[9fans] How to read/write pixels from Memimage

[9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

Hello. I'm porting Bell Labs' Pico image language over to Plan 9 to  
use the Plan 9 native image format (image(6)), and I chose to use  
Memimage. I'd like to know how to get the 32-bit RGB pixel  
information out of a Memimage and into a properly allocated array of  
u32ints, and to do the reverse. Is that possible? Thanks.

Re: [9fans] How to read/write pixels from Memimage

[Russ Cox]

> Hello. I'm porting Bell Labs' Pico image language over to Plan 9 to  
> use the Plan 9 native image format (image(6)), and I chose to use  
> Memimage. I'd like to know how to get the 32-bit RGB pixel  
> information out of a Memimage and into a properly allocated array of  
> u32ints, and to do the reverse. Is that possible? Thanks.

A good way to answer questions like these is to 

	cd /sys/src/cmd
	grep -n Memimage *.c */*.c */*/*.c

Looking through the output, all of these:

	/sys/src/cmd/crop.c
	/sys/src/cmd/resample.c
	/sys/src/cmd/jpg/writepng.c

are examples of programs that use memimages
in a byte-at-a-time fashion.

Russ

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

However, I want to be able to read 32-bit color values. How does  
Memimage store colors and will the code I use work? I can change it  
to 24-bit if necessary, but that would make it slower to read files.  
If I do what these programs say, how will my u32int be arranged?

On Jan 26, 2008, at 9:05 AM, Russ Cox wrote:

>> Hello. I'm porting Bell Labs' Pico image language over to Plan 9 to
>> use the Plan 9 native image format (image(6)), and I chose to use
>> Memimage. I'd like to know how to get the 32-bit RGB pixel
>> information out of a Memimage and into a properly allocated array of
>> u32ints, and to do the reverse. Is that possible? Thanks.
>
> A good way to answer questions like these is to
>
> 	cd /sys/src/cmd
> 	grep -n Memimage *.c */*.c */*/*.c
>
> Looking through the output, all of these:
>
> 	/sys/src/cmd/crop.c
> 	/sys/src/cmd/resample.c
> 	/sys/src/cmd/jpg/writepng.c
>
> are examples of programs that use memimages
> in a byte-at-a-time fashion.
>
> Russ
>

Re: [9fans] How to read/write pixels from Memimage

[Russ Cox]

> However, I want to be able to read 32-bit color values. How does  
> Memimage store colors and 

You should be able to answer this by reading the programs
I pointed out as well as color(6), image(6), and memdraw(2).
If all else fails, you could try writing a simple program and
see how far you get.

> will the code I use work? 

Probably not at first (whose code does?), but eventually.

> I can change it  to 24-bit if necessary, but that would make it slower to read files.  
> If I do what these programs say, how will my u32int be arranged?

The answers to these questions are *in* those programs
(not to mention the manual pages!).  Given that your reply
came in four minutes after my post, it sounds like you didn't
actually read the programs I pointed out.  The first one on
the list, /sys/src/cmd/crop.c, picks up a Plan 9 image file
and then walks over every 32-bit pixel looking for a given color.
And it's only 211 lines.  Between that and the documentation
I'd think that would be enough to answer questions about
how colors are stored.  At the least it should prompt 
questions that are more specific than "how do I do this?"

Read /sys/src/cmd/crop.c, color(6), image(6), and memdraw(2).

Russ

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

The code led me to confusion, so I just decided to use wordaddr. It  
works properly now,  and my Pico is now up at /n/sources/contrib/ 
pietro/pico9.tgz (note the 9). Details are in README.Plan9, test  
files are in /lib/face/.

