Werner Almesberger | 1 Nov 03:31 2011
Picon

Re: M1 with MIDI

I wrote:
> http://downloads.qi-hardware.com/people/werner/m1/demo/MVI_1743.MOV

I've deleted it now because of the non-Free audio track. Today's
http://downloads.qi-hardware.com/people/werner/m1/demo/MVI_1747.MOV
is a worthy substitute.

- Werner
Werner Almesberger | 1 Nov 11:44 2011
Picon

autoVJ ?

I'm wondering if we should perhaps call M1's ability to generate
nice-looking effects on its own an "autoVJ" feature, much like an
autopilot assists pilots and handles boring routine tasks for them,
but doesn't replace them.

For example, for this video, I set up the controls such that the
pattern was near a "tipping point" (*) (setup not shown), then
didn't touch the controls while the sound was playing:

http://downloads.qi-hardware.com/people/werner/m1/demo/MVI_1756.MOV

The audio track is "Emergence vs Accension" by medicisoundsystem,
featuring Snowflake, and licensed under CC-BY:

http://ccmixter.org/files/medicisoundsystem/30478
http://ccmixter.org/content/medicisoundsystem/medicisoundsystem_-_Emergence_vs_Accension.mp3

Sorry for the poor sound quality. Seems that my camera's microphone
doesn't "hear" my speakers very well when I position it close to the
screen.

You may also notice that it's actually the same vocals as in
MVI_1747.MOV but with a different melody. Seems everyone at ccmixter
likes "snowflake" a lot :-)

The patterns are a bit repetitive but should be quite sufficient to
afford the VJ a restroom break or a chat with admiring fans.

(*) The modified "Tornado" patch basically has two modes:

(Continue reading)

Werner Almesberger | 1 Nov 11:54 2011
Picon

Re: bug: M1 hangs after heavy MIDI traffic

S?bastien Bourdeauducq wrote:
> Try GDB:

Thanks ! I tried to attach gdb while in the crashed state but didn't
get a response. Now running under gdb since last evening. So far,
MIDI hasn't slowed down much, but I'm sure the hang will come back.
They always do ;-)

- Werner
Sébastien Bourdeauducq | 1 Nov 11:54 2011

Re: bug: M1 hangs after heavy MIDI traffic

On 11/01/2011 11:54 AM, Werner Almesberger wrote:
> S?bastien Bourdeauducq wrote:
>> Try GDB:
>
> Thanks ! I tried to attach gdb while in the crashed state but didn't
> get a response.

This shouldn't happen. Does your GDB work when you're not in the crashed 
state? All commands (backtrace, continue, break...) are supposed to 
work, all the time. The only requirement is that the SoC bitstream is 
loaded.

Note that if you're not using the flterm pass-through, you may need a 
patch for GDB to send the break command correctly (it's upstream, but 
only in versions 7.3 and newer).

S.
Werner Almesberger | 1 Nov 12:08 2011
Picon

Re: bug: M1 hangs after heavy MIDI traffic

S?bastien Bourdeauducq wrote:
> This shouldn't happen. Does your GDB work when you're not in the
> crashed state?

I tried ^C, continue, bt, and in the non-crashed state,
everything worked perfectly.

> Note that if you're not using the flterm pass-through, you may need
> a patch for GDB to send the break command correctly (it's upstream,
> but only in versions 7.3 and newer).

I'm using gdb 7.2 from the SDK, without flterm. Given that ^C and
"continue" work, I suppose this means that it has the patch ?

SDK, I think this one (in case there are more versions):
http://fidelio.qi-hardware.com/~xiangfu/build-milkymist/milkymist-firmware-07062011-0000/Flickernoise-lm32-rtems-4.11-SDK-for-Linux-x86_64.tar.bz2

- Werner
Sébastien Bourdeauducq | 1 Nov 12:06 2011

Re: bug: M1 hangs after heavy MIDI traffic

On 11/01/2011 12:08 PM, Werner Almesberger wrote:
> I'm using gdb 7.2 from the SDK, without flterm. Given that ^C and
> "continue" work, I suppose this means that it has the patch ?

Yes, the patch is there:
https://github.com/milkymist/scripts/tree/master/compile-lm32-rtems/milkymist-one-patches
Werner Almesberger | 1 Nov 12:24 2011
Picon

bug ? texturing artefacts in corners when rotating

I noticed that, for high values of "rot", I get fairly large
"titles" in the corners where the texture doesn't match the rest:

http://downloads.qi-hardware.com/people/werner/m1/midi/corner-artefact.jpg

When modulating "zoom", I can put them to good use, but they look
a little odd when running with fixed zoom/rot and a "crowded"
screen. Bug or feature ?

