Perihelion tutorial 6

From Atari Wiki
Revision as of 15:34, 12 October 2011 by Admin (talk | contribs)
Jump to navigation Jump to search

The Atari ST M68000 tutorial part 6 – of seeing behind the curtain of an execution and getting intimate with files


Hiya’all, it’s been a little while since the last tutorial. Mainly because I wanted to code a little bit for myself and not only write stuff. This tutorial will NOT be about scrolling, unfortunately, but it will cover the theoretical base which you’ll need to be able to do the scrolling as will be covered in the next tutorial. However, after this tutorial, you may figure it out by yourself. Of course, the thing you have to do to scroll, is to just move the correct screen memory bytes to the correct place. This will be covered in depth in the next tutorial, promise.

We’re now beginning to get past the most fundamental theory, and so our code is getting to be more and more advanced. This in turn means that often, a program will assemble without errors, but it still won’t work the way we want it to. Something somewhere is not as we thought it would be, a variable might not be assigned the correct value, a mathematical equation might not produce what we thought and so on; endless possibilities. This is where the debugger comes in. Debugger? says you. To illustrate, let me tell you this fairy tale.

In the olden days, there was a big computer. So big it was that two men could not put their arms around it. The computer stood in the big country that lies west of here, and all day long it crunched numbers. It was very happy. Then, one day, it could not crunch numbers any more, something was wrong and the computer fell sick. All the people in white robes, that saw to the computers every need, were greatly distressed. No one knew what was wrong. So, in a last desperate effort, they opened up the poor computer to have a look inside. They found that a little bug had flown in, and that was the root to the sickness. So, the people in the white robes removed the bug, and the computer was again healthy. It was all smiles and could once again crunch numbers all day long. Thus endeth the tale. (since this is a fairy tale, I make no claims that the exact facts are true, but like all legends, it contain a grain of truth)

Debugging, is the art of removing errors from source code. This is actually very hard, and one can probably be as skilled in debugging as writing code in the first place. Debugging usually takes at least half the time of developing a program, so good planning and lots of time in the debugger is a good thing indeed. Nowadays, bugs are errors in the source code, rather than actually physical bugs. Debugging is getting rid of bugs, creating error free code, and a debugger is a tool that helps you with this process. Devpac comes with a debugger, called MonST, I guess it stands for MONitor ST.

After you’ve assembled to memory, instead of pressing alt+x and run the program, you can press alt+m and run the MonST, henceforth refered to as the debugger. Lots of information will pop at you, and after you’ve come over the shock, you’ll start to make quick sense of it. There are three “windows”, areas rather, registers, disassembly pc and the memory. The disassembly pc area is your actual source code, the other two should speak for themselves. When you are in debugger mode, instructions will be executed one at a time, this allows you to see how each instructions change the content of memory and registers. I’ll go through each area and what you do with it.

Registers, here you have the content of all data registers, all address registers, the status register and the program counter. All values are given in hex, which makes every to digits one byte, and each digit one nibble. As you can see, there are eight digits for each data register, which makes sense since you can store a longword in a data register. When data registers are beginning to get filled with values, there will pop up some symbols, sometimes strange, to the right of the register. These symbols are the ASCII equivalents for each byte in the data register. We haven’t talked about ASCII I think, but it’s the way to represent characters with numbers I mentioned back in tutorial one. For example, the number $41 is the letter ‘A’.

The address registers are to the right of the data registers, and work pretty much the same. To the right of the address registers, are the memory content that the address register points to. Since there are four digits to every group, each group is a word. Thus, to the right of each address register, is the memory content of the first five words that the address register points to. To the right of the memory content, you’ll also see ASCII representations of the content, just as with the data registers.

Below the data registers, are the status register and the program counter. The status register haven’t been mentioned much either, but it takes note of several statuses of the ST, for now, it will probably be 0300 and you’ll se a ‘U’ to the right of it. The U means User mode, and that’s what we’re in now until we change it to Super visor. The status register will also keep track if a mathematical operation results in an overflow and so on. An overflow is when the number generated is bigger than can be stored, for example, adding two data registers with very big values will generate a value to big to store in one data register, so data loss will occur. Below the status register is the program counter, and to the right of the program counter you’ll see the instruction that it points to.

