[9fans] actionfs

[9fans] actionfs

[mattmobile]

Hi,

this one was an experiment

/n/sources/contrib/maht/actionfs.c

invoked with a regex like  actionfs (file.mpg).([0-9]+).(ppm)

if you then

cat /n/actionfs/file.mpg.100.ppm

actionfs responds with the output from executing

/bin/action-read $fd file.mpg.100.ppm file.mpg 100 ppm

where $fd will be an fd to write to

i.e. trivially action-read would be something like

----
#!/bin/rc

fd = $1
shift

echo $* > /fd/$fd

-----

The coresponding action-write also works


----
#!/bin/rc

fd = $1
shift

cat /fd/$fd > /dev/null # or whatever

-----

I wrote it specifically to extract individual frames from video files using ffmpeg on Linux and
bring them into Plan9 for processing but generalized the arguments in case I thought of something
interesting later.

My first round of experiment went like this

cpu% cat /bin/action-read
#!/bin/rc

# expect fd fullname videoname frameno
fname = `{echo -n $3 | tr ! '/'}
{
	ssh storm single_frame $fname $4
} > /fd/$1


cpu% cat /n/storm/home/maht/bin/single_frame
#!/usr/local/plan9/bin/rc

# expect filename frameno

timer = `{echo $2  | awk ' { printf "%d.%02d\n",  $1/ 25, 4 * ($1 % 25) }'}
{
	ffmpeg -i $1 -t 00.001 -ss $timer /tmp/frame_$pid ^_%d.ppm
	cat /tmp/frame_$pid ^_1.ppm
	rm -f frame_$pid ^_1.ppm
	rm -f frame_$pid ^_2.ppm  # stupid ffmpeg outputs 2 frames (sometimes)
}  >[2] /dev/null


I was then using imgfs to calculate the average rgb value to look for black frames but (unsurprisingly) it was taking too long (4 secs per frame) esp. as the Plan9 I was using is in Qemu, cue installing Plan 9 on my terminal.

The ffmpeg part on the Linux side (2Ghz Opteron) was taking 1 second on its own so I have to come up with some sort of look ahead cache which is contrary to the idea, I may as well just convert the whole file to ppms at the start! I've not looked if it is I/O or CPU - perhaps a bit of both.

I've not got round to doing it on my fresh terminal yet. I've got a new 3.2Ghz Dual Xeon server to migrate to and a Quad Core terminal to play with so we'll see how that works out.

I was hoping to get Xcpu in there but I couldn't see how to get the Plan9 part working though I have the Linux bits up.

I have a couple of decent OSX boxes available too (one PPC one Intel) but I gave up getting it to compile :)

too many projects .....

matt

Re: [9fans] actionfs

[Bruce Ellis]

to make "score" i converted over 30,000 frames from mpg to tiff, using
inferno. it took a while, as it was 10 years ago, but nice to wake up
to.

brucee

On Mon, Feb 23, 2009 at 11:13 AM,  <mattmobile@proweb.co.uk> wrote:
> Hi,
>
> this one was an experiment
>
> /n/sources/contrib/maht/actionfs.c
>
> invoked with a regex like  actionfs (file.mpg).([0-9]+).(ppm)
>
> if you then
>
> cat /n/actionfs/file.mpg.100.ppm
>
> actionfs responds with the output from executing
>
> /bin/action-read $fd file.mpg.100.ppm file.mpg 100 ppm
>
> where $fd will be an fd to write to
>
> i.e. trivially action-read would be something like
>
> ----
> #!/bin/rc
>
> fd = $1
> shift
>
> echo $* > /fd/$fd
>
> -----
>
> The coresponding action-write also works
>
>
> ----
> #!/bin/rc
>
> fd = $1
> shift
>
> cat /fd/$fd > /dev/null # or whatever
>
> -----
>
> I wrote it specifically to extract individual frames from video files using ffmpeg on Linux and
> bring them into Plan9 for processing but generalized the arguments in case I thought of something
> interesting later.
>
> My first round of experiment went like this
>
> cpu% cat /bin/action-read
> #!/bin/rc
>
> # expect fd fullname videoname frameno
> fname = `{echo -n $3 | tr ! '/'}
> {
>        ssh storm single_frame $fname $4
> } > /fd/$1
>
>
> cpu% cat /n/storm/home/maht/bin/single_frame
> #!/usr/local/plan9/bin/rc
>
> # expect filename frameno
>
> timer = `{echo $2  | awk ' { printf "%d.%02d\n",  $1/ 25, 4 * ($1 % 25) }'}
> {
>        ffmpeg -i $1 -t 00.001 -ss $timer /tmp/frame_$pid ^_%d.ppm
>        cat /tmp/frame_$pid ^_1.ppm
>        rm -f frame_$pid ^_1.ppm
>        rm -f frame_$pid ^_2.ppm  # stupid ffmpeg outputs 2 frames (sometimes)
> }  >[2] /dev/null
>
>
> I was then using imgfs to calculate the average rgb value to look for black frames but (unsurprisingly) it was taking too long (4 secs per frame) esp. as the Plan9 I was using is in Qemu, cue installing Plan 9 on my terminal.
>
> The ffmpeg part on the Linux side (2Ghz Opteron) was taking 1 second on its own so I have to come up with some sort of look ahead cache which is contrary to the idea, I may as well just convert the whole file to ppms at the start! I've not looked if it is I/O or CPU - perhaps a bit of both.
>
> I've not got round to doing it on my fresh terminal yet. I've got a new 3.2Ghz Dual Xeon server to migrate to and a Quad Core terminal to play with so we'll see how that works out.
>
> I was hoping to get Xcpu in there but I couldn't see how to get the Plan9 part working though I have the Linux bits up.
>
> I have a couple of decent OSX boxes available too (one PPC one Intel) but I gave up getting it to compile :)
>
> too many projects .....
>
> matt
>
>

