jpeg decoding in full asm

[Thorham]

Originally Posted by meynaf

I wondered if it couldn't be the source of your curious machine behaviors. No driver needed on the amiga side ?

Nope. It just accepts the amigas video output.

Originally Posted by meynaf

It can be understood like this : when you write a value to memory, you don't need it to be actually written before going on. On the other hand, how can the program continue without knowing what we have read ?

Good explanation. Should have figured it out, though

Originally Posted by meynaf

So "than" and "then" are the same in Dutch, am I right ?

Yes, we use the word 'dan' for both cases.

Originally Posted by meynaf

How unfortunate. Now what's left for you to do is to write a faster triangular one
Note that if you can't beat mine (and you won't, heheh ) there is still the 2:1 version to check (also triangular interpolation but writes 2 horizontal pixels and 1 vertical). A more common case than I first expected.

Ah, very interesting. I think I'll have a few thoughts about it. A 2x1 versions should be a bit less complicated, that's for sure!

Originally Posted by meynaf

Resurrected ! It's miraculous. You're a wizard, man

Hey, it was only the cable!

[meynaf]

Originally Posted by Thorham

Ah, very interesting. I think I'll have a few thoughts about it. A 2x1 versions should be a bit less complicated, that's for sure!

Try to beat mine then :

; upsample 2 horizontal, 1 vertical
; a5=input_data, a6=output_data, d7=nb cols, d6=nb rows
h2v1_fancy_upsample
 subq.w #3,d7            ; for dbf, and minus 1st/last col.
.yloop
 move.l (a5)+,a0        ; a0 = input_data[inrow]
 move.l (a6)+,a1        ; a1 = output_data[outrow]

; special case of 1st column
 moveq #0,d1
 moveq #0,d0
 move.b (a0)+,d1
 move.b (a0)+,d0
 move.b d1,(a1)+        ; 1st pix : as is
 move.l d1,d2
 add.l d2,d2
 add.l d1,d2            ; *3
 add.l d0,d2            ; + next
 addq.l #2,d2            ; round with +2
 lsr.l #2,d2            ; /4
 move.b d2,(a1)+        ; 2nd pix

; general case
; d0    current pixel (do *3)
; d1    previous pixel
; d2    next pixel (the one we read within the loop)
; d3    tmp
 move.w d7,d5
 moveq #0,d2            ; can be out of the loop
.xloop
 move.l d0,d3
 add.l d3,d3
 add.l d0,d3            ; 3c
 add.l d3,d1            ; 3c + 1l
 addq.l #1,d1            ; round with +1
 lsr.l #2,d1            ; /4
 move.b d1,(a1)+        ; 1st pix
 move.l d0,d1            ; save center pix for next left pix
 move.b (a0)+,d2
 move.l d2,d0            ; save right pix for next center pix
 add.l d2,d3
 addq.l #2,d3            ; round with +2
 lsr.l #2,d3            ; /4
 move.b d3,(a1)+        ; 2nd pix
 dbf d5,.xloop

; special case of last column
 move.l d0,d2
 add.l d2,d2
 add.l d0,d2            ; *3
 add.l d1,d2            ; +old
 addq.l #1,d2            ; round with +1
 lsr.l #2,d2            ; /4
 move.b d2,(a1)+        ; 1 before last pix
 move.b d0,(a1)+        ; last pix (sent directly)

; line loop
 subq.w #1,d6
 bne .yloop
 rts

Anyway you're perfectly right : it was much easier to do !

Originally Posted by Thorham

Hey, it was only the cable!

Don't underestimate yourself

[Thorham]

That code sure looks hard to beat. But I have a question: What are the rounding adds for? They are present in your 2x2 code, too, and seeing what you have to do, I just don't get why you need those rounding adds

[meynaf]

Originally Posted by Thorham

That code sure looks hard to beat. But I have a question: What are the rounding adds for? They are present in your 2x2 code, too, and seeing what you have to do, I just don't get why you need those rounding adds

In fact I did the same as in the original code. The rounding adds are... errrh... to round the value.

In jdsample.c it was explained like this :

A note about the "bias" calculations: when rounding fractional values to
integer, we do not want to always round 0.5 up to the next integer.
If we did that, we'd introduce a noticeable bias towards larger values.
Instead, this code is arranged so that 0.5 will be rounded up or down at
alternate pixel locations (a simple ordered dither pattern).

