[Bug target/98557] New: [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[Bug target/98557] New: [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

            Bug ID: 98557
           Summary: [11 Regression] ICE in insert_operand_rank, at
                    tree-ssa-reassoc.c:367 on arm-linux-gnueabihf
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at debian dot org
  Target Milestone: ---

Created attachment 49896
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49896&action=edit
preprocessed source

seen with trunk 20210102 on arm-linux-gnueabihf

// during GIMPLE pass: reassoc
// gen/wxe_funcs.cpp: In member function 'void
WxeApp::wxe_dispatch(wxeCommand&)':
// gen/wxe_funcs.cpp:40:6: internal compiler error: in insert_operand_rank, at
tree-ssa-reassoc.c:367
//    40 | void WxeApp::wxe_dispatch(wxeCommand& Ecmd)
//       |      ^~~~~~
// 0xb4bea3 insert_operand_rank
//      ../../src/gcc/tree-ssa-reassoc.c:367
// 0xb4bea3 insert_operand_rank
//      ../../src/gcc/tree-ssa-reassoc.c:365
// 0xb4bea3 get_rank
//      ../../src/gcc/tree-ssa-reassoc.c:455
// 0xb4ba7b add_to_ops_vec
//      ../../src/gcc/tree-ssa-reassoc.c:569
// 0xb48bd3 reassociate_bb
//      ../../src/gcc/tree-ssa-reassoc.c:6492
// 0xb48d53 reassociate_bb
//      ../../src/gcc/tree-ssa-reassoc.c:6656
// 0xb48d53 reassociate_bb
//      ../../src/gcc/tree-ssa-reassoc.c:6656
// 0xb48e1b reassociate_bb
//      ../../src/gcc/tree-ssa-reassoc.c:6656
// 0xb47d93 do_reassoc
//      ../../src/gcc/tree-ssa-reassoc.c:6768
// 0xb47d93 execute_reassoc
//      ../../src/gcc/tree-ssa-reassoc.c:6855
// 0xb47d93 execute
//      ../../src/gcc/tree-ssa-reassoc.c:6894
// Please submit a full bug report,
// with preprocessed source if appropriate.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Can't reproduce with a cross compiler :/

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

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

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

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Dup of PR98514 ?

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Martin Liška from comment #1)
> Can't reproduce with a cross compiler :/

If it is a dup, you wouldn't reproduce that with a cross, unless it was a cross
from 32-bit host.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #4 from Matthias Klose <doko at debian dot org> ---
I'll recheck with an updated build from trunk

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.0
             Status|UNCONFIRMED                 |WAITING
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-01-07

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #5 from Matthias Klose <doko at debian dot org> ---
fyi, creduce segfaults on arm-linux-gnueabihf trying to reduce this:

$ creduce ./cr2.sh wxe_funcs.ii
===< 26689 >===
running 4 interestingness tests in parallel
===< pass_unifdef :: 0 >===
===< pass_comments :: 0 >===
(0.1 %, 5992612 bytes)
(0.1 %, 5989171 bytes)
===< pass_ifs :: 0 >===
===< pass_includes :: 0 >===
===< pass_line_markers :: 0 >===
(5.2 %, 5683057 bytes)
===< pass_blank :: 0 >===
(6.0 %, 5637660 bytes)
===< pass_clang_binsrch :: replace-function-def-with-decl >===
clang_delta: /usr/lib/llvm-11/include/llvm/ADT/PointerIntPair.h:178: static
intptr_t llvm::PointerIntPairInfo<PointerT, IntBits,
PtrTraits>::updatePointer(intptr_t, PointerT) [with PointerT = clang::Stmt*;
unsigned int IntBits = 1; PtrTraits =
llvm::PointerLikeTypeTraits<clang::Stmt*>; intptr_t = int]: Assertion `(PtrWord
& ~PointerBitMask) == 0 && "Pointer is not sufficiently aligned"' failed.
Aborted (core dumped)
Use of uninitialized value $line in pattern match (m//) at
/usr/share/creduce/perl/pass_clang_binsrch.pm line 36.
===< pass_clang_binsrch :: remove-unused-function >===
clang_delta: /usr/lib/llvm-11/include/llvm/ADT/PointerIntPair.h:178: static
intptr_t llvm::PointerIntPairInfo<PointerT, IntBits,
PtrTraits>::updatePointer(intptr_t, PointerT) [with PointerT = clang::Stmt*;
unsigned int IntBits = 1; PtrTraits =
llvm::PointerLikeTypeTraits<clang::Stmt*>; intptr_t = int]: Assertion `(PtrWord
& ~PointerBitMask) == 0 && "Pointer is not sufficiently aligned"' failed.
Aborted (core dumped)
Use of uninitialized value $line in pattern match (m//) at
/usr/share/creduce/perl/pass_clang_binsrch.pm line 36.
===< pass_lines :: 0 >===
(5.9 %, 5642847 bytes)
[...]