- Werner
Sébastien Bourdeauducq | 1 Nov 12:27 2011

Re: bug ? texturing artefacts in corners when rotating

    !!!     "It's not a BUG,
   /o o\  /  it's a FEATURE!"
  (  >  )
   \ - /
   _] [_     (jcooley 1991)

When distorting the current picture, when texture (i.e. source) 
coordinates go off-screen, two things can happen:
1) the coordinates wrap around (i.e. the texture repeats)
2) the coordinates are clamped

Right now, you are seeing behaviour #1. You can go to #2 with this line:
bTexWrap=0
(it also works in per-frame equations, if you want to play with it)

You can also add a border in combination with bTexWrap=0, to make those 
area black. Try this:
ob_a=1
ob_size=0.01
(There are also ob_[r|g|b] to change the color, defaults are 0)

S.

On 11/01/2011 12:24 PM, Werner Almesberger wrote:
> I noticed that, for high values of "rot", I get fairly large
> "titles" in the corners where the texture doesn't match the rest:
>
> http://downloads.qi-hardware.com/people/werner/m1/midi/corner-artefact.jpg
>
> When modulating "zoom", I can put them to good use, but they look
(Continue reading)

Werner Almesberger | 1 Nov 15:06 2011
Picon

Re: bug ? texturing artefacts in corners when rotating

S?bastien Bourdeauducq wrote:
> Right now, you are seeing behaviour #1. You can go to #2 with this line:
> bTexWrap=0
> (it also works in per-frame equations, if you want to play with it)

Cool, thanks ! Of course I want to play with it ;-) Here's a short
video:

http://downloads.qi-hardware.com/people/werner/m1/demo/MVI_1765.MOV

It's rotated. To unrotate, play with mplayer -vf rotate=2

Audio: "Love Tractor Part 1" from "Doghouse Riley" (the full song
uses samples from Lessig, but the extract in the video is just an
instrumental part). CC-BY.

http://ccmixter.org/files/Doghouse/18888
http://ccmixter.org/content/Doghouse/Doghouse_-_Love_Tractor_Part_1.mp3

I assigned bTexWrap to a button on the nK2. So the default would be
clamping and it would only wrap (as it did in the videos before)
when I press the button.

The effect is most visible when inverting the zoom, so that the
screen content is pulled towards the center:

00:03 smooth traces from the border (with clamping)
00:06 oops, missed the border, let's try this again ...
00:10 distortion producing "kaleidoscope" patterns (without clamping)
00:13 and once more
(Continue reading)

Werner Almesberger | 1 Nov 18:44 2011
Picon

Re: bug: M1 hangs after heavy MIDI traffic

I wrote:
> Now running under gdb since last evening.

It now felt ready for another hang. The GDB I had already attached
responded to ^C and produces this stack trace:

Program received signal SIGTRAP, Trace/breakpoint trap.
...
(gdb) bt f
#0  memcpy (dst0=0x40a52dc4, src0=0x10, len0=4234967967)
    at ../../../../../../../../../gcc-4.5.2/newlib/libc/string/memcpy.c:87
        dst = 0x40a52dc4 "U\231\252f\f\205"
        src = 0x10 "U\231\252f\f\205"
        aligned_dst = 0x4438b224
        aligned_src = 0x3938470
#1  0x401039e8 in _CORE_message_queue_Copy_buffer (
    the_message_queue=0x408d96bc, id=<value optimized out>, 
    buffer=<value optimized out>, size_p=<value optimized out>, wait=36, 
    timeout=<value optimized out>)
    at ../../cpukit/../../../milkymist/lib/include/rtems/score/coremsg.inl:101
No locals.
#2  _CORE_message_queue_Seize (the_message_queue=0x408d96bc, 
    id=<value optimized out>, buffer=<value optimized out>, 
    size_p=<value optimized out>, wait=36, timeout=<value optimized out>)
    at ../../../../../../rtems-milkymist/c/src/../../cpukit/score/src/coremsgseize.c:83
        level = 4294967295
        executing = <value optimized out>
#3  0x40101c5c in rtems_message_queue_receive (id=<value optimized out>, 
    buffer=0x40a52dc4, size=0x40a52da0, option_set=0, timeout=0)
    at ../../../../../../rtems-milkymist/c/src/../../cpukit/rtems/src/msgqreceive.c:83
(Continue reading)


Gmane