% cat bubblesort.c
#include <stdio.h>
#include <stdlib.h>

#define COUNT 1000

void swap (int *Array, int i, int j) {
        int temp;
        temp = Array[i];
        Array[i] = Array[j];
        Array[j] = temp;
}

void bubblesort(int *Array) {
        int i, j;

        for (i = 0; i < COUNT; ++i) {
                for (j = (i + 1); j < COUNT; ++j) {
                        if (Array[i] > Array[j]) {
                                swap (Array, i, j);
                        }
                }
        }
}

int main() {
        int i, *Array;

        Array = (int *) malloc (sizeof (int) * COUNT);

        for (i = COUNT; i > 0; --i)
                Array[COUNT - i] = i;

        bubblesort(Array);

        for (i = 0; i < COUNT; ++i)
                printf("\nArray[%d] = %d", i, Array[i]);

        return (0);
}

% cc -o bubblesort -xO4 -xprofile=collect bubblesort.c

% ./bubblesort
Array[0] = 1
Array[1] = 2
..
..
Array[9998] = 9999
Array[9999] = 10000

% ls -dF *.profile
bubblesort.profile/

%ls -dF /tmp/*.profile
No match

% setenv SUN_PROFDATA_DIR /tmp

% ./bubblesort
Array[0] = 1
Array[1] = 2
..
..

% ls -dF /tmp/*.profile
/tmp/bubblesort.profile/

1. Code layout
   • Place code in a way that the frequently executed code in a routine is grouped together
     eg., Block chaining, block ordering, block splitting, register allocation
   • Place code to improve instruction cache (I$) utilization, instruction prefetching etc.,
2. Loop unrolling, loop unswitching
   
3. Inline routines that are frequently called, to both remove the cost of calling the routine, and potentially to enable further optimisation of the inlined code

% cc -o bubblesort -xO4 -xprofile=use:bubblesort.profile bubblesort.c
% ./bubblesort
Array[0] = 1
Array[1] = 2
..
..

• 
  % {f90,CC} -xprofile=use:<feedback dir> -Qoption cg -assembly,-Qcg-V
  % cc -xprofile=use:<feedback dir> -Wc,-assembly,-Qcg-V

  
  The -assembly option will generate a .s file with the same basename and dirname as the object file (e.g., bubblesort.o will be accompanied by bubblesort.s in the same directory). The -Qcg-V option adds more information as assembler comments to the generated .s file. If -xprofile=use has been specified, this information includes execution counts derived from the <profile data directory>

• eg.,

  % cc -xprofile=use:bubblesort -Wc,-assembly,-Qcg-V -xO4 bubblesort.c
  ...
  ...
  !   13                              !void bubblesort(int *Array) {
  !   14                              !   int i, j;
  !   16                              !   for (i = 0; i < COUNT; ++i) {
  
  !
  ! SUBROUTINE bubblesort
  !
  ! OFFSET    SOURCE LINE LABEL   INSTRUCTION     (ISSUE TIME)    (COMPLETION TIME)
  
                                          .global bubblesort
  
  
                          bubblesort:             /* frequency 1.0 confidence 1.0 */
  /* 000000         16 ( 0  1) */         sethi   %hi(0x2400),%o5 ! const ! ssar:9216
  /* 0x0004            ( 0  1) */         or      %g0,0,%o2
  /* 0x0008            ( 1  2) */         add     %o5,783,%o3 ! const ! ssar:9999 ! urw: 32
  /* 0x000c            ( 1  2) */         add     %o5,784,%o1 ! const ! ssar:10000 ! urw: 32
  
  !   17                              !           for (j = (i + 1); j < COUNT; ++j) {
  
  /* 0x0010         17 ( 2  3) */         or      %g0,1,%o5
  
  ! Registers live out of bubblesort:
  ! o0 o1 o2 o3 o5 sp o7 fp gsr
  !
  
  ! predecessor blocks : bubblesort .L900000205
  
                          .L900000206:            /* frequency 10000.0 confidence 1.0 */
  /* 0x0014         17 ( 0  1) */         cmp     %o5,%o1 ! urw: 32
  /* 0x0018            ( 0  1) */         bge,pn  %icc,.L77000113 ! tprob=0.00 ! urw: 32
  /* 0x001c            ( 0  1) */         add     %o0,4,%g3 ! hoisted ! urw: 32
  
  ! Registers live out of .L900000206:
  ! g3 o0 o1 o2 o3 o5 sp o7 fp gsr
  !
  ...
  ...