[Bug rtl-optimization/43491] New: Unnecessary temporary for global register variable

[Bug rtl-optimization/43491] New: Unnecessary temporary for global register variable

[mirq-gccboogs at rere dot qmqm dot pl]

gcc sometimes allocates a temporary register for a variable that is global and
has a fixed register. This happens when:
 a. the global is a register-fixed variable
 b. global is a pointer to structure
 c. an address of structure's field is passed as argument to inlined function
 d. the global is marked as constant


code:

struct b {
        unsigned g,h,j;
};

register struct b *const reg asm("r4");

static inline int diff(unsigned *p)
{
        return *p;
}

void c(void);

void d(void)
{
        while (diff(&reg->j))
                c();
}


generates:

d:
        @ args = 0, pretend = 0, frame = 0
        @ frame_needed = 0, uses_anonymous_args = 0
        ldr     r3, [r4, #8]    @ first ok
        push    {r5, lr}
        mov     r5, r4          @ an temporary even when r4 is marked const
        cbz     r3, .L1
.L4:   
        bl      c
        ldr     r3, [r5, #8]    @ accesses via temporary
        cmp     r3, #0
        bne     .L4
.L1:   
        pop     {r5, pc}


-- 
           Summary: Unnecessary temporary for global register variable
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: mirq-gccboogs at rere dot qmqm dot pl
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: arm-none-eabi


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43491

[Bug rtl-optimization/43491] Unnecessary temporary for global register variable

[ibolton at gcc dot gnu dot org]

------- Comment #1 from ibolton at gcc dot gnu dot org  2010-09-08 16:21 -------
reg is assigned to a temporary (reg.0) at the very first tree pass, as shown by
this 004t.gimple dump:

d ()
{
  struct b * const reg.0;
  unsigned int * D.2019;
  int D.2020;

  goto <D.1276>;
  <D.1275>:
  c ();
  <D.1276>:
  reg.0 = reg;
  D.2019 = &reg.0->j;
  D.2020 = diff (D.2019);
  if (D.2020 != 0) goto <D.1275>; else goto <D.1277>;
  <D.1277>:
}

I'm thinking that this is perfectly normal thing to do, and that the redundant
move is meant to disappear in a later pass.  My guess is that IRA is choosing
not to assign the pseudo to r4, but I do not know why at the moment.


-- 

ibolton at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
           Keywords|                            |missed-optimization, ra
      Known to fail|                            |4.5.3 4.6.0
   Last reconfirmed|0000-00-00 00:00:00         |2010-09-08 16:21:50
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43491