[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[sen2403 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

Eason Lai <sen2403 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sen2403 at hotmail dot com

--- Comment #8 from Eason Lai <sen2403 at hotmail dot com> ---
This problem can be reproduced by arm-none-eabi-gcc 10.2.1-q4 toolchain.

The weak definition in LTO object is been inline directly even there is a
strong definition in normal object. And, the problem is gone if changing
optimization from (-Os) to (-O1 or -O0).

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[amonakov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

Alexander Monakov <amonakov at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amonakov at gcc dot gnu.org

--- Comment #9 from Alexander Monakov <amonakov at gcc dot gnu.org> ---
I can reproduce it with gcc-10.2. Why is main 'overwritable', but foo is
'available'?

cat /tmp/cchaUSjV.ltrans0.o.079i.inline

;; Function main (main, funcdef_no=0, decl_uid=4385, cgraph_uid=1,
symbol_order=1) (executed once)

weakdef.c:5:5: note: Inlining foo/0 to main/1 with frequency 1.00
foo/0 (foo) @0x7f51a512d168
  Type: function definition analyzed
  Visibility: preempted_reg external public weak
  References:
  Referring:
  Function foo/0 is inline copy in main/1
  Availability: available
  Unit id: 2
  Function flags: count:1073741824 (estimated locally) body nonfreeing_fn
  Called by: main/1 (inlined) (1073741824 (estimated locally),1.00 per call)
  Calls:
main/1 (main) @0x7f51a512d000
  Type: function definition analyzed
  Visibility: externally_visible prevailing_def public
  References:
  Referring:
  Availability: overwritable
  Unit id: 2
  Function flags: count:1073741824 (estimated locally) body
only_called_at_startup nonfreeing_fn executed_once
  Called by:
  Calls: foo/0 (inlined) (1073741824 (estimated locally),1.00 per call)

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[sen2403 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

--- Comment #10 from Eason Lai <sen2403 at hotmail dot com> ---
Add my test program.

$cat boo.c
#include "boo.h"

int get_t(void)
{
    return 1;
}

$cat main.c
#include "boo.h"

__attribute__((weak)) int get_t(void)
{
    return 0;
}

int a;
int main(void)
{
    a = get_t();
    return a;
}

$arm-none-eabi-gcc -mlittle-endian -mthumb -mcpu=cortex-m33 -Os -flto -c main.c
-o main.o
#arm-none-eabi-gcc -mlittle-endian -mthumb -mcpu=cortex-m33 -Os -fno-lto -c
boo.c -o boo.o
#arm-none-eabi-gcc --specs=nano.specs -lnosys -nostartfiles -Wl,--gc-sections 
-Wl,-Ttest.ld boo.o main.o -o main -save-temps
$arm-none-eabi-objdump -d main

main:     file format elf32-littlearm


Disassembly of section .test:

00001000 <main>:
    1000:       2000            movs    r0, #0
    1002:       4770            bx      lr

$cat "./-lm.res"
1
main.o 3
190 a62a5458565feecb PREEMPTED_REG get_t
192 a62a5458565feecb PREVAILING_DEF main
196 a62a5458565feecb PREVAILING_DEF_IRONLY a

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[amonakov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

--- Comment #11 from Alexander Monakov <amonakov at gcc dot gnu.org> ---
Marxin, you've marked this as WAITING, can you please re-evaluate? The nice
testcase from comment #2 is reproducible on trunk as well.

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[sen2403 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

--- Comment #12 from Eason Lai <sen2403 at hotmail dot com> ---
Hope this information could help.

I added "-fopt-info-inline-optimized-missed=inline.txt" in the CFLAGS to see
what happens between -Os and -O1.

Here is the output when using "-O1":
missed:   not inlinable: main/5 -> get_t/4, function not inline candidate

Here is the output when using "-Os":
main.c:13:9: missed:   not inlinable: main/5 -> get_t/4, function body can be
overwritten at link time
optimized:  Inlined get_t/4 into main/5 which now has time 3.000000 and size 4,
net change of +0.

It happens in ipa-inline.c::inline_small_functions when using "-O2" or above.

[Bug lto/91299] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[sen2403 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

--- Comment #13 from Eason Lai <sen2403 at hotmail dot com> ---
The inline rule in ipa-inline.c::inline_small_functions can be bypassed by
adding "noinline" attribute as shown below.

__attribute__((weak, noinline)) int get_t(void)
{
    return 0;
}

It's an alternative solution for me.

[Bug lto/91299] [10/11/12/13 Regression] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[amonakov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

Alexander Monakov <amonakov at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
            Summary|LTO inlines a weak          |[10/11/12/13 Regression]
                   |definition in presence of a |LTO inlines a weak
                   |non-weak definition from an |definition in presence of a
                   |ELF file                    |non-weak definition from an
                   |                            |ELF file

--- Comment #14 from Alexander Monakov <amonakov at gcc dot gnu.org> ---
gcc-4.9 used to get this right, so let's play the regression card? This should
not be in WAITING.

[Bug lto/91299] [10/11/12/13 Regression] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Target Milestone|---                         |10.5
                 CC|                            |hubicka at gcc dot gnu.org
           Priority|P3                          |P2

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
1
a-weakdef.o 2
195 5359582ee909af0f PREEMPTED_REG foo
198 5359582ee909af0f PREVAILING_DEF main


Considering foo/0 with 3 size
 to be inlined into main/1 in unknown:-1
 Estimated badness is -inf, frequency 1.00.
    Badness calculation for main/1 -> foo/0
      size growth -1, time 0.000000 unspec 2.000000  big_speedup
      -inf: Growth -1 <= 0
      Adjusted by hints -inf
Updated mod-ref summary for main/1
  loads:
    Limits: 32 bases, 16 refs
  stores:
    Limits: 32 bases, 16 refs
optimized:  Inlined foo/0 into main/1 which now has time 2.000000 and size 3,
net change of -1.

but

foo/0 (foo) @0x7ffff65b5110
  Type: function definition analyzed
  Visibility: externally_visible preempted_reg external public weak
  References:
  Referring:
  Read from file: a-weakdef.o
  Availability: available
  Unit id: 1
  Function flags: count:1073741824 (estimated locally)
  Called by: main/1 (1073741824 (estimated locally),1.00 per call)
  Calls:

Honza?  This looks like a bug in handling of PREEMTED_REG?  Or do we apply
some sort of ODR argument here?  We also see 'foo' as 'const' at IPA time
(but at local time we see it's interposable and thus do not analyze it)

[Bug lto/91299] [11/12/13/14 Regression] LTO inlines a weak definition in presence of a non-weak definition from an ELF file

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91299

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.5                        |11.5

--- Comment #16 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 10 branch is being closed.