand the crazy addrec here.
33 define void @int323() nounwind {
37 for.cond: ; preds = %lbl_264, %for.inc, %entry
38 %g_263.tmp.1 = phi i8 [ undef, %entry ], [ %g_263.tmp.1, %for.cond ]
39 %p_95.addr.0 = phi i8 [ 0, %entry ], [ %add, %for.cond ]
40 %add = add i8 %p_95.addr.0, 1 ; <i8> [#uses=1]
41 br i1 undef, label %for.cond, label %lbl_264
43 lbl_264: ; preds = %if.end, %lbl_264.preheader
44 %g_263.tmp.0 = phi i8 [ %g_263.tmp.1, %for.cond ] ; <i8> [#uses=1]
45 %tmp7 = load i16* undef ; <i16> [#uses=1]
46 %conv8 = trunc i16 %tmp7 to i8 ; <i8> [#uses=1]
47 %mul.i = mul i8 %p_95.addr.0, %p_95.addr.0 ; <i8> [#uses=1]
48 %mul.i18 = mul i8 %mul.i, %conv8 ; <i8> [#uses=1]
49 %tobool12 = icmp eq i8 %mul.i18, 0 ; <i1> [#uses=1]
53 ; LSR ends up going into conservative pruning mode; don't prune the solution
54 ; so far that it becomes unsolvable though.
57 %struct.Bu = type { i32, i32, i32 }
59 define void @_Z3fooP2Bui(%struct.Bu* nocapture %bu) {
63 for.body: ; preds = %for.inc131, %entry
64 %indvar = phi i64 [ %indvar.next, %for.inc131 ], [ 0, %entry ] ; <i64> [#uses=3]
65 br i1 undef, label %for.inc131, label %lor.lhs.false
67 lor.lhs.false: ; preds = %for.body
68 %tmp15 = add i64 %indvar, 1 ; <i64> [#uses=1]
69 %tmp17 = add i64 %indvar, 2 ; <i64> [#uses=1]
70 %tmp19 = add i64 %indvar, 3 ; <i64> [#uses=1]
71 %tmp21 = add i64 %indvar, 4 ; <i64> [#uses=1]
72 %tmp23 = add i64 %indvar, 5 ; <i64> [#uses=1]
73 %tmp25 = add i64 %indvar, 6 ; <i64> [#uses=1]
74 %tmp27 = add i64 %indvar, 7 ; <i64> [#uses=1]
75 %tmp29 = add i64 %indvar, 8 ; <i64> [#uses=1]
76 %tmp31 = add i64 %indvar, 9 ; <i64> [#uses=1]
77 %tmp35 = add i64 %indvar, 11 ; <i64> [#uses=1]
78 %tmp37 = add i64 %indvar, 12 ; <i64> [#uses=1]
79 %tmp39 = add i64 %indvar, 13 ; <i64> [#uses=1]
80 %tmp41 = add i64 %indvar, 14 ; <i64> [#uses=1]
81 %tmp43 = add i64 %indvar, 15 ; <i64> [#uses=1]
82 %tmp45 = add i64 %indvar, 16 ; <i64> [#uses=1]
83 %tmp47 = add i64 %indvar, 17 ; <i64> [#uses=1]
84 %mul = trunc i64 %indvar to i32 ; <i32> [#uses=1]
85 %add22 = trunc i64 %tmp15 to i32 ; <i32> [#uses=1]
86 %add28 = trunc i64 %tmp17 to i32 ; <i32> [#uses=1]
87 %add34 = trunc i64 %tmp19 to i32 ; <i32> [#uses=1]
88 %add40 = trunc i64 %tmp21 to i32 ; <i32> [#uses=1]
89 %add46 = trunc i64 %tmp23 to i32 ; <i32> [#uses=1]
90 %add52 = trunc i64 %tmp25 to i32 ; <i32> [#uses=1]
91 %add58 = trunc i64 %tmp27 to i32 ; <i32> [#uses=1]
92 %add64 = trunc i64 %tmp29 to i32 ; <i32> [#uses=1]
93 %add70 = trunc i64 %tmp31 to i32 ; <i32> [#uses=1]
94 %add82 = trunc i64 %tmp35 to i32 ; <i32> [#uses=1]
95 %add88 = trunc i64 %tmp37 to i32 ; <i32> [#uses=1]
