[Bug tree-optimization/108354] New: Dead Code Elimination Regression at -O2 (trunk vs. 12.2.0)

[Bug tree-optimization/108354] New: Dead Code Elimination Regression at -O2 (trunk vs. 12.2.0)

[yann at ywg dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

            Bug ID: 108354
           Summary: Dead Code Elimination Regression at -O2 (trunk vs.
                    12.2.0)
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: yann at ywg dot ch
  Target Milestone: ---

Created attachment 54229
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54229&action=edit
presented code as file

cat case.c #617
int b;
int *c;
int e;
static int *f = &e;
int g;
void foo();
short(a)(short h, short i) { return h - i; }
int(d)(int h) { return h == 83647 ? 0 : -h; }
int main() {
  short j;
  int *k = &e, *l = &b;
  *f = 0 == c;
  j = a(0 != 2, *k);
  if (d(j ^ (0 == l || *k)) != *k)
    ;
  else
    foo();
  c = &g;
}

`gcc-cb93c5f8008b95743b741d6f1842f9be50c6985c (trunk) -O2` can not eliminate
`foo` but `gcc-releases/gcc-12.2.0 -O2` can.

`gcc-cb93c5f8008b95743b741d6f1842f9be50c6985c (trunk) -O2 -S -o /dev/stdout
case.c`
--------- OUTPUT ---------
main:
.LFB2:
        .cfi_startproc
        xorl    %eax, %eax
        cmpq    $0, c(%rip)
        setne   %dl
        sete    %al
        movzbl  %dl, %edx
        movl    %eax, e(%rip)
        orl     %eax, %edx
        je      .L12
        movq    $g, c(%rip)
        xorl    %eax, %eax
        ret
.L12:
        pushq   %rax
        .cfi_def_cfa_offset 16
        xorl    %eax, %eax
        call    foo
        xorl    %eax, %eax
        movq    $g, c(%rip)
        popq    %rdx
        .cfi_def_cfa_offset 8
        ret
---------- END OUTPUT ---------


`gcc-releases/gcc-12.2.0 -O2 -S -o /dev/stdout case.c`
--------- OUTPUT ---------
main:
.LFB2:
        .cfi_startproc
        xorl    %eax, %eax
        cmpq    $0, c(%rip)
        movq    $g, c(%rip)
        sete    %al
        movl    %eax, e(%rip)
        xorl    %eax, %eax
        ret
---------- END OUTPUT ---------


Bisects to:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=b3e98eb3396e16ae8b20c94916bc2bd7862d2c97

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Dead Code Elimination       |[13 Regression] Dead Code
                   |Regression at -O2 (trunk    |Elimination Regression at
                   |vs. 12.2.0)                 |-O2 since
                   |                            |r13-89-gb3e98eb3396e16
     Ever confirmed|0                           |1
   Target Milestone|---                         |13.0
             Status|UNCONFIRMED                 |NEW
                 CC|                            |marxin at gcc dot gnu.org,
                   |                            |sayle at gcc dot gnu.org
   Last reconfirmed|                            |2023-01-10

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |pinskia at gcc dot gnu.org
             Status|NEW                         |ASSIGNED
         Depends on|                            |96921

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Looks like:
  _4 = (int) _2;
...
  i.1_15 = (unsigned short) _2;
  _16 = 1 - i.1_15;
  _6 = (int) _16;
...
  _8 = _4 | _6;



Simplify to just
  _4 = (int) _2;
...
_t = _2 ^ 1;
_6 = (int)_t;
...
  _8 = _4 | _6;

Which then simplifies down to:
_8 = 1; (because (_2 ^ 1) | (_2) == 1)

Which then would be optimized.
Let me look.


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96921
[Bug 96921] Failure to optimize combined boolean not patterns

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is a specific case of PR 91213 and the case which will also fix PR 96921:

/* 1 - a is a ^ 1 if a had a bool range. */
(simplify
 (minus integer_onep@0 SSA_NAME@1)
  (if (INTEGRAL_TYPE_P (type)
       && ssa_name_has_boolean_range (@1))
   (bit_xor @1 @0)))

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> This is a specific case of PR 91213 and the case which will also fix PR
> 96921:
> 
> /* 1 - a is a ^ 1 if a had a bool range. */
> (simplify
>  (minus integer_onep@0 SSA_NAME@1)
>   (if (INTEGRAL_TYPE_P (type)
>        && ssa_name_has_boolean_range (@1))
>    (bit_xor @1 @0)))

I might be running into an Ada build failure with this patch; I am double
checking this now. It also could be a latent bug ...

