1# RUN: llvm-mc -mcpu=skylake -filetype=obj -triple x86_64-pc-linux-gnu -x86-pad-max-prefix-size=0 %s | llvm-objdump -d - | FileCheck %s --check-prefix=NOPAD
2# RUN: llvm-mc -mcpu=skylake -filetype=obj -triple x86_64-pc-linux-gnu -x86-pad-max-prefix-size=0 -x86-pad-for-align=1 %s | llvm-objdump -d - | FileCheck %s
3
4# This test exercises only the padding via relaxation logic.  The  interaction
5# etween prefix padding and relaxation logic can be seen in align-via-padding.s
6
7  .file "test.c"
8  .text
9  .section  .text
10
11# NOPAD-LABEL: <.text>:
12# NOPAD-NEXT:     0: eb 1f           jmp 0x21 <foo>
13# NOPAD-NEXT:     2: eb 1d           jmp 0x21 <foo>
14# NOPAD-NEXT:     4: eb 1b           jmp 0x21 <foo>
15# NOPAD-NEXT:     6: eb 19           jmp 0x21 <foo>
16# NOPAD-NEXT:     8: eb 17           jmp 0x21 <foo>
17# NOPAD-NEXT:     a: eb 15           jmp 0x21 <foo>
18# NOPAD-NEXT:     c: eb 13           jmp 0x21 <foo>
19# NOPAD-NEXT:     e: 66 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00  nopw    %cs:(%rax,%rax)
20# NOPAD-NEXT:    1d: 0f 1f 00        nopl (%rax)
21# NOPAD-NEXT:    20: cc              int3
22
23  # Demonstrate that we can relax instructions to provide padding, not
24  # just insert nops.  jmps are being used for ease of demonstration.
25  # CHECK: .text
26  # CHECK: 0: eb 1f                         jmp 0x21 <foo>
27  # CHECK: 2: e9 1a 00 00 00                jmp 0x21 <foo>
28  # CHECK: 7: e9 15 00 00 00                jmp 0x21 <foo>
29  # CHECK: c: e9 10 00 00 00                jmp 0x21 <foo>
30  # CHECK: 11: e9 0b 00 00 00               jmp 0x21 <foo>
31  # CHECK: 16: e9 06 00 00 00               jmp 0x21 <foo>
32  # CHECK: 1b: e9 01 00 00 00               jmp 0x21 <foo>
33  # CHECK: 20: cc                           int3
34  .p2align 4
35  jmp foo
36  jmp foo
37  jmp foo
38  jmp foo
39  jmp foo
40  jmp foo
41  jmp foo
42  .p2align 5
43  int3
44foo:
45  ret
46
47  # Check that we're not shifting aroudn the offsets of labels - doing
48  # that would require a further round of relaxation
49  # CHECK: <bar>:
50  # CHECK: 22: eb fe                          jmp 0x22 <bar>
51  # CHECK: 24: 66 66 66 2e 0f 1f 84 00 00 00 00 00 nopw %cs:(%rax,%rax)
52  # CHECK: 30: 0f 0b                          ud2
53
54bar:
55  jmp bar
56nobypass:
57  .p2align 4
58  ud2
59
60
61  # Canonical toy loop to show benefit - we can align the loop header with
62  # fewer nops by relaxing the branch, even though we don't need to
63  # CHECK: <loop_preheader>:
64  # CHECK: 45: 48 85 c0                       testq %rax, %rax
65  # CHECK: 48: 0f 8e 22 00 00 00              jle 0x70 <loop_exit>
66  # CHECK: 4e: 66 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 nopw %cs:(%rax,%rax)
67  # CHECK: 5d: 0f 1f 00                       nopl (%rax)
68  # CHECK: <loop_header>:
69  # CHECK: 60: 48 83 e8 01                    subq $1, %rax
70  # CHECK: 64: 48 85 c0                       testq %rax, %rax
71  # CHECK: 67: 7e 07                          jle 0x70 <loop_exit>
72  # CHECK: 69: e9 f2 ff ff ff                 jmp 0x60 <loop_header>
73  # CHECK: 6e: 66 90                          nop
74  # CHECK: <loop_exit>:
75  # CHECK: 70: c3                             retq
76  .p2align 5
77  .skip 5
78loop_preheader:
79  testq %rax, %rax
80  jle loop_exit
81  .p2align 5
82loop_header:
83  subq $1, %rax
84  testq %rax, %rax
85  jle loop_exit
86  jmp loop_header
87  .p2align 4
88loop_exit:
89  ret
90