News:

Masm32 SDK description, downloads and other helpful links
Message to All Guests
NB: Posting URL's See here: Posted URL Change

Main Menu

We are hobby coders

Started by jj2007, May 15, 2021, 07:59:58 AM

Previous topic - Next topic

jj2007

Just found this video explaining what Intel scientists plan for GTA V... impressive :rolleyes:

HSE

Equations in Assembly: SmplMath

Biterider

Hi
Incredible amount of work for just a game...

Biterider

LiaoMi

Quote from: jj2007 on May 15, 2021, 07:59:58 AM
Just found this video explaining what Intel scientists plan for GTA V... impressive :rolleyes:

This is not a difficult task, multi-stage image processing according to the rules. Hopefully they won't write this in python :biggrin:

https://www.youtube.com/watch?v=pp7HdI0-MIo
https://www.youtube.com/watch?v=MjViy6kyiqs
https://www.youtube.com/watch?v=eFK7Iy8enqM
https://www.youtube.com/watch?v=KZ7BnJb30Cc
https://www.youtube.com/watch?v=0X1RtXCvPFQ
https://www.youtube.com/watch?v=rkSzRbM4lIM

To do this, you need to start using engines that will definitely not be in assembler - http://masm32.com/board/index.php?topic=9260.0
The question is what place does assembly language occupy in graphics processing, neural networks, and analysis of data structures.

Basic work - Michael Abrash's Graphics Programming Black Book Special Paperback – 1 Sept. 1997
Print length - 1342 pages
Michael Abrash's Graphics Programming Black Book is a collection of the author's previous books on assembly language and graphics programming, as well as past columns for Dr. Dobb's magazine. Though much of the book (such as 8088/286/386 assembly language optimization and VGA graphics programming) is out-of-date by now, the reader can see some of the larger tendencies in the author's work over the years--a reliance on profiling in order to test code, and creative thinking to improve performance wherever possible. This text features assembler optimization for a variety of problems including searching algorithms, and records the author's approaches to optimizing code for the evolving line of Intel CPUs, from the 8088 on to the early Pentium lines. The last few chapters of this book are more relevant, and include a series of explorations of some of the technology behind the popular Doom and Quake 3-D games by id Corporation (where the author worked). Optimized solutions to 3-D graphics problems from texture mapping, hidden surface removal, and Binary Space Partitioning (BSP) trees are explained. Current gaming and 3-D technology, such as Direct3D and VRML is left out, but it's clear that game programmers like the author will continue to push the limits of current hardware technology in inventive ways. This book is clearly targeted at game developers and serious assembly language programmers, not for the general reader.
PDF - https://archive.org/details/gpbb20/gpbb0 or https://www.drdobbs.com/parallel/graphics-programming-black-book/184404919
HTML - https://www.jagregory.com/abrash-black-book/
Epub - https://github.com/jagregory/abrash-black-book/releases/download/v1.1/black-book.epub
Markdown source for Michael Abrash's Graphics Programming Black Book - https://github.com/jagregory/abrash-black-book
CDROM ISO - https://archive.org/details/blackbook_201910, https://archive.org/download/blackbook_201910

Does Google implement TensorFlow in Assembly language?
https://www.quora.com/Does-Google-implement-TensorFlow-in-Assembly-language

QuoteBut in contrary to Antonio Nesic it is not, because an Assembly language base of AI would not speed that whole thing up by factor 5 to 1000 (~30 optimizing C->ASM, ~30 HLL->C, both lower estimate, ~200 for cache optimization and tight ASM loops inside cache vs distributed, scattered OOP code, ~factor 10k if databases are involved, estimate of 1k is very conservative). But because people are ignorant and especially when it comes to AI you have all those HLL freaks, that actually have no idea how slow their toy and prototype-languages make their project.

And AI almost never sees any optimization in it's life. I worked around '96 on a C++ based semantic network software ("Peirce") at the Uni-HH. And the crap I had to see was really bad and unbearable code.

Well, it worked, but it was some of the worst code I've ever seen.

Has an AI ever been written completely in assembly language?
https://www.quora.com/Has-an-AI-ever-been-written-completely-in-assembly-language

As with GPUs, neural processors will require their own language to implement the logic.