That's why I alternatively add 8 and 7 in the 2:2 code, and I do a similar thing in the 2:1. If the IJG does that, there's probably a good reason.
Anyway removing those adds won't accelerate the program by much so they're ok for me.

[Thorham]

Originally Posted by meynaf

In fact I did the same as in the original code. The rounding adds are... errrh... to round the value.

In jdsample.c it was explained like this :

So they're for dithering? Hmm, I can try this in my freebasic code on the peecee, I'm really wondering if this improves the quality of the process.

Originally Posted by meynaf

That's why I alternatively add 8 and 7 in the 2:2 code, and I do a similar thing in the 2:1. If the IJG does that, there's probably a good reason.
Anyway removing those adds won't accelerate the program by much so they're ok for me.

But have you tried, just to see the difference? If not, I'll still try it out myself. You see, I think it's a little odd that they're using those adds for interpolation.

[meynaf]

Originally Posted by Thorham

So they're for dithering? Hmm, I can try this in my freebasic code on the peecee, I'm really wondering if this improves the quality of the process.

They're not for dithering. They're to round values without rounding 0.5 to 1 all the time. Correct rounding instead of truncating usually gets a slightly better quality, so it's good to do if it doesn't cost too much.

Originally Posted by Thorham

But have you tried, just to see the difference? If not, I'll still try it out myself. You see, I think it's a little odd that they're using those adds for interpolation.

Try it if you want. You probably will see no difference at all, both in speed and quality.


Anyway here is the project file for StormC 3.0.

You may wish to manually edit it (it's a text file) to change the paths in it (for now it goes in ram:src/).

Included is a reference executable file (a.out) built by it.

Note : as a bonus you'll get a slightly faster version of the viewer because it produced better code
(not to mention the exe being one kilobyte smaller : we're now under 32k )

[Thorham]

Originally Posted by meynaf

They're not for dithering. They're to round values without rounding 0.5 to 1 all the time. Correct rounding instead of truncating usually gets a slightly better quality, so it's good to do if it doesn't cost too much.

Yes, I was wondering why it was needed... In this case it only uses a couple of cheap adds, nothing to worry about.

Originally Posted by meynaf

Try it if you want. You probably will see no difference at all, both in speed and quality.


Anyway here is the project file for StormC 3.0.

You may wish to manually edit it (it's a text file) to change the paths in it (for now it goes in ram:src/).

Included is a reference executable file (a.out) built by it.

Note : as a bonus you'll get a slightly faster version of the viewer because it produced better code
(not to mention the exe being one kilobyte smaller : we're now under 32k )

Thank you, thank you Much appreciated. At first it wouldn't work, though. I checked everything once, twice even three times and more, until I found out Storm likes the project file in the directory where the source sits! Well, that's to be expected if one never really uses Storm! And the only reason for that, is that I haven't figured out how to use the great FrexxEd programmers editor with it.

Anyway, that was the ONLY problem, and it works like a charm. Thanks again

[meynaf]

If you spot something in the code that you want to modify and check, now you can do it. Happy crashi... errrhhh, happy playing !

Maybe I've missed something in the compiled part that can be simplified a lot, please let me know whatever you find. And remember : that C code has to get out !

[Thorham]

Originally Posted by meynaf

If you spot something in the code that you want to modify and check, now you can do it. Happy crashi... errrhhh, happy playing !

Yes, I can now indeed do work on that source code properly, and it gives me a good reason to use Storm3, so yeah, I'll give it a go.

Originally Posted by meynaf

Maybe I've missed something in the compiled part that can be simplified a lot, please let me know whatever you find. And remember : that C code has to get out !

There's a very good chance that there are indeed things that can be simplified. As you know, I have to restrict doing work on this to the evenings, so unfortunately I can't work on this as much as I would want.

[meynaf]

Originally Posted by Thorham

Yes, I can now indeed do work on that source code properly, and it gives me a good reason to use Storm3, so yeah, I'll give it a go.

I'm looking forward...

Originally Posted by Thorham

There's a very good chance that there are indeed things that can be simplified. As you know, I have to restrict doing work on this to the evenings, so unfortunately I can't work on this as much as I would want.