The disassembly area is the code you’re currently debugging. It will look just like your source code. You can scroll up and down the code, and a little arrow will indicate your current position. To execute a line of code, press alt+z, to skip a line of code, press alt+s. Usually, you’ll want to skip jumping into the initialise subroutine, because this takes some time and might also put the ST in low resolution, making it hard to see anything. You’ll usually want to go to the mathematical equations directly, to see what happens. There's also a very nice way to jump straight to a position of your choosing. You can put "flags" in your source code, by entering the "command" illegal, then, when in debugger mode, hit ctrl+r. This will execute all commands from your current position to the next illegal position, you'll have to skip past the illegal instruction to continue, using alt+s. A great way for executing an entire loop without stepping through it all.

The memory area is most interesting, this is where the entire content of the memory is listed. By pressing m, you can type in the name of any memory tag (variable) that you are using, and see what the memory that it points to contains. If you’re smart, you’ll immediately type in ff8240, which will take you directly to the palette. Unfortunately, that will get you little, since this is protected memory, you’ll only see *’s.

You can change between these areas by pressing tab, and you can only issue commands in the active area. When you are done debugging, you don’t have to wait for the whole program to execute and terminate, just hit control+c, twice. Now this is useful, right? The best way to get to the workings of the debugger is, like always in programming, to get to it; debug some simple piece of code and see what happens to the registers and memory. Oh, yes, in the memory area, you can also type in aN (where n is 0-7) to get directly to the memory area pointed to by an address register.

Now, onto file formats! A file is simply a collection of data. There really is no such thing as a .pi1 (Degas Elite) file, or an .mp3 file. A file contains data, so, this data is interpreted. Different things will happen depending on how you interpret the data. Let’s say, for example, that we have a file containing only a byte, and it holds this data

           %01000001   

Easy, says some paint program, these are the first eight pixels in monochrome mode. Pixel number 2 and 8 is supposed to be black, the rest are white. No, says the text editor, %01000001 is $41, which corresponds to ASCII character ‘A’. This is the letter A. Nonsense, says the home taxation program, %01000001 is a control code in my program that says this file represents a terminated account… and so on. Programs interpret files, and do something with the information. Since programs are also files themselves, interpreted by the operation system, which is itself also files more or less, the whole shit is build on subjective opinions on what to do with the data presented.

Given the information above, one might think that it’s a good way to know how different programs interpret data, this is the knowledge of file formats. In order to understand this, we will examine a very simple file format, the Degas Elite .pi1 file format. It’s almost to simple really, but it’s useful and we’re going to use it in our next tutorial. Usually, files have so called file headers, which give some information about the file. For example, a Windows BMP file, starts with the ASCII codes for ‘B’ and ‘M’, which makes sense and gives a signal of what kind of file it is. It’s of a little nerdy interest to know that each .exe file on the PC, starts with the ASCII codes for the letters ‘MZ’, which was some hot shot in Microsoft back when they defined the file format (and perhaps still). A good example of a file header could perhaps be the resolution of an image, or the font type in a word processor file.

In order to examine files correctly, we need a so called hex editor. A normal text editor will not do, since the text editor would interpret data as ASCII code, we want a program that just presents the data in the file, and does not interpret it in any way. With this hex editor, you can “hack” files yourself. Say, for example, that you want a program that converts one graphic file format to another, you’d need a knowledge of both file formats. Sit down with a paint program, and a hex editor. Do some small changes in the paint program, and watch what’s changing in the file with the hex editor. This is tedious work, at best, and you’re probably better off trying to locate the information somewhere. So, in order for you to begin and try out your efforts, I will tell you how the .pi1 files look like.

First, there are two bytes giving the resolution, in low resolution, it's just 0, in medium, 1 and in high resolution 2. Then comes 32 bytes containing the palette data for the picture. After that comes the pixel information, looking exactly the way it does in the screen memory. After the pixel information, there are some 32bytes containing animation data, but we don't have to care about that. And that is that. Very simple file format indeed. So, how big is a .pi1 file then, only knowing the above? 32066 bytes. 32000 bytes for the pixel information, 32 bytes for the palette, two extra bytes in the beginning of the file plus the uninteresting 32 last bytes. Here's a little program that will display a .pi1 file.


