GCC Bugzilla Bug 16876

On AMD64, PPC, PPC64, i386: 
 
gcc -O3 -c parse_opts.i 
parse_opts.i: In function `f': 
parse_opts.i:11: internal compiler error: in gen_lowpart, at emit-rtl.c:1436 
 
The code is invalid (a bit), but it should not crash. 
 
reproduced with mainline from some days ago, 3.3-hammer branch, 3.4 branch 
too.

Created an attachment (id=6881) [edit]
parse_opts.i

compile with -O3 or -finline-functions

Confirmed, the code is undefined.  With the mainline we get a different ICE:
pr16876.c: In function `f':
pr16876.c:8: internal compiler error: in fold_convert, at fold-const.c:2000
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.

But this is because of changes by me and other people.

Really this is a regression from 2.95.3 and 3.0.4 but that is because of the use of the RTL inliner with 
those versions instead of the tree inline.

: Search converges between 2001-11-11-trunk (#45) and 2001-11-18-trunk (#46).

The problem in mainline is that 'tree type' of RECORD_TYPE enters
fold_convert()
and triggers gcc_unreachable() in line 2003.

However Borland C++ exits with error: "Call to function 'g' with no prototype."
on line where g(0) is called.

*** Bug 20141 has been marked as a duplicate of this bug. ***

*** Bug 20680 has been marked as a duplicate of this bug. ***

See also my notes <http://gcc.gnu.org/ml/gcc-patches/2004-10/msg00772.html> on
what should receive warnings (possibly compiled into aborts), what should
receive errors and how c_convert_parm_for_inlining (a) is too late for such
diagnostics and (b) fails actually to carry out the conversion or diagnostics
in certain cases.

*** Bug 24210 has been marked as a duplicate of this bug. ***

P2 is the right setting for this bug; it's popping up often enough to make it
worth fixing.  Disabling the inlining in the case that the argument type
doesn't match (closely enough) the parameter type seems reasonable to me, even
without introducing a subsequent runtime abort.  That would be equivalent to
the behavior if the function called appeared in a different translation unit.

rth assigned this to himself:
http://gcc.gnu.org/ml/gcc-bugs/2005-11/msg02843.html

A progress report would be nice ;-)

This issue will not be resolved in GCC 4.1.0; retargeted at GCC 4.1.1.

Will not be fixed in 4.1.1; adjust target milestone to 4.1.2.

Unassigning rth, since he's obviously not actually interested in fixing this.

The failures in 3.4 and later are in fold_const, so the gen_lowpart problem is
now avoided by, well, ICEing earlier :)

Created an attachment (id=12130) [edit]
patch

I think this should do the trick.

Subject: Bug 16876

Author: amylaar
Date: Fri Aug 25 18:51:57 2006
New Revision: 116424

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=116424
Log:
gcc:
        PR tree-optimization/16876
        * c-typeck.c (c_convert_parm_for_inlining): Don't take early
        exit if PARM doesn't match VALUE.

testsuite:
        gcc.dg/noncompile/pr16876.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/noncompile/pr16876.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/c-typeck.c
    trunk/gcc/testsuite/ChangeLog

A backport of the patch bootstraps & regtests ok on the 4.1 branch, but I'm
unsure we want the side-effects of the rejects-valid, because this is valid
code according to jsm (only undefined at runtime).

(In reply to comment #17)
> A backport of the patch bootstraps & regtests ok on the 4.1 branch.
Except it introduces PR 29478.

Subject: Bug 16876

Author: rguenth
Date: Sat Jun 23 18:17:57 2007
New Revision: 125974

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=125974
Log:
2007-06-23  Richard Guenther  <rguenther@suse.de>

        PR tree-optimization/16876
        PR middle-end/29478
        * tree.h (CALL_CANNOT_INLINE_P): New macro to access static_flag
        for CALL_EXPRs.
        * tree-inline.c (initialize_inlined_parameters): Do not call
        lang_hooks.tree_inlining.convert_parm_for_inlining.
        * cgraphbuild.c (initialize_inline_failed): Set inline failed
        reason for mismatched types.
        * gimplify.c (gimplify_call_expr): Verify the call expression
        arguments match the called function type signature.  Otherwise
        mark the call expression to be not considered for inlining
        using CALL_CANNOT_INLINE_P flag.
        * ipa-inline.c (cgraph_mark_inline): Honor CALL_CANNOT_INLINE_P on the
        edges call expression.
        (cgraph_decide_inlining_of_small_function): Likewise.
        (cgraph_decide_inlining): Likewise.
        * c-objc-common.h (LANG_HOOKS_TREE_INLINING_CONVERT_PARM_FOR_INLINING):
        Remove define.
        * c-tree.h (c_convert_parm_for_inlining): Remove declaration.
        * c-typeck.c (c_convert_parm_for_inlining): Remove.
        * langhooks-def.h (lhd_tree_inlining_convert_parm_for_inlining):
        Remove declaration.
        (LANG_HOOKS_TREE_INLINING_CONVERT_PARM_FOR_INLINING): Remove define.
        * langhooks.c (lhd_tree_inlining_convert_parm_for_inlining):
        Remove.
        * langhooks.h (struct lang_hooks_for_tree_inlining): Remove
        convert_parm_for_inlining member.

        * gcc.dg/pr29254.c: The warning is bogus.
        * gcc.dg/warn-1.c: Likewise.
        * gcc.dg/assign-warn-3.c: Likewise.
        * gcc.dg/noncompile/pr16876.c: The testcase is bogus, remove.

Removed:
    trunk/gcc/testsuite/gcc.dg/noncompile/pr16876.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/c-objc-common.h
    trunk/gcc/c-tree.h
    trunk/gcc/c-typeck.c
    trunk/gcc/cgraphbuild.c
    trunk/gcc/gimplify.c
    trunk/gcc/ipa-inline.c
    trunk/gcc/langhooks-def.h
    trunk/gcc/langhooks.c
    trunk/gcc/langhooks.h
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.dg/assign-warn-3.c
    trunk/gcc/testsuite/gcc.dg/pr29254.c
    trunk/gcc/testsuite/gcc.dg/warn-1.c
    trunk/gcc/tree-inline.c
    trunk/gcc/tree.h

As the code is deemed valid, we still have a rejects-valid bug on the 4.2
branch.

Change target milestone to 4.2.3, as 4.2.2 has been released.

4.2.3 is being released now, changing milestones of open bugs to 4.2.4.

4.2.4 is being released, changing milestones to 4.2.5.

Closing 4.1 branch.

Closing 4.2 branch.  Original bug was fixed for 4.2.0, rejects-valid
for 4.3.0.