Gcc LTO: Limit Optimization Scope

Question

Creating an LTO for a fairly large shared library (many template instances) takes quite a while (> 10 minutes). Now I know a few things about the library and I can specify some kind of "black list" in the form of object files that do not need to be analyzed together (because there are no calls between them that need to be embedded either), or I could specify groups of objects files to be analyzed together. Is this possible somehow (without lib separation)?

Answer 1

There is a small function ldcalled -r/ --relocatable, which can be used to combine several object files into one, which can subsequently be associated with the final product. If you can get LTO here, but not later, you can have the type of “partial” LTO that you are looking for.

Sadly ld -rit won’t work; it simply combines all the LTO information that will be processed later. But calling it using the gcc ( gcc -r) driver seems to work:

ac

int a() {
    return 42;
}

bc

int a(void);

int b() {
    return a();
}

c.c

int b(void);

int c() {
    return b();
}

dc

int c(void);

int main() {
    return c();
}

$ gcc -O3 -flto -c [a-d].c
$ gcc -O3 -r -nostdlib a.o b.o -o g1.o
$ gcc -O3 -r -nostdlib c.o d.o -o g2.o
$ gcc -O3 -fno-lto g1.o g2.o
$ objdump -d a.out
...
00000000000004f0 <main>:
 4f0:   e9 1b 01 00 00          jmpq   610 <b>
...
0000000000000610 <b>:
 610:   b8 2a 00 00 00          mov    $0x2a,%eax
 615:   c3                      retq   
...

So, I main()got optimization before return b();, but b()got optimization before return 42;, but there was no interprocedural optimization between the two groups.

Answer 2

, a.c b.c , c.c d.c . -combine GCC :

$ gcc -O3 -c -combine a.c b.c -o group1.o
$ gcc -O3 -c -combine c.c d.c -o group2.o

, LTO, -combine .

Edit

-combine C. #include :

// file group1.cpp
#include "a.cpp"
#include "b.cpp"

// file group2.cpp
#include "c.cpp"
#include "d.cpp"

LTO :

g++ -O3 group1.cpp group2.cpp

LTO.

, , . . , . .

Answer 3

You can completely exclude the object file from the process of optimizing link time by simply building it without -flto.