[Bug tree-optimization/37239] New: peeling last iteration of a <= loop

[bonzini at gnu dot org]

If the condition of the loop is tested within the loop with == or !=, it may be
beneficial to peel off the final iteration of the loop by changing the
condition to <.

This happens in the attached benchmark's heapsort function where

    while ((maxIdx += maxIdx) <= last) { 
        if (maxIdx != last && numbers[maxIdx] < numbers[maxIdx + 1]) maxIdx++;
        if (tmp >= numbers[maxIdx]) break;
        numbers[top] = numbers[maxIdx];
        top = maxIdx;
    }

can become

    while ((maxIdx += maxIdx) <= last) { 
        if (numbers[maxIdx] < numbers[maxIdx + 1]) maxIdx++;
        if (tmp >= numbers[maxIdx]) break;
        numbers[top] = numbers[maxIdx];
        top = maxIdx;
    }
    if (maxIdx == last && tmp < numbers[maxIdx]) {
        numbers[top] = numbers[maxIdx];
        top = maxIdx;
    }

enabling in turn if-conversion of the first branch.

Performance of the benchmark is (-O3)

    basic               2.990
    peeling only        2.730
    if-conversion only  2.290
    peel+if-convert     2.010   (faster than quicksort!!)

ICC does this optimization.


-- 
           Summary: peeling last iteration of a <= loop
           Product: gcc
           Version: 4.3.2
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: bonzini at gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37239