[Bug tree-optimization/19347] New: Invariant load not moved out of loop

[Bug tree-optimization/19347] New: Invariant load not moved out of loop

[irar at il dot ibm dot com]

In mesa benchmark (osmesa.c:678) 

         GLuint i, n, *ptr4;
         n = osmesa->rowlength * osmesa->height;
         ptr4 = (GLuint *) osmesa->buffer;
         for (i=0;i<n;i++) {
            *ptr4++ = osmesa->clearpixel;
         }   

The load of osmesa->clearpixel is not taken outside the loop by LIM because of 
aliasing limitations. This in turn also prevents vectorization.

In this particular case we can actually get the load moved out of the loop even 
without resolving the aliasing issue (which requires whole-program), on 
account that even if the store aliases the load, it will not alter the value 
loaded (because we store the same value that we loaded).

I'm looking into this in the context of the vectorizer.

-- 
           Summary: Invariant load not moved out of loop
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: irar at il dot ibm dot com
        ReportedBy: irar at il dot ibm dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: powerpc-apple-darwin7.0.0
  GCC host triplet: powerpc-apple-darwin7.0.0
GCC target triplet: powerpc-apple-darwin7.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

[Bug tree-optimization/19347] Invariant load not moved out of loop

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-01-09 15:55 -------
Confirmed but add a full testcase here, thanks.

Actually I don't think it would require whole program analysis to do this in LIM (or LICM), we just need 
more information (which might be done on the structure-aliasing branch).

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |missed-optimization
   Last reconfirmed|0000-00-00 00:00:00         |2005-01-09 15:55:15
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

[Bug tree-optimization/19347] Invariant load not moved out of loop

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
   Last reconfirmed|2005-01-09 15:55:15         |2005-04-11 06:09:59
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

[Bug tree-optimization/19347] Invariant load not moved out of loop

[phosit at autistici dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

--- Comment #10 from Phosit <phosit at autistici dot org> ---
The analysis in my previous comment is wrong.
I don't know why there is no alias-check at -O2.
Also the loop _is_ removed at -O3.

[Bug tree-optimization/19347] Invariant load not moved out of loop

[phosit at autistici dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

Phosit <phosit at autistici dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |phosit at autistici dot org

--- Comment #9 from Phosit <phosit at autistici dot org> ---
When there are few iterations (likely when it is unroled) there is no
alias-check:

void fill(int* data, int& value)
{
    for(unsigned long i{0}; i != 4; ++i)
        data[i] = value;
}

--- at -O2 results in:
fill(int*, int&):
        movd    xmm1, DWORD PTR [rsi]
        pshufd  xmm0, xmm1, 0xe0
        movq    QWORD PTR [rdi], xmm0
        movd    xmm0, DWORD PTR [rsi]
        pshufd  xmm0, xmm0, 0xe0
        movq    QWORD PTR [rdi+8], xmm0
        ret

---
At -O3 the loop isn't removed and (thus) there is an alias-check.

[Bug tree-optimization/19347] Invariant load not moved out of loop

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2009-04-03 12:06 -------
The store to *ptr aliases the load from osmesa->clearpixel - there
is no (easy) way to otherwise prove that it is not (at least with the
testcase).
We do not see where osmesa->buffer points to (it may point to
&osmesa->clearpixel).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347

[Bug tree-optimization/19347] Invariant load not moved out of loop

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2005-10-24 00:20 -------
This is the normal, we need offset/varaible aliasing.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dberlin at gcc dot gnu dot
                   |                            |org
           Keywords|                            |alias
   Last reconfirmed|2005-05-04 18:15:55         |2005-10-24 00:20:32
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19347