Vyznam dual core

[Petrik]

Podle Eagla si zasifrovany 2GB soubor na kompu s 1GB RAM nejenom neotevru, ale ani neoscanuju antivirem. S tim odposlouchanim 1/10 sifrovane komunikace a nasledneho dekodovani jsem to tedy nejak nepochopil, ale to bude asi tim, ze mam tu snizenou inteligenci. Ja myslim ze sifrovancimu algoritmu je uplne sumak, zda sifruje telnet komunikaci, komunikaci po nejakem portu, stremovanou hudbu ci nejaky soubor...

Ocividne stale odmitas precist si ten clanek o valve engine, tak alespon uryvek, abys tu porad nepsal, ze to nejde:
http://arstechnica.com/articles/paed...ulticore.ars/1

Almost all game programming, from the simplest BASIC game all the way up to Half Life 2, uses a main game loop from which all events are calculated. The drawing portion of the main loop used in Valve's games broke down roughly as follows:

1. Build world lists (background, scenery, etc.)
2. Build object lists (players, nonplayer characters, items)
3. Graphical simulation (particles, ropes, sprites)
4. Update animations (such as skeletal animation used for character movement)
5. Compute shadows
6. Draw (send display commands to the graphics card)

This list had to be executed once for every "view" currently visible in the game: once for the player's own point of view, and again for every water reflection or displays on in-game TV monitors.

The new pipeline was rearranged as follows:

1. Construct scene rendering lists for multiple scenes in parallel (world, water reflections, and TV monitors)
2. Overlap graphics simulations
3. Compute character skeletal transformations for all characters in all scenes in parallel
4. Compute shadows for all characters in parallel
5. Allow multiple threads to draw in parallel (this required a rewrite of the graphics libraries that live directly above Direct3D)

The new pipeline featured so much multithreading that special efforts were needed to avoid the dreaded deadlocks and other pitfalls mentioned earlier. To accomplish this, programmer Leonard made use of a technique called lock-free algorithms. He implemented a spin lock to replace the more traditional mutexes (mutual-exclusion algorithms) and semaphores that are used to flag threads as being "safe" to multithread. The spin loop utilizes a new interlock instruction that is built into the CPU. However, there were still too many deadlocks. Tom examined the threads' activity with profiling tools, and it turned out that 95 percent of the time the threads were reading memory while only spending 5 percent of their time writing. A read/write lock allowed many threads to read the same bit of memory, but only one thread to write to it.

Zaver:
The end results of Valve's efforts were even better than they had initially hoped. Not only was the speedup on the four-core Kentsfield chips nearly linear in most cases (an average of 3.2 times over a single core) but having four CPUs instead of two made it possible to do things that simply couldn't be done with fewer cores.

Porad si myslis, Svaco, ze AMD neudela QC do destopu? Jinak treba tebou vzpominany virtualdub JE MT.

[Eagle]

1. Construct scene rendering lists for multiple scenes in parallel (world, water reflections, and TV monitors)

Když tam víc scén nebude, nárůst výkonu = 0.

2. Overlap graphics simulations

Tohle nevím, co znamená.

3. Compute character skeletal transformations for all characters in all scenes in parallel

Tohle se dá přeložit více způsoby. Pokud se tím myslí dělat sekvenčně v rámci jednotlivých scén, tak opět více scén = nárůst výkonu 0. Pokud se tím myslí paralelně i v rámci jednoho světa, tak co se asi tak stane, když se dvě postavy srazí? Zadělávání si na deathlock.

4. Compute shadows for all characters in parallel

A když se budou stíny překrývat, tak co? Opět zadělávání si na livelock nebo deathlock.

5. Allow multiple threads to draw in parallel (this required a rewrite of the graphics libraries that live directly above Direct3D)

Kreslení je limitováno rychlostí nanášení textur grafickou kartou.

The new pipeline featured so much multithreading that special efforts were needed to avoid the dreaded deadlocks and other pitfalls mentioned earlier. To accomplish this, programmer Leonard made use of a technique called lock-free algorithms. He implemented a spin lock to replace the more traditional mutexes (mutual-exclusion algorithms) and semaphores that are used to flag threads as being "safe" to multithread. The spin loop utilizes a new interlock instruction that is built into the CPU. However, there were still too many deadlocks. Tom examined the threads' activity with profiling tools, and it turned out that 95 percent of the time the threads were reading memory while only spending 5 percent of their time writing. A read/write lock allowed many threads to read the same bit of memory, but only one thread to write to it.

Tohle už jsme tady probíralo a vyšlo nám, že buďto je to udělané jinak nebo je to totální kravina, protože načítání zastaralých dat by byla prasárna nade vše.

Zaver:
The end results of Valve's efforts were even better than they had initially hoped. Not only was the speedup on the four-core Kentsfield chips nearly linear in most cases (an average of 3.2 times over a single core) but having four CPUs instead of two made it possible to do things that simply couldn't be done with fewer cores.

Neříká nic o tom, nakolik je to čisté.

ad Half-Life: Hry jsou především o tom, jak to zaonačit, aby to připomínalo realitu, protože to prodává a generuje prachy. O kvalitě programování to už dávno není. A tenhle úryvek v podstatě jen říká, že budou chtít cosi přidat, což přináší víc prachů a ztěžuje život konkurenci. Neříká to nic o tom, zda je to na současném HW proveditelné bez prasáren.