Bug 105263 - [10 Regression] ICE: gimple check: expected gimple_assign(error_mark), have gimple_nop() in gimple_assign_rhs1, at gimple.h:2655 with _Decimal64 -ffast-math since r7-950-g8a85cee26eabf5cf
Summary: [10 Regression] ICE: gimple check: expected gimple_assign(error_mark), have g...
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: middle-end (show other bugs)
Version: 12.0
: P2 normal
Target Milestone: 11.4
Assignee: Richard Biener
URL:
Keywords: ice-on-valid-code
Depends on:
Blocks:
 
Reported: 2022-04-13 12:18 UTC by Zdenek Sojka
Modified: 2023-07-07 09:59 UTC (History)
1 user (show)

See Also:
Host: x86_64-pc-linux-gnu
Target:
Build:
Known to work: 11.3.1, 12.0, 5.5.0, 6.5.0
Known to fail: 10.3.1, 10.5.0, 11.3.0, 12.0, 7.5.0, 8.5.0, 9.4.1
Last reconfirmed: 2022-04-13 00:00:00

Attachments
reduced testcase (128 bytes, text/plain)
2022-04-13 12:18 UTC, Zdenek Sojka 		Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Zdenek Sojka 2022-04-13 12:18:15 UTC 
Created attachment 52800 [details]
reduced testcase

Compiler output:
$ x86_64-pc-linux-gnu-gcc -O -ffast-math testcase.c 
during GIMPLE pass: reassoc
testcase.c: In function 'foo':
testcase.c:7:1: internal compiler error: gimple check: expected gimple_assign(error_mark), have gimple_nop() in gimple_assign_rhs1, at gimple.h:2655
    7 | foo (U u)
      | ^~~
0x6e6c4d gimple_check_failed(gimple const*, char const*, int, char const*, gimple_code, tree_code)
        /repo/gcc-trunk/gcc/gimple.cc:1292
0x7d830d GIMPLE_CHECK2<const gassign*>
        /repo/gcc-trunk/gcc/gimple.h:74
0x7d833d GIMPLE_CHECK2<const gassign*>
        /repo/gcc-trunk/gcc/gimple.h:2653
0x7d833d gimple_assign_rhs1
        /repo/gcc-trunk/gcc/gimple.h:2655
0x7d886f gimple_assign_rhs1
        /repo/gcc-trunk/gcc/tree.h:3571
0x7d886f rewrite_expr_tree
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:5248
0x155514e rewrite_expr_tree
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:5339
0x155514e rewrite_expr_tree
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:5339
0x15665fa reassociate_bb
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:6848
0x156593a reassociate_bb
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:6897
0x1567ede do_reassoc
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:7009
0x1567ede execute_reassoc
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:7098
0x1567ede execute
        /repo/gcc-trunk/gcc/tree-ssa-reassoc.cc:7139
Please submit a full bug report, with preprocessed source (by using -freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

$ x86_64-pc-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk/binary-latest-amd64/bin/x86_64-pc-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-r12-8113-20220412190241-gaa7874596b9-checking-yes-rtl-df-extra-amd64/bin/../libexec/gcc/x86_64-pc-linux-gnu/12.0.1/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++ --enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra --with-cloog --with-ppl --with-isl --build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu --target=x86_64-pc-linux-gnu --with-ld=/usr/bin/x86_64-pc-linux-gnu-ld --with-as=/usr/bin/x86_64-pc-linux-gnu-as --disable-libstdcxx-pch --prefix=/repo/gcc-trunk//binary-trunk-r12-8113-20220412190241-gaa7874596b9-checking-yes-rtl-df-extra-amd64
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.0.1 20220412 (experimental) (GCC)
Comment 1 Martin Liška 2022-04-13 12:44:40 UTC 
Started with r7-950-g8a85cee26eabf5cf.
Comment 2 Richard Biener 2022-04-13 12:52:00 UTC 
Confirmed.

5336      /* Recurse on the LHS of the binary operator, which is guaranteed to
5337         be the non-leaf side.  */
5338      tree new_rhs1
5339        = rewrite_expr_tree (SSA_NAME_DEF_STMT (rhs1), rhs_code, opindex + 1, ops,
5340                             changed || oe->op != rhs2 || next_changed,
5341                             false);

but in this case it's the leaf side...

