[Bug tree-optimization/47579] New: STL size() == 0 does unnecessary shift

[Bug tree-optimization/47579] New: STL size() == 0 does unnecessary shift

[ian at airs dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47579

           Summary: STL size() == 0 does unnecessary shift
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ian@airs.com


Consider this C++ code:

#include <vector>
extern void b1(), b2();
void foo(const std::vector<int>& v) { if (v.size() == 0) b1(); else b2(); }

When I compile it with current mainline with -O2 on x86_64, I get this:

        movq    8(%rdi), %rax
        subq    (%rdi), %rax
        sarq    $2, %rax
        testq   %rax, %rax
        ...

That sarq instruction is useless.  We know that the two values being subtracted
are both aligned pointers, so we should know that the two least significant
bits of the result are zero.  And that should be enough to let us know that we
don't need to shift before comparing for equality with zero.

[Bug tree-optimization/47579] STL size() == 0 does unnecessary shift

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47579

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-02-01 22:54:19 UTC ---
I think v.size() should be simply optimized to work on char * pointer
differences.  Recognizing the optimization opportunity requires to see
dereferences of the pointers to deduce their alignment.

[Bug tree-optimization/47579] STL size() == 0 does unnecessary shift

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47579

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-01-06
     Ever Confirmed|0                           |1

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-01-06 00:32:36 UTC ---
Two more interesting testcases:
void f(int *a, int *b)  {if((b-a)==0) b1(); else b2();}
void f1(int *a, int *b)
{
  std::size_t t = a-b;
  if (t == 0)  b1(); else b2();
}
--- CUT ---
f produces:
    .cfi_startproc
    subq    %rdi, %rsi
    addq    $3, %rsi
    cmpq    $6, %rsi
    jbe    .L7
    jmp    _Z2b2v
    .p2align 4,,10
    .p2align 3
.L7:
    jmp    _Z2b1v
    .cfi_endproc

While f1 produces:
    .cfi_startproc
    subq    %rsi, %rdi
    sarq    $2, %rdi
    testq    %rdi, %rdi
    je    .L10
    jmp    _Z2b2v
    .p2align 4,,10
    .p2align 3
.L10:
    jmp    _Z2b1v
    .cfi_endproc
Which is the same as your foo.  So we should be able to do the same as f for
foo.
f:
  b.2_2 = (long int) b_1(D);
  a.3_4 = (long int) a_3(D);
  D.8126_5 = b.2_2 - a.3_4;
  D.8127_6 = (long unsigned int) D.8126_5;
  D.8128_7 = D.8127_6 + 3;
  if (D.8128_7 <= 6)
    goto <bb 3>;
  else
    goto <bb 4>;

f1 (and foo really):
  a.0_2 = (long int) a_1(D);
  b.1_4 = (long int) b_3(D);
  D.8119_5 = a.0_2 - b.1_4;
  D.8120_6 = D.8119_5 /[ex] 4;
  if (D.8120_6 == 0)

[Bug tree-optimization/47579] STL size() == 0 does unnecessary shift

[vanyacpp at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47579

Ivan Sorokin <vanyacpp at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vanyacpp at gmail dot com

--- Comment #3 from Ivan Sorokin <vanyacpp at gmail dot com> ---
Since 7.1 GCC doesn't produce any shifts on the test code as well as on the
examples from comment #2. https://godbolt.org/z/f48EqP

I think the bug can be closed now.

[Bug tree-optimization/47579] STL size() == 0 does unnecessary shift

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47579

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
           Severity|normal                      |enhancement
      Known to fail|                            |6.1.0
         Resolution|---                         |FIXED
   Target Milestone|---                         |7.0
      Known to work|                            |7.1.0

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed for GCC 7 by r7-4718-g40fd269ab128d1 .