HSE

Quote from: LiaoMi on May 15, 2021, 09:26:42 PM
Has an AI ever been written completely in assembly language?

http://masm32.com/board/index.php?topic=8583.0  :biggrin:
Equations in Assembly: SmplMath

LiaoMi

Quote from: HSE on May 16, 2021, 01:45:29 AM
Quote from: LiaoMi on May 15, 2021, 09:26:42 PM
Has an AI ever been written completely in assembly language?

http://masm32.com/board/index.php?topic=8583.0  :biggrin:

Hi HSE,

cool project, at that time I did not have time to test it properly.

QuoteHas an AI ever been written completely in assembly language?
Probably it meant a fully scalable system. In a system where you can dynamically perform a variety of calculations with the ability to connect to other projects.

HSE

Quote from: LiaoMi on May 16, 2021, 03:13:30 AM
QuoteHas an AI ever been written completely in assembly language?
Probably it meant a fully scalable system. In a system where you can dynamically perform a variety of calculations with the ability to connect to other projects.

:biggrin:  Yes. It's a wrong question. Who maked that question for sure don't know that AI it's a very generic concept that include from relativily simple things to very complex ones.
Equations in Assembly: SmplMath

daydreamer

LiaoMi
producing things like several 500poly meshes and some textures are hardly worth put thru all kinds of techniques of reducing it to only visible polys on modern hardware,you probably faster to let gpu brute force
iits moved from cpu caps to get gpu caps,to see if it has instancing and other things and lfast enough and enough vram to handle the game

so I dont have the productivity of a 100 member development team,to make a very big game,to fill many GB data,so I never can make anything that use full potential of a 2021 gaming pc

but I enjoyed assembly size reduction from start and now its also fun to first make lowpoly 3dmodels and reduce polys

but use highend 2021 gaming pc,helps when developing 3d be faster and easier

AI ala neural networks emulating,wonder if you add a drunk macro where it selfadjusts you could emulate drunk behaviour for example a human 3dmodel or robot walking around controlled by AI,is adjusted by drunk macro to overcompensate,you would see "a drunk walking around"? :badgrin:
my none asm creations
https://masm32.com/board/index.php?topic=6937.msg74303#msg74303
I am an Invoker
"An Invoker is a mage who specializes in the manipulation of raw and elemental energies."
Like SIMD coding

LiaoMi

Why Deep Learning Needs Assembler Hackers - https://petewarden.com/2017/01/03/why-deep-learning-needs-assembler-hackers/
An Engineer's Guide to GEMM - https://petewarden.com/2015/10/25/an-engineers-guide-to-gemm/
Why GEMM is at the heart of deep learning - https://petewarden.com/2015/04/20/why-gemm-is-at-the-heart-of-deep-learning/

LiaoMi

Quote from: daydreamer on May 16, 2021, 08:42:07 PM
LiaoMi
producing things like several 500poly meshes and some textures are hardly worth put thru all kinds of techniques of reducing it to only visible polys on modern hardware,you probably faster to let gpu brute force
iits moved from cpu caps to get gpu caps,to see if it has instancing and other things and lfast enough and enough vram to handle the game

so I dont have the productivity of a 100 member development team,to make a very big game,to fill many GB data,so I never can make anything that use full potential of a 2021 gaming pc

but I enjoyed assembly size reduction from start and now its also fun to first make lowpoly 3dmodels and reduce polys

but use highend 2021 gaming pc,helps when developing 3d be faster and easier

AI ala neural networks emulating,wonder if you add a drunk macro where it selfadjusts you could emulate drunk behaviour for example a human 3dmodel or robot walking around controlled by AI,is adjusted by drunk macro to overcompensate,you would see "a drunk walking around"? :badgrin:

Hi daydreamer,

https://www.sciencemag.org/news/2018/04/watch-artificial-intelligence-create-3d-model-person-just-few-seconds-video
https://www.youtube.com/watch?v=nPOawky2eNk
https://www.dpreview.com/news/6843849919/this-ai-that-can-generate-a-3d-walking-model-from-a-single-still-image-or-painting
https://www.youtube.com/watch?v=G63goXc5MyU

It seems to me that today no one does it manually, the finished models are limited to the surface topology, and the neural network works with this later.