[Bug tree-optimization/18687] New: [4.0 Regression] ~50% compile time regression

[Bug tree-optimization/18687] New: [4.0 Regression] ~50% compile time regression

[belyshev at lubercy dot com]

These two files taken from linux kernel (hashes.c) and zlib (infcodes.c)

hashes.c:

        3.4.4   4.0.0   delta

-O0     0.045   0.049    9%
-O1     0.067   0.144  115%
-O2     0.108   0.177   64%

infcodes.c:

        3.4.4   4.0.0   delta

-O0     0.080   0.086    7%
-O1     0.120   0.223   86%
-O2     0.196   0.295   51%

-- 
           Summary: [4.0 Regression] ~50% compile time regression
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Keywords: compile-time-hog
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: belyshev at lubercy dot com
                CC: gcc-bugs at gcc dot gnu dot org
 BugsThisDependsOn: 15678


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-11-27 04:31 -------
Created an attachment (id=7616)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7616&action=view)
3338 bytes


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-11-27 04:32 -------
Created an attachment (id=7617)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7617&action=view)
9535 bytes


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-27 15:47 -------
Hmm, These are really short compile times to begin with.  Almost anything this short could be 
considered just error but then again, the compile time for me on ppc-darwin was about the same time 
as 3.3 as 4.0.0.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-11-27 18:44 -------
Created an attachment (id=7618)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=7618&action=view)
perl script (463 bytes)

Use this script to get more accurate timings. (install
http://search.cpan.org/~jhi/BSD-Resource-1.24/ if you want to get even more
accurate timings).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-30 13:09 -------
Confirmed, how many new builtins do we have between 3.4.x and 4.0.0, this will most likely account for 
the -O0 time.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-11-30 13:09:22
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-30 13:10 -------
In both of my profiles for -O0 and -O1, I see that is_attribute_p is always high up there.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-30 14:32 -------
At -O2 and -O3 also I see is_attribute_p is always taking 2% of the time.
At -O0:
- is_attribute_p -  6%
- decode_reg_name - 2.1%
at -O1:
- is_attribute_p -  2.5%
- decode_reg_name 0.8%
at -O2:
 - is_attribute_p -  1.9%
- decode_reg_name  - 0.5%
at -O3 (this was a bigger sample than the obove ones):
- is_attribute_p -  1.8%
- decode_reg_name - 0.6%


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-30 14:57 -------
I have a fix for decode_reg_name and there is a branch with the fix for is_attribute_p, maybe that 
should be merged in ...

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-30 15:17 -------
The patch to fix decode_reg_name is here: <http://gcc.gnu.org/ml/gcc-patches/2004-11/
msg02683.html>, I don't know how much this buys us but should buys us at -O0 about 2%.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-02 14:06 -------
Can you try again after my patch to varasm.c, I cannot get any accrete results on these tests.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-12-02 15:27 -------
Patch to speedup lookup_attribute here: <http://gcc.gnu.org/ml/gcc-patches/2004-12/
msg00164.html>, this should help but not all 6% at -O0.

Note for -O0, we are slowing down because there are more attributes and more builtins functions for 
4.0.0 compared with 3.4.x.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-12-03 00:24 -------
        3.4.4   4.0.0   delta

hashes.c:
-O0     0.0515  0.0526    2%           
-O1     0.078	0.157   101%
-O2     0.123	0.198    61%

infcodes.c:
-O0     0.0907  0.0933    3%  
-O1     0.137 	0.248    81%
-O2     0.224 	0.328    46%

Average speedup (not only -O0) is 6%.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2004-11-30 13:09:22         |2004-12-03 00:24:07
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at depni dot sinp dot msu dot ru]

------- Additional Comments From belyshev at depni dot sinp dot msu dot ru  2005-02-10 15:55 -------
                3.4.4           4.0.0           delta

hashes100.c:

-O0              3.66            3.81             4%
-O1              6.43           11.47            78%
-O2             11.40           17.89            57%
-O3             12.00           18.35            53%


infcodes100.c:

-O0              7.24            7.66             6%
-O1             12.43           22.24            79%
-O2             21.28           31.54            48%
-O3             22.37           32.02            43%


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|patch                       |
      Known to fail|                            |4.0.0
      Known to work|                            |3.4.4
   Last reconfirmed|2004-12-22 21:50:22         |2005-02-10 15:55:22
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-10 00:51 -------
(In reply to comment #22)
So does someone do more testings again?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-02-10 00:33 -------
Subject: Bug 18687

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rakdver@gcc.gnu.org	2005-02-10 00:32:49

Modified files:
	gcc            : ChangeLog tree-flow.h tree-ssa-loop-ivcanon.c 
	                 tree-ssa-loop-ivopts.c tree-ssa-loop-niter.c 
	                 tree.c tree.h 