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> (In reply to Andrew Pinski from comment #2)
> > This is a specific case of PR 91213 and the case which will also fix PR
> > 96921:
> > 
> > /* 1 - a is a ^ 1 if a had a bool range. */
> > (simplify
> >  (minus integer_onep@0 SSA_NAME@1)
> >   (if (INTEGRAL_TYPE_P (type)
> >        && ssa_name_has_boolean_range (@1))
> >    (bit_xor @1 @0)))
> 
> I might be running into an Ada build failure with this patch; I am double
> checking this now. It also could be a latent bug ...

I added an extra check for TYPE_UNSIGNED and there is still an issue.
I suspect there might be a latent bug happening somewhere. 

apinski@xeond:~/src/upstream-gcc/gcc/objdir/gcc/ada/tools$ ../../xgcc -B../../
-c -g -O2   -W -Wall  -gnatpg -gnata -I- -I../rts -I.
-I/home/apinski/src/upstream-gcc/gcc/gcc/ada
/home/apinski/src/upstream-gcc/gcc/gcc/ada/gnatlink.adb -o gnatlink.o -v
Reading specs from ../../specs
COLLECT_GCC=../../xgcc
Target: x86_64-pc-linux-gnu
Configured with: /home/apinski/src/upstream-gcc/gcc/configure
--prefix=/home/apinski/upstream-gcc
--enable-languages=c,c++,fortran,lto,objc,go,ada
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 13.0.1 20230131 (experimental) [master 733988099bc] (GCC)
COLLECT_GCC_OPTIONS='-B' '../../' '-c' '-g' '-O2' '-Wextra' '-Wall' '-gnatpg'
'-gnata' '-I' '-' '-I' '../rts' '-I' '.' '-I'
'/home/apinski/src/upstream-gcc/gcc/gcc/ada' '-o' 'gnatlink.o' '-v'
'-mtune=generic' '-march=x86-64'
 ../../gnat1 -I - -I ../rts -I . -I /home/apinski/src/upstream-gcc/gcc/gcc/ada
-quiet -O2 -Wextra -Wall -dumpbase gnatlink.adb -dumpbase-ext .adb -gnatwa -g
-gnatpg -gnata -mtune=generic -march=x86-64 -gnatO gnatlink.o
/home/apinski/src/upstream-gcc/gcc/gcc/ada/gnatlink.adb -o /tmp/ccn5ePu7.s

raised STORAGE_ERROR : stack overflow or erroneous memory access

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #4)
> raised STORAGE_ERROR : stack overflow or erroneous memory access

0x00000000015769b4 in get_range_query (fun=<optimized out>) at
/home/apinski/src/upstream-gcc/gcc/gcc/value-query.h:143
143       return fun->x_range_query ? fun->x_range_query : &global_ranges;
(gdb) bt
#0  0x00000000015769b4 in get_range_query (fun=<optimized out>) at
/home/apinski/src/upstream-gcc/gcc/gcc/value-query.h:143
#1  ssa_name_has_boolean_range(tree_node*) () at
/home/apinski/src/upstream-gcc/gcc/gcc/tree-ssanames.cc:534
#2  0x0000000001ae01a6 in generic_simplify_MINUS_EXPR(unsigned int, tree_code,
tree_node*, tree_node*, tree_node*) [clone .constprop.0] () at
/home/apinski/src/upstream-gcc/gcc/gcc/tree.h:3654
#3  0x0000000000fb1286 in fold_binary_loc(unsigned int, tree_code, tree_node*,
tree_node*, tree_node*) () at
/home/apinski/src/upstream-gcc/gcc/gcc/fold-const.cc:10926
#4  0x0000000000fb8eeb in fold_build2_loc (loc=loc@entry=0,
code=code@entry=MINUS_EXPR, type=0x7ffff725c000, op0=op0@entry=0x7ffff7242d50,
op1=op1@entry=0x7ffff52a6870) at
/home/apinski/src/upstream-gcc/gcc/gcc/fold-const.cc:13824
#5  0x0000000000f7373f in get_inner_reference(tree_node*, poly_int_pod<1u,
long>*, poly_int_pod<1u, long>*, tree_node**, machine_mode*, int*, int*, int*)
() at /home/apinski/src/upstream-gcc/gcc/gcc/tree.h:3645
#6  0x00000000010a2546 in (anonymous
namespace)::isra_analyze_call(cgraph_edge*) () at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2167
#7  0x00000000010a32a8 in (anonymous
namespace)::isra_analyze_all_outgoing_calls (node=<error reading variable:
dwarf2_find_location_expression: Corrupted DWARF expression.>) at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2202
#8  ipa_sra_summarize_function(cgraph_node*) () at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:4540
#9  0x00000000010a5558 in (anonymous namespace)::ipa_sra_generate_summary() ()
at /home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2716
#10 0x0000000001242f23 in execute_ipa_summary_passes (ipa_pass=0x3a90670) at
/home/apinski/src/upstream-gcc/gcc/gcc/passes.cc:2297
(gdb) p cfun
$3 = (function *) 0x0

Ok, let me mark the pattern only for GIMPLE because it looks like we are
calling fold without setting cfun and still passing a SSA_NAME.
I suspect we might run into this issue with other code too but at least I hide
the latent bug ...

