Re: Timing information for CFG manipulations

[Zack Weinberg <zack@codesourcery.com>]

On Sat, Oct 13, 2001 at 10:32:06PM -0500, Brad Lucier wrote:
> Jan:
> 
> Here is timing information for today's CVS version of 3.1 configured
> with --enable-languages='c' --enable-checking=no and built with
> 'make bootstrap BOOT_CFLAGS="-O2 -g -pg" BOOT_LDFLAGS="-O2 -g -pg"'
> when called on the file at
> 
> http://www.math.purdue.edu/~lucier/_num.i.gz
...


>  44.47    348.50   348.50  7450804     0.05     0.05  remove_edge
>  32.36    602.16   253.66     9556    26.54    26.54  sbitmap_vector_alloc

...


>                 0.08    0.00       3/9556        flow_loops_find [59]
>                 0.16    0.00       6/9556        estimate_probability [33]
>                 0.16    0.00       6/9556        if_convert [25]
>                 0.32    0.00      12/9556        optimize_mode_switching [19]
>                 1.43    0.00      54/9556        pre_edge_lcm [24]
>               251.51    0.00    9475/9556        make_edges [10]
> [13]    32.4  253.66    0.00    9556         sbitmap_vector_alloc [13]

Two things:

- Can you repeat this test with cc1 linked against a profile-
instrumented libc, and libiberty re-built with CFLAGS="-O2 -g -pg"?
I'm suspicious that lots of time is being spent inside malloc.

- Have you tried Daniel Berlin's ebitmap patches?

zw