Re: [PATCH] gdb.trace: Add a testcase for tdesc in tfile.

[Antoine Tremblay <antoine.tremblay@ericsson.com>]

Marcin Kościelnicki writes:

> On 12/02/16 19:49, Antoine Tremblay wrote:
>>
>> Marcin Kościelnicki writes:
>>
>>> On 12/02/16 19:31, Antoine Tremblay wrote:
>>>>
>>>> Marcin Kościelnicki writes:
>>>>
>>>>> On 11/02/16 14:00, Pedro Alves wrote:
>>>>>> On 02/11/2016 10:14 AM, Marcin Kościelnicki wrote:
>>>>>>> This tests whether $ymm15 can be correctly collected and printed from
>>>>>>> tfile.  It covers:
>>>>>>>
>>>>>>> - storing tdesc in tfile (without that, $ymm15 doesn't exist)
>>>>>>> - ax_pseudo_register_collect for x86 (without that, $ymm15 cannot be
>>>>>>>      collected)
>>>>>>> - register order in tfile_fetch_registers (without that, $ymm15h is
>>>>>>>      fetched from wrong position)
>>>>>>> - off-by-one in tfile_fetch_registers (without that, $ymm15h is
>>>>>>>      incorrectly considered to be out of bounds)
>>>>>>> - using proper tdesc in encoding tracepoint actions (without that,
>>>>>>>      internal error happens due to $ymm15h being
>>>>>>
>>>>>> OK once prereqs are in.
>>>>>>
>>>>>> Thanks,
>>>>>> Pedro Alves
>>>>>>
>>>>>
>>>>> Thanks, pushed.
>>>>
>>>> Hi,
>>>>     I've been trying to run this test on x86 but I get the following error
>>>>     while compiling tfile-avx.c :
>>>>
>>>>    binutils-gdb/build-x86/gdb/testsuite/../../../gdb/testsuite/gdb.trace/tfile-avx.c:38:19: error: invalid register name for 'a'
>>>>     register __v8si a asm("ymm15") = {
>>>>                      ^
>>>>
>>>>     I've also noticed the same error on the buildbot results see:
>>>>     http://gdb-build.sergiodj.net/builders/Debian-x86_64-m64/builds/2928/steps/test%20gdb/logs/stdio
>>>>
>>>>     My cpu (Intel(R) Core(TM) i7-4600M ) supports avx, cat /proc/cpuinfo
>>>>     shows avx and a gdb print $ymm15 returns something...
>>>>
>>>>     This is with gcc 4.8.4...
>>>>
>>>>     Am I missing something?
>>>>
>>>> Regards,
>>>> Antoine
>>>>
>>>
>>> Ugh.  It seems you need a newer gcc to recognize "ymm15" as a register
>>> name - 4.8.2 seems to want it called "xmm15" - sort of incorrect, but
>>> close enough.  gcc 5.3 still accepts that, so perhaps we should change
>>> it to xmm15 for the sake of older compilers, even if it harms readability?
>>
>> Would xmm15 still work on newer gccs ? If so I would guess it's a good
>> idea to change it given that our own buildbot test machines seem to test
>> with an older gcc...?
>
> Both xmm15 and ymm15 work just fine on gcc 4.9.3 and 5.3.0, which are 
> the only ones I have around right now.  So, let's do it.

Great, sorry I had missed that in your original mail.

>>
>> Maybe adding a note of it in the test... or if there's a way to check
>> for the gcc version ?
>
> I'll throw in a comment.
>>
>> I actually can't find the gcc doc where those names are defined at the
>> moment would you have that handy by any chance?
>
> gcc/config/i386/i386.h, ADDITIONAL_REGISTER_NAMES.
>>

Thanks!
Antoine