llvm/test/MC/X86/align-via-padding.s

   1 # RUN: llvm-mc -mcpu=skylake -filetype=obj -triple x86_64-pc-linux-gnu %s -x86-pad-max-prefix-size=5 -x86-pad-for-align=1 | llvm-objdump --no-print-imm-hex -d - | FileCheck %s
   2
   3 # This test file highlights the interactions between prefix padding and
   4 # relaxation padding.
   5
   6   .file "test.c"
   7   .text
   8   .section  .text
   9   # We can both relax and prefix for padding purposes, but the moment, we
  10   # can't prefix without first having relaxed.
  11   # CHECK: .text
  12   # CHECK:  0: eb 1f                         jmp
  13   # CHECK:  2: eb 1d                         jmp
  14   # CHECK:  4: eb 1b                         jmp
  15   # CHECK:  6: eb 19                         jmp
  16   # CHECK:  8: eb 17                         jmp
  17   # CHECK:  a: 2e 2e 2e 2e 2e e9 0d 00 00 00 jmp
  18   # CHECK: 14: 2e 2e 2e 2e 2e e9 03 00 00 00 jmp
  19   # CHECK: 1e: 66 90                         nop
  20   # CHECK: 20: cc                            int3
  21   .p2align 4
  22   jmp foo
  23   jmp foo
  24   jmp foo
  25   jmp foo
  26   jmp foo
  27   jmp foo
  28   jmp foo
  29   .p2align 5
  30   int3
  31 foo:
  32   ret
  33
  34   # Canonical toy loop to show benefit - we can align the loop header with
  35   # fewer nops by relaxing the branch, even though we don't need to
  36   # CHECK: <loop_preheader>:
  37   # CHECK: 45: 48 85 c0                       testq %rax, %rax
  38   # CHECK: 48: 2e 2e 2e 2e 0f 8e 1e 00 00 00  jle 0x70 <loop_exit>
  39   # CHECK: 52: 66 66 66 66 66 2e 0f 1f 84 00 00 00 00 00        nopw    %cs:(%rax,%rax)
  40   # CHECK: <loop_header>:
  41   # CHECK: 60: 48 83 e8 01                    subq $1, %rax
  42   # CHECK: 64: 48 85 c0                       testq %rax, %rax
  43   # CHECK: 67: 7e 07                          jle 0x70 <loop_exit>
  44   # CHECK: 69: 2e 2e e9 f0 ff ff ff           jmp
  45   # CHECK: <loop_exit>:
  46   # CHECK: 70: c3                             retq
  47   .p2align 5
  48   .skip 5
  49 loop_preheader:
  50   testq %rax, %rax
  51   jle loop_exit
  52   .p2align 5
  53 loop_header:
  54   subq $1, %rax
  55   testq %rax, %rax
  56   jle loop_exit
  57   jmp loop_header
  58   .p2align 4
  59 loop_exit:
  60   ret
  61
  62   # Correctness cornercase - can't prefix pad jmp without having relaxed it
  63   # first as doing so would make the relative offset too large
  64   # CHECK: fd: cc                             int3
  65   # CHECK: fe: eb 80                          jmp 0x80 <loop_exit+0x10>
  66   # CHECK: 100: cc                              int3
  67 .p2align 5
  68 .L1:
  69 .rept 126
  70   int3
  71 .endr
  72   jmp .L1
  73 .rept 30
  74   int3
  75 .endr
  76 .p2align 5