Comments on: Graphics Tech: Shadow Maps (part 1) http://the-witness.net/news/?p=77 An exploration-puzzle game on an uninhabited island. The development blog. Tue, 07 Sep 2010 08:17:02 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-507 Jonathan Blow Fri, 25 Jun 2010 05:53:50 +0000 http://the-witness.net/news/?p=77#comment-507 In this game, we don't have a dynamic day cycle, so I just don't worry about it. I have noticed that for example Assassin's Creed 2 does what you are talking about (with the accompanying pops). If you are willing to burn some memory and bandwidth, then you can of course just keep the most recent two sets of shadow maps around and interpolate between them to get rid of those pops. I think that would feel pretty stable. In this game, we don’t have a dynamic day cycle, so I just don’t worry about it. I have noticed that for example Assassin’s Creed 2 does what you are talking about (with the accompanying pops). If you are willing to burn some memory and bandwidth, then you can of course just keep the most recent two sets of shadow maps around and interpolate between them to get rid of those pops. I think that would feel pretty stable.

]]> By: BoyC http://the-witness.net/news/?p=77&cpage=1#comment-506 BoyC Thu, 24 Jun 2010 14:58:32 +0000 http://the-witness.net/news/?p=77#comment-506 Very nice article, there are too few of these - when you're new to the subject you need to do a lot of digging first to find something this clear :) I was wondering how you'd handle the stability issues with a dynamic light say a day-night cycle? I couldn't find anything on that subject for now. We have a CSM implementation for our outdoor scenes with dynamic lighting and the best solution I could find around the issue was to only update the shadow casting light direction every 15 seconds or so. This introduced highly visible pops in the shadows of course, but it's a lot better than the constant shimmering from the moving light. Very nice article, there are too few of these – when you’re new to the subject you need to do a lot of digging first to find something this clear :)
I was wondering how you’d handle the stability issues with a dynamic light say a day-night cycle? I couldn’t find anything on that subject for now.
We have a CSM implementation for our outdoor scenes with dynamic lighting and the best solution I could find around the issue was to only update the shadow casting light direction every 15 seconds or so. This introduced highly visible pops in the shadows of course, but it’s a lot better than the constant shimmering from the moving light.

]]> By: Tim http://the-witness.net/news/?p=77&cpage=1#comment-397 Tim Sun, 23 May 2010 10:16:34 +0000 http://the-witness.net/news/?p=77#comment-397 Great read and is helping me a lot. I've implemented a PSSM, but I find the shimmering when I move or look around annoying. This is exactly what I needed to move in the right direction. Please keep posting :) Great read and is helping me a lot. I’ve implemented a PSSM, but I find the shimmering when I move or look around annoying. This is exactly what I needed to move in the right direction.

Please keep posting :)

]]> By: Sebastian Sylvan http://the-witness.net/news/?p=77&cpage=1#comment-170 Sebastian Sylvan Sat, 27 Mar 2010 23:33:34 +0000 http://the-witness.net/news/?p=77#comment-170 Jonathan, the big win is that we get pretty good shadow map resolution, the downside is that your transition regions are circular (in principle, for any given view it's obviously a box since the shadow space will clamp to the sphere) so you could get lower res shadows around the edges of the screen compare to the centre for objects at the same distance but to be honest that's actually pretty awesome because important stuff generally happens near the centre of the frustum so you'd want to skew resolution to that area, and again for any given view you typically get the shadow frustum for the higher res cascade cover a large part of the "low res" region in the overlapping area anyway so it's not generally a problem except for very specific views (this advantage would go away if you just test the bounding sphere). For testing the coordinates I just transform into the space of the first shadow so that the x,y coordinates are in -1,1 range, and the z is in 0,1, then I take the abs (free) which pushes all valid coordinates to the 0,1 range, any invalid coordinate would now be >1, and then I do max4 (Xbox 360) to get the biggest one which avoids checking each of them, if this is greater than 1, then we're outside the frustum. Currently I just transform the coordinate again for the second cascade (we're pretty fetch bound so no need to get fancy with the ALU cycles at the moment), but in since your cascades are all aligned the differences between coordinates in one of them to the next is just a scale and offset, so a single mad would actually do to move to the next one. Jonathan, the big win is that we get pretty good shadow map resolution, the downside is that your transition regions are circular (in principle, for any given view it’s obviously a box since the shadow space will clamp to the sphere) so you could get lower res shadows around the edges of the screen compare to the centre for objects at the same distance but to be honest that’s actually pretty awesome because important stuff generally happens near the centre of the frustum so you’d want to skew resolution to that area, and again for any given view you typically get the shadow frustum for the higher res cascade cover a large part of the “low res” region in the overlapping area anyway so it’s not generally a problem except for very specific views (this advantage would go away if you just test the bounding sphere).