I've commented out / suppressed a lot of unneeded stuff already, but I couldn't remove their manager/controller objects (well, not all of them).
My work on this is restricted too : I won't be home this week-end and my side of the code isn't going to evolve until the next week-end.

[Thorham]

Originally Posted by meynaf

I've commented out / suppressed a lot of unneeded stuff already, but I couldn't remove their manager/controller objects (well, not all of them).
My work on this is restricted too : I won't be home this week-end and my side of the code isn't going to evolve until the next week-end.

Sounds uncool. I would really hate not having my miggy around. As said I'll look at it, and actually I have already, but I didn't find anything as of yet. I'm sure I will, though. By the way, is the wrapper complete? I might just try my hand at rewriting some c functions... Should be interesting.

[meynaf]

Originally Posted by Thorham

Sounds uncool. I would really hate not having my miggy around.

Well, ok, I won't have my miggy around, but there are others things apart the miggy in life

Originally Posted by Thorham

As said I'll look at it, and actually I have already, but I didn't find anything as of yet. I'm sure I will, though. By the way, is the wrapper complete? I might just try my hand at rewriting some c functions... Should be interesting.

What do you call the wrapper ?

[Thorham]

Originally Posted by meynaf

Well, ok, I won't have my miggy around, but there are others things apart the miggy in life

That's right. There's my peecee Tons of beer and weed (this is the Netherlands) and lot's of gaming (Diablo 2 Lod addict). Wow, I guess that sums it up

Originally Posted by meynaf

What do you call the wrapper ?

Maybe I called it something it it's not. I just meant the c parts which handle the asm integration. Or is the asm linked into the code directly? Well, since I finally have enough free time again, maybe I should take a look for my self, eh?

[meynaf]

Originally Posted by Thorham

That's right. There's my peecee Tons of beer and weed (this is the Netherlands) and lot's of gaming (Diablo 2 Lod addict). Wow, I guess that sums it up

No girlfriend ?

Originally Posted by Thorham

Maybe I called it something it it's not. I just meant the c parts which handle the asm integration. Or is the asm linked into the code directly? Well, since I finally have enough free time again, maybe I should take a look for my self, eh?

Oh, yes, there is such a wrapper but it's in asm, not in c. It is located in jpeg.s. However there is no wrapper for the part we were talking about (huffman decoding).
I have some code for that but it simply won't integrate

[Thorham]

Originally Posted by meynaf

No girlfriend ?

How can you even ask Yep, I'm a real nerd when it comes down to that. If it's 'girl friend out' or 'computer out' than in my case, you just know the woman is going to have to leave

Originally Posted by meynaf

Oh, yes, there is such a wrapper but it's in asm, not in c. It is located in jpeg.s. However there is no wrapper for the part we were talking about (huffman decoding).

Right, thats cleared up well. Just wanted to know so I can try some stuff if needed.

Originally Posted by meynaf

I have some code for that but it simply won't integrate

Why not? Could you post that code, please?

[meynaf]

Originally Posted by Thorham

How can you even ask Yep, I'm a real nerd when it comes down to that. If it's 'girl friend out' or 'computer out' than in my case, you just know the woman is going to have to leave

How could I even ask ? I just felt RECKLESS
For me if the woman wants a cuddle or such, then the computer will remain turned off. But if it's her out or computer out, then she's probably not the right person

Originally Posted by Thorham

Right, thats cleared up well. Just wanted to know so I can try some stuff if needed.

A wrapper here has to forward the parameters (from stack to registers), save non-scratch regs (usually d2-d7/a2-a6 or so), and load my global bss pointer (a5). It may also move the return value to d0 if it was elsewhere.

Originally Posted by Thorham

Why not? Could you post that code, please?

I'm not doing the things the same as they did. My huffman tables are incompatible with theirs, I'm not reading bits the same way, not to mention the whole thing still being a work in progress.