cvise doesn't show any segfaults, but is skipping these two passes as well:

00:04:32 INFO ===< BlankPass >===
00:05:52 INFO (5.9%, 5645707 bytes, 113225 lines)
00:07:15 DEBUG Creating pass root folder: /tmp/cvise-6w2dcu0a
00:07:15 INFO ===< ClangBinarySearchPass::replace-function-def-with-decl >===
00:07:23 DEBUG available transformation opportunities for c++98: 0, took: 8.22
s
00:07:33 DEBUG available transformation opportunities for c++11: 0, took: 9.38
s
00:07:41 DEBUG available transformation opportunities for c++14: 0, took: 7.99
s
00:07:50 DEBUG available transformation opportunities for c++17: 0, took: 9.73
s
00:08:02 DEBUG available transformation opportunities for c++20: 0, took: 11.24
s
00:08:02 INFO using C++ standard: c++98 with 0 transformation opportunities
00:08:12 DEBUG Creating pass root folder: /tmp/cvise-zq6tn5o2
00:08:12 INFO ===< ClangBinarySearchPass::remove-unused-function >===
00:08:20 DEBUG available transformation opportunities for c++98: 0, took: 8.21
s
00:08:31 DEBUG available transformation opportunities for c++11: 0, took: 10.69
s
00:08:46 DEBUG available transformation opportunities for c++14: 0, took: 15.39
s
00:08:57 DEBUG available transformation opportunities for c++17: 0, took: 10.66
s
00:09:05 DEBUG available transformation opportunities for c++20: 0, took: 7.99
s
00:09:05 INFO using C++ standard: c++98 with 0 transformation opportunities
00:09:13 DEBUG Creating pass root folder: /tmp/cvise-ftvo6ifx
00:09:13 INFO ===< LinesPass::0 >===

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #6 from Matthias Klose <doko at debian dot org> ---
Created attachment 49907
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=49907&action=edit
preprocessed source

attaching the 50% reduced file, still showing the clang_delta issue.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |marxin at gcc dot gnu.org

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
I bet it's fixed (as PR98514 is fixed), I'll verify that later this week.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #8 from Martin Liška <marxin at gcc dot gnu.org> ---
Note that C-Vise now reports similarly to C-Reduce.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
It doesn't surprise me that creduce/cvise is uneasy about it, because that
testcase really can't be reduced.
You'd need to reduce by hand, e.g. to:
#define A(n) if (a[n]) a[n]++;
#define B(n) A(n##0) A(n##1) A(n##2) A(n##3) A(n##4) A(n##5) A(n##6) A(n##7)
A(n##8) A(n##9)
#define C(n) B(n##0) B(n##1) B(n##2) B(n##3) B(n##4) B(n##5) B(n##6) B(n##7)
B(n##8) B(n##9)
#define D(n) C(n##0) C(n##1) C(n##2) C(n##3) C(n##4) C(n##5) C(n##6) C(n##7)
C(n##8) C(n##9)
#define E(n) D(n##0) D(n##1) D(n##2) D(n##3) D(n##4) D(n##5) D(n##6) D(n##7)
D(n##8) D(n##9)