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Full backtrace:
(gdb) bt
#0  0x000000000158208f in get_range_query (fun=0x0) at
/home/apinski/src/upstream-gcc/gcc/gcc/value-query.h:143
#1  0x0000000001e4a9f0 in ssa_name_has_boolean_range (op=0x7ffff52a6870) at
/home/apinski/src/upstream-gcc/gcc/gcc/tree-ssanames.cc:534
#2  0x00000000026a6e33 in generic_simplify_MINUS_EXPR (loc=0, code=MINUS_EXPR,
type=0x7ffff725c000, _p0=0x7ffff7242d50, _p1=0x7ffff52a6870) at
generic-match.cc:36418
#3  0x00000000027c9064 in generic_simplify (loc=0, code=MINUS_EXPR,
type=0x7ffff725c000, _p0=0x7ffff7242d50, _p1=0x7ffff52a6870) at
generic-match.cc:103671
#4  0x00000000015670c0 in fold_binary_loc (loc=0, code=MINUS_EXPR,
type=0x7ffff725c000, op0=0x7ffff7242d50, op1=0x7ffff52a6870) at
/home/apinski/src/upstream-gcc/gcc/gcc/fold-const.cc:10926
#5  0x0000000001575b6a in fold_build2_loc (loc=0, code=MINUS_EXPR,
type=0x7ffff725c000, op0=0x7ffff7242d50, op1=0x7ffff52a6870) at
/home/apinski/src/upstream-gcc/gcc/gcc/fold-const.cc:13824
#6  0x000000000150714d in get_inner_reference (exp=0x7ffff5a89540,
pbitsize=0x7fffffffd5a8, pbitpos=0x7fffffffd5b0, poffset=0x7fffffffd568,
pmode=0x7fffffffd54c, punsignedp=0x7fffffffd550, preversep=0x7fffffffd554,
pvolatilep=0x7fffffffd560) at
/home/apinski/src/upstream-gcc/gcc/gcc/expr.cc:7995
#7  0x00000000016dac07 in (anonymous namespace)::isra_analyze_call
(cs=0x7ffff4e59888) at /home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2167
#8  0x00000000016dad39 in (anonymous
namespace)::isra_analyze_all_outgoing_calls (node=0x7ffff5a38cc0) at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2202
#9  0x00000000016e2295 in ipa_sra_summarize_function (node=0x7ffff5a38cc0) at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:4540
#10 0x00000000016dc4e0 in (anonymous namespace)::ipa_sra_generate_summary () at
/home/apinski/src/upstream-gcc/gcc/gcc/ipa-sra.cc:2716
#11 0x000000000195ef94 in execute_ipa_summary_passes (ipa_pass=0x52a9270) at
/home/apinski/src/upstream-gcc/gcc/gcc/passes.cc:2297
#12 0x00000000013b8699 in ipa_passes () at
/home/apinski/src/upstream-gcc/gcc/gcc/cgraphunit.cc:2226
#13 0x00000000013b8987 in symbol_table::compile (this=0x7ffff7246000) at
/home/apinski/src/upstream-gcc/gcc/gcc/cgraphunit.cc:2322
#14 0x00000000013b8f7c in symbol_table::finalize_compilation_unit
(this=0x7ffff7246000) at
/home/apinski/src/upstream-gcc/gcc/gcc/cgraphunit.cc:2574
#15 0x0000000001ac4ae9 in compile_file () at
/home/apinski/src/upstream-gcc/gcc/gcc/toplev.cc:471
#16 0x0000000001ac7b29 in do_compile (no_backend=false) at
/home/apinski/src/upstream-gcc/gcc/gcc/toplev.cc:2125
#17 0x0000000001ac7f11 in toplev::main (this=0x7fffffffdba2, argc=28,
argv=0x7fffffffdca8) at /home/apinski/src/upstream-gcc/gcc/gcc/toplev.cc:2277
#18 0x00000000034a7ea1 in main (argc=28, argv=0x7fffffffdca8) at
/home/apinski/src/upstream-gcc/gcc/gcc/main.cc:39

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

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

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.

[Bug tree-optimization/108354] [13 Regression] Dead Code Elimination Regression at -O2 since r13-89-gb3e98eb3396e16

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108354
Bug 108354 depends on bug 96921, which changed state.

Bug 96921 Summary: Failure to optimize combined boolean not patterns
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96921

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