[Bug tree-optimization/32826] New: Reduction into a global variable causes a Load Hit Store Hazard (for the Cell)

[Bug tree-optimization/32826] New: Reduction into a global variable causes a Load Hit Store Hazard (for the Cell)

[pinskia at gcc dot gnu dot org]

Testcase (compile at -O2 -maltivec -ftree-vectorize):
int a[16*100];
int e;
float f(void)
{
  int i;
  int e1;
  e1 = e;
  for(i = 0;i<16*100;i++)
    e1 += a[i];
  e = e1;
}

----------- cut ------
Currently you get:
        stvewx v1,0,r2
        lis r2,ha16(_e)
        lwz r0,-20(r1)   <---- LHS hazard
        stw r0,lo16(_e)(r2)

Even though the elements of v1 will all be the same, so GCC could do:
        lis r2,ha16(_e)
        add r2, lo16(_e)(r2)
        stvewx v1,0,r2


-- 
           Summary: Reduction into a global variable causes a Load Hit Store
                    Hazard (for the Cell)
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org
GCC target triplet: powerpc*-*-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32826

[Bug tree-optimization/32826] Reduction into a global variable causes a Load Hit Store Hazard (for the Cell)

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2007-07-19 17:15 -------
Note I copied the wrong testcase, the correct testcase is:
int a[16*100];
int e;
float f(void)
{
  int i;
  int e1;
  e1 = 0;
  for(i = 0;i<16*100;i++)
    e1 += a[i];
  e = e1;
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32826

[Bug tree-optimization/32826] Reduction into a global variable causes a Load Hit Store Hazard (for the Cell)

[tehila at il dot ibm dot com]

------- Comment #2 from tehila at il dot ibm dot com  2007-07-26 10:46 -------
(In reply to comment #2)
Just want a clarification:
I see you're compiling on PPU (since you're using -maltivec).
Does this problematic also on SPU? Does SPU has this LHS hazard?

Another question:
lwz r0,-20(r1)   <---- LHS hazard
stw r0,lo16(_e)(r2)

The problem here is these 2 insns, right?
The store that is right after (or too close to) the load ?


-- 

tehila at il dot ibm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tehila at il dot ibm dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32826

[Bug tree-optimization/32826] Reduction into a global variable causes a Load Hit Store Hazard (for the Cell)

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2007-07-26 17:02 -------
(In reply to comment #2)
> (In reply to comment #2)
> Just want a clarification:
> I see you're compiling on PPU (since you're using -maltivec).
> Does this problematic also on SPU?
No because extraction from a vector is a simple rotate.
> Does SPU has this LHS hazard?
No because there is no store queue on the SPU.
> 
> Another question:
> lwz r0,-20(r1)   <---- LHS hazard
> stw r0,lo16(_e)(r2)
> 
> The problem here is these 2 insns, right?

No, it is the load after the store to the same address:
        stvewx v1,0,r2
        lwz r0,-20(r1)   <---- LHS hazard

r2 at this point is equal to -20(r1).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32826