Re: [9fans] actionfs

[roger peppe]

2009/2/23 Bruce Ellis <bruce.ellis@gmail.com>:
> to make "score" i converted over 30,000 frames from mpg to tiff, using
> inferno. it took a while, as it was 10 years ago, but nice to wake up
> to.

what's "score"?

Re: [9fans] actionfs

[Bruce Ellis]

Score!

http://www.chunder.com/stuff/score.html

On Tue, Feb 24, 2009 at 12:41 AM, roger peppe <rogpeppe@gmail.com> wrote:
> 2009/2/23 Bruce Ellis <bruce.ellis@gmail.com>:
>> to make "score" i converted over 30,000 frames from mpg to tiff, using
>> inferno. it took a while, as it was 10 years ago, but nice to wake up
>> to.
>
> what's "score"?
>
>

[9fans] (no subject)

[cej]

[-- Attachment #1: Type: text/plain, Size: 2593 bytes --]




-----Original Message-----
From: 9fans-bounces@9fans.net on behalf of mattmobile@proweb.co.uk
Sent: Mon 2/23/2009 1:13 AM
To: 9fans@9fans.net
Subject: [9fans] actionfs
 
Hi,

this one was an experiment

/n/sources/contrib/maht/actionfs.c

invoked with a regex like  actionfs (file.mpg).([0-9]+).(ppm)

if you then 

cat /n/actionfs/file.mpg.100.ppm

actionfs responds with the output from executing

/bin/action-read $fd file.mpg.100.ppm file.mpg 100 ppm

where $fd will be an fd to write to 

i.e. trivially action-read would be something like

----
#!/bin/rc

fd = $1
shift

echo $* > /fd/$fd

-----

The coresponding action-write also works


----
#!/bin/rc

fd = $1
shift

cat /fd/$fd > /dev/null # or whatever

-----

I wrote it specifically to extract individual frames from video files using ffmpeg on Linux and 
bring them into Plan9 for processing but generalized the arguments in case I thought of something 
interesting later.

My first round of experiment went like this

cpu% cat /bin/action-read
#!/bin/rc

# expect fd fullname videoname frameno
fname = `{echo -n $3 | tr ! '/'}
{
	ssh storm single_frame $fname $4 
} > /fd/$1


cpu% cat /n/storm/home/maht/bin/single_frame
#!/usr/local/plan9/bin/rc

# expect filename frameno

timer = `{echo $2  | awk ' { printf "%d.%02d\n",  $1/ 25, 4 * ($1 % 25) }'}
{
	ffmpeg -i $1 -t 00.001 -ss $timer /tmp/frame_$pid ^_%d.ppm
	cat /tmp/frame_$pid ^_1.ppm 
	rm -f frame_$pid ^_1.ppm
	rm -f frame_$pid ^_2.ppm  # stupid ffmpeg outputs 2 frames (sometimes)
}  >[2] /dev/null


I was then using imgfs to calculate the average rgb value to look for black frames but (unsurprisingly) it was taking too long (4 secs per frame) esp. as the Plan9 I was using is in Qemu, cue installing Plan 9 on my terminal.

The ffmpeg part on the Linux side (2Ghz Opteron) was taking 1 second on its own so I have to come up with some sort of look ahead cache which is contrary to the idea, I may as well just convert the whole file to ppms at the start! I've not looked if it is I/O or CPU - perhaps a bit of both.

I've not got round to doing it on my fresh terminal yet. I've got a new 3.2Ghz Dual Xeon server to migrate to and a Quad Core terminal to play with so we'll see how that works out.

I was hoping to get Xcpu in there but I couldn't see how to get the Plan9 part working though I have the Linux bits up.

I have a couple of decent OSX boxes available too (one PPC one Intel) but I gave up getting it to compile :)

too many projects .....

matt



[-- Attachment #2: winmail.dat --]
[-- Type: application/ms-tnef, Size: 3785 bytes --]