i
do you want to make fully optimised modular os?
if yes this tread is for you

optimisation is the ability of shorting the time of execution and the size of the code

so you need to show me only one optimised code and lets go
ok!!
if you can write a fully optimised modular function i invite you to help

i know it's ridiculous but it's the beginning of my project

i am working on a command interpret to use the memory of the hard disk and navigate in its file system
for the moment i'am limited by the bios capabylities
to make my disk driver i need some documents about the ATA to Memory with DMA 16bits
i don't find anything about DMA 16bit specifications so if you have some
i thank you

and some other about the different files systems FAT,FAT32,NTFS because for the moment i am trying to create my own but it's not compatible with the partition tables and the current OS

I did, thats why i made one
http://www.dex4u.com/

Good luck.

i'll test your dex4u and after i'll see if i am interested in your project
today i'am develloping a file system fully protected it's hard but possible
bye

is this somehow alive?

5 days... i quess not. what to do with this? even the name doesn't tell me anything about what this is supposed to be

Joined - posted - got lost
mark it [DEAD]

as it was the summer and i haven't internet at home i cannot follow the news of fasm site

i just can say that i have found some data structures totaly evolutive and backward and forward compatible

for exemple with the button function:


button:
.x dd ?
.y dd ?
.xl dd ?
.yl dd ?
.size dd .end-.size-4
data
.end:

with structures like this, the code can be updated easily



for the moment i am working on a routine to make windows and panels
for this i use a meta structure with a routine who organize the pointers to windows and panels structures and makes focus on the instance on the bottom of the meta structure because i display the instances in the order of the meta structure

so the top most window is the last pointer

other ideas but no more time to devellop by
and thank you for the insterest you give to me

goto assembleur en france

I'm rather interested in the most optimal way of passing these x1,y1,x2,y2 to programs, kernel, API, etc.

One not-so-ingenious way is to pass it through XMM:
xmm0=param1 shl 16*(N-1)+param2 shl 16*(N-2)+....+paramN shl 16*0
you can extract words with: pextrw r32,xmm0,{word index}
SSE is fast, but not existent on all systems and its limited to 8*8 (or 16* parameters, but not N parameters

one year, and it is more achieved.
you can see all the work made on this project in my various posts.

i now that exemples are very poor, i don't code so many usefull codes, but it will become ok one day, if i only works on it frequently.

it is hard to follow a straight line, without zig zags.
i cannot code (not so poor code) more than 3 hours now, at the start, i coded (very poor code) more than 10 hours in one pass.

first very important thing.
the first dword, it is always the function index or pointer.
and the second is the start of function specific structure.

for exemple, a graphic node can call non graphic items.
and items are not always with the structure i used at the begining.
there was many steps before this first "ok" version 0.01 oolib.

then, it is possible to contain items into a windows, compose windows with various items. like buttons, faders, images, boxes, lines, collisions, cursors, oscilloscopes, etc...
and all is modular.
it need now to be fully implemented, with file system r/w, memory management, item creation, etc...

it need at least 2 years to be able to code a serious thing with this lib, but when this lib will be ok, it will help to fast create interactive programs and gui.

From what I've seen of the benchmarks, having ring 0 provide graphics primitive routines is very inefficient. Each call takes hundreds of cycles up just doing cache/page flushing and loading up the instruction pipeline.

Instead, I've understood it to be much faster to build a todo list in ring 3, pass it to ring 0, and have mostly the graphics card do it. The complicated part of this is that each graphics card will do it differently, and unless you've got experience with a broad range of cards as you try to abstract it you'll get some things too specific to your card and some things too general.

My preference for a todo-list approach is to define type (rectangle, triangle, etc), and then a variably sized structure. The type should define the size being used so you don't need to pad it out (the data is packed).

In lieu of such a todo-list approach, the next best thing is to serve up a framebuffer. I personally agree with DirectFB's primitives for this purpose, though it is a bit abstract.

It's my opinion that if someone's doing graphics, they're technically inclined enough they can cope with passing an array* around and don't need it wrapped in a C++ class.

I believe, edfed, that you're correct in approaching the OS as a set of discrete working parts. The graphics system does not need any more tight interface to the CPU or memory than any other system.

I personally believe that Xen is the new multiplexing layer (exokernel), and windows and linux are just overstuffed libOS's. With that in mind, what specialized systems could you provide libraries for inside the Xen Hypervisor?



Regards,
Dan

And decent graphics APIs tend to be written so this command-buffer system is possible to implement - GDI, OpenGL, DirectX are all proper abstraction (GdiFlush(), GlEnd(), ...) - whether an implementation of the API does i this way is another matter.

From a quick glimpse of the Intel integrated graphics documentation, it also seems like these command buffers are DMA'ed to the card, and I wouldn't be surprised if other vendors use the same approach.

lucky guess on my part.

"Computer optimization is an exercise in caching"

Hi there!

> dd 0x90909090 ; problem solved

How can you distinguish "nop" from "xchg ax, ax"?

Sincerely yours,
Sergio.

If it is 32 or 64 bit code then the encodings are different.

It doesn't matter. It was chosen so it doesn't matter.

0x90909090 is a philosophy. If you can't open it, you don't own it. It's about controlling your software, and not the other way around.

Nowadays the game between the reversers and the obfuscators has become more complex. One can't just use 0x90909090 anymore as often code is scrambled, junked, xor'd or whatnot. (who else noticed previously 100mb programs inflating to 2gb?).

Ultimately the reversers always win, because programs are boxes of data that must ultimately partially exist in plaintext form when unfolded in a way identical to the target platform.

I shall admit to the near-complete success of a solution; however, put in mind that closed source software is a multi-billion dollar industry as I refrain from disclosing my method and divulge a sense of paranoia about the issue.

Best Regards,
Dan

What I mean was that (in 16-bit) nop is always db 0x90, and xor ax,ax can be db 0x90 or db 0x87,0xC0

Typo: he means "xchg ax,ax" here.