- 精华
- 0
- 帖子
- 750
- 威望
- 0 点
- 积分
- 734 点
- 种子
- 0 点
- 注册时间
- 2006-8-16
- 最后登录
- 2007-8-2
|
[翻译]PSINext对Heavenly_Sword程序员的采访(技术化)
原文：http://psinext.e-mpire.com/index ... 712a227326d6f25b8a2
By Carl Bender - June 21, 2006 翻译by Roger Wang
PSINext对PS3有一些比较技术化的，比较让普通玩家望而却步的文章，这是其中之一。另外两篇我也希望有机会能翻出来。希望更多的玩家可以看看这样的东西。会很有收获的。
PSINext：高动态范围光照－或HDR（译者按：HDRL和HDR是两个相关，但完全不同的概念，前者建立在后者的基础上，但复杂得多）－是一个我们最近越来越经常听到的名词。你能解释一下HDR到底是什么东西，还有它到底对游戏玩家有什么意义么？
Marco：是的，这是一个现在十分常见并且经常被错用的名词。
当我们渲染一张HDR画面的时候，画面上每个像素的几乎所有信息都被保存了下来，所以我们可以用最完整的明暗和色彩信息来捕捉一个场景。糟糕的是，目前的低成本显示技术（常见的LCD，CRT等）无法正确的显示HDR画面，因此我们要通过一个名叫“色调映射”的过程来把我们的画面映射成一张LDR（低动态范围）画面，才能在屏幕上显示。
色调映射还可以模仿人眼对不同光照环境的适应过程。对于玩家来说，总而言之就是它可以带来更好的视觉体验，尽管游戏开发者还需要一段时间来学习它。
PSINext：你能举几个游戏中的场景的例子，来说明HDR的好处么？
Marco：相比LDR，任何很高和很低的亮度的场景，包括对比度很高的场景，都能从HDR获益。
HDR可以改善那种所有东西都显得太亮或太暗的的画面，从而保留画面所有的亮度和色彩信息。这是让画面显得生动和华丽的重要元素。
此外，特别的色调映射操作还能够给开发者一个额外的叙事工具，因为它可以模拟一些人眼的视觉行为。重要是，要知道HDR绝不仅仅是一些眩目的光晕效果－它要比这多得多。在我们的真实生活中，根本是没有LDR画面这么一回事的。
PSINext：现在人们一提到HDR的时候，就会想到GPU硬件上的FP16或FP32的HDR渲染支持。Ninja Theory为什么采用了NAO32，这能算得上是“真正的”HDR么？
Marco：FP16和FP32渲染格式让开发者可以去收集每个像素更多的信息（分别是每个像素8和16字节）；所以我们可以很简单的实现渲染和储存HDR画面。糟糕的是，这些帧缓存格式速度太慢了，因为它们需要更多的内存带宽和更多内存：一张FP16的4x反锯齿的720p图像要将近30MB的内存！
重要的一点是，只要我们能找到一种不损失太多信息的编码方式，那么怎么存储HDR图像并没有什么关系，RGB色彩空间用来编码HDR图像并不很有效，所以在做了一些研究之后，我们找到了另外一种有效得多的色彩空间。这个色彩空间叫CIE Luv，它用三个数值来表示色彩：一个是非标准化的用来表示色彩烈度的数值（亮度），另外两个则是0－1之间的标准化值。这个色彩空间的发明者Gregory Ward是HDR图像的开拓者之一，他在很多年前就发明了用这种色彩空间来存储HDR图像的格式LogLuv。我们为了满足自己的要求，在这个基础上做了一些改进
PSINext：这么说NAO32就是一种在保持图像质量的前提下节省内存的方法，是么？
Marco：没错。使用NAO32的基本目的就是用shader计算能力来换取内存空间和带宽（这些资源在主机上非常宝贵）。所以我们没有直接用FP16或FP32来存储HDR图像，而是利用了RSX的像素shader单元来把RGB色彩转换成CIE Luv色彩，后者只需要一个普通的RGB8的内存空间（每个像素4个字节，是FP16的一半）。
这种格式的质量相当出色。尽管它只用了普通HDR方法的一半内存空间和带宽，但它其实并没有在图像质量上做出任何牺牲。
不过这也没什么神奇的：我们只是把HDR的渲染代价从内存转移到了shaders上，所以我们的shaders变长了一点（3到5个周期）。我们觉得这是一个很好的交易。此外，它还能让我们在不能硬件支持FP16或FP32的反锯齿的GPU上实现HDR的反锯齿。
我们还开发了一种更快的3字节的格式，叫NAO24。它支持的HDR更窄，精确度更低。尽管它的质量在大多数时候足够好了，但是我们还是不愿意妥协，所以最后并没有采用它。
我最后的回答就是：NAO32绝对称得上是真正的HDR。
PSINext：NAO32在内存上的优势和无损的质量似乎很有推广的价值。你认为这种格式是不是可能进入桌面，甚至是微软或任天堂的主机？
Marco：任天堂的主机我不知道，不过在微软的主机上，NAO32可能一定优势，如果开发者想要一个比FP10质量高得多，但内存消耗差不多的格式的话。但是他们要付出一些shader能力，而且还会失去用eDRAM做混合的功能。所以目前来说，我觉得NAO32在RSX上要比在Xenos上更合适。
PSINext：我们已经谈了不少NAO32的优点了。那么缺点呢？
Marco：它也有缺点。GPU一般用硬件来做RGB空间的混合，所以对于NAO32的帧缓存，硬件混合就没法用了－会产生出错误的结果。。
克服这个问题有几种办法，比如说在一个像素shader里模拟混合操作，或者就在RGBA8帧缓存里用LDR来做混合。最好的当然就是GPU能硬件支持CIE Luv帧缓存。
PSINExt：在E3上玩了这个游戏之后，我们发现了大量的特殊效果，包括HDR和软阴影。你能告诉我们Ninja Theory为这个游戏选定的AA和AF设定么？
Marco：好。阴影是完全动态的，所有物体都能投射/接受阴影。软阴影是每个像素12个抖动取样。反锯齿是4X（多采样），但效果并不如我们想象的那么好；我们还会继续对其加以改进，希望在今后几个月里会有进步。
PSINext：分辨率和帧频的目标如何呢？
Marco：我们的分辨率目标是720p和4X AA，并且已经实现了。帧频的目标还没有确定。虽然E3上的demo是在30帧以上，但我想最后的游戏将会是30fps。这会让我们能加入更多的效果。
PSINext：NAO32的一大好处就是节省带宽。你觉不觉得RSX的128位内存带宽和缺少eDRAM会使其不如Xenos？
Marco：完全不是；事实上对于很多帧缓存效果，我觉得RSX要比Xenos更好一点。我不想谈论细节，只是想指出RSX其实连接了两条总线，而非一条。
PSINext：在和你的同时Dean和Mike在E3上谈过之后，我知道Ninja Theory现在在使用SPE来实现诸如音效，物理，动画之类的东西。那么你觉得现在Cell的潜力还有多少呢？
Marco：我现在没法给你任何准确的数字，但是我可以肯定的一点是，现在开发者们用到的还是Cell的冰山一角。
我们都需要学习Cell的工作方式和如何得到最好的性能。这不是一朝一夕的事情，但玩家们一定会得到一些真正的惊喜。
http://www.ps4focus.nl/sitee/nieuws/bekijk.php?id=3054
Quote:
LONDON--Four members of the Ninja Theory development team revealed a little bit more of the promising third-person action game Heavenly Sword during a presentation at the Game Developers Conference London today.
Duncan Frostick (combat code Ninja), Nina Kristensen (chief development Ninja), SaiTong Man (combat design Ninja), and Guy Midgely (lead animation Ninja) talked through some of the gameplay aspects and showed off some demo videos to prove their points.
