Re: lto pseudo-object files and fixed registers

[Richard Guenther <richard dot guenther at gmail dot com>]

On Tue, Feb 7, 2012 at 1:57 PM, Konstantin Vladimirov
<konstantin.vladimirov@gmail.com> wrote:
> Hi,
>
> That is good solution, thanks.
>
> But what if I want to compile e.o and d.o with cross-module inlining
> (but also with fixed regs and so, without lto, as you are suggesting)?
> On gcc-4.3.3, I had "combine" option for such cases. Is it completely
> impossible in gcc 4.6.2?

You then can do

 gcc $OPTIONS -flto a.c -o a.o
 gcc $OPTIONS -flto b.c -o b.o
 gcc $OPTIONS -ffixed-r9 -ffixed-r10 -flto d.c -o d.o
 gcc $OPTIONS -ffixed-r9 -ffixed-r10 -flto e.c -o e.o
 gcc $OPTIONS -flto a.o b.o -o non-fixed-reg-part.o -r -nostdlib
 gcc $OPTIONS -flto -ffixed-r9 -ffixed-r10 d.o e.o -o fixed-reg-part.o
-r -nostdlib
 gcc non-fixed-reg-part.o fixed-reg-part.o

thus, optimize both pieces via partial LTO linking (-r, maybe the -nostdlib is
not needed) and then do the final link separately.

Richard.

> ---
> With best regards, Konstantin
>
> On Tue, Feb 7, 2012 at 4:37 PM, Richard Guenther
> <richard.guenther@gmail.com> wrote:
>> On Tue, Feb 7, 2012 at 1:26 PM, Konstantin Vladimirov
>> <konstantin.vladimirov@gmail.com> wrote:
>>> Hi,
>>>
>>> Consider some project, consisting of files: a.c, b.c, d.c and e.c
>>>
>>> Compiler is gcc 4.6.2
>>>
>>> Files a.c and b.c are performance bottlenecks and requires heavy
>>> cross-module inline, so must be compiled with -flto option
>>> Files d.c and e.c is preffered to be compiled with lto option too, but
>>> they are of special usage, and requires some registers (say r9 and
>>> r10) to be fixed (with -ffixed-<reg> option) during compilation.
>>>
>>> All these files forms single binary.
>>>
>>> Now the problem is: if I compiling
>>>
>>> gcc $OPTIONS -flto a.c -o a.o
>>> gcc $OPTIONS -flto b.c -o b.o
>>> gcc $OPTIONS -ffixed-r9 -ffixed-r10 -flto d.c -o d.o
>>> gcc $OPTIONS -ffixed-r9 -ffixed-r10 -flto e.c -o d.o
>>>
>>> and then
>>>
>>> gcc $OPTIONS -flto a.o b.o d.o e.o -o a.out
>>>
>>> Then registers inside d.o and e.o are being reallocated at link time,
>>> and r9, r10 are used in the d.o and e.o parts in the resulted binary.
>>> Also I can not specify fixed regs to final link, because this will fix
>>> registers in a.o and b.o parts, that will affect performance.
>>>
>>> The best way for me seems to somehow separately link pseudo-object
>>> files a.o and b.o with -flto to simple object (say x.o), and then link
>>> e.o and d.o to single, say, y.o, and then call linker to finally link
>>> binary without cross-module optimizations. But I can not find
>>> possibility to do it, and I doubt if this at all conforms with lto
>>> ideology.
>>
>> That's indeed the way to go, but you don't need anything fancy like
>> partial linking.
>>
>> ?gcc $OPTIONS -flto a.c -o a.o
>> ?gcc $OPTIONS -flto b.c -o b.o
>> ?gcc $OPTIONS -ffixed-r9 -ffixed-r10 d.c -o d.o
>> ?gcc $OPTIONS -ffixed-r9 -ffixed-r10 e.c -o e.o
>> ?gcc $OPTIONS -flto a.o b.o d.o e.o
>>
>> should do it automatically. ?Just make sure to not compile d.o and e.o
>> with -flto.
>>
>> Richard.
>>
>>> So, I want to perform link-time optimizations between a.o and b.o, and
>>> don't want them between (a.o or b.o) and (d.o or e.o)
>>>
>>> How can I approach this?
>>>
>>> Thanks in advance for everyone, who will help.
>>>
>>> ---
>>> With best regards, Konstantin