Re: Timing information for CFG manipulations

[Jan Hubicka <jh at suse dot cz>]

> dino01% /u/lucier/local/gcc-3.1/lib/gcc-lib/i686-pc-linux-gnu/3.1/cc1 -fpic -fomit-frame-pointer -O1 -fno-math-errno -fno-strict-aliasing -mcpu=athlon -march=athlon _num.i
>  __sgn __sgnf __sgnl atan2 atan2f atan2l __atan2l fmod fmodf fmodl sqrt sqrtf sqrtl __sqrtl fabs fabsf fabsl __fabsl atan atanf atanl __sgn1l floor floorf floorl ceil ceilf ceill ldexp log1p log1pf log1pl asinh asinhf asinhl acosh acoshf acoshl atanh atanhf atanhl hypot hypotf hypotl logb logbf logbl drem dremf dreml __finite ___H__20___num {GC 25431k -> 7824k} {GC 10944k -> 7883k} {GC 10373k -> 7769k} {GC 13951k -> 8584k} {GC 14266k -> 9195k} ___init_proc {GC 12104k -> 9290k} ____20___num
> Execution times (seconds)
>  garbage collection    :   0.67 ( 0%) usr   0.00 ( 0%) sys   0.72 ( 0%) wall
>  cfg construction      :   6.37 ( 4%) usr   0.26 ( 6%) sys   6.66 ( 4%) wall
>  cfg cleanup           :  33.59 (21%) usr   0.01 ( 0%) sys  33.69 (20%) wall

I do have patch to cfg_cleanup speedup in the bootstrap now.
Esentially there is quite a lot to squeze out by caching forwarder_block_p.
I also can reorganize try_forward_edges to try forward incomming edges of forwarder block
instead of outgoing edges of each block if the first change won't be enought.

>  preprocessing         :   0.17 ( 0%) usr   0.11 ( 2%) sys   0.44 ( 0%) wall
>  lexical analysis      :   0.29 ( 0%) usr   0.22 ( 5%) sys   0.53 ( 0%) wall
>  parser                :   1.19 ( 1%) usr   0.15 ( 3%) sys   1.28 ( 1%) wall
>  varconst              :   0.07 ( 0%) usr   0.00 ( 0%) sys   0.03 ( 0%) wall
>  jump                  :   0.63 ( 0%) usr   0.00 ( 0%) sys   0.66 ( 0%) wall
>  CSE                   :   0.72 ( 0%) usr   0.00 ( 0%) sys   0.69 ( 0%) wall
>  global CSE            :  32.87 (21%) usr   0.57 (13%) sys  33.53 (20%) wall

This is the mode switching code.  Can someone familiar with the code more than I do
comment whether there are trivial ways to speed it up?


Honza