Compilation performance comparison of 3.5.0 and TreeSSA trees onMICO sources as requested in: [tree-ssa] Merge status 2004-05-03

[Karel Gardas <kgardas at objectsecurity dot com>]

On Mon, 3 May 2004, Diego Novillo wrote:

>       * POOMA, DLV and MICO compile times and memory consumption within
>         1-5% of mainline.
>
>         NOT SURE.  I *think* we are within range, but I don't have
>         recent figures.  Could the application owners check that I'm not
>         babbling nonsense?

MICO doesn't look good. I've compared todays source trees:

gcc version 3.5.0 20040504 (experimental)
gcc version 3.5-tree-ssa 20040504 (merged 20040428)

and results look (again the same benchmark which I have posted several
times here, i.e. compilation of MICO orb directory):

File		350-O0	TSSA-O0	Delta%	350-O2	TSSA-O2	Delta%

os-unix.cc	4.12	5.02	-17.93	4.58	5.52	-17.03
dii.cc		12.66	14.94	-15.26	16.35	19.85	-17.63
typecode.cc	8.97	11.13	-19.41	31.6	30.02	5.26
any.cc		6.84	7.8	-12.31	12.64	14	-9.71
codec.cc	5.86	6.7	-12.54	9.35	11	-15
buffer.cc	3.34	3.84	-13.02	3.71	4.16	-10.82
context.cc	3.51	4.02	-12.69	4.19	4.76	-11.97
except.cc	4.38	5.3	-17.36	5.9	7.11	-17.02
dispatch.cc	4.44	5.42	-18.08	4.95	5.99	-17.36
string.cc	3.35	3.78	-11.38	3.48	3.89	-10.54
object.cc	4.65	5.78	-19.55	7.09	8.23	-13.85
address.cc	4.9	5.97	-17.92	7.23	7.36	-1.77
ior.cc		12.31	14.67	-16.09	16.94	18.62	-9.02
orb.cc		16.62	19.87	-16.36	35.94	35.72	0.62
boa.cc		9.13	10.81	-15.54	14.28	15.73	-9.22
dsi.cc		10.19	12.82	-20.51	12.56	14.69	-14.5
transport.cc	4.08	5.04	-19.05	4.51	5.41	-16.64
t..port/tcp.cc	4.01	4.91	-18.33	4.46	5.34	-16.48
t..port/udp.cc	4.11	5.02	-18.13	4.68	5.57	-15.98
t..port/unix.cc	4.09	4.92	-16.87	4.4	5.26	-16.35
iop.cc		16.04	19.47	-17.62	28.61	33.64	-14.95
util.cc		5.95	6.93	-14.14	10.04	11.03	-8.98
basic_seq.cc	3.77	4.26	-11.5	3.88	4.47	-13.2
fast_array.cc	3.87	4.68	-17.31	3.91	4.75	-17.68
ssl.cc		9.24	10.84	-14.76	9.02	10.49	-14.01
fixed.cc	3.75	4.3	-12.79	4.28	4.92	-13.01
intercept.cc	10.16	12.75	-20.31	12.01	14.53	-17.34
codeset.cc	5.92	6.85	-13.58	10.14	11.16	-9.14
queue.cc	4.34	5.3	-18.11	4.71	5.73	-17.8
static.cc	20.09	24.32	-17.39	28.63	35.84	-20.12
current.cc	8.84	10.56	-16.29	8.66	10.17	-14.85
policy_impl.cc	12.62	15.41	-18.11	14.97	18.21	-17.79
service_info.cc	8.74	10.43	-16.2	8.54	10.09	-15.36
ioptypes.cc	10.51	13.31	-21.04	13.71	15.93	-13.94
ssliop.cc	8.97	10.67	-15.93	8.77	10.27	-14.61
value.cc	10.54	13.27	-20.57	12.52	14.19	-11.77
valuetype.cc	9.81	11.85	-17.22	11.19	13.75	-18.62
v..type_impl.cc	12.45	14.92	-16.55	13.56	17.02	-20.33
dynany_impl.cc	9.98	13.87	-28.05	23.52	28.91	-18.64
policy2.cc	8.95	10.72	-16.51	9.01	10.6	-15
tckind.cc	8.76	10.49	-16.49	8.56	10.09	-15.16
orb_excepts.cc	8.93	10.63	-15.99	8.87	10.5	-15.52
policy.cc	8.9	10.68	-16.67	8.91	10.38	-14.16
poa.cc		12.86	15.57	-17.41	17.45	21.03	-17.02
poa_base.cc	10.12	12.4	-18.39	11.68	14.33	-18.49
poa_impl.cc	17.29	21.18	-18.37	28.8	32.84	-12.3
dynany.cc	10.15	12.19	-16.74	11.41	14.06	-18.85
uni_base64.cc	0.13	0.15	-13.33	0.26	0.3	-13.33
uni_unicode.cc	0.2	0.25	-20	0.44	0.51	-13.73
uni_fromuni.cc	0.4	0.55	-27.27	1.11	1.43	-22.38
uni_touni.cc	0.43	0.62	-30.65	1.31	1.54	-14.94
except2.cc	6.67	7.71	-13.49	13	14.1	-7.8
pi.cc		10.82	13.7	-21.02	13.59	15.94	-14.74
pi_impl.cc	18.22	23.36	-22	29.91	34.78	-14
typecode_seq.cc	9.21	11.04	-16.58	9.49	11.26	-15.72
timebase.cc	8.82	10.55	-16.4	8.73	10.26	-14.91
ir.cc		45.74	67.7	-32.44	108.48	133.17	-18.54
ir_base.cc	11.14	14.31	-22.15	15.91	20.05	-20.65
imr.cc		14.47	19.19	-24.6	24.54	29.09	-15.64
mtdebug.cc	3.84	4.62	-16.88	3.73	4.52	-17.48

Sum		523.2	649.36	-19.43	774.7	904.11	-14.31


Conclusion: compile time regressions for both -O0 and -O2 compilation.

Cheers,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com