Log message:
	PR tree-optimization/18687
	* tree-flow.h (find_loop_niter): Declare.
	* tree-ssa-loop-ivcanon.c (canonicalize_loop_induction_variables):
	Try using scev even for loops with more than one exit.
	* tree-ssa-loop-ivopts.c (struct loop_data): Removed niter field.
	(struct ivopts_data): Added niters field.
	(struct nfe_cache_elt): New.
	(nfe_hash, nfe_eq, niter_for_exit, niter_for_single_dom_exit): New
	functions.
	(tree_ssa_iv_optimize_init): Initialize niters cache.
	(determine_number_of_iterations): Removed.
	(find_induction_variables): Do not call determine_number_of_iterations.
	Access niters for single exit through niter_for_single_dom_exit.
	(add_iv_outer_candidates): Access niters for single exit through
	niter_for_single_dom_exit.
	(may_eliminate_iv): Take data argument.  Use niter_for_exit.  Do not use
	number_of_iterations_cond.
	(iv_period): New function.
	(determine_use_iv_cost_condition): Pass data to may_eliminate_iv.
	(may_replace_final_value): Take data argument.  Use
	niter_for_single_dom_exit.
	(determine_use_iv_cost_outer): Pass data to may_replace_final_value.
	(rewrite_use_compare): Pass data to may_eliminate_iv.
	(rewrite_use_outer): Pass data to may_replace_final_value.
	(free_loop_data): Clean up the niters cache.
	(tree_ssa_iv_optimize_finalize): Free the niters cache.
	(tree_ssa_iv_optimize_loop): Do not call loop_commit_inserts.
	* tree-ssa-loop-niter.c (find_loop_niter): New function.
	(find_loop_niter_by_eval): Use tree_int_cst_lt.
	(num_ending_zeros): Moved to tree.c.
	* tree.h (num_ending_zeros): Declare.
	* tree.c (num_ending_zeros): Moved from tree.c.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7427&r2=2.7428
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-flow.h.diff?cvsroot=gcc&r1=2.80&r2=2.81
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-loop-ivcanon.c.diff?cvsroot=gcc&r1=2.5&r2=2.6
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-loop-ivopts.c.diff?cvsroot=gcc&r1=2.44&r2=2.45
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-ssa-loop-niter.c.diff?cvsroot=gcc&r1=2.18&r2=2.19
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree.c.diff?cvsroot=gcc&r1=1.463&r2=1.464
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree.h.diff?cvsroot=gcc&r1=1.687&r2=1.688



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2005-02-08 16:43 -------
Patch (improving ivopts performance on the testcase by some 40%):

http://gcc.gnu.org/ml/gcc-patches/2005-02/msg00307.html

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2005-02-08 00:29 -------
Out of the time spent in ivopts:

-- this patch saves some 25%:
   http://gcc.gnu.org/ml/gcc-patches/2004-10/msg00136.html

-- another 25% are caused by inefficient use of # of iterations analysis;
   I am thinking about solution

-- yet another 25% are spent in scalar evolutions analysis; I am working on
   speeding this up

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-30 23:42 -------
IVopts takes a lot of time (in several runs): 
 tree iv optimization  :   1.12 ( 9%) usr   0.16 (23%) sys   1.34 (10%) wall 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-30 22:55 -------
The profile for infcodes100.c is essentially the same. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-30 22:53 -------
Profile for hashes100.c on an x86-64 with "-O2 -m32": 
 
CPU: Hammer, speed 1394.98 MHz (estimated) 
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit 
mask of 0x00 (No unit mask) count 4000 
Counted DATA_CACHE_MISSES events (Data cache misses) with a unit mask of 0x00 
(No unit mask) count 1000 
samples  %        samples  %        image name       symbol name 
27235     2.3397  158       0.5518  cc1              constrain_operands 
21936     1.8845  53        0.1851  cc1              record_reg_classes 
21735     1.8672  399       1.3936  cc1              reg_scan_mark_refs 
20966     1.8012  1179      4.1179  libc.so.6        __GI_memset 
20786     1.7857  418       1.4600  cc1              ggc_alloc_stat 
20199     1.7353  168       0.5868  cc1              fold 
18374     1.5785  589       2.0572  libc.so.6        _int_malloc 
16467     1.4147  177       0.6182  cc1              iterative_hash_expr 
15562     1.3369  324       1.1316  cc1              htab_find_slot_with_hash 
15080     1.2955  842       2.9409  cc1              note_stores 
13709     1.1777  78        0.2724  cc1              find_reloads 
13054     1.1214  45        0.1572  cc1              nonzero_bits1 
12297     1.0564  84        0.2934  cc1              mark_used_regs 
11402     0.9795  68        0.2375  cc1              refers_to_regno_p 
11335     0.9738  95        0.3318  cc1              cse_insn 
11088     0.9526  71        0.2480  cc1              ix86_decompose_address 
10401     0.8935  70        0.2445  cc1              for_each_rtx_1 
9731      0.8360  354       1.2364  cc1              count_reg_usage 
9176      0.7883  202       0.7055  cc1              bitmap_set_bit 
8578      0.7369  284       0.9919  cc1              bitmap_bit_p 
8137      0.6990  208       0.7265  cc1              extract_insn 
8019      0.6889  30        0.1048  cc1             
reload_cse_simplify_operands 
7755      0.6662  541       1.8896  cc1              pool_alloc 
7341      0.6307  143       0.4995  cc1              et_splay 
6992      0.6007  319       1.1142  cc1              init_alias_analysis 
 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-01-28 00:55 -------
Did anyone try to profile these already? 
 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |stevenb at suse dot de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687

[Bug tree-optimization/18687] [4.0 Regression] ~50% compile time regression

[belyshev at depni dot sinp dot msu dot ru]

------- Additional Comments From belyshev at depni dot sinp dot msu dot ru  2004-12-22 21:50 -------
Hope these two larger test cases will help profiling.

		3.4.4		4.0.0		delta

hashes100.c:

-O0		 3.7		 3.8		  3%
-O1		 6.4		14.5		126%
-O2		11.3		18.7		 65%
-O3		11.9		19.0		 60%


infcodes100.c:

-O0		 7.4		 7.6		  3%
-O1		12.6		25.4		101%
-O2		21.4		33.8		 58%
-O3		22.4		34.7		 55%


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2004-12-03 00:24:07         |2004-12-22 21:50:22
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18687