Re: [dpdk-dev] Compiling files with .S with GCC

["Wiles, Keith" <keith.wiles@intel.com>]

On 5/4/15, 8:42 AM, "Olivier MATZ" <olivier.matz@6wind.com> wrote:

>
>
>On 05/04/2015 04:49 PM, Wiles, Keith wrote:
>> 
>> 
>> On 5/4/15, 1:37 AM, "Olivier MATZ" <olivier.matz@6wind.com> wrote:
>> 
>>> Hello,
>>>
>>> On 04/26/2015 06:55 PM, Wiles, Keith wrote:
>>>>
>>>>
>>>> On 4/26/15, 11:53 AM, "Wiles, Keith" <keith.wiles@intel.com> wrote:
>>>>
>>>>> Hi All,
>>>>>
>>>>> I noticed in my builds with foo.S file I would get a warning from the
>>>>> compiler the foo_s.o.tmp linker file will not be used as the code is
>>>>> not
>>>>> linked. A strange error and the foo_s.o file would not be created. In
>>>>> the
>>>>> mk/internal/rte.compile-pre.mk we have the following
>>>>>
>>>>> # command to assemble a .S file to generate an object
>>>>> ifeq ($(USE_HOST),1)
>>>>> S_TO_O = $(CPP) $(HOST_CPPFLAGS) $($(@)_CPPFLAGS)
>>>>> $(HOST_EXTRA_CPPFLAGS)
>>>>> $< $(@).tmp && \
>>>>>    $(HOSTAS) $(HOST_ASFLAGS) $($(@)_ASFLAGS) $(HOST_EXTRA_ASFLAGS) -o
>>>>> $@
>>>>> $(@).tmp
>>>>> S_TO_O_STR = $(subst ','\'',$(S_TO_O)) #'# fix syntax highlight
>>>>> S_TO_O_DISP =  $(if $(V),"$(S_TO_O_STR)","  HOSTAS $(@)")
>>>>> else
>>>>> S_TO_O = $(CPP) $(CPPFLAGS) $($(@)_CPPFLAGS) $(EXTRA_CPPFLAGS) $< -o
>>>>> $(@).tmp && \
>>>>>    $(AS) $(ASFLAGS) $($(@)_ASFLAGS) $(EXTRA_ASFLAGS) -o $@ $(@).tmp
>>>>> S_TO_O_STR = $(subst ','\'',$(S_TO_O)) #'# fix syntax highlight
>>>>> S_TO_O_DISP =  $(if $(V),"$(S_TO_O_STR)","  AS $(@)")
>>>>> endif
>>>>>
>>>>> I had to change the above Œ.tmp¹ strings to Œ.s¹ to remove this
>>>>>warning
>>>>> and get the foo_s.o file created. Using the .s the file name becomes
>>>>> foo_s.o.s
>>>>
>>>> Compiler used on Ubuntu 14.04 'gcc (Ubuntu 4.8.2-19ubuntu1) 4.8.2'
>>>>
>>>>>
>>>>> Did not try this with clang or any other compiler, but I expect we
>>>>>can
>>>>> safely change the Œ.tmp¹ to Œ.s¹ IMO
>>>>>
>>>>> Anyone else notice this problem?
>>>
>>> I tested a similar use-case and I don't reproduce the issue.
>>> I don't don't understand why replacing $(@).tmp by $(@).s would
>>> solve it.
>> 
>> In the case I saw the compiler seemed to want the .s to under stand it
>>was
>> not some other type of file. Not sure why the error occurred in my
>> external build. We can leave it using .tmp if you like, but I feel .s
>> would have been the correct name for the file.
>
>Can you send a log of the error you've seen?
>Can you reproduce it outside the dpdk framework by trying to assemble
>a file called foo.tmp?
>This would show that the problem really comes from the extension of
>the file.

OK will try, but out of town this week.
>
>
>> 
>> If you think it needs to be changed to .s then you can submit a patch
>>or I
>> can or just leave it along your option.
>
>Until we really understand the cause of the issue, I don't think this
>should be changed. But I'm not opposed to change it if we have a good
>justification.
>
>
>Regards,
>Olivier
>
>
>
>> 
>> Regards,
>> ++Keith
>>>
>>> The file $(@).tmp is used as a temporary file to store the preprocessed
>>> version of the $(@).s file. I agree that using the .s extension is not
>>> a bad choice (see below), but to me it is not the proper solution to
>>> your problem.
>>>
>>>From https://gcc.gnu.org/onlinedocs/gcc/Overall-Options.html
>>>
>>>  file.s
>>>     Assembler code.
>>>  file.S
>>>  file.sx
>>>     Assembler code that must be preprocessed.
>>>
>>> Regards,
>>> Olivier
>>