For testing the coordinates I just transform into the space of the first shadow so that the x,y coordinates are in -1,1 range, and the z is in 0,1, then I take the abs (free) which pushes all valid coordinates to the 0,1 range, any invalid coordinate would now be >1, and then I do max4 (Xbox 360) to get the biggest one which avoids checking each of them, if this is greater than 1, then we’re outside the frustum.

Currently I just transform the coordinate again for the second cascade (we’re pretty fetch bound so no need to get fancy with the ALU cycles at the moment), but in since your cascades are all aligned the differences between coordinates in one of them to the next is just a scale and offset, so a single mad would actually do to move to the next one.

]]> By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-169 Jonathan Blow Sat, 27 Mar 2010 22:05:16 +0000 http://the-witness.net/news/?p=77#comment-169 Hmm, one could actually just test against the sphere instead of testing texture coordinates. I wonder if that would be faster. Hmm, one could actually just test against the sphere instead of testing texture coordinates. I wonder if that would be faster.

]]> By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-168 Jonathan Blow Sat, 27 Mar 2010 21:39:01 +0000 http://the-witness.net/news/?p=77#comment-168 Sebastian, that's an interesting way to think about it that obviously works. It might introduce some further dependency on camera orientation, but that already exists in the Valient scheme if one uses view-depth as the way to decide which shadow map to use (as suggested in the GPG6 article). Did you guys find a big win doing it this way? In terms of checking the coordinates... how do you do that efficiently? You need to check if either u or v is outside of either 0 or 1, right? I guess if you rotate the global shadow map depending on which way the camera is facing, you might be able to just test against 1. But that may interfere with the other optimization that I am going to post about next... Sebastian, that’s an interesting way to think about it that obviously works. It might introduce some further dependency on camera orientation, but that already exists in the Valient scheme if one uses view-depth as the way to decide which shadow map to use (as suggested in the GPG6 article).

Did you guys find a big win doing it this way?

In terms of checking the coordinates… how do you do that efficiently? You need to check if either u or v is outside of either 0 or 1, right? I guess if you rotate the global shadow map depending on which way the camera is facing, you might be able to just test against 1. But that may interfere with the other optimization that I am going to post about next…

]]> By: Sebastian Sylvan http://the-witness.net/news/?p=77&cpage=1#comment-166 Sebastian Sylvan Sat, 27 Mar 2010 17:26:33 +0000 http://the-witness.net/news/?p=77#comment-166 The way we arrange our bounding spheres is that we don't give you a splitting plane to specify them, the only parameters we give the artists is the radius of each bounding sphere. Then we arrange each of them so that their "back" touches the far intersection of the previous one with the frustum. This way you get minimal overlap between bounding spheres, while not having any gaps between them. The important realisation was that we don't really care that much about having a straight splitting plane involved here at all, so the important thing isn't to match a sphere to splitting plane, the important thing is to push all of our spheres forward as much as possible while not leaving any gaps. So the first sphere is pushed forward as much as possible while still containing the near plane. Then we compute the far intersection of this sphere with the frustum, and the second sphere is pushed forward as mush as possible while still containing that intersection, and so on. In the shader we just check the shadow-space coordinate for each cascade to see if we need to move to the next one, so you always choose the higher res map when there's overlap. The way we arrange our bounding spheres is that we don’t give you a splitting plane to specify them, the only parameters we give the artists is the radius of each bounding sphere. Then we arrange each of them so that their “back” touches the far intersection of the previous one with the frustum. This way you get minimal overlap between bounding spheres, while not having any gaps between them.

The important realisation was that we don’t really care that much about having a straight splitting plane involved here at all, so the important thing isn’t to match a sphere to splitting plane, the important thing is to push all of our spheres forward as much as possible while not leaving any gaps. So the first sphere is pushed forward as much as possible while still containing the near plane. Then we compute the far intersection of this sphere with the frustum, and the second sphere is pushed forward as mush as possible while still containing that intersection, and so on.

In the shader we just check the shadow-space coordinate for each cascade to see if we need to move to the next one, so you always choose the higher res map when there’s overlap.

]]> By: Dennis Steller http://the-witness.net/news/?p=77&cpage=1#comment-152 Dennis Steller Fri, 19 Mar 2010 02:49:46 +0000 http://the-witness.net/news/?p=77#comment-152 It's really good to see this kind of information up. I dont understand anything written in this entry but knowing that your taking so much care with the development of your next game is enough assurance for me. I loved how Braid make me think. And hearing about the creation of your next game makes me excited. Keep it up! I know this game will change the way I play games the same way Braid did! It’s really good to see this kind of information up. I dont understand anything written in this entry but knowing that your taking so much care with the development of your next game is enough assurance for me. I loved how Braid make me think. And hearing about the creation of your next game makes me excited. Keep it up! I know this game will change the way I play games the same way Braid did!