jsr initialise movem.l picture+2,d0-d7 put picture palette in d0-d7 movem.l


d0-d7,$ff8240


move palette from d0-d7


move.w #2,-(a7) get physbase trap #14 addq.l #2,a7 move.l d0,a0 a0 points to screen memory move.l


#picture+34,a1


a1 points to picture


move.l #7999,d0 7999 longwords to a screen loop move.l (a1)+,(a0)+ move one longword to screen dbf


d0,loop


move.w #7,-(a7) wait keypress trap #1 addq.l


#2,a7


jsr


restore


clr.l -(a7) trap #1

include initlib.s

section data picture incbin jet_li.pi1


There are three new instructions here, movem and incbin and include. Include is the easy one, just consider it as though you had pasted the entire contents of the initlib.s file on the include line. As you will see, when you assemble the code, this takes a while since the Atari needs to read the file each time. Therefore, I strongly suggest you actually do paste the file in, instead of just including it. Your choice.

Incbin, as you may have guessed, is the way to include files, they fall under the section data. This puts the entire contents of the file in memory. In this particular case, I put the entire contents of the .pi1 file called jet_li.pi1 at the memory position I choose to call picture. You can achieve the same result by hand copying the content of jet_li.pi1. Something like

picture dc.b 0,0,0,0,$07,$11 …(this is the beginning of the file)

Movem MOVEs Multiple data from memory to registers or the other way around. It can only move words and longwords. As you can see, I move the memory from picture+2 into the data registers. This is great since all eight data registers can hold all in all 32 bytes of data, since each colour is 2 bytes of data, this means that the entire palette of 16*2 bytes of data fits precisely into the eight data registers. The reason for picture+2 is that we want to skip the first two bytes, since they only contain resolution information. After filling the data registers with the palette, we just smack it in at the correct starting address.

Then, it’s a question of putting the screen memory pointer in a0, and the start of the pixel part of the picture in a1. The picture+34 is because this is where the pixel part starts, 2 resolution bytes plus 32 palette bytes is 34 bytes that should be skipped in order to reach the pixel part. As shown in the previous tutorial, the screen size is 8000 longwords. I just loop through that amount, copying the content from the picture into the screen memory. Easy? This is a small loader for .pi1 files. If you assemble this piece of code as a .prg file, you’ll notice that the program size will be 32652. Most of this is the .pi1 file itself, our added code is only 32652 – 32066 = 586 bytes. We now have a self-loading .pi1 image, nice.

If you think it would be amusing, you can add this little loader to all your .pi1 files, in this way, you’ll never have to go through Degas to watch them; they load themselves. Of course, you’ll get a .prg file instead of a .pi1 file, meaning that you can’t edit it with Degas. But then you could write your own program for extracting the image information and turn it into a .pi1 file again. Fun, right? Note; you don’t have to keep the original .pi1 file for this “loader” to work, since the .prg file contains the data it needs for the image.

While we’re on the topic, I will mention, briefly, compression. You must know what file compression is, it’s making a file smaller, but usually useless, until you decompress, or unpack, it again. How does this work? The file can’t just shrink, can it? Well, more or less, it actually can. Consider this information.


           %00000000
           %11111111


The first byte is all 0’s, and the second one all 1’s. Suppose we replace the information given with


           08
           18


and tell the program that after each 1’s or 0’s, there will be a number that tells how many 1’s or 0’s there will be. If we have a file with big areas of similar data, for example 50 bytes of 0’s and then 70 bytes of 1’s, this so called compression algorithm would compress this information into four bytes. It would look like this


           050
           170


or, just to give you some bit mathematics, we say that the high bit of each byte controls whether it should be 1’s or 0’s, and the next seven bits tell how many of each kind should follow, it would look like this.


           %00110010                50 0’s
           %11000110                70 1’s


That was that on compression. The above is a very simple compression algorithm and if you use it, you may end up with files bigger than they were from the beginning. I know file compression was a bit sketchy, but if you get the part of how files work, the compression part shouldn’t be that hard. Also, file compression might be covered more extensively later. So far, I know very little myself since I haven’t used it for anything. I have no idea how good file compression algorithms look or anything, so don’t ask. This is just the theoretical base. Study carefully, since I’m going to use a .pi1 file for the font in the upcoming scroller.


Warrior Munk of poSTmortem, 2002-04-22 “Great! I love fighting.”

- Fong Sai-Yuk

Last edited 2002-06-14