On Jan 26, 2008, at 9:55 AM, Russ Cox wrote:

>> However, I want to be able to read 32-bit color values. How does
>> Memimage store colors and
>
> You should be able to answer this by reading the programs
> I pointed out as well as color(6), image(6), and memdraw(2).
> If all else fails, you could try writing a simple program and
> see how far you get.
>
>> will the code I use work?
>
> Probably not at first (whose code does?), but eventually.
>
>> I can change it  to 24-bit if necessary, but that would make it  
>> slower to read files.
>> If I do what these programs say, how will my u32int be arranged?
>
> The answers to these questions are *in* those programs
> (not to mention the manual pages!).  Given that your reply
> came in four minutes after my post, it sounds like you didn't
> actually read the programs I pointed out.  The first one on
> the list, /sys/src/cmd/crop.c, picks up a Plan 9 image file
> and then walks over every 32-bit pixel looking for a given color.
> And it's only 211 lines.  Between that and the documentation
> I'd think that would be enough to answer questions about
> how colors are stored.  At the least it should prompt
> questions that are more specific than "how do I do this?"
>
> Read /sys/src/cmd/crop.c, color(6), image(6), and memdraw(2).
>
> Russ
>

Re: [9fans] How to read/write pixels from Memimage

[Skip Tavakkolian]

> The code led me to confusion, so I just decided to use wordaddr. It  
> works properly now,  and my Pico is now up at /n/sources/contrib/ 
> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test  
> files are in /lib/face/.

some book examples are failing:

cpu% 8.out
1:	new[x,y] = x % y
8.out 181613: suicide: sys: trap: divide error pc=0x000023fd

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

That transformation is in the book? It divides by zero for y == 0. I  
have to look.

I'm working on an update. It adds:
	r x y w h file (done)
		Reads a certain rectangle of a file. This makes crop(1)'s -r option  
three lines of rc or /bin/sh:
			echo 'r '$1' '$2' '$3' '$4' '$5'
			new = old
			w '$6 | pico
	Simple variables (started)
	Undo command u (done)
	Use of RGB24 instead of RGBA32 to make the examples work (done)
	Using Brdstr instead of Bgetc/Bungetc (this is not working)

On Jan 27, 2008, at 1:05 PM, Skip Tavakkolian wrote:

>> The code led me to confusion, so I just decided to use wordaddr. It
>> works properly now,  and my Pico is now up at /n/sources/contrib/
>> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test
>> files are in /lib/face/.
>
> some book examples are failing:
>
> cpu% 8.out
> 1:	new[x,y] = x % y
> 8.out 181613: suicide: sys: trap: divide error pc=0x000023fd
>

[9fans] pico

[Steve Simon]

FYI

Byron Rakitzis (of posix rc fame) produced a version popi with the JIT
compiler, though sadly his where for cpus which are common cpus these days.

Also possibly of interest is the paper from the 10th edition manual
http://plan9.bell-labs.com/10thEdMan/pico.pdf

And a few fun pictures that where generated
http://plan9.bell-labs.com/10thEdMan/v2pix.html

-Steve

Re: [9fans] pico

[Russ Cox]

> Byron Rakitzis (of posix rc fame) produced a version popi with the JIT
> compiler, though sadly his where for cpus which are common cpus these days.

Computers and compilers are fast enough now
that you can get away with just feeding code
into a C compiler instead of writing a full JIT.
And there's no porting to do!

I just put a pico on sources that does this - 738 lines,
not many of which are the "JIT".

9fs sources
cd /n/sources/contrib/rsc/pico
mk demo

Russ

Re: [9fans] pico

[Pietro Gagliardi]

Wow, that's very impressive!

	open: /tmp/something does not exist (2x)
	no image lerp
	no image doug

But you have a preview, which I was going to add soon.

I am sticking with my code interpreter because I know how to use one  
(I toiled over hoc. and fossil ate my code up last December) and it  
doesn't require hooking to a C compiler. I do like the preprocessor  
idea though.

