Re: [Qemu-trivial] [Qemu-devel] [PULL 00/13] Trivial branch patches

[Markus Armbruster]

Alex Bennée <address@hidden> writes:

> Markus Armbruster <address@hidden> writes:
>
>> Alex Bennée <address@hidden> writes:
>>
>>> Peter Maydell <address@hidden> writes:
>>>
>>>> On Thu, 2 May 2019 at 19:58, Laurent Vivier <address@hidden> wrote:
>>>>>
>>>>> The following changes since commit 
>>>>> 8482ff2eb3bb95020eb2f370a9b3ea26511e41df:
>>>>>
>>>>>   Merge remote-tracking branch 'remotes/jnsnow/tags/bitmaps-pull-request' 
>>>>> into staging (2019-05-02 12:04:51 +0100)
>>>>>
>>>>> are available in the Git repository at:
>>>>>
>>>>>   git://github.com/vivier/qemu.git tags/trivial-branch-pull-request
>>>>>
>>>>> for you to fetch changes up to f7c03a68b814254df414977ff299468fbf0fa1c0:
>>>>>
>>>>>   sockets: avoid string truncation warnings when copying UNIX path 
>>>>> (2019-05-02 20:50:47 +0200)
>>>>>
>>>>> ----------------------------------------------------------------
>>>>> Pull request trivial branch 2019-05-02
>>>>>
>>>>> ----------------------------------------------------------------
>>>>
>>>> Markus's "Clean up includes" patch breaks 'make check' on
>>>> big-endian hosts:
>>>> cd /home/pm215/qemu/build/all/tests/fp && ./fp-test -s -l 1 -r all  
>>>> i32_to_f16 i
>>>> 64_to_f16 i32_to_f32 i64_to_f32 i32_to_f64 i64_to_f64 i32_to_f128 
>>>> i64_to_f128 >
>>>>  int-to-float.out 2>&1 || (cat  int-to-float.out && exit 1;)
>>>>>> Testing i32_to_f16, rounding near_even
>>>> ^M372 tests total.
>>>> ^M372 tests performed.
>>>> In 372 tests, no errors found in i32_to_f16, rounding near_even.
>>>>>> Testing i32_to_f16, rounding minMag
>>>> [...]
>>>> ^MErrors found in i32_to_f128:
>>>> 1006FFFF  => +0000.000000000000401B006FFFF00000 .....
>>>>         expected +401B.006FFFF000000000000000000000 .....
>>>> ^M0001DDEB  => +0000.000000000000400FDDEB00000000 .....
>>>>         expected +400F.DDEB000000000000000000000000 .....
>>>> ^MFFF6FFFC  => +0000.000000000000C012200080000000 .....
>>>>         expected -4012.2000800000000000000000000000 .....
>>>> ^MFFFFFFF8  => +0000.000000000000C002000000000000 .....
>>>>         expected -4002.0000000000000000000000000000 .....
>>>> [etc]
>>>
>>> Considering the header clean-ups moved bswap related stuff and the
>>> patterns look incorrectly swapped something has gone fishy.
>>>
>>> If it's just stuff touching f128 then we have some magic in
>>> softfloat-types:
>>>
>>>  typedef struct {
>>>  #ifdef HOST_WORDS_BIGENDIAN
>>>      uint64_t high, low;
>>>  #else
>>>      uint64_t low, high;
>>>  #endif
>>>  } float128;
>>>
>>> but I would have though HOST_WORDS_BIGENDIAN is in the config. I shall
>>> have a poke once I get onto the s390 machine.
>>
>> Any luck?
>>
>> In my own poking, I stumbled over
>>
>>     #ifndef HOST_WORDS_BIGENDIAN
>>     #define LITTLEENDIAN 1
>>     /* otherwise do not define it */
>>     #endif
>>
>> in platform.h.  Is LITTLEENDIAN unused, or am I confused?
>
> I couldn't replicate the bug with access to my s390 account so I think
> it might be existing build artefact related which is odd.

Any ideas on how to debug this further?