1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mattr=+avx512f | FileCheck %s  --check-prefix=CHECK --check-prefix=KNL
3; RUN: llc < %s -mattr=+avx512f -mattr=+avx512vl -mattr=+avx512bw -mattr=+avx512dq | FileCheck %s  --check-prefix=CHECK --check-prefix=SKX
4
5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6target triple = "x86_64-unknown-linux-gnu"
7
8define void @func() {
9; CHECK-LABEL: func:
10; CHECK:       # %bb.0: # %bb1
11; CHECK-NEXT:    xorl %eax, %eax
12; CHECK-NEXT:    testb %al, %al
13; CHECK-NEXT:    je .LBB0_1
14; CHECK-NEXT:  # %bb.4: # %L_30
15; CHECK-NEXT:    retq
16; CHECK-NEXT:  .LBB0_1: # %bb56
17; CHECK-NEXT:    xorl %eax, %eax
18; CHECK-NEXT:    jmp .LBB0_2
19; CHECK-NEXT:    .p2align 4, 0x90
20; CHECK-NEXT:  .LBB0_3: # %bb35
21; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
22; CHECK-NEXT:    testb %al, %al
23; CHECK-NEXT:  .LBB0_2: # %bb33
24; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
25; CHECK-NEXT:    testb %al, %al
26; CHECK-NEXT:    jne .LBB0_2
27; CHECK-NEXT:    jmp .LBB0_3
28bb1:
29  br i1 undef, label %L_10, label %L_10
30
31L_10:                                             ; preds = %bb1, %bb1
32  br i1 undef, label %L_30, label %bb56
33
34bb56:                                             ; preds = %L_10
35  br label %bb33
36
37bb33:                                             ; preds = %bb51, %bb56
38  %r111 = load i64, i64* undef, align 8
39  br i1 undef, label %bb51, label %bb35
40
41bb35:                                             ; preds = %bb33
42  br i1 undef, label %L_19, label %bb37
43
44bb37:                                             ; preds = %bb35
45  %r128 = and i64 %r111, 576460752303423488
46  %phitmp = icmp eq i64 %r128, 0
47  br label %L_19
48
49L_19:                                             ; preds = %bb37, %bb35
50  %"$V_S25.0" = phi i1 [ %phitmp, %bb37 ], [ true, %bb35 ]
51  br i1 undef, label %bb51, label %bb42
52
53bb42:                                             ; preds = %L_19
54  %r136 = select i1 %"$V_S25.0", i32* undef, i32* undef
55  br label %bb51
56
57bb51:                                             ; preds = %bb42, %L_19, %bb33
58  br i1 false, label %L_30, label %bb33
59
60L_30:                                             ; preds = %bb51, %L_10
61  ret void
62}
63
64; The following test generates suboptimal code on AVX-512
65; PR 28175
66define i64 @func2(i1 zeroext %i, i32 %j) {
67; CHECK-LABEL: func2:
68; CHECK:       # %bb.0: # %entry
69; CHECK-NEXT:    testl %esi, %esi
70; CHECK-NEXT:    je .LBB1_1
71; CHECK-NEXT:  # %bb.2: # %if.then
72; CHECK-NEXT:    jmp bar # TAILCALL
73; CHECK-NEXT:  .LBB1_1: # %if.end
74; CHECK-NEXT:    movzbl %dil, %eax
75; CHECK-NEXT:    orq $-2, %rax
76; CHECK-NEXT:    retq
77entry:
78  %tobool = icmp eq i32 %j, 0
79  br i1 %tobool, label %if.end, label %if.then
80
81if.then:                                          ; preds = %entry
82  %call = tail call i64 @bar()
83  br label %return
84
85if.end:                                           ; preds = %entry
86  %conv = zext i1 %i to i64
87  %or = or i64 %conv, -2
88  br label %return
89
90return:                                           ; preds = %if.end, %if.then
91  %or.sink = phi i64 [ %or, %if.end ], [ %call, %if.then ]
92  ret i64 %or.sink
93}
94
95declare i64 @bar()
96