[Bug target/97827] New: [11 Regression] bootstrap error building the amdgcn-amdhsa offload compiler

[Bug target/97827] New: [11 Regression] bootstrap error building the amdgcn-amdhsa offload compiler

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

            Bug ID: 97827
           Summary: [11 Regression] bootstrap error building the
                    amdgcn-amdhsa offload compiler
           Product: gcc
           Version: 10.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at debian dot org
  Target Milestone: ---

builds fine with trunk 20201112, fails with trunk 20201113. newlib 3.3.0 is
used.

[...]
/tmp/ccpzv56q.s:1129:2: error: changed section flags for .rodata.cst8,
expected: 0x12
        .section        .rodata.cst8
        ^
/tmp/ccpzv56q.s:1129:2: error: changed section entsize for .rodata.cst8,
expected: 8
        .section        .rodata.cst8
        ^
make[10]: *** [Makefile:980: lib_a-wcstod.o] Error 1
make[10]: *** Waiting for unfinished jobs....
/tmp/ccDmKs3n.s:4450:2: error: changed section flags for .rodata.cst8,
expected: 0x12
        .section        .rodata.cst8
        ^
/tmp/ccDmKs3n.s:4450:2: error: changed section entsize for .rodata.cst8,
expected: 8
        .section        .rodata.cst8
        ^
/tmp/ccDmKs3n.s:4643:2: error: changed section flags for .rodata.cst8,
expected: 0x12
        .section        .rodata.cst8
        ^
/tmp/ccDmKs3n.s:4643:2: error: changed section entsize for .rodata.cst8,
expected: 8
        .section        .rodata.cst8
        ^
make[10]: *** [Makefile:944: lib_a-strtod.o] Error 1
make[10]: Leaving directory
'/packages/gcc/11/gcc-11-11-20201113/build-gcn/amdgcn-amdhsa/gfx900/newlib/libc/stdlib'

[Bug target/97827] [11 Regression] bootstrap error building the amdgcn-amdhsa offload compiler

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |gcn
            Version|10.2.1                      |11.0
   Target Milestone|---                         |11.0

[Bug target/97827] [11 Regression] bootstrap error building the amdgcn-amdhsa offload compiler

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #1 from Matthias Klose <doko at debian dot org> ---
my bad, this is not a regression.  Seen when building the offload compiler with
LLVM 11 instead of LLVM 10 or 9

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ams at gcc dot gnu.org,
                   |                            |burnus at gcc dot gnu.org

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Changed in LLVM in https://reviews.llvm.org/D73999
cf. https://lists.llvm.org/pipermail/llvm-dev/2020-February/139390.html
and related Binutils discussion:
https://sourceware.org/legacy-ml/binutils/2020-02/msg00093.html

LLVM commit 75af9da755721123e62b45cd0bc0c5e688a9722a:
"   GNU as started to emit warnings for changed sh_type or sh_flags in 2000.
    GNU as>=2.35 will emit errors for most sh_type/sh_flags change, and error
for entsize change."

Cf. also http://web.mit.edu/rhel-doc/3/rhel-as-en-3/section.html



A grep shows:
        .section        .rodata.cst8
        .section        .rodata.cst8
        .section        .rodata.cst8,"aM",@progbits,8

Codewise, we have:
  mergeable_constant_section (mode=E_DFmode, align=64, flags=0)
which calls:
  884           flags |= (align / 8) | SECTION_MERGE;
  885           return get_section (name, flags, NULL);
with flags == 32776.

In the latter:
      flags |= SECTION_NAMED;
  → 2129928
and as "slot = section_htab->find_slot_with_hash" → *slot == NULL:
      sect = ggc_alloc<section> ();
      sect->named.common.flags = flags;
      sect->named.name = ggc_strdup (name);
      sect->named.decl = decl;

However, once the same function is called, we have:
  sect->common.flags == 3178504
while still/again:
  flags == 2129928

Next comes the call to:
  switch_to_section
which ends with
  new_section->common.flags |= SECTION_DECLARED;
which is
  3178504


If we look at default_elf_asm_named_section, we find:

  /* If we have already declared this section, we can use an
     abbreviated form to switch back to it -- unless this section is
     part of a COMDAT groups, in which case GAS requires the full
     declaration every time.  */
  if (!(HAVE_COMDAT_GROUP && (flags & SECTION_LINKONCE))
      && (flags & SECTION_DECLARED))
    { 
      fprintf (asm_out_file, "\t.section\t%s\n", name);
      return;
    }

