[m-dev.] MLDS back-end performance: a preliminary benchmark

[Fergus Henderson]

I ran the N-queens benchmark (for N = 10,
and finding the first solution only).

I compared three versions: one using the MLDS back-end
to generate code using gcc nested functions, one using
the MLDS back-end to generate standard ANSI C code,
and our current compiler.  The version using gcc nested
functions is more elegant and simpler to generate, but it
is significantly less efficient than the version using
standard ANSI C.

Anyway, the quick summary is that the MLDS back-end
is at very least very competetive with our current back-end,
even for nondeterministic code.  In fact on this benchmark,
looking at elapsed time rather than just user time,
it beats our current `--grade asm_fast.gc -O6' back-end
by more than 40%!

The following times are on hg
(a 500MHz Pentium III running Red Hat Linux 6.0).

Name			Options
----			-------
queens10.hlc.gcc	--high-level-c --gcc-nested-functions --inline-alloc
queens10.hlc		--high-level-c --inline-alloc
queens10		(none)
queens10.hlc.gcc.O6	--high-level-c --gcc-nested-functions -O6
queens10.hlc.O6		--high-level-c -O6
queens10.O6		-O6

User   System Elapsed CPU
time   time   time    usage	Page faults
----   ------ ------- -----	-----------

queens10.hlc.gcc
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.48 100.0%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
0.480u 0.000s 0:00.47 102.1%	0+0k 0+0io 33pf+0w
queens10.hlc.gcc.O6
0.450u 0.000s 0:00.45 100.0%	0+0k 0+0io 32pf+0w
0.450u 0.000s 0:00.44 102.2%	0+0k 0+0io 32pf+0w
0.450u 0.000s 0:00.44 102.2%	0+0k 0+0io 32pf+0w
0.450u 0.000s 0:00.44 102.2%	0+0k 0+0io 32pf+0w
0.440u 0.000s 0:00.44 100.0%	0+0k 0+0io 32pf+0w
0.460u 0.000s 0:00.46 100.0%	0+0k 0+0io 32pf+0w
0.450u 0.000s 0:00.44 102.2%	0+0k 0+0io 32pf+0w
0.450u 0.000s 0:00.44 102.2%	0+0k 0+0io 32pf+0w
queens10.hlc
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.15 100.0%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.15 100.0%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 32pf+0w
queens10.hlc.O6
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
0.140u 0.000s 0:00.14 100.0%	0+0k 0+0io 34pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
0.140u 0.000s 0:00.14 100.0%	0+0k 0+0io 34pf+0w
0.140u 0.010s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
0.150u 0.000s 0:00.14 107.1%	0+0k 0+0io 34pf+0w
queens10
0.200u 0.050s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.230u 0.020s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.190u 0.060s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.230u 0.020s 0:00.25 100.0%	0+0k 0+0io 52pf+0w
0.230u 0.020s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.210u 0.040s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.190u 0.060s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
0.240u 0.010s 0:00.24 104.1%	0+0k 0+0io 52pf+0w
queens10.O6
0.180u 0.030s 0:00.20 105.0%	0+0k 0+0io 51pf+0w
0.140u 0.060s 0:00.20 100.0%	0+0k 0+0io 51pf+0w
0.150u 0.060s 0:00.20 105.0%	0+0k 0+0io 51pf+0w
0.170u 0.040s 0:00.21 100.0%	0+0k 0+0io 51pf+0w
0.180u 0.030s 0:00.20 105.0%	0+0k 0+0io 51pf+0w
0.170u 0.040s 0:00.20 105.0%	0+0k 0+0io 51pf+0w
0.170u 0.040s 0:00.21 100.0%	0+0k 0+0io 51pf+0w
0.190u 0.020s 0:00.20 105.0%	0+0k 0+0io 51pf+0w
-- 
Fergus Henderson <fjh at cs.mu.oz.au>  |  "I have always known that the pursuit
WWW: <http://www.cs.mu.oz.au/~fjh>  |  of excellence is a lethal habit"
PGP: finger fjh at 128.250.37.3        |     -- the last words of T. S. Garp.
--------------------------------------------------------------------------
mercury-developers mailing list
Post messages to:       mercury-developers at cs.mu.oz.au
Administrative Queries: owner-mercury-developers at cs.mu.oz.au
Subscriptions:          mercury-developers-request at cs.mu.oz.au
--------------------------------------------------------------------------