This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: big slowdown in egcs-1.1.2->gcc-2.95 on alpha
- To: amylaar@cygnus.co.uk (Joern Rennecke)
- Subject: Re: big slowdown in egcs-1.1.2->gcc-2.95 on alpha
- From: Brad Lucier <lucier@math.purdue.edu>
- Date: Fri, 6 Aug 1999 23:27:32 -0500 (EST)
- Cc: lucier@math.purdue.edu (Brad Lucier), amylaar@cygnus.co.uk, gcc@gcc.gnu.org, gcc-bugs@gcc.gnu.org, staff@math.purdue.edu, hosking@cs.purdue.edu, wilker@math.purdue.edu, bernds@cygnus.com
>
> > The biggest time sink seems to be the quadratic algorithm in prune_preferences
> > in global.c.
>
> Hmm, I wonder if the CONFLICT_P tests take a lot of time, or if the rest
> of the operations in the inner loop cost more.
>
> I wonder if this works:
>
Sorry, no, it made little difference:
Flat profile:
Each sample counts as 0.000976562 seconds.
% cumulative self self total
time seconds seconds calls ms/call ms/call name
35.63 55.32 55.32 16 3457.52 3458.44 prune_preferences
12.51 74.75 19.43 391960848 0.00 0.00 bitmap_bit_p
7.40 86.24 11.49 202789 0.06 0.06 record_one_conflict
7.20 97.41 11.17 24 465.58 1218.03 build_insn_chain
1.55 99.83 2.41 8 301.76 19347.54 yyparse
1.49 102.13 2.31 27806760 0.00 0.00 count_pseudo
Here are some more details about global_alloc, prune_preferences,
and build_insn_chain.
-----------------------------------------------
0.08 99.53 24/24 rest_of_compilation [5]
[6] 64.1 0.08 99.53 24 global_alloc [6]
55.32 0.01 16/16 prune_preferences [7]
11.17 18.06 24/24 build_insn_chain [8]
0.24 10.63 24/24 reload [13]
0.12 2.22 16/16 global_conflicts [33]
1.66 0.00 1436/1436 find_reg [39]
0.08 0.01 16/16 expand_preferences [256]
0.00 0.00 24/220059 xmalloc [479]
0.00 0.00 48/5488 get_insns [995]
-----------------------------------------------
55.32 0.01 16/16 global_alloc [6]
[7] 35.6 55.32 0.01 16 prune_preferences [7]
0.01 0.00 78844/334068 reg_preferred_class [306]
-----------------------------------------------
11.17 18.06 24/24 global_alloc [6]
[8] 18.8 11.17 18.06 24 build_insn_chain [8]
17.40 0.00 350981800/391960848 bitmap_bit_p [9]
0.02 0.43 174187/5402847 note_stores [10]
0.08 0.01 194131/194131 new_insn_chain [263]
0.05 0.00 388262/574314 bitmap_copy [293]
0.04 0.00 574446/16720630 bitmap_set_bit [48]
0.01 0.02 88014/88014 reg_dies [385]
0.00 0.00 34596/3151441 bitmap_clear [204]
0.00 0.00 24/1004613 bitmap_initialize [401]
-----------------------------------------------
Brad Lucier lucier@math.purdue.edu