[PATCH v2 13/14] x86: convert testcases to use .insn

[Jan Beulich]

On 20.04.2023 11:09, Clément Chigot wrote:
> On Thu, Apr 20, 2023 at 11:01 AM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> On 20.04.2023 10:56, Clément Chigot wrote:
>>>> --- a/gas/testsuite/gas/i386/x86-64-opcode.s
>>>> +++ b/gas/testsuite/gas/i386/x86-64-opcode.s
>>>> @@ -458,16 +458,16 @@
>>>>         int3
>>>>         int    $0x90
>>>>
>>>> -       .byte 0xf6, 0xc9, 0x01
>>>> -       .byte 0x66, 0xf7, 0xc9, 0x02, 0x00
>>>> -       .byte 0xf7, 0xc9, 0x04, 0x00, 0x00, 0x00
>>>> -       .byte 0x48, 0xf7, 0xc9, 0x08, 0x00, 0x00, 0x00
>>>> -       .byte 0xc0, 0xf0, 0x02
>>>> -       .byte 0xc1, 0xf0, 0x01
>>>> -       .byte 0x48, 0xc1, 0xf0, 0x01
>>>> -       .byte 0xd0, 0xf0
>>>> -       .byte 0xd1, 0xf0
>>>> -       .byte 0x48, 0xd1, 0xf0
>>>> -       .byte 0xd2, 0xf0
>>>> -       .byte 0xd3, 0xf0
>>>> -       .byte 0x48, 0xd3, 0xf0
>>>> +       .insn 0xf6/1, $1, %cl
>>>> +       .insn 0xf7/1, $2{:u16}, %cx
>>>> +       .insn 0xf7/1, $4{:u32}, %ecx
>>>> +       .insn 0xf7/1, $8{:s32}, %rcx
>>>> +       .insn 0xc0/6, $2, %al
>>>> +       .insn 0xc1/6, $1, %eax
>>>> +       .insn 0xc1/6, $1, %rax
>>>> +       .insn 0xd0/6, %al
>>>> +       .insn 0xd1/6, %eax
>>>> +       .insn 0xd1/6, %rax
>>>> +       .insn 0xd2/6, %al
>>>> +       .insn 0xd3/6, %eax
>>>> +       .insn 0xd3/6, %rax
>>>
>>> The test is failing on my side when building with --target=x86_64-elf.
>>> I'm not sure what's wrong yet but gas seems to ignore everything after "/":
>>>   | $ ../../binutils/objdump  -drw tmpdir/x86-64-opcode.o
>>>   |  ...
>>>   |  4ea: f6 f7                div    %bh
>>>   |  4ec: f7 f7                div    %edi
>>>   |  4ee: c0 c1 c1              rol    $0xc1,%cl
>>>   |  4f1: d0 d1                rcl    %cl
>>>   |  4f3: d1 d2                rcl    %edx
>>>   |  4f5: d3 d3                rcl    %cl,%ebx
>>
>> Right, and I think I did address all of these issues (there were more than
>> just here) in what was committed (and in fact already in v2), by passing
>> --divide to as. Can you confirm --divide does not take the intended effect
>> in that case?
> 
> --divide is not passed to x86_64-opcode test.

Well, you continue to supply ambiguous information up to here; it only
becomes clear ...

> But adding it resolves the issue:
>   | $ ../as-new  --x32 --divide -J  -o tmpdir/x86-64-opcode.o

... here that what you mean is the ilp32/x86-64-opcode test (which is a
clone of the x86-64-opcode one). So yes, I did overlook the need to add
--divide there as well.

Jan

> .../x86-64-opcode.s
>   | $ ../../binutils/objdump  -drw tmpdir/x86-64-opcode.o
>   |  4ea: f6 c9 01              test   $0x1,%cl
>   |  4ed: 66 f7 c9 02 00        test   $0x2,%cx
>   |  4f2: f7 c9 04 00 00 00    test   $0x4,%ecx
>   |  4f8: 48 f7 c9 08 00 00 00 test   $0x8,%rcx
>   |  4ff: c0 f0 02              shl    $0x2,%al
>   |  502: c1 f0 01              shl    $0x1,%eax
>   |  505: 48 c1 f0 01          shl    $0x1,%rax
>   |  509: d0 f0                shl    %al
>   |  50b: d1 f0                shl    %eax
>   |  50d: 48 d1 f0              shl    %rax
>   |  510: d2 f0                shl    %cl,%al
>   |  512: d3 f0                shl    %cl,%eax
>   |  514: 48 d3 f0              shl    %cl,%rax