Anyway it looks like that :

; get the value from a huffman code
get_huffcode
 moveq #8,d5
 bsr readnbits   ; returns d1
 move.w (a1,d1.w*2),d5
 bpl.s .norm   ; >=0 = bits ok, <0=n°extra-table-2
; more than 8 bits needed (according to jdhuff.c it's only 5% cases)
.re
 neg.w d5
 subq.w #2,d5   ; ffff=invalid code
 bmi.s .invalid
 lea (a2,d5.w*2),a3
 moveq #1,d5
 bsr readnbits
; chose between the two combinations
 move.w (a3,d1.w*2),d5
 bmi.s .re
.norm
 move.b d5,d1   ; final value (yep it's only a byte)
 lsr.w #8,d5
 add.b d5,d2   ; reinject unused bits in d2
 bmi .hit   ; if still negative : not enough data !
 rts
; here it ain't funny anymore : the file is corrupt
; instead of gently inserting zeroes in the stream and emitting a warning like
; the jpeg lib does, we go into an error
; (if the file is broken then fixing it is none of our business)
.hit
 lea jerr15(pc),a3
 bra failure
.invalid
 lea jerr16(pc),a3
 bra failure

[Thorham]

Originally Posted by meynaf

For me if the woman wants a cuddle or such, then the computer will remain turned off. But if it's her out or computer out, then she's probably not the right person

Good point. If she were like that, than she'd be too much of a controll freak for me.

Originally Posted by meynaf

A wrapper here has to forward the parameters (from stack to registers), save non-scratch regs (usually d2-d7/a2-a6 or so), and load my global bss pointer (a5). It may also move the return value to d0 if it was elsewhere.

Good to know. I just hope I'm going to need to know!

Originally Posted by meynaf

I'm not doing the things the same as they did. My huffman tables are incompatible with theirs, I'm not reading bits the same way, not to mention the whole thing still being a work in progress.

Thanks for posting the code Shame it's not doing things the same way, but I guess it makes sense to me why it won't work now. This huffman decoding stuff is something I have to do for the png part, which is something I haven't really come round to doing yet. Of course, it doesn't have to be finished tomorrow. Guess I'm going to get to it soon. If you want to know why I've put if off, it's because I've been working on my Japanese dictionary program too much. Thank goodness that thing doesn't require much work anymore, as it's eating my time. You know how it is: Once you start you can't stop!

Originally Posted by meynaf

There's not much in store yet, but that damn c code will be dropped one day or another. I have an asm-only jpeg.s which already parses markers.

That's great, you're making good progress. I suppose I can download that version from your site?

Originally Posted by meynaf

It would be really cool but it simply can't be done. Remember : we're on a peecee, where you don't know a thing about what sort of hardware is there. So the os is needed for hardware abstraction layers. Yet another reason why I like the miggy

But you don't need an entire os just to handle some hardware. The emulator should be able to handle the drivers. They'd have to be adapted for the emu, but that should still be possible.

Originally Posted by meynaf

Thanks. A change here just means quitting my job though (and this is gonna happen soon !).

Good luck.

Originally Posted by meynaf

Mine is an old AT and is really noisy but I wouldn't dare to try to mod another one. I've got two left hands when it comes to soldering
(not to mention I don't have the required hardware anyway)

You could just try to replace the fan. It's something I might do instead of buying a new unit. With a little luck, it doesn't need soldering.

Originally Posted by meynaf

Sorry, yes, but no regrets. I once heard japanese people talking and I didn't like the way it sounded. Just monotonic, incomprehensible sounds to me...

Funny, I think it sounds kind of cool. It's really the writing system that fascinates me, hence the dictionary program.

Originally Posted by meynaf

For win forms things are usually easy. But did you try web forms ?

No, I haven't. I'm not even sure VB express supports that. The web developer does, most likely. But I just haven't needed those, yet.

Originally Posted by meynaf

They're probably not that incompatible if you code correctly
(and 300 bucks are way too much for such a cockroach nest IMO)

Cockroach nest Three hundred bucks is too much money for any program to me, I suppose one tenth of that price would be a bit better

Originally Posted by meynaf

For using existing components which may be written in either languages. A little too much code copy-paste I guess

Cut'n paste. I do that all the time. Must admit, though that oop saves you a lot of it sometimes.

Originally Posted by meynaf

Hmm I've just thought about this... Maybe we should continue this discussion here :
http://eab.abime.net/showthread.php?t=33576
as there's not much left to say about the ham8 renderer now...

That's a very good idea. There really is nothing that can be improved anymore, so continuing that thread is a bit pointless.

[meynaf]

Originally Posted by Thorham

Good point. If she were like that, than she'd be too much of a controll freak for me.

So far I didn't have the problem anyway. The few girls I've bring home seemed to like the miggy.