unsigned
foo (int n, int m, int *a)
{
  unsigned int b, c, d, e, f, g, h, i;
  b = a[m] ? a[m + 4] : a[m + 5];
  c = a[m + 1];
  d = a[m + 2];
  e = a[m + 3];
  f = c * b;
  g = d * b * b;
  h = e * b * b * b;
  i = f + g + h;
  E(1) E(2)
  b = a[m] ? a[m + 4] : a[m + 5];
  c = a[m + 1];
  d = a[m + 2];
  e = a[m + 3];
  f = c * b;
  g = d * b * b;
  h = e * b * b * b;
  return (f + g + h) * i;
}
which indeed ICEs with unfixed gcc on 32-bit hosts at -O2, as this testcase has
more than 40000 basic blocks and uses phi_rank/bb_rank of the bbs with negative
ranks.
during GIMPLE pass: reassoc
pr98557.c: In function ‘foo’:
pr98557.c:8:1: internal compiler error: in insert_operand_rank, at
tree-ssa-reassoc.c:367
    8 | foo (int n, int m, int *a)
      | ^~~
0x8380d90 insert_operand_rank
        ../../gcc/tree-ssa-reassoc.c:367
0x8380d90 insert_operand_rank
        ../../gcc/tree-ssa-reassoc.c:365
0x8d3ea02 get_rank
        ../../gcc/tree-ssa-reassoc.c:455
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:322
0x8d3eb14 propagate_rank
        ../../gcc/tree-ssa-reassoc.c:348
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:442
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:322
0x8d3eb14 propagate_rank
        ../../gcc/tree-ssa-reassoc.c:348
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:442
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:322
0x8d3eb14 propagate_rank
        ../../gcc/tree-ssa-reassoc.c:348
0x8d3eb14 get_rank
        ../../gcc/tree-ssa-reassoc.c:442
0x8d3ecb6 get_rank
        ../../gcc/vec.h:290
0x8d3ecb6 add_to_ops_vec
        ../../gcc/tree-ssa-reassoc.c:569
0x8d3f58e linearize_expr_tree
        ../../gcc/tree-ssa-reassoc.c:5711
0x8d3f443 linearize_expr_tree
        ../../gcc/tree-ssa-reassoc.c:5753
0x8d3f443 linearize_expr_tree
        ../../gcc/tree-ssa-reassoc.c:5753
0x8d45f2c reassociate_bb
        ../../gcc/tree-ssa-reassoc.c:6492
0x8d461f2 reassociate_bb
        ../../gcc/tree-ssa-reassoc.c:6656
0x8d484b5 do_reassoc
        ../../gcc/tree-ssa-reassoc.c:6768
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

But I'm not sure the testcase is desirable for the testsuite, because it takes
to compile 105 seconds on a fast machine.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #10 from Matthias Klose <doko at debian dot org> ---
to reproduce the clang_delta segfault:

$ cat wxe_funcs.ii
static __typeof() a     __attribute__((__weakref__("pthread_mutex_destroy")))

$ /usr/lib/arm-linux-gnueabihf/clang_delta
--query-instances=replace-function-def-with-decl wxe_funcs.ii
clang_delta: /usr/lib/llvm-11/include/llvm/ADT/PointerIntPair.h:178: static
intptr_t llvm::PointerIntPairInfo<PointerT, IntBits,
PtrTraits>::updatePointer(intptr_t, PointerT) [with PointerT = clang::Stmt*;
unsigned int IntBits = 1; PtrTraits =
llvm::PointerLikeTypeTraits<clang::Stmt*>; intptr_t = int]: Assertion `(PtrWord
& ~PointerBitMask) == 0 && "Pointer is not sufficiently aligned"' failed.
Aborted (core dumped)

