[Bug tree-optimization/104530] New: [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[Bug tree-optimization/104530] New: [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[theodort at inf dot ethz.ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

            Bug ID: 104530
           Summary: [12 Regression] Dead Code Elimination Regression at
                    -O3 (trunk vs. 11.2.0)
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: theodort at inf dot ethz.ch
  Target Milestone: ---

Dead Code Elimination Regression at -O3 (trunk vs.
a7fede6704dd207fb32b97bc30f945acc7b953c5) 138751
---------------
cat case.c #138751
void foo(void);

static int a, *b = &a, c, d = 1;

int main() {
    c = 0 == b;
    a = *b;
    if (c % d)
        for (; d; --d)
            foo();
    b = 0;
}

gcc-58aeb75d4097010ad9bb72b964265b18ab284f93 (trunk) -O3 can not eliminate foo
but gcc-11.2.0 -O3 can.

gcc-58aeb75d4097010ad9bb72b964265b18ab284f93 (trunk) -O3 -S -o /dev/stdout
case.c
--------- OUTPUT ---------
main:
.LFB0:
        .cfi_startproc
        movq    b(%rip), %rax
        testq   %rax, %rax
        movl    (%rax), %edx
        sete    %al
        movzbl  %al, %eax
        movl    %edx, a(%rip)
        cltd
        idivl   d(%rip)
        testl   %edx, %edx
        je      .L9
        pushq   %rax
        .cfi_def_cfa_offset 16
        .p2align 4,,10
        .p2align 3
.L3:
        call    foo
        subl    $1, d(%rip)
        jne     .L3
        xorl    %ecx, %ecx
        xorl    %eax, %eax
        movq    %rcx, b(%rip)
        popq    %rsi
        .cfi_def_cfa_offset 8
        ret
.L9:
        xorl    %edx, %edx
        xorl    %eax, %eax
        movq    %rdx, b(%rip)
        ret
---------- END OUTPUT ---------


gcc-11.2.0 -O3 -S -o /dev/stdout case.c
--------- OUTPUT ---------

main:
.LFB0:
        .cfi_startproc
        movq    b(%rip), %rax
        movq    $0, b(%rip)
        movl    (%rax), %eax
        movl    %eax, a(%rip)
        xorl    %eax, %eax
        ret
---------- END OUTPUT ---------


Bisects to:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=c6bb1db76b3ac127aff7dacf391fc1798a94bb7d

----- Build information -----
----- 58aeb75d4097010ad9bb72b964265b18ab284f93 (trunk)
Target: x86_64-pc-linux-gnu
Configured with: ../configure --disable-multilib --disable-bootstrap
--enable-languages=c,c++
--prefix=/zdata/compiler_cache/gcc-58aeb75d4097010ad9bb72b964265b18ab284f93
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.0.1 20220213 (experimental) (GCC)

----- releases/gcc-11.2.0
Target: x86_64-pc-linux-gnu
Configured with: ../configure --disable-multilib --disable-bootstrap
--enable-languages=c,c++
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.0.0   (GCC)

[Bug tree-optimization/104530] [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.0
                 CC|                            |jakub at gcc dot gnu.org
           Priority|P3                          |P1

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r12-7128-gc6bb1db76b3ac127aff7dacf391fc1798a94bb7d
The comparison is done before dereferencing pointer, but unlike the PR104288
case there is no call that could fail to return in between.
I think PR104288 has been fixed on the 11 branch too, so the question is if
this still works there.

[Bug tree-optimization/104530] [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

--- Comment #2 from Andrew Macleod <amacleod at redhat dot com> ---
mmm. tricky.  

Imports: b.0_1  d.3_7
Exports: b.0_1  _2  _3  d.3_7  _8
         _2 : b.0_1(I)
         _3 : b.0_1(I)  _2
         _8 : b.0_1(I)  _2  _3  d.3_7(I)

    <bb 2> :
    b.0_1 = b;
    _2 = b.0_1 == 0B;
    _3 = (int) _2;
    c = _3;
    _5 = *b.0_1;        <<-- from this point b.0_1 is [+1, +INF]
    a = _5;
    d.3_7 = d;
    _8 = _3 % d.3_7;
    if (_8 != 0)
      goto <bb 6>; [INV]
    else
      goto <bb 5>; [INV]

This is a bit at conflict with the "lets be safe" approach without pre-scanning
the block for lack of throwing stmts.   Although that's really just a half
measure since you can construct more complicated blocks with a mix.

Im experimenting with marking values as stale for all exports of the block
which are dependant on the value we set to non null. ie, _2, _3 and _8 

Then when we ask for a range of that name AFTER the de-reference, it will
recompute it.  The use of _3 in computing _8 is then "stale", and will be
re-evaluated.

Simple in concept, but I'm encountering an interaction with how we currently
set the global values for the end of the pass, and although I can recompute _2
fine, it is also setting the global value of _2 to [0,0], which is not what we
want.  Works for this case, but returns us to the same situation we had before
if there is a call in the block.

Im continuing to look at it.  Easy fix in the next release since we are going
to change the way we export globals outside the pass... but we shall see about
this release whether I can keep it contained.

[Bug tree-optimization/104530] [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
What could be done in passes like evrp or vrp* but not in unrelated other
passes that use the range infrastructure would be to set uids on stmts
depending on presence of calls that might not always return in the middle of
basic blocks (and perhaps other stmts, wonder about e.g. inline asm),
and optimize whenever the comparison and dereference have the same uid (aka
there is nothing in between that would make the optimization problematic).

[Bug tree-optimization/104530] [12 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P2

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Downgrading, lets defer to GCC 13 (we can still backport if the solution is
suitable).

[Bug tree-optimization/104530] [12/13 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|12.0                        |12.2

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 12.1 is being released, retargeting bugs to GCC 12.2.

[Bug tree-optimization/104530] [12/13 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

--- Comment #7 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Andrew Macleod <amacleod@gcc.gnu.org>:

https://gcc.gnu.org/g:c838119946c9f75f1e42f4320275355822cc86fc

commit r13-3765-gc838119946c9f75f1e42f4320275355822cc86fc
Author: Andrew MacLeod <amacleod@redhat.com>
Date:   Mon Nov 7 15:07:35 2022 -0500

    Add transitive inferred range processing.

    Rewalk statements at the end of a block to see if any inferred ranges
    affect earlier calculations and register those as inferred ranges.

            gcc/
            PR tree-optimization/104530
            * gimple-range-cache.cc (ranger_cache::register_inferred_value):
            New.  Split from:
            (ranger_cache::apply_inferred_ranges): Move setting cache to
            separate function.
            * gimple-range-cache.h (register_inferred_value): New prototype.
            * gimple-range-infer.cc (infer_range_manager::has_range_p): New.
            * gimple-range-infer.h (has_range_p): New prototype.
            * gimple-range.cc (register_transitive_inferred_ranges): New.
            * gimple-range.h (register_transitive_inferred_ranges): New proto.
            * tree-vrp.cc (rvrp_folder::fold_stmt): Check for transitive
inferred
            ranges at the end of the block before folding final stmt.

            gcc/testsuite/
            * gcc.dg/pr104530.c: New.

[Bug tree-optimization/104530] [12/13 Regression] Dead Code Elimination Regression at -O3 (trunk vs. 11.2.0)

[amacleod at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104530

Andrew Macleod <amacleod at redhat dot com> changed:

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

--- Comment #8 from Andrew Macleod <amacleod at redhat dot com> ---
fixed.