Bug 108582 - [12 Regression] ICE on valid code at -Os and above with "-fno-tree-ccp -fno-tree-dce": tree check: expected class ‘type’, have ‘exceptional’ (error_mark) in useless_type_conversion_p, at gimple-expr.cc:87
Summary: [12 Regression] ICE on valid code at -Os and above with "-fno-tree-ccp -fno-t...
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: tree-optimization (show other bugs)
Version: 13.0
: P2 normal
Target Milestone: 12.3
Assignee: Andrew Pinski
URL:
Keywords: ice-on-valid-code, wrong-code
Depends on:
Blocks:
 
Reported: 2023-01-28 13:48 UTC by Zhendong Su
Modified: 2023-02-07 20:22 UTC (History)
0 users

See Also:
Host:
Target:
Build:
Known to work: 11.1.0, 13.0
Known to fail: 12.2.0
Last reconfirmed: 2023-01-28 00:00:00

Attachments
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 Zhendong Su 2023-01-28 13:48:17 UTC 
It appears to be a recent regression from 12.2. 

Compiler Explorer: https://godbolt.org/z/91oGh1W8K


[617] % gcctk -v
Using built-in specs.
COLLECT_GCC=gcctk
COLLECT_LTO_WRAPPER=/local/suz-local/software/local/gcc-trunk/libexec/gcc/x86_64-pc-linux-gnu/13.0.1/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../gcc-trunk/configure --disable-bootstrap --enable-checking=yes --prefix=/local/suz-local/software/local/gcc-trunk --enable-sanitizers --enable-languages=c,c++ --disable-werror --disable-multilib
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 13.0.1 20230128 (experimental) [master r12-4647-g3f861a5c8fd] (GCC)
[618] %
[618] % gcctk -O3 small.c; ./a.out
[619] %
[619] % gcctk -O3 -fno-tree-ccp -fno-tree-dce small.c
during GIMPLE pass: phiopt
small.c: In function ‘main’:
small.c:3:5: internal compiler error: tree check: expected class ‘type’, have ‘exceptional’ (error_mark) in useless_type_conversion_p, at gimple-expr.cc:87
    3 | int main() {
      |     ^~~~
0x6e9981 tree_class_check_failed(tree_node const*, tree_code_class, char const*, int, char const*)
        ../../gcc-trunk/gcc/tree.cc:8959
0xbad6e3 tree_class_check(tree_node*, tree_code_class, char const*, int, char const*)
        ../../gcc-trunk/gcc/tree.h:3653
0xbad6e3 useless_type_conversion_p(tree_node*, tree_node*)
        ../../gcc-trunk/gcc/gimple-expr.cc:87
0xfb106c verify_gimple_assign_binary
        ../../gcc-trunk/gcc/tree-cfg.cc:4292
0xfc1897 verify_gimple_assign
        ../../gcc-trunk/gcc/tree-cfg.cc:4829
0xfc1897 verify_gimple_stmt
        ../../gcc-trunk/gcc/tree-cfg.cc:5088
0xfc9448 verify_gimple_in_cfg(function*, bool, bool)
        ../../gcc-trunk/gcc/tree-cfg.cc:5583
0xe5a333 execute_function_todo
        ../../gcc-trunk/gcc/passes.cc:2091
0xe5adb2 execute_todo
        ../../gcc-trunk/gcc/passes.cc:2145
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.
[620] %
[620] % cat small.c
int a, c, d, e, f;
char b;
int main() {
  int g = 1;
  char h[1] = {0};
  while (a) {
    if (f) {
      b = 0;
      if (d)
        continue;
    }
    if (a < 1) {
      g = 0;
      goto L;
    }
  }
  while (c) {
    char *j = h;
    int k;
  L:
    if (e && !g)
      k |= 2 | (*j < 0);
  }
  return 0;
}
Comment 1 Andrew Pinski 2023-01-28 18:10:08 UTC 
#5  0x000000000162e9cc in verify_gimple_assign (stmt=0x7ffff741c2c0) at /home/apinski/src/upstream-gcc/gcc/gcc/tree-cfg.cc:4829
4829          return verify_gimple_assign_binary (stmt);
(gdb) p debug_gimple_stmt(stmt)
_4 = _13 & _28;