Originally Posted by Thorham

Good to know. I just hope I'm going to need to know!

Unsure, as I'm working on a rewrite right now...

Originally Posted by Thorham

Thanks for posting the code Shame it's not doing things the same way, but I guess it makes sense to me why it won't work now. This huffman decoding stuff is something I have to do for the png part, which is something I haven't really come round to doing yet. Of course, it doesn't have to be finished tomorrow. Guess I'm going to get to it soon. If you want to know why I've put if off, it's because I've been working on my Japanese dictionary program too much. Thank goodness that thing doesn't require much work anymore, as it's eating my time. You know how it is: Once you start you can't stop!

My huffman code uses one single table where the c code uses several. It is good for me as I'm really out of registers in this part.

Note : the code I posted here is (already) obsolete. I used the fact that A2 always points A1+$200 to use A1 instead (I've added $100 to the values which indicate we should peek in the second part of the table). Not sure if it's clear, but that means you can replace lea (a2,d5.w*2),a3 / move.w (a3,d1.w*2),d5 by lea (a1,d5.w*2),a2 / move.w (a2,d1.w*2),d5.

And, yes, once started I can't stop too. I started that rewrite to see how far I could get ; it's not the first time I try it but with my actual code and knowledge I think I can do it all.

Originally Posted by Thorham

That's great, you're making good progress. I suppose I can download that version from your site?

You really want to download that thing which does nothing but display an error message saying the next thing to do is not implemented yet ?
And which will be already outdated by the time you get it ?

Originally Posted by Thorham

But you don't need an entire os just to handle some hardware. The emulator should be able to handle the drivers. They'd have to be adapted for the emu, but that should still be possible.

And you will have to write several hundreds of drivers for your emulator.

Originally Posted by Thorham

Good luck.

Sure I will need it. But it also means you'll see me less often here.

Originally Posted by Thorham

You could just try to replace the fan. It's something I might do instead of buying a new unit. With a little luck, it doesn't need soldering.

I could try. But it's the only power supply I have so I prefer not to risk it.

Originally Posted by Thorham

Funny, I think it sounds kind of cool. It's really the writing system that fascinates me, hence the dictionary program.

What is your dictionary program exactly doing ?

Originally Posted by Thorham

Cockroach nest Three hundred bucks is too much money for any program to me, I suppose one tenth of that price would be a bit better

What has always amazed me is that it's the very first software company in the world who produces such (high-priced) bullshit.

Originally Posted by Thorham

Cut'n paste. I do that all the time. Must admit, though that oop saves you a lot of it sometimes.

You even cut'n paste code that wasn't yours at first place ?

[Thorham]

Originally Posted by meynaf

So far I didn't have the problem anyway. The few girls I've bring home seemed to like the miggy.

Really Wow! You'd think they would think 'What's the point?'.

Originally Posted by meynaf

Unsure, as I'm working on a rewrite right now...

Yup, that's enough to make my newly acquired knowledge useless again

Originally Posted by meynaf

My huffman code uses one single table where the c code uses several. It is good for me as I'm really out of registers in this part.

Note : the code I posted here is (already) obsolete. I used the fact that A2 always points A1+$200 to use A1 instead (I've added $100 to the values which indicate we should peek in the second part of the table). Not sure if it's clear, but that means you can replace lea (a2,d5.w*2),a3 / move.w (a3,d1.w*2),d5 by lea (a1,d5.w*2),a2 / move.w (a2,d1.w*2),d5.

I'm going to have to write one, too. I've finally done enough on my dictionary program to give that a rest. Since it's for png unpacking, it's probably not going to be compatible. Shame to have to have two of them in one viewer

Originally Posted by meynaf

And, yes, once started I can't stop too. I started that rewrite to see how far I could get ; it's not the first time I try it but with my actual code and knowledge I think I can do it all.

You can do it, don't underestimate yourself!

Originally Posted by meynaf

You really want to download that thing which does nothing but display an error message saying the next thing to do is not implemented yet ?
And which will be already outdated by the time you get it ?

On second thought, maybe not...

Originally Posted by meynaf

And you will have to write several hundreds of drivers for your emulator.

Or just reuse existing drivers. You would just have to pick the best ones. And when push comes to shove, you can always optimize the most important ones. But I do agree it would be a very laborious task to do alone...