(gdb) p debug_gimple_stmt (stmt)
u_3 = u_2(D) * u_2(D);

(gdb) p debug_generic_expr (ops.m_vec->m_vecdata[0]->op)
u_2(D)
$11 = void
(gdb) p debug_generic_expr (ops.m_vec->m_vecdata[1]->op)
u_2(D)
$12 = void
(gdb) p debug_generic_expr (ops.m_vec->m_vecdata[2]->op)
_5
$13 = void
(gdb) p debug_generic_expr (ops.m_vec->m_vecdata[3]->op)
{ -1.0e+0 }

U foo (U u)
{
  _Decimal64 _1;
  U _5;
  vector(1) _Decimal64 _6;

  <bb 2> [local count: 1073741824]:
  u_3 = u_2(D) * u_2(D);
  _1 = BIT_FIELD_REF <v(address-taken), 64, 64>;
  _5 = {_1};
  _6 = -_5;
  u_7 = u_3 * _6;
  <retval> = u_7;
  return <retval>;

so basically the multiplication by -1 trick to handle the negation
breaks the left associativity rewrite because we need to go too deep.

Tricky.

Supposedly we have to pop the -1 before rewriting the expression
but we fail.  We fail because build_minus_one_cst doesn't build something
real_minus_onep, in this case the VECTOR_CST part, recognizes.  And
that's because of

3156    bool
3157    real_minus_onep (const_tree expr)
...
3164          return real_equal (&TREE_REAL_CST (expr), &dconstm1)
3165                 && !(DECIMAL_FLOAT_MODE_P (TYPE_MODE (TREE_TYPE (expr))));

means we have to exclude _Decimal modes from this trick.
Comment 3 GCC Commits 2022-04-13 13:59:35 UTC 
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:ca145c6306f19272ac8756d88c4eba0bfdf01dfb

commit r12-8142-gca145c6306f19272ac8756d88c4eba0bfdf01dfb
Author: Richard Biener <rguenther@suse.de>
Date:   Wed Apr 13 14:53:40 2022 +0200

    tree-optimization/105263 - reassoc and DFP
    
    reassoc has certain tricks which in the end depend on the ability
    to undo them.  For DFP creating a -1. constant is easy but
    re-identifying is appearantly not - real_minus_onep rejects those
    outright for DFP.  So we have to disable (at least) this one trick.
    
    2022-04-13  Richard Biener  <rguenther@suse.de>
    
            PR tree-optimization/105263
            * tree-ssa-reassoc.cc (try_special_add_to_ops): Do not consume
            negates in multiplication chains with DFP.
    
            * gcc.dg/pr105263.c: New testcase.
Comment 4 Richard Biener 2022-04-13 14:00:03 UTC 
Fixed on trunk sofar.
Comment 5 GCC Commits 2022-05-19 14:03:21 UTC 
The releases/gcc-11 branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:b2557f9081e5bf440abe3fc184cbdaea0447ec64

commit r11-10017-gb2557f9081e5bf440abe3fc184cbdaea0447ec64
Author: Richard Biener <rguenther@suse.de>
Date:   Wed Apr 13 14:53:40 2022 +0200

    tree-optimization/105263 - reassoc and DFP
    
    reassoc has certain tricks which in the end depend on the ability
    to undo them.  For DFP creating a -1. constant is easy but
    re-identifying is appearantly not - real_minus_onep rejects those
    outright for DFP.  So we have to disable (at least) this one trick.
    
    2022-04-13  Richard Biener  <rguenther@suse.de>
    
            PR tree-optimization/105263
            * tree-ssa-reassoc.c (try_special_add_to_ops): Do not consume
            negates in multiplication chains with DFP.
    
            * gcc.dg/pr105263.c: New testcase.
    
    (cherry picked from commit ca145c6306f19272ac8756d88c4eba0bfdf01dfb)
Comment 6 Richard Biener 2022-05-27 09:48:01 UTC 
GCC 9 branch is being closed
Comment 7 Jakub Jelinek 2022-06-28 10:49:04 UTC 
GCC 10.4 is being released, retargeting bugs to GCC 10.5.
Comment 8 Richard Biener 2023-07-07 09:59:10 UTC 
Fixed in 11.4.