On Jan 28, 2008, at 6:36 PM, Russ Cox wrote:

>> Byron Rakitzis (of posix rc fame) produced a version popi with the  
>> JIT
>> compiler, though sadly his where for cpus which are common cpus  
>> these days.
>
> Computers and compilers are fast enough now
> that you can get away with just feeding code
> into a C compiler instead of writing a full JIT.
> And there's no porting to do!
>
> I just put a pico on sources that does this - 738 lines,
> not many of which are the "JIT".
>
> 9fs sources
> cd /n/sources/contrib/rsc/pico
> mk demo
>
> Russ
>

Re: [9fans] pico

[Bruce Ellis]

well i agree that russ did a good job but JITs are fun and often
make things 10 times faster.

brucee

On Jan 29, 2008 11:30 AM, Pietro Gagliardi <pietro10@mac.com> wrote:
> Wow, that's very impressive!
>
>        open: /tmp/something does not exist (2x)
>        no image lerp
>        no image doug
>
> But you have a preview, which I was going to add soon.
>
> I am sticking with my code interpreter because I know how to use one
> (I toiled over hoc. and fossil ate my code up last December) and it
> doesn't require hooking to a C compiler. I do like the preprocessor
> idea though.
>
>
> On Jan 28, 2008, at 6:36 PM, Russ Cox wrote:
>
> >> Byron Rakitzis (of posix rc fame) produced a version popi with the
> >> JIT
> >> compiler, though sadly his where for cpus which are common cpus
> >> these days.
> >
> > Computers and compilers are fast enough now
> > that you can get away with just feeding code
> > into a C compiler instead of writing a full JIT.
> > And there's no porting to do!
> >
> > I just put a pico on sources that does this - 738 lines,
> > not many of which are the "JIT".
> >
> > 9fs sources
> > cd /n/sources/contrib/rsc/pico
> > mk demo
> >
> > Russ
> >
>
>

Re: [9fans] pico

[Pietro Gagliardi]

In the latest update, I tried adding differentiating between color  
and b/w images. However, I can't test anything because every time I  
try a line like

	x new = dennis

I get something that ends in "(double-free?)" and then the program  
crashes, but something like

	x new
	x new =

do call the error() function. The lexer did not change since I  
started this update, but it did change when I improved symbol handling.

On Jan 29, 2008, at 12:30 AM, Bruce Ellis wrote:

> well i agree that russ did a good job but JITs are fun and often
> make things 10 times faster.
>
> brucee
>
> On Jan 29, 2008 11:30 AM, Pietro Gagliardi <pietro10@mac.com> wrote:
>> Wow, that's very impressive!
>>
>>        open: /tmp/something does not exist (2x)
>>        no image lerp
>>        no image doug
>>
>> But you have a preview, which I was going to add soon.
>>
>> I am sticking with my code interpreter because I know how to use one
>> (I toiled over hoc. and fossil ate my code up last December) and it
>> doesn't require hooking to a C compiler. I do like the preprocessor
>> idea though.
>>
>>
>> On Jan 28, 2008, at 6:36 PM, Russ Cox wrote:
>>
>>>> Byron Rakitzis (of posix rc fame) produced a version popi with the
>>>> JIT
>>>> compiler, though sadly his where for cpus which are common cpus
>>>> these days.
>>>
>>> Computers and compilers are fast enough now
>>> that you can get away with just feeding code
>>> into a C compiler instead of writing a full JIT.
>>> And there's no porting to do!
>>>
>>> I just put a pico on sources that does this - 738 lines,
>>> not many of which are the "JIT".
>>>
>>> 9fs sources
>>> cd /n/sources/contrib/rsc/pico
>>> mk demo
>>>
>>> Russ
>>>
>>
>>

Re: [9fans] pico

[Pietro Gagliardi]

This only happens with black/white images, I ran a color image (boyd)  
through and it worked fine. I think it's with the run function...