]]> By: Thomas Bowen http://the-witness.net/news/?p=77&cpage=1#comment-151 Thomas Bowen Wed, 17 Mar 2010 19:11:39 +0000 http://the-witness.net/news/?p=77#comment-151 So Jon, is the engine for this game being written from the ground up or is it based on an existing engine? Really interesting entry btw. So Jon, is the engine for this game being written from the ground up or is it based on an existing engine? Really interesting entry btw.

]]> By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-138 Jonathan Blow Tue, 16 Mar 2010 06:03:42 +0000 http://the-witness.net/news/?p=77#comment-138 It's a 16:10 monitor. 2560x1600. It’s a 16:10 monitor. 2560×1600.

]]> By: alastair john jack http://the-witness.net/news/?p=77&cpage=1#comment-137 alastair john jack Tue, 16 Mar 2010 04:27:58 +0000 http://the-witness.net/news/?p=77#comment-137 What size is your monitor Jon? It looks 4:3. What size is your monitor Jon? It looks 4:3.

]]> By: Andreas Brinck http://the-witness.net/news/?p=77&cpage=1#comment-131 Andreas Brinck Tue, 09 Mar 2010 10:56:57 +0000 http://the-witness.net/news/?p=77#comment-131 Johan Andersson at DICE has made a presentation of how the shadow map utilization was increased in their frostbite engine: http://www.slideshare.net/repii/02-g-d-c09-shadow-and-decals-frostbite-final3flat Johan Andersson at DICE has made a presentation of how the shadow map utilization was increased in their frostbite engine:

http://www.slideshare.net/repii/02-g-d-c09-shadow-and-decals-frostbite-final3flat

]]> By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-121 Jonathan Blow Fri, 05 Mar 2010 16:38:29 +0000 http://the-witness.net/news/?p=77#comment-121 The shadow maps are currently updated every frame. There's a Z bias in the shadow map depth test that is proportional to that particular shadow map's texel size. I think it is 0.7 of a texel or thereabouts. Currently we render shadows using the parts of geometry that are front-facing to the light source, so that equates to the front face's position plus 0.7 of a texel in Z. The shadow maps are currently updated every frame.

There’s a Z bias in the shadow map depth test that is proportional to that particular shadow map’s texel size. I think it is 0.7 of a texel or thereabouts. Currently we render shadows using the parts of geometry that are front-facing to the light source, so that equates to the front face’s position plus 0.7 of a texel in Z.

]]> By: SylHar http://the-witness.net/news/?p=77&cpage=1#comment-119 SylHar Fri, 05 Mar 2010 13:54:30 +0000 http://the-witness.net/news/?p=77#comment-119 <> I'm happy to hear that. As I always had disappointing result with my own shadow map. I'm using parallel split shadow map. It's working but I still have few artifact here and there. So I'm waiting impatiently for the second part. Question : how often do you update your shadow map with the movement of your camera ? What kind of bias do you use ? <>
I’m happy to hear that. As I always had disappointing result with my own shadow map.
I’m using parallel split shadow map. It’s working but I still have few artifact here and there.
So I’m waiting impatiently for the second part.

Question : how often do you update your shadow map with the movement of your camera ? What kind of bias do you use ?

]]> By: justin? http://the-witness.net/news/?p=77&cpage=1#comment-115 justin? Thu, 04 Mar 2010 16:24:12 +0000 http://the-witness.net/news/?p=77#comment-115 i just finished Braid like 3 hours ago and it is the best game ever. and yep the princes is the atomic bomb but also a girl. i think it uses the human relationship as a simbol for... well alot of stuff, it could be interpret as lot of things because the end is very open ended so people can draw conclusions that are very difirent, but there is always the interpretation of the creator and i think that is the most important one. mr. Bowl you must feel realy good that Braid has done so much in so little time, imagine how it will be interprept in the future... i didn't use a guide at all like you said and i'm glad i didn't... ohh and ofcourse i will get the Witness, i hope it comes to PSN. its looking really mystirius and that technique you talk about looks as if a human being would look at the real world, i love when developers share what goes into their game i just finished Braid like 3 hours ago and it is the best game ever. and yep the princes is the atomic bomb but also a girl. i think it uses the human relationship as a simbol for… well alot of stuff, it could be interpret as lot of things because the end is very open ended so people can draw conclusions that are very difirent, but there is always the interpretation of the creator and i think that is the most important one.

mr. Bowl you must feel realy good that Braid has done so much in so little time, imagine how it will be interprept in the future… i didn’t use a guide at all like you said and i’m glad i didn’t… ohh and ofcourse i will get the Witness, i hope it comes to PSN. its looking really mystirius and that technique you talk about looks as if a human being would look at the real world, i love when developers share what goes into their game

