[Bug middle-end/105638] New: Redundant stores aren't removed by DSE

[Bug middle-end/105638] New: Redundant stores aren't removed by DSE

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105638

            Bug ID: 105638
           Summary: Redundant stores aren't removed by DSE
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hjl.tools at gmail dot com
  Target Milestone: ---

For

$ cat foo.cpp
#include <stdint.h>
#include <vector>
#include <tr1/array>

class FastBoard {
public:
    typedef std::pair<int, int> movescore_t;
    typedef std::tr1::array<movescore_t, 24> scoredlist_t;

protected:
    std::vector<int> m_critical;

    int m_boardsize;    
};

class FastState {
public:        
    FastBoard board;

    int movenum;              
protected:
    FastBoard::scoredlist_t scoredmoves;
};

class KoState : public FastState {
private:         
    std::vector<uint64_t> ko_hash_history;   
    std::vector<uint64_t> hash_history;     
};

class GameState : public KoState {
public:                    
    void foo ();      
private:
    std::vector<KoState> game_history;                          
};

void GameState::foo() {
    game_history.resize(movenum);
}
$ g++ -O2 -march=skylake foo.cpp -S

generates:

...
        movl    $280, %edx
        xorl    %esi, %esi
        call    memset
        movq    %rax, %rcx
        vpxor   %xmm0, %xmm0, %xmm0
        addq    $280, %rcx
        vmovdqu %xmm0, 36(%rax)
        vmovdqu %xmm0, 52(%rax)
        vmovdqu %xmm0, 68(%rax)
        vmovdqu %xmm0, 84(%rax)
        vmovdqu %xmm0, 100(%rax)
        vmovdqu %xmm0, 116(%rax)
        vmovdqu %xmm0, 132(%rax)
        vmovdqu %xmm0, 148(%rax)
        vmovdqu %xmm0, 164(%rax)
        vmovdqu %xmm0, 180(%rax)
        vmovdqu %xmm0, 196(%rax)
        vmovdqu %xmm0, 212(%rax)
...

Here memset has cleared 280 bytes starting from RAX.  There is no need to
clear these bytes again.  The optimized tree dump shows:

  <bb 14> [local count: 444773291]:
  # __cur_154 = PHI <__cur_42(14), _6(13)>
  # __n_155 = PHI <__n_41(14), __n_20(D)(13)>
  *__cur_154 = {};
  MEM[(int * *)__cur_154] = 0B; 
  MEM[(int * *)__cur_154 + 8B] = 0B; 
  MEM[(int * *)__cur_154 + 16B] = 0B; 
  MEM[(struct array *)__cur_154 + 36B]._M_instance = {}; 
  MEM <vector(4) long unsigned int> [(long unsigned int * *)__cur_154 + 232B] =
{ 0, 0, 0, 0 }; 
  MEM[(long unsigned int * *)__cur_154 + 264B] = 0B; 
  MEM[(long unsigned int * *)__cur_154 + 272B] = 0B;

Some of them are removed by RTL DSE.  But vector stores aren't.  Should
SSA DSE remove them?

[Bug middle-end/105638] Redundant stores aren't removed by DSE

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105638

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2022-05-18
     Ever confirmed|0                           |1
           Keywords|                            |missed-optimization

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
DSE doesn't remove redundant stores, in some cases it can prune an earlier
store but that's only handled in a limited set of cases.  The issue with
removing redundant stores is that side-effects on the effective dynamic type
have to be preserved so removing earlier stores is much better.

[Bug middle-end/105638] Redundant stores aren't removed by DSE

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105638

--- Comment #2 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by H.J. Lu <hjl@gcc.gnu.org>:

https://gcc.gnu.org/g:a743a72714fc4a9d7036d28d0cacdf2a3621f629

commit r13-921-ga743a72714fc4a9d7036d28d0cacdf2a3621f629
Author: H.J. Lu <hjl.tools@gmail.com>
Date:   Wed May 18 13:00:47 2022 -0700

    DSE: Use the constant store source if possible

    RTL DSE tracks redundant constant stores within a basic block.  When RTL
    loop invariant motion hoists a constant initialization out of the loop
    into a separate basic block, the constant store value becomes unknown
    within the original basic block.  When recording store for RTL DSE, check
    if the source register is set only once to a constant by a non-partial
    unconditional load.  If yes, record the constant as the constant store
    source.  It eliminates unrolled zero stores after memset 0 in a loop
    where a vector register is used as the zero store source.

    gcc/

            PR rtl-optimization/105638
            * df-core.cc (df_find_single_def_src): Moved and renamed from
            find_single_def_src in loop-iv.cc.  Change the argument to rtx
            and use rtx_equal_p.  Return null for partial or conditional
            defs.
            * df.h (df_find_single_def_src): New prototype.
            * dse.cc (record_store): Use the constant source if the source
            register is set only once.
            * loop-iv.cc (find_single_def_src): Moved to df-core.cc.
            (replace_single_def_regs): Replace find_single_def_src with
            df_find_single_def_src.

    gcc/testsuite/

            PR rtl-optimization/105638
            * g++.target/i386/pr105638.C: New test.

[Bug middle-end/105638] Redundant stores aren't removed by DSE

[hjl.tools at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105638

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|NEW                         |RESOLVED
   Target Milestone|---                         |13.0

--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
Fixed.