[Bug rtl-optimization/56711] New: spectaculary bad code generated for __uint128_t

[Bug rtl-optimization/56711] New: spectaculary bad code generated for __uint128_t

[felix-gcc at fefe dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56711

             Bug #: 56711
           Summary: spectaculary bad code generated for __uint128_t
    Classification: Unclassified
           Product: gcc
           Version: 4.7.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: felix-gcc@fefe.de


Consider this function:

size_t scan_ulong(const char* src,unsigned long int* dest) {
  register const char *tmp=src;
  register unsigned long int l=0;
  register unsigned char c;
  while ((c=*tmp-'0')<10) {
    __uint128_t x=(__uint128_t)l*10+c;
    if ((unsigned long)x != x) break;
    l=(unsigned long)x;
    ++tmp;
  }
  if (tmp-src) *dest=l;
  return tmp-src;
}

I'm compiling this with gcc -Os -c test.c on an amd64-linux box.
The code gcc generates is 92 bytes long, the one from clang only 65.  What is
happening here?  What are all that code doing that gcc is generating there?

[Bug tree-optimization/56711] missed optimization for __uint128_t of (unsigned long long)x != x

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56711

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-linux-gnu
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |missed-optimization
   Last reconfirmed|                            |2013-03-24
          Component|rtl-optimization            |tree-optimization
     Ever Confirmed|0                           |1
            Summary|spectaculary bad code       |missed optimization for
                   |generated for __uint128_t   |__uint128_t of (unsigned
                   |                            |long long)x != x
           Severity|normal                      |enhancement

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2013-03-24 17:22:57 UTC ---
>The code gcc generates is 92 bytes long, the one from clang only 65. 

If you are going to compare GCC to clang with respect of code size then you
should analyze the code at least.  It is very obvious what GCC is not doing.

Well looking at the generated code GCC does not optimize:
    if ((unsigned long)x != x) break;

Into just:
x>>64 == 0

An another example is:
int f(unsigned a)
{
  return (unsigned short)a != a;
}

int g(unsigned a)
{
  return a>>16 == 0;
}

[Bug tree-optimization/56711] missed optimization for __uint128_t of (unsigned long long)x != x

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56711

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

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

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I am going to look into implementing this.
Shouldn't be hard:
 _5 = a_2(D) & 18446744069414584320;
  _1 = _5 != 0;

Note for aarch64 we want to do the opposite on the RTL level though.
Clang is able to produce the shift for x86_64 but for armv8, these two
functions have different code gen :).

[Bug tree-optimization/56711] missed optimization for __uint128_t of (unsigned long long)x != x

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56711

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=94589
         Resolution|---                         |FIXED
   Target Milestone|---                         |12.0
             Status|ASSIGNED                    |RESOLVED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
f5f183843540 f7854e2faf76


This was fixed already for GCC 12 by:
r12-733-gf5f1838435400b