some wall clock time measurements (with CpCmpu enabled)

[Salva Peiró]

these are some wall clock time measurements I've gathered
after being able to enable the mpu & caches on the DS port,
I would have prefered to use time(1) or another profiling tool
but I can't get time(1) to print floating point numbers correctly;
I'm looking into the time(1) issue [1].

Some time measurements without CpCmpu and caches
% date # boot .nds file
Fri Jun 6 21:13:06 CEST 2008
% date # dsinit.b prompt: ``root from (kernel)''
Fri Jun 6 21:13:15 CEST 2008
% date # wm/wm finished loading $home/lib/wmsetup
Fri Jun 6 21:14:08 CEST 2008

total: 62 secs

Some time measurements with CpCmpu and caches working (jit disabled)
% date # boot .nds file
Fri Jun 6 21:00:47 CEST 2008
% date # dsinit.b prompt: ``root from (kernel)''
Fri Jun 6 21:00:50 CEST 2008
% date # wm/wm finished loading $home/lib/wmsetup
Fri Jun 6 21:00:57 CEST 2008

total: 10 secs

So it has become 6x times quicker.

The thing is that with the mpu enabled,
interacting with wm/{wm,sh,keyboard} and other programs is a lot faster,
and the touchscreen is a lot more responsive making the whole thing a
lot more usable.
I'd say (from my perception) that it's similar in speed launching emu
on a standard pc (2 or 3 years old).

That's good news to me because I was worried about the performance
one could achieve running Inferno on the DS, needless to say now all
my worries about that have dissapeared.

[1] '`status update (rev 142) : arm7 print(), wifi, and time''
http://groups.google.com/group/inferno-ds/browse_thread/thread/acedc438fae01829

--
salva

[Salva Peiró]

i want to test the changes a bit, so i'm not going to push them to the repo
in the mean time i've attached them to this mail, for people to try them.
I was able to confirm that they worked for r4ds, gnm and ttds.

--
salva

[Caerwyn Jones]

>> The thing is that with the mpu enabled,
>> interacting with wm/{wm,sh,keyboard} and other programs is a lot faster,
>> and the touchscreen is a lot more responsive making the whole thing a
>> lot more usable.
>> I'd say (from my perception) that it's similar in speed launching emu
>> on a standard pc (2 or 3 years old).

Thanks Salva for persevering with this. These changes really are
impressive. Inferno is now quite snappy on the DS. Once we (and by we
I mean you :) have solved the card storage issue inferno on the DS
will be an amazing little platform.

Caerwyn

[Salva Peiró]

> i want to test the changes a bit, so i'm not going to push them to the repo
> in the mean time i've attached them to this mail, for people to try them.
> I was able to confirm that they worked for r4ds, gnm and ttds.

Since the last week the above patch with the mpu changes it's into the
repo (rev 149),
and devnds.c:/^ndsinit has been modified (rev 150) as it was causing
problems when
booting from slot-2 (gba) devices, like m3cf.

--
salva

[Salva Peiró]

On Sun, Jun 8, 2008 at 4:17 PM, Caerwyn Jones <caer...@gmail.com> wrote:
>
> Thanks Salva for persevering with this. These changes really are
> impressive. Inferno is now quite snappy on the DS. Once we (and by we
> I mean you :) have solved the card storage issue inferno on the DS
> will be an amazing little platform.
>
> Caerwyn
>

i've been able to read the first sector (mbr) of a sd card
using a r4ds by compiling the r4tf.c io storage functions.
For details about this see devdldi.c:/^dldiinit and devdldi.c:/^mbrtest [r159],
next step is mounting a partition using dossrv/kfs.

I've made things in a way to make easy to add support for other cards, like GnM.

[GnM] http://dldi.drunkencoders.com/index.php?title=Datel_Games_%27n%27_Music_(microSD)
[r159] http://code.google.com/p/inferno-ds/source/detail?r=159
--
salva