Originally Posted by meynaf

Sure I will need it. But it also means you'll see me less often here.

Hey, 'less posting' equ 'more coding'

Originally Posted by meynaf

I could try. But it's the only power supply I have so I prefer not to risk it.

With two left hands for these things, I can see why. But I just can't resist to do it one day. I mean, the noise

Originally Posted by meynaf

What is your dictionary program exactly doing ?

It let's you look up the meanings and readings of the ideographic characters, including searching on grade, etc, and it has a word dictionary with similar features. I started doing it after finding tons of dreadful programs on the net, and thinking 'I can do that better'. Sounds familiar, doesn't it It's based on the free kanjidic and edict dictionaries, which are quite extensive. The thing to do now, is to write a search engine dll in c++ to avoid the clr, as the vb express version of the search algorithms are a little slow. I could use a hash table, but it would only make whole word search faster. A lot faster, but still, you can also search on any match, begins with and ends with, and I want those to be fast, too.

Originally Posted by meynaf

What has always amazed me is that it's the very first software company in the world who produces such (high-priced) bullshit.

Leave it to money hungry Bill Gates to start a company like that. You know when someone produces bovine excrement like ms dos, they're going to take crap software to a whole new level. Don't get me wrong, getting lot's of money is cool, but only if the product I would offer is of a high quality standard.

Originally Posted by meynaf

You even cut'n paste code that wasn't yours at first place ?

Not if I can help it, no. Using other peoples code is a no go area for me, unless the code is open source, and the program I'm writing is also open source. But I still don't like it, and will always try to solve something by myself first, even if it takes me a long time.

[meynaf]

Originally Posted by Thorham

Really Wow! You'd think they would think 'What's the point?'.

They could have thought "what's the point ?" if they had enough knowledge about computers, but they hadn't

Originally Posted by Thorham

Yup, that's enough to make my newly acquired knowledge useless again

Originally Posted by Thorham

I'm going to have to write one, too. I've finally done enough on my dictionary program to give that a rest. Since it's for png unpacking, it's probably not going to be compatible. Shame to have to have two of them in one viewer

You're right : there'll probably be no common code between jpeg and png. The bit reading stuff could have been compatible if there wasn't that marker hit detection stuff right in the middle of it.

Alternatively if you want to see some bit reading stuff for inspiration, you can find some in flac.s of my sound player.

Originally Posted by Thorham

You can do it, don't underestimate yourself!

Apparently yes. I have successfully displayed a grayscale picture.

Originally Posted by Thorham

On second thought, maybe not...

Now it does a little bit more but it's not much more useful.

Originally Posted by Thorham

Or just reuse existing drivers. You would just have to pick the best ones. And when push comes to shove, you can always optimize the most important ones. But I do agree it would be a very laborious task to do alone...

You can't reuse existing drivers. They're meant for a particular operating system, using them would be like porting a program to another plaform.

Originally Posted by Thorham

Hey, 'less posting' equ 'more coding'

That's right !

Originally Posted by Thorham

It let's you look up the meanings and readings of the ideographic characters, including searching on grade, etc, and it has a word dictionary with similar features. I started doing it after finding tons of dreadful programs on the net, and thinking 'I can do that better'. Sounds familiar, doesn't it It's based on the free kanjidic and edict dictionaries, which are quite extensive. The thing to do now, is to write a search engine dll in c++ to avoid the clr, as the vb express version of the search algorithms are a little slow. I could use a hash table, but it would only make whole word search faster. A lot faster, but still, you can also search on any match, begins with and ends with, and I want those to be fast, too.

If your dictionnary is fixed (does not dynamically evolve while running) then a dichotomic search will be better than a hash table. Just a suggestion

Originally Posted by Thorham

Leave it to money hungry Bill Gates to start a company like that. You know when someone produces bovine excrement like ms dos, they're going to take crap software to a whole new level. Don't get me wrong, getting lot's of money is cool, but only if the product I would offer is of a high quality standard.

Originally Posted by Thorham

Not if I can help it, no. Using other peoples code is a no go area for me, unless the code is open source, and the program I'm writing is also open source. But I still don't like it, and will always try to solve something by myself first, even if it takes me a long time.

I don't like using other people's code either. But the reuse of my own code is something I like a lot, look in my sysbtm.i and you'll see a lot of it !