1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -instcombine -S -data-layout="e-m:e-i64:64-f80:128-n8:16:32:64-S128" | FileCheck %s
3
4@a = global [1000 x float] zeroinitializer, align 16
5@b = global [1000 x float] zeroinitializer, align 16
6
7define void @_Z3foov() {
8; CHECK-LABEL: @_Z3foov(
9; CHECK-NEXT:  entry:
10; CHECK-NEXT:    br label [[FOR_COND:%.*]]
11; CHECK:       for.cond:
12; CHECK-NEXT:    [[I_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]
13; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[I_0]], 1000
14; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP:%.*]]
15; CHECK:       for.cond.cleanup:
16; CHECK-NEXT:    ret void
17; CHECK:       for.body:
18; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[I_0]] to i64
19; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds [1000 x float], [1000 x float]* @a, i64 0, i64 [[TMP0]]
20; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds [1000 x float], [1000 x float]* @b, i64 0, i64 [[TMP0]]
21; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[ARRAYIDX]], align 4
22; CHECK-NEXT:    [[TMP2:%.*]] = load float, float* [[ARRAYIDX2]], align 4
23; CHECK-NEXT:    [[CMP_I:%.*]] = fcmp fast olt float [[TMP1]], [[TMP2]]
24; CHECK-NEXT:    [[TMP3:%.*]] = select i1 [[CMP_I]], float [[TMP2]], float [[TMP1]]
25; CHECK-NEXT:    store float [[TMP3]], float* [[ARRAYIDX]], align 4
26; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I_0]], 1
27; CHECK-NEXT:    br label [[FOR_COND]]
28;
29entry:
30  br label %for.cond
31
32for.cond:                                         ; preds = %for.body, %entry
33  %i.0 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
34  %cmp = icmp ult i32 %i.0, 1000
35  br i1 %cmp, label %for.body, label %for.cond.cleanup
36
37for.cond.cleanup:                                 ; preds = %for.cond
38  ret void
39
40for.body:                                         ; preds = %for.cond
41  %0 = zext i32 %i.0 to i64
42  %arrayidx = getelementptr inbounds [1000 x float], [1000 x float]* @a, i64 0, i64 %0
43  %arrayidx2 = getelementptr inbounds [1000 x float], [1000 x float]* @b, i64 0, i64 %0
44  %1 = load float, float* %arrayidx, align 4
45  %2 = load float, float* %arrayidx2, align 4
46  %cmp.i = fcmp fast olt float %1, %2
47  %__b.__a.i = select i1 %cmp.i, float* %arrayidx2, float* %arrayidx
48  %3 = bitcast float* %__b.__a.i to i32*
49  %4 = load i32, i32* %3, align 4
50  %5 = bitcast float* %arrayidx to i32*
51  store i32 %4, i32* %5, align 4
52  %inc = add nuw nsw i32 %i.0, 1
53  br label %for.cond
54}
55
56define i32 @store_bitcasted_load(i1 %cond, float* dereferenceable(4) %addr1, float* dereferenceable(4) %addr2) {
57; CHECK-LABEL: @store_bitcasted_load(
58; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[COND:%.*]], float* [[ADDR1:%.*]], float* [[ADDR2:%.*]]
59; CHECK-NEXT:    [[BC1:%.*]] = bitcast float* [[SEL]] to i32*
60; CHECK-NEXT:    [[LD:%.*]] = load i32, i32* [[BC1]], align 4
61; CHECK-NEXT:    ret i32 [[LD]]
62;
63  %sel = select i1 %cond, float* %addr1, float* %addr2
64  %bc1 = bitcast float* %sel to i32*
65  %ld = load i32, i32* %bc1
66  ret i32 %ld
67}
68
69define void @bitcasted_store(i1 %cond, float* %loadaddr1, float* %loadaddr2, float* %storeaddr) {
70; CHECK-LABEL: @bitcasted_store(
71; CHECK-NEXT:    [[SEL:%.*]] = select i1 [[COND:%.*]], float* [[LOADADDR1:%.*]], float* [[LOADADDR2:%.*]]
72; CHECK-NEXT:    [[INT_LOAD_ADDR:%.*]] = bitcast float* [[SEL]] to i32*
73; CHECK-NEXT:    [[LD:%.*]] = load i32, i32* [[INT_LOAD_ADDR]], align 4
74; CHECK-NEXT:    [[INT_STORE_ADDR:%.*]] = bitcast float* [[STOREADDR:%.*]] to i32*
75; CHECK-NEXT:    store i32 [[LD]], i32* [[INT_STORE_ADDR]], align 4
76; CHECK-NEXT:    ret void
77;
78  %sel = select i1 %cond, float* %loadaddr1, float* %loadaddr2
79  %int_load_addr = bitcast float* %sel to i32*
80  %ld = load i32, i32* %int_load_addr
81  %int_store_addr = bitcast float* %storeaddr to i32*
82  store i32 %ld, i32* %int_store_addr
83  ret void
84}
85
86define void @bitcasted_minmax_with_select_of_pointers(float* %loadaddr1, float* %loadaddr2, float* %storeaddr) {
87; CHECK-LABEL: @bitcasted_minmax_with_select_of_pointers(
88; CHECK-NEXT:    [[LD1:%.*]] = load float, float* [[LOADADDR1:%.*]], align 4
89; CHECK-NEXT:    [[LD2:%.*]] = load float, float* [[LOADADDR2:%.*]], align 4
90; CHECK-NEXT:    [[COND:%.*]] = fcmp ogt float [[LD1]], [[LD2]]
91; CHECK-NEXT:    [[LD3:%.*]] = select i1 [[COND]], float [[LD1]], float [[LD2]]
92; CHECK-NEXT:    store float [[LD3]], float* [[STOREADDR:%.*]], align 4
93; CHECK-NEXT:    ret void
94;
95  %ld1 = load float, float* %loadaddr1, align 4
96  %ld2 = load float, float* %loadaddr2, align 4
97  %cond = fcmp ogt float %ld1, %ld2
98  %sel = select i1 %cond, float* %loadaddr1, float* %loadaddr2
99  %int_load_addr = bitcast float* %sel to i32*
100  %ld = load i32, i32* %int_load_addr, align 4
101  %int_store_addr = bitcast float* %storeaddr to i32*
102  store i32 %ld, i32* %int_store_addr, align 4
103  ret void
104}
105