xref: /llvm-project/polly/test/CodeGen/non_affine_float_compare.ll (revision 4b3a878e8a24c13f8dc435e91304d8692d8cb42b)
1; RUN: opt %loadNPMPolly -passes=polly-codegen \
2; RUN:     -polly-allow-nonaffine-branches -S -verify-dom-info \
3; RUN:     < %s | FileCheck %s
4;
5;    void f(float *A) {
6;      for (int i = 0; i < 1024; i++)
7;        if (A[i] == A[i - 1])
8;          A[i]++;
9;       A[i]++;
10;    }
11;
12;
13; CHECK: polly.stmt.bb2:
14; CHECK:   %[[offset:.*]] = shl nuw nsw i64 %polly.indvar, 2
15; CHECK:   %scevgep[[R0:[0-9]*]] = getelementptr i8, ptr %A, i64 %[[offset]]
16; CHECK:   %tmp3_p_scalar_ = load float, ptr %scevgep[[R0]], align 4, !alias.scope !2, !noalias !5
17; CHECK:   %[[offset2:.*]] = shl nuw nsw i64 %polly.indvar, 2
18; CHECK:   %scevgep[[R2:[0-9]*]] = getelementptr i8, ptr %scevgep{{[0-9]*}}, i64 %[[offset2]]
19; CHECK:   %tmp6_p_scalar_ = load float, ptr %scevgep[[R2]], align 4, !alias.scope !2, !noalias !5
20; CHECK:   %p_tmp7 = fcmp oeq float %tmp3_p_scalar_, %tmp6_p_scalar_
21; CHECK:   br i1 %p_tmp7, label %polly.stmt.bb8, label %polly.stmt.bb12.[[R:[a-zA-Z_.0-9]*]]
22
23; CHECK: polly.stmt.bb8:
24; CHECK:   %[[offset3:.*]] = shl nuw nsw i64 %polly.indvar, 2
25; CHECK:   %scevgep[[R3:[0-9]*]] = getelementptr i8, ptr %A, i64 %[[offset3]]
26; CHECK:   %tmp10_p_scalar_ = load float, ptr %scevgep[[R3]], align 4, !alias.scope !2, !noalias !5
27; CHECK:   %p_tmp11 = fadd float %tmp10_p_scalar_, 1.000000e+00
28; CHECK:   store float %p_tmp11, ptr %scevgep[[R3]], align 4, !alias.scope !2, !noalias !5
29; CHECK:   br label %polly.stmt.bb12.[[R]]
30
31; CHECK: polly.stmt.bb12.[[R]]:
32; CHECK:   br label %polly.stmt.bb12
33
34; CHECK: polly.stmt.bb12:
35; CHECK:   %[[offset4:.*]] = shl nuw nsw i64 %polly.indvar, 2
36; CHECK:   %scevgep[[R4:[0-9]*]] = getelementptr i8, ptr %A, i64 %[[offset4]]
37; CHECK:   %tmp10b_p_scalar_ = load float, ptr %scevgep[[R4]], align 4, !alias.scope !2, !noalias !5
38; CHECK:   %p_tmp11b = fadd float %tmp10b_p_scalar_, 1.000000e+00
39; CHECK:   store float %p_tmp11b, ptr %scevgep[[R4]], align 4, !alias.scope !2, !noalias !5
40; CHECK:   %polly.indvar_next = add nsw i64 %polly.indvar, 1
41; CHECK:   %polly.loop_cond = icmp sle i64 %polly.indvar_next, 1023
42; CHECK:   br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit
43
44target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
45
46define void @f(ptr %A) {
47bb:
48  br label %bb1
49
50bb1:                                              ; preds = %bb13, %bb
51  %indvars.iv = phi i64 [ %indvars.iv.next, %bb13 ], [ 0, %bb ]
52  %exitcond = icmp ne i64 %indvars.iv, 1024
53  br i1 %exitcond, label %bb2, label %bb14
54
55bb2:                                              ; preds = %bb1
56  %tmp = getelementptr inbounds float, ptr %A, i64 %indvars.iv
57  %tmp3 = load float, ptr %tmp, align 4
58  %tmp4 = add nsw i64 %indvars.iv, -1
59  %tmp5 = getelementptr inbounds float, ptr %A, i64 %tmp4
60  %tmp6 = load float, ptr %tmp5, align 4
61  %tmp7 = fcmp oeq float %tmp3, %tmp6
62  br i1 %tmp7, label %bb8, label %bb12
63
64bb8:                                              ; preds = %bb2
65  %tmp9 = getelementptr inbounds float, ptr %A, i64 %indvars.iv
66  %tmp10 = load float, ptr %tmp9, align 4
67  %tmp11 = fadd float %tmp10, 1.000000e+00
68  store float %tmp11, ptr %tmp9, align 4
69  br label %bb12
70
71bb12:                                             ; preds = %bb8, %bb2
72  %tmp9b = getelementptr inbounds float, ptr %A, i64 %indvars.iv
73  %tmp10b = load float, ptr %tmp9b, align 4
74  %tmp11b = fadd float %tmp10b, 1.000000e+00
75  store float %tmp11b, ptr %tmp9b, align 4
76  br label %bb13
77
78bb13:                                             ; preds = %bb12
79  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
80  br label %bb1
81
82bb14:                                             ; preds = %bb1
83  ret void
84}
85