#4  0x000000000162c6b7 in verify_gimple_assign_binary (stmt=0x7ffff741c2c0) at /home/apinski/src/upstream-gcc/gcc/gcc/tree-cfg.cc:4292
4292      if (!useless_type_conversion_p (lhs_type, rhs1_type)
(gdb) p debug_generic_expr(lhs_type)
int
$2 = void
(gdb) p debug_generic_expr(rhs1_type)
<<< error >>>
$3 = void
Comment 2 Andrew Pinski 2023-01-28 18:12:04 UTC 
Mine I think ...
Comment 3 Andrew Pinski 2023-01-28 18:12:55 UTC 
PHI g_9 reduced for COND_EXPR in block 7 to g_8.
statement un-sinked:
k_22 = k_11 | iftmp.5_13;
Removing basic block 8
;; basic block 8, loop depth 0
;;  pred:       7
# iftmp.5_13 = PHI <2(7)>
;;  succ:       9

Looks like match-and-simplify phiopt does not take into account PHInodes ...
Comment 4 Andrew Pinski 2023-01-28 18:17:28 UTC 
This should fix the issue:
apinski@xeond:~/src/upstream-gcc/gcc/gcc$ git diff tree-ssa-phiopt.cc
diff --git a/gcc/tree-ssa-phiopt.cc b/gcc/tree-ssa-phiopt.cc
index c3a889dc593..a7ab6ce4ad9 100644
--- a/gcc/tree-ssa-phiopt.cc
+++ b/gcc/tree-ssa-phiopt.cc
@@ -1002,6 +1002,11 @@ match_simplify_replacement (basic_block cond_bb, basic_block middle_bb,
       if (!single_pred_p (middle_bb))
        return false;

+      /* The middle bb cannot have phi nodes as we don't
+        move those assignments yet. */
+      if (!gimple_seq_empty_p (phi_nodes (middle_bb)))
+       return false;
+
       stmt_to_move = last_and_only_stmt (middle_bb);
       if (!stmt_to_move)
        return false;
Comment 5 Andrew Pinski 2023-01-28 18:50:47 UTC 
Note this is a latent bug since it was introduced by r12-1309-gc4574d23cb0734 .
I don't know what changed that cause the IR to be different on the trunk but I just know it is a latent bug.
Comment 6 Andrew Pinski 2023-01-28 22:25:14 UTC 
Patch submitted:
https://gcc.gnu.org/pipermail/gcc-patches/2023-January/610817.html
Comment 7 GCC Commits 2023-01-30 12:46:08 UTC 
The trunk branch has been updated by Andrew Pinski <pinskia@gcc.gnu.org>:

https://gcc.gnu.org/g:876b3e0514bc8cb2256c44db56255403bedfa52d

commit r13-5493-g876b3e0514bc8cb2256c44db56255403bedfa52d
Author: Andrew Pinski <apinski@marvell.com>
Date:   Sat Jan 28 18:27:08 2023 +0000

    Fix PR 108582: ICE due to PHI-OPT removing a still in use ssa_name.
    
    This patch adds a check in match_simplify_replacement to make sure the middlebb
    does not have any phi-nodes as we don't currently move those.
    This was just a thinko from before.
    
    Ok? Bootstrapped and tested on x86_64-linux-gnu with no regressions?
    
            PR tree-optimization/108582
    
    gcc/ChangeLog:
    
            * tree-ssa-phiopt.cc (match_simplify_replacement): Add check
            for middlebb to have no phi nodes.
    
    gcc/testsuite/ChangeLog:
    
            * gcc.dg/pr108582-1.c: New test.
Comment 8 Andrew Pinski 2023-01-30 12:47:12 UTC 
Fixed on the trunk, still latent in GCC 12, will be backporting the patch next week.
Comment 9 GCC Commits 2023-02-07 20:21:53 UTC 
The releases/gcc-12 branch has been updated by Andrew Pinski <pinskia@gcc.gnu.org>:

https://gcc.gnu.org/g:6f23c9077feebb29c2a28ffe89b287286df27d6d

commit r12-9114-g6f23c9077feebb29c2a28ffe89b287286df27d6d
Author: Andrew Pinski <apinski@marvell.com>
Date:   Sat Jan 28 18:27:08 2023 +0000

    Fix PR 108582: ICE due to PHI-OPT removing a still in use ssa_name.
    
    This patch adds a check in match_simplify_replacement to make sure the middlebb
    does not have any phi-nodes as we don't currently move those.
    This was just a thinko from before.
    
    Committed on the GCC 12 branch after a bootstrap/test on x86_64-linux-gnu.
    
            PR tree-optimization/108582
    
    gcc/ChangeLog:
    
            * tree-ssa-phiopt.cc (match_simplify_replacement): Add check
            for middlebb to have no phi nodes.
    
    gcc/testsuite/ChangeLog:
    
            * gcc.dg/pr108582-1.c: New test.
    
    (cherry picked from commit 876b3e0514bc8cb2256c44db56255403bedfa52d)
Comment 10 Andrew Pinski 2023-02-07 20:22:28 UTC 
Fixed.