This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [lto]: Patch to restructure file specific information.
Jan Hubicka wrote:
>> Jan Hubicka wrote:
>>
>>>> i had seen them at the last summit. Mark Mitchell had fantasized to me
>>>> once about wanting a patch that would clean out all of the temporary
>>>> information that the front ends produced right after gimplifying. I
>>>> think that such a patch should be strongly encouraged after we get rid
>>>> of the lang hooks and other odds and ends like the front ends not
>>>> gimplifiying the static initializers until late.
>>>>
>>>> However, with lto, that pie chart means nothing. There, for large
>>>> programs, there will be no front end garbage, it will all be function
>>>> bodies and what is necessary to compile the function currently under the
>>>> microscope.
>>>>
>>>>
>>> I hope so, that is why I want to regenerate them. The table mostly
>>> suggested that type and symbol tables are places where a lot of memory
>>> goes, but it might be because we hold around dead stuff that won't be
>>> preserved across LTO if we are cureful.
>>>
>>> Just out of curiosity, do you know how much memory was needed for LTO
>>> bootstrap and how does it compare with --combine?
>>>
>>> Honza
>>>
>>>
>>>> kenny
>>>>
>>>>
>> we cannot bootstrap because of the problems listed on the wiki. We are
>> able to run the c spec marks becasee these contain no gcc extensions.
>> but we cannot lto the libraries.
>>
>
> Ah, i tought there was a point when LTO was bootstrapping but it is not
> anymore. Comparing memory footprint on any of bigger SPEC testcase
> would be interesting too. I will play a bit with this.
>
> Honza
>
>> kenny
>>
no, what it will do is compile all of the gcc i. files in isolation.
the test for combining things was to make the c spec marks all work that
way.
kenny