[Bug tree-optimization/36010] Loop interchange not performed

[Bug tree-optimization/36010] Loop interchange not performed

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is still true, you need still need  -fno-tree-pre -fno-tree-loop-im to get
the loop to interchanged.

-O2 -fno-tree-pre -fno-tree-loop-im -floop-interchange works while just -O2
-floop-interchange does not.

[Bug tree-optimization/36010] Loop interchange not performed

[aagarwa at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

Ajit Kumar Agarwal <aagarwa at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aagarwa at gcc dot gnu.org

--- Comment #5 from Ajit Kumar Agarwal <aagarwa at gcc dot gnu.org> ---
Use the following flags-fassociative-math -fno-signed-zeros -fno-trapping-math
to make loop-interchange work.

Following code in gcc/gimple-loop-interchange.cc
@@ -514,8 +514,8 @@ loop_cand::analyze_iloop_reduction_var (tree var)
       if (! (associative_tree_code (code)
             || (code == MINUS_EXPR
                 && use_p->use == gimple_assign_rhs1_ptr (ass)))
         || (FLOAT_TYPE_P (TREE_TYPE (var))
           && ! flag_associative_math))
         return false;
     }
   else

Because of the flag_associative_math conditions at line no: 514 it returns
false and loop interchange doesn't work. Using the above flags make
flag_associative_math as true and loop interchange works.

[Bug tree-optimization/36010] Loop interchange not performed

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rguenth at gcc dot gnu.org

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, the condition seems to be superfluous - we don't (in this case?) associate
when moving the load/store back into the loop or when we perform the
interchange?

In fact the comment before the block suggests that.

But yes, with -O2 -floop-interchange -ffast-math we now interchange the
original testcase (and the case from bwaves).  This was fixed with
r8-5164-gfbdec14e80e939 which also introduced this check for associativity.

[Bug tree-optimization/36010] New: Loop interchange not performed

[rguenth at gcc dot gnu dot org]

For the testcase

double a[16][64], y[64], x[16];
void foo(void)
{
  int i, j;
  for (j = 0; j < 64; ++j)
    for (i = 0; i < 16; ++i)
      y[j] = y[j] + a[i][j] * x[i];
}

with -O2 -fno-tree-pre -fno-tree-loop-im -ftree-loop-linear loop interchange
is performed but with PRE or lim moving the load (and the store) from/to
y[j] out of the innermost loop the interchange is no longer performed because

      before = gcc_loopnest_to_lambda_loopnest (loop_nest, &oldivs,
                                                &invariants, &lambda_obstack);

returns NULL.


-- 
           Summary: Loop interchange not performed
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

[Bug tree-optimization/36010] Loop interchange not performed

[rguenth at gcc dot gnu dot org]

------- Comment #1 from rguenth at gcc dot gnu dot org  2008-04-22 15:19 -------
perfect_nestify needs to be enhanced to handle this I guess.  In fact we cannot
distribute these loops but need to undo what PRE/lim do.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

[Bug tree-optimization/36010] Loop interchange not performed

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2008-04-22 15:22 -------
Bwaves from SPEC2006 has

...
               do l=1,nb
                  y(l,i,j,k)=0.0d0
                  do m=1,nb
                     y(l,i,j,k)=y(l,i,j,k)+
     1               a(l,m,i,j,k)*x(m,i,j,k)+
     2               axp(l,m,i,j,k)*x(m,ip1,j,k)+
     3               ayp(l,m,i,j,k)*x(m,i,jp1,k)+
     4               azp(l,m,i,j,k)*x(m,i,j,kp1)+
     5               axm(l,m,i,j,k)*x(m,im1,j,k)+
     6               aym(l,m,i,j,k)*x(m,i,jm1,k)+
     7               azm(l,m,i,j,k)*x(m,i,j,km1)
                  enddo
               enddo

which we don't distribute either (even with PRE/lim turned off).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

[Bug tree-optimization/36010] Loop interchange not performed

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
           Severity|normal                      |enhancement


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010

[Bug tree-optimization/36010] Loop interchange not performed

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2008-12-28 06:33 -------
Confirmed.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-12-28 06:33:38
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36010