On Jan 29, 2008, at 3:33 PM, Pietro Gagliardi wrote:

> In the latest update, I tried adding differentiating between color  
> and b/w images. However, I can't test anything because every time I  
> try a line like
>
> 	x new = dennis
>
> I get something that ends in "(double-free?)" and then the program  
> crashes, but something like
>
> 	x new
> 	x new =
>
> do call the error() function. The lexer did not change since I  
> started this update, but it did change when I improved symbol  
> handling.
>
> On Jan 29, 2008, at 12:30 AM, Bruce Ellis wrote:
>
>> well i agree that russ did a good job but JITs are fun and often
>> make things 10 times faster.
>>
>> brucee
>>
>> On Jan 29, 2008 11:30 AM, Pietro Gagliardi <pietro10@mac.com> wrote:
>>> Wow, that's very impressive!
>>>
>>>        open: /tmp/something does not exist (2x)
>>>        no image lerp
>>>        no image doug
>>>
>>> But you have a preview, which I was going to add soon.
>>>
>>> I am sticking with my code interpreter because I know how to use one
>>> (I toiled over hoc. and fossil ate my code up last December) and it
>>> doesn't require hooking to a C compiler. I do like the preprocessor
>>> idea though.
>>>
>>>
>>> On Jan 28, 2008, at 6:36 PM, Russ Cox wrote:
>>>
>>>>> Byron Rakitzis (of posix rc fame) produced a version popi with the
>>>>> JIT
>>>>> compiler, though sadly his where for cpus which are common cpus
>>>>> these days.
>>>>
>>>> Computers and compilers are fast enough now
>>>> that you can get away with just feeding code
>>>> into a C compiler instead of writing a full JIT.
>>>> And there's no porting to do!
>>>>
>>>> I just put a pico on sources that does this - 738 lines,
>>>> not many of which are the "JIT".
>>>>
>>>> 9fs sources
>>>> cd /n/sources/contrib/rsc/pico
>>>> mk demo
>>>>
>>>> Russ
>>>>
>>>
>>>
>

Re: [9fans] pico

[Russ Cox]

> I get something that ends in "(double-free?)" and then the program  
> crashes, but something like

This usually means you have freed the same pointer twice
or you are passing a pointer to free that was not returned
by malloc.  If you run acid to get a stack trace to see the
context of the free, that is often enough to identify the
problem.

% acid pid
acid: stk()
...
^D
%

Russ

Re: [9fans] pico

[Pietro Gagliardi]

In this case, it only told me where the program crashed, which told  
me nothing on why (because a different line or a color image worked).

On Jan 29, 2008, at 5:46 PM, Russ Cox wrote:

>> I get something that ends in "(double-free?)" and then the program
>> crashes, but something like
>
> This usually means you have freed the same pointer twice
> or you are passing a pointer to free that was not returned
> by malloc.  If you run acid to get a stack trace to see the
> context of the free, that is often enough to identify the
> problem.
>
> % acid pid
> acid: stk()
> ...
> ^D
> %
>
> Russ
>

Re: [9fans] pico

[Bruce Ellis]

work it out.  it's called debugging.

brucee

On Jan 30, 2008 12:13 PM, Pietro Gagliardi <pietro10@mac.com> wrote:
> In this case, it only told me where the program crashed, which told
> me nothing on why (because a different line or a color image worked).
>
>
> On Jan 29, 2008, at 5:46 PM, Russ Cox wrote:
>
> >> I get something that ends in "(double-free?)" and then the program
> >> crashes, but something like
> >
> > This usually means you have freed the same pointer twice
> > or you are passing a pointer to free that was not returned
> > by malloc.  If you run acid to get a stack trace to see the
> > context of the free, that is often enough to identify the
> > problem.
> >
> > % acid pid
> > acid: stk()
> > ...
> > ^D
> > %
> >
> > Russ
> >
>
>

Re: [9fans] pico

[Pietro Gagliardi]

[-- Attachment #1: Type: text/plain, Size: 256 bytes --]

What do you think I've been doing? :-)

The problem seems to be in run(), because after run() returns,  
poolcheck fails. I'll go check it out.