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Now filled  https://bugs.llvm.org/show_bug.cgi?id=48201  for this

LLVM and GNU as ("gas") tried to align, cf. discussion
  https://sourceware.org/legacy-ml/binutils/2020-02/msg00091.html
and the GAS patch to turn a warning to an error:
  https://sourceware.org/legacy-ml/binutils/2020-02/msg00129.html


However, while llvm/lib/MC/MCParser/ELFAsmParser.cpp always issues an error:

+  if (Section->getType() != Type)
+    Error(loc, "changed section type for " + SectionName + ", expected: 0x" +
+                   utohexstr(Section->getType()));
etc.

The GNU assembler only does so if flags have been specified,
gas/config/obj-elf.c:

      if (attr != 0)
        { 
          /* If section attributes are specified the second time we see a
             particular section, then check that they are the same as we
             saw the first time.  */
          if (((old_sec->flags ^ flags)
...

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Draft patch (untested):

diff --git a/gcc/varasm.c b/gcc/varasm.c
index ada99940f65..51a507393a8 100644
--- a/gcc/varasm.c
+++ b/gcc/varasm.c
@@ -6738,9 +6738,11 @@ default_elf_asm_named_section (const char *name,
unsigned int flags,
   /* If we have already declared this section, we can use an
      abbreviated form to switch back to it -- unless this section is
      part of a COMDAT groups, in which case GAS requires the full
-     declaration every time.  */
+     declaration every time.  LLVM's MC linker requires that the
+     flags are identical, thus avoid the abbreviated form with MERGE.  */
   if (!(HAVE_COMDAT_GROUP && (flags & SECTION_LINKONCE))
-      && (flags & SECTION_DECLARED))
+      && (flags & SECTION_DECLARED)
+      && !(flags & SECTION_MERGE))
     {
       fprintf (asm_out_file, "\t.section\t%s\n", name);
       return;

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Patch: https://gcc.gnu.org/pipermail/gcc-patches/2020-November/559487.html

First review regards this as LLVM bug.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Might not be a LLVM bug if llvm-as doesn't pretend to be assembler compatible
with other assemblers.  But in that case we should treat it as a target with
specific assembler requirements (e.g. like Darwin).

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Submitted LLVM patch at https://reviews.llvm.org/D92052
If it gets accepted for LLVM + backported to 11, we are done.

Otherwise, we have to proceed as suggested in the email thread.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |WONTFIX

--- Comment #8 from Tobias Burnus <burnus at gcc dot gnu.org> ---
FIXED in LLVM 'main' in commit 1deff4009e0ae661b03682901bf6932297ce7ea1
→ https://reviews.llvm.org/D92052 + https://bugs.llvm.org/show_bug.cgi?id=48201

Close as WONTFIX on the GCC side.

I try to get it backported to LLVM 11.0.1 or, failing that, LLVM 11.0.2.

Thanks Matthias for pointing out the issue.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[i at maskray dot me]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Fangrui Song <i at maskray dot me> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |i at maskray dot me

--- Comment #9 from Fangrui Song <i at maskray dot me> ---
I want to know whether this is really a wontfix on GCC's side.

Richard Sandiford on
https://gcc.gnu.org/pipermail/gcc-patches/2020-November/559572.html

"I'm not saying we should bend over backwards to support difficult
quirks.  But here we're talking about a choice between (a) doing
something that works “everywhere” unconditionally (and keeping things
simple) vs. (b) having both code that takes a shortcut and code that
doesn't take a shortcut and trying to predict which one we should do."

This makes a lot of sense to me. For the LLVM "fix", we had not known this PR
before https://reviews.llvm.org/D92052#2452577
To me personally, I might have a different opinion if I knew this is not an
entire dead end on gcc -S output.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[i at maskray dot me]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #10 from Fangrui Song <i at maskray dot me> ---
Note: the section key is not just (name, group name "G"). It is a quadruple:
(name, group name "G", linked-to "o", unique ID)

Keeping just name works for the simplest case.