]]> By: Benoit FOULETIER http://the-witness.net/news/?p=77&cpage=1#comment-112 Benoit FOULETIER Thu, 04 Mar 2010 09:40:57 +0000 http://the-witness.net/news/?p=77#comment-112 Thanks Jonathan, I've been using cascading shadows for years and never knew how they were made! Very clear article, concise and complete. For those who want more of these "open development" blogs, you might be interested in Wolfire's blog that has been going on for over a year... and btw Tyler, they discussed the dilemma of spoiling the freshness just a few days ago! http://blog.wolfire.com/2010/03/The-dilemma-of-open-development-and-polish Thanks Jonathan, I’ve been using cascading shadows for years and never knew how they were made! Very clear article, concise and complete.

For those who want more of these “open development” blogs, you might be interested in Wolfire’s blog that has been going on for over a year… and btw Tyler, they discussed the dilemma of spoiling the freshness just a few days ago! http://blog.wolfire.com/2010/03/The-dilemma-of-open-development-and-polish

]]> By: Jonathan Blow http://the-witness.net/news/?p=77&cpage=1#comment-109 Jonathan Blow Thu, 04 Mar 2010 06:34:02 +0000 http://the-witness.net/news/?p=77#comment-109 It is your call! I would guess that if you follow this blog from now until the game's release, that is a *lot* of information, and there isn't any way that the game will feel totally fresh. However, I am going to try to stay away from spoilers and that kind of thing. It is your call! I would guess that if you follow this blog from now until the game’s release, that is a *lot* of information, and there isn’t any way that the game will feel totally fresh.

However, I am going to try to stay away from spoilers and that kind of thing.

]]> By: Tyler http://the-witness.net/news/?p=77&cpage=1#comment-108 Tyler Thu, 04 Mar 2010 05:54:18 +0000 http://the-witness.net/news/?p=77#comment-108 I have a question regarding following the developer blogs. I am looking forward to think game more than any other game that has ever come out. With that being said, when I play The Witness for the first time, I would like it all to feel fresh and new, I am afraid however that if I follow this blog to closely it wont feel fresh the first time I play it. I am worried I will just start to play through what I have already seen via blogs. Do you think that I don't have anything to worry about, and the experience wont be hampered by that much of a degree? I do realize that following the blog will take some of it away, but if it isn't by much I will continue to follow you blog very closely. Thank you for your time. I have a question regarding following the developer blogs.
I am looking forward to think game more than any other game that has ever come out. With that being said, when I play The Witness for the first time, I would like it all to feel fresh and new, I am afraid however that if I follow this blog to closely it wont feel fresh the first time I play it. I am worried I will just start to play through what I have already seen via blogs.
Do you think that I don’t have anything to worry about, and the experience wont be hampered by that much of a degree?

I do realize that following the blog will take some of it away, but if it isn’t by much I will continue to follow you blog very closely.

Thank you for your time.

]]> By: Benjamin DeLillo http://the-witness.net/news/?p=77&cpage=1#comment-103 Benjamin DeLillo Wed, 03 Mar 2010 23:18:27 +0000 http://the-witness.net/news/?p=77#comment-103 Very nice article, this is nearly the level of technical detail I like to see in a development blog. I hope the next article will really dive into the nuts and bolts of implementation. Very nice article, this is nearly the level of technical detail I like to see in a development blog. I hope the next article will really dive into the nuts and bolts of implementation.

]]> By: Leonard McCoy http://the-witness.net/news/?p=77&cpage=1#comment-101 Leonard McCoy Wed, 03 Mar 2010 22:28:56 +0000 http://the-witness.net/news/?p=77#comment-101 To say that I'd understood everything in this article would be a huge exaggeration. But, nonetheless, I find it hugely interesting for the same reason I read EVERY of Richard Leadbetter's tech articles: I simply want to know how the clock ticks behind the cover. Keep them coming. To say that I’d understood everything in this article would be a huge exaggeration. But, nonetheless, I find it hugely interesting for the same reason I read EVERY of Richard Leadbetter’s tech articles: I simply want to know how the clock ticks behind the cover.

Keep them coming.

]]> By: Asteao http://the-witness.net/news/?p=77&cpage=1#comment-100 Asteao Wed, 03 Mar 2010 20:25:30 +0000 http://the-witness.net/news/?p=77#comment-100 This is awesome, its cool to see developers talk about the technologies behind their games. It inspires people to look into a career in game development and it helps beginners get a grasp of what problems they may encounter and how to go about solving them. I look forward reading more about this stuff. This is awesome, its cool to see developers talk about the technologies behind their games. It inspires people to look into a career in game development and it helps beginners get a grasp of what problems they may encounter and how to go about solving them. I look forward reading more about this stuff.

]]>