96 %add94 = trunc i64 %tmp39 to i32 ; <i32> [#uses=1]
97 %add100 = trunc i64 %tmp41 to i32 ; <i32> [#uses=1]
98 %add106 = trunc i64 %tmp43 to i32 ; <i32> [#uses=1]
99 %add112 = trunc i64 %tmp45 to i32 ; <i32> [#uses=1]
100 %add118 = trunc i64 %tmp47 to i32 ; <i32> [#uses=1]
101 %tmp10 = getelementptr %struct.Bu* %bu, i64 %indvar, i32 2 ; <i32*> [#uses=1]
102 %tmp11 = load i32* %tmp10 ; <i32> [#uses=0]
103 tail call void undef(i32 %add22)
104 tail call void undef(i32 %add28)
105 tail call void undef(i32 %add34)
106 tail call void undef(i32 %add40)
107 tail call void undef(i32 %add46)
108 tail call void undef(i32 %add52)
109 tail call void undef(i32 %add58)
110 tail call void undef(i32 %add64)
111 tail call void undef(i32 %add70)
112 tail call void undef(i32 %add82)
113 tail call void undef(i32 %add88)
114 tail call void undef(i32 %add94)
115 tail call void undef(i32 %add100)
116 tail call void undef(i32 %add106)
117 tail call void undef(i32 %add112)
118 tail call void undef(i32 %add118)
119 br label %for.body123
121 for.body123: ; preds = %for.body123, %lor.lhs.false
122 %j.03 = phi i32 [ 0, %lor.lhs.false ], [ %inc, %for.body123 ] ; <i32> [#uses=2]
123 %add129 = add i32 %mul, %j.03 ; <i32> [#uses=1]
124 tail call void undef(i32 %add129)
125 %inc = add nsw i32 %j.03, 1 ; <i32> [#uses=1]
126 br i1 undef, label %for.inc131, label %for.body123
128 for.inc131: ; preds = %for.body123, %for.body
129 %indvar.next = add i64 %indvar, 1 ; <i64> [#uses=1]
130 br i1 undef, label %for.end134, label %for.body
132 for.end134: ; preds = %for.inc131
136 ; LSR needs to remember inserted instructions even in postinc mode, because
137 ; there could be multiple subexpressions within a single expansion which
138 ; require insert point adjustment.
141 define fastcc i32 @GetOptimum() nounwind {
145 bb1: ; preds = %bb1, %bb
146 %t = phi i32 [ 0, %bb ], [ %t2, %bb1 ] ; <i32> [#uses=1]
147 %t2 = add i32 %t, undef ; <i32> [#uses=3]
148 br i1 undef, label %bb1, label %bb3
151 %t4 = add i32 undef, -1 ; <i32> [#uses=1]
154 bb5: ; preds = %bb16, %bb3
155 %t6 = phi i32 [ %t17, %bb16 ], [ 0, %bb3 ] ; <i32> [#uses=3]
156 %t7 = add i32 undef, %t6 ; <i32> [#uses=2]
157 %t8 = add i32 %t4, %t6 ; <i32> [#uses=1]
158 br i1 undef, label %bb9, label %bb10
163 bb10: ; preds = %bb9, %bb5
164 br i1 undef, label %bb11, label %bb16
166 bb11: ; preds = %bb10
167 %t12 = icmp ugt i32 %t7, %t2 ; <i1> [#uses=1]
168 %t13 = select i1 %t12, i32 %t2, i32 %t7 ; <i32> [#uses=1]
171 bb14: ; preds = %bb11
172 store i32 %t13, i32* null
175 bb16: ; preds = %bb10
176 %t17 = add i32 %t6, 1 ; <i32> [#uses=1]