If GCC decides to support PR95095  -fno-unique-section-names, unique ID can be
common. https://sourceware.org/bugzilla/show_bug.cgi?id=25490#c3 added the
support for `.section ....,unique` to GNU as.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #11 from Tobias Burnus <burnus at gcc dot gnu.org> ---
For completeness, the LLVM 'main' patch was backported/cherry-picked for
LLVM 11.0.1 with commit 700baa009dc685a0adc5f94d258be4ae24742471

Regarding the .section discussion, see also last few comments in
https://reviews.llvm.org/D92052

Thanks again for the bug report report, Matthias – and to those providing
background information and helping to reach a good solution :-)

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #12 from Matthias Klose <doko at debian dot org> ---
Fyi, this is still seen with LLVM 11.0.1 rc2

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
     Ever confirmed|0                           |1
         Resolution|WONTFIX                     |---
   Last reconfirmed|                            |2020-12-27

--- Comment #13 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Matthias Klose from comment #12)
> Fyi, this is still seen with LLVM 11.0.1 rc2

But the fix (commit 700baa009dc685a0adc5f94d258be4ae24742471) is included in
llvmorg-11.0.1-rc2.

If the issue is still present, something else goes wrong. What's the error
message? Is it really identical to the one of comment 0?

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #14 from Matthias Klose <doko at debian dot org> ---
[...]
/tmp/ccQFzQVM.s:1082:2: error: changed section flags for .rodata.cst8,
expected:
0x12
        .section        .rodata.cst8
        ^
/tmp/ccQFzQVM.s:1082:2: error: changed section entsize for .rodata.cst8,
expected: 8
        .section        .rodata.cst8
        ^
make[10]: *** [Makefile:980: lib_a-wcstod.o] Error 1

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #15 from Tobias Burnus <burnus at gcc dot gnu.org> ---
I unfortunately missed in my the LLVM patch that '.rodata' implies flags and
messed up the check. Should by fixed by: https://reviews.llvm.org/D94072

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|REOPENED                    |RESOLVED

--- Comment #16 from Tobias Burnus <burnus at gcc dot gnu.org> ---
(In reply to Tobias Burnus from comment #15)
> I unfortunately missed in my the LLVM patch that '.rodata' implies flags and
> messed up the check. Should by fixed by: https://reviews.llvm.org/D94072

Now merged to LLVM 12/trunk (see ↑ or next link); for LLVM 11 backporting, see:
https://bugs.llvm.org/show_bug.cgi?id=48922

For related Debian bug, see
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=975692

Close as FIXED – hopefully, now for real.

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[ams at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

Andrew Stubbs <ams at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xw111luoye at gmail dot com

--- Comment #17 from Andrew Stubbs <ams at gcc dot gnu.org> ---
*** Bug 95023 has been marked as a duplicate of this bug. ***

[Bug target/97827] bootstrap error building the amdgcn-amdhsa offload compiler with LLVM 11

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97827

--- Comment #18 from Tobias Burnus <burnus at gcc dot gnu.org> ---
POST SCRIPT / STATUS UPDATE (This PR remains FIXED)

(In reply to Tobias Burnus from comment #16)
> > I unfortunately missed in my the LLVM patch that '.rodata' implies flags and
> > messed up the check. Should by fixed by: https://reviews.llvm.org/D94072
> 
> Now merged to LLVM 12/trunk (see ↑ or next link); for LLVM 11 backporting,
> see: https://bugs.llvm.org/show_bug.cgi?id=48922

For completeness:

It turned out that it missed the LLVM 12 branch point - such that is is only in
LLVM 13+ and not LLVM 12. (The LLVM 11 request was also ignored.)

 * * *

Additionally, a very hard to hit corner case was missed that became easier to
trigger due to an LLVM 13 change. - That one is still rare to hit (and does not
prevent bootstrapping).
It was fixed in LLVM 'main' and backported to LLVM releases/13.x (just after
13.0.1-RC1); thus, it will be in RC2 (if any) and 13.0.1.

Cf. https://reviews.llvm.org/D114717
main:
 
https://github.com/llvm/llvm-project/commit/c01c62c76c60a5a5da0496e41faae907944c92dd
LLVM 13.x backport (for 13.0.1):
  https://github.com/llvm/llvm-project/issues/52607
 
https://github.com/llvm/llvm-project/commit/5932c004778cf251302db0d46d1dfb247325ed3f

Thus: FIXED (only) in LLVM 13 - but best to use at least LLVM release/13.x or
13.0.1(-rc2).

* * *

GCC 12 also has several changes (mostly: configure checks to detect and act on
the LLVM version) - to be able to use LLVM 13; several were backported to GCC
11 (releases/gcc-11).

* * *

Crossref: PR 100181 (-> PR 100181 comment 10) is another GCN-related issue,
which was solved in LLVM 13