On Jan 30, 2008, at 1:53 AM, Bruce Ellis wrote:

> work it out.  it's called debugging.
>
> brucee


[-- Attachment #2: Type: text/html, Size: 903 bytes --]

Re: [9fans] pico

[Pietro Gagliardi]

[-- Attachment #1: Type: text/plain, Size: 514 bytes --]

Well, I got the memory tampering bugs fixed, but a black and white  
image now does not show correctly. Gonna go back and check and fix.  
Will report when I put a new version up.

On Jan 30, 2008, at 6:42 PM, Pietro Gagliardi wrote:

> What do you think I've been doing? :-)
>
> The problem seems to be in run(), because after run() returns,  
> poolcheck fails. I'll go check it out.
>
> On Jan 30, 2008, at 1:53 AM, Bruce Ellis wrote:
>
>> work it out.  it's called debugging.
>>
>> brucee
>


[-- Attachment #2: Type: text/html, Size: 1410 bytes --]

Re: [9fans] pico

[Bruce Ellis]

more debugging, please tell us about it.  yawn.

On Jan 31, 2008 10:55 AM, Pietro Gagliardi <pietro10@mac.com> wrote:
>
> Well, I got the memory tampering bugs fixed, but a black and white image now
> does not show correctly. Gonna go back and check and fix. Will report when I
> put a new version up.
>
>
>
> On Jan 30, 2008, at 6:42 PM, Pietro Gagliardi wrote:
>
> What do you think I've been doing? :-)
>
> The problem seems to be in run(), because after run() returns, poolcheck
> fails. I'll go check it out.
>
> On Jan 30, 2008, at 1:53 AM, Bruce Ellis wrote:
>
> work it out.  it's called debugging.
>
> brucee
>
>

Re: [9fans] pico

[Pietro Gagliardi]

Okay, I got everything fixed a few days ago and added some more
features (simple arrays, r "long/file/name.1", etc.). The new pico is
in the usual place.

I'm redesigning my website, so I'm going to have my software
available from there too when that's done.

On Jan 31, 2008, at 2:41 AM, Bruce Ellis wrote:

> more debugging, please tell us about it.  yawn.
>
> On Jan 31, 2008 10:55 AM, Pietro Gagliardi <pietro10@mac.com> wrote:
>>
>> Well, I got the memory tampering bugs fixed, but a black and white
>> image now
>> does not show correctly. Gonna go back and check and fix. Will
>> report when I
>> put a new version up.
>>
>>
>>
>> On Jan 30, 2008, at 6:42 PM, Pietro Gagliardi wrote:
>>
>> What do you think I've been doing? :-)
>>
>> The problem seems to be in run(), because after run() returns,
>> poolcheck
>> fails. I'll go check it out.
>>
>> On Jan 30, 2008, at 1:53 AM, Bruce Ellis wrote:
>>
>> work it out.  it's called debugging.
>>
>> brucee
>>
>>

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

Well I just found it, and I couldn't believe it. I'm going to look at  
Gerard's original source for the VAX and see what he does on divide  
by zero.

On Jan 27, 2008, at 1:05 PM, Skip Tavakkolian wrote:

>> The code led me to confusion, so I just decided to use wordaddr. It
>> works properly now,  and my Pico is now up at /n/sources/contrib/
>> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test
>> files are in /lib/face/.
>
> some book examples are failing:
>
> cpu% 8.out
> 1:	new[x,y] = x % y
> 8.out 181613: suicide: sys: trap: divide error pc=0x000023fd
>

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

Okay, I just found out that division by zero meant a black pixel  
would be placed, so I am going to add that and upload a new version.

On Jan 27, 2008, at 1:05 PM, Skip Tavakkolian wrote:

>> The code led me to confusion, so I just decided to use wordaddr. It
>> works properly now,  and my Pico is now up at /n/sources/contrib/
>> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test
>> files are in /lib/face/.
>
> some book examples are failing:
>
> cpu% 8.out
> 1:	new[x,y] = x % y
> 8.out 181613: suicide: sys: trap: divide error pc=0x000023fd
>

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

It's up.