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[doko at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #11 from Matthias Klose <doko at debian dot org> ---
$ valgrind /usr/lib/arm-linux-gnueabihf/clang_delta
--query-instances=replace-function-def-with-decl wxe_funcs.ii
==48917== Memcheck, a memory error detector
==48917== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al.
==48917== Using Valgrind-3.16.1 and LibVEX; rerun with -h for copyright info
==48917== Command: /usr/lib/arm-linux-gnueabihf/clang_delta
--query-instances=replace-function-def-with-decl wxe_funcs.ii
==48917== 
clang_delta: /usr/lib/llvm-11/include/llvm/ADT/PointerIntPair.h:178: static
intptr_t llvm::PointerIntPairInfo<PointerT, IntBits,
PtrTraits>::updatePointer(intptr_t, PointerT) [with PointerT = clang::Stmt*;
unsigned int IntBits = 1; PtrTraits =
llvm::PointerLikeTypeTraits<clang::Stmt*>; intptr_t = int]: Assertion `(PtrWord
& ~PointerBitMask) == 0 && "Pointer is not sufficiently aligned"' failed.
==48917== 
==48917== Process terminating with default action of signal 6 (SIGABRT)
==48917==    at 0xB442216: __libc_do_syscall (libc-do-syscall.S:47)
==48917==    by 0xB451427: __libc_signal_restore_set (internal-signals.h:104)
==48917==    by 0xB451427: raise (raise.c:47)
==48917==    by 0xB441D55: abort (abort.c:79)
==48917==    by 0xB44C647: __assert_fail_base (assert.c:92)
==48917==    by 0xB44C6D1: __assert_fail (assert.c:101)
==48917==    by 0x969EAF: llvm::PointerIntPairInfo<clang::Stmt*, 1u,
llvm::PointerLikeTypeTraits<clang::Stmt*> >::updatePointer(int, clang::Stmt*)
[clone .part.0] (PointerIntPair.h:178)
==48917==    by 0x972259: updatePointer (SmallVector.h:345)
==48917==    by 0x972259: setPointerAndInt (PointerIntPair.h:76)
==48917==    by 0x972259: PointerIntPair (PointerIntPair.h:54)
==48917==    by 0x972259:
clang::RecursiveASTVisitor<ReplaceFunctionDefWithDeclCollectionVisitor>::TraverseStmt(clang::Stmt*,
llvm::SmallVectorImpl<llvm::PointerIntPair<clang::Stmt*, 1u, bool,
llvm::PointerLikeTypeTraits<clang::Stmt*>,
llvm::PointerIntPairInfo<clang::Stmt*, 1u,
llvm::PointerLikeTypeTraits<clang::Stmt*> > > >*) (RecursiveASTVisitor.h:576)
==48917==    by 0x97F703: TraverseVarDecl (RecursiveASTVisitor.h:2106)
==48917==    by 0x97F703:
clang::RecursiveASTVisitor<ReplaceFunctionDefWithDeclCollectionVisitor>::TraverseVarDecl(clang::VarDecl*)
(RecursiveASTVisitor.h:2106)
==48917==    by 0x96B6EB:
clang::RecursiveASTVisitor<ReplaceFunctionDefWithDeclCollectionVisitor>::TraverseDeclContextHelper(clang::DeclContext*)
[clone .part.0] (RecursiveASTVisitor.h:1383)
==48917==    by 0x96F179: TraverseDeclContextHelper
(RecursiveASTVisitor.h:1377)
==48917==    by 0x96F179: TraverseTranslationUnitDecl
(RecursiveASTVisitor.h:1484)
==48917==    by 0x96F179:
clang::RecursiveASTVisitor<ReplaceFunctionDefWithDeclCollectionVisitor>::TraverseDecl(clang::Decl*)
(DeclNodes.inc:595)
==48917==    by 0x96B661:
ReplaceFunctionDefWithDecl::HandleTranslationUnit(clang::ASTContext&)
(ReplaceFunctionDefWithDecl.cpp:71)
==48917==    by 0x4E58C47: clang::ParseAST(clang::Sema&, bool, bool)
(clang/lib/Parse/ParseAST.cpp:171)
==48917== 
==48917== HEAP SUMMARY:
==48917==     in use at exit: 875,117 bytes in 3,394 blocks
==48917==   total heap usage: 3,646 allocs, 252 frees, 989,805 bytes allocated
==48917== 
==48917== LEAK SUMMARY:
==48917==    definitely lost: 0 bytes in 0 blocks
==48917==    indirectly lost: 0 bytes in 0 blocks
==48917==      possibly lost: 0 bytes in 0 blocks
==48917==    still reachable: 875,117 bytes in 3,394 blocks
==48917==         suppressed: 0 bytes in 0 blocks
==48917== Rerun with --leak-check=full to see details of leaked memory
==48917== 
==48917== For lists of detected and suppressed errors, rerun with: -s
==48917== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
Aborted (core dumped)

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So fixed now?  clang_delta bugs should be tracked elsewhere.

[Bug target/98557] [11 Regression] ICE in insert_operand_rank, at tree-ssa-reassoc.c:367 on arm-linux-gnueabihf

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98557

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #13 from Martin Liška <marxin at gcc dot gnu.org> ---
Yes, it should be fixed.