Man explained that the team found inspiration for the upcoming PlayStation 3 game from kung fu movies and current popular fighting games. "We looked at games like Virtua Fighter and Soul Calibur because they have essentially very simple designs." Frostick added: "We worked on a flowing pose-to-pose animation. The inspiration came from kung fu films, not hard, Western-style animation."
Man said that one of the driving forces behind the game was to create something that looked "pretty" and action sequences that flowed together perfectly. "Way back when we first started working on Heavenly Sword, the main thing for me was aesthetics; it had to look amazing. Also, the gameplay had to be completely accessible to the casual gamer, but it had to have depth for the hardcore gamer."
Frostick revealed that the main character would become stronger and gain attack moves as the game progressed, which suggested the game will feature role-playing elements. "Like most fighting games, we worked on the basis that through the game, the bigger and better the moves become. There are two types of synchronised attacks. There are counters and what we call super styles. The super style stuff looks spectacular. These get more and more kind of ridiculous as you go through the game and get more and more powerful."
Midgely went on to talk about how important the flow of the game is to make physically impossible fighting moves look real. "We're taking reality and amping it up to the next level. The players always feel like they are in the game. You don't want to be enjoying playing a game, or watching a kung-fu movie, and think, 'Wait, that's not right.' It takes away enjoyment from the game."
The game, as has been seen from previous videos and demos, will be fast paced--Midgely confirmed that the animations will be running at 60 frames per second, while most games today run at 30. He added that there will also be a slowing effect within the game, which may work out something like 'bullet time,' now common in games.
No visuals of game bosses were shown, but the team was quick to assert that their moves will be equally impressive as those of the main character. Frostick elaborated on some of the boss-related elements: "There are some bosses who do truly spectacular moves. And there are armies, 1,000 people tearing through a valley trying to get you. There is also another element to the game, more cinematic elements, environmental interactions, power jumps, etc."
He went on to quantify the number of attack animations: "For the hero, there are literally hundreds of different attack animations. I'd say around 1,000 animations at the moment."
One revelation that surprised many attending the session was the lack of a collision-based system within the game, but Frostick defended the choice. "We use transitions to take the hero away from the wall into a safe area with cool animations. I'm pretty happy with that solution."
The quartet also explained the creative process for the game. "We were an autonomous unit and the studio heads have faith in what we do. We don't have to go through several people to get what we want. It's talking around the table rather than having to go through lots of red tape to make something happen."
Although some of the fighting sequences in the demo were obviously violent, including one in which a creature was sliced from the groin up, Midgely claimed that the game is likely to receive a T-for-Teen rating in the US.. |
|
川公网安备 51019002005286号
楼主
发表于 2006-11-19 13:35 · 北京