On Jan 27, 2008, at 6:44 PM, Pietro Gagliardi wrote:

> Okay, I just found out that division by zero meant a black pixel  
> would be placed, so I am going to add that and upload a new version.
>
> On Jan 27, 2008, at 1:05 PM, Skip Tavakkolian wrote:
>
>>> The code led me to confusion, so I just decided to use wordaddr. It
>>> works properly now,  and my Pico is now up at /n/sources/contrib/
>>> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test
>>> files are in /lib/face/.
>>
>> some book examples are failing:
>>
>> cpu% 8.out
>> 1:	new[x,y] = x % y
>> 8.out 181613: suicide: sys: trap: divide error pc=0x000023fd
>>
>

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

Another question: in RGB24, how are colors composited: 0x00RRGGBB or  
0xRRGGBB00? I can't seem to get a picture that is solid red; just a  
wacky line art picture.

On Jan 27, 2008, at 7:18 PM, Pietro Gagliardi wrote:

> It's up.
>
> On Jan 27, 2008, at 6:44 PM, Pietro Gagliardi wrote:
>
>> Okay, I just found out that division by zero meant a black pixel  
>> would be placed, so I am going to add that and upload a new version.
>>
>> On Jan 27, 2008, at 1:05 PM, Skip Tavakkolian wrote:
>>
>>>> The code led me to confusion, so I just decided to use wordaddr. It
>>>> works properly now,  and my Pico is now up at /n/sources/contrib/
>>>> pietro/pico9.tgz (note the 9). Details are in README.Plan9, test
>>>> files are in /lib/face/.
>>>
>>> some book examples are failing:
>>>
>>> cpu% 8.out
>>> 1:	new[x,y] = x % y
>>> 8.out 181613: suicide: sys: trap: divide error pc=0x000023fd
>>>
>>
>

Re: [9fans] How to read/write pixels from Memimage

[erik quanstrom]

> Another question: in RGB24, how are colors composited: 0x00RRGGBB or  
> 0xRRGGBB00? I can't seem to get a picture that is solid red; just a  
> wacky line art picture.

read image(6).  all the details are spelled out there.  (you likely want an alpha
channel in your image, so that each pixel is 4 bytes.)

- erik

Re: [9fans] How to read/write pixels from Memimage

[Pietro Gagliardi]

[-- Attachment #1: Type: text/plain, Size: 410 bytes --]

On Jan 27, 2008, at 8:22 PM, erik quanstrom wrote:

> (you likely want an alpha
> channel in your image, so that each pixel is 4 bytes.)

Apparently, my RGB24 manipulation was wrong, and now it works like a  
charm. The next step is to support a construct like new.red and to  
eliminate the use of undocumented stuff (I peeked in /sys/include/ 
draw.h and /sys/src/libbio/brdstr.c for some hacks).


[-- Attachment #2: Type: text/html, Size: 902 bytes --]

Re: [9fans] How to read/write pixels from Memimage

[mattmobile]

9fans is not a blog, try blogger.com
>
>
>> (you likely want an alpha
>>
>> channel in your image, so that each pixel is 4 bytes.)
>>
>
> Apparently, my RGB24 manipulation was wrong, and now it works like a 
> charm. The next step is to support a construct like new.red and to 
> eliminate the use of undocumented stuff (I peeked in 
> /sys/include/draw.h and /sys/src/libbio/brdstr.c for some hacks).
>

Re: [9fans] How to read/write pixels from Memimage

[ron minnich]

On Jan 28, 2008 1:24 AM,  <mattmobile@proweb.co.uk> wrote:
> 9fans is not a blog, try blogger.com

now now now :-)

Why give the guy a hard time, he's actually discussing real coding he
is doing :-)

ron