[LLVMbugs] [Bug 864] NEW: LSR should emit expressions in proper order for CSE to happen

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=864

           Summary: LSR should emit expressions in proper order for CSE to
                    happen
           Product: libraries
           Version: 1.0
          Platform: All
        OS/Version: All
            Status: NEW
          Severity: enhancement
          Priority: P2
         Component: Scalar Optimizations
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sabre at nondot.org
                CC: natebegeman at mac.com


Consider:

int foo(int X, int Y, int Z, int* P1, int* P2, int* P3, int* P4) {
  int i;
  for (i = 0; i != 1000; ++i) {
    int t = *P3;
    int t2 = i*4+t;
    *P1 = t2+X;
    *P2 = t2+Y;
  }
}

LSR breaks down the two expression uses (the sstore to P1/P2) into stores two two different loop 
invariant bases (X and Y), and then inserts code for each base.  Unfortunately, it inserts the expressions 
with this order of operations:

    *P1 = (i*4+X)+t;
    *P2 = (i*4+Y)+t;

Because of this, it is implicitly un-CSE-ing the add "i*4+t", causing the output machine code two have 
two adds for this op.

-Chris



------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.