Jaguar gcc3 has some 30-40 optimizations that are not in standard
3.1/3.2, both
generic and PowerPC-specific, so I think you can expect to see different
performance numbers (hopefully better ones :-) ).
Stan
Jonas Maebe wrote:
> Hello,
>
> The program at the end of this mail is 10% slower when compiled with
> GCC 3.x (both 3.0.4 and 3.1.1 tested, using -O6, on a PPC604e/180MHz)
> than when compiled with 2.95. The reasons are
>
> a) several superfluous "fmr" instructions are generated by gcc3 (which
> aren't generated by gcc 2.95):
>
> lfs 0,200(11)
> lfs 1,-200(11)
> lfs 10,4(11)
> lfs 8,-4(11)
> fadds 9,0,1
> fadds 7,9,10
> fadds 6,7,8
> >> fmr 13,6
> fmul 5,13,11
> frsp 4,5
> stfs 4,4(6)
>
>
> b) the static prediction bit for the loop-branches are wrong (they are
> set to be predicted "not taken). Changing this doesn't really improve
> performance on my test machine, but it should still be fixed of course
> (note that I've only verified this with gcc 3.0.4, I don't have access
> to 3.1.1 currently):
>
> .L46:
> li 5,1
> li 4,50
> .L45:
> li 8,8
> slwi 6,4,2
> mtctr 8
> add 7,6,0
> addi 6,7,4
> .L65:
> lfs 10,200(6)
>
> ...
>
> cmpwi 0,5,48
> ble- 0,.L45
> addi 3,3,1
> cmpwi 0,3,2999
> ble- 0,.L46
>
>
> From the dialect, you can see these tests were performed under Linux.
> I don't have access a Mac OS X machine with gcc 3.x (only with 2.95.2,
> I'm still under 10.1.5 and I don't want to install the beta tools
> since I'll soon upgrade to 10.2), so it's possible neither of these
> problems shows up in the Apple version of gcc3.
>
>
> Jonas
>
>
> -------------------------------------
> /*
> ************************************************************************
> * laplace.c: Solution of Laplace equation with finite differences *
> * *
> * taken from: "Projects in Computational Physics" by Landau and Paez *
> * copyrighted by John Wiley and Sons, New York *
> * *
> * written by: students in PH465/565, Computational Physics, *
> * at Oregon State University *
> * code copyrighted by RH Landau *
> * supported by: US National Science Foundation, Northwest Alliance *
> * for Computational Science and Engineering (NACSE), *
> * US Department of Energy *
> * *
> * UNIX (DEC OSF, IBM AIX): cc laplace.c *
> * *
> * comment: Output data is saved in 3D grid format used by gnuplot *
> ************************************************************************
> */
> #include <stdio.h>
>
> #define max 50 /* number of grid points */
>
> main()
> {
> float x, p[max][max];
> int i, j, iter, y;
> #if 0
> FILE *output; /* save data in laplace.dat */
> output = fopen("laplace.dat","w");
> #endif
>
> for(i=0; i<max; i++) /* clear the array */
> {
> for (j=0; j<max; j++) p[i][j] = 0;
> }
>
> for(i=0; i<max; i++) p[i][0] = 100.0; /* p[i][0] = 100 V
> */
>
> for(iter=0; iter<3000; iter++) /* iterations */
> {
> for(i=1; i<(max-1); i++) /* x-direction */
> {
> for(j=1; j<(max-1); j++) /* y-direction */
> {
> p[i][j] = 0.25*(p[i+1][j]+p[i-1][j]+p[i][j+1]+p[i][j-1]);
> }
> }
> }
>
> #if 0
> printf("Start writing data ...\n");
> for (i=0; i<max ; i++) /* write data gnuplot 3D format */
> {
> for (j=0; j<max; j++)
> {
> fprintf(output, "%f\n",p[i][j]);
> }
> fprintf(output, "\n"); /* empty line for gnuplot */
> }
> printf("data stored in laplace.dat\n");
> fclose(output);
> #endif
> }
>